]> git.ipfire.org Git - thirdparty/kernel/stable.git/blob - include/linux/skbuff.h
KVM: x86/mmu: Remove unnecessary ‘NULL’ values from sptep
[thirdparty/kernel/stable.git] / include / linux / skbuff.h
1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3 * Definitions for the 'struct sk_buff' memory handlers.
4 *
5 * Authors:
6 * Alan Cox, <gw4pts@gw4pts.ampr.org>
7 * Florian La Roche, <rzsfl@rz.uni-sb.de>
8 */
9
10 #ifndef _LINUX_SKBUFF_H
11 #define _LINUX_SKBUFF_H
12
13 #include <linux/kernel.h>
14 #include <linux/compiler.h>
15 #include <linux/time.h>
16 #include <linux/bug.h>
17 #include <linux/bvec.h>
18 #include <linux/cache.h>
19 #include <linux/rbtree.h>
20 #include <linux/socket.h>
21 #include <linux/refcount.h>
22
23 #include <linux/atomic.h>
24 #include <asm/types.h>
25 #include <linux/spinlock.h>
26 #include <net/checksum.h>
27 #include <linux/rcupdate.h>
28 #include <linux/dma-mapping.h>
29 #include <linux/netdev_features.h>
30 #include <net/flow_dissector.h>
31 #include <linux/in6.h>
32 #include <linux/if_packet.h>
33 #include <linux/llist.h>
34 #include <net/flow.h>
35 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
36 #include <linux/netfilter/nf_conntrack_common.h>
37 #endif
38 #include <net/net_debug.h>
39 #include <net/dropreason-core.h>
40
41 /**
42 * DOC: skb checksums
43 *
44 * The interface for checksum offload between the stack and networking drivers
45 * is as follows...
46 *
47 * IP checksum related features
48 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
49 *
50 * Drivers advertise checksum offload capabilities in the features of a device.
51 * From the stack's point of view these are capabilities offered by the driver.
52 * A driver typically only advertises features that it is capable of offloading
53 * to its device.
54 *
55 * .. flat-table:: Checksum related device features
56 * :widths: 1 10
57 *
58 * * - %NETIF_F_HW_CSUM
59 * - The driver (or its device) is able to compute one
60 * IP (one's complement) checksum for any combination
61 * of protocols or protocol layering. The checksum is
62 * computed and set in a packet per the CHECKSUM_PARTIAL
63 * interface (see below).
64 *
65 * * - %NETIF_F_IP_CSUM
66 * - Driver (device) is only able to checksum plain
67 * TCP or UDP packets over IPv4. These are specifically
68 * unencapsulated packets of the form IPv4|TCP or
69 * IPv4|UDP where the Protocol field in the IPv4 header
70 * is TCP or UDP. The IPv4 header may contain IP options.
71 * This feature cannot be set in features for a device
72 * with NETIF_F_HW_CSUM also set. This feature is being
73 * DEPRECATED (see below).
74 *
75 * * - %NETIF_F_IPV6_CSUM
76 * - Driver (device) is only able to checksum plain
77 * TCP or UDP packets over IPv6. These are specifically
78 * unencapsulated packets of the form IPv6|TCP or
79 * IPv6|UDP where the Next Header field in the IPv6
80 * header is either TCP or UDP. IPv6 extension headers
81 * are not supported with this feature. This feature
82 * cannot be set in features for a device with
83 * NETIF_F_HW_CSUM also set. This feature is being
84 * DEPRECATED (see below).
85 *
86 * * - %NETIF_F_RXCSUM
87 * - Driver (device) performs receive checksum offload.
88 * This flag is only used to disable the RX checksum
89 * feature for a device. The stack will accept receive
90 * checksum indication in packets received on a device
91 * regardless of whether NETIF_F_RXCSUM is set.
92 *
93 * Checksumming of received packets by device
94 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
95 *
96 * Indication of checksum verification is set in &sk_buff.ip_summed.
97 * Possible values are:
98 *
99 * - %CHECKSUM_NONE
100 *
101 * Device did not checksum this packet e.g. due to lack of capabilities.
102 * The packet contains full (though not verified) checksum in packet but
103 * not in skb->csum. Thus, skb->csum is undefined in this case.
104 *
105 * - %CHECKSUM_UNNECESSARY
106 *
107 * The hardware you're dealing with doesn't calculate the full checksum
108 * (as in %CHECKSUM_COMPLETE), but it does parse headers and verify checksums
109 * for specific protocols. For such packets it will set %CHECKSUM_UNNECESSARY
110 * if their checksums are okay. &sk_buff.csum is still undefined in this case
111 * though. A driver or device must never modify the checksum field in the
112 * packet even if checksum is verified.
113 *
114 * %CHECKSUM_UNNECESSARY is applicable to following protocols:
115 *
116 * - TCP: IPv6 and IPv4.
117 * - UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
118 * zero UDP checksum for either IPv4 or IPv6, the networking stack
119 * may perform further validation in this case.
120 * - GRE: only if the checksum is present in the header.
121 * - SCTP: indicates the CRC in SCTP header has been validated.
122 * - FCOE: indicates the CRC in FC frame has been validated.
123 *
124 * &sk_buff.csum_level indicates the number of consecutive checksums found in
125 * the packet minus one that have been verified as %CHECKSUM_UNNECESSARY.
126 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
127 * and a device is able to verify the checksums for UDP (possibly zero),
128 * GRE (checksum flag is set) and TCP, &sk_buff.csum_level would be set to
129 * two. If the device were only able to verify the UDP checksum and not
130 * GRE, either because it doesn't support GRE checksum or because GRE
131 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
132 * not considered in this case).
133 *
134 * - %CHECKSUM_COMPLETE
135 *
136 * This is the most generic way. The device supplied checksum of the _whole_
137 * packet as seen by netif_rx() and fills in &sk_buff.csum. This means the
138 * hardware doesn't need to parse L3/L4 headers to implement this.
139 *
140 * Notes:
141 *
142 * - Even if device supports only some protocols, but is able to produce
143 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
144 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
145 *
146 * - %CHECKSUM_PARTIAL
147 *
148 * A checksum is set up to be offloaded to a device as described in the
149 * output description for CHECKSUM_PARTIAL. This may occur on a packet
150 * received directly from another Linux OS, e.g., a virtualized Linux kernel
151 * on the same host, or it may be set in the input path in GRO or remote
152 * checksum offload. For the purposes of checksum verification, the checksum
153 * referred to by skb->csum_start + skb->csum_offset and any preceding
154 * checksums in the packet are considered verified. Any checksums in the
155 * packet that are after the checksum being offloaded are not considered to
156 * be verified.
157 *
158 * Checksumming on transmit for non-GSO
159 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
160 *
161 * The stack requests checksum offload in the &sk_buff.ip_summed for a packet.
162 * Values are:
163 *
164 * - %CHECKSUM_PARTIAL
165 *
166 * The driver is required to checksum the packet as seen by hard_start_xmit()
167 * from &sk_buff.csum_start up to the end, and to record/write the checksum at
168 * offset &sk_buff.csum_start + &sk_buff.csum_offset.
169 * A driver may verify that the
170 * csum_start and csum_offset values are valid values given the length and
171 * offset of the packet, but it should not attempt to validate that the
172 * checksum refers to a legitimate transport layer checksum -- it is the
173 * purview of the stack to validate that csum_start and csum_offset are set
174 * correctly.
175 *
176 * When the stack requests checksum offload for a packet, the driver MUST
177 * ensure that the checksum is set correctly. A driver can either offload the
178 * checksum calculation to the device, or call skb_checksum_help (in the case
179 * that the device does not support offload for a particular checksum).
180 *
181 * %NETIF_F_IP_CSUM and %NETIF_F_IPV6_CSUM are being deprecated in favor of
182 * %NETIF_F_HW_CSUM. New devices should use %NETIF_F_HW_CSUM to indicate
183 * checksum offload capability.
184 * skb_csum_hwoffload_help() can be called to resolve %CHECKSUM_PARTIAL based
185 * on network device checksumming capabilities: if a packet does not match
186 * them, skb_checksum_help() or skb_crc32c_help() (depending on the value of
187 * &sk_buff.csum_not_inet, see :ref:`crc`)
188 * is called to resolve the checksum.
189 *
190 * - %CHECKSUM_NONE
191 *
192 * The skb was already checksummed by the protocol, or a checksum is not
193 * required.
194 *
195 * - %CHECKSUM_UNNECESSARY
196 *
197 * This has the same meaning as CHECKSUM_NONE for checksum offload on
198 * output.
199 *
200 * - %CHECKSUM_COMPLETE
201 *
202 * Not used in checksum output. If a driver observes a packet with this value
203 * set in skbuff, it should treat the packet as if %CHECKSUM_NONE were set.
204 *
205 * .. _crc:
206 *
207 * Non-IP checksum (CRC) offloads
208 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
209 *
210 * .. flat-table::
211 * :widths: 1 10
212 *
213 * * - %NETIF_F_SCTP_CRC
214 * - This feature indicates that a device is capable of
215 * offloading the SCTP CRC in a packet. To perform this offload the stack
216 * will set csum_start and csum_offset accordingly, set ip_summed to
217 * %CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication
218 * in the skbuff that the %CHECKSUM_PARTIAL refers to CRC32c.
219 * A driver that supports both IP checksum offload and SCTP CRC32c offload
220 * must verify which offload is configured for a packet by testing the
221 * value of &sk_buff.csum_not_inet; skb_crc32c_csum_help() is provided to
222 * resolve %CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
223 *
224 * * - %NETIF_F_FCOE_CRC
225 * - This feature indicates that a device is capable of offloading the FCOE
226 * CRC in a packet. To perform this offload the stack will set ip_summed
227 * to %CHECKSUM_PARTIAL and set csum_start and csum_offset
228 * accordingly. Note that there is no indication in the skbuff that the
229 * %CHECKSUM_PARTIAL refers to an FCOE checksum, so a driver that supports
230 * both IP checksum offload and FCOE CRC offload must verify which offload
231 * is configured for a packet, presumably by inspecting packet headers.
232 *
233 * Checksumming on output with GSO
234 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
235 *
236 * In the case of a GSO packet (skb_is_gso() is true), checksum offload
237 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
238 * gso_type is %SKB_GSO_TCPV4 or %SKB_GSO_TCPV6, TCP checksum offload as
239 * part of the GSO operation is implied. If a checksum is being offloaded
240 * with GSO then ip_summed is %CHECKSUM_PARTIAL, and both csum_start and
241 * csum_offset are set to refer to the outermost checksum being offloaded
242 * (two offloaded checksums are possible with UDP encapsulation).
243 */
244
245 /* Don't change this without changing skb_csum_unnecessary! */
246 #define CHECKSUM_NONE 0
247 #define CHECKSUM_UNNECESSARY 1
248 #define CHECKSUM_COMPLETE 2
249 #define CHECKSUM_PARTIAL 3
250
251 /* Maximum value in skb->csum_level */
252 #define SKB_MAX_CSUM_LEVEL 3
253
254 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
255 #define SKB_WITH_OVERHEAD(X) \
256 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
257
258 /* For X bytes available in skb->head, what is the minimal
259 * allocation needed, knowing struct skb_shared_info needs
260 * to be aligned.
261 */
262 #define SKB_HEAD_ALIGN(X) (SKB_DATA_ALIGN(X) + \
263 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
264
265 #define SKB_MAX_ORDER(X, ORDER) \
266 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
267 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
268 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
269
270 /* return minimum truesize of one skb containing X bytes of data */
271 #define SKB_TRUESIZE(X) ((X) + \
272 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
273 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
274
275 struct ahash_request;
276 struct net_device;
277 struct scatterlist;
278 struct pipe_inode_info;
279 struct iov_iter;
280 struct napi_struct;
281 struct bpf_prog;
282 union bpf_attr;
283 struct skb_ext;
284 struct ts_config;
285
286 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
287 struct nf_bridge_info {
288 enum {
289 BRNF_PROTO_UNCHANGED,
290 BRNF_PROTO_8021Q,
291 BRNF_PROTO_PPPOE
292 } orig_proto:8;
293 u8 pkt_otherhost:1;
294 u8 in_prerouting:1;
295 u8 bridged_dnat:1;
296 u8 sabotage_in_done:1;
297 __u16 frag_max_size;
298 struct net_device *physindev;
299
300 /* always valid & non-NULL from FORWARD on, for physdev match */
301 struct net_device *physoutdev;
302 union {
303 /* prerouting: detect dnat in orig/reply direction */
304 __be32 ipv4_daddr;
305 struct in6_addr ipv6_daddr;
306
307 /* after prerouting + nat detected: store original source
308 * mac since neigh resolution overwrites it, only used while
309 * skb is out in neigh layer.
310 */
311 char neigh_header[8];
312 };
313 };
314 #endif
315
316 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
317 /* Chain in tc_skb_ext will be used to share the tc chain with
318 * ovs recirc_id. It will be set to the current chain by tc
319 * and read by ovs to recirc_id.
320 */
321 struct tc_skb_ext {
322 union {
323 u64 act_miss_cookie;
324 __u32 chain;
325 };
326 __u16 mru;
327 __u16 zone;
328 u8 post_ct:1;
329 u8 post_ct_snat:1;
330 u8 post_ct_dnat:1;
331 u8 act_miss:1; /* Set if act_miss_cookie is used */
332 u8 l2_miss:1; /* Set by bridge upon FDB or MDB miss */
333 };
334 #endif
335
336 struct sk_buff_head {
337 /* These two members must be first to match sk_buff. */
338 struct_group_tagged(sk_buff_list, list,
339 struct sk_buff *next;
340 struct sk_buff *prev;
341 );
342
343 __u32 qlen;
344 spinlock_t lock;
345 };
346
347 struct sk_buff;
348
349 #ifndef CONFIG_MAX_SKB_FRAGS
350 # define CONFIG_MAX_SKB_FRAGS 17
351 #endif
352
353 #define MAX_SKB_FRAGS CONFIG_MAX_SKB_FRAGS
354
355 extern int sysctl_max_skb_frags;
356
357 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
358 * segment using its current segmentation instead.
359 */
360 #define GSO_BY_FRAGS 0xFFFF
361
362 typedef struct bio_vec skb_frag_t;
363
364 /**
365 * skb_frag_size() - Returns the size of a skb fragment
366 * @frag: skb fragment
367 */
368 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
369 {
370 return frag->bv_len;
371 }
372
373 /**
374 * skb_frag_size_set() - Sets the size of a skb fragment
375 * @frag: skb fragment
376 * @size: size of fragment
377 */
378 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
379 {
380 frag->bv_len = size;
381 }
382
383 /**
384 * skb_frag_size_add() - Increments the size of a skb fragment by @delta
385 * @frag: skb fragment
386 * @delta: value to add
387 */
388 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
389 {
390 frag->bv_len += delta;
391 }
392
393 /**
394 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta
395 * @frag: skb fragment
396 * @delta: value to subtract
397 */
398 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
399 {
400 frag->bv_len -= delta;
401 }
402
403 /**
404 * skb_frag_must_loop - Test if %p is a high memory page
405 * @p: fragment's page
406 */
407 static inline bool skb_frag_must_loop(struct page *p)
408 {
409 #if defined(CONFIG_HIGHMEM)
410 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p))
411 return true;
412 #endif
413 return false;
414 }
415
416 /**
417 * skb_frag_foreach_page - loop over pages in a fragment
418 *
419 * @f: skb frag to operate on
420 * @f_off: offset from start of f->bv_page
421 * @f_len: length from f_off to loop over
422 * @p: (temp var) current page
423 * @p_off: (temp var) offset from start of current page,
424 * non-zero only on first page.
425 * @p_len: (temp var) length in current page,
426 * < PAGE_SIZE only on first and last page.
427 * @copied: (temp var) length so far, excluding current p_len.
428 *
429 * A fragment can hold a compound page, in which case per-page
430 * operations, notably kmap_atomic, must be called for each
431 * regular page.
432 */
433 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
434 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
435 p_off = (f_off) & (PAGE_SIZE - 1), \
436 p_len = skb_frag_must_loop(p) ? \
437 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
438 copied = 0; \
439 copied < f_len; \
440 copied += p_len, p++, p_off = 0, \
441 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
442
443 /**
444 * struct skb_shared_hwtstamps - hardware time stamps
445 * @hwtstamp: hardware time stamp transformed into duration
446 * since arbitrary point in time
447 * @netdev_data: address/cookie of network device driver used as
448 * reference to actual hardware time stamp
449 *
450 * Software time stamps generated by ktime_get_real() are stored in
451 * skb->tstamp.
452 *
453 * hwtstamps can only be compared against other hwtstamps from
454 * the same device.
455 *
456 * This structure is attached to packets as part of the
457 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
458 */
459 struct skb_shared_hwtstamps {
460 union {
461 ktime_t hwtstamp;
462 void *netdev_data;
463 };
464 };
465
466 /* Definitions for tx_flags in struct skb_shared_info */
467 enum {
468 /* generate hardware time stamp */
469 SKBTX_HW_TSTAMP = 1 << 0,
470
471 /* generate software time stamp when queueing packet to NIC */
472 SKBTX_SW_TSTAMP = 1 << 1,
473
474 /* device driver is going to provide hardware time stamp */
475 SKBTX_IN_PROGRESS = 1 << 2,
476
477 /* generate hardware time stamp based on cycles if supported */
478 SKBTX_HW_TSTAMP_USE_CYCLES = 1 << 3,
479
480 /* generate wifi status information (where possible) */
481 SKBTX_WIFI_STATUS = 1 << 4,
482
483 /* determine hardware time stamp based on time or cycles */
484 SKBTX_HW_TSTAMP_NETDEV = 1 << 5,
485
486 /* generate software time stamp when entering packet scheduling */
487 SKBTX_SCHED_TSTAMP = 1 << 6,
488 };
489
490 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
491 SKBTX_SCHED_TSTAMP)
492 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \
493 SKBTX_HW_TSTAMP_USE_CYCLES | \
494 SKBTX_ANY_SW_TSTAMP)
495
496 /* Definitions for flags in struct skb_shared_info */
497 enum {
498 /* use zcopy routines */
499 SKBFL_ZEROCOPY_ENABLE = BIT(0),
500
501 /* This indicates at least one fragment might be overwritten
502 * (as in vmsplice(), sendfile() ...)
503 * If we need to compute a TX checksum, we'll need to copy
504 * all frags to avoid possible bad checksum
505 */
506 SKBFL_SHARED_FRAG = BIT(1),
507
508 /* segment contains only zerocopy data and should not be
509 * charged to the kernel memory.
