Merge tag 'trace-printf-v6.13' of git://git.kernel.org/pub/scm/linux/kernel/git/trace...
[drm/drm-misc.git] / include / linux / skbuff.h
blob58009fa661026bd9a9a1448cc7fdaea29b4901ef
1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3 * Definitions for the 'struct sk_buff' memory handlers.
5 * Authors:
6 * Alan Cox, <gw4pts@gw4pts.ampr.org>
7 * Florian La Roche, <rzsfl@rz.uni-sb.de>
8 */
10 #ifndef _LINUX_SKBUFF_H
11 #define _LINUX_SKBUFF_H
13 #include <linux/kernel.h>
14 #include <linux/compiler.h>
15 #include <linux/time.h>
16 #include <linux/bug.h>
17 #include <linux/bvec.h>
18 #include <linux/cache.h>
19 #include <linux/rbtree.h>
20 #include <linux/socket.h>
21 #include <linux/refcount.h>
23 #include <linux/atomic.h>
24 #include <asm/types.h>
25 #include <linux/spinlock.h>
26 #include <net/checksum.h>
27 #include <linux/rcupdate.h>
28 #include <linux/dma-mapping.h>
29 #include <linux/netdev_features.h>
30 #include <net/flow_dissector.h>
31 #include <linux/in6.h>
32 #include <linux/if_packet.h>
33 #include <linux/llist.h>
34 #include <linux/page_frag_cache.h>
35 #include <net/flow.h>
36 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
37 #include <linux/netfilter/nf_conntrack_common.h>
38 #endif
39 #include <net/net_debug.h>
40 #include <net/dropreason-core.h>
41 #include <net/netmem.h>
43 /**
44 * DOC: skb checksums
46 * The interface for checksum offload between the stack and networking drivers
47 * is as follows...
49 * IP checksum related features
50 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
52 * Drivers advertise checksum offload capabilities in the features of a device.
53 * From the stack's point of view these are capabilities offered by the driver.
54 * A driver typically only advertises features that it is capable of offloading
55 * to its device.
57 * .. flat-table:: Checksum related device features
58 * :widths: 1 10
60 * * - %NETIF_F_HW_CSUM
61 * - The driver (or its device) is able to compute one
62 * IP (one's complement) checksum for any combination
63 * of protocols or protocol layering. The checksum is
64 * computed and set in a packet per the CHECKSUM_PARTIAL
65 * interface (see below).
67 * * - %NETIF_F_IP_CSUM
68 * - Driver (device) is only able to checksum plain
69 * TCP or UDP packets over IPv4. These are specifically
70 * unencapsulated packets of the form IPv4|TCP or
71 * IPv4|UDP where the Protocol field in the IPv4 header
72 * is TCP or UDP. The IPv4 header may contain IP options.
73 * This feature cannot be set in features for a device
74 * with NETIF_F_HW_CSUM also set. This feature is being
75 * DEPRECATED (see below).
77 * * - %NETIF_F_IPV6_CSUM
78 * - Driver (device) is only able to checksum plain
79 * TCP or UDP packets over IPv6. These are specifically
80 * unencapsulated packets of the form IPv6|TCP or
81 * IPv6|UDP where the Next Header field in the IPv6
82 * header is either TCP or UDP. IPv6 extension headers
83 * are not supported with this feature. This feature
84 * cannot be set in features for a device with
85 * NETIF_F_HW_CSUM also set. This feature is being
86 * DEPRECATED (see below).
88 * * - %NETIF_F_RXCSUM
89 * - Driver (device) performs receive checksum offload.
90 * This flag is only used to disable the RX checksum
91 * feature for a device. The stack will accept receive
92 * checksum indication in packets received on a device
93 * regardless of whether NETIF_F_RXCSUM is set.
95 * Checksumming of received packets by device
96 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
98 * Indication of checksum verification is set in &sk_buff.ip_summed.
99 * Possible values are:
101 * - %CHECKSUM_NONE
103 * Device did not checksum this packet e.g. due to lack of capabilities.
104 * The packet contains full (though not verified) checksum in packet but
105 * not in skb->csum. Thus, skb->csum is undefined in this case.
107 * - %CHECKSUM_UNNECESSARY
109 * The hardware you're dealing with doesn't calculate the full checksum
110 * (as in %CHECKSUM_COMPLETE), but it does parse headers and verify checksums
111 * for specific protocols. For such packets it will set %CHECKSUM_UNNECESSARY
112 * if their checksums are okay. &sk_buff.csum is still undefined in this case
113 * though. A driver or device must never modify the checksum field in the
114 * packet even if checksum is verified.
116 * %CHECKSUM_UNNECESSARY is applicable to following protocols:
118 * - TCP: IPv6 and IPv4.
119 * - UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
120 * zero UDP checksum for either IPv4 or IPv6, the networking stack
121 * may perform further validation in this case.
122 * - GRE: only if the checksum is present in the header.
123 * - SCTP: indicates the CRC in SCTP header has been validated.
124 * - FCOE: indicates the CRC in FC frame has been validated.
126 * &sk_buff.csum_level indicates the number of consecutive checksums found in
127 * the packet minus one that have been verified as %CHECKSUM_UNNECESSARY.
128 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
129 * and a device is able to verify the checksums for UDP (possibly zero),
130 * GRE (checksum flag is set) and TCP, &sk_buff.csum_level would be set to
131 * two. If the device were only able to verify the UDP checksum and not
132 * GRE, either because it doesn't support GRE checksum or because GRE
133 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
134 * not considered in this case).
136 * - %CHECKSUM_COMPLETE
138 * This is the most generic way. The device supplied checksum of the _whole_
139 * packet as seen by netif_rx() and fills in &sk_buff.csum. This means the
140 * hardware doesn't need to parse L3/L4 headers to implement this.
142 * Notes:
144 * - Even if device supports only some protocols, but is able to produce
145 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
146 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
148 * - %CHECKSUM_PARTIAL
150 * A checksum is set up to be offloaded to a device as described in the
151 * output description for CHECKSUM_PARTIAL. This may occur on a packet
152 * received directly from another Linux OS, e.g., a virtualized Linux kernel
153 * on the same host, or it may be set in the input path in GRO or remote
154 * checksum offload. For the purposes of checksum verification, the checksum
155 * referred to by skb->csum_start + skb->csum_offset and any preceding
156 * checksums in the packet are considered verified. Any checksums in the
157 * packet that are after the checksum being offloaded are not considered to
158 * be verified.
160 * Checksumming on transmit for non-GSO
161 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
163 * The stack requests checksum offload in the &sk_buff.ip_summed for a packet.
164 * Values are:
166 * - %CHECKSUM_PARTIAL
168 * The driver is required to checksum the packet as seen by hard_start_xmit()
169 * from &sk_buff.csum_start up to the end, and to record/write the checksum at
170 * offset &sk_buff.csum_start + &sk_buff.csum_offset.
171 * A driver may verify that the
172 * csum_start and csum_offset values are valid values given the length and
173 * offset of the packet, but it should not attempt to validate that the
174 * checksum refers to a legitimate transport layer checksum -- it is the
175 * purview of the stack to validate that csum_start and csum_offset are set
176 * correctly.
178 * When the stack requests checksum offload for a packet, the driver MUST
179 * ensure that the checksum is set correctly. A driver can either offload the
180 * checksum calculation to the device, or call skb_checksum_help (in the case
181 * that the device does not support offload for a particular checksum).
183 * %NETIF_F_IP_CSUM and %NETIF_F_IPV6_CSUM are being deprecated in favor of
184 * %NETIF_F_HW_CSUM. New devices should use %NETIF_F_HW_CSUM to indicate
185 * checksum offload capability.
186 * skb_csum_hwoffload_help() can be called to resolve %CHECKSUM_PARTIAL based
187 * on network device checksumming capabilities: if a packet does not match
188 * them, skb_checksum_help() or skb_crc32c_help() (depending on the value of
189 * &sk_buff.csum_not_inet, see :ref:`crc`)
190 * is called to resolve the checksum.
192 * - %CHECKSUM_NONE
194 * The skb was already checksummed by the protocol, or a checksum is not
195 * required.
197 * - %CHECKSUM_UNNECESSARY
199 * This has the same meaning as CHECKSUM_NONE for checksum offload on
200 * output.
202 * - %CHECKSUM_COMPLETE
204 * Not used in checksum output. If a driver observes a packet with this value
205 * set in skbuff, it should treat the packet as if %CHECKSUM_NONE were set.
207 * .. _crc:
209 * Non-IP checksum (CRC) offloads
210 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
212 * .. flat-table::
213 * :widths: 1 10
215 * * - %NETIF_F_SCTP_CRC
216 * - This feature indicates that a device is capable of
217 * offloading the SCTP CRC in a packet. To perform this offload the stack
218 * will set csum_start and csum_offset accordingly, set ip_summed to
219 * %CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication
220 * in the skbuff that the %CHECKSUM_PARTIAL refers to CRC32c.
221 * A driver that supports both IP checksum offload and SCTP CRC32c offload
222 * must verify which offload is configured for a packet by testing the
223 * value of &sk_buff.csum_not_inet; skb_crc32c_csum_help() is provided to
224 * resolve %CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
226 * * - %NETIF_F_FCOE_CRC
227 * - This feature indicates that a device is capable of offloading the FCOE
228 * CRC in a packet. To perform this offload the stack will set ip_summed
229 * to %CHECKSUM_PARTIAL and set csum_start and csum_offset
230 * accordingly. Note that there is no indication in the skbuff that the
231 * %CHECKSUM_PARTIAL refers to an FCOE checksum, so a driver that supports
232 * both IP checksum offload and FCOE CRC offload must verify which offload
233 * is configured for a packet, presumably by inspecting packet headers.
235 * Checksumming on output with GSO
236 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
238 * In the case of a GSO packet (skb_is_gso() is true), checksum offload
239 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
240 * gso_type is %SKB_GSO_TCPV4 or %SKB_GSO_TCPV6, TCP checksum offload as
241 * part of the GSO operation is implied. If a checksum is being offloaded
242 * with GSO then ip_summed is %CHECKSUM_PARTIAL, and both csum_start and
243 * csum_offset are set to refer to the outermost checksum being offloaded
244 * (two offloaded checksums are possible with UDP encapsulation).
247 /* Don't change this without changing skb_csum_unnecessary! */
248 #define CHECKSUM_NONE 0
249 #define CHECKSUM_UNNECESSARY 1
250 #define CHECKSUM_COMPLETE 2
251 #define CHECKSUM_PARTIAL 3
253 /* Maximum value in skb->csum_level */
254 #define SKB_MAX_CSUM_LEVEL 3
256 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
257 #define SKB_WITH_OVERHEAD(X) \
258 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
260 /* For X bytes available in skb->head, what is the minimal
261 * allocation needed, knowing struct skb_shared_info needs
262 * to be aligned.
264 #define SKB_HEAD_ALIGN(X) (SKB_DATA_ALIGN(X) + \
265 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
267 #define SKB_MAX_ORDER(X, ORDER) \
268 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
269 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
270 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
272 /* return minimum truesize of one skb containing X bytes of data */
273 #define SKB_TRUESIZE(X) ((X) + \
274 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
275 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
277 struct ahash_request;
278 struct net_device;
279 struct scatterlist;
280 struct pipe_inode_info;
281 struct iov_iter;
282 struct napi_struct;
283 struct bpf_prog;
284 union bpf_attr;
285 struct skb_ext;
286 struct ts_config;
288 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
289 struct nf_bridge_info {
290 enum {
291 BRNF_PROTO_UNCHANGED,
292 BRNF_PROTO_8021Q,
293 BRNF_PROTO_PPPOE
294 } orig_proto:8;
295 u8 pkt_otherhost:1;
296 u8 in_prerouting:1;
297 u8 bridged_dnat:1;
298 u8 sabotage_in_done:1;
299 __u16 frag_max_size;
300 int physinif;
302 /* always valid & non-NULL from FORWARD on, for physdev match */
303 struct net_device *physoutdev;
304 union {
305 /* prerouting: detect dnat in orig/reply direction */
306 __be32 ipv4_daddr;
307 struct in6_addr ipv6_daddr;
309 /* after prerouting + nat detected: store original source
310 * mac since neigh resolution overwrites it, only used while
311 * skb is out in neigh layer.
313 char neigh_header[8];
316 #endif
318 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
319 /* Chain in tc_skb_ext will be used to share the tc chain with
320 * ovs recirc_id. It will be set to the current chain by tc
321 * and read by ovs to recirc_id.
323 struct tc_skb_ext {
324 union {
325 u64 act_miss_cookie;
326 __u32 chain;
328 __u16 mru;
329 __u16 zone;
330 u8 post_ct:1;
331 u8 post_ct_snat:1;
332 u8 post_ct_dnat:1;
333 u8 act_miss:1; /* Set if act_miss_cookie is used */
334 u8 l2_miss:1; /* Set by bridge upon FDB or MDB miss */
336 #endif
338 struct sk_buff_head {
339 /* These two members must be first to match sk_buff. */
340 struct_group_tagged(sk_buff_list, list,
341 struct sk_buff *next;
342 struct sk_buff *prev;
345 __u32 qlen;
346 spinlock_t lock;
349 struct sk_buff;
351 #ifndef CONFIG_MAX_SKB_FRAGS
352 # define CONFIG_MAX_SKB_FRAGS 17
353 #endif
355 #define MAX_SKB_FRAGS CONFIG_MAX_SKB_FRAGS
357 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
358 * segment using its current segmentation instead.
360 #define GSO_BY_FRAGS 0xFFFF
362 typedef struct skb_frag {
363 netmem_ref netmem;
364 unsigned int len;
365 unsigned int offset;
366 } skb_frag_t;
369 * skb_frag_size() - Returns the size of a skb fragment
370 * @frag: skb fragment
372 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
374 return frag->len;
378 * skb_frag_size_set() - Sets the size of a skb fragment
379 * @frag: skb fragment
380 * @size: size of fragment
382 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
384 frag->len = size;
388 * skb_frag_size_add() - Increments the size of a skb fragment by @delta
389 * @frag: skb fragment
390 * @delta: value to add
392 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
394 frag->len += delta;
398 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta
399 * @frag: skb fragment
400 * @delta: value to subtract
402 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
404 frag->len -= delta;
408 * skb_frag_must_loop - Test if %p is a high memory page
409 * @p: fragment's page
411 static inline bool skb_frag_must_loop(struct page *p)
413 #if defined(CONFIG_HIGHMEM)
414 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p))
415 return true;
416 #endif
417 return false;
421 * skb_frag_foreach_page - loop over pages in a fragment
423 * @f: skb frag to operate on
424 * @f_off: offset from start of f->netmem
425 * @f_len: length from f_off to loop over
426 * @p: (temp var) current page
427 * @p_off: (temp var) offset from start of current page,
428 * non-zero only on first page.
429 * @p_len: (temp var) length in current page,
430 * < PAGE_SIZE only on first and last page.
431 * @copied: (temp var) length so far, excluding current p_len.
433 * A fragment can hold a compound page, in which case per-page
434 * operations, notably kmap_atomic, must be called for each
435 * regular page.
437 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
438 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
439 p_off = (f_off) & (PAGE_SIZE - 1), \
440 p_len = skb_frag_must_loop(p) ? \
441 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
442 copied = 0; \
443 copied < f_len; \
444 copied += p_len, p++, p_off = 0, \
445 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
448 * struct skb_shared_hwtstamps - hardware time stamps
449 * @hwtstamp: hardware time stamp transformed into duration
450 * since arbitrary point in time
451 * @netdev_data: address/cookie of network device driver used as
452 * reference to actual hardware time stamp
454 * Software time stamps generated by ktime_get_real() are stored in
455 * skb->tstamp.
457 * hwtstamps can only be compared against other hwtstamps from
458 * the same device.
460 * This structure is attached to packets as part of the
461 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
463 struct skb_shared_hwtstamps {
464 union {
465 ktime_t hwtstamp;
466 void *netdev_data;
470 /* Definitions for tx_flags in struct skb_shared_info */
471 enum {
472 /* generate hardware time stamp */
473 SKBTX_HW_TSTAMP = 1 << 0,
475 /* generate software time stamp when queueing packet to NIC */
476 SKBTX_SW_TSTAMP = 1 << 1,
478 /* device driver is going to provide hardware time stamp */
479 SKBTX_IN_PROGRESS = 1 << 2,
481 /* generate hardware time stamp based on cycles if supported */
482 SKBTX_HW_TSTAMP_USE_CYCLES = 1 << 3,
484 /* generate wifi status information (where possible) */
485 SKBTX_WIFI_STATUS = 1 << 4,
487 /* determine hardware time stamp based on time or cycles */
488 SKBTX_HW_TSTAMP_NETDEV = 1 << 5,
490 /* generate software time stamp when entering packet scheduling */
491 SKBTX_SCHED_TSTAMP = 1 << 6,
494 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
495 SKBTX_SCHED_TSTAMP)
496 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \
497 SKBTX_HW_TSTAMP_USE_CYCLES | \
498 SKBTX_ANY_SW_TSTAMP)
500 /* Definitions for flags in struct skb_shared_info */
501 enum {
502 /* use zcopy routines */
503 SKBFL_ZEROCOPY_ENABLE = BIT(0),
505 /* This indicates at least one fragment might be overwritten
506 * (as in vmsplice(), sendfile() ...)
