master
1/* $NetBSD: tcp_vtw.h,v 1.10 2022/12/11 08:09:20 mlelstv Exp $ */
2/*
3 * Copyright (c) 2011 The NetBSD Foundation, Inc.
4 * All rights reserved.
5 *
6 * This code is derived from software contributed to The NetBSD Foundation
7 * by Coyote Point Systems, Inc.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
19 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
20 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
21 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
22 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
23 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
24 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
25 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
26 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
27 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28 * POSSIBILITY OF SUCH DAMAGE.
29 */
30
31/*
32 * Vestigial time-wait.
33 *
34 * This implementation uses cache-efficient techniques, which will
35 * appear somewhat peculiar. The main philosophy is to optimise the
36 * amount of information available within a cache line. Cache miss is
37 * expensive. So we employ ad-hoc techniques to pull a series of
38 * linked-list follows into a cache line. One cache line, multiple
39 * linked-list equivalents.
40 *
41 * One such ad-hoc technique is fat pointers. Additional degrees of
42 * ad-hoqueness result from having to hand tune it for pointer size
43 * and for cache line size.
44 *
45 * The 'fat pointer' approach aggregates, for x86_32, 15 linked-list
46 * data structures into one cache line. The additional 32 bits in the
47 * cache line are used for linking fat pointers, and for
48 * allocation/bookkeeping.
49 *
50 * The 15 32-bit tags encode the pointers to the linked list elements,
51 * and also encode the results of a search comparison.
52 *
53 * First, some more assumptions/restrictions.
54 *
55 * All the fat pointers are from a contiguous allocation arena. Thus,
56 * we can refer to them by offset from a base, not as full pointers.
57 *
58 * All the linked list data elements are also from a contiguous
59 * allocation arena, again so that we can refer to them as offset from
60 * a base.
61 *
62 * In order to add a data element to a fat pointer, a key value is
63 * computed, based on unique data within the data element. It is the
64 * linear searching of the linked lists of these elements based on
65 * these unique data that are being optimised here.
66 *
67 * Lets call the function that computes the key k(e), where e is the
68 * data element. In this example, k(e) returns 32-bits.
69 *
70 * Consider a set E (say of order 15) of data elements. Let K be
71 * the set of the k(e) for e in E.
72 *
73 * Let O be the set of the offsets from the base of the data elements in E.
74 *
75 * For each x in K, for each matching o in O, let t be x ^ o. These
76 * are the tags. (More or less).
77 *
78 * In order to search all the data elements in E, we compute the
79 * search key, and one at a time, XOR the key into the tags. If any
80 * result is a valid data element index, we have a possible match. If
81 * not, there is no match.
82 *
83 * The no-match cases mean we do not have to de-reference the pointer
84 * to the data element in question. We save cache miss penalty and
85 * cache load decreases. Only in the case of a valid looking data
86 * element index, do we have to look closer.
87 *
88 * Thus, in the absence of false positives, 15 data elements can be
89 * searched with one cache line fill, as opposed to 15 cache line
90 * fills for the usual implementation.
91 *
92 * The vestigial time waits (vtw_t), the data elements in the above, are
93 * searched by faddr, fport, laddr, lport. The key is a function of
94 * these values.
95 *
96 * We hash these keys into the traditional hash chains to reduce the
97 * search time, and use fat pointers to reduce the cache impacts of
98 * searching.
99 *
100 * The vtw_t are, per requirement, in a contiguous chunk. Allocation
101 * is done with a clock hand, and all vtw_t within one allocation
102 * domain have the same lifetime, so they will always be sorted by
103 * age.
104 *
105 * A vtw_t will be allocated, timestamped, and have a fixed future
106 * expiration. It will be added to a hash bucket implemented with fat
107 * pointers, which means that a cache line will be allocated in the
108 * hash bucket, placed at the head (more recent in time) and the vtw_t
109 * will be added to this. As more entries are added, the fat pointer
110 * cache line will fill, requiring additional cache lines for fat
111 * pointers to be allocated. These will be added at the head, and the
112 * aged entries will hang down, tapeworm like. As the vtw_t entries
113 * expire, the corresponding slot in the fat pointer will be
114 * reclaimed, and eventually the cache line will completely empty and
115 * be re-cycled, if not at the head of the chain.
116 *
117 * At times, a time-wait timer is restarted. This corresponds to
118 * deleting the current entry and re-adding it.
119 *
120 * Most of the time, they are just placed here to die.
121 */
122#ifndef _NETINET_TCP_VTW_H
123#define _NETINET_TCP_VTW_H
124
125#include <sys/types.h>
126#include <sys/socket.h>
127#include <sys/sysctl.h>
128#include <net/if.h>
129#include <netinet/in.h>
130#include <netinet/in_systm.h>
131#include <netinet/ip.h>
132#include <netinet/in_pcb.h>
133#include <netinet/in_var.h>
134#include <netinet/ip_var.h>
135#include <netinet/in.h>
136#include <netinet/tcp.h>
137#include <netinet/tcp_timer.h>
138#include <netinet/tcp_var.h>
139#include <netinet6/in6.h>
140#include <netinet/ip6.h>
141#include <netinet6/ip6_var.h>
142#include <netinet6/in6_pcb.h>
143#include <netinet6/ip6_var.h>
144#include <netinet6/in6_var.h>
145#include <netinet/icmp6.h>
146
147#define VTW_NCLASS (1+3) /* # different classes */
148
149/*
150 * fat pointers, MI.