510 */
511 SKBFL_PURE_ZEROCOPY = BIT(2),
512
513 SKBFL_DONT_ORPHAN = BIT(3),
514
515 /* page references are managed by the ubuf_info, so it's safe to
516 * use frags only up until ubuf_info is released
517 */
518 SKBFL_MANAGED_FRAG_REFS = BIT(4),
519 };
520
521 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG)
522 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \
523 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS)
524
525 /*
526 * The callback notifies userspace to release buffers when skb DMA is done in
527 * lower device, the skb last reference should be 0 when calling this.
528 * The zerocopy_success argument is true if zero copy transmit occurred,
529 * false on data copy or out of memory error caused by data copy attempt.
530 * The ctx field is used to track device context.
531 * The desc field is used to track userspace buffer index.
532 */
533 struct ubuf_info {
534 void (*callback)(struct sk_buff *, struct ubuf_info *,
535 bool zerocopy_success);
536 refcount_t refcnt;
537 u8 flags;
538 };
539
540 struct ubuf_info_msgzc {
541 struct ubuf_info ubuf;
542
543 union {
544 struct {
545 unsigned long desc;
546 void *ctx;
547 };
548 struct {
549 u32 id;
550 u16 len;
551 u16 zerocopy:1;
552 u32 bytelen;
553 };
554 };
555
556 struct mmpin {
557 struct user_struct *user;
558 unsigned int num_pg;
559 } mmp;
560 };
561
562 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
563 #define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \
564 ubuf)
565
566 int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
567 void mm_unaccount_pinned_pages(struct mmpin *mmp);
568
569 /* This data is invariant across clones and lives at
570 * the end of the header data, ie. at skb->end.
571 */
572 struct skb_shared_info {
573 __u8 flags;
574 __u8 meta_len;
575 __u8 nr_frags;
576 __u8 tx_flags;
577 unsigned short gso_size;
578 /* Warning: this field is not always filled in (UFO)! */
579 unsigned short gso_segs;
580 struct sk_buff *frag_list;
581 struct skb_shared_hwtstamps hwtstamps;
582 unsigned int gso_type;
583 u32 tskey;
584
585 /*
586 * Warning : all fields before dataref are cleared in __alloc_skb()
587 */
588 atomic_t dataref;
589 unsigned int xdp_frags_size;
590
591 /* Intermediate layers must ensure that destructor_arg
592 * remains valid until skb destructor */
593 void * destructor_arg;
594
595 /* must be last field, see pskb_expand_head() */
596 skb_frag_t frags[MAX_SKB_FRAGS];
597 };
598
599 /**
600 * DOC: dataref and headerless skbs
601 *
602 * Transport layers send out clones of payload skbs they hold for
603 * retransmissions. To allow lower layers of the stack to prepend their headers
604 * we split &skb_shared_info.dataref into two halves.
605 * The lower 16 bits count the overall number of references.
606 * The higher 16 bits indicate how many of the references are payload-only.
607 * skb_header_cloned() checks if skb is allowed to add / write the headers.
608 *
609 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr
610 * (via __skb_header_release()). Any clone created from marked skb will get
611 * &sk_buff.hdr_len populated with the available headroom.
612 * If there's the only clone in existence it's able to modify the headroom
613 * at will. The sequence of calls inside the transport layer is::
614 *
615 * <alloc skb>
616 * skb_reserve()
617 * __skb_header_release()
618 * skb_clone()
619 * // send the clone down the stack
620 *
621 * This is not a very generic construct and it depends on the transport layers
622 * doing the right thing. In practice there's usually only one payload-only skb.
623 * Having multiple payload-only skbs with different lengths of hdr_len is not
624 * possible. The payload-only skbs should never leave their owner.
625 */
626 #define SKB_DATAREF_SHIFT 16
627 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
628
629
630 enum {
631 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
632 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
633 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
634 };
635
636 enum {
637 SKB_GSO_TCPV4 = 1 << 0,
638
639 /* This indicates the skb is from an untrusted source. */
640 SKB_GSO_DODGY = 1 << 1,
641
642 /* This indicates the tcp segment has CWR set. */
643 SKB_GSO_TCP_ECN = 1 << 2,
644
645 SKB_GSO_TCP_FIXEDID = 1 << 3,
646
647 SKB_GSO_TCPV6 = 1 << 4,
648
649 SKB_GSO_FCOE = 1 << 5,
650
651 SKB_GSO_GRE = 1 << 6,
652
653 SKB_GSO_GRE_CSUM = 1 << 7,
654
655 SKB_GSO_IPXIP4 = 1 << 8,
656
657 SKB_GSO_IPXIP6 = 1 << 9,
658
659 SKB_GSO_UDP_TUNNEL = 1 << 10,
660
661 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
662
663 SKB_GSO_PARTIAL = 1 << 12,
664
665 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
666
667 SKB_GSO_SCTP = 1 << 14,
668
669 SKB_GSO_ESP = 1 << 15,
670
671 SKB_GSO_UDP = 1 << 16,
672
673 SKB_GSO_UDP_L4 = 1 << 17,
674
675 SKB_GSO_FRAGLIST = 1 << 18,
676 };
677
678 #if BITS_PER_LONG > 32
679 #define NET_SKBUFF_DATA_USES_OFFSET 1
680 #endif
681
682 #ifdef NET_SKBUFF_DATA_USES_OFFSET
683 typedef unsigned int sk_buff_data_t;
684 #else
685 typedef unsigned char *sk_buff_data_t;
686 #endif
687
688 /**
689 * DOC: Basic sk_buff geometry
690 *
691 * struct sk_buff itself is a metadata structure and does not hold any packet
692 * data. All the data is held in associated buffers.
693 *
694 * &sk_buff.head points to the main "head" buffer. The head buffer is divided
695 * into two parts:
696 *
697 * - data buffer, containing headers and sometimes payload;
698 * this is the part of the skb operated on by the common helpers
699 * such as skb_put() or skb_pull();
700 * - shared info (struct skb_shared_info) which holds an array of pointers
701 * to read-only data in the (page, offset, length) format.
702 *
703 * Optionally &skb_shared_info.frag_list may point to another skb.
704 *
705 * Basic diagram may look like this::
706 *
707 * ---------------
708 * | sk_buff |
709 * ---------------
710 * ,--------------------------- + head
711 * / ,----------------- + data
712 * / / ,----------- + tail
713 * | | | , + end
714 * | | | |
715 * v v v v
716 * -----------------------------------------------
717 * | headroom | data | tailroom | skb_shared_info |
718 * -----------------------------------------------
719 * + [page frag]
720 * + [page frag]
721 * + [page frag]
722 * + [page frag] ---------
723 * + frag_list --> | sk_buff |
724 * ---------
725 *
726 */
727
728 /**
729 * struct sk_buff - socket buffer
730 * @next: Next buffer in list
731 * @prev: Previous buffer in list
732 * @tstamp: Time we arrived/left
733 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point
734 * for retransmit timer
735 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
736 * @list: queue head
737 * @ll_node: anchor in an llist (eg socket defer_list)
738 * @sk: Socket we are owned by
739 * @ip_defrag_offset: (aka @sk) alternate use of @sk, used in
740 * fragmentation management
741 * @dev: Device we arrived on/are leaving by
742 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL
743 * @cb: Control buffer. Free for use by every layer. Put private vars here
744 * @_skb_refdst: destination entry (with norefcount bit)
745 * @sp: the security path, used for xfrm
746 * @len: Length of actual data
747 * @data_len: Data length
748 * @mac_len: Length of link layer header
749 * @hdr_len: writable header length of cloned skb
750 * @csum: Checksum (must include start/offset pair)
751 * @csum_start: Offset from skb->head where checksumming should start
752 * @csum_offset: Offset from csum_start where checksum should be stored
753 * @priority: Packet queueing priority
754 * @ignore_df: allow local fragmentation
755 * @cloned: Head may be cloned (check refcnt to be sure)
756 * @ip_summed: Driver fed us an IP checksum
757 * @nohdr: Payload reference only, must not modify header
758 * @pkt_type: Packet class
759 * @fclone: skbuff clone status
760 * @ipvs_property: skbuff is owned by ipvs
761 * @inner_protocol_type: whether the inner protocol is
762 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO
763 * @remcsum_offload: remote checksum offload is enabled
764 * @offload_fwd_mark: Packet was L2-forwarded in hardware
765 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
766 * @tc_skip_classify: do not classify packet. set by IFB device
767 * @tc_at_ingress: used within tc_classify to distinguish in/egress
768 * @redirected: packet was redirected by packet classifier
769 * @from_ingress: packet was redirected from the ingress path
770 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h
771 * @peeked: this packet has been seen already, so stats have been
772 * done for it, don't do them again
773 * @nf_trace: netfilter packet trace flag
774 * @protocol: Packet protocol from driver
775 * @destructor: Destruct function
776 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
777 * @_sk_redir: socket redirection information for skmsg
778 * @_nfct: Associated connection, if any (with nfctinfo bits)
779 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
780 * @skb_iif: ifindex of device we arrived on
781 * @tc_index: Traffic control index
782 * @hash: the packet hash
783 * @queue_mapping: Queue mapping for multiqueue devices
784 * @head_frag: skb was allocated from page fragments,
785 * not allocated by kmalloc() or vmalloc().
786 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
787 * @pp_recycle: mark the packet for recycling instead of freeing (implies
788 * page_pool support on driver)
789 * @active_extensions: active extensions (skb_ext_id types)
790 * @ndisc_nodetype: router type (from link layer)
791 * @ooo_okay: allow the mapping of a socket to a queue to be changed
792 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
793 * ports.
794 * @sw_hash: indicates hash was computed in software stack
795 * @wifi_acked_valid: wifi_acked was set
796 * @wifi_acked: whether frame was acked on wifi or not
797 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
798 * @encapsulation: indicates the inner headers in the skbuff are valid
799 * @encap_hdr_csum: software checksum is needed
800 * @csum_valid: checksum is already valid
801 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
802 * @csum_complete_sw: checksum was completed by software
803 * @csum_level: indicates the number of consecutive checksums found in
804 * the packet minus one that have been verified as
805 * CHECKSUM_UNNECESSARY (max 3)
806 * @dst_pending_confirm: need to confirm neighbour
807 * @decrypted: Decrypted SKB
808 * @slow_gro: state present at GRO time, slower prepare step required
809 * @mono_delivery_time: When set, skb->tstamp has the
810 * delivery_time in mono clock base (i.e. EDT). Otherwise, the
811 * skb->tstamp has the (rcv) timestamp at ingress and
812 * delivery_time at egress.
813 * @napi_id: id of the NAPI struct this skb came from
814 * @sender_cpu: (aka @napi_id) source CPU in XPS
815 * @alloc_cpu: CPU which did the skb allocation.
816 * @secmark: security marking
817 * @mark: Generic packet mark
818 * @reserved_tailroom: (aka @mark) number of bytes of free space available
819 * at the tail of an sk_buff
820 * @vlan_all: vlan fields (proto & tci)
821 * @vlan_proto: vlan encapsulation protocol
822 * @vlan_tci: vlan tag control information
823 * @inner_protocol: Protocol (encapsulation)
824 * @inner_ipproto: (aka @inner_protocol) stores ipproto when
825 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO;
826 * @inner_transport_header: Inner transport layer header (encapsulation)
827 * @inner_network_header: Network layer header (encapsulation)
828 * @inner_mac_header: Link layer header (encapsulation)
829 * @transport_header: Transport layer header
830 * @network_header: Network layer header
831 * @mac_header: Link layer header
832 * @kcov_handle: KCOV remote handle for remote coverage collection
833 * @tail: Tail pointer
834 * @end: End pointer
835 * @head: Head of buffer
836 * @data: Data head pointer
837 * @truesize: Buffer size
838 * @users: User count - see {datagram,tcp}.c
839 * @extensions: allocated extensions, valid if active_extensions is nonzero
840 */
841
842 struct sk_buff {
843 union {
844 struct {
845 /* These two members must be first to match sk_buff_head. */
846 struct sk_buff *next;
847 struct sk_buff *prev;
848
849 union {
850 struct net_device *dev;
851 /* Some protocols might use this space to store information,
852 * while device pointer would be NULL.
853 * UDP receive path is one user.
854 */
855 unsigned long dev_scratch;
856 };
857 };
858 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */
859 struct list_head list;
860 struct llist_node ll_node;
861 };
862
863 union {
864 struct sock *sk;
865 int ip_defrag_offset;
866 };
867
868 union {
869 ktime_t tstamp;
870 u64 skb_mstamp_ns; /* earliest departure time */
871 };
872 /*
873 * This is the control buffer. It is free to use for every
874 * layer. Please put your private variables there. If you
875 * want to keep them across layers you have to do a skb_clone()
876 * first. This is owned by whoever has the skb queued ATM.
877 */
878 char cb[48] __aligned(8);
879
880 union {
881 struct {
882 unsigned long _skb_refdst;
883 void (*destructor)(struct sk_buff *skb);
884 };
885 struct list_head tcp_tsorted_anchor;
886 #ifdef CONFIG_NET_SOCK_MSG
887 unsigned long _sk_redir;
888 #endif
889 };
890
891 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
892 unsigned long _nfct;
893 #endif
894 unsigned int len,
895 data_len;
896 __u16 mac_len,
897 hdr_len;
898
899 /* Following fields are _not_ copied in __copy_skb_header()
900 * Note that queue_mapping is here mostly to fill a hole.
901 */
902 __u16 queue_mapping;
903
904 /* if you move cloned around you also must adapt those constants */
905 #ifdef __BIG_ENDIAN_BITFIELD
906 #define CLONED_MASK (1 << 7)
907 #else
908 #define CLONED_MASK 1
909 #endif
910 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset)
911
912 /* private: */
913 __u8 __cloned_offset[0];
914 /* public: */
915 __u8 cloned:1,
916 nohdr:1,
917 fclone:2,
918 peeked:1,
919 head_frag:1,
920 pfmemalloc:1,
921 pp_recycle:1; /* page_pool recycle indicator */
922 #ifdef CONFIG_SKB_EXTENSIONS
923 __u8 active_extensions;
924 #endif
925
926 /* Fields enclosed in headers group are copied
927 * using a single memcpy() in __copy_skb_header()
928 */
929 struct_group(headers,
930
931 /* private: */
932 __u8 __pkt_type_offset[0];
933 /* public: */
934 __u8 pkt_type:3; /* see PKT_TYPE_MAX */
935 __u8 ignore_df:1;
936 __u8 dst_pending_confirm:1;
937 __u8 ip_summed:2;
938 __u8 ooo_okay:1;
939
940 /* private: */
941 __u8 __mono_tc_offset[0];
942 /* public: */
943 __u8 mono_delivery_time:1; /* See SKB_MONO_DELIVERY_TIME_MASK */
944 #ifdef CONFIG_NET_XGRESS
945 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */
946 __u8 tc_skip_classify:1;
947 #endif
948 __u8 remcsum_offload:1;
949 __u8 csum_complete_sw:1;
950 __u8 csum_level:2;
951 __u8 inner_protocol_type:1;
952
953 __u8 l4_hash:1;
954 __u8 sw_hash:1;
955 #ifdef CONFIG_WIRELESS
956 __u8 wifi_acked_valid:1;
957 __u8 wifi_acked:1;
958 #endif
959 __u8 no_fcs:1;
960 /* Indicates the inner headers are valid in the skbuff. */
961 __u8 encapsulation:1;
962 __u8 encap_hdr_csum:1;
963 __u8 csum_valid:1;
964 #ifdef CONFIG_IPV6_NDISC_NODETYPE
965 __u8 ndisc_nodetype:2;
966 #endif
967
968 #if IS_ENABLED(CONFIG_IP_VS)
969 __u8 ipvs_property:1;
970 #endif
971 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
972 __u8 nf_trace:1;
973 #endif
974 #ifdef CONFIG_NET_SWITCHDEV
975 __u8 offload_fwd_mark:1;
976 __u8 offload_l3_fwd_mark:1;
977 #endif
978 __u8 redirected:1;
979 #ifdef CONFIG_NET_REDIRECT
980 __u8 from_ingress:1;
981 #endif
982 #ifdef CONFIG_NETFILTER_SKIP_EGRESS
983 __u8 nf_skip_egress:1;
984 #endif
985 #ifdef CONFIG_TLS_DEVICE
986 __u8 decrypted:1;
987 #endif
988 __u8 slow_gro:1;
989 #if IS_ENABLED(CONFIG_IP_SCTP)
990 __u8 csum_not_inet:1;
991 #endif
992
993 #if defined(CONFIG_NET_SCHED) || defined(CONFIG_NET_XGRESS)
994 __u16 tc_index; /* traffic control index */
995 #endif
996
997 u16 alloc_cpu;
998
999 union {
1000 __wsum csum;
1001 struct {
1002 __u16 csum_start;
1003 __u16 csum_offset;
1004 };
1005 };
1006 __u32 priority;
1007 int skb_iif;
1008 __u32 hash;
1009 union {
1010 u32 vlan_all;
1011 struct {
1012 __be16 vlan_proto;
1013 __u16 vlan_tci;
1014 };
1015 };
1016 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
1017 union {
1018 unsigned int napi_id;
1019 unsigned int sender_cpu;
1020 };
1021 #endif
1022 #ifdef CONFIG_NETWORK_SECMARK
1023 __u32 secmark;
1024 #endif
1025
1026 union {
1027 __u32 mark;
1028 __u32 reserved_tailroom;
1029 };
1030
1031 union {
1032 __be16 inner_protocol;
1033 __u8 inner_ipproto;
1034 };
1035
1036 __u16 inner_transport_header;
1037 __u16 inner_network_header;
1038 __u16 inner_mac_header;
1039
1040 __be16 protocol;
1041 __u16 transport_header;
1042 __u16 network_header;
1043 __u16 mac_header;
1044
1045 #ifdef CONFIG_KCOV
1046 u64 kcov_handle;
1047 #endif
1048
1049 ); /* end headers group */
1050
1051 /* These elements must be at the end, see alloc_skb() for details. */
1052 sk_buff_data_t tail;
1053 sk_buff_data_t end;
1054 unsigned char *head,
1055 *data;
1056 unsigned int truesize;
1057 refcount_t users;
1058
1059 #ifdef CONFIG_SKB_EXTENSIONS
1060 /* only useable after checking ->active_extensions != 0 */
1061 struct skb_ext *extensions;
1062 #endif
1063 };
1064
1065 /* if you move pkt_type around you also must adapt those constants */
1066 #ifdef __BIG_ENDIAN_BITFIELD
1067 #define PKT_TYPE_MAX (7 << 5)
1068 #else
1069 #define PKT_TYPE_MAX 7
1070 #endif
1071 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset)
1072
1073 /* if you move tc_at_ingress or mono_delivery_time
1074 * around, you also must adapt these constants.