507 * If we need to compute a TX checksum, we'll need to copy
508 * all frags to avoid possible bad checksum
510 SKBFL_SHARED_FRAG = BIT(1),
512 /* segment contains only zerocopy data and should not be
513 * charged to the kernel memory.
515 SKBFL_PURE_ZEROCOPY = BIT(2),
517 SKBFL_DONT_ORPHAN = BIT(3),
519 /* page references are managed by the ubuf_info, so it's safe to
520 * use frags only up until ubuf_info is released
522 SKBFL_MANAGED_FRAG_REFS = BIT(4),
525 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG)
526 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \
527 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS)
529 struct ubuf_info_ops {
530 void (*complete)(struct sk_buff *, struct ubuf_info *,
531 bool zerocopy_success);
532 /* has to be compatible with skb_zcopy_set() */
533 int (*link_skb)(struct sk_buff *skb, struct ubuf_info *uarg);
537 * The callback notifies userspace to release buffers when skb DMA is done in
538 * lower device, the skb last reference should be 0 when calling this.
539 * The zerocopy_success argument is true if zero copy transmit occurred,
540 * false on data copy or out of memory error caused by data copy attempt.
541 * The ctx field is used to track device context.
542 * The desc field is used to track userspace buffer index.
544 struct ubuf_info {
545 const struct ubuf_info_ops *ops;
546 refcount_t refcnt;
547 u8 flags;
550 struct ubuf_info_msgzc {
551 struct ubuf_info ubuf;
553 union {
554 struct {
555 unsigned long desc;
556 void *ctx;
558 struct {
559 u32 id;
560 u16 len;
561 u16 zerocopy:1;
562 u32 bytelen;
566 struct mmpin {
567 struct user_struct *user;
568 unsigned int num_pg;
569 } mmp;
572 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
573 #define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \
574 ubuf)
576 int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
577 void mm_unaccount_pinned_pages(struct mmpin *mmp);
579 /* Preserve some data across TX submission and completion.
581 * Note, this state is stored in the driver. Extending the layout
582 * might need some special care.
584 struct xsk_tx_metadata_compl {
585 __u64 *tx_timestamp;
588 /* This data is invariant across clones and lives at
589 * the end of the header data, ie. at skb->end.
591 struct skb_shared_info {
592 __u8 flags;
593 __u8 meta_len;
594 __u8 nr_frags;
595 __u8 tx_flags;
596 unsigned short gso_size;
597 /* Warning: this field is not always filled in (UFO)! */
598 unsigned short gso_segs;
599 struct sk_buff *frag_list;
600 union {
601 struct skb_shared_hwtstamps hwtstamps;
602 struct xsk_tx_metadata_compl xsk_meta;
604 unsigned int gso_type;
605 u32 tskey;
608 * Warning : all fields before dataref are cleared in __alloc_skb()
610 atomic_t dataref;
611 unsigned int xdp_frags_size;
613 /* Intermediate layers must ensure that destructor_arg
614 * remains valid until skb destructor */
615 void * destructor_arg;
617 /* must be last field, see pskb_expand_head() */
618 skb_frag_t frags[MAX_SKB_FRAGS];
622 * DOC: dataref and headerless skbs
624 * Transport layers send out clones of payload skbs they hold for
625 * retransmissions. To allow lower layers of the stack to prepend their headers
626 * we split &skb_shared_info.dataref into two halves.
627 * The lower 16 bits count the overall number of references.
628 * The higher 16 bits indicate how many of the references are payload-only.
629 * skb_header_cloned() checks if skb is allowed to add / write the headers.
631 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr
632 * (via __skb_header_release()). Any clone created from marked skb will get
633 * &sk_buff.hdr_len populated with the available headroom.
634 * If there's the only clone in existence it's able to modify the headroom
635 * at will. The sequence of calls inside the transport layer is::
637 * <alloc skb>
638 * skb_reserve()
639 * __skb_header_release()
640 * skb_clone()
641 * // send the clone down the stack
643 * This is not a very generic construct and it depends on the transport layers
644 * doing the right thing. In practice there's usually only one payload-only skb.
645 * Having multiple payload-only skbs with different lengths of hdr_len is not
646 * possible. The payload-only skbs should never leave their owner.
648 #define SKB_DATAREF_SHIFT 16
649 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
652 enum {
653 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
654 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
655 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
658 enum {
659 SKB_GSO_TCPV4 = 1 << 0,
661 /* This indicates the skb is from an untrusted source. */
662 SKB_GSO_DODGY = 1 << 1,
664 /* This indicates the tcp segment has CWR set. */
665 SKB_GSO_TCP_ECN = 1 << 2,
667 SKB_GSO_TCP_FIXEDID = 1 << 3,
669 SKB_GSO_TCPV6 = 1 << 4,
671 SKB_GSO_FCOE = 1 << 5,
673 SKB_GSO_GRE = 1 << 6,
675 SKB_GSO_GRE_CSUM = 1 << 7,
677 SKB_GSO_IPXIP4 = 1 << 8,
679 SKB_GSO_IPXIP6 = 1 << 9,
681 SKB_GSO_UDP_TUNNEL = 1 << 10,
683 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
685 SKB_GSO_PARTIAL = 1 << 12,
687 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
689 SKB_GSO_SCTP = 1 << 14,
691 SKB_GSO_ESP = 1 << 15,
693 SKB_GSO_UDP = 1 << 16,
695 SKB_GSO_UDP_L4 = 1 << 17,
697 SKB_GSO_FRAGLIST = 1 << 18,
700 #if BITS_PER_LONG > 32
701 #define NET_SKBUFF_DATA_USES_OFFSET 1
702 #endif
704 #ifdef NET_SKBUFF_DATA_USES_OFFSET
705 typedef unsigned int sk_buff_data_t;
706 #else
707 typedef unsigned char *sk_buff_data_t;
708 #endif
710 enum skb_tstamp_type {
711 SKB_CLOCK_REALTIME,
712 SKB_CLOCK_MONOTONIC,
713 SKB_CLOCK_TAI,
714 __SKB_CLOCK_MAX = SKB_CLOCK_TAI,
718 * DOC: Basic sk_buff geometry
720 * struct sk_buff itself is a metadata structure and does not hold any packet
721 * data. All the data is held in associated buffers.
723 * &sk_buff.head points to the main "head" buffer. The head buffer is divided
724 * into two parts:
726 * - data buffer, containing headers and sometimes payload;
727 * this is the part of the skb operated on by the common helpers
728 * such as skb_put() or skb_pull();
729 * - shared info (struct skb_shared_info) which holds an array of pointers
730 * to read-only data in the (page, offset, length) format.
732 * Optionally &skb_shared_info.frag_list may point to another skb.
734 * Basic diagram may look like this::
736 * ---------------
737 * | sk_buff |
738 * ---------------
739 * ,--------------------------- + head
740 * / ,----------------- + data
741 * / / ,----------- + tail
742 * | | | , + end
743 * | | | |
744 * v v v v
745 * -----------------------------------------------
746 * | headroom | data | tailroom | skb_shared_info |
747 * -----------------------------------------------
748 * + [page frag]
749 * + [page frag]
750 * + [page frag]
751 * + [page frag] ---------
752 * + frag_list --> | sk_buff |
753 * ---------
758 * struct sk_buff - socket buffer
759 * @next: Next buffer in list
760 * @prev: Previous buffer in list
761 * @tstamp: Time we arrived/left
762 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point
763 * for retransmit timer
764 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
765 * @list: queue head
766 * @ll_node: anchor in an llist (eg socket defer_list)
767 * @sk: Socket we are owned by
768 * @dev: Device we arrived on/are leaving by
769 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL
770 * @cb: Control buffer. Free for use by every layer. Put private vars here
771 * @_skb_refdst: destination entry (with norefcount bit)
772 * @len: Length of actual data
773 * @data_len: Data length
774 * @mac_len: Length of link layer header
775 * @hdr_len: writable header length of cloned skb
776 * @csum: Checksum (must include start/offset pair)
777 * @csum_start: Offset from skb->head where checksumming should start
778 * @csum_offset: Offset from csum_start where checksum should be stored
779 * @priority: Packet queueing priority
780 * @ignore_df: allow local fragmentation
781 * @cloned: Head may be cloned (check refcnt to be sure)
782 * @ip_summed: Driver fed us an IP checksum
783 * @nohdr: Payload reference only, must not modify header
784 * @pkt_type: Packet class
785 * @fclone: skbuff clone status
786 * @ipvs_property: skbuff is owned by ipvs
787 * @inner_protocol_type: whether the inner protocol is
788 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO
789 * @remcsum_offload: remote checksum offload is enabled
790 * @offload_fwd_mark: Packet was L2-forwarded in hardware
791 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
792 * @tc_skip_classify: do not classify packet. set by IFB device
793 * @tc_at_ingress: used within tc_classify to distinguish in/egress
794 * @redirected: packet was redirected by packet classifier
795 * @from_ingress: packet was redirected from the ingress path
796 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h
797 * @peeked: this packet has been seen already, so stats have been
798 * done for it, don't do them again
799 * @nf_trace: netfilter packet trace flag
800 * @protocol: Packet protocol from driver
801 * @destructor: Destruct function
802 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
803 * @_sk_redir: socket redirection information for skmsg
804 * @_nfct: Associated connection, if any (with nfctinfo bits)
805 * @skb_iif: ifindex of device we arrived on
806 * @tc_index: Traffic control index
807 * @hash: the packet hash
808 * @queue_mapping: Queue mapping for multiqueue devices
809 * @head_frag: skb was allocated from page fragments,
810 * not allocated by kmalloc() or vmalloc().
811 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
812 * @pp_recycle: mark the packet for recycling instead of freeing (implies
813 * page_pool support on driver)
814 * @active_extensions: active extensions (skb_ext_id types)
815 * @ndisc_nodetype: router type (from link layer)
816 * @ooo_okay: allow the mapping of a socket to a queue to be changed
817 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
818 * ports.
819 * @sw_hash: indicates hash was computed in software stack
820 * @wifi_acked_valid: wifi_acked was set
821 * @wifi_acked: whether frame was acked on wifi or not
822 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
823 * @encapsulation: indicates the inner headers in the skbuff are valid
824 * @encap_hdr_csum: software checksum is needed
825 * @csum_valid: checksum is already valid
826 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
827 * @csum_complete_sw: checksum was completed by software
828 * @csum_level: indicates the number of consecutive checksums found in
829 * the packet minus one that have been verified as
830 * CHECKSUM_UNNECESSARY (max 3)
831 * @unreadable: indicates that at least 1 of the fragments in this skb is
832 * unreadable.
833 * @dst_pending_confirm: need to confirm neighbour
834 * @decrypted: Decrypted SKB
835 * @slow_gro: state present at GRO time, slower prepare step required
836 * @tstamp_type: When set, skb->tstamp has the
837 * delivery_time clock base of skb->tstamp.
838 * @napi_id: id of the NAPI struct this skb came from
839 * @sender_cpu: (aka @napi_id) source CPU in XPS
840 * @alloc_cpu: CPU which did the skb allocation.
841 * @secmark: security marking
842 * @mark: Generic packet mark
843 * @reserved_tailroom: (aka @mark) number of bytes of free space available
844 * at the tail of an sk_buff
845 * @vlan_all: vlan fields (proto & tci)
846 * @vlan_proto: vlan encapsulation protocol
847 * @vlan_tci: vlan tag control information
848 * @inner_protocol: Protocol (encapsulation)
849 * @inner_ipproto: (aka @inner_protocol) stores ipproto when
850 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO;
851 * @inner_transport_header: Inner transport layer header (encapsulation)
852 * @inner_network_header: Network layer header (encapsulation)
853 * @inner_mac_header: Link layer header (encapsulation)
854 * @transport_header: Transport layer header
855 * @network_header: Network layer header
856 * @mac_header: Link layer header
857 * @kcov_handle: KCOV remote handle for remote coverage collection
858 * @tail: Tail pointer
859 * @end: End pointer
860 * @head: Head of buffer
861 * @data: Data head pointer
862 * @truesize: Buffer size
863 * @users: User count - see {datagram,tcp}.c
864 * @extensions: allocated extensions, valid if active_extensions is nonzero
867 struct sk_buff {
868 union {
869 struct {
870 /* These two members must be first to match sk_buff_head. */
871 struct sk_buff *next;
872 struct sk_buff *prev;
874 union {
875 struct net_device *dev;
876 /* Some protocols might use this space to store information,
877 * while device pointer would be NULL.
878 * UDP receive path is one user.
880 unsigned long dev_scratch;
883 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
884 struct list_head list;
885 struct llist_node ll_node;
888 struct sock *sk;
890 union {
891 ktime_t tstamp;
892 u64 skb_mstamp_ns; /* earliest departure time */
895 * This is the control buffer. It is free to use for every
896 * layer. Please put your private variables there. If you
897 * want to keep them across layers you have to do a skb_clone()
898 * first. This is owned by whoever has the skb queued ATM.
900 char cb[48] __aligned(8);
902 union {
903 struct {
904 unsigned long _skb_refdst;
905 void (*destructor)(struct sk_buff *skb);
907 struct list_head tcp_tsorted_anchor;
908 #ifdef CONFIG_NET_SOCK_MSG
909 unsigned long _sk_redir;
910 #endif
913 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
914 unsigned long _nfct;
915 #endif
916 unsigned int len,
917 data_len;
918 __u16 mac_len,
919 hdr_len;
921 /* Following fields are _not_ copied in __copy_skb_header()
922 * Note that queue_mapping is here mostly to fill a hole.
924 __u16 queue_mapping;
926 /* if you move cloned around you also must adapt those constants */
927 #ifdef __BIG_ENDIAN_BITFIELD
928 #define CLONED_MASK (1 << 7)
929 #else
930 #define CLONED_MASK 1
931 #endif
932 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset)
934 /* private: */
935 __u8 __cloned_offset[0];
936 /* public: */
937 __u8 cloned:1,
938 nohdr:1,
939 fclone:2,
940 peeked:1,
941 head_frag:1,
942 pfmemalloc:1,
943 pp_recycle:1; /* page_pool recycle indicator */
944 #ifdef CONFIG_SKB_EXTENSIONS
945 __u8 active_extensions;
946 #endif
948 /* Fields enclosed in headers group are copied
949 * using a single memcpy() in __copy_skb_header()
951 struct_group(headers,
953 /* private: */
954 __u8 __pkt_type_offset[0];
955 /* public: */
956 __u8 pkt_type:3; /* see PKT_TYPE_MAX */
957 __u8 ignore_df:1;
958 __u8 dst_pending_confirm:1;
959 __u8 ip_summed:2;
960 __u8 ooo_okay:1;
962 /* private: */
963 __u8 __mono_tc_offset[0];
964 /* public: */
965 __u8 tstamp_type:2; /* See skb_tstamp_type */
966 #ifdef CONFIG_NET_XGRESS
967 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */
968 __u8 tc_skip_classify:1;
969 #endif
970 __u8 remcsum_offload:1;
971 __u8 csum_complete_sw:1;
972 __u8 csum_level:2;
973 __u8 inner_protocol_type:1;
975 __u8 l4_hash:1;
976 __u8 sw_hash:1;
977 #ifdef CONFIG_WIRELESS
978 __u8 wifi_acked_valid:1;
979 __u8 wifi_acked:1;
980 #endif
981 __u8 no_fcs:1;
982 /* Indicates the inner headers are valid in the skbuff. */
983 __u8 encapsulation:1;
984 __u8 encap_hdr_csum:1;
985 __u8 csum_valid:1;
986 #ifdef CONFIG_IPV6_NDISC_NODETYPE
987 __u8 ndisc_nodetype:2;
988 #endif
990 #if IS_ENABLED(CONFIG_IP_VS)
991 __u8 ipvs_property:1;
992 #endif
993 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
994 __u8 nf_trace:1;
995 #endif
996 #ifdef CONFIG_NET_SWITCHDEV
997 __u8 offload_fwd_mark:1;
998 __u8 offload_l3_fwd_mark:1;
999 #endif
1000 __u8 redirected:1;
1001 #ifdef CONFIG_NET_REDIRECT
1002 __u8 from_ingress:1;
1003 #endif
1004 #ifdef CONFIG_NETFILTER_SKIP_EGRESS
1005 __u8 nf_skip_egress:1;
1006 #endif
1007 #ifdef CONFIG_SKB_DECRYPTED
1008 __u8 decrypted:1;
1009 #endif
1010 __u8 slow_gro:1;
1011 #if IS_ENABLED(CONFIG_IP_SCTP)
1012 __u8 csum_not_inet:1;
1013 #endif
1014 __u8 unreadable:1;
1015 #if defined(CONFIG_NET_SCHED) || defined(CONFIG_NET_XGRESS)
1016 __u16 tc_index; /* traffic control index */
1017 #endif
1019 u16 alloc_cpu;
1021 union {
1022 __wsum csum;
1023 struct {
1024 __u16 csum_start;
1025 __u16 csum_offset;
1028 __u32 priority;
1029 int skb_iif;
1030 __u32 hash;
1031 union {
1032 u32 vlan_all;
1033 struct {
1034 __be16 vlan_proto;
1035 __u16 vlan_tci;
1038 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
1039 union {
1040 unsigned int napi_id;
1041 unsigned int sender_cpu;
1043 #endif
1044 #ifdef CONFIG_NETWORK_SECMARK
1045 __u32 secmark;
1046 #endif
1048 union {
1049 __u32 mark;
1050 __u32 reserved_tailroom;
1053 union {
1054 __be16 inner_protocol;
1055 __u8 inner_ipproto;
1058 __u16 inner_transport_header;
1059 __u16 inner_network_header;
1060 __u16 inner_mac_header;
1062 __be16 protocol;
1063 __u16 transport_header;
1064 __u16 network_header;
1065 __u16 mac_header;
1067 #ifdef CONFIG_KCOV
1068 u64 kcov_handle;
1069 #endif
1071 ); /* end headers group */
1073 /* These elements must be at the end, see alloc_skb() for details. */
1074 sk_buff_data_t tail;
1075 sk_buff_data_t end;
1076 unsigned char *head,
1077 *data;
1078 unsigned int truesize;
1079 refcount_t users;
1081 #ifdef CONFIG_SKB_EXTENSIONS
1082 /* only usable after checking ->active_extensions != 0 */
1083 struct skb_ext *extensions;
1084 #endif
1087 /* if you move pkt_type around you also must adapt those constants */
1088 #ifdef __BIG_ENDIAN_BITFIELD
1089 #define PKT_TYPE_MAX (7 << 5)
1090 #else
1091 #define PKT_TYPE_MAX 7
1092 #endif
1093 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset)
1095 /* if you move tc_at_ingress or tstamp_type
1096 * around, you also must adapt these constants.