151 */
152struct fatp_mi;
153
154#if CACHE_LINE_SIZE == 128
155typedef uint64_t fatp_word_t;
156#else
157typedef uint32_t fatp_word_t;
158#endif
159
160typedef struct fatp_mi fatp_t;
161
162/* Supported cacheline sizes: 32 64 128 bytes. See fatp_key(),
163 * fatp_slot_from_key(), fatp_xtra[].
164 */
165#define FATP_NTAGS (CACHE_LINE_SIZE / sizeof(fatp_word_t) - 1)
166#define FATP_NXT_WIDTH (sizeof(fatp_word_t) * NBBY - FATP_NTAGS)
167
168#define FATP_MAX (1 << (FATP_NXT_WIDTH < 31 ? FATP_NXT_WIDTH : 31))
169
170/* Worked example: ULP32 with 64-byte cacheline (32-bit x86):
171 * 15 tags per cacheline. At most 2^17 fat pointers per fatp_ctl_t.
172 * The comments on the fatp_mi members, below, correspond to the worked
173 * example.
174 */
175struct fatp_mi {
176 fatp_word_t inuse : FATP_NTAGS; /* (1+15)*4 == CL_SIZE */
177 fatp_word_t nxt : FATP_NXT_WIDTH;/* at most 2^17 fat pointers */
178 fatp_word_t tag[FATP_NTAGS]; /* 15 tags per CL */
179};
180
181static __inline int
182fatp_ntags(void)
183{
184 return FATP_NTAGS;
185}
186
187static __inline int
188fatp_full(fatp_t *fp)
189{
190 fatp_t full;
191
192 full.inuse = (1U << FATP_NTAGS) - 1U;
193
194 return (fp->inuse == full.inuse);
195}
196
197struct vtw_common;
198struct vtw_v4;
199struct vtw_v6;
200struct vtw_ctl;
201
202/*!\brief common to all vtw
203 */
204typedef struct vtw_common {
205 struct timeval expire; /* date of birth+msl */
206 uint32_t key; /* hash key: full hash */
207 uint32_t port_key; /* hash key: local port hash */
208 uint32_t rcv_nxt;
209 uint32_t rcv_wnd;
210 uint32_t snd_nxt;
211 uint32_t snd_scale : 8; /* window scaling for send win */
212 uint32_t msl_class : 2; /* TCP MSL class {0,1,2,3} */
213 uint32_t reuse_port : 1;
214 uint32_t reuse_addr : 1;
215 uint32_t v6only : 1;
216 uint32_t hashed : 1; /* reachable via FATP */
217 uint32_t uid;
218} vtw_t;
219
220/*!\brief vestigial timewait for IPv4
221 */
222typedef struct vtw_v4 {
223 vtw_t common; /* must be first */
224 uint16_t lport;
225 uint16_t fport;
226 uint32_t laddr;
227 uint32_t faddr;
228} vtw_v4_t;
229
230/*!\brief vestigial timewait for IPv6
231 */
232typedef struct vtw_v6 {
233 vtw_t common; /* must be first */
234 uint16_t lport;
235 uint16_t fport;
236 struct in6_addr laddr;
237 struct in6_addr faddr;
238} vtw_v6_t;
239
240struct fatp_ctl;
241typedef struct vtw_ctl vtw_ctl_t;
242typedef struct fatp_ctl fatp_ctl_t;
243
244/*
245 * The vestigial time waits are kept in a contiguous chunk.
246 * Allocation and free pointers run as clock hands thru this array.
247 */
248struct vtw_ctl {
249 fatp_ctl_t *fat; /* collection of fatp to use */
250 vtw_ctl_t *ctl; /* <! controller's controller */
251 union {
252 vtw_t *v; /* common */
253 struct vtw_v4 *v4; /* IPv4 resources */
254 struct vtw_v6 *v6; /* IPv6 resources */
255 } base, /* base of vtw_t array */
256 /**/ lim, /* extent of vtw_t array */
257 /**/ alloc, /* allocation pointer */
258 /**/ oldest; /* ^ to oldest */
259 uint32_t nfree; /* # free */
260 uint32_t nalloc; /* # allocated */
261 uint32_t idx_mask; /* mask capturing all index bits*/
262 uint32_t is_v4 : 1;
263 uint32_t is_v6 : 1;
264 uint32_t idx_bits: 6;
265 uint32_t clidx : 3; /* <! class index */
266};
267
268/*!\brief Collections of fat pointers.