1075 */
1076 #ifdef __BIG_ENDIAN_BITFIELD
1077 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 7)
1078 #define TC_AT_INGRESS_MASK (1 << 6)
1079 #else
1080 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 0)
1081 #define TC_AT_INGRESS_MASK (1 << 1)
1082 #endif
1083 #define SKB_BF_MONO_TC_OFFSET offsetof(struct sk_buff, __mono_tc_offset)
1084
1085 #ifdef __KERNEL__
1086 /*
1087 * Handling routines are only of interest to the kernel
1088 */
1089
1090 #define SKB_ALLOC_FCLONE 0x01
1091 #define SKB_ALLOC_RX 0x02
1092 #define SKB_ALLOC_NAPI 0x04
1093
1094 /**
1095 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
1096 * @skb: buffer
1097 */
1098 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
1099 {
1100 return unlikely(skb->pfmemalloc);
1101 }
1102
1103 /*
1104 * skb might have a dst pointer attached, refcounted or not.
1105 * _skb_refdst low order bit is set if refcount was _not_ taken
1106 */
1107 #define SKB_DST_NOREF 1UL
1108 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
1109
1110 /**
1111 * skb_dst - returns skb dst_entry
1112 * @skb: buffer
1113 *
1114 * Returns skb dst_entry, regardless of reference taken or not.
1115 */
1116 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
1117 {
1118 /* If refdst was not refcounted, check we still are in a
1119 * rcu_read_lock section
1120 */
1121 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
1122 !rcu_read_lock_held() &&
1123 !rcu_read_lock_bh_held());
1124 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
1125 }
1126
1127 /**
1128 * skb_dst_set - sets skb dst
1129 * @skb: buffer
1130 * @dst: dst entry
1131 *
1132 * Sets skb dst, assuming a reference was taken on dst and should
1133 * be released by skb_dst_drop()
1134 */
1135 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
1136 {
1137 skb->slow_gro |= !!dst;
1138 skb->_skb_refdst = (unsigned long)dst;
1139 }
1140
1141 /**
1142 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
1143 * @skb: buffer
1144 * @dst: dst entry
1145 *
1146 * Sets skb dst, assuming a reference was not taken on dst.
1147 * If dst entry is cached, we do not take reference and dst_release
1148 * will be avoided by refdst_drop. If dst entry is not cached, we take
1149 * reference, so that last dst_release can destroy the dst immediately.
1150 */
1151 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
1152 {
1153 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
1154 skb->slow_gro |= !!dst;
1155 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
1156 }
1157
1158 /**
1159 * skb_dst_is_noref - Test if skb dst isn't refcounted
1160 * @skb: buffer
1161 */
1162 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
1163 {
1164 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
1165 }
1166
1167 /**
1168 * skb_rtable - Returns the skb &rtable
1169 * @skb: buffer
1170 */
1171 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
1172 {
1173 return (struct rtable *)skb_dst(skb);
1174 }
1175
1176 /* For mangling skb->pkt_type from user space side from applications
1177 * such as nft, tc, etc, we only allow a conservative subset of
1178 * possible pkt_types to be set.
1179 */
1180 static inline bool skb_pkt_type_ok(u32 ptype)
1181 {
1182 return ptype <= PACKET_OTHERHOST;
1183 }
1184
1185 /**
1186 * skb_napi_id - Returns the skb's NAPI id
1187 * @skb: buffer
1188 */
1189 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
1190 {
1191 #ifdef CONFIG_NET_RX_BUSY_POLL
1192 return skb->napi_id;
1193 #else
1194 return 0;
1195 #endif
1196 }
1197
1198 static inline bool skb_wifi_acked_valid(const struct sk_buff *skb)
1199 {
1200 #ifdef CONFIG_WIRELESS
1201 return skb->wifi_acked_valid;
1202 #else
1203 return 0;
1204 #endif
1205 }
1206
1207 /**
1208 * skb_unref - decrement the skb's reference count
1209 * @skb: buffer
1210 *
1211 * Returns true if we can free the skb.
1212 */
1213 static inline bool skb_unref(struct sk_buff *skb)
1214 {
1215 if (unlikely(!skb))
1216 return false;
1217 if (likely(refcount_read(&skb->users) == 1))
1218 smp_rmb();
1219 else if (likely(!refcount_dec_and_test(&skb->users)))
1220 return false;
1221
1222 return true;
1223 }
1224
1225 void __fix_address
1226 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason);
1227
1228 /**
1229 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason
1230 * @skb: buffer to free
1231 */
1232 static inline void kfree_skb(struct sk_buff *skb)
1233 {
1234 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED);
1235 }
1236
1237 void skb_release_head_state(struct sk_buff *skb);
1238 void kfree_skb_list_reason(struct sk_buff *segs,
1239 enum skb_drop_reason reason);
1240 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1241 void skb_tx_error(struct sk_buff *skb);
1242
1243 static inline void kfree_skb_list(struct sk_buff *segs)
1244 {
1245 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED);
1246 }
1247
1248 #ifdef CONFIG_TRACEPOINTS
1249 void consume_skb(struct sk_buff *skb);
1250 #else
1251 static inline void consume_skb(struct sk_buff *skb)
1252 {
1253 return kfree_skb(skb);
1254 }
1255 #endif
1256
1257 void __consume_stateless_skb(struct sk_buff *skb);
1258 void __kfree_skb(struct sk_buff *skb);
1259 extern struct kmem_cache *skbuff_cache;
1260
1261 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1262 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1263 bool *fragstolen, int *delta_truesize);
1264
1265 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1266 int node);
1267 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1268 struct sk_buff *build_skb(void *data, unsigned int frag_size);
1269 struct sk_buff *build_skb_around(struct sk_buff *skb,
1270 void *data, unsigned int frag_size);
1271 void skb_attempt_defer_free(struct sk_buff *skb);
1272
1273 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size);
1274 struct sk_buff *slab_build_skb(void *data);
1275
1276 /**
1277 * alloc_skb - allocate a network buffer
1278 * @size: size to allocate
1279 * @priority: allocation mask
1280 *
1281 * This function is a convenient wrapper around __alloc_skb().
1282 */
1283 static inline struct sk_buff *alloc_skb(unsigned int size,
1284 gfp_t priority)
1285 {
1286 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1287 }
1288
1289 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1290 unsigned long data_len,
1291 int max_page_order,
1292 int *errcode,
1293 gfp_t gfp_mask);
1294 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1295
1296 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
1297 struct sk_buff_fclones {
1298 struct sk_buff skb1;
1299
1300 struct sk_buff skb2;
1301
1302 refcount_t fclone_ref;
1303 };
1304
1305 /**
1306 * skb_fclone_busy - check if fclone is busy
1307 * @sk: socket
1308 * @skb: buffer
1309 *
1310 * Returns true if skb is a fast clone, and its clone is not freed.
1311 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1312 * so we also check that this didnt happen.
1313 */
1314 static inline bool skb_fclone_busy(const struct sock *sk,
1315 const struct sk_buff *skb)
1316 {
1317 const struct sk_buff_fclones *fclones;
1318
1319 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1320
1321 return skb->fclone == SKB_FCLONE_ORIG &&
1322 refcount_read(&fclones->fclone_ref) > 1 &&
1323 READ_ONCE(fclones->skb2.sk) == sk;
1324 }
1325
1326 /**
1327 * alloc_skb_fclone - allocate a network buffer from fclone cache
1328 * @size: size to allocate
1329 * @priority: allocation mask
1330 *
1331 * This function is a convenient wrapper around __alloc_skb().
1332 */
1333 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1334 gfp_t priority)
1335 {
1336 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1337 }
1338
1339 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1340 void skb_headers_offset_update(struct sk_buff *skb, int off);
1341 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1342 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1343 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1344 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1345 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1346 gfp_t gfp_mask, bool fclone);
1347 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1348 gfp_t gfp_mask)
1349 {
1350 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1351 }
1352
1353 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1354 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1355 unsigned int headroom);
1356 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom);
1357 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1358 int newtailroom, gfp_t priority);
1359 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1360 int offset, int len);
1361 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1362 int offset, int len);
1363 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1364 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1365
1366 /**
1367 * skb_pad - zero pad the tail of an skb
1368 * @skb: buffer to pad
1369 * @pad: space to pad
1370 *
1371 * Ensure that a buffer is followed by a padding area that is zero
1372 * filled. Used by network drivers which may DMA or transfer data
1373 * beyond the buffer end onto the wire.
1374 *
1375 * May return error in out of memory cases. The skb is freed on error.
1376 */
1377 static inline int skb_pad(struct sk_buff *skb, int pad)
1378 {
1379 return __skb_pad(skb, pad, true);
1380 }
1381 #define dev_kfree_skb(a) consume_skb(a)
1382
1383 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1384 int offset, size_t size, size_t max_frags);
1385
1386 struct skb_seq_state {
1387 __u32 lower_offset;
1388 __u32 upper_offset;
1389 __u32 frag_idx;
1390 __u32 stepped_offset;
1391 struct sk_buff *root_skb;
1392 struct sk_buff *cur_skb;
1393 __u8 *frag_data;
1394 __u32 frag_off;
1395 };
1396
1397 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1398 unsigned int to, struct skb_seq_state *st);
1399 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1400 struct skb_seq_state *st);
1401 void skb_abort_seq_read(struct skb_seq_state *st);
1402
1403 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1404 unsigned int to, struct ts_config *config);
1405
1406 /*
1407 * Packet hash types specify the type of hash in skb_set_hash.
1408 *
1409 * Hash types refer to the protocol layer addresses which are used to
1410 * construct a packet's hash. The hashes are used to differentiate or identify
1411 * flows of the protocol layer for the hash type. Hash types are either
1412 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1413 *
1414 * Properties of hashes:
1415 *
1416 * 1) Two packets in different flows have different hash values
1417 * 2) Two packets in the same flow should have the same hash value
1418 *
1419 * A hash at a higher layer is considered to be more specific. A driver should
1420 * set the most specific hash possible.
1421 *
1422 * A driver cannot indicate a more specific hash than the layer at which a hash
1423 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1424 *
1425 * A driver may indicate a hash level which is less specific than the
1426 * actual layer the hash was computed on. For instance, a hash computed
1427 * at L4 may be considered an L3 hash. This should only be done if the
1428 * driver can't unambiguously determine that the HW computed the hash at
1429 * the higher layer. Note that the "should" in the second property above
1430 * permits this.
1431 */
1432 enum pkt_hash_types {
1433 PKT_HASH_TYPE_NONE, /* Undefined type */
1434 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1435 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1436 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1437 };
1438
1439 static inline void skb_clear_hash(struct sk_buff *skb)
1440 {
1441 skb->hash = 0;
1442 skb->sw_hash = 0;
1443 skb->l4_hash = 0;
1444 }
1445
1446 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1447 {
1448 if (!skb->l4_hash)
1449 skb_clear_hash(skb);
1450 }
1451
1452 static inline void
1453 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1454 {
1455 skb->l4_hash = is_l4;
1456 skb->sw_hash = is_sw;
1457 skb->hash = hash;
1458 }
1459
1460 static inline void
1461 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1462 {
1463 /* Used by drivers to set hash from HW */
1464 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1465 }
1466
1467 static inline void
1468 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1469 {
1470 __skb_set_hash(skb, hash, true, is_l4);
1471 }
1472
1473 void __skb_get_hash(struct sk_buff *skb);
1474 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1475 u32 skb_get_poff(const struct sk_buff *skb);
1476 u32 __skb_get_poff(const struct sk_buff *skb, const void *data,
1477 const struct flow_keys_basic *keys, int hlen);
1478 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1479 const void *data, int hlen_proto);
1480
1481 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1482 int thoff, u8 ip_proto)
1483 {
1484 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1485 }
1486
1487 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1488 const struct flow_dissector_key *key,
1489 unsigned int key_count);
1490
1491 struct bpf_flow_dissector;
1492 u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1493 __be16 proto, int nhoff, int hlen, unsigned int flags);
1494
1495 bool __skb_flow_dissect(const struct net *net,
1496 const struct sk_buff *skb,
1497 struct flow_dissector *flow_dissector,
1498 void *target_container, const void *data,
1499 __be16 proto, int nhoff, int hlen, unsigned int flags);
1500
1501 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1502 struct flow_dissector *flow_dissector,
1503 void *target_container, unsigned int flags)
1504 {
1505 return __skb_flow_dissect(NULL, skb, flow_dissector,
1506 target_container, NULL, 0, 0, 0, flags);
1507 }
1508
1509 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1510 struct flow_keys *flow,
1511 unsigned int flags)
1512 {
1513 memset(flow, 0, sizeof(*flow));
1514 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1515 flow, NULL, 0, 0, 0, flags);
1516 }
1517
1518 static inline bool
1519 skb_flow_dissect_flow_keys_basic(const struct net *net,
1520 const struct sk_buff *skb,
1521 struct flow_keys_basic *flow,
1522 const void *data, __be16 proto,
1523 int nhoff, int hlen, unsigned int flags)
1524 {
1525 memset(flow, 0, sizeof(*flow));
1526 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1527 data, proto, nhoff, hlen, flags);
1528 }
1529
1530 void skb_flow_dissect_meta(const struct sk_buff *skb,
1531 struct flow_dissector *flow_dissector,
1532 void *target_container);
1533
1534 /* Gets a skb connection tracking info, ctinfo map should be a
1535 * map of mapsize to translate enum ip_conntrack_info states
1536 * to user states.
1537 */
1538 void
1539 skb_flow_dissect_ct(const struct sk_buff *skb,
1540 struct flow_dissector *flow_dissector,
1541 void *target_container,
1542 u16 *ctinfo_map, size_t mapsize,
1543 bool post_ct, u16 zone);
1544 void
1545 skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1546 struct flow_dissector *flow_dissector,
1547 void *target_container);
1548
1549 void skb_flow_dissect_hash(const struct sk_buff *skb,
1550 struct flow_dissector *flow_dissector,
1551 void *target_container);
1552
1553 static inline __u32 skb_get_hash(struct sk_buff *skb)
1554 {
1555 if (!skb->l4_hash && !skb->sw_hash)
1556 __skb_get_hash(skb);
1557
1558 return skb->hash;
1559 }
1560
1561 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1562 {
1563 if (!skb->l4_hash && !skb->sw_hash) {
1564 struct flow_keys keys;
1565 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1566
1567 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1568 }
1569
1570 return skb->hash;
1571 }
1572
1573 __u32 skb_get_hash_perturb(const struct sk_buff *skb,
1574 const siphash_key_t *perturb);
1575
1576 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1577 {
1578 return skb->hash;
1579 }
1580
1581 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1582 {
1583 to->hash = from->hash;
1584 to->sw_hash = from->sw_hash;
1585 to->l4_hash = from->l4_hash;
1586 };
1587
1588 static inline int skb_cmp_decrypted(const struct sk_buff *skb1,
1589 const struct sk_buff *skb2)
1590 {
1591 #ifdef CONFIG_TLS_DEVICE
1592 return skb2->decrypted - skb1->decrypted;
1593 #else
1594 return 0;
1595 #endif
1596 }
1597
1598 static inline void skb_copy_decrypted(struct sk_buff *to,
1599 const struct sk_buff *from)
1600 {
1601 #ifdef CONFIG_TLS_DEVICE
1602 to->decrypted = from->decrypted;
1603 #endif
1604 }
1605
1606 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1607 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1608 {
1609 return skb->head + skb->end;
1610 }
1611
1612 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1613 {
1614 return skb->end;
1615 }
1616
1617 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1618 {
1619 skb->end = offset;
1620 }
1621 #else
1622 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1623 {
1624 return skb->end;
1625 }
1626
1627 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1628 {
1629 return skb->end - skb->head;
1630 }
1631
1632 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
1633 {
1634 skb->end = skb->head + offset;
1635 }
1636 #endif
1637
1638 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1639 struct ubuf_info *uarg);
1640
1641 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
1642
1643 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
1644 bool success);
1645
1646 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk,
1647 struct sk_buff *skb, struct iov_iter *from,
1648 size_t length);
1649
1650 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb,
1651 struct msghdr *msg, int len)
1652 {
1653 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len);
1654 }
1655
1656 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1657 struct msghdr *msg, int len,
1658 struct ubuf_info *uarg);
1659
1660 /* Internal */
1661 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1662
1663 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1664 {
1665 return &skb_shinfo(skb)->hwtstamps;
1666 }
1667
1668 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1669 {
1670 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE;
1671
1672 return is_zcopy ? skb_uarg(skb) : NULL;
1673 }
1674
1675 static inline bool skb_zcopy_pure(const struct sk_buff *skb)
1676 {
1677 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY;
1678 }
1679
1680 static inline bool skb_zcopy_managed(const struct sk_buff *skb)
1681 {
1682 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS;
1683 }
1684
1685 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1,
1686 const struct sk_buff *skb2)
1687 {
1688 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2);
1689 }
1690
1691 static inline void net_zcopy_get(struct ubuf_info *uarg)
1692 {
1693 refcount_inc(&uarg->refcnt);
1694 }
1695
1696 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg)
1697 {
1698 skb_shinfo(skb)->destructor_arg = uarg;
1699 skb_shinfo(skb)->flags |= uarg->flags;
1700 }
1701
1702 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1703 bool *have_ref)
1704 {
1705 if (skb && uarg && !skb_zcopy(skb)) {
1706 if (unlikely(have_ref && *have_ref))
1707 *have_ref = false;
1708 else
1709 net_zcopy_get(uarg);
1710 skb_zcopy_init(skb, uarg);
1711 }
1712 }
1713
1714 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1715 {
1716 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1717 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG;
1718 }
1719
1720 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1721 {
1722 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1723 }
1724
1725 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1726 {
1727 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1728 }
1729
1730 static inline void net_zcopy_put(struct ubuf_info *uarg)
1731 {
1732 if (uarg)
1733 uarg->callback(NULL, uarg, true);
1734 }
1735
1736 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1737 {
1738 if (uarg) {
1739 if (uarg->callback == msg_zerocopy_callback)
1740 msg_zerocopy_put_abort(uarg, have_uref);
1741 else if (have_uref)
1742 net_zcopy_put(uarg);
1743 }
1744 }
1745
1746 /* Release a reference on a zerocopy structure */
1747 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success)
1748 {
1749 struct ubuf_info *uarg = skb_zcopy(skb);
1750
1751 if (uarg) {
1752 if (!skb_zcopy_is_nouarg(skb))
1753 uarg->callback(skb, uarg, zerocopy_success);
1754
1755 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY;
1756 }
1757 }
1758
1759 void __skb_zcopy_downgrade_managed(struct sk_buff *skb);
1760
1761 static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb)
1762 {
1763 if (unlikely(skb_zcopy_managed(skb)))
1764 __skb_zcopy_downgrade_managed(skb);
1765 }
1766
1767 static inline void skb_mark_not_on_list(struct sk_buff *skb)
1768 {
1769 skb->next = NULL;
1770 }
1771
1772 static inline void skb_poison_list(struct sk_buff *skb)
1773 {
1774 #ifdef CONFIG_DEBUG_NET
1775 skb->next = SKB_LIST_POISON_NEXT;
1776 #endif
1777 }
1778
1779 /* Iterate through singly-linked GSO fragments of an skb. */
1780 #define skb_list_walk_safe(first, skb, next_skb) \
1781 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \
1782 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL)
1783
1784 static inline void skb_list_del_init(struct sk_buff *skb)
1785 {
1786 __list_del_entry(&skb->list);
1787 skb_mark_not_on_list(skb);
1788 }
1789
1790 /**
1791 * skb_queue_empty - check if a queue is empty
1792 * @list: queue head
1793 *
1794 * Returns true if the queue is empty, false otherwise.