1098 #ifdef __BIG_ENDIAN_BITFIELD
1099 #define SKB_TSTAMP_TYPE_MASK (3 << 6)
1100 #define SKB_TSTAMP_TYPE_RSHIFT (6)
1101 #define TC_AT_INGRESS_MASK (1 << 5)
1102 #else
1103 #define SKB_TSTAMP_TYPE_MASK (3)
1104 #define TC_AT_INGRESS_MASK (1 << 2)
1105 #endif
1106 #define SKB_BF_MONO_TC_OFFSET offsetof(struct sk_buff, __mono_tc_offset)
1108 #ifdef __KERNEL__
1110 * Handling routines are only of interest to the kernel
1113 #define SKB_ALLOC_FCLONE 0x01
1114 #define SKB_ALLOC_RX 0x02
1115 #define SKB_ALLOC_NAPI 0x04
1118 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
1119 * @skb: buffer
1121 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
1123 return unlikely(skb->pfmemalloc);
1127 * skb might have a dst pointer attached, refcounted or not.
1128 * _skb_refdst low order bit is set if refcount was _not_ taken
1130 #define SKB_DST_NOREF 1UL
1131 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
1134 * skb_dst - returns skb dst_entry
1135 * @skb: buffer
1137 * Returns skb dst_entry, regardless of reference taken or not.
1139 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
1141 /* If refdst was not refcounted, check we still are in a
1142 * rcu_read_lock section
1144 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
1145 !rcu_read_lock_held() &&
1146 !rcu_read_lock_bh_held());
1147 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
1151 * skb_dst_set - sets skb dst
1152 * @skb: buffer
1153 * @dst: dst entry
1155 * Sets skb dst, assuming a reference was taken on dst and should
1156 * be released by skb_dst_drop()
1158 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
1160 skb->slow_gro |= !!dst;
1161 skb->_skb_refdst = (unsigned long)dst;
1165 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
1166 * @skb: buffer
1167 * @dst: dst entry
1169 * Sets skb dst, assuming a reference was not taken on dst.
1170 * If dst entry is cached, we do not take reference and dst_release
1171 * will be avoided by refdst_drop. If dst entry is not cached, we take
1172 * reference, so that last dst_release can destroy the dst immediately.
1174 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
1176 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
1177 skb->slow_gro |= !!dst;
1178 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
1182 * skb_dst_is_noref - Test if skb dst isn't refcounted
1183 * @skb: buffer
1185 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
1187 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
1190 /* For mangling skb->pkt_type from user space side from applications
1191 * such as nft, tc, etc, we only allow a conservative subset of
1192 * possible pkt_types to be set.
1194 static inline bool skb_pkt_type_ok(u32 ptype)
1196 return ptype <= PACKET_OTHERHOST;
1200 * skb_napi_id - Returns the skb's NAPI id
1201 * @skb: buffer
1203 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
1205 #ifdef CONFIG_NET_RX_BUSY_POLL
1206 return skb->napi_id;
1207 #else
1208 return 0;
1209 #endif
1212 static inline bool skb_wifi_acked_valid(const struct sk_buff *skb)
1214 #ifdef CONFIG_WIRELESS
1215 return skb->wifi_acked_valid;
1216 #else
1217 return 0;
1218 #endif
1222 * skb_unref - decrement the skb's reference count
1223 * @skb: buffer
1225 * Returns true if we can free the skb.
1227 static inline bool skb_unref(struct sk_buff *skb)
1229 if (unlikely(!skb))
1230 return false;
1231 if (!IS_ENABLED(CONFIG_DEBUG_NET) && likely(refcount_read(&skb->users) == 1))
1232 smp_rmb();
1233 else if (likely(!refcount_dec_and_test(&skb->users)))
1234 return false;
1236 return true;
1239 static inline bool skb_data_unref(const struct sk_buff *skb,
1240 struct skb_shared_info *shinfo)
1242 int bias;
1244 if (!skb->cloned)
1245 return true;
1247 bias = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
1249 if (atomic_read(&shinfo->dataref) == bias)
1250 smp_rmb();
1251 else if (atomic_sub_return(bias, &shinfo->dataref))
1252 return false;
1254 return true;
1257 void __fix_address sk_skb_reason_drop(struct sock *sk, struct sk_buff *skb,
1258 enum skb_drop_reason reason);
1260 static inline void
1261 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason)
1263 sk_skb_reason_drop(NULL, skb, reason);
1267 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason
1268 * @skb: buffer to free
1270 static inline void kfree_skb(struct sk_buff *skb)
1272 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED);
1275 void skb_release_head_state(struct sk_buff *skb);
1276 void kfree_skb_list_reason(struct sk_buff *segs,
1277 enum skb_drop_reason reason);
1278 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1279 void skb_tx_error(struct sk_buff *skb);
1281 static inline void kfree_skb_list(struct sk_buff *segs)
1283 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED);
1286 #ifdef CONFIG_TRACEPOINTS
1287 void consume_skb(struct sk_buff *skb);
1288 #else
1289 static inline void consume_skb(struct sk_buff *skb)
1291 return kfree_skb(skb);
1293 #endif
1295 void __consume_stateless_skb(struct sk_buff *skb);
1296 void __kfree_skb(struct sk_buff *skb);
1298 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1299 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1300 bool *fragstolen, int *delta_truesize);
1302 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1303 int node);
1304 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1305 struct sk_buff *build_skb(void *data, unsigned int frag_size);
1306 struct sk_buff *build_skb_around(struct sk_buff *skb,
1307 void *data, unsigned int frag_size);
1308 void skb_attempt_defer_free(struct sk_buff *skb);
1310 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size);
1311 struct sk_buff *slab_build_skb(void *data);
1314 * alloc_skb - allocate a network buffer
1315 * @size: size to allocate
1316 * @priority: allocation mask
1318 * This function is a convenient wrapper around __alloc_skb().
1320 static inline struct sk_buff *alloc_skb(unsigned int size,
1321 gfp_t priority)
1323 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1326 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1327 unsigned long data_len,
1328 int max_page_order,
1329 int *errcode,
1330 gfp_t gfp_mask);
1331 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1333 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
1334 struct sk_buff_fclones {
1335 struct sk_buff skb1;
1337 struct sk_buff skb2;
1339 refcount_t fclone_ref;
1343 * skb_fclone_busy - check if fclone is busy
1344 * @sk: socket
1345 * @skb: buffer
1347 * Returns true if skb is a fast clone, and its clone is not freed.
1348 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1349 * so we also check that didn't happen.
1351 static inline bool skb_fclone_busy(const struct sock *sk,
1352 const struct sk_buff *skb)
1354 const struct sk_buff_fclones *fclones;
1356 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1358 return skb->fclone == SKB_FCLONE_ORIG &&
1359 refcount_read(&fclones->fclone_ref) > 1 &&
1360 READ_ONCE(fclones->skb2.sk) == sk;
1364 * alloc_skb_fclone - allocate a network buffer from fclone cache
1365 * @size: size to allocate
1366 * @priority: allocation mask
1368 * This function is a convenient wrapper around __alloc_skb().
1370 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1371 gfp_t priority)
1373 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1376 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1377 void skb_headers_offset_update(struct sk_buff *skb, int off);
1378 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1379 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1380 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1381 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1382 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1383 gfp_t gfp_mask, bool fclone);
1384 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1385 gfp_t gfp_mask)
1387 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1390 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1391 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1392 unsigned int headroom);
1393 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom);
1394 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1395 int newtailroom, gfp_t priority);
1396 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1397 int offset, int len);
1398 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1399 int offset, int len);
1400 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1401 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1404 * skb_pad - zero pad the tail of an skb
1405 * @skb: buffer to pad
1406 * @pad: space to pad
1408 * Ensure that a buffer is followed by a padding area that is zero
1409 * filled. Used by network drivers which may DMA or transfer data
1410 * beyond the buffer end onto the wire.
1412 * May return error in out of memory cases. The skb is freed on error.
1414 static inline int skb_pad(struct sk_buff *skb, int pad)
1416 return __skb_pad(skb, pad, true);
1418 #define dev_kfree_skb(a) consume_skb(a)
1420 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1421 int offset, size_t size, size_t max_frags);
1423 struct skb_seq_state {
1424 __u32 lower_offset;
1425 __u32 upper_offset;
1426 __u32 frag_idx;
1427 __u32 stepped_offset;
1428 struct sk_buff *root_skb;
1429 struct sk_buff *cur_skb;
1430 __u8 *frag_data;
1431 __u32 frag_off;
1434 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1435 unsigned int to, struct skb_seq_state *st);
1436 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1437 struct skb_seq_state *st);
1438 void skb_abort_seq_read(struct skb_seq_state *st);
1439 int skb_copy_seq_read(struct skb_seq_state *st, int offset, void *to, int len);
1441 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1442 unsigned int to, struct ts_config *config);
1445 * Packet hash types specify the type of hash in skb_set_hash.
1447 * Hash types refer to the protocol layer addresses which are used to
1448 * construct a packet's hash. The hashes are used to differentiate or identify
1449 * flows of the protocol layer for the hash type. Hash types are either
1450 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1452 * Properties of hashes:
1454 * 1) Two packets in different flows have different hash values
1455 * 2) Two packets in the same flow should have the same hash value
1457 * A hash at a higher layer is considered to be more specific. A driver should
1458 * set the most specific hash possible.
1460 * A driver cannot indicate a more specific hash than the layer at which a hash
1461 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1463 * A driver may indicate a hash level which is less specific than the
1464 * actual layer the hash was computed on. For instance, a hash computed
1465 * at L4 may be considered an L3 hash. This should only be done if the
1466 * driver can't unambiguously determine that the HW computed the hash at
1467 * the higher layer. Note that the "should" in the second property above
1468 * permits this.
1470 enum pkt_hash_types {
1471 PKT_HASH_TYPE_NONE, /* Undefined type */
1472 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1473 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1474 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1477 static inline void skb_clear_hash(struct sk_buff *skb)
1479 skb->hash = 0;
1480 skb->sw_hash = 0;
1481 skb->l4_hash = 0;
1484 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1486 if (!skb->l4_hash)
1487 skb_clear_hash(skb);
1490 static inline void
1491 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1493 skb->l4_hash = is_l4;
1494 skb->sw_hash = is_sw;
1495 skb->hash = hash;
1498 static inline void
1499 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1501 /* Used by drivers to set hash from HW */
1502 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1505 static inline void
1506 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1508 __skb_set_hash(skb, hash, true, is_l4);
1511 u32 __skb_get_hash_symmetric_net(const struct net *net, const struct sk_buff *skb);
1513 static inline u32 __skb_get_hash_symmetric(const struct sk_buff *skb)
1515 return __skb_get_hash_symmetric_net(NULL, skb);
1518 void __skb_get_hash_net(const struct net *net, struct sk_buff *skb);
1519 u32 skb_get_poff(const struct sk_buff *skb);
1520 u32 __skb_get_poff(const struct sk_buff *skb, const void *data,
1521 const struct flow_keys_basic *keys, int hlen);
1522 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1523 const void *data, int hlen_proto);
1525 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1526 int thoff, u8 ip_proto)
1528 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1531 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1532 const struct flow_dissector_key *key,
1533 unsigned int key_count);
1535 struct bpf_flow_dissector;
1536 u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1537 __be16 proto, int nhoff, int hlen, unsigned int flags);
1539 bool __skb_flow_dissect(const struct net *net,
1540 const struct sk_buff *skb,
1541 struct flow_dissector *flow_dissector,
1542 void *target_container, const void *data,
1543 __be16 proto, int nhoff, int hlen, unsigned int flags);
1545 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1546 struct flow_dissector *flow_dissector,
1547 void *target_container, unsigned int flags)
1549 return __skb_flow_dissect(NULL, skb, flow_dissector,
1550 target_container, NULL, 0, 0, 0, flags);
1553 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1554 struct flow_keys *flow,
1555 unsigned int flags)
1557 memset(flow, 0, sizeof(*flow));
1558 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1559 flow, NULL, 0, 0, 0, flags);
1562 static inline bool
1563 skb_flow_dissect_flow_keys_basic(const struct net *net,
1564 const struct sk_buff *skb,
1565 struct flow_keys_basic *flow,
1566 const void *data, __be16 proto,
1567 int nhoff, int hlen, unsigned int flags)
1569 memset(flow, 0, sizeof(*flow));
1570 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1571 data, proto, nhoff, hlen, flags);
1574 void skb_flow_dissect_meta(const struct sk_buff *skb,
1575 struct flow_dissector *flow_dissector,
1576 void *target_container);
1578 /* Gets a skb connection tracking info, ctinfo map should be a
1579 * map of mapsize to translate enum ip_conntrack_info states
1580 * to user states.
1582 void
1583 skb_flow_dissect_ct(const struct sk_buff *skb,
1584 struct flow_dissector *flow_dissector,
1585 void *target_container,
1586 u16 *ctinfo_map, size_t mapsize,
1587 bool post_ct, u16 zone);
1588 void
1589 skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1590 struct flow_dissector *flow_dissector,
1591 void *target_container);
1593 void skb_flow_dissect_hash(const struct sk_buff *skb,
1594 struct flow_dissector *flow_dissector,
1595 void *target_container);
1597 static inline __u32 skb_get_hash_net(const struct net *net, struct sk_buff *skb)
1599 if (!skb->l4_hash && !skb->sw_hash)
1600 __skb_get_hash_net(net, skb);
1602 return skb->hash;
1605 static inline __u32 skb_get_hash(struct sk_buff *skb)
1607 if (!skb->l4_hash && !skb->sw_hash)
1608 __skb_get_hash_net(NULL, skb);
1610 return skb->hash;
1613 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1615 if (!skb->l4_hash && !skb->sw_hash) {
1616 struct flow_keys keys;
1617 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1619 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1622 return skb->hash;
1625 __u32 skb_get_hash_perturb(const struct sk_buff *skb,
1626 const siphash_key_t *perturb);
1628 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1630 return skb->hash;
1633 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1635 to->hash = from->hash;
1636 to->sw_hash = from->sw_hash;
1637 to->l4_hash = from->l4_hash;
1640 static inline int skb_cmp_decrypted(const struct sk_buff *skb1,
1641 const struct sk_buff *skb2)
1643 #ifdef CONFIG_SKB_DECRYPTED
1644 return skb2->decrypted - skb1->decrypted;
1645 #else
1646 return 0;
1647 #endif
1650 static inline bool skb_is_decrypted(const struct sk_buff *skb)
1652 #ifdef CONFIG_SKB_DECRYPTED
1653 return skb->decrypted;
1654 #else
1655 return false;
1656 #endif
1659 static inline void skb_copy_decrypted(struct sk_buff *to,
1660 const struct sk_buff *from)
1662 #ifdef CONFIG_SKB_DECRYPTED
1663 to->decrypted = from->decrypted;
1664 #endif
1667 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1668 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1670 return skb->head + skb->end;
1673 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1675 return skb->end;
1678 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1680 skb->end = offset;
1682 #else
1683 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1685 return skb->end;
1688 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1690 return skb->end - skb->head;
1693 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1695 skb->end = skb->head + offset;
1697 #endif
1699 extern const struct ubuf_info_ops msg_zerocopy_ubuf_ops;
1701 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1702 struct ubuf_info *uarg);
1704 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
1706 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk,
1707 struct sk_buff *skb, struct iov_iter *from,
1708 size_t length);
1710 int zerocopy_fill_skb_from_iter(struct sk_buff *skb,
1711 struct iov_iter *from, size_t length);
1713 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb,
1714 struct msghdr *msg, int len)
1716 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len);
1719 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1720 struct msghdr *msg, int len,
1721 struct ubuf_info *uarg);
1723 /* Internal */
1724 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1726 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1728 return &skb_shinfo(skb)->hwtstamps;
1731 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1733 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE;
1735 return is_zcopy ? skb_uarg(skb) : NULL;
1738 static inline bool skb_zcopy_pure(const struct sk_buff *skb)
1740 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY;
1743 static inline bool skb_zcopy_managed(const struct sk_buff *skb)
1745 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS;
1748 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1,
1749 const struct sk_buff *skb2)
1751 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2);
1754 static inline void net_zcopy_get(struct ubuf_info *uarg)
1756 refcount_inc(&uarg->refcnt);
1759 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg)
1761 skb_shinfo(skb)->destructor_arg = uarg;
1762 skb_shinfo(skb)->flags |= uarg->flags;
1765 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1766 bool *have_ref)
1768 if (skb && uarg && !skb_zcopy(skb)) {
1769 if (unlikely(have_ref && *have_ref))
1770 *have_ref = false;
1771 else
1772 net_zcopy_get(uarg);
1773 skb_zcopy_init(skb, uarg);
1777 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1779 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1780 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG;
1783 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1785 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1788 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1790 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1793 static inline void net_zcopy_put(struct ubuf_info *uarg)
1795 if (uarg)
1796 uarg->ops->complete(NULL, uarg, true);
1799 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1801 if (uarg) {
1802 if (uarg->ops == &msg_zerocopy_ubuf_ops)
1803 msg_zerocopy_put_abort(uarg, have_uref);
1804 else if (have_uref)
1805 net_zcopy_put(uarg);
1809 /* Release a reference on a zerocopy structure */
1810 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success)
1812 struct ubuf_info *uarg = skb_zcopy(skb);
1814 if (uarg) {
1815 if (!skb_zcopy_is_nouarg(skb))
1816 uarg->ops->complete(skb, uarg, zerocopy_success);
1818 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY;
1822 void __skb_zcopy_downgrade_managed(struct sk_buff *skb);
1824 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb)
1826 if (unlikely(skb_zcopy_managed(skb)))
1827 __skb_zcopy_downgrade_managed(skb);
1830 /* Return true if frags in this skb are readable by the host. */
1831 static inline bool skb_frags_readable(const struct sk_buff *skb)
1833 return !skb->unreadable;
1836 static inline void skb_mark_not_on_list(struct sk_buff *skb)
1838 skb->next = NULL;
1841 static inline void skb_poison_list(struct sk_buff *skb)
1843 #ifdef CONFIG_DEBUG_NET
1844 skb->next = SKB_LIST_POISON_NEXT;
1845 #endif
1848 /* Iterate through singly-linked GSO fragments of an skb. */
1849 #define skb_list_walk_safe(first, skb, next_skb) \
1850 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1851 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1853 static inline void skb_list_del_init(struct sk_buff *skb)
1855 __list_del_entry(&skb->list);
1856 skb_mark_not_on_list(skb);
1860 * skb_queue_empty - check if a queue is empty
1861 * @list: queue head
1863 * Returns true if the queue is empty, false otherwise.
