Home | History | Annotate | Line # | Download | only in netinet
      1 /*	$NetBSD: tcp_vtw.h,v 1.11 2024/10/07 23:17:00 jakllsch 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  */
    152 struct fatp_mi;
    153 
    154 #if CACHE_LINE_SIZE >= 128
    155 typedef uint64_t fatp_word_t;
    156 #else
    157 typedef uint32_t fatp_word_t;
    158 #endif
    159 
    160 typedef 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  */
    175 struct 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 
    181 static __inline int
    182 fatp_ntags(void)
    183 {
    184 	return FATP_NTAGS;
    185 }
    186 
    187 static __inline int
    188 fatp_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 
    197 struct vtw_common;
    198 struct vtw_v4;
    199 struct vtw_v6;
    200 struct vtw_ctl;
    201 
    202 /*!\brief common to all vtw
    203  */
    204 typedef 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  */
    222 typedef 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  */
    232 typedef 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 
    240 struct fatp_ctl;
    241 typedef struct vtw_ctl		vtw_ctl_t;
    242 typedef 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  */
    248 struct 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  */
    270 struct 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  */
    284 struct 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 
    300 typedef struct vtw_stats	vtw_stats_t;
    301 
    302 /*!\brief	follow fatp next 'pointer'
    303  */
    304 static __inline fatp_t *
    305 fatp_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  */
    312 static __inline uint32_t
    313 fatp_index(fatp_ctl_t *fat, fatp_t *fp)
    314 {
    315 	return fp ? 1 + (fp - fat->base) : 0;
    316 }
    317 
    318 
    319 static __inline uint32_t
    320 v4_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 
    326 static __inline uint32_t
    327 v6_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 
    337 static __inline uint32_t
    338 v4_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 
    352 static __inline uint32_t
    353 v6_port_tag(uint16_t lport)
    354 {
    355 	return v4_port_tag(lport);
    356 }
    357 
    358 struct tcpcb;
    359 struct tcphdr;
    360 
    361 int  vtw_add(int, struct tcpcb *);
    362 void vtw_del(vtw_ctl_t *, vtw_t *);
    363 int 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);
    366 struct ip6_hdr;
    367 struct in6_addr;
    368 
    369 int 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 
    373 typedef 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
    394 void vtw_restart(vestigial_inpcb_t*);
    395 int vtw_earlyinit(void);
    396 int sysctl_tcp_vtw_enable(SYSCTLFN_PROTO);
    397 #endif /* _KERNEL */
    398 
    399 #ifdef VTW_DEBUG
    400 typedef 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 
    410 int vtw_debug_add(int af, sin_either_t *, sin_either_t *, int, int);
    411 
    412 typedef 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 */
    421