1795 */
1796 static inline int skb_queue_empty(const struct sk_buff_head *list)
1797 {
1798 return list->next == (const struct sk_buff *) list;
1799 }
1800
1801 /**
1802 * skb_queue_empty_lockless - check if a queue is empty
1803 * @list: queue head
1804 *
1805 * Returns true if the queue is empty, false otherwise.
1806 * This variant can be used in lockless contexts.
1807 */
1808 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list)
1809 {
1810 return READ_ONCE(list->next) == (const struct sk_buff *) list;
1811 }
1812
1813
1814 /**
1815 * skb_queue_is_last - check if skb is the last entry in the queue
1816 * @list: queue head
1817 * @skb: buffer
1818 *
1819 * Returns true if @skb is the last buffer on the list.
1820 */
1821 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1822 const struct sk_buff *skb)
1823 {
1824 return skb->next == (const struct sk_buff *) list;
1825 }
1826
1827 /**
1828 * skb_queue_is_first - check if skb is the first entry in the queue
1829 * @list: queue head
1830 * @skb: buffer
1831 *
1832 * Returns true if @skb is the first buffer on the list.
1833 */
1834 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1835 const struct sk_buff *skb)
1836 {
1837 return skb->prev == (const struct sk_buff *) list;
1838 }
1839
1840 /**
1841 * skb_queue_next - return the next packet in the queue
1842 * @list: queue head
1843 * @skb: current buffer
1844 *
1845 * Return the next packet in @list after @skb. It is only valid to
1846 * call this if skb_queue_is_last() evaluates to false.
1847 */
1848 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1849 const struct sk_buff *skb)
1850 {
1851 /* This BUG_ON may seem severe, but if we just return then we
1852 * are going to dereference garbage.
1853 */
1854 BUG_ON(skb_queue_is_last(list, skb));
1855 return skb->next;
1856 }
1857
1858 /**
1859 * skb_queue_prev - return the prev packet in the queue
1860 * @list: queue head
1861 * @skb: current buffer
1862 *
1863 * Return the prev packet in @list before @skb. It is only valid to
1864 * call this if skb_queue_is_first() evaluates to false.
1865 */
1866 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1867 const struct sk_buff *skb)
1868 {
1869 /* This BUG_ON may seem severe, but if we just return then we
1870 * are going to dereference garbage.
1871 */
1872 BUG_ON(skb_queue_is_first(list, skb));
1873 return skb->prev;
1874 }
1875
1876 /**
1877 * skb_get - reference buffer
1878 * @skb: buffer to reference
1879 *
1880 * Makes another reference to a socket buffer and returns a pointer
1881 * to the buffer.
1882 */
1883 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1884 {
1885 refcount_inc(&skb->users);
1886 return skb;
1887 }
1888
1889 /*
1890 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1891 */
1892
1893 /**
1894 * skb_cloned - is the buffer a clone
1895 * @skb: buffer to check
1896 *
1897 * Returns true if the buffer was generated with skb_clone() and is
1898 * one of multiple shared copies of the buffer. Cloned buffers are
1899 * shared data so must not be written to under normal circumstances.
1900 */
1901 static inline int skb_cloned(const struct sk_buff *skb)
1902 {
1903 return skb->cloned &&
1904 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1905 }
1906
1907 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1908 {
1909 might_sleep_if(gfpflags_allow_blocking(pri));
1910
1911 if (skb_cloned(skb))
1912 return pskb_expand_head(skb, 0, 0, pri);
1913
1914 return 0;
1915 }
1916
1917 /* This variant of skb_unclone() makes sure skb->truesize
1918 * and skb_end_offset() are not changed, whenever a new skb->head is needed.
1919 *
1920 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X))
1921 * when various debugging features are in place.
1922 */
1923 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri);
1924 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
1925 {
1926 might_sleep_if(gfpflags_allow_blocking(pri));
1927
1928 if (skb_cloned(skb))
1929 return __skb_unclone_keeptruesize(skb, pri);
1930 return 0;
1931 }
1932
1933 /**
1934 * skb_header_cloned - is the header a clone
1935 * @skb: buffer to check
1936 *
1937 * Returns true if modifying the header part of the buffer requires
1938 * the data to be copied.
1939 */
1940 static inline int skb_header_cloned(const struct sk_buff *skb)
1941 {
1942 int dataref;
1943
1944 if (!skb->cloned)
1945 return 0;
1946
1947 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1948 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1949 return dataref != 1;
1950 }
1951
1952 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1953 {
1954 might_sleep_if(gfpflags_allow_blocking(pri));
1955
1956 if (skb_header_cloned(skb))
1957 return pskb_expand_head(skb, 0, 0, pri);
1958
1959 return 0;
1960 }
1961
1962 /**
1963 * __skb_header_release() - allow clones to use the headroom
1964 * @skb: buffer to operate on
1965 *
1966 * See "DOC: dataref and headerless skbs".
1967 */
1968 static inline void __skb_header_release(struct sk_buff *skb)
1969 {
1970 skb->nohdr = 1;
1971 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1972 }
1973
1974
1975 /**
1976 * skb_shared - is the buffer shared
1977 * @skb: buffer to check
1978 *
1979 * Returns true if more than one person has a reference to this
1980 * buffer.
1981 */
1982 static inline int skb_shared(const struct sk_buff *skb)
1983 {
1984 return refcount_read(&skb->users) != 1;
1985 }
1986
1987 /**
1988 * skb_share_check - check if buffer is shared and if so clone it
1989 * @skb: buffer to check
1990 * @pri: priority for memory allocation
1991 *
1992 * If the buffer is shared the buffer is cloned and the old copy
1993 * drops a reference. A new clone with a single reference is returned.
1994 * If the buffer is not shared the original buffer is returned. When
1995 * being called from interrupt status or with spinlocks held pri must
1996 * be GFP_ATOMIC.
1997 *
1998 * NULL is returned on a memory allocation failure.
1999 */
2000 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
2001 {
2002 might_sleep_if(gfpflags_allow_blocking(pri));
2003 if (skb_shared(skb)) {
2004 struct sk_buff *nskb = skb_clone(skb, pri);
2005
2006 if (likely(nskb))
2007 consume_skb(skb);
2008 else
2009 kfree_skb(skb);
2010 skb = nskb;
2011 }
2012 return skb;
2013 }
2014
2015 /*
2016 * Copy shared buffers into a new sk_buff. We effectively do COW on
2017 * packets to handle cases where we have a local reader and forward
2018 * and a couple of other messy ones. The normal one is tcpdumping
2019 * a packet thats being forwarded.
2020 */
2021
2022 /**
2023 * skb_unshare - make a copy of a shared buffer
2024 * @skb: buffer to check
2025 * @pri: priority for memory allocation
2026 *
2027 * If the socket buffer is a clone then this function creates a new
2028 * copy of the data, drops a reference count on the old copy and returns
2029 * the new copy with the reference count at 1. If the buffer is not a clone
2030 * the original buffer is returned. When called with a spinlock held or
2031 * from interrupt state @pri must be %GFP_ATOMIC
2032 *
2033 * %NULL is returned on a memory allocation failure.
2034 */
2035 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
2036 gfp_t pri)
2037 {
2038 might_sleep_if(gfpflags_allow_blocking(pri));
2039 if (skb_cloned(skb)) {
2040 struct sk_buff *nskb = skb_copy(skb, pri);
2041
2042 /* Free our shared copy */
2043 if (likely(nskb))
2044 consume_skb(skb);
2045 else
2046 kfree_skb(skb);
2047 skb = nskb;
2048 }
2049 return skb;
2050 }
2051
2052 /**
2053 * skb_peek - peek at the head of an &sk_buff_head
2054 * @list_: list to peek at
2055 *
2056 * Peek an &sk_buff. Unlike most other operations you _MUST_
2057 * be careful with this one. A peek leaves the buffer on the
2058 * list and someone else may run off with it. You must hold
2059 * the appropriate locks or have a private queue to do this.
2060 *
2061 * Returns %NULL for an empty list or a pointer to the head element.
2062 * The reference count is not incremented and the reference is therefore
2063 * volatile. Use with caution.
2064 */
2065 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
2066 {
2067 struct sk_buff *skb = list_->next;
2068
2069 if (skb == (struct sk_buff *)list_)
2070 skb = NULL;
2071 return skb;
2072 }
2073
2074 /**
2075 * __skb_peek - peek at the head of a non-empty &sk_buff_head
2076 * @list_: list to peek at
2077 *
2078 * Like skb_peek(), but the caller knows that the list is not empty.
2079 */
2080 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
2081 {
2082 return list_->next;
2083 }
2084
2085 /**
2086 * skb_peek_next - peek skb following the given one from a queue
2087 * @skb: skb to start from
2088 * @list_: list to peek at
2089 *
2090 * Returns %NULL when the end of the list is met or a pointer to the
2091 * next element. The reference count is not incremented and the
2092 * reference is therefore volatile. Use with caution.
2093 */
2094 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
2095 const struct sk_buff_head *list_)
2096 {
2097 struct sk_buff *next = skb->next;
2098
2099 if (next == (struct sk_buff *)list_)
2100 next = NULL;
2101 return next;
2102 }
2103
2104 /**
2105 * skb_peek_tail - peek at the tail of an &sk_buff_head
2106 * @list_: list to peek at
2107 *
2108 * Peek an &sk_buff. Unlike most other operations you _MUST_
2109 * be careful with this one. A peek leaves the buffer on the
2110 * list and someone else may run off with it. You must hold
2111 * the appropriate locks or have a private queue to do this.
2112 *
2113 * Returns %NULL for an empty list or a pointer to the tail element.
2114 * The reference count is not incremented and the reference is therefore
2115 * volatile. Use with caution.
2116 */
2117 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
2118 {
2119 struct sk_buff *skb = READ_ONCE(list_->prev);
2120
2121 if (skb == (struct sk_buff *)list_)
2122 skb = NULL;
2123 return skb;
2124
2125 }
2126
2127 /**
2128 * skb_queue_len - get queue length
2129 * @list_: list to measure
2130 *
2131 * Return the length of an &sk_buff queue.
2132 */
2133 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
2134 {
2135 return list_->qlen;
2136 }
2137
2138 /**
2139 * skb_queue_len_lockless - get queue length
2140 * @list_: list to measure
2141 *
2142 * Return the length of an &sk_buff queue.
2143 * This variant can be used in lockless contexts.
2144 */
2145 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_)
2146 {
2147 return READ_ONCE(list_->qlen);
2148 }
2149
2150 /**
2151 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
2152 * @list: queue to initialize
2153 *
2154 * This initializes only the list and queue length aspects of
2155 * an sk_buff_head object. This allows to initialize the list
2156 * aspects of an sk_buff_head without reinitializing things like
2157 * the spinlock. It can also be used for on-stack sk_buff_head
2158 * objects where the spinlock is known to not be used.
2159 */
2160 static inline void __skb_queue_head_init(struct sk_buff_head *list)
2161 {
2162 list->prev = list->next = (struct sk_buff *)list;
2163 list->qlen = 0;
2164 }
2165
2166 /*
2167 * This function creates a split out lock class for each invocation;
2168 * this is needed for now since a whole lot of users of the skb-queue
2169 * infrastructure in drivers have different locking usage (in hardirq)
2170 * than the networking core (in softirq only). In the long run either the
2171 * network layer or drivers should need annotation to consolidate the
2172 * main types of usage into 3 classes.
2173 */
2174 static inline void skb_queue_head_init(struct sk_buff_head *list)
2175 {
2176 spin_lock_init(&list->lock);
2177 __skb_queue_head_init(list);
2178 }
2179
2180 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
2181 struct lock_class_key *class)
2182 {
2183 skb_queue_head_init(list);
2184 lockdep_set_class(&list->lock, class);
2185 }
2186
2187 /*
2188 * Insert an sk_buff on a list.
2189 *
2190 * The "__skb_xxxx()" functions are the non-atomic ones that
2191 * can only be called with interrupts disabled.
2192 */
2193 static inline void __skb_insert(struct sk_buff *newsk,
2194 struct sk_buff *prev, struct sk_buff *next,
2195 struct sk_buff_head *list)
2196 {
2197 /* See skb_queue_empty_lockless() and skb_peek_tail()
2198 * for the opposite READ_ONCE()
2199 */
2200 WRITE_ONCE(newsk->next, next);
2201 WRITE_ONCE(newsk->prev, prev);
2202 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk);
2203 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk);
2204 WRITE_ONCE(list->qlen, list->qlen + 1);
2205 }
2206
2207 static inline void __skb_queue_splice(const struct sk_buff_head *list,
2208 struct sk_buff *prev,
2209 struct sk_buff *next)
2210 {
2211 struct sk_buff *first = list->next;
2212 struct sk_buff *last = list->prev;
2213
2214 WRITE_ONCE(first->prev, prev);
2215 WRITE_ONCE(prev->next, first);
2216
2217 WRITE_ONCE(last->next, next);
2218 WRITE_ONCE(next->prev, last);
2219 }
2220
2221 /**
2222 * skb_queue_splice - join two skb lists, this is designed for stacks
2223 * @list: the new list to add
2224 * @head: the place to add it in the first list
2225 */
2226 static inline void skb_queue_splice(const struct sk_buff_head *list,
2227 struct sk_buff_head *head)
2228 {
2229 if (!skb_queue_empty(list)) {
2230 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2231 head->qlen += list->qlen;
2232 }
2233 }
2234
2235 /**
2236 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
2237 * @list: the new list to add
2238 * @head: the place to add it in the first list
2239 *
2240 * The list at @list is reinitialised
2241 */
2242 static inline void skb_queue_splice_init(struct sk_buff_head *list,
2243 struct sk_buff_head *head)
2244 {
2245 if (!skb_queue_empty(list)) {
2246 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
2247 head->qlen += list->qlen;
2248 __skb_queue_head_init(list);
2249 }
2250 }
2251
2252 /**
2253 * skb_queue_splice_tail - join two skb lists, each list being a queue
2254 * @list: the new list to add
2255 * @head: the place to add it in the first list
2256 */
2257 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
2258 struct sk_buff_head *head)
2259 {
2260 if (!skb_queue_empty(list)) {
2261 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2262 head->qlen += list->qlen;
2263 }
2264 }
2265
2266 /**
2267 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
2268 * @list: the new list to add
2269 * @head: the place to add it in the first list
2270 *
2271 * Each of the lists is a queue.
2272 * The list at @list is reinitialised
2273 */
2274 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
2275 struct sk_buff_head *head)
2276 {
2277 if (!skb_queue_empty(list)) {
2278 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
2279 head->qlen += list->qlen;
2280 __skb_queue_head_init(list);
2281 }
2282 }
2283
2284 /**
2285 * __skb_queue_after - queue a buffer at the list head
2286 * @list: list to use
2287 * @prev: place after this buffer
2288 * @newsk: buffer to queue
2289 *
2290 * Queue a buffer int the middle of a list. This function takes no locks
2291 * and you must therefore hold required locks before calling it.
2292 *
2293 * A buffer cannot be placed on two lists at the same time.
2294 */
2295 static inline void __skb_queue_after(struct sk_buff_head *list,
2296 struct sk_buff *prev,
2297 struct sk_buff *newsk)
2298 {
2299 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list);
2300 }
2301
2302 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
2303 struct sk_buff_head *list);
2304
2305 static inline void __skb_queue_before(struct sk_buff_head *list,
2306 struct sk_buff *next,
2307 struct sk_buff *newsk)
2308 {
2309 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list);
2310 }
2311
2312 /**
2313 * __skb_queue_head - queue a buffer at the list head
2314 * @list: list to use
2315 * @newsk: buffer to queue
2316 *
2317 * Queue a buffer at the start of a list. This function takes no locks
2318 * and you must therefore hold required locks before calling it.
2319 *
2320 * A buffer cannot be placed on two lists at the same time.
2321 */
2322 static inline void __skb_queue_head(struct sk_buff_head *list,
2323 struct sk_buff *newsk)
2324 {
2325 __skb_queue_after(list, (struct sk_buff *)list, newsk);
2326 }
2327 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
2328
2329 /**
2330 * __skb_queue_tail - queue a buffer at the list tail
2331 * @list: list to use
2332 * @newsk: buffer to queue
2333 *
2334 * Queue a buffer at the end of a list. This function takes no locks
2335 * and you must therefore hold required locks before calling it.
2336 *
2337 * A buffer cannot be placed on two lists at the same time.
2338 */
2339 static inline void __skb_queue_tail(struct sk_buff_head *list,
2340 struct sk_buff *newsk)
2341 {
2342 __skb_queue_before(list, (struct sk_buff *)list, newsk);
2343 }
2344 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
2345
2346 /*
2347 * remove sk_buff from list. _Must_ be called atomically, and with
2348 * the list known..
2349 */
2350 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2351 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2352 {
2353 struct sk_buff *next, *prev;
2354
2355 WRITE_ONCE(list->qlen, list->qlen - 1);
2356 next = skb->next;
2357 prev = skb->prev;
2358 skb->next = skb->prev = NULL;
2359 WRITE_ONCE(next->prev, prev);
2360 WRITE_ONCE(prev->next, next);
2361 }
2362
2363 /**
2364 * __skb_dequeue - remove from the head of the queue
2365 * @list: list to dequeue from
2366 *
2367 * Remove the head of the list. This function does not take any locks
2368 * so must be used with appropriate locks held only. The head item is
2369 * returned or %NULL if the list is empty.
2370 */
2371 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2372 {
2373 struct sk_buff *skb = skb_peek(list);
2374 if (skb)
2375 __skb_unlink(skb, list);
2376 return skb;
2377 }
2378 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2379
2380 /**
2381 * __skb_dequeue_tail - remove from the tail of the queue
2382 * @list: list to dequeue from
2383 *
2384 * Remove the tail of the list. This function does not take any locks
2385 * so must be used with appropriate locks held only. The tail item is
2386 * returned or %NULL if the list is empty.