1865 static inline int skb_queue_empty(const struct sk_buff_head *list)
1867 return list->next == (const struct sk_buff *) list;
1871 * skb_queue_empty_lockless - check if a queue is empty
1872 * @list: queue head
1874 * Returns true if the queue is empty, false otherwise.
1875 * This variant can be used in lockless contexts.
1877 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1879 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1884 * skb_queue_is_last - check if skb is the last entry in the queue
1885 * @list: queue head
1886 * @skb: buffer
1888 * Returns true if @skb is the last buffer on the list.
1890 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1891 const struct sk_buff *skb)
1893 return skb->next == (const struct sk_buff *) list;
1897 * skb_queue_is_first - check if skb is the first entry in the queue
1898 * @list: queue head
1899 * @skb: buffer
1901 * Returns true if @skb is the first buffer on the list.
1903 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1904 const struct sk_buff *skb)
1906 return skb->prev == (const struct sk_buff *) list;
1910 * skb_queue_next - return the next packet in the queue
1911 * @list: queue head
1912 * @skb: current buffer
1914 * Return the next packet in @list after @skb. It is only valid to
1915 * call this if skb_queue_is_last() evaluates to false.
1917 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1918 const struct sk_buff *skb)
1920 /* This BUG_ON may seem severe, but if we just return then we
1921 * are going to dereference garbage.
1923 BUG_ON(skb_queue_is_last(list, skb));
1924 return skb->next;
1928 * skb_queue_prev - return the prev packet in the queue
1929 * @list: queue head
1930 * @skb: current buffer
1932 * Return the prev packet in @list before @skb. It is only valid to
1933 * call this if skb_queue_is_first() evaluates to false.
1935 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1936 const struct sk_buff *skb)
1938 /* This BUG_ON may seem severe, but if we just return then we
1939 * are going to dereference garbage.
1941 BUG_ON(skb_queue_is_first(list, skb));
1942 return skb->prev;
1946 * skb_get - reference buffer
1947 * @skb: buffer to reference
1949 * Makes another reference to a socket buffer and returns a pointer
1950 * to the buffer.
1952 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1954 refcount_inc(&skb->users);
1955 return skb;
1959 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1963 * skb_cloned - is the buffer a clone
1964 * @skb: buffer to check
1966 * Returns true if the buffer was generated with skb_clone() and is
1967 * one of multiple shared copies of the buffer. Cloned buffers are
1968 * shared data so must not be written to under normal circumstances.
1970 static inline int skb_cloned(const struct sk_buff *skb)
1972 return skb->cloned &&
1973 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1976 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1978 might_sleep_if(gfpflags_allow_blocking(pri));
1980 if (skb_cloned(skb))
1981 return pskb_expand_head(skb, 0, 0, pri);
1983 return 0;
1986 /* This variant of skb_unclone() makes sure skb->truesize
1987 * and skb_end_offset() are not changed, whenever a new skb->head is needed.
1989 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X))
1990 * when various debugging features are in place.
1992 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri);
1993 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
1995 might_sleep_if(gfpflags_allow_blocking(pri));
1997 if (skb_cloned(skb))
1998 return __skb_unclone_keeptruesize(skb, pri);
1999 return 0;
2003 * skb_header_cloned - is the header a clone
2004 * @skb: buffer to check
2006 * Returns true if modifying the header part of the buffer requires
2007 * the data to be copied.
2009 static inline int skb_header_cloned(const struct sk_buff *skb)
2011 int dataref;
2013 if (!skb->cloned)
2014 return 0;
2016 dataref = atomic_read(&skb_shinfo(skb)->dataref);
2017 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
2018 return dataref != 1;
2021 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
2023 might_sleep_if(gfpflags_allow_blocking(pri));
2025 if (skb_header_cloned(skb))
2026 return pskb_expand_head(skb, 0, 0, pri);
2028 return 0;
2032 * __skb_header_release() - allow clones to use the headroom
2033 * @skb: buffer to operate on
2035 * See "DOC: dataref and headerless skbs".
2037 static inline void __skb_header_release(struct sk_buff *skb)
2039 skb->nohdr = 1;
2040 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
2045 * skb_shared - is the buffer shared
2046 * @skb: buffer to check
2048 * Returns true if more than one person has a reference to this
2049 * buffer.
2051 static inline int skb_shared(const struct sk_buff *skb)
2053 return refcount_read(&skb->users) != 1;
2057 * skb_share_check - check if buffer is shared and if so clone it
2058 * @skb: buffer to check
2059 * @pri: priority for memory allocation
2061 * If the buffer is shared the buffer is cloned and the old copy
2062 * drops a reference. A new clone with a single reference is returned.
2063 * If the buffer is not shared the original buffer is returned. When
2064 * being called from interrupt status or with spinlocks held pri must
2065 * be GFP_ATOMIC.
2067 * NULL is returned on a memory allocation failure.
2069 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
2071 might_sleep_if(gfpflags_allow_blocking(pri));
2072 if (skb_shared(skb)) {
2073 struct sk_buff *nskb = skb_clone(skb, pri);
2075 if (likely(nskb))
2076 consume_skb(skb);
2077 else
2078 kfree_skb(skb);
2079 skb = nskb;
2081 return skb;
2085 * Copy shared buffers into a new sk_buff. We effectively do COW on
2086 * packets to handle cases where we have a local reader and forward
2087 * and a couple of other messy ones. The normal one is tcpdumping
2088 * a packet that's being forwarded.
2092 * skb_unshare - make a copy of a shared buffer
2093 * @skb: buffer to check
2094 * @pri: priority for memory allocation
2096 * If the socket buffer is a clone then this function creates a new
2097 * copy of the data, drops a reference count on the old copy and returns
2098 * the new copy with the reference count at 1. If the buffer is not a clone
2099 * the original buffer is returned. When called with a spinlock held or
2100 * from interrupt state @pri must be %GFP_ATOMIC
2102 * %NULL is returned on a memory allocation failure.
2104 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
2105 gfp_t pri)
2107 might_sleep_if(gfpflags_allow_blocking(pri));
2108 if (skb_cloned(skb)) {
2109 struct sk_buff *nskb = skb_copy(skb, pri);
2111 /* Free our shared copy */
2112 if (likely(nskb))
2113 consume_skb(skb);
2114 else
2115 kfree_skb(skb);
2116 skb = nskb;
2118 return skb;
2122 * skb_peek - peek at the head of an &sk_buff_head
2123 * @list_: list to peek at
2125 * Peek an &sk_buff. Unlike most other operations you _MUST_
2126 * be careful with this one. A peek leaves the buffer on the
2127 * list and someone else may run off with it. You must hold
2128 * the appropriate locks or have a private queue to do this.
2130 * Returns %NULL for an empty list or a pointer to the head element.
2131 * The reference count is not incremented and the reference is therefore
2132 * volatile. Use with caution.
2134 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
2136 struct sk_buff *skb = list_->next;
2138 if (skb == (struct sk_buff *)list_)
2139 skb = NULL;
2140 return skb;
2144 * __skb_peek - peek at the head of a non-empty &sk_buff_head
2145 * @list_: list to peek at
2147 * Like skb_peek(), but the caller knows that the list is not empty.
2149 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
2151 return list_->next;
2155 * skb_peek_next - peek skb following the given one from a queue
2156 * @skb: skb to start from
2157 * @list_: list to peek at
2159 * Returns %NULL when the end of the list is met or a pointer to the
2160 * next element. The reference count is not incremented and the
2161 * reference is therefore volatile. Use with caution.
2163 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
2164 const struct sk_buff_head *list_)
2166 struct sk_buff *next = skb->next;
2168 if (next == (struct sk_buff *)list_)
2169 next = NULL;
2170 return next;
2174 * skb_peek_tail - peek at the tail of an &sk_buff_head
2175 * @list_: list to peek at
2177 * Peek an &sk_buff. Unlike most other operations you _MUST_
2178 * be careful with this one. A peek leaves the buffer on the
2179 * list and someone else may run off with it. You must hold
2180 * the appropriate locks or have a private queue to do this.
2182 * Returns %NULL for an empty list or a pointer to the tail element.
2183 * The reference count is not incremented and the reference is therefore
2184 * volatile. Use with caution.
2186 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
2188 struct sk_buff *skb = READ_ONCE(list_->prev);
2190 if (skb == (struct sk_buff *)list_)
2191 skb = NULL;
2192 return skb;
2197 * skb_queue_len - get queue length
2198 * @list_: list to measure
2200 * Return the length of an &sk_buff queue.
2202 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
2204 return list_->qlen;
2208 * skb_queue_len_lockless - get queue length
2209 * @list_: list to measure
2211 * Return the length of an &sk_buff queue.
2212 * This variant can be used in lockless contexts.
2214 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_)
2216 return READ_ONCE(list_->qlen);
2220 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
2221 * @list: queue to initialize
2223 * This initializes only the list and queue length aspects of
2224 * an sk_buff_head object. This allows to initialize the list
2225 * aspects of an sk_buff_head without reinitializing things like
2226 * the spinlock. It can also be used for on-stack sk_buff_head
2227 * objects where the spinlock is known to not be used.
2229 static inline void __skb_queue_head_init(struct sk_buff_head *list)
2231 list->prev = list->next = (struct sk_buff *)list;
2232 list->qlen = 0;
2236 * This function creates a split out lock class for each invocation;
2237 * this is needed for now since a whole lot of users of the skb-queue
2238 * infrastructure in drivers have different locking usage (in hardirq)
2239 * than the networking core (in softirq only). In the long run either the
2240 * network layer or drivers should need annotation to consolidate the
2241 * main types of usage into 3 classes.
2243 static inline void skb_queue_head_init(struct sk_buff_head *list)
2245 spin_lock_init(&list->lock);
2246 __skb_queue_head_init(list);
2249 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
2250 struct lock_class_key *class)
2252 skb_queue_head_init(list);
2253 lockdep_set_class(&list->lock, class);
2257 * Insert an sk_buff on a list.
2259 * The "__skb_xxxx()" functions are the non-atomic ones that
2260 * can only be called with interrupts disabled.
2262 static inline void __skb_insert(struct sk_buff *newsk,
2263 struct sk_buff *prev, struct sk_buff *next,
2264 struct sk_buff_head *list)
2266 /* See skb_queue_empty_lockless() and skb_peek_tail()
2267 * for the opposite READ_ONCE()
2269 WRITE_ONCE(newsk->next, next);
2270 WRITE_ONCE(newsk->prev, prev);
2271 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk);
2272 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk);
2273 WRITE_ONCE(list->qlen, list->qlen + 1);
2276 static inline void __skb_queue_splice(const struct sk_buff_head *list,
2277 struct sk_buff *prev,
2278 struct sk_buff *next)
2280 struct sk_buff *first = list->next;
2281 struct sk_buff *last = list->prev;
2283 WRITE_ONCE(first->prev, prev);
2284 WRITE_ONCE(prev->next, first);
2286 WRITE_ONCE(last->next, next);
2287 WRITE_ONCE(next->prev, last);
2291 * skb_queue_splice - join two skb lists, this is designed for stacks
2292 * @list: the new list to add
2293 * @head: the place to add it in the first list
2295 static inline void skb_queue_splice(const struct sk_buff_head *list,
2296 struct sk_buff_head *head)
2298 if (!skb_queue_empty(list)) {
2299 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2300 head->qlen += list->qlen;
2305 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
2306 * @list: the new list to add
2307 * @head: the place to add it in the first list
2309 * The list at @list is reinitialised
2311 static inline void skb_queue_splice_init(struct sk_buff_head *list,
2312 struct sk_buff_head *head)
2314 if (!skb_queue_empty(list)) {
2315 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2316 head->qlen += list->qlen;
2317 __skb_queue_head_init(list);
2322 * skb_queue_splice_tail - join two skb lists, each list being a queue
2323 * @list: the new list to add
2324 * @head: the place to add it in the first list
2326 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
2327 struct sk_buff_head *head)
2329 if (!skb_queue_empty(list)) {
2330 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2331 head->qlen += list->qlen;
2336 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
2337 * @list: the new list to add
2338 * @head: the place to add it in the first list
2340 * Each of the lists is a queue.
2341 * The list at @list is reinitialised
2343 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
2344 struct sk_buff_head *head)
2346 if (!skb_queue_empty(list)) {
2347 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2348 head->qlen += list->qlen;
2349 __skb_queue_head_init(list);
2354 * __skb_queue_after - queue a buffer at the list head
2355 * @list: list to use
2356 * @prev: place after this buffer
2357 * @newsk: buffer to queue
2359 * Queue a buffer int the middle of a list. This function takes no locks
2360 * and you must therefore hold required locks before calling it.
2362 * A buffer cannot be placed on two lists at the same time.
2364 static inline void __skb_queue_after(struct sk_buff_head *list,
2365 struct sk_buff *prev,
2366 struct sk_buff *newsk)
2368 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list);
2371 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
2372 struct sk_buff_head *list);
2374 static inline void __skb_queue_before(struct sk_buff_head *list,
2375 struct sk_buff *next,
2376 struct sk_buff *newsk)
2378 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list);
2382 * __skb_queue_head - queue a buffer at the list head
2383 * @list: list to use
2384 * @newsk: buffer to queue
2386 * Queue a buffer at the start of a list. This function takes no locks
2387 * and you must therefore hold required locks before calling it.
2389 * A buffer cannot be placed on two lists at the same time.
2391 static inline void __skb_queue_head(struct sk_buff_head *list,
2392 struct sk_buff *newsk)
2394 __skb_queue_after(list, (struct sk_buff *)list, newsk);
2396 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
2399 * __skb_queue_tail - queue a buffer at the list tail
2400 * @list: list to use
2401 * @newsk: buffer to queue
2403 * Queue a buffer at the end of a list. This function takes no locks
2404 * and you must therefore hold required locks before calling it.
2406 * A buffer cannot be placed on two lists at the same time.
2408 static inline void __skb_queue_tail(struct sk_buff_head *list,
2409 struct sk_buff *newsk)
2411 __skb_queue_before(list, (struct sk_buff *)list, newsk);
2413 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
2416 * remove sk_buff from list. _Must_ be called atomically, and with
2417 * the list known..
2419 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2420 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2422 struct sk_buff *next, *prev;
2424 WRITE_ONCE(list->qlen, list->qlen - 1);
2425 next = skb->next;
2426 prev = skb->prev;
2427 skb->next = skb->prev = NULL;
2428 WRITE_ONCE(next->prev, prev);
2429 WRITE_ONCE(prev->next, next);
2433 * __skb_dequeue - remove from the head of the queue
2434 * @list: list to dequeue from
2436 * Remove the head of the list. This function does not take any locks
2437 * so must be used with appropriate locks held only. The head item is
2438 * returned or %NULL if the list is empty.