269 */
270struct fatp_ctl {
271 vtw_ctl_t *vtw; /* associated VTWs */
272 fatp_t *base; /* base of fatp_t array */
273 fatp_t *lim; /* extent of fatp_t array */
274 fatp_t *free; /* free list */
275 uint32_t mask; /* hash mask */
276 uint32_t nfree; /* # free */
277 uint32_t nalloc; /* # allocated */
278 fatp_t **hash; /* hash anchors */
279 fatp_t **port; /* port hash anchors */
280};
281
282/*!\brief stats
283 */
284struct vtw_stats {
285 uint64_t ins; /* <! inserts */
286 uint64_t del; /* <! deleted */
287 uint64_t kill; /* <! assassination */
288 uint64_t look[2]; /* <! lookup: full hash, port hash */
289 uint64_t hit[2]; /* <! lookups that hit */
290 uint64_t miss[2]; /* <! lookups that miss */
291 uint64_t probe[2]; /* <! hits+miss */
292 uint64_t losing[2]; /* <! misses requiring dereference */
293 uint64_t max_chain[2]; /* <! max fatp chain traversed */
294 uint64_t max_probe[2]; /* <! max probes in any one chain */
295 uint64_t max_loss[2]; /* <! max losing probes in any one
296 * chain
297 */
298};
299
300typedef struct vtw_stats vtw_stats_t;
301
302/*!\brief follow fatp next 'pointer'
303 */
304static __inline fatp_t *
305fatp_next(fatp_ctl_t *fat, fatp_t *fp)
306{
307 return fp->nxt ? fat->base + fp->nxt-1 : 0;
308}
309
310/*!\brief determine a collection-relative fat pointer index.
311 */
312static __inline uint32_t
313fatp_index(fatp_ctl_t *fat, fatp_t *fp)
314{
315 return fp ? 1 + (fp - fat->base) : 0;
316}
317
318
319static __inline uint32_t
320v4_tag(uint32_t faddr, uint32_t fport, uint32_t laddr, uint32_t lport)
321{
322 return (ntohl(faddr) + ntohs(fport)
323 + ntohl(laddr) + ntohs(lport));
324}
325
326static __inline uint32_t
327v6_tag(const struct in6_addr *faddr, uint16_t fport,
328 const struct in6_addr *laddr, uint16_t lport)
329{
330#ifdef IN6_HASH
331 return IN6_HASH(faddr, fport, laddr, lport);
332#else
333 return 0;
334#endif
335}
336
337static __inline uint32_t
338v4_port_tag(uint16_t lport)
339{
340 uint32_t tag = lport ^ (lport << 11);
341
342 tag ^= tag << 3;
343 tag += tag >> 5;
344 tag ^= tag << 4;
345 tag += tag >> 17;
346 tag ^= tag << 25;
347 tag += tag >> 6;
348
349 return tag;
350}
351
352static __inline uint32_t
353v6_port_tag(uint16_t lport)
354{
355 return v4_port_tag(lport);
356}
357
358struct tcpcb;
359struct tcphdr;
360
361int vtw_add(int, struct tcpcb *);
362void vtw_del(vtw_ctl_t *, vtw_t *);
363int vtw_lookup_v4(const struct ip *ip, const struct tcphdr *th,
364 uint32_t faddr, uint16_t fport,
365 uint32_t laddr, uint16_t lport);
366struct ip6_hdr;
367struct in6_addr;
368
369int vtw_lookup_v6(const struct ip6_hdr *ip, const struct tcphdr *th,
370 const struct in6_addr *faddr, uint16_t fport,
371 const struct in6_addr *laddr, uint16_t lport);
372
373typedef struct vestigial_inpcb {
374 union {
375 struct in_addr v4;
376 struct in6_addr v6;
377 } faddr, laddr;
378 uint16_t fport, lport;
379 uint32_t valid : 1;
380 uint32_t v4 : 1;
381 uint32_t reuse_addr : 1;
382 uint32_t reuse_port : 1;
383 uint32_t v6only : 1;
384 uint32_t more_tbd : 1;
385 uint32_t uid;
386 uint32_t rcv_nxt;
387 uint32_t rcv_wnd;
388 uint32_t snd_nxt;
389 struct vtw_common *vtw;
390 struct vtw_ctl *ctl;
391} vestigial_inpcb_t;
392
393#ifdef _KERNEL
394void vtw_restart(vestigial_inpcb_t*);
395int vtw_earlyinit(void);
396int sysctl_tcp_vtw_enable(SYSCTLFN_PROTO);
397#endif /* _KERNEL */
398
399#ifdef VTW_DEBUG
400typedef struct sin_either {
401 uint8_t sin_len;
402 uint8_t sin_family;
403 uint16_t sin_port;
404 union {
405 struct in_addr v4;
406 struct in6_addr v6;
407 } sin_addr;
408} sin_either_t;
409
410int vtw_debug_add(int af, sin_either_t *, sin_either_t *, int, int);
411
412typedef struct vtw_sysargs {
413 uint32_t op;
414 sin_either_t fa;
415 sin_either_t la;
416} vtw_sysargs_t;
417
418#endif /* VTW_DEBUG */
419
420#endif /* _NETINET_TCP_VTW_H */