2387 */
2388 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2389 {
2390 struct sk_buff *skb = skb_peek_tail(list);
2391 if (skb)
2392 __skb_unlink(skb, list);
2393 return skb;
2394 }
2395 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2396
2397
2398 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2399 {
2400 return skb->data_len;
2401 }
2402
2403 static inline unsigned int skb_headlen(const struct sk_buff *skb)
2404 {
2405 return skb->len - skb->data_len;
2406 }
2407
2408 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2409 {
2410 unsigned int i, len = 0;
2411
2412 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2413 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2414 return len;
2415 }
2416
2417 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2418 {
2419 return skb_headlen(skb) + __skb_pagelen(skb);
2420 }
2421
2422 static inline void skb_frag_fill_page_desc(skb_frag_t *frag,
2423 struct page *page,
2424 int off, int size)
2425 {
2426 frag->bv_page = page;
2427 frag->bv_offset = off;
2428 skb_frag_size_set(frag, size);
2429 }
2430
2431 static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo,
2432 int i, struct page *page,
2433 int off, int size)
2434 {
2435 skb_frag_t *frag = &shinfo->frags[i];
2436
2437 skb_frag_fill_page_desc(frag, page, off, size);
2438 }
2439
2440 /**
2441 * skb_len_add - adds a number to len fields of skb
2442 * @skb: buffer to add len to
2443 * @delta: number of bytes to add
2444 */
2445 static inline void skb_len_add(struct sk_buff *skb, int delta)
2446 {
2447 skb->len += delta;
2448 skb->data_len += delta;
2449 skb->truesize += delta;
2450 }
2451
2452 /**
2453 * __skb_fill_page_desc - initialise a paged fragment in an skb
2454 * @skb: buffer containing fragment to be initialised
2455 * @i: paged fragment index to initialise
2456 * @page: the page to use for this fragment
2457 * @off: the offset to the data with @page
2458 * @size: the length of the data
2459 *
2460 * Initialises the @i'th fragment of @skb to point to &size bytes at
2461 * offset @off within @page.
2462 *
2463 * Does not take any additional reference on the fragment.
2464 */
2465 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2466 struct page *page, int off, int size)
2467 {
2468 __skb_fill_page_desc_noacc(skb_shinfo(skb), i, page, off, size);
2469
2470 /* Propagate page pfmemalloc to the skb if we can. The problem is
2471 * that not all callers have unique ownership of the page but rely
2472 * on page_is_pfmemalloc doing the right thing(tm).
2473 */
2474 page = compound_head(page);
2475 if (page_is_pfmemalloc(page))
2476 skb->pfmemalloc = true;
2477 }
2478
2479 /**
2480 * skb_fill_page_desc - initialise a paged fragment in an skb
2481 * @skb: buffer containing fragment to be initialised
2482 * @i: paged fragment index to initialise
2483 * @page: the page to use for this fragment
2484 * @off: the offset to the data with @page
2485 * @size: the length of the data
2486 *
2487 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2488 * @skb to point to @size bytes at offset @off within @page. In
2489 * addition updates @skb such that @i is the last fragment.
2490 *
2491 * Does not take any additional reference on the fragment.
2492 */
2493 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2494 struct page *page, int off, int size)
2495 {
2496 __skb_fill_page_desc(skb, i, page, off, size);
2497 skb_shinfo(skb)->nr_frags = i + 1;
2498 }
2499
2500 /**
2501 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb
2502 * @skb: buffer containing fragment to be initialised
2503 * @i: paged fragment index to initialise
2504 * @page: the page to use for this fragment
2505 * @off: the offset to the data with @page
2506 * @size: the length of the data
2507 *
2508 * Variant of skb_fill_page_desc() which does not deal with
2509 * pfmemalloc, if page is not owned by us.
2510 */
2511 static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i,
2512 struct page *page, int off,
2513 int size)
2514 {
2515 struct skb_shared_info *shinfo = skb_shinfo(skb);
2516
2517 __skb_fill_page_desc_noacc(shinfo, i, page, off, size);
2518 shinfo->nr_frags = i + 1;
2519 }
2520
2521 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2522 int size, unsigned int truesize);
2523
2524 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2525 unsigned int truesize);
2526
2527 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
2528
2529 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2530 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2531 {
2532 return skb->head + skb->tail;
2533 }
2534
2535 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2536 {
2537 skb->tail = skb->data - skb->head;
2538 }
2539
2540 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2541 {
2542 skb_reset_tail_pointer(skb);
2543 skb->tail += offset;
2544 }
2545
2546 #else /* NET_SKBUFF_DATA_USES_OFFSET */
2547 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2548 {
2549 return skb->tail;
2550 }
2551
2552 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2553 {
2554 skb->tail = skb->data;
2555 }
2556
2557 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2558 {
2559 skb->tail = skb->data + offset;
2560 }
2561
2562 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
2563
2564 static inline void skb_assert_len(struct sk_buff *skb)
2565 {
2566 #ifdef CONFIG_DEBUG_NET
2567 if (WARN_ONCE(!skb->len, "%s\n", __func__))
2568 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
2569 #endif /* CONFIG_DEBUG_NET */
2570 }
2571
2572 /*
2573 * Add data to an sk_buff
2574 */
2575 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2576 void *skb_put(struct sk_buff *skb, unsigned int len);
2577 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2578 {
2579 void *tmp = skb_tail_pointer(skb);
2580 SKB_LINEAR_ASSERT(skb);
2581 skb->tail += len;
2582 skb->len += len;
2583 return tmp;
2584 }
2585
2586 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2587 {
2588 void *tmp = __skb_put(skb, len);
2589
2590 memset(tmp, 0, len);
2591 return tmp;
2592 }
2593
2594 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2595 unsigned int len)
2596 {
2597 void *tmp = __skb_put(skb, len);
2598
2599 memcpy(tmp, data, len);
2600 return tmp;
2601 }
2602
2603 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2604 {
2605 *(u8 *)__skb_put(skb, 1) = val;
2606 }
2607
2608 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2609 {
2610 void *tmp = skb_put(skb, len);
2611
2612 memset(tmp, 0, len);
2613
2614 return tmp;
2615 }
2616
2617 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2618 unsigned int len)
2619 {
2620 void *tmp = skb_put(skb, len);
2621
2622 memcpy(tmp, data, len);
2623
2624 return tmp;
2625 }
2626
2627 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2628 {
2629 *(u8 *)skb_put(skb, 1) = val;
2630 }
2631
2632 void *skb_push(struct sk_buff *skb, unsigned int len);
2633 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2634 {
2635 skb->data -= len;
2636 skb->len += len;
2637 return skb->data;
2638 }
2639
2640 void *skb_pull(struct sk_buff *skb, unsigned int len);
2641 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2642 {
2643 skb->len -= len;
2644 if (unlikely(skb->len < skb->data_len)) {
2645 #if defined(CONFIG_DEBUG_NET)
2646 skb->len += len;
2647 pr_err("__skb_pull(len=%u)\n", len);
2648 skb_dump(KERN_ERR, skb, false);
2649 #endif
2650 BUG();
2651 }
2652 return skb->data += len;
2653 }
2654
2655 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2656 {
2657 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2658 }
2659
2660 void *skb_pull_data(struct sk_buff *skb, size_t len);
2661
2662 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2663
2664 static inline enum skb_drop_reason
2665 pskb_may_pull_reason(struct sk_buff *skb, unsigned int len)
2666 {
2667 if (likely(len <= skb_headlen(skb)))
2668 return SKB_NOT_DROPPED_YET;
2669
2670 if (unlikely(len > skb->len))
2671 return SKB_DROP_REASON_PKT_TOO_SMALL;
2672
2673 if (unlikely(!__pskb_pull_tail(skb, len - skb_headlen(skb))))
2674 return SKB_DROP_REASON_NOMEM;
2675
2676 return SKB_NOT_DROPPED_YET;
2677 }
2678
2679 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len)
2680 {
2681 return pskb_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET;
2682 }
2683
2684 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2685 {
2686 if (!pskb_may_pull(skb, len))
2687 return NULL;
2688
2689 skb->len -= len;
2690 return skb->data += len;
2691 }
2692
2693 void skb_condense(struct sk_buff *skb);
2694
2695 /**
2696 * skb_headroom - bytes at buffer head
2697 * @skb: buffer to check
2698 *
2699 * Return the number of bytes of free space at the head of an &sk_buff.
2700 */
2701 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2702 {
2703 return skb->data - skb->head;
2704 }
2705
2706 /**
2707 * skb_tailroom - bytes at buffer end
2708 * @skb: buffer to check
2709 *
2710 * Return the number of bytes of free space at the tail of an sk_buff
2711 */
2712 static inline int skb_tailroom(const struct sk_buff *skb)
2713 {
2714 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2715 }
2716
2717 /**
2718 * skb_availroom - bytes at buffer end
2719 * @skb: buffer to check
2720 *
2721 * Return the number of bytes of free space at the tail of an sk_buff
2722 * allocated by sk_stream_alloc()
2723 */
2724 static inline int skb_availroom(const struct sk_buff *skb)
2725 {
2726 if (skb_is_nonlinear(skb))
2727 return 0;
2728
2729 return skb->end - skb->tail - skb->reserved_tailroom;
2730 }
2731
2732 /**
2733 * skb_reserve - adjust headroom
2734 * @skb: buffer to alter
2735 * @len: bytes to move
2736 *
2737 * Increase the headroom of an empty &sk_buff by reducing the tail
2738 * room. This is only allowed for an empty buffer.
2739 */
2740 static inline void skb_reserve(struct sk_buff *skb, int len)
2741 {
2742 skb->data += len;
2743 skb->tail += len;
2744 }
2745
2746 /**
2747 * skb_tailroom_reserve - adjust reserved_tailroom
2748 * @skb: buffer to alter
2749 * @mtu: maximum amount of headlen permitted
2750 * @needed_tailroom: minimum amount of reserved_tailroom
2751 *
2752 * Set reserved_tailroom so that headlen can be as large as possible but
2753 * not larger than mtu and tailroom cannot be smaller than
2754 * needed_tailroom.
2755 * The required headroom should already have been reserved before using
2756 * this function.
2757 */
2758 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2759 unsigned int needed_tailroom)
2760 {
2761 SKB_LINEAR_ASSERT(skb);
2762 if (mtu < skb_tailroom(skb) - needed_tailroom)
2763 /* use at most mtu */
2764 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2765 else
2766 /* use up to all available space */
2767 skb->reserved_tailroom = needed_tailroom;
2768 }
2769
2770 #define ENCAP_TYPE_ETHER 0
2771 #define ENCAP_TYPE_IPPROTO 1
2772
2773 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2774 __be16 protocol)
2775 {
2776 skb->inner_protocol = protocol;
2777 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2778 }
2779
2780 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2781 __u8 ipproto)
2782 {
2783 skb->inner_ipproto = ipproto;
2784 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2785 }
2786
2787 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2788 {
2789 skb->inner_mac_header = skb->mac_header;
2790 skb->inner_network_header = skb->network_header;
2791 skb->inner_transport_header = skb->transport_header;
2792 }
2793
2794 static inline void skb_reset_mac_len(struct sk_buff *skb)
2795 {
2796 skb->mac_len = skb->network_header - skb->mac_header;
2797 }
2798
2799 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2800 *skb)
2801 {
2802 return skb->head + skb->inner_transport_header;
2803 }
2804
2805 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2806 {
2807 return skb_inner_transport_header(skb) - skb->data;
2808 }
2809
2810 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2811 {
2812 skb->inner_transport_header = skb->data - skb->head;
2813 }
2814
2815 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2816 const int offset)
2817 {
2818 skb_reset_inner_transport_header(skb);
2819 skb->inner_transport_header += offset;
2820 }
2821
2822 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2823 {
2824 return skb->head + skb->inner_network_header;
2825 }
2826
2827 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2828 {
2829 skb->inner_network_header = skb->data - skb->head;
2830 }
2831
2832 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2833 const int offset)
2834 {
2835 skb_reset_inner_network_header(skb);
2836 skb->inner_network_header += offset;
2837 }
2838
2839 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2840 {
2841 return skb->head + skb->inner_mac_header;
2842 }
2843
2844 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2845 {
2846 skb->inner_mac_header = skb->data - skb->head;
2847 }
2848
2849 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2850 const int offset)
2851 {
2852 skb_reset_inner_mac_header(skb);
2853 skb->inner_mac_header += offset;
2854 }
2855 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2856 {
2857 return skb->transport_header != (typeof(skb->transport_header))~0U;
2858 }
2859
2860 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2861 {
2862 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb));
2863 return skb->head + skb->transport_header;
2864 }
2865
2866 static inline void skb_reset_transport_header(struct sk_buff *skb)
2867 {
2868 skb->transport_header = skb->data - skb->head;
2869 }
2870
2871 static inline void skb_set_transport_header(struct sk_buff *skb,
2872 const int offset)
2873 {
2874 skb_reset_transport_header(skb);
2875 skb->transport_header += offset;
2876 }
2877
2878 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2879 {
2880 return skb->head + skb->network_header;
2881 }
2882
2883 static inline void skb_reset_network_header(struct sk_buff *skb)
2884 {
2885 skb->network_header = skb->data - skb->head;
2886 }
2887
2888 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2889 {
2890 skb_reset_network_header(skb);
2891 skb->network_header += offset;
2892 }
2893
2894 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2895 {
2896 return skb->mac_header != (typeof(skb->mac_header))~0U;
2897 }
2898
2899 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2900 {
2901 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
2902 return skb->head + skb->mac_header;
2903 }
2904
2905 static inline int skb_mac_offset(const struct sk_buff *skb)
2906 {
2907 return skb_mac_header(skb) - skb->data;
2908 }
2909
2910 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2911 {
2912 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb));
2913 return skb->network_header - skb->mac_header;
2914 }
2915
2916 static inline void skb_unset_mac_header(struct sk_buff *skb)
2917 {
2918 skb->mac_header = (typeof(skb->mac_header))~0U;
2919 }
2920
2921 static inline void skb_reset_mac_header(struct sk_buff *skb)
2922 {
2923 skb->mac_header = skb->data - skb->head;
2924 }
2925
2926 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2927 {
2928 skb_reset_mac_header(skb);
2929 skb->mac_header += offset;
2930 }
2931
2932 static inline void skb_pop_mac_header(struct sk_buff *skb)
2933 {
2934 skb->mac_header = skb->network_header;
2935 }
2936
2937 static inline void skb_probe_transport_header(struct sk_buff *skb)
2938 {
2939 struct flow_keys_basic keys;
2940
2941 if (skb_transport_header_was_set(skb))
2942 return;
2943
2944 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
2945 NULL, 0, 0, 0, 0))
2946 skb_set_transport_header(skb, keys.control.thoff);
2947 }
2948
2949 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2950 {
2951 if (skb_mac_header_was_set(skb)) {
2952 const unsigned char *old_mac = skb_mac_header(skb);
2953
2954 skb_set_mac_header(skb, -skb->mac_len);
2955 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2956 }
2957 }
2958
2959 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2960 {
2961 return skb->csum_start - skb_headroom(skb);
2962 }
2963
2964 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2965 {
2966 return skb->head + skb->csum_start;
2967 }
2968
2969 static inline int skb_transport_offset(const struct sk_buff *skb)
2970 {
2971 return skb_transport_header(skb) - skb->data;
2972 }
2973
2974 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2975 {
2976 return skb->transport_header - skb->network_header;
2977 }
2978
2979 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2980 {
2981 return skb->inner_transport_header - skb->inner_network_header;
2982 }
2983
2984 static inline int skb_network_offset(const struct sk_buff *skb)
2985 {
2986 return skb_network_header(skb) - skb->data;
2987 }
2988
2989 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2990 {
2991 return skb_inner_network_header(skb) - skb->data;
2992 }
2993
2994 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2995 {
2996 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2997 }
2998
2999 /*
3000 * CPUs often take a performance hit when accessing unaligned memory
3001 * locations. The actual performance hit varies, it can be small if the
3002 * hardware handles it or large if we have to take an exception and fix it
3003 * in software.
3004 *
3005 * Since an ethernet header is 14 bytes network drivers often end up with
3006 * the IP header at an unaligned offset. The IP header can be aligned by
3007 * shifting the start of the packet by 2 bytes. Drivers should do this
3008 * with:
3009 *
3010 * skb_reserve(skb, NET_IP_ALIGN);
3011 *
3012 * The downside to this alignment of the IP header is that the DMA is now
3013 * unaligned. On some architectures the cost of an unaligned DMA is high
3014 * and this cost outweighs the gains made by aligning the IP header.
3015 *
3016 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
3017 * to be overridden.
3018 */
3019 #ifndef NET_IP_ALIGN
3020 #define NET_IP_ALIGN 2
3021 #endif
3022
3023 /*
3024 * The networking layer reserves some headroom in skb data (via
3025 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
3026 * the header has to grow. In the default case, if the header has to grow
3027 * 32 bytes or less we avoid the reallocation.
3028 *
3029 * Unfortunately this headroom changes the DMA alignment of the resulting
3030 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
3031 * on some architectures. An architecture can override this value,
3032 * perhaps setting it to a cacheline in size (since that will maintain
3033 * cacheline alignment of the DMA). It must be a power of 2.
3034 *
3035 * Various parts of the networking layer expect at least 32 bytes of
3036 * headroom, you should not reduce this.
3037 *
3038 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
3039 * to reduce average number of cache lines per packet.
3040 * get_rps_cpu() for example only access one 64 bytes aligned block :
3041 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
3042 */
3043 #ifndef NET_SKB_PAD
3044 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
3045 #endif
3046
3047 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
3048
3049 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
3050 {
3051 if (WARN_ON(skb_is_nonlinear(skb)))
3052 return;
3053 skb->len = len;
3054 skb_set_tail_pointer(skb, len);
3055 }
3056
3057 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
3058 {
3059 __skb_set_length(skb, len);
3060 }
3061
3062 void skb_trim(struct sk_buff *skb, unsigned int len);
3063
3064 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
3065 {
3066 if (skb->data_len)
3067 return ___pskb_trim(skb, len);
3068 __skb_trim(skb, len);
3069 return 0;
3070 }
3071
3072 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
3073 {
3074 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
3075 }
3076
3077 /**
3078 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
3079 * @skb: buffer to alter
3080 * @len: new length
3081 *
3082 * This is identical to pskb_trim except that the caller knows that
3083 * the skb is not cloned so we should never get an error due to out-
3084 * of-memory.
3085 */
3086 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
3087 {
3088 int err = pskb_trim(skb, len);
3089 BUG_ON(err);
3090 }
3091
3092 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
3093 {
3094 unsigned int diff = len - skb->len;
3095
3096 if (skb_tailroom(skb) < diff) {
3097 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
3098 GFP_ATOMIC);
3099 if (ret)
3100 return ret;
3101 }
3102 __skb_set_length(skb, len);
3103 return 0;
3104 }
3105
3106 /**
3107 * skb_orphan - orphan a buffer
3108 * @skb: buffer to orphan
3109 *
3110 * If a buffer currently has an owner then we call the owner's
3111 * destructor function and make the @skb unowned. The buffer continues
3112 * to exist but is no longer charged to its former owner.