2440 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2442 struct sk_buff *skb = skb_peek(list);
2443 if (skb)
2444 __skb_unlink(skb, list);
2445 return skb;
2447 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2450 * __skb_dequeue_tail - remove from the tail of the queue
2451 * @list: list to dequeue from
2453 * Remove the tail of the list. This function does not take any locks
2454 * so must be used with appropriate locks held only. The tail item is
2455 * returned or %NULL if the list is empty.
2457 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2459 struct sk_buff *skb = skb_peek_tail(list);
2460 if (skb)
2461 __skb_unlink(skb, list);
2462 return skb;
2464 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2467 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2469 return skb->data_len;
2472 static inline unsigned int skb_headlen(const struct sk_buff *skb)
2474 return skb->len - skb->data_len;
2477 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2479 unsigned int i, len = 0;
2481 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2482 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2483 return len;
2486 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2488 return skb_headlen(skb) + __skb_pagelen(skb);
2491 static inline void skb_frag_fill_netmem_desc(skb_frag_t *frag,
2492 netmem_ref netmem, int off,
2493 int size)
2495 frag->netmem = netmem;
2496 frag->offset = off;
2497 skb_frag_size_set(frag, size);
2500 static inline void skb_frag_fill_page_desc(skb_frag_t *frag,
2501 struct page *page,
2502 int off, int size)
2504 skb_frag_fill_netmem_desc(frag, page_to_netmem(page), off, size);
2507 static inline void __skb_fill_netmem_desc_noacc(struct skb_shared_info *shinfo,
2508 int i, netmem_ref netmem,
2509 int off, int size)
2511 skb_frag_t *frag = &shinfo->frags[i];
2513 skb_frag_fill_netmem_desc(frag, netmem, off, size);
2516 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo,
2517 int i, struct page *page,
2518 int off, int size)
2520 __skb_fill_netmem_desc_noacc(shinfo, i, page_to_netmem(page), off,
2521 size);
2525 * skb_len_add - adds a number to len fields of skb
2526 * @skb: buffer to add len to
2527 * @delta: number of bytes to add
2529 static inline void skb_len_add(struct sk_buff *skb, int delta)
2531 skb->len += delta;
2532 skb->data_len += delta;
2533 skb->truesize += delta;
2537 * __skb_fill_netmem_desc - initialise a fragment in an skb
2538 * @skb: buffer containing fragment to be initialised
2539 * @i: fragment index to initialise
2540 * @netmem: the netmem to use for this fragment
2541 * @off: the offset to the data with @page
2542 * @size: the length of the data
2544 * Initialises the @i'th fragment of @skb to point to &size bytes at
2545 * offset @off within @page.
2547 * Does not take any additional reference on the fragment.
2549 static inline void __skb_fill_netmem_desc(struct sk_buff *skb, int i,
2550 netmem_ref netmem, int off, int size)
2552 struct page *page;
2554 __skb_fill_netmem_desc_noacc(skb_shinfo(skb), i, netmem, off, size);
2556 if (netmem_is_net_iov(netmem)) {
2557 skb->unreadable = true;
2558 return;
2561 page = netmem_to_page(netmem);
2563 /* Propagate page pfmemalloc to the skb if we can. The problem is
2564 * that not all callers have unique ownership of the page but rely
2565 * on page_is_pfmemalloc doing the right thing(tm).
2567 page = compound_head(page);
2568 if (page_is_pfmemalloc(page))
2569 skb->pfmemalloc = true;
2572 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2573 struct page *page, int off, int size)
2575 __skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size);
2578 static inline void skb_fill_netmem_desc(struct sk_buff *skb, int i,
2579 netmem_ref netmem, int off, int size)
2581 __skb_fill_netmem_desc(skb, i, netmem, off, size);
2582 skb_shinfo(skb)->nr_frags = i + 1;
2586 * skb_fill_page_desc - initialise a paged fragment in an skb
2587 * @skb: buffer containing fragment to be initialised
2588 * @i: paged fragment index to initialise
2589 * @page: the page to use for this fragment
2590 * @off: the offset to the data with @page
2591 * @size: the length of the data
2593 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2594 * @skb to point to @size bytes at offset @off within @page. In
2595 * addition updates @skb such that @i is the last fragment.
2597 * Does not take any additional reference on the fragment.
2599 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2600 struct page *page, int off, int size)
2602 skb_fill_netmem_desc(skb, i, page_to_netmem(page), off, size);
2606 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb
2607 * @skb: buffer containing fragment to be initialised
2608 * @i: paged fragment index to initialise
2609 * @page: the page to use for this fragment
2610 * @off: the offset to the data with @page
2611 * @size: the length of the data
2613 * Variant of skb_fill_page_desc() which does not deal with
2614 * pfmemalloc, if page is not owned by us.
2616 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i,
2617 struct page *page, int off,
2618 int size)
2620 struct skb_shared_info *shinfo = skb_shinfo(skb);
2622 __skb_fill_page_desc_noacc(shinfo, i, page, off, size);
2623 shinfo->nr_frags = i + 1;
2626 void skb_add_rx_frag_netmem(struct sk_buff *skb, int i, netmem_ref netmem,
2627 int off, int size, unsigned int truesize);
2629 static inline void skb_add_rx_frag(struct sk_buff *skb, int i,
2630 struct page *page, int off, int size,
2631 unsigned int truesize)
2633 skb_add_rx_frag_netmem(skb, i, page_to_netmem(page), off, size,
2634 truesize);
2637 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2638 unsigned int truesize);
2640 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2642 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2643 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2645 return skb->head + skb->tail;
2648 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2650 skb->tail = skb->data - skb->head;
2653 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2655 skb_reset_tail_pointer(skb);
2656 skb->tail += offset;
2659 #else /* NET_SKBUFF_DATA_USES_OFFSET */
2660 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2662 return skb->tail;
2665 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2667 skb->tail = skb->data;
2670 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2672 skb->tail = skb->data + offset;
2675 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
2677 static inline void skb_assert_len(struct sk_buff *skb)
2679 #ifdef CONFIG_DEBUG_NET
2680 if (WARN_ONCE(!skb->len, "%s\n", __func__))
2681 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
2682 #endif /* CONFIG_DEBUG_NET */
2685 #if defined(CONFIG_FAIL_SKB_REALLOC)
2686 void skb_might_realloc(struct sk_buff *skb);
2687 #else
2688 static inline void skb_might_realloc(struct sk_buff *skb) {}
2689 #endif
2692 * Add data to an sk_buff
2694 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2695 void *skb_put(struct sk_buff *skb, unsigned int len);
2696 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2698 void *tmp = skb_tail_pointer(skb);
2699 SKB_LINEAR_ASSERT(skb);
2700 skb->tail += len;
2701 skb->len += len;
2702 return tmp;
2705 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2707 void *tmp = __skb_put(skb, len);
2709 memset(tmp, 0, len);
2710 return tmp;
2713 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2714 unsigned int len)
2716 void *tmp = __skb_put(skb, len);
2718 memcpy(tmp, data, len);
2719 return tmp;
2722 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2724 *(u8 *)__skb_put(skb, 1) = val;
2727 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2729 void *tmp = skb_put(skb, len);
2731 memset(tmp, 0, len);
2733 return tmp;
2736 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2737 unsigned int len)
2739 void *tmp = skb_put(skb, len);
2741 memcpy(tmp, data, len);
2743 return tmp;
2746 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2748 *(u8 *)skb_put(skb, 1) = val;
2751 void *skb_push(struct sk_buff *skb, unsigned int len);
2752 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2754 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX);
2756 skb->data -= len;
2757 skb->len += len;
2758 return skb->data;
2761 void *skb_pull(struct sk_buff *skb, unsigned int len);
2762 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2764 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX);
2766 skb->len -= len;
2767 if (unlikely(skb->len < skb->data_len)) {
2768 #if defined(CONFIG_DEBUG_NET)
2769 skb->len += len;
2770 pr_err("__skb_pull(len=%u)\n", len);
2771 skb_dump(KERN_ERR, skb, false);
2772 #endif
2773 BUG();
2775 return skb->data += len;
2778 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2780 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2783 void *skb_pull_data(struct sk_buff *skb, size_t len);
2785 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2787 static inline enum skb_drop_reason
2788 pskb_may_pull_reason(struct sk_buff *skb, unsigned int len)
2790 DEBUG_NET_WARN_ON_ONCE(len > INT_MAX);
2791 skb_might_realloc(skb);
2793 if (likely(len <= skb_headlen(skb)))
2794 return SKB_NOT_DROPPED_YET;
2796 if (unlikely(len > skb->len))
2797 return SKB_DROP_REASON_PKT_TOO_SMALL;
2799 if (unlikely(!__pskb_pull_tail(skb, len - skb_headlen(skb))))
2800 return SKB_DROP_REASON_NOMEM;
2802 return SKB_NOT_DROPPED_YET;
2805 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len)
2807 return pskb_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET;
2810 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2812 if (!pskb_may_pull(skb, len))
2813 return NULL;
2815 skb->len -= len;
2816 return skb->data += len;
2819 void skb_condense(struct sk_buff *skb);
2822 * skb_headroom - bytes at buffer head
2823 * @skb: buffer to check
2825 * Return the number of bytes of free space at the head of an &sk_buff.
2827 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2829 return skb->data - skb->head;
2833 * skb_tailroom - bytes at buffer end
2834 * @skb: buffer to check
2836 * Return the number of bytes of free space at the tail of an sk_buff
2838 static inline int skb_tailroom(const struct sk_buff *skb)
2840 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2844 * skb_availroom - bytes at buffer end
2845 * @skb: buffer to check
2847 * Return the number of bytes of free space at the tail of an sk_buff
2848 * allocated by sk_stream_alloc()
2850 static inline int skb_availroom(const struct sk_buff *skb)
2852 if (skb_is_nonlinear(skb))
2853 return 0;
2855 return skb->end - skb->tail - skb->reserved_tailroom;
2859 * skb_reserve - adjust headroom
2860 * @skb: buffer to alter
2861 * @len: bytes to move
2863 * Increase the headroom of an empty &sk_buff by reducing the tail
2864 * room. This is only allowed for an empty buffer.
2866 static inline void skb_reserve(struct sk_buff *skb, int len)
2868 skb->data += len;
2869 skb->tail += len;
2873 * skb_tailroom_reserve - adjust reserved_tailroom
2874 * @skb: buffer to alter
2875 * @mtu: maximum amount of headlen permitted
2876 * @needed_tailroom: minimum amount of reserved_tailroom
2878 * Set reserved_tailroom so that headlen can be as large as possible but
2879 * not larger than mtu and tailroom cannot be smaller than
2880 * needed_tailroom.
2881 * The required headroom should already have been reserved before using
2882 * this function.
2884 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2885 unsigned int needed_tailroom)
2887 SKB_LINEAR_ASSERT(skb);
2888 if (mtu < skb_tailroom(skb) - needed_tailroom)
2889 /* use at most mtu */
2890 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2891 else
2892 /* use up to all available space */
2893 skb->reserved_tailroom = needed_tailroom;
2896 #define ENCAP_TYPE_ETHER 0
2897 #define ENCAP_TYPE_IPPROTO 1
2899 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2900 __be16 protocol)
2902 skb->inner_protocol = protocol;
2903 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2906 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2907 __u8 ipproto)
2909 skb->inner_ipproto = ipproto;
2910 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2913 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2915 skb->inner_mac_header = skb->mac_header;
2916 skb->inner_network_header = skb->network_header;
2917 skb->inner_transport_header = skb->transport_header;
2920 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2922 return skb->mac_header != (typeof(skb->mac_header))~0U;
2925 static inline void skb_reset_mac_len(struct sk_buff *skb)
2927 if (!skb_mac_header_was_set(skb)) {
2928 DEBUG_NET_WARN_ON_ONCE(1);
2929 skb->mac_len = 0;
2930 } else {
2931 skb->mac_len = skb->network_header - skb->mac_header;
2935 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2936 *skb)
2938 return skb->head + skb->inner_transport_header;
2941 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2943 return skb_inner_transport_header(skb) - skb->data;
2946 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2948 long offset = skb->data - skb->head;
2950 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_transport_header))offset);
2951 skb->inner_transport_header = offset;
2954 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2955 const int offset)
2957 skb_reset_inner_transport_header(skb);
2958 skb->inner_transport_header += offset;
2961 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2963 return skb->head + skb->inner_network_header;
2966 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2968 long offset = skb->data - skb->head;
2970 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_network_header))offset);
2971 skb->inner_network_header = offset;
2974 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2975 const int offset)
2977 skb_reset_inner_network_header(skb);
2978 skb->inner_network_header += offset;
2981 static inline bool skb_inner_network_header_was_set(const struct sk_buff *skb)
2983 return skb->inner_network_header > 0;
2986 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2988 return skb->head + skb->inner_mac_header;
2991 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2993 long offset = skb->data - skb->head;
2995 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->inner_mac_header))offset);
2996 skb->inner_mac_header = offset;
2999 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
3000 const int offset)
3002 skb_reset_inner_mac_header(skb);
3003 skb->inner_mac_header += offset;
3005 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
3007 return skb->transport_header != (typeof(skb->transport_header))~0U;
3010 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
3012 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb));
3013 return skb->head + skb->transport_header;
3016 static inline void skb_reset_transport_header(struct sk_buff *skb)
3018 long offset = skb->data - skb->head;
3020 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->transport_header))offset);
3021 skb->transport_header = offset;
3024 static inline void skb_set_transport_header(struct sk_buff *skb,
3025 const int offset)
3027 skb_reset_transport_header(skb);
3028 skb->transport_header += offset;
3031 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
3033 return skb->head + skb->network_header;
3036 static inline void skb_reset_network_header(struct sk_buff *skb)
3038 long offset = skb->data - skb->head;
3040 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->network_header))offset);
3041 skb->network_header = offset;
3044 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
3046 skb_reset_network_header(skb);
3047 skb->network_header += offset;
3050 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
3052 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
3053 return skb->head + skb->mac_header;
3056 static inline int skb_mac_offset(const struct sk_buff *skb)
3058 return skb_mac_header(skb) - skb->data;
3061 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
3063 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
3064 return skb->network_header - skb->mac_header;
3067 static inline void skb_unset_mac_header(struct sk_buff *skb)
3069 skb->mac_header = (typeof(skb->mac_header))~0U;
3072 static inline void skb_reset_mac_header(struct sk_buff *skb)
3074 long offset = skb->data - skb->head;
3076 DEBUG_NET_WARN_ON_ONCE(offset != (typeof(skb->mac_header))offset);
3077 skb->mac_header = offset;
3080 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
3082 skb_reset_mac_header(skb);
3083 skb->mac_header += offset;
3086 static inline void skb_pop_mac_header(struct sk_buff *skb)
3088 skb->mac_header = skb->network_header;
3091 static inline void skb_probe_transport_header(struct sk_buff *skb)
3093 struct flow_keys_basic keys;
3095 if (skb_transport_header_was_set(skb))
3096 return;
3098 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
3099 NULL, 0, 0, 0, 0))
3100 skb_set_transport_header(skb, keys.control.thoff);
3103 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
3105 if (skb_mac_header_was_set(skb)) {
3106 const unsigned char *old_mac = skb_mac_header(skb);
3108 skb_set_mac_header(skb, -skb->mac_len);
3109 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
3113 /* Move the full mac header up to current network_header.
3114 * Leaves skb->data pointing at offset skb->mac_len into the mac_header.
3115 * Must be provided the complete mac header length.
3117 static inline void skb_mac_header_rebuild_full(struct sk_buff *skb, u32 full_mac_len)
3119 if (skb_mac_header_was_set(skb)) {
3120 const unsigned char *old_mac = skb_mac_header(skb);
3122 skb_set_mac_header(skb, -full_mac_len);
3123 memmove(skb_mac_header(skb), old_mac, full_mac_len);
3124 __skb_push(skb, full_mac_len - skb->mac_len);
3128 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
3130 return skb->csum_start - skb_headroom(skb);
3133 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
3135 return skb->head + skb->csum_start;
3138 static inline int skb_transport_offset(const struct sk_buff *skb)
3140 return skb_transport_header(skb) - skb->data;
3143 static inline u32 skb_network_header_len(const struct sk_buff *skb)
3145 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb));
3146 return skb->transport_header - skb->network_header;
3149 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
3151 return skb->inner_transport_header - skb->inner_network_header;
3154 static inline int skb_network_offset(const struct sk_buff *skb)
3156 return skb_network_header(skb) - skb->data;
3159 static inline int skb_inner_network_offset(const struct sk_buff *skb)
3161 return skb_inner_network_header(skb) - skb->data;
3164 static inline enum skb_drop_reason
3165 pskb_network_may_pull_reason(struct sk_buff *skb, unsigned int len)
3167 return pskb_may_pull_reason(skb, skb_network_offset(skb) + len);
3170 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
3172 return pskb_network_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET;
3176 * CPUs often take a performance hit when accessing unaligned memory
3177 * locations. The actual performance hit varies, it can be small if the
3178 * hardware handles it or large if we have to take an exception and fix it
3179 * in software.