3113 */
3114 static inline void skb_orphan(struct sk_buff *skb)
3115 {
3116 if (skb->destructor) {
3117 skb->destructor(skb);
3118 skb->destructor = NULL;
3119 skb->sk = NULL;
3120 } else {
3121 BUG_ON(skb->sk);
3122 }
3123 }
3124
3125 /**
3126 * skb_orphan_frags - orphan the frags contained in a buffer
3127 * @skb: buffer to orphan frags from
3128 * @gfp_mask: allocation mask for replacement pages
3129 *
3130 * For each frag in the SKB which needs a destructor (i.e. has an
3131 * owner) create a copy of that frag and release the original
3132 * page by calling the destructor.
3133 */
3134 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
3135 {
3136 if (likely(!skb_zcopy(skb)))
3137 return 0;
3138 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN)
3139 return 0;
3140 return skb_copy_ubufs(skb, gfp_mask);
3141 }
3142
3143 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
3144 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
3145 {
3146 if (likely(!skb_zcopy(skb)))
3147 return 0;
3148 return skb_copy_ubufs(skb, gfp_mask);
3149 }
3150
3151 /**
3152 * __skb_queue_purge_reason - empty a list
3153 * @list: list to empty
3154 * @reason: drop reason
3155 *
3156 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3157 * the list and one reference dropped. This function does not take the
3158 * list lock and the caller must hold the relevant locks to use it.
3159 */
3160 static inline void __skb_queue_purge_reason(struct sk_buff_head *list,
3161 enum skb_drop_reason reason)
3162 {
3163 struct sk_buff *skb;
3164
3165 while ((skb = __skb_dequeue(list)) != NULL)
3166 kfree_skb_reason(skb, reason);
3167 }
3168
3169 static inline void __skb_queue_purge(struct sk_buff_head *list)
3170 {
3171 __skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE);
3172 }
3173
3174 void skb_queue_purge_reason(struct sk_buff_head *list,
3175 enum skb_drop_reason reason);
3176
3177 static inline void skb_queue_purge(struct sk_buff_head *list)
3178 {
3179 skb_queue_purge_reason(list, SKB_DROP_REASON_QUEUE_PURGE);
3180 }
3181
3182 unsigned int skb_rbtree_purge(struct rb_root *root);
3183 void skb_errqueue_purge(struct sk_buff_head *list);
3184
3185 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3186
3187 /**
3188 * netdev_alloc_frag - allocate a page fragment
3189 * @fragsz: fragment size
3190 *
3191 * Allocates a frag from a page for receive buffer.
3192 * Uses GFP_ATOMIC allocations.
3193 */
3194 static inline void *netdev_alloc_frag(unsigned int fragsz)
3195 {
3196 return __netdev_alloc_frag_align(fragsz, ~0u);
3197 }
3198
3199 static inline void *netdev_alloc_frag_align(unsigned int fragsz,
3200 unsigned int align)
3201 {
3202 WARN_ON_ONCE(!is_power_of_2(align));
3203 return __netdev_alloc_frag_align(fragsz, -align);
3204 }
3205
3206 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
3207 gfp_t gfp_mask);
3208
3209 /**
3210 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
3211 * @dev: network device to receive on
3212 * @length: length to allocate
3213 *
3214 * Allocate a new &sk_buff and assign it a usage count of one. The
3215 * buffer has unspecified headroom built in. Users should allocate
3216 * the headroom they think they need without accounting for the
3217 * built in space. The built in space is used for optimisations.
3218 *
3219 * %NULL is returned if there is no free memory. Although this function
3220 * allocates memory it can be called from an interrupt.
3221 */
3222 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
3223 unsigned int length)
3224 {
3225 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
3226 }
3227
3228 /* legacy helper around __netdev_alloc_skb() */
3229 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
3230 gfp_t gfp_mask)
3231 {
3232 return __netdev_alloc_skb(NULL, length, gfp_mask);
3233 }
3234
3235 /* legacy helper around netdev_alloc_skb() */
3236 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
3237 {
3238 return netdev_alloc_skb(NULL, length);
3239 }
3240
3241
3242 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
3243 unsigned int length, gfp_t gfp)
3244 {
3245 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
3246
3247 if (NET_IP_ALIGN && skb)
3248 skb_reserve(skb, NET_IP_ALIGN);
3249 return skb;
3250 }
3251
3252 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
3253 unsigned int length)
3254 {
3255 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
3256 }
3257
3258 static inline void skb_free_frag(void *addr)
3259 {
3260 page_frag_free(addr);
3261 }
3262
3263 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask);
3264
3265 static inline void *napi_alloc_frag(unsigned int fragsz)
3266 {
3267 return __napi_alloc_frag_align(fragsz, ~0u);
3268 }
3269
3270 static inline void *napi_alloc_frag_align(unsigned int fragsz,
3271 unsigned int align)
3272 {
3273 WARN_ON_ONCE(!is_power_of_2(align));
3274 return __napi_alloc_frag_align(fragsz, -align);
3275 }
3276
3277 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
3278 unsigned int length, gfp_t gfp_mask);
3279 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
3280 unsigned int length)
3281 {
3282 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
3283 }
3284 void napi_consume_skb(struct sk_buff *skb, int budget);
3285
3286 void napi_skb_free_stolen_head(struct sk_buff *skb);
3287 void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason);
3288
3289 /**
3290 * __dev_alloc_pages - allocate page for network Rx
3291 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3292 * @order: size of the allocation
3293 *
3294 * Allocate a new page.
3295 *
3296 * %NULL is returned if there is no free memory.
3297 */
3298 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
3299 unsigned int order)
3300 {
3301 /* This piece of code contains several assumptions.
3302 * 1. This is for device Rx, therefor a cold page is preferred.
3303 * 2. The expectation is the user wants a compound page.
3304 * 3. If requesting a order 0 page it will not be compound
3305 * due to the check to see if order has a value in prep_new_page
3306 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
3307 * code in gfp_to_alloc_flags that should be enforcing this.
3308 */
3309 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
3310
3311 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
3312 }
3313
3314 static inline struct page *dev_alloc_pages(unsigned int order)
3315 {
3316 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
3317 }
3318
3319 /**
3320 * __dev_alloc_page - allocate a page for network Rx
3321 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
3322 *
3323 * Allocate a new page.
3324 *
3325 * %NULL is returned if there is no free memory.
3326 */
3327 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
3328 {
3329 return __dev_alloc_pages(gfp_mask, 0);
3330 }
3331
3332 static inline struct page *dev_alloc_page(void)
3333 {
3334 return dev_alloc_pages(0);
3335 }
3336
3337 /**
3338 * dev_page_is_reusable - check whether a page can be reused for network Rx
3339 * @page: the page to test
3340 *
3341 * A page shouldn't be considered for reusing/recycling if it was allocated
3342 * under memory pressure or at a distant memory node.
3343 *
3344 * Returns false if this page should be returned to page allocator, true
3345 * otherwise.
3346 */
3347 static inline bool dev_page_is_reusable(const struct page *page)
3348 {
3349 return likely(page_to_nid(page) == numa_mem_id() &&
3350 !page_is_pfmemalloc(page));
3351 }
3352
3353 /**
3354 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
3355 * @page: The page that was allocated from skb_alloc_page
3356 * @skb: The skb that may need pfmemalloc set
3357 */
3358 static inline void skb_propagate_pfmemalloc(const struct page *page,
3359 struct sk_buff *skb)
3360 {
3361 if (page_is_pfmemalloc(page))
3362 skb->pfmemalloc = true;
3363 }
3364
3365 /**
3366 * skb_frag_off() - Returns the offset of a skb fragment
3367 * @frag: the paged fragment
3368 */
3369 static inline unsigned int skb_frag_off(const skb_frag_t *frag)
3370 {
3371 return frag->bv_offset;
3372 }
3373
3374 /**
3375 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta
3376 * @frag: skb fragment
3377 * @delta: value to add
3378 */
3379 static inline void skb_frag_off_add(skb_frag_t *frag, int delta)
3380 {
3381 frag->bv_offset += delta;
3382 }
3383
3384 /**
3385 * skb_frag_off_set() - Sets the offset of a skb fragment
3386 * @frag: skb fragment
3387 * @offset: offset of fragment
3388 */
3389 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset)
3390 {
3391 frag->bv_offset = offset;
3392 }
3393
3394 /**
3395 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment
3396 * @fragto: skb fragment where offset is set
3397 * @fragfrom: skb fragment offset is copied from
3398 */
3399 static inline void skb_frag_off_copy(skb_frag_t *fragto,
3400 const skb_frag_t *fragfrom)
3401 {
3402 fragto->bv_offset = fragfrom->bv_offset;
3403 }
3404
3405 /**
3406 * skb_frag_page - retrieve the page referred to by a paged fragment
3407 * @frag: the paged fragment
3408 *
3409 * Returns the &struct page associated with @frag.
3410 */
3411 static inline struct page *skb_frag_page(const skb_frag_t *frag)
3412 {
3413 return frag->bv_page;
3414 }
3415
3416 /**
3417 * __skb_frag_ref - take an addition reference on a paged fragment.
3418 * @frag: the paged fragment
3419 *
3420 * Takes an additional reference on the paged fragment @frag.
3421 */
3422 static inline void __skb_frag_ref(skb_frag_t *frag)
3423 {
3424 get_page(skb_frag_page(frag));
3425 }
3426
3427 /**
3428 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
3429 * @skb: the buffer
3430 * @f: the fragment offset.
3431 *
3432 * Takes an additional reference on the @f'th paged fragment of @skb.
3433 */
3434 static inline void skb_frag_ref(struct sk_buff *skb, int f)
3435 {
3436 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
3437 }
3438
3439 bool napi_pp_put_page(struct page *page, bool napi_safe);
3440
3441 static inline void
3442 napi_frag_unref(skb_frag_t *frag, bool recycle, bool napi_safe)
3443 {
3444 struct page *page = skb_frag_page(frag);
3445
3446 #ifdef CONFIG_PAGE_POOL
3447 if (recycle && napi_pp_put_page(page, napi_safe))
3448 return;
3449 #endif
3450 put_page(page);
3451 }
3452
3453 /**
3454 * __skb_frag_unref - release a reference on a paged fragment.
3455 * @frag: the paged fragment
3456 * @recycle: recycle the page if allocated via page_pool
3457 *
3458 * Releases a reference on the paged fragment @frag
3459 * or recycles the page via the page_pool API.
3460 */
3461 static inline void __skb_frag_unref(skb_frag_t *frag, bool recycle)
3462 {
3463 napi_frag_unref(frag, recycle, false);
3464 }
3465
3466 /**
3467 * skb_frag_unref - release a reference on a paged fragment of an skb.
3468 * @skb: the buffer
3469 * @f: the fragment offset
3470 *
3471 * Releases a reference on the @f'th paged fragment of @skb.
3472 */
3473 static inline void skb_frag_unref(struct sk_buff *skb, int f)
3474 {
3475 struct skb_shared_info *shinfo = skb_shinfo(skb);
3476
3477 if (!skb_zcopy_managed(skb))
3478 __skb_frag_unref(&shinfo->frags[f], skb->pp_recycle);
3479 }
3480
3481 /**
3482 * skb_frag_address - gets the address of the data contained in a paged fragment
3483 * @frag: the paged fragment buffer
3484 *
3485 * Returns the address of the data within @frag. The page must already
3486 * be mapped.
3487 */
3488 static inline void *skb_frag_address(const skb_frag_t *frag)
3489 {
3490 return page_address(skb_frag_page(frag)) + skb_frag_off(frag);
3491 }
3492
3493 /**
3494 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
3495 * @frag: the paged fragment buffer
3496 *
3497 * Returns the address of the data within @frag. Checks that the page
3498 * is mapped and returns %NULL otherwise.
3499 */
3500 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
3501 {
3502 void *ptr = page_address(skb_frag_page(frag));
3503 if (unlikely(!ptr))
3504 return NULL;
3505
3506 return ptr + skb_frag_off(frag);
3507 }
3508
3509 /**
3510 * skb_frag_page_copy() - sets the page in a fragment from another fragment
3511 * @fragto: skb fragment where page is set
3512 * @fragfrom: skb fragment page is copied from
3513 */
3514 static inline void skb_frag_page_copy(skb_frag_t *fragto,
3515 const skb_frag_t *fragfrom)
3516 {
3517 fragto->bv_page = fragfrom->bv_page;
3518 }
3519
3520 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
3521
3522 /**
3523 * skb_frag_dma_map - maps a paged fragment via the DMA API
3524 * @dev: the device to map the fragment to
3525 * @frag: the paged fragment to map
3526 * @offset: the offset within the fragment (starting at the
3527 * fragment's own offset)
3528 * @size: the number of bytes to map
3529 * @dir: the direction of the mapping (``PCI_DMA_*``)
3530 *
3531 * Maps the page associated with @frag to @device.
3532 */
3533 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3534 const skb_frag_t *frag,
3535 size_t offset, size_t size,
3536 enum dma_data_direction dir)
3537 {
3538 return dma_map_page(dev, skb_frag_page(frag),
3539 skb_frag_off(frag) + offset, size, dir);
3540 }
3541
3542 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3543 gfp_t gfp_mask)
3544 {
3545 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3546 }
3547
3548
3549 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3550 gfp_t gfp_mask)
3551 {
3552 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3553 }
3554
3555
3556 /**
3557 * skb_clone_writable - is the header of a clone writable
3558 * @skb: buffer to check
3559 * @len: length up to which to write
3560 *
3561 * Returns true if modifying the header part of the cloned buffer
3562 * does not requires the data to be copied.
3563 */
3564 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3565 {
3566 return !skb_header_cloned(skb) &&
3567 skb_headroom(skb) + len <= skb->hdr_len;
3568 }
3569
3570 static inline int skb_try_make_writable(struct sk_buff *skb,
3571 unsigned int write_len)
3572 {
3573 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3574 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3575 }
3576
3577 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3578 int cloned)
3579 {
3580 int delta = 0;
3581
3582 if (headroom > skb_headroom(skb))
3583 delta = headroom - skb_headroom(skb);
3584
3585 if (delta || cloned)
3586 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3587 GFP_ATOMIC);
3588 return 0;
3589 }
3590
3591 /**
3592 * skb_cow - copy header of skb when it is required
3593 * @skb: buffer to cow
3594 * @headroom: needed headroom
3595 *
3596 * If the skb passed lacks sufficient headroom or its data part
3597 * is shared, data is reallocated. If reallocation fails, an error
3598 * is returned and original skb is not changed.
3599 *
3600 * The result is skb with writable area skb->head...skb->tail
3601 * and at least @headroom of space at head.
3602 */
3603 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3604 {
3605 return __skb_cow(skb, headroom, skb_cloned(skb));
3606 }
3607
3608 /**
3609 * skb_cow_head - skb_cow but only making the head writable
3610 * @skb: buffer to cow
3611 * @headroom: needed headroom
3612 *
3613 * This function is identical to skb_cow except that we replace the
3614 * skb_cloned check by skb_header_cloned. It should be used when
3615 * you only need to push on some header and do not need to modify
3616 * the data.
3617 */
3618 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3619 {
3620 return __skb_cow(skb, headroom, skb_header_cloned(skb));
3621 }
3622
3623 /**
3624 * skb_padto - pad an skbuff up to a minimal size
3625 * @skb: buffer to pad
3626 * @len: minimal length
3627 *
3628 * Pads up a buffer to ensure the trailing bytes exist and are
3629 * blanked. If the buffer already contains sufficient data it
3630 * is untouched. Otherwise it is extended. Returns zero on
3631 * success. The skb is freed on error.
3632 */
3633 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3634 {
3635 unsigned int size = skb->len;
3636 if (likely(size >= len))
3637 return 0;
3638 return skb_pad(skb, len - size);
3639 }
3640
3641 /**
3642 * __skb_put_padto - increase size and pad an skbuff up to a minimal size
3643 * @skb: buffer to pad
3644 * @len: minimal length
3645 * @free_on_error: free buffer on error
3646 *
3647 * Pads up a buffer to ensure the trailing bytes exist and are
3648 * blanked. If the buffer already contains sufficient data it
3649 * is untouched. Otherwise it is extended. Returns zero on
3650 * success. The skb is freed on error if @free_on_error is true.
3651 */
3652 static inline int __must_check __skb_put_padto(struct sk_buff *skb,
3653 unsigned int len,
3654 bool free_on_error)
3655 {
3656 unsigned int size = skb->len;
3657
3658 if (unlikely(size < len)) {
3659 len -= size;
3660 if (__skb_pad(skb, len, free_on_error))
3661 return -ENOMEM;
3662 __skb_put(skb, len);
3663 }
3664 return 0;
3665 }
3666
3667 /**
3668 * skb_put_padto - increase size and pad an skbuff up to a minimal size
3669 * @skb: buffer to pad
3670 * @len: minimal length
3671 *
3672 * Pads up a buffer to ensure the trailing bytes exist and are
3673 * blanked. If the buffer already contains sufficient data it
3674 * is untouched. Otherwise it is extended. Returns zero on
3675 * success. The skb is freed on error.
3676 */
3677 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len)
3678 {
3679 return __skb_put_padto(skb, len, true);
3680 }
3681
3682 static inline int skb_add_data(struct sk_buff *skb,
3683 struct iov_iter *from, int copy)
3684 {
3685 const int off = skb->len;
3686
3687 if (skb->ip_summed == CHECKSUM_NONE) {
3688 __wsum csum = 0;
3689 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3690 &csum, from)) {
3691 skb->csum = csum_block_add(skb->csum, csum, off);
3692 return 0;
3693 }
3694 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3695 return 0;
3696
3697 __skb_trim(skb, off);
3698 return -EFAULT;
3699 }
3700
3701 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3702 const struct page *page, int off)
3703 {
3704 if (skb_zcopy(skb))
3705 return false;
3706 if (i) {
3707 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
3708
3709 return page == skb_frag_page(frag) &&
3710 off == skb_frag_off(frag) + skb_frag_size(frag);
3711 }
3712 return false;
3713 }
3714
3715 static inline int __skb_linearize(struct sk_buff *skb)
3716 {
3717 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3718 }
3719
3720 /**
3721 * skb_linearize - convert paged skb to linear one
3722 * @skb: buffer to linarize
3723 *
3724 * If there is no free memory -ENOMEM is returned, otherwise zero
3725 * is returned and the old skb data released.
3726 */
3727 static inline int skb_linearize(struct sk_buff *skb)
3728 {
3729 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3730 }
3731
3732 /**
3733 * skb_has_shared_frag - can any frag be overwritten
3734 * @skb: buffer to test
3735 *
3736 * Return true if the skb has at least one frag that might be modified
3737 * by an external entity (as in vmsplice()/sendfile())
3738 */
3739 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3740 {
3741 return skb_is_nonlinear(skb) &&
3742 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG;
3743 }
3744
3745 /**
3746 * skb_linearize_cow - make sure skb is linear and writable
3747 * @skb: buffer to process
3748 *
3749 * If there is no free memory -ENOMEM is returned, otherwise zero
3750 * is returned and the old skb data released.