3181 * Since an ethernet header is 14 bytes network drivers often end up with
3182 * the IP header at an unaligned offset. The IP header can be aligned by
3183 * shifting the start of the packet by 2 bytes. Drivers should do this
3184 * with:
3186 * skb_reserve(skb, NET_IP_ALIGN);
3188 * The downside to this alignment of the IP header is that the DMA is now
3189 * unaligned. On some architectures the cost of an unaligned DMA is high
3190 * and this cost outweighs the gains made by aligning the IP header.
3192 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
3193 * to be overridden.
3195 #ifndef NET_IP_ALIGN
3196 #define NET_IP_ALIGN 2
3197 #endif
3200 * The networking layer reserves some headroom in skb data (via
3201 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
3202 * the header has to grow. In the default case, if the header has to grow
3203 * 32 bytes or less we avoid the reallocation.
3205 * Unfortunately this headroom changes the DMA alignment of the resulting
3206 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
3207 * on some architectures. An architecture can override this value,
3208 * perhaps setting it to a cacheline in size (since that will maintain
3209 * cacheline alignment of the DMA). It must be a power of 2.
3211 * Various parts of the networking layer expect at least 32 bytes of
3212 * headroom, you should not reduce this.
3214 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
3215 * to reduce average number of cache lines per packet.
3216 * get_rps_cpu() for example only access one 64 bytes aligned block :
3217 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
3219 #ifndef NET_SKB_PAD
3220 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
3221 #endif
3223 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
3225 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
3227 if (WARN_ON(skb_is_nonlinear(skb)))
3228 return;
3229 skb->len = len;
3230 skb_set_tail_pointer(skb, len);
3233 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
3235 __skb_set_length(skb, len);
3238 void skb_trim(struct sk_buff *skb, unsigned int len);
3240 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
3242 if (skb->data_len)
3243 return ___pskb_trim(skb, len);
3244 __skb_trim(skb, len);
3245 return 0;
3248 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
3250 skb_might_realloc(skb);
3251 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
3255 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
3256 * @skb: buffer to alter
3257 * @len: new length
3259 * This is identical to pskb_trim except that the caller knows that
3260 * the skb is not cloned so we should never get an error due to out-
3261 * of-memory.
3263 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
3265 int err = pskb_trim(skb, len);
3266 BUG_ON(err);
3269 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
3271 unsigned int diff = len - skb->len;
3273 if (skb_tailroom(skb) < diff) {
3274 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
3275 GFP_ATOMIC);
3276 if (ret)
3277 return ret;
3279 __skb_set_length(skb, len);
3280 return 0;
3284 * skb_orphan - orphan a buffer
3285 * @skb: buffer to orphan
3287 * If a buffer currently has an owner then we call the owner's
3288 * destructor function and make the @skb unowned. The buffer continues
3289 * to exist but is no longer charged to its former owner.
3291 static inline void skb_orphan(struct sk_buff *skb)
3293 if (skb->destructor) {
3294 skb->destructor(skb);
3295 skb->destructor = NULL;
3296 skb->sk = NULL;
3297 } else {
3298 BUG_ON(skb->sk);
3303 * skb_orphan_frags - orphan the frags contained in a buffer
3304 * @skb: buffer to orphan frags from
3305 * @gfp_mask: allocation mask for replacement pages
3307 * For each frag in the SKB which needs a destructor (i.e. has an
3308 * owner) create a copy of that frag and release the original
3309 * page by calling the destructor.
3311 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
3313 if (likely(!skb_zcopy(skb)))
3314 return 0;
3315 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN)
3316 return 0;
3317 return skb_copy_ubufs(skb, gfp_mask);
3320 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
3321 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
3323 if (likely(!skb_zcopy(skb)))
3324 return 0;
3325 return skb_copy_ubufs(skb, gfp_mask);
3329 * __skb_queue_purge_reason - empty a list
3330 * @list: list to empty
3331 * @reason: drop reason
3333 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3334 * the list and one reference dropped. This function does not take the
3335 * list lock and the caller must hold the relevant locks to use it.
3337 static inline void __skb_queue_purge_reason(struct sk_buff_head *list,
3338 enum skb_drop_reason reason)
3340 struct sk_buff *skb;
3342 while ((skb = __skb_dequeue(list)) != NULL)
3343 kfree_skb_reason(skb, reason);
3346 static inline void __skb_queue_purge(struct sk_buff_head *list)
3348 __skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE);
3351 void skb_queue_purge_reason(struct sk_buff_head *list,
3352 enum skb_drop_reason reason);
3354 static inline void skb_queue_purge(struct sk_buff_head *list)
3356 skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE);
3359 unsigned int skb_rbtree_purge(struct rb_root *root);
3360 void skb_errqueue_purge(struct sk_buff_head *list);
3362 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3365 * netdev_alloc_frag - allocate a page fragment
3366 * @fragsz: fragment size
3368 * Allocates a frag from a page for receive buffer.
3369 * Uses GFP_ATOMIC allocations.
3371 static inline void *netdev_alloc_frag(unsigned int fragsz)
3373 return __netdev_alloc_frag_align(fragsz, ~0u);
3376 static inline void *netdev_alloc_frag_align(unsigned int fragsz,
3377 unsigned int align)
3379 WARN_ON_ONCE(!is_power_of_2(align));
3380 return __netdev_alloc_frag_align(fragsz, -align);
3383 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
3384 gfp_t gfp_mask);
3387 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
3388 * @dev: network device to receive on
3389 * @length: length to allocate
3391 * Allocate a new &sk_buff and assign it a usage count of one. The
3392 * buffer has unspecified headroom built in. Users should allocate
3393 * the headroom they think they need without accounting for the
3394 * built in space. The built in space is used for optimisations.
3396 * %NULL is returned if there is no free memory. Although this function
3397 * allocates memory it can be called from an interrupt.
3399 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
3400 unsigned int length)
3402 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
3405 /* legacy helper around __netdev_alloc_skb() */
3406 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
3407 gfp_t gfp_mask)
3409 return __netdev_alloc_skb(NULL, length, gfp_mask);
3412 /* legacy helper around netdev_alloc_skb() */
3413 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
3415 return netdev_alloc_skb(NULL, length);
3419 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
3420 unsigned int length, gfp_t gfp)
3422 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
3424 if (NET_IP_ALIGN && skb)
3425 skb_reserve(skb, NET_IP_ALIGN);
3426 return skb;
3429 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
3430 unsigned int length)
3432 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
3435 static inline void skb_free_frag(void *addr)
3437 page_frag_free(addr);
3440 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3442 static inline void *napi_alloc_frag(unsigned int fragsz)
3444 return __napi_alloc_frag_align(fragsz, ~0u);
3447 static inline void *napi_alloc_frag_align(unsigned int fragsz,
3448 unsigned int align)
3450 WARN_ON_ONCE(!is_power_of_2(align));
3451 return __napi_alloc_frag_align(fragsz, -align);
3454 struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int length);
3455 void napi_consume_skb(struct sk_buff *skb, int budget);
3457 void napi_skb_free_stolen_head(struct sk_buff *skb);
3458 void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason);
3461 * __dev_alloc_pages - allocate page for network Rx
3462 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3463 * @order: size of the allocation
3465 * Allocate a new page.
3467 * %NULL is returned if there is no free memory.
3469 static inline struct page *__dev_alloc_pages_noprof(gfp_t gfp_mask,
3470 unsigned int order)
3472 /* This piece of code contains several assumptions.
3473 * 1. This is for device Rx, therefore a cold page is preferred.
3474 * 2. The expectation is the user wants a compound page.
3475 * 3. If requesting a order 0 page it will not be compound
3476 * due to the check to see if order has a value in prep_new_page
3477 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
3478 * code in gfp_to_alloc_flags that should be enforcing this.
3480 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
3482 return alloc_pages_node_noprof(NUMA_NO_NODE, gfp_mask, order);
3484 #define __dev_alloc_pages(...) alloc_hooks(__dev_alloc_pages_noprof(__VA_ARGS__))
3487 * This specialized allocator has to be a macro for its allocations to be
3488 * accounted separately (to have a separate alloc_tag).
3490 #define dev_alloc_pages(_order) __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, _order)
3493 * __dev_alloc_page - allocate a page for network Rx
3494 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3496 * Allocate a new page.
3498 * %NULL is returned if there is no free memory.
3500 static inline struct page *__dev_alloc_page_noprof(gfp_t gfp_mask)
3502 return __dev_alloc_pages_noprof(gfp_mask, 0);
3504 #define __dev_alloc_page(...) alloc_hooks(__dev_alloc_page_noprof(__VA_ARGS__))
3507 * This specialized allocator has to be a macro for its allocations to be
3508 * accounted separately (to have a separate alloc_tag).
3510 #define dev_alloc_page() dev_alloc_pages(0)
3513 * dev_page_is_reusable - check whether a page can be reused for network Rx
3514 * @page: the page to test
3516 * A page shouldn't be considered for reusing/recycling if it was allocated
3517 * under memory pressure or at a distant memory node.
3519 * Returns false if this page should be returned to page allocator, true
3520 * otherwise.
3522 static inline bool dev_page_is_reusable(const struct page *page)
3524 return likely(page_to_nid(page) == numa_mem_id() &&
3525 !page_is_pfmemalloc(page));
3529 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
3530 * @page: The page that was allocated from skb_alloc_page
3531 * @skb: The skb that may need pfmemalloc set
3533 static inline void skb_propagate_pfmemalloc(const struct page *page,
3534 struct sk_buff *skb)
3536 if (page_is_pfmemalloc(page))
3537 skb->pfmemalloc = true;
3541 * skb_frag_off() - Returns the offset of a skb fragment
3542 * @frag: the paged fragment
3544 static inline unsigned int skb_frag_off(const skb_frag_t *frag)
3546 return frag->offset;
3550 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
3551 * @frag: skb fragment
3552 * @delta: value to add
3554 static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
3556 frag->offset += delta;
3560 * skb_frag_off_set() - Sets the offset of a skb fragment
3561 * @frag: skb fragment
3562 * @offset: offset of fragment
3564 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
3566 frag->offset = offset;
3570 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
3571 * @fragto: skb fragment where offset is set
3572 * @fragfrom: skb fragment offset is copied from
3574 static inline void skb_frag_off_copy(skb_frag_t *fragto,
3575 const skb_frag_t *fragfrom)
3577 fragto->offset = fragfrom->offset;
3580 /* Return: true if the skb_frag contains a net_iov. */
3581 static inline bool skb_frag_is_net_iov(const skb_frag_t *frag)
3583 return netmem_is_net_iov(frag->netmem);
3587 * skb_frag_net_iov - retrieve the net_iov referred to by fragment
3588 * @frag: the fragment
3590 * Return: the &struct net_iov associated with @frag. Returns NULL if this
3591 * frag has no associated net_iov.
3593 static inline struct net_iov *skb_frag_net_iov(const skb_frag_t *frag)
3595 if (!skb_frag_is_net_iov(frag))
3596 return NULL;
3598 return netmem_to_net_iov(frag->netmem);
3602 * skb_frag_page - retrieve the page referred to by a paged fragment
3603 * @frag: the paged fragment
3605 * Return: the &struct page associated with @frag. Returns NULL if this frag
3606 * has no associated page.
3608 static inline struct page *skb_frag_page(const skb_frag_t *frag)
3610 if (skb_frag_is_net_iov(frag))
3611 return NULL;
3613 return netmem_to_page(frag->netmem);
3617 * skb_frag_netmem - retrieve the netmem referred to by a fragment
3618 * @frag: the fragment
3620 * Return: the &netmem_ref associated with @frag.
3622 static inline netmem_ref skb_frag_netmem(const skb_frag_t *frag)
3624 return frag->netmem;
3627 int skb_pp_cow_data(struct page_pool *pool, struct sk_buff **pskb,
3628 unsigned int headroom);
3629 int skb_cow_data_for_xdp(struct page_pool *pool, struct sk_buff **pskb,
3630 struct bpf_prog *prog);
3633 * skb_frag_address - gets the address of the data contained in a paged fragment
3634 * @frag: the paged fragment buffer
3636 * Returns the address of the data within @frag. The page must already
3637 * be mapped.
3639 static inline void *skb_frag_address(const skb_frag_t *frag)
3641 if (!skb_frag_page(frag))
3642 return NULL;
3644 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
3648 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
3649 * @frag: the paged fragment buffer
3651 * Returns the address of the data within @frag. Checks that the page
3652 * is mapped and returns %NULL otherwise.
3654 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
3656 void *ptr = page_address(skb_frag_page(frag));
3657 if (unlikely(!ptr))
3658 return NULL;
3660 return ptr + skb_frag_off(frag);
3664 * skb_frag_page_copy() - sets the page in a fragment from another fragment
3665 * @fragto: skb fragment where page is set
3666 * @fragfrom: skb fragment page is copied from
3668 static inline void skb_frag_page_copy(skb_frag_t *fragto,
3669 const skb_frag_t *fragfrom)
3671 fragto->netmem = fragfrom->netmem;
3674 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3677 * skb_frag_dma_map - maps a paged fragment via the DMA API
3678 * @dev: the device to map the fragment to
3679 * @frag: the paged fragment to map
3680 * @offset: the offset within the fragment (starting at the
3681 * fragment's own offset)
3682 * @size: the number of bytes to map
3683 * @dir: the direction of the mapping (``PCI_DMA_*``)
3685 * Maps the page associated with @frag to @device.
3687 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3688 const skb_frag_t *frag,
3689 size_t offset, size_t size,
3690 enum dma_data_direction dir)
3692 return dma_map_page(dev, skb_frag_page(frag),
3693 skb_frag_off(frag) + offset, size, dir);
3696 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3697 gfp_t gfp_mask)
3699 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3703 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3704 gfp_t gfp_mask)
3706 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3711 * skb_clone_writable - is the header of a clone writable
3712 * @skb: buffer to check
3713 * @len: length up to which to write
3715 * Returns true if modifying the header part of the cloned buffer
3716 * does not requires the data to be copied.
3718 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3720 return !skb_header_cloned(skb) &&
3721 skb_headroom(skb) + len <= skb->hdr_len;
3724 static inline int skb_try_make_writable(struct sk_buff *skb,
3725 unsigned int write_len)
3727 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3728 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3731 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3732 int cloned)
3734 int delta = 0;
3736 if (headroom > skb_headroom(skb))
3737 delta = headroom - skb_headroom(skb);
3739 if (delta || cloned)
3740 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3741 GFP_ATOMIC);
3742 return 0;
3746 * skb_cow - copy header of skb when it is required
3747 * @skb: buffer to cow
3748 * @headroom: needed headroom
3750 * If the skb passed lacks sufficient headroom or its data part
3751 * is shared, data is reallocated. If reallocation fails, an error
3752 * is returned and original skb is not changed.
3754 * The result is skb with writable area skb->head...skb->tail
3755 * and at least @headroom of space at head.
3757 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3759 return __skb_cow(skb, headroom, skb_cloned(skb));
3763 * skb_cow_head - skb_cow but only making the head writable
3764 * @skb: buffer to cow
3765 * @headroom: needed headroom
3767 * This function is identical to skb_cow except that we replace the
3768 * skb_cloned check by skb_header_cloned. It should be used when
3769 * you only need to push on some header and do not need to modify
3770 * the data.
3772 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3774 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3778 * skb_padto - pad an skbuff up to a minimal size
3779 * @skb: buffer to pad
3780 * @len: minimal length
3782 * Pads up a buffer to ensure the trailing bytes exist and are
3783 * blanked. If the buffer already contains sufficient data it
3784 * is untouched. Otherwise it is extended. Returns zero on
3785 * success. The skb is freed on error.
3787 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3789 unsigned int size = skb->len;
3790 if (likely(size >= len))
3791 return 0;
3792 return skb_pad(skb, len - size);
3796 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3797 * @skb: buffer to pad
3798 * @len: minimal length
3799 * @free_on_error: free buffer on error
3801 * Pads up a buffer to ensure the trailing bytes exist and are
3802 * blanked. If the buffer already contains sufficient data it
3803 * is untouched. Otherwise it is extended. Returns zero on
3804 * success. The skb is freed on error if @free_on_error is true.
3806 static inline int __must_check __skb_put_padto(struct sk_buff *skb,
3807 unsigned int len,
3808 bool free_on_error)
3810 unsigned int size = skb->len;
3812 if (unlikely(size < len)) {
3813 len -= size;
3814 if (__skb_pad(skb, len, free_on_error))
3815 return -ENOMEM;
3816 __skb_put(skb, len);
3818 return 0;
3822 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3823 * @skb: buffer to pad
3824 * @len: minimal length
3826 * Pads up a buffer to ensure the trailing bytes exist and are
3827 * blanked. If the buffer already contains sufficient data it
3828 * is untouched. Otherwise it is extended. Returns zero on
3829 * success. The skb is freed on error.
3831 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len)
3833 return __skb_put_padto(skb, len, true);
3836 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum, struct iov_iter *i)
3837 __must_check;
3839 static inline int skb_add_data(struct sk_buff *skb,
3840 struct iov_iter *from, int copy)
3842 const int off = skb->len;
3844 if (skb->ip_summed == CHECKSUM_NONE) {
3845 __wsum csum = 0;
3846 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3847 &csum, from)) {
3848 skb->csum = csum_block_add(skb->csum, csum, off);
3849 return 0;
3851 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3852 return 0;
3854 __skb_trim(skb, off);
3855 return -EFAULT;
3858 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3859 const struct page *page, int off)
3861 if (skb_zcopy(skb))
3862 return false;
3863 if (i) {
3864 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
3866 return page == skb_frag_page(frag) &&
3867 off == skb_frag_off(frag) + skb_frag_size(frag);
3869 return false;
3872 static inline int __skb_linearize(struct sk_buff *skb)
3874 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3878 * skb_linearize - convert paged skb to linear one
3879 * @skb: buffer to linarize
3881 * If there is no free memory -ENOMEM is returned, otherwise zero
3882 * is returned and the old skb data released.
3884 static inline int skb_linearize(struct sk_buff *skb)
3886 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3890 * skb_has_shared_frag - can any frag be overwritten
3891 * @skb: buffer to test
3893 * Return true if the skb has at least one frag that might be modified
3894 * by an external entity (as in vmsplice()/sendfile())
3896 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3898 return skb_is_nonlinear(skb) &&
3899 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG;
3903 * skb_linearize_cow - make sure skb is linear and writable
3904 * @skb: buffer to process
3906 * If there is no free memory -ENOMEM is returned, otherwise zero
3907 * is returned and the old skb data released.
3909 static inline int skb_linearize_cow(struct sk_buff *skb)
3911 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3912 __skb_linearize(skb) : 0;
3915 static __always_inline void
3916 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3917 unsigned int off)
3919 if (skb->ip_summed == CHECKSUM_COMPLETE)
3920 skb->csum = csum_block_sub(skb->csum,
3921 csum_partial(start, len, 0), off);
3922 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3923 skb_checksum_start_offset(skb) < 0)
3924 skb->ip_summed = CHECKSUM_NONE;
3928 * skb_postpull_rcsum - update checksum for received skb after pull
3929 * @skb: buffer to update
3930 * @start: start of data before pull
3931 * @len: length of data pulled
3933 * After doing a pull on a received packet, you need to call this to
3934 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3935 * CHECKSUM_NONE so that it can be recomputed from scratch.
3937 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3938 const void *start, unsigned int len)
3940 if (skb->ip_summed == CHECKSUM_COMPLETE)
3941 skb->csum = wsum_negate(csum_partial(start, len,
3942 wsum_negate(skb->csum)));
3943 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3944 skb_checksum_start_offset(skb) < 0)
3945 skb->ip_summed = CHECKSUM_NONE;
3948 static __always_inline void
3949 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3950 unsigned int off)
3952 if (skb->ip_summed == CHECKSUM_COMPLETE)
3953 skb->csum = csum_block_add(skb->csum,
3954 csum_partial(start, len, 0), off);
3958 * skb_postpush_rcsum - update checksum for received skb after push
3959 * @skb: buffer to update
3960 * @start: start of data after push
3961 * @len: length of data pushed
3963 * After doing a push on a received packet, you need to call this to
3964 * update the CHECKSUM_COMPLETE checksum.
3966 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3967 const void *start, unsigned int len)
3969 __skb_postpush_rcsum(skb, start, len, 0);
3972 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3975 * skb_push_rcsum - push skb and update receive checksum
3976 * @skb: buffer to update
3977 * @len: length of data pulled
3979 * This function performs an skb_push on the packet and updates
3980 * the CHECKSUM_COMPLETE checksum. It should be used on
3981 * receive path processing instead of skb_push unless you know
3982 * that the checksum difference is zero (e.g., a valid IP header)
3983 * or you are setting ip_summed to CHECKSUM_NONE.
3985 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3987 skb_push(skb, len);
3988 skb_postpush_rcsum(skb, skb->data, len);
3989 return skb->data;
3992 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3994 * pskb_trim_rcsum - trim received skb and update checksum
3995 * @skb: buffer to trim
3996 * @len: new length
3998 * This is exactly the same as pskb_trim except that it ensures the
3999 * checksum of received packets are still valid after the operation.
4000 * It can change skb pointers.
4003 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
4005 skb_might_realloc(skb);
4006 if (likely(len >= skb->len))
4007 return 0;
4008 return pskb_trim_rcsum_slow(skb, len);
4011 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
4013 if (skb->ip_summed == CHECKSUM_COMPLETE)
4014 skb->ip_summed = CHECKSUM_NONE;
4015 __skb_trim(skb, len);
4016 return 0;
4019 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
4021 if (skb->ip_summed == CHECKSUM_COMPLETE)
4022 skb->ip_summed = CHECKSUM_NONE;
4023 return __skb_grow(skb, len);
4026 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
4027 #define skb_rb_first(root) rb_to_skb(rb_first(root))
4028 #define skb_rb_last(root) rb_to_skb(rb_last(root))
4029 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
4030 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
4032 #define skb_queue_walk(queue, skb) \
4033 for (skb = (queue)->next; \
4034 skb != (struct sk_buff *)(queue); \
4035 skb = skb->next)
4037 #define skb_queue_walk_safe(queue, skb, tmp) \
4038 for (skb = (queue)->next, tmp = skb->next; \
4039 skb != (struct sk_buff *)(queue); \
4040 skb = tmp, tmp = skb->next)
4042 #define skb_queue_walk_from(queue, skb) \
4043 for (; skb != (struct sk_buff *)(queue); \
4044 skb = skb->next)
4046 #define skb_rbtree_walk(skb, root) \
4047 for (skb = skb_rb_first(root); skb != NULL; \
4048 skb = skb_rb_next(skb))
4050 #define skb_rbtree_walk_from(skb) \
4051 for (; skb != NULL; \
4052 skb = skb_rb_next(skb))
4054 #define skb_rbtree_walk_from_safe(skb, tmp) \
4055 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
4056 skb = tmp)
4058 #define skb_queue_walk_from_safe(queue, skb, tmp) \
4059 for (tmp = skb->next; \
4060 skb != (struct sk_buff *)(queue); \
4061 skb = tmp, tmp = skb->next)
4063 #define skb_queue_reverse_walk(queue, skb) \
4064 for (skb = (queue)->prev; \
4065 skb != (struct sk_buff *)(queue); \
4066 skb = skb->prev)
4068 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
4069 for (skb = (queue)->prev, tmp = skb->prev; \
4070 skb != (struct sk_buff *)(queue); \
4071 skb = tmp, tmp = skb->prev)
4073 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
4074 for (tmp = skb->prev; \
4075 skb != (struct sk_buff *)(queue); \
4076 skb = tmp, tmp = skb->prev)
4078 static inline bool skb_has_frag_list(const struct sk_buff *skb)
4080 return skb_shinfo(skb)->frag_list != NULL;
4083 static inline void skb_frag_list_init(struct sk_buff *skb)
4085 skb_shinfo(skb)->frag_list = NULL;
4088 #define skb_walk_frags(skb, iter) \
4089 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
4092 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue,
4093 int *err, long *timeo_p,
4094 const struct sk_buff *skb);
4095 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
4096 struct sk_buff_head *queue,
4097 unsigned int flags,
4098 int *off, int *err,
4099 struct sk_buff **last);
4100 struct sk_buff *__skb_try_recv_datagram(struct sock *sk,
4101 struct sk_buff_head *queue,
4102 unsigned int flags, int *off, int *err,
4103 struct sk_buff **last);
4104 struct sk_buff *__skb_recv_datagram(struct sock *sk,
4105 struct sk_buff_head *sk_queue,
4106 unsigned int flags, int *off, int *err);
4107 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err);
4108 __poll_t datagram_poll(struct file *file, struct socket *sock,
4109 struct poll_table_struct *wait);
4110 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
4111 struct iov_iter *to, int size);
4112 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
4113 struct msghdr *msg, int size)
4115 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
4117 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
4118 struct msghdr *msg);
4119 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
4120 struct iov_iter *to, int len,
4121 struct ahash_request *hash);
4122 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
4123 struct iov_iter *from, int len);
4124 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
4125 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
4126 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
4127 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
4128 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
4129 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
4130 int len);
4131 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
4132 struct pipe_inode_info *pipe, unsigned int len,
4133 unsigned int flags);
4134 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
4135 int len);
4136 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
4137 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
4138 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
4139 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
4140 int len, int hlen);
4141 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
4142 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
4143 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
4144 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
4145 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features,
4146 unsigned int offset);
4147 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
4148 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len);
4149 int skb_ensure_writable_head_tail(struct sk_buff *skb, struct net_device *dev);
4150 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
4151 int skb_vlan_pop(struct sk_buff *skb);
4152 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
4153 int skb_eth_pop(struct sk_buff *skb);
4154 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
4155 const unsigned char *src);
4156 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
4157 int mac_len, bool ethernet);
4158 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
4159 bool ethernet);
4160 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
4161 int skb_mpls_dec_ttl(struct sk_buff *skb);
4162 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
4163 gfp_t gfp);
4165 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
4167 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
4170 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
4172 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
4175 struct skb_checksum_ops {
4176 __wsum (*update)(const void *mem, int len, __wsum wsum);
4177 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
4180 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
4182 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
4183 __wsum csum, const struct skb_checksum_ops *ops);
4184 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
4185 __wsum csum);
4187 static inline void * __must_check
4188 __skb_header_pointer(const struct sk_buff *skb, int offset, int len,
4189 const void *data, int hlen, void *buffer)
4191 if (likely(hlen - offset >= len))
4192 return (void *)data + offset;
4194 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0))
4195 return NULL;
4197 return buffer;
4200 static inline void * __must_check
4201 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
4203 return __skb_header_pointer(skb, offset, len, skb->data,
4204 skb_headlen(skb), buffer);
4207 static inline void * __must_check
4208 skb_pointer_if_linear(const struct sk_buff *skb, int offset, int len)
4210 if (likely(skb_headlen(skb) - offset >= len))
4211 return skb->data + offset;
4212 return NULL;
4216 * skb_needs_linearize - check if we need to linearize a given skb
4217 * depending on the given device features.
4218 * @skb: socket buffer to check
4219 * @features: net device features
4221 * Returns true if either:
4222 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
4223 * 2. skb is fragmented and the device does not support SG.
4225 static inline bool skb_needs_linearize(struct sk_buff *skb,
4226 netdev_features_t features)
4228 return skb_is_nonlinear(skb) &&
4229 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
4230 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
4233 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
4234 void *to,
4235 const unsigned int len)
4237 memcpy(to, skb->data, len);
4240 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
4241 const int offset, void *to,
4242 const unsigned int len)
4244 memcpy(to, skb->data + offset, len);
4247 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
4248 const void *from,
4249 const unsigned int len)
4251 memcpy(skb->data, from, len);
4254 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
4255 const int offset,
4256 const void *from,
4257 const unsigned int len)
4259 memcpy(skb->data + offset, from, len);
4262 void skb_init(void);
4264 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
4266 return skb->tstamp;
4270 * skb_get_timestamp - get timestamp from a skb
4271 * @skb: skb to get stamp from
4272 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
4274 * Timestamps are stored in the skb as offsets to a base timestamp.
4275 * This function converts the offset back to a struct timeval and stores
4276 * it in stamp.
4278 static inline void skb_get_timestamp(const struct sk_buff *skb,
4279 struct __kernel_old_timeval *stamp)
4281 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
4284 static inline void skb_get_new_timestamp(const struct sk_buff *skb,
4285 struct __kernel_sock_timeval *stamp)
4287 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4289 stamp->tv_sec = ts.tv_sec;
4290 stamp->tv_usec = ts.tv_nsec / 1000;
4293 static inline void skb_get_timestampns(const struct sk_buff *skb,
4294 struct __kernel_old_timespec *stamp)
4296 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4298 stamp->tv_sec = ts.tv_sec;
4299 stamp->tv_nsec = ts.tv_nsec;
4302 static inline void skb_get_new_timestampns(const struct sk_buff *skb,
4303 struct __kernel_timespec *stamp)
4305 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4307 stamp->tv_sec = ts.tv_sec;
4308 stamp->tv_nsec = ts.tv_nsec;
4311 static inline void __net_timestamp(struct sk_buff *skb)
4313 skb->tstamp = ktime_get_real();
4314 skb->tstamp_type = SKB_CLOCK_REALTIME;
4317 static inline ktime_t net_timedelta(ktime_t t)
4319 return ktime_sub(ktime_get_real(), t);
4322 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt,
4323 u8 tstamp_type)
4325 skb->tstamp = kt;
4327 if (kt)
4328 skb->tstamp_type = tstamp_type;
4329 else
4330 skb->tstamp_type = SKB_CLOCK_REALTIME;
4333 static inline void skb_set_delivery_type_by_clockid(struct sk_buff *skb,
4334 ktime_t kt, clockid_t clockid)
4336 u8 tstamp_type = SKB_CLOCK_REALTIME;
4338 switch (clockid) {
4339 case CLOCK_REALTIME:
4340 break;
4341 case CLOCK_MONOTONIC:
4342 tstamp_type = SKB_CLOCK_MONOTONIC;
4343 break;
4344 case CLOCK_TAI:
4345 tstamp_type = SKB_CLOCK_TAI;
4346 break;
4347 default:
4348 WARN_ON_ONCE(1);
4349 kt = 0;
4352 skb_set_delivery_time(skb, kt, tstamp_type);
4355 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key);
4357 /* It is used in the ingress path to clear the delivery_time.
4358 * If needed, set the skb->tstamp to the (rcv) timestamp.
4360 static inline void skb_clear_delivery_time(struct sk_buff *skb)
4362 if (skb->tstamp_type) {
4363 skb->tstamp_type = SKB_CLOCK_REALTIME;
4364 if (static_branch_unlikely(&netstamp_needed_key))
4365 skb->tstamp = ktime_get_real();
4366 else
4367 skb->tstamp = 0;
4371 static inline void skb_clear_tstamp(struct sk_buff *skb)
4373 if (skb->tstamp_type)
4374 return;
4376 skb->tstamp = 0;
4379 static inline ktime_t skb_tstamp(const struct sk_buff *skb)
4381 if (skb->tstamp_type)
4382 return 0;
4384 return skb->tstamp;
4387 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond)
4389 if (skb->tstamp_type != SKB_CLOCK_MONOTONIC && skb->tstamp)
4390 return skb->tstamp;
4392 if (static_branch_unlikely(&netstamp_needed_key) || cond)
4393 return ktime_get_real();
4395 return 0;
4398 static inline u8 skb_metadata_len(const struct sk_buff *skb)
4400 return skb_shinfo(skb)->meta_len;
4403 static inline void *skb_metadata_end(const struct sk_buff *skb)
4405 return skb_mac_header(skb);
4408 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
4409 const struct sk_buff *skb_b,
4410 u8 meta_len)
4412 const void *a = skb_metadata_end(skb_a);
4413 const void *b = skb_metadata_end(skb_b);
4414 u64 diffs = 0;
4416 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) ||
4417 BITS_PER_LONG != 64)
4418 goto slow;
4420 /* Using more efficient variant than plain call to memcmp(). */
4421 switch (meta_len) {
4422 #define __it(x, op) (x -= sizeof(u##op))
4423 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
4424 case 32: diffs |= __it_diff(a, b, 64);
4425 fallthrough;
4426 case 24: diffs |= __it_diff(a, b, 64);
4427 fallthrough;
4428 case 16: diffs |= __it_diff(a, b, 64);
4429 fallthrough;
4430 case 8: diffs |= __it_diff(a, b, 64);
4431 break;
4432 case 28: diffs |= __it_diff(a, b, 64);
4433 fallthrough;
4434 case 20: diffs |= __it_diff(a, b, 64);
4435 fallthrough;
4436 case 12: diffs |= __it_diff(a, b, 64);
4437 fallthrough;
4438 case 4: diffs |= __it_diff(a, b, 32);
4439 break;
4440 default:
4441 slow:
4442 return memcmp(a - meta_len, b - meta_len, meta_len);
4444 return diffs;
4447 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
4448 const struct sk_buff *skb_b)
4450 u8 len_a = skb_metadata_len(skb_a);
4451 u8 len_b = skb_metadata_len(skb_b);
4453 if (!(len_a | len_b))
4454 return false;
4456 return len_a != len_b ?
4457 true : __skb_metadata_differs(skb_a, skb_b, len_a);
4460 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
4462 skb_shinfo(skb)->meta_len = meta_len;
4465 static inline void skb_metadata_clear(struct sk_buff *skb)
4467 skb_metadata_set(skb, 0);
4470 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
4472 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
4474 void skb_clone_tx_timestamp(struct sk_buff *skb);
4475 bool skb_defer_rx_timestamp(struct sk_buff *skb);
4477 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
4479 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
4483 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
4485 return false;
4488 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
4491 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
4493 * PHY drivers may accept clones of transmitted packets for
4494 * timestamping via their phy_driver.txtstamp method. These drivers
4495 * must call this function to return the skb back to the stack with a
4496 * timestamp.
4498 * @skb: clone of the original outgoing packet
4499 * @hwtstamps: hardware time stamps
4502 void skb_complete_tx_timestamp(struct sk_buff *skb,
4503 struct skb_shared_hwtstamps *hwtstamps);
4505 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb,
4506 struct skb_shared_hwtstamps *hwtstamps,
4507 struct sock *sk, int tstype);
4510 * skb_tstamp_tx - queue clone of skb with send time stamps
4511 * @orig_skb: the original outgoing packet
4512 * @hwtstamps: hardware time stamps, may be NULL if not available
4514 * If the skb has a socket associated, then this function clones the
4515 * skb (thus sharing the actual data and optional structures), stores
4516 * the optional hardware time stamping information (if non NULL) or
4517 * generates a software time stamp (otherwise), then queues the clone
4518 * to the error queue of the socket. Errors are silently ignored.