3751 */
3752 static inline int skb_linearize_cow(struct sk_buff *skb)
3753 {
3754 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3755 __skb_linearize(skb) : 0;
3756 }
3757
3758 static __always_inline void
3759 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3760 unsigned int off)
3761 {
3762 if (skb->ip_summed == CHECKSUM_COMPLETE)
3763 skb->csum = csum_block_sub(skb->csum,
3764 csum_partial(start, len, 0), off);
3765 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3766 skb_checksum_start_offset(skb) < 0)
3767 skb->ip_summed = CHECKSUM_NONE;
3768 }
3769
3770 /**
3771 * skb_postpull_rcsum - update checksum for received skb after pull
3772 * @skb: buffer to update
3773 * @start: start of data before pull
3774 * @len: length of data pulled
3775 *
3776 * After doing a pull on a received packet, you need to call this to
3777 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3778 * CHECKSUM_NONE so that it can be recomputed from scratch.
3779 */
3780 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3781 const void *start, unsigned int len)
3782 {
3783 if (skb->ip_summed == CHECKSUM_COMPLETE)
3784 skb->csum = wsum_negate(csum_partial(start, len,
3785 wsum_negate(skb->csum)));
3786 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3787 skb_checksum_start_offset(skb) < 0)
3788 skb->ip_summed = CHECKSUM_NONE;
3789 }
3790
3791 static __always_inline void
3792 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3793 unsigned int off)
3794 {
3795 if (skb->ip_summed == CHECKSUM_COMPLETE)
3796 skb->csum = csum_block_add(skb->csum,
3797 csum_partial(start, len, 0), off);
3798 }
3799
3800 /**
3801 * skb_postpush_rcsum - update checksum for received skb after push
3802 * @skb: buffer to update
3803 * @start: start of data after push
3804 * @len: length of data pushed
3805 *
3806 * After doing a push on a received packet, you need to call this to
3807 * update the CHECKSUM_COMPLETE checksum.
3808 */
3809 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3810 const void *start, unsigned int len)
3811 {
3812 __skb_postpush_rcsum(skb, start, len, 0);
3813 }
3814
3815 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3816
3817 /**
3818 * skb_push_rcsum - push skb and update receive checksum
3819 * @skb: buffer to update
3820 * @len: length of data pulled
3821 *
3822 * This function performs an skb_push on the packet and updates
3823 * the CHECKSUM_COMPLETE checksum. It should be used on
3824 * receive path processing instead of skb_push unless you know
3825 * that the checksum difference is zero (e.g., a valid IP header)
3826 * or you are setting ip_summed to CHECKSUM_NONE.
3827 */
3828 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3829 {
3830 skb_push(skb, len);
3831 skb_postpush_rcsum(skb, skb->data, len);
3832 return skb->data;
3833 }
3834
3835 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3836 /**
3837 * pskb_trim_rcsum - trim received skb and update checksum
3838 * @skb: buffer to trim
3839 * @len: new length
3840 *
3841 * This is exactly the same as pskb_trim except that it ensures the
3842 * checksum of received packets are still valid after the operation.
3843 * It can change skb pointers.
3844 */
3845
3846 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3847 {
3848 if (likely(len >= skb->len))
3849 return 0;
3850 return pskb_trim_rcsum_slow(skb, len);
3851 }
3852
3853 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3854 {
3855 if (skb->ip_summed == CHECKSUM_COMPLETE)
3856 skb->ip_summed = CHECKSUM_NONE;
3857 __skb_trim(skb, len);
3858 return 0;
3859 }
3860
3861 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3862 {
3863 if (skb->ip_summed == CHECKSUM_COMPLETE)
3864 skb->ip_summed = CHECKSUM_NONE;
3865 return __skb_grow(skb, len);
3866 }
3867
3868 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3869 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3870 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3871 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3872 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3873
3874 #define skb_queue_walk(queue, skb) \
3875 for (skb = (queue)->next; \
3876 skb != (struct sk_buff *)(queue); \
3877 skb = skb->next)
3878
3879 #define skb_queue_walk_safe(queue, skb, tmp) \
3880 for (skb = (queue)->next, tmp = skb->next; \
3881 skb != (struct sk_buff *)(queue); \
3882 skb = tmp, tmp = skb->next)
3883
3884 #define skb_queue_walk_from(queue, skb) \
3885 for (; skb != (struct sk_buff *)(queue); \
3886 skb = skb->next)
3887
3888 #define skb_rbtree_walk(skb, root) \
3889 for (skb = skb_rb_first(root); skb != NULL; \
3890 skb = skb_rb_next(skb))
3891
3892 #define skb_rbtree_walk_from(skb) \
3893 for (; skb != NULL; \
3894 skb = skb_rb_next(skb))
3895
3896 #define skb_rbtree_walk_from_safe(skb, tmp) \
3897 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3898 skb = tmp)
3899
3900 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3901 for (tmp = skb->next; \
3902 skb != (struct sk_buff *)(queue); \
3903 skb = tmp, tmp = skb->next)
3904
3905 #define skb_queue_reverse_walk(queue, skb) \
3906 for (skb = (queue)->prev; \
3907 skb != (struct sk_buff *)(queue); \
3908 skb = skb->prev)
3909
3910 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3911 for (skb = (queue)->prev, tmp = skb->prev; \
3912 skb != (struct sk_buff *)(queue); \
3913 skb = tmp, tmp = skb->prev)
3914
3915 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3916 for (tmp = skb->prev; \
3917 skb != (struct sk_buff *)(queue); \
3918 skb = tmp, tmp = skb->prev)
3919
3920 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3921 {
3922 return skb_shinfo(skb)->frag_list != NULL;
3923 }
3924
3925 static inline void skb_frag_list_init(struct sk_buff *skb)
3926 {
3927 skb_shinfo(skb)->frag_list = NULL;
3928 }
3929
3930 #define skb_walk_frags(skb, iter) \
3931 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3932
3933
3934 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue,
3935 int *err, long *timeo_p,
3936 const struct sk_buff *skb);
3937 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3938 struct sk_buff_head *queue,
3939 unsigned int flags,
3940 int *off, int *err,
3941 struct sk_buff **last);
3942 struct sk_buff *__skb_try_recv_datagram(struct sock *sk,
3943 struct sk_buff_head *queue,
3944 unsigned int flags, int *off, int *err,
3945 struct sk_buff **last);
3946 struct sk_buff *__skb_recv_datagram(struct sock *sk,
3947 struct sk_buff_head *sk_queue,
3948 unsigned int flags, int *off, int *err);
3949 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err);
3950 __poll_t datagram_poll(struct file *file, struct socket *sock,
3951 struct poll_table_struct *wait);
3952 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3953 struct iov_iter *to, int size);
3954 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3955 struct msghdr *msg, int size)
3956 {
3957 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3958 }
3959 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3960 struct msghdr *msg);
3961 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
3962 struct iov_iter *to, int len,
3963 struct ahash_request *hash);
3964 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3965 struct iov_iter *from, int len);
3966 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3967 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3968 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3969 static inline void skb_free_datagram_locked(struct sock *sk,
3970 struct sk_buff *skb)
3971 {
3972 __skb_free_datagram_locked(sk, skb, 0);
3973 }
3974 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3975 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3976 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3977 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3978 int len);
3979 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3980 struct pipe_inode_info *pipe, unsigned int len,
3981 unsigned int flags);
3982 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3983 int len);
3984 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
3985 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3986 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3987 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3988 int len, int hlen);
3989 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3990 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3991 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3992 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3993 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features,
3994 unsigned int offset);
3995 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3996 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len);
3997 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3998 int skb_vlan_pop(struct sk_buff *skb);
3999 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
4000 int skb_eth_pop(struct sk_buff *skb);
4001 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
4002 const unsigned char *src);
4003 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
4004 int mac_len, bool ethernet);
4005 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
4006 bool ethernet);
4007 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
4008 int skb_mpls_dec_ttl(struct sk_buff *skb);
4009 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
4010 gfp_t gfp);
4011
4012 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
4013 {
4014 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
4015 }
4016
4017 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
4018 {
4019 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
4020 }
4021
4022 struct skb_checksum_ops {
4023 __wsum (*update)(const void *mem, int len, __wsum wsum);
4024 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
4025 };
4026
4027 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
4028
4029 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
4030 __wsum csum, const struct skb_checksum_ops *ops);
4031 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
4032 __wsum csum);
4033
4034 static inline void * __must_check
4035 __skb_header_pointer(const struct sk_buff *skb, int offset, int len,
4036 const void *data, int hlen, void *buffer)
4037 {
4038 if (likely(hlen - offset >= len))
4039 return (void *)data + offset;
4040
4041 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0))
4042 return NULL;
4043
4044 return buffer;
4045 }
4046
4047 static inline void * __must_check
4048 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
4049 {
4050 return __skb_header_pointer(skb, offset, len, skb->data,
4051 skb_headlen(skb), buffer);
4052 }
4053
4054 static inline void * __must_check
4055 skb_pointer_if_linear(const struct sk_buff *skb, int offset, int len)
4056 {
4057 if (likely(skb_headlen(skb) - offset >= len))
4058 return skb->data + offset;
4059 return NULL;
4060 }
4061
4062 /**
4063 * skb_needs_linearize - check if we need to linearize a given skb
4064 * depending on the given device features.
4065 * @skb: socket buffer to check
4066 * @features: net device features
4067 *
4068 * Returns true if either:
4069 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
4070 * 2. skb is fragmented and the device does not support SG.
4071 */
4072 static inline bool skb_needs_linearize(struct sk_buff *skb,
4073 netdev_features_t features)
4074 {
4075 return skb_is_nonlinear(skb) &&
4076 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
4077 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
4078 }
4079
4080 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
4081 void *to,
4082 const unsigned int len)
4083 {
4084 memcpy(to, skb->data, len);
4085 }
4086
4087 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
4088 const int offset, void *to,
4089 const unsigned int len)
4090 {
4091 memcpy(to, skb->data + offset, len);
4092 }
4093
4094 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
4095 const void *from,
4096 const unsigned int len)
4097 {
4098 memcpy(skb->data, from, len);
4099 }
4100
4101 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
4102 const int offset,
4103 const void *from,
4104 const unsigned int len)
4105 {
4106 memcpy(skb->data + offset, from, len);
4107 }
4108
4109 void skb_init(void);
4110
4111 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
4112 {
4113 return skb->tstamp;
4114 }
4115
4116 /**
4117 * skb_get_timestamp - get timestamp from a skb
4118 * @skb: skb to get stamp from
4119 * @stamp: pointer to struct __kernel_old_timeval to store stamp in
4120 *
4121 * Timestamps are stored in the skb as offsets to a base timestamp.
4122 * This function converts the offset back to a struct timeval and stores
4123 * it in stamp.
4124 */
4125 static inline void skb_get_timestamp(const struct sk_buff *skb,
4126 struct __kernel_old_timeval *stamp)
4127 {
4128 *stamp = ns_to_kernel_old_timeval(skb->tstamp);
4129 }
4130
4131 static inline void skb_get_new_timestamp(const struct sk_buff *skb,
4132 struct __kernel_sock_timeval *stamp)
4133 {
4134 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4135
4136 stamp->tv_sec = ts.tv_sec;
4137 stamp->tv_usec = ts.tv_nsec / 1000;
4138 }
4139
4140 static inline void skb_get_timestampns(const struct sk_buff *skb,
4141 struct __kernel_old_timespec *stamp)
4142 {
4143 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4144
4145 stamp->tv_sec = ts.tv_sec;
4146 stamp->tv_nsec = ts.tv_nsec;
4147 }
4148
4149 static inline void skb_get_new_timestampns(const struct sk_buff *skb,
4150 struct __kernel_timespec *stamp)
4151 {
4152 struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
4153
4154 stamp->tv_sec = ts.tv_sec;
4155 stamp->tv_nsec = ts.tv_nsec;
4156 }
4157
4158 static inline void __net_timestamp(struct sk_buff *skb)
4159 {
4160 skb->tstamp = ktime_get_real();
4161 skb->mono_delivery_time = 0;
4162 }
4163
4164 static inline ktime_t net_timedelta(ktime_t t)
4165 {
4166 return ktime_sub(ktime_get_real(), t);
4167 }
4168
4169 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt,
4170 bool mono)
4171 {
4172 skb->tstamp = kt;
4173 skb->mono_delivery_time = kt && mono;
4174 }
4175
4176 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key);
4177
4178 /* It is used in the ingress path to clear the delivery_time.
4179 * If needed, set the skb->tstamp to the (rcv) timestamp.
4180 */
4181 static inline void skb_clear_delivery_time(struct sk_buff *skb)
4182 {
4183 if (skb->mono_delivery_time) {
4184 skb->mono_delivery_time = 0;
4185 if (static_branch_unlikely(&netstamp_needed_key))
4186 skb->tstamp = ktime_get_real();
4187 else
4188 skb->tstamp = 0;
4189 }
4190 }
4191
4192 static inline void skb_clear_tstamp(struct sk_buff *skb)
4193 {
4194 if (skb->mono_delivery_time)
4195 return;
4196
4197 skb->tstamp = 0;
4198 }
4199
4200 static inline ktime_t skb_tstamp(const struct sk_buff *skb)
4201 {
4202 if (skb->mono_delivery_time)
4203 return 0;
4204
4205 return skb->tstamp;
4206 }
4207
4208 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond)
4209 {
4210 if (!skb->mono_delivery_time && skb->tstamp)
4211 return skb->tstamp;
4212
4213 if (static_branch_unlikely(&netstamp_needed_key) || cond)
4214 return ktime_get_real();
4215
4216 return 0;
4217 }
4218
4219 static inline u8 skb_metadata_len(const struct sk_buff *skb)
4220 {
4221 return skb_shinfo(skb)->meta_len;
4222 }
4223
4224 static inline void *skb_metadata_end(const struct sk_buff *skb)
4225 {
4226 return skb_mac_header(skb);
4227 }
4228
4229 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
4230 const struct sk_buff *skb_b,
4231 u8 meta_len)
4232 {
4233 const void *a = skb_metadata_end(skb_a);
4234 const void *b = skb_metadata_end(skb_b);
4235 /* Using more efficient varaiant than plain call to memcmp(). */
4236 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
4237 u64 diffs = 0;
4238
4239 switch (meta_len) {
4240 #define __it(x, op) (x -= sizeof(u##op))
4241 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
4242 case 32: diffs |= __it_diff(a, b, 64);
4243 fallthrough;
4244 case 24: diffs |= __it_diff(a, b, 64);
4245 fallthrough;
4246 case 16: diffs |= __it_diff(a, b, 64);
4247 fallthrough;
4248 case 8: diffs |= __it_diff(a, b, 64);
4249 break;
4250 case 28: diffs |= __it_diff(a, b, 64);
4251 fallthrough;
4252 case 20: diffs |= __it_diff(a, b, 64);
4253 fallthrough;
4254 case 12: diffs |= __it_diff(a, b, 64);
4255 fallthrough;
4256 case 4: diffs |= __it_diff(a, b, 32);
4257 break;
4258 }
4259 return diffs;
4260 #else
4261 return memcmp(a - meta_len, b - meta_len, meta_len);
4262 #endif
4263 }
4264
4265 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
4266 const struct sk_buff *skb_b)
4267 {
4268 u8 len_a = skb_metadata_len(skb_a);
4269 u8 len_b = skb_metadata_len(skb_b);
4270
4271 if (!(len_a | len_b))
4272 return false;
4273
4274 return len_a != len_b ?
4275 true : __skb_metadata_differs(skb_a, skb_b, len_a);
4276 }
4277
4278 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
4279 {
4280 skb_shinfo(skb)->meta_len = meta_len;
4281 }
4282
4283 static inline void skb_metadata_clear(struct sk_buff *skb)
4284 {
4285 skb_metadata_set(skb, 0);
4286 }
4287
4288 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
4289
4290 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
4291
4292 void skb_clone_tx_timestamp(struct sk_buff *skb);
4293 bool skb_defer_rx_timestamp(struct sk_buff *skb);
4294
4295 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
4296
4297 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
4298 {
4299 }
4300
4301 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
4302 {
4303 return false;
4304 }
4305
4306 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
4307
4308 /**
4309 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
4310 *
4311 * PHY drivers may accept clones of transmitted packets for
4312 * timestamping via their phy_driver.txtstamp method. These drivers
4313 * must call this function to return the skb back to the stack with a
4314 * timestamp.
4315 *
4316 * @skb: clone of the original outgoing packet
4317 * @hwtstamps: hardware time stamps
4318 *
4319 */
4320 void skb_complete_tx_timestamp(struct sk_buff *skb,
4321 struct skb_shared_hwtstamps *hwtstamps);
4322
4323 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb,
4324 struct skb_shared_hwtstamps *hwtstamps,
4325 struct sock *sk, int tstype);
4326
4327 /**
4328 * skb_tstamp_tx - queue clone of skb with send time stamps
4329 * @orig_skb: the original outgoing packet
4330 * @hwtstamps: hardware time stamps, may be NULL if not available
4331 *
4332 * If the skb has a socket associated, then this function clones the
4333 * skb (thus sharing the actual data and optional structures), stores
4334 * the optional hardware time stamping information (if non NULL) or
4335 * generates a software time stamp (otherwise), then queues the clone
4336 * to the error queue of the socket. Errors are silently ignored.
4337 */
4338 void skb_tstamp_tx(struct sk_buff *orig_skb,
4339 struct skb_shared_hwtstamps *hwtstamps);
4340
4341 /**
4342 * skb_tx_timestamp() - Driver hook for transmit timestamping
4343 *
4344 * Ethernet MAC Drivers should call this function in their hard_xmit()
4345 * function immediately before giving the sk_buff to the MAC hardware.
4346 *
4347 * Specifically, one should make absolutely sure that this function is
4348 * called before TX completion of this packet can trigger. Otherwise
4349 * the packet could potentially already be freed.
4350 *
4351 * @skb: A socket buffer.
4352 */
4353 static inline void skb_tx_timestamp(struct sk_buff *skb)
4354 {
4355 skb_clone_tx_timestamp(skb);
4356 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
4357 skb_tstamp_tx(skb, NULL);
4358 }
4359
4360 /**
4361 * skb_complete_wifi_ack - deliver skb with wifi status
4362 *
4363 * @skb: the original outgoing packet
4364 * @acked: ack status
4365 *
4366 */
4367 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
4368
4369 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
4370 __sum16 __skb_checksum_complete(struct sk_buff *skb);
4371
4372 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
4373 {
4374 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
4375 skb->csum_valid ||
4376 (skb->ip_summed == CHECKSUM_PARTIAL &&
4377 skb_checksum_start_offset(skb) >= 0));
4378 }
4379
4380 /**
4381 * skb_checksum_complete - Calculate checksum of an entire packet
4382 * @skb: packet to process
4383 *
4384 * This function calculates the checksum over the entire packet plus
4385 * the value of skb->csum. The latter can be used to supply the
4386 * checksum of a pseudo header as used by TCP/UDP. It returns the
4387 * checksum.