4520 void skb_tstamp_tx(struct sk_buff *orig_skb,
4521 struct skb_shared_hwtstamps *hwtstamps);
4524 * skb_tx_timestamp() - Driver hook for transmit timestamping
4526 * Ethernet MAC Drivers should call this function in their hard_xmit()
4527 * function immediately before giving the sk_buff to the MAC hardware.
4529 * Specifically, one should make absolutely sure that this function is
4530 * called before TX completion of this packet can trigger. Otherwise
4531 * the packet could potentially already be freed.
4533 * @skb: A socket buffer.
4535 static inline void skb_tx_timestamp(struct sk_buff *skb)
4537 skb_clone_tx_timestamp(skb);
4538 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
4539 skb_tstamp_tx(skb, NULL);
4543 * skb_complete_wifi_ack - deliver skb with wifi status
4545 * @skb: the original outgoing packet
4546 * @acked: ack status
4549 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
4551 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
4552 __sum16 __skb_checksum_complete(struct sk_buff *skb);
4554 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
4556 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
4557 skb->csum_valid ||
4558 (skb->ip_summed == CHECKSUM_PARTIAL &&
4559 skb_checksum_start_offset(skb) >= 0));
4563 * skb_checksum_complete - Calculate checksum of an entire packet
4564 * @skb: packet to process
4566 * This function calculates the checksum over the entire packet plus
4567 * the value of skb->csum. The latter can be used to supply the
4568 * checksum of a pseudo header as used by TCP/UDP. It returns the
4569 * checksum.
4571 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
4572 * this function can be used to verify that checksum on received
4573 * packets. In that case the function should return zero if the
4574 * checksum is correct. In particular, this function will return zero
4575 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
4576 * hardware has already verified the correctness of the checksum.
4578 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
4580 return skb_csum_unnecessary(skb) ?
4581 0 : __skb_checksum_complete(skb);
4584 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
4586 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4587 if (skb->csum_level == 0)
4588 skb->ip_summed = CHECKSUM_NONE;
4589 else
4590 skb->csum_level--;
4594 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
4596 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4597 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
4598 skb->csum_level++;
4599 } else if (skb->ip_summed == CHECKSUM_NONE) {
4600 skb->ip_summed = CHECKSUM_UNNECESSARY;
4601 skb->csum_level = 0;
4605 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb)
4607 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4608 skb->ip_summed = CHECKSUM_NONE;
4609 skb->csum_level = 0;
4613 /* Check if we need to perform checksum complete validation.
4615 * Returns true if checksum complete is needed, false otherwise
4616 * (either checksum is unnecessary or zero checksum is allowed).
4618 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
4619 bool zero_okay,
4620 __sum16 check)
4622 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
4623 skb->csum_valid = 1;
4624 __skb_decr_checksum_unnecessary(skb);
4625 return false;
4628 return true;
4631 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly
4632 * in checksum_init.
4634 #define CHECKSUM_BREAK 76
4636 /* Unset checksum-complete
4638 * Unset checksum complete can be done when packet is being modified
4639 * (uncompressed for instance) and checksum-complete value is
4640 * invalidated.
4642 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
4644 if (skb->ip_summed == CHECKSUM_COMPLETE)
4645 skb->ip_summed = CHECKSUM_NONE;
4648 /* Validate (init) checksum based on checksum complete.
4650 * Return values:
4651 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
4652 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
4653 * checksum is stored in skb->csum for use in __skb_checksum_complete
4654 * non-zero: value of invalid checksum
4657 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
4658 bool complete,
4659 __wsum psum)
4661 if (skb->ip_summed == CHECKSUM_COMPLETE) {
4662 if (!csum_fold(csum_add(psum, skb->csum))) {
4663 skb->csum_valid = 1;
4664 return 0;
4668 skb->csum = psum;
4670 if (complete || skb->len <= CHECKSUM_BREAK) {
4671 __sum16 csum;
4673 csum = __skb_checksum_complete(skb);
4674 skb->csum_valid = !csum;
4675 return csum;
4678 return 0;
4681 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
4683 return 0;
4686 /* Perform checksum validate (init). Note that this is a macro since we only
4687 * want to calculate the pseudo header which is an input function if necessary.
4688 * First we try to validate without any computation (checksum unnecessary) and
4689 * then calculate based on checksum complete calling the function to compute
4690 * pseudo header.
4692 * Return values:
4693 * 0: checksum is validated or try to in skb_checksum_complete
4694 * non-zero: value of invalid checksum
4696 #define __skb_checksum_validate(skb, proto, complete, \
4697 zero_okay, check, compute_pseudo) \
4698 ({ \
4699 __sum16 __ret = 0; \
4700 skb->csum_valid = 0; \
4701 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
4702 __ret = __skb_checksum_validate_complete(skb, \
4703 complete, compute_pseudo(skb, proto)); \
4704 __ret; \
4707 #define skb_checksum_init(skb, proto, compute_pseudo) \
4708 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
4710 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
4711 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
4713 #define skb_checksum_validate(skb, proto, compute_pseudo) \
4714 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
4716 #define skb_checksum_validate_zero_check(skb, proto, check, \
4717 compute_pseudo) \
4718 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
4720 #define skb_checksum_simple_validate(skb) \
4721 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
4723 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
4725 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
4728 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
4730 skb->csum = ~pseudo;
4731 skb->ip_summed = CHECKSUM_COMPLETE;
4734 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \
4735 do { \
4736 if (__skb_checksum_convert_check(skb)) \
4737 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
4738 } while (0)
4740 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
4741 u16 start, u16 offset)
4743 skb->ip_summed = CHECKSUM_PARTIAL;
4744 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
4745 skb->csum_offset = offset - start;
4748 /* Update skbuf and packet to reflect the remote checksum offload operation.
4749 * When called, ptr indicates the starting point for skb->csum when
4750 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
4751 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4753 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4754 int start, int offset, bool nopartial)
4756 __wsum delta;
4758 if (!nopartial) {
4759 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4760 return;
4763 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4764 __skb_checksum_complete(skb);
4765 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4768 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4770 /* Adjust skb->csum since we changed the packet */
4771 skb->csum = csum_add(skb->csum, delta);
4774 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4776 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4777 return (void *)(skb->_nfct & NFCT_PTRMASK);
4778 #else
4779 return NULL;
4780 #endif
4783 static inline unsigned long skb_get_nfct(const struct sk_buff *skb)
4785 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4786 return skb->_nfct;
4787 #else
4788 return 0UL;
4789 #endif
4792 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct)
4794 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4795 skb->slow_gro |= !!nfct;
4796 skb->_nfct = nfct;
4797 #endif
4800 #ifdef CONFIG_SKB_EXTENSIONS
4801 enum skb_ext_id {
4802 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4803 SKB_EXT_BRIDGE_NF,
4804 #endif
4805 #ifdef CONFIG_XFRM
4806 SKB_EXT_SEC_PATH,
4807 #endif
4808 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4809 TC_SKB_EXT,
4810 #endif
4811 #if IS_ENABLED(CONFIG_MPTCP)
4812 SKB_EXT_MPTCP,
4813 #endif
4814 #if IS_ENABLED(CONFIG_MCTP_FLOWS)
4815 SKB_EXT_MCTP,
4816 #endif
4817 SKB_EXT_NUM, /* must be last */
4821 * struct skb_ext - sk_buff extensions
4822 * @refcnt: 1 on allocation, deallocated on 0
4823 * @offset: offset to add to @data to obtain extension address
4824 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4825 * @data: start of extension data, variable sized
4827 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4828 * to use 'u8' types while allowing up to 2kb worth of extension data.
4830 struct skb_ext {
4831 refcount_t refcnt;
4832 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4833 u8 chunks; /* same */
4834 char data[] __aligned(8);
4837 struct skb_ext *__skb_ext_alloc(gfp_t flags);
4838 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
4839 struct skb_ext *ext);
4840 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4841 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4842 void __skb_ext_put(struct skb_ext *ext);
4844 static inline void skb_ext_put(struct sk_buff *skb)
4846 if (skb->active_extensions)
4847 __skb_ext_put(skb->extensions);
4850 static inline void __skb_ext_copy(struct sk_buff *dst,
4851 const struct sk_buff *src)
4853 dst->active_extensions = src->active_extensions;
4855 if (src->active_extensions) {
4856 struct skb_ext *ext = src->extensions;
4858 refcount_inc(&ext->refcnt);
4859 dst->extensions = ext;
4863 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4865 skb_ext_put(dst);
4866 __skb_ext_copy(dst, src);
4869 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4871 return !!ext->offset[i];
4874 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4876 return skb->active_extensions & (1 << id);
4879 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4881 if (skb_ext_exist(skb, id))
4882 __skb_ext_del(skb, id);
4885 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4887 if (skb_ext_exist(skb, id)) {
4888 struct skb_ext *ext = skb->extensions;
4890 return (void *)ext + (ext->offset[id] << 3);
4893 return NULL;
4896 static inline void skb_ext_reset(struct sk_buff *skb)
4898 if (unlikely(skb->active_extensions)) {
4899 __skb_ext_put(skb->extensions);
4900 skb->active_extensions = 0;
4904 static inline bool skb_has_extensions(struct sk_buff *skb)
4906 return unlikely(skb->active_extensions);
4908 #else
4909 static inline void skb_ext_put(struct sk_buff *skb) {}
4910 static inline void skb_ext_reset(struct sk_buff *skb) {}
4911 static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4912 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4913 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4914 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; }
4915 #endif /* CONFIG_SKB_EXTENSIONS */
4917 static inline void nf_reset_ct(struct sk_buff *skb)
4919 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4920 nf_conntrack_put(skb_nfct(skb));
4921 skb->_nfct = 0;
4922 #endif
4925 static inline void nf_reset_trace(struct sk_buff *skb)
4927 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
4928 skb->nf_trace = 0;
4929 #endif
4932 static inline void ipvs_reset(struct sk_buff *skb)
4934 #if IS_ENABLED(CONFIG_IP_VS)
4935 skb->ipvs_property = 0;
4936 #endif
4939 /* Note: This doesn't put any conntrack info in dst. */
4940 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4941 bool copy)
4943 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4944 dst->_nfct = src->_nfct;
4945 nf_conntrack_get(skb_nfct(src));
4946 #endif
4947 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
4948 if (copy)
4949 dst->nf_trace = src->nf_trace;
4950 #endif
4953 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4955 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4956 nf_conntrack_put(skb_nfct(dst));
4957 #endif
4958 dst->slow_gro = src->slow_gro;
4959 __nf_copy(dst, src, true);
4962 #ifdef CONFIG_NETWORK_SECMARK
4963 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4965 to->secmark = from->secmark;
4968 static inline void skb_init_secmark(struct sk_buff *skb)
4970 skb->secmark = 0;
4972 #else
4973 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4976 static inline void skb_init_secmark(struct sk_buff *skb)
4978 #endif
4980 static inline int secpath_exists(const struct sk_buff *skb)
4982 #ifdef CONFIG_XFRM
4983 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4984 #else
4985 return 0;
4986 #endif
4989 static inline bool skb_irq_freeable(const struct sk_buff *skb)
4991 return !skb->destructor &&
4992 !secpath_exists(skb) &&
4993 !skb_nfct(skb) &&
4994 !skb->_skb_refdst &&
4995 !skb_has_frag_list(skb);
4998 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
5000 skb->queue_mapping = queue_mapping;
5003 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
5005 return skb->queue_mapping;
5008 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
5010 to->queue_mapping = from->queue_mapping;
5013 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
5015 skb->queue_mapping = rx_queue + 1;
5018 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
5020 return skb->queue_mapping - 1;
5023 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
5025 return skb->queue_mapping != 0;
5028 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
5030 skb->dst_pending_confirm = val;
5033 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
5035 return skb->dst_pending_confirm != 0;
5038 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
5040 #ifdef CONFIG_XFRM
5041 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
5042 #else
5043 return NULL;
5044 #endif
5047 static inline bool skb_is_gso(const struct sk_buff *skb)
5049 return skb_shinfo(skb)->gso_size;
5052 /* Note: Should be called only if skb_is_gso(skb) is true */
5053 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
5055 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
5058 /* Note: Should be called only if skb_is_gso(skb) is true */
5059 static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
5061 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
5064 /* Note: Should be called only if skb_is_gso(skb) is true */
5065 static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
5067 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
5070 static inline void skb_gso_reset(struct sk_buff *skb)
5072 skb_shinfo(skb)->gso_size = 0;
5073 skb_shinfo(skb)->gso_segs = 0;
5074 skb_shinfo(skb)->gso_type = 0;
5077 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
5078 u16 increment)
5080 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
5081 return;
5082 shinfo->gso_size += increment;
5085 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
5086 u16 decrement)
5088 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
5089 return;
5090 shinfo->gso_size -= decrement;
5093 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
5095 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
5097 /* LRO sets gso_size but not gso_type, whereas if GSO is really
5098 * wanted then gso_type will be set. */
5099 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5101 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
5102 unlikely(shinfo->gso_type == 0)) {
5103 __skb_warn_lro_forwarding(skb);
5104 return true;
5106 return false;
5109 static inline void skb_forward_csum(struct sk_buff *skb)
5111 /* Unfortunately we don't support this one. Any brave souls? */
5112 if (skb->ip_summed == CHECKSUM_COMPLETE)
5113 skb->ip_summed = CHECKSUM_NONE;
5117 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
5118 * @skb: skb to check
5120 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
5121 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
5122 * use this helper, to document places where we make this assertion.
5124 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
5126 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE);
5129 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
5131 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
5132 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5133 unsigned int transport_len,
5134 __sum16(*skb_chkf)(struct sk_buff *skb));
5137 * skb_head_is_locked - Determine if the skb->head is locked down
5138 * @skb: skb to check
5140 * The head on skbs build around a head frag can be removed if they are
5141 * not cloned. This function returns true if the skb head is locked down
5142 * due to either being allocated via kmalloc, or by being a clone with
5143 * multiple references to the head.
5145 static inline bool skb_head_is_locked(const struct sk_buff *skb)
5147 return !skb->head_frag || skb_cloned(skb);
5150 /* Local Checksum Offload.
5151 * Compute outer checksum based on the assumption that the
5152 * inner checksum will be offloaded later.
5153 * See Documentation/networking/checksum-offloads.rst for
5154 * explanation of how this works.
5155 * Fill in outer checksum adjustment (e.g. with sum of outer
5156 * pseudo-header) before calling.
5157 * Also ensure that inner checksum is in linear data area.
5159 static inline __wsum lco_csum(struct sk_buff *skb)
5161 unsigned char *csum_start = skb_checksum_start(skb);
5162 unsigned char *l4_hdr = skb_transport_header(skb);
5163 __wsum partial;
5165 /* Start with complement of inner checksum adjustment */
5166 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
5167 skb->csum_offset));
5169 /* Add in checksum of our headers (incl. outer checksum
5170 * adjustment filled in by caller) and return result.
5172 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
5175 static inline bool skb_is_redirected(const struct sk_buff *skb)
5177 return skb->redirected;
5180 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress)
5182 skb->redirected = 1;
5183 #ifdef CONFIG_NET_REDIRECT
5184 skb->from_ingress = from_ingress;
5185 if (skb->from_ingress)
5186 skb_clear_tstamp(skb);
5187 #endif
5190 static inline void skb_reset_redirect(struct sk_buff *skb)
5192 skb->redirected = 0;
5195 static inline void skb_set_redirected_noclear(struct sk_buff *skb,
5196 bool from_ingress)
5198 skb->redirected = 1;
5199 #ifdef CONFIG_NET_REDIRECT
5200 skb->from_ingress = from_ingress;
5201 #endif
5204 static inline bool skb_csum_is_sctp(struct sk_buff *skb)
5206 #if IS_ENABLED(CONFIG_IP_SCTP)
5207 return skb->csum_not_inet;
5208 #else
5209 return 0;
5210 #endif
5213 static inline void skb_reset_csum_not_inet(struct sk_buff *skb)
5215 skb->ip_summed = CHECKSUM_NONE;
5216 #if IS_ENABLED(CONFIG_IP_SCTP)
5217 skb->csum_not_inet = 0;
5218 #endif
5221 static inline void skb_set_kcov_handle(struct sk_buff *skb,
5222 const u64 kcov_handle)
5224 #ifdef CONFIG_KCOV
5225 skb->kcov_handle = kcov_handle;
5226 #endif
5229 static inline u64 skb_get_kcov_handle(struct sk_buff *skb)
5231 #ifdef CONFIG_KCOV
5232 return skb->kcov_handle;
5233 #else
5234 return 0;
5235 #endif
5238 static inline void skb_mark_for_recycle(struct sk_buff *skb)
5240 #ifdef CONFIG_PAGE_POOL
5241 skb->pp_recycle = 1;
5242 #endif
5245 ssize_t skb_splice_from_iter(struct sk_buff *skb, struct iov_iter *iter,
5246 ssize_t maxsize, gfp_t gfp);
5248 #endif /* __KERNEL__ */
5249 #endif /* _LINUX_SKBUFF_H */