4388 *
4389 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
4390 * this function can be used to verify that checksum on received
4391 * packets. In that case the function should return zero if the
4392 * checksum is correct. In particular, this function will return zero
4393 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
4394 * hardware has already verified the correctness of the checksum.
4395 */
4396 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
4397 {
4398 return skb_csum_unnecessary(skb) ?
4399 0 : __skb_checksum_complete(skb);
4400 }
4401
4402 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
4403 {
4404 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4405 if (skb->csum_level == 0)
4406 skb->ip_summed = CHECKSUM_NONE;
4407 else
4408 skb->csum_level--;
4409 }
4410 }
4411
4412 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
4413 {
4414 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4415 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
4416 skb->csum_level++;
4417 } else if (skb->ip_summed == CHECKSUM_NONE) {
4418 skb->ip_summed = CHECKSUM_UNNECESSARY;
4419 skb->csum_level = 0;
4420 }
4421 }
4422
4423 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb)
4424 {
4425 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4426 skb->ip_summed = CHECKSUM_NONE;
4427 skb->csum_level = 0;
4428 }
4429 }
4430
4431 /* Check if we need to perform checksum complete validation.
4432 *
4433 * Returns true if checksum complete is needed, false otherwise
4434 * (either checksum is unnecessary or zero checksum is allowed).
4435 */
4436 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
4437 bool zero_okay,
4438 __sum16 check)
4439 {
4440 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
4441 skb->csum_valid = 1;
4442 __skb_decr_checksum_unnecessary(skb);
4443 return false;
4444 }
4445
4446 return true;
4447 }
4448
4449 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly
4450 * in checksum_init.
4451 */
4452 #define CHECKSUM_BREAK 76
4453
4454 /* Unset checksum-complete
4455 *
4456 * Unset checksum complete can be done when packet is being modified
4457 * (uncompressed for instance) and checksum-complete value is
4458 * invalidated.
4459 */
4460 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
4461 {
4462 if (skb->ip_summed == CHECKSUM_COMPLETE)
4463 skb->ip_summed = CHECKSUM_NONE;
4464 }
4465
4466 /* Validate (init) checksum based on checksum complete.
4467 *
4468 * Return values:
4469 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
4470 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
4471 * checksum is stored in skb->csum for use in __skb_checksum_complete
4472 * non-zero: value of invalid checksum
4473 *
4474 */
4475 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
4476 bool complete,
4477 __wsum psum)
4478 {
4479 if (skb->ip_summed == CHECKSUM_COMPLETE) {
4480 if (!csum_fold(csum_add(psum, skb->csum))) {
4481 skb->csum_valid = 1;
4482 return 0;
4483 }
4484 }
4485
4486 skb->csum = psum;
4487
4488 if (complete || skb->len <= CHECKSUM_BREAK) {
4489 __sum16 csum;
4490
4491 csum = __skb_checksum_complete(skb);
4492 skb->csum_valid = !csum;
4493 return csum;
4494 }
4495
4496 return 0;
4497 }
4498
4499 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
4500 {
4501 return 0;
4502 }
4503
4504 /* Perform checksum validate (init). Note that this is a macro since we only
4505 * want to calculate the pseudo header which is an input function if necessary.
4506 * First we try to validate without any computation (checksum unnecessary) and
4507 * then calculate based on checksum complete calling the function to compute
4508 * pseudo header.
4509 *
4510 * Return values:
4511 * 0: checksum is validated or try to in skb_checksum_complete
4512 * non-zero: value of invalid checksum
4513 */
4514 #define __skb_checksum_validate(skb, proto, complete, \
4515 zero_okay, check, compute_pseudo) \
4516 ({ \
4517 __sum16 __ret = 0; \
4518 skb->csum_valid = 0; \
4519 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
4520 __ret = __skb_checksum_validate_complete(skb, \
4521 complete, compute_pseudo(skb, proto)); \
4522 __ret; \
4523 })
4524
4525 #define skb_checksum_init(skb, proto, compute_pseudo) \
4526 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
4527
4528 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
4529 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
4530
4531 #define skb_checksum_validate(skb, proto, compute_pseudo) \
4532 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
4533
4534 #define skb_checksum_validate_zero_check(skb, proto, check, \
4535 compute_pseudo) \
4536 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
4537
4538 #define skb_checksum_simple_validate(skb) \
4539 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
4540
4541 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
4542 {
4543 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
4544 }
4545
4546 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
4547 {
4548 skb->csum = ~pseudo;
4549 skb->ip_summed = CHECKSUM_COMPLETE;
4550 }
4551
4552 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \
4553 do { \
4554 if (__skb_checksum_convert_check(skb)) \
4555 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
4556 } while (0)
4557
4558 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
4559 u16 start, u16 offset)
4560 {
4561 skb->ip_summed = CHECKSUM_PARTIAL;
4562 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
4563 skb->csum_offset = offset - start;
4564 }
4565
4566 /* Update skbuf and packet to reflect the remote checksum offload operation.
4567 * When called, ptr indicates the starting point for skb->csum when
4568 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
4569 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
4570 */
4571 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
4572 int start, int offset, bool nopartial)
4573 {
4574 __wsum delta;
4575
4576 if (!nopartial) {
4577 skb_remcsum_adjust_partial(skb, ptr, start, offset);
4578 return;
4579 }
4580
4581 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
4582 __skb_checksum_complete(skb);
4583 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
4584 }
4585
4586 delta = remcsum_adjust(ptr, skb->csum, start, offset);
4587
4588 /* Adjust skb->csum since we changed the packet */
4589 skb->csum = csum_add(skb->csum, delta);
4590 }
4591
4592 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
4593 {
4594 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4595 return (void *)(skb->_nfct & NFCT_PTRMASK);
4596 #else
4597 return NULL;
4598 #endif
4599 }
4600
4601 static inline unsigned long skb_get_nfct(const struct sk_buff *skb)
4602 {
4603 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4604 return skb->_nfct;
4605 #else
4606 return 0UL;
4607 #endif
4608 }
4609
4610 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct)
4611 {
4612 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
4613 skb->slow_gro |= !!nfct;
4614 skb->_nfct = nfct;
4615 #endif
4616 }
4617
4618 #ifdef CONFIG_SKB_EXTENSIONS
4619 enum skb_ext_id {
4620 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4621 SKB_EXT_BRIDGE_NF,
4622 #endif
4623 #ifdef CONFIG_XFRM
4624 SKB_EXT_SEC_PATH,
4625 #endif
4626 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4627 TC_SKB_EXT,
4628 #endif
4629 #if IS_ENABLED(CONFIG_MPTCP)
4630 SKB_EXT_MPTCP,
4631 #endif
4632 #if IS_ENABLED(CONFIG_MCTP_FLOWS)
4633 SKB_EXT_MCTP,
4634 #endif
4635 SKB_EXT_NUM, /* must be last */
4636 };
4637
4638 /**
4639 * struct skb_ext - sk_buff extensions
4640 * @refcnt: 1 on allocation, deallocated on 0
4641 * @offset: offset to add to @data to obtain extension address
4642 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4643 * @data: start of extension data, variable sized
4644 *
4645 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4646 * to use 'u8' types while allowing up to 2kb worth of extension data.
4647 */
4648 struct skb_ext {
4649 refcount_t refcnt;
4650 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4651 u8 chunks; /* same */
4652 char data[] __aligned(8);
4653 };
4654
4655 struct skb_ext *__skb_ext_alloc(gfp_t flags);
4656 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
4657 struct skb_ext *ext);
4658 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4659 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4660 void __skb_ext_put(struct skb_ext *ext);
4661
4662 static inline void skb_ext_put(struct sk_buff *skb)
4663 {
4664 if (skb->active_extensions)
4665 __skb_ext_put(skb->extensions);
4666 }
4667
4668 static inline void __skb_ext_copy(struct sk_buff *dst,
4669 const struct sk_buff *src)
4670 {
4671 dst->active_extensions = src->active_extensions;
4672
4673 if (src->active_extensions) {
4674 struct skb_ext *ext = src->extensions;
4675
4676 refcount_inc(&ext->refcnt);
4677 dst->extensions = ext;
4678 }
4679 }
4680
4681 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4682 {
4683 skb_ext_put(dst);
4684 __skb_ext_copy(dst, src);
4685 }
4686
4687 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4688 {
4689 return !!ext->offset[i];
4690 }
4691
4692 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4693 {
4694 return skb->active_extensions & (1 << id);
4695 }
4696
4697 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4698 {
4699 if (skb_ext_exist(skb, id))
4700 __skb_ext_del(skb, id);
4701 }
4702
4703 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4704 {
4705 if (skb_ext_exist(skb, id)) {
4706 struct skb_ext *ext = skb->extensions;
4707
4708 return (void *)ext + (ext->offset[id] << 3);
4709 }
4710
4711 return NULL;
4712 }
4713
4714 static inline void skb_ext_reset(struct sk_buff *skb)
4715 {
4716 if (unlikely(skb->active_extensions)) {
4717 __skb_ext_put(skb->extensions);
4718 skb->active_extensions = 0;
4719 }
4720 }
4721
4722 static inline bool skb_has_extensions(struct sk_buff *skb)
4723 {
4724 return unlikely(skb->active_extensions);
4725 }
4726 #else
4727 static inline void skb_ext_put(struct sk_buff *skb) {}
4728 static inline void skb_ext_reset(struct sk_buff *skb) {}
4729 static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4730 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4731 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4732 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; }
4733 #endif /* CONFIG_SKB_EXTENSIONS */
4734
4735 static inline void nf_reset_ct(struct sk_buff *skb)
4736 {
4737 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4738 nf_conntrack_put(skb_nfct(skb));
4739 skb->_nfct = 0;
4740 #endif
4741 }
4742
4743 static inline void nf_reset_trace(struct sk_buff *skb)
4744 {
4745 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
4746 skb->nf_trace = 0;
4747 #endif
4748 }
4749
4750 static inline void ipvs_reset(struct sk_buff *skb)
4751 {
4752 #if IS_ENABLED(CONFIG_IP_VS)
4753 skb->ipvs_property = 0;
4754 #endif
4755 }
4756
4757 /* Note: This doesn't put any conntrack info in dst. */
4758 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4759 bool copy)
4760 {
4761 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4762 dst->_nfct = src->_nfct;
4763 nf_conntrack_get(skb_nfct(src));
4764 #endif
4765 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES)
4766 if (copy)
4767 dst->nf_trace = src->nf_trace;
4768 #endif
4769 }
4770
4771 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4772 {
4773 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4774 nf_conntrack_put(skb_nfct(dst));
4775 #endif
4776 dst->slow_gro = src->slow_gro;
4777 __nf_copy(dst, src, true);
4778 }
4779
4780 #ifdef CONFIG_NETWORK_SECMARK
4781 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4782 {
4783 to->secmark = from->secmark;
4784 }
4785
4786 static inline void skb_init_secmark(struct sk_buff *skb)
4787 {
4788 skb->secmark = 0;
4789 }
4790 #else
4791 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4792 { }
4793
4794 static inline void skb_init_secmark(struct sk_buff *skb)
4795 { }
4796 #endif
4797
4798 static inline int secpath_exists(const struct sk_buff *skb)
4799 {
4800 #ifdef CONFIG_XFRM
4801 return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4802 #else
4803 return 0;
4804 #endif
4805 }
4806
4807 static inline bool skb_irq_freeable(const struct sk_buff *skb)
4808 {
4809 return !skb->destructor &&
4810 !secpath_exists(skb) &&
4811 !skb_nfct(skb) &&
4812 !skb->_skb_refdst &&
4813 !skb_has_frag_list(skb);
4814 }
4815
4816 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4817 {
4818 skb->queue_mapping = queue_mapping;
4819 }
4820
4821 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4822 {
4823 return skb->queue_mapping;
4824 }
4825
4826 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4827 {
4828 to->queue_mapping = from->queue_mapping;
4829 }
4830
4831 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4832 {
4833 skb->queue_mapping = rx_queue + 1;
4834 }
4835
4836 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4837 {
4838 return skb->queue_mapping - 1;
4839 }
4840
4841 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4842 {
4843 return skb->queue_mapping != 0;
4844 }
4845
4846 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4847 {
4848 skb->dst_pending_confirm = val;
4849 }
4850
4851 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4852 {
4853 return skb->dst_pending_confirm != 0;
4854 }
4855
4856 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4857 {
4858 #ifdef CONFIG_XFRM
4859 return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4860 #else
4861 return NULL;
4862 #endif
4863 }
4864
4865 static inline bool skb_is_gso(const struct sk_buff *skb)
4866 {
4867 return skb_shinfo(skb)->gso_size;
4868 }
4869
4870 /* Note: Should be called only if skb_is_gso(skb) is true */
4871 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4872 {
4873 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4874 }
4875
4876 /* Note: Should be called only if skb_is_gso(skb) is true */
4877 static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4878 {
4879 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4880 }
4881
4882 /* Note: Should be called only if skb_is_gso(skb) is true */
4883 static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4884 {
4885 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4886 }
4887
4888 static inline void skb_gso_reset(struct sk_buff *skb)
4889 {
4890 skb_shinfo(skb)->gso_size = 0;
4891 skb_shinfo(skb)->gso_segs = 0;
4892 skb_shinfo(skb)->gso_type = 0;
4893 }
4894
4895 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4896 u16 increment)
4897 {
4898 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4899 return;
4900 shinfo->gso_size += increment;
4901 }
4902
4903 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4904 u16 decrement)
4905 {
4906 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4907 return;
4908 shinfo->gso_size -= decrement;
4909 }
4910
4911 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4912
4913 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4914 {
4915 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4916 * wanted then gso_type will be set. */
4917 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4918
4919 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4920 unlikely(shinfo->gso_type == 0)) {
4921 __skb_warn_lro_forwarding(skb);
4922 return true;
4923 }
4924 return false;
4925 }
4926
4927 static inline void skb_forward_csum(struct sk_buff *skb)
4928 {
4929 /* Unfortunately we don't support this one. Any brave souls? */
4930 if (skb->ip_summed == CHECKSUM_COMPLETE)
4931 skb->ip_summed = CHECKSUM_NONE;
4932 }
4933
4934 /**
4935 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4936 * @skb: skb to check
4937 *
4938 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4939 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4940 * use this helper, to document places where we make this assertion.
4941 */
4942 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4943 {
4944 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE);
4945 }
4946
4947 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4948
4949 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4950 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4951 unsigned int transport_len,
4952 __sum16(*skb_chkf)(struct sk_buff *skb));
4953
4954 /**
4955 * skb_head_is_locked - Determine if the skb->head is locked down
4956 * @skb: skb to check
4957 *
4958 * The head on skbs build around a head frag can be removed if they are
4959 * not cloned. This function returns true if the skb head is locked down
4960 * due to either being allocated via kmalloc, or by being a clone with
4961 * multiple references to the head.
4962 */
4963 static inline bool skb_head_is_locked(const struct sk_buff *skb)
4964 {
4965 return !skb->head_frag || skb_cloned(skb);
4966 }
4967
4968 /* Local Checksum Offload.
4969 * Compute outer checksum based on the assumption that the
4970 * inner checksum will be offloaded later.
4971 * See Documentation/networking/checksum-offloads.rst for
4972 * explanation of how this works.
4973 * Fill in outer checksum adjustment (e.g. with sum of outer
4974 * pseudo-header) before calling.
4975 * Also ensure that inner checksum is in linear data area.
4976 */
4977 static inline __wsum lco_csum(struct sk_buff *skb)
4978 {
4979 unsigned char *csum_start = skb_checksum_start(skb);
4980 unsigned char *l4_hdr = skb_transport_header(skb);
4981 __wsum partial;
4982
4983 /* Start with complement of inner checksum adjustment */
4984 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4985 skb->csum_offset));
4986
4987 /* Add in checksum of our headers (incl. outer checksum
4988 * adjustment filled in by caller) and return result.
4989 */
4990 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
4991 }
4992
4993 static inline bool skb_is_redirected(const struct sk_buff *skb)
4994 {
4995 return skb->redirected;
4996 }
4997
4998 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress)
4999 {
5000 skb->redirected = 1;
5001 #ifdef CONFIG_NET_REDIRECT
5002 skb->from_ingress = from_ingress;
5003 if (skb->from_ingress)
5004 skb_clear_tstamp(skb);
5005 #endif
5006 }
5007
5008 static inline void skb_reset_redirect(struct sk_buff *skb)
5009 {
5010 skb->redirected = 0;
5011 }
5012
5013 static inline void skb_set_redirected_noclear(struct sk_buff *skb,
5014 bool from_ingress)
5015 {
5016 skb->redirected = 1;
5017 #ifdef CONFIG_NET_REDIRECT
5018 skb->from_ingress = from_ingress;
5019 #endif
5020 }
5021
5022 static inline bool skb_csum_is_sctp(struct sk_buff *skb)
5023 {
5024 #if IS_ENABLED(CONFIG_IP_SCTP)
5025 return skb->csum_not_inet;
5026 #else
5027 return 0;
5028 #endif
5029 }
5030
5031 static inline void skb_reset_csum_not_inet(struct sk_buff *skb)
5032 {
5033 skb->ip_summed = CHECKSUM_NONE;
5034 #if IS_ENABLED(CONFIG_IP_SCTP)
5035 skb->csum_not_inet = 0;
5036 #endif
5037 }
5038
5039 static inline void skb_set_kcov_handle(struct sk_buff *skb,
5040 const u64 kcov_handle)
5041 {
5042 #ifdef CONFIG_KCOV
5043 skb->kcov_handle = kcov_handle;
5044 #endif
5045 }
5046
5047 static inline u64 skb_get_kcov_handle(struct sk_buff *skb)
5048 {
5049 #ifdef CONFIG_KCOV
5050 return skb->kcov_handle;
5051 #else
5052 return 0;
5053 #endif
5054 }
5055
5056 static inline void skb_mark_for_recycle(struct sk_buff *skb)
5057 {
5058 #ifdef CONFIG_PAGE_POOL
5059 skb->pp_recycle = 1;
5060 #endif
5061 }
5062
5063 ssize_t skb_splice_from_iter(struct sk_buff *skb, struct iov_iter *iter,
5064 ssize_t maxsize, gfp_t gfp);
5065
5066 #endif /* __KERNEL__ */
5067 #endif /* _LINUX_SKBUFF_H */