dn.c

这是一个同样来自贝尔实验室的和UNIX有着渊源的操作系统, 其简洁的设计和实现易于我们学习和理解

#include <u.h>
#include <libc.h>
#include <ip.h>
#include <pool.h>
#include <ctype.h>
#include "dns.h"

/*
 *  Hash table for domain names.  The hash is based only on the
 *  first element of the domain name.
 */
DN	*ht[HTLEN];


static struct
{
	Lock;
	ulong	names;	/* names allocated */
	ulong	oldest;	/* longest we'll leave a name around */
	int	active;
	int	mutex;
	int	id;
} dnvars;

/* names of RR types */
char *rrtname[] =
{
[Ta]		"ip",
[Tns]		"ns",
[Tmd]		"md",
[Tmf]		"mf",
[Tcname]	"cname",
[Tsoa]		"soa",
[Tmb]		"mb",
[Tmg]		"mg",
[Tmr]		"mr",
[Tnull]		"null",
[Twks]		"wks",
[Tptr]		"ptr",
[Thinfo]	"hinfo",
[Tminfo]	"minfo",
[Tmx]		"mx",
[Ttxt]		"txt",
[Trp]		"rp",
[Tafsdb]	"afsdb",
[Tx25]		"x.25",
[Tisdn]		"isdn",
[Trt]		"rt",
[Tnsap]		"nsap",
[Tnsapptr]	"nsap-ptr",
[Tsig]		"sig",
[Tkey]		"key",
[Tpx]		"px",
[Tgpos]		"gpos",
[Taaaa]		"ipv6",
[Tloc]		"loc",
[Tnxt]		"nxt",
[Teid]		"eid",
[Tnimloc]	"nimrod",
[Tsrv]		"srv",
[Tatma]		"atma",
[Tnaptr]	"naptr",
[Tkx]		"kx",
[Tcert]		"cert",
[Ta6]		"a6",
[Tdname]	"dname",
[Tsink]		"sink",
[Topt]		"opt",
[Tapl]		"apl",
[Tds]		"ds",
[Tsshfp]	"sshfp",
[Tipseckey]	"ipseckey",
[Trrsig]	"rrsig",
[Tnsec]		"nsec",
[Tdnskey]	"dnskey",
[Tspf]		"spf",
[Tuinfo]	"uinfo",
[Tuid]		"uid",
[Tgid]		"gid",
[Tunspec]	"unspec",
[Ttkey]		"tkey",
[Ttsig]		"tsig",
[Tixfr]		"ixfr",
[Taxfr]		"axfr",
[Tmailb]	"mailb",
[Tmaila]	"maila",
[Tall]		"all",
		0,
};

/* names of response codes */
char *rname[Rmask+1] =
{
[Rok]			"ok",
[Rformat]		"format error",
[Rserver]		"server failure",
[Rname]			"bad name",
[Runimplimented]	"unimplemented",
[Rrefused]		"we don't like you",
[Ryxdomain]		"name should not exist",
[Ryxrrset]		"rr set should not exist",
[Rnxrrset]		"rr set should exist",
[Rnotauth]		"not authorative",
[Rnotzone]		"not in zone",
[Rbadvers]		"bad opt version",
/* [Rbadsig]		"bad signature", */
[Rbadkey]		"bad key",
[Rbadtime]		"bad signature time",
[Rbadmode]		"bad mode",
[Rbadname]		"duplicate key name",
[Rbadalg]		"bad algorithm",
};

/* names of op codes */
char *opname[] =
{
[Oquery]	"query",
[Oinverse]	"inverse query (retired)",
[Ostatus]	"status",
[Oupdate]	"update",
};

Lock	dnlock;

static int sencodefmt(Fmt*);

/*
 *  set up a pipe to use as a lock
 */
void
dninit(void)
{
	fmtinstall('E', eipfmt);
	fmtinstall('I', eipfmt);
	fmtinstall('V', eipfmt);
	fmtinstall('R', rrfmt);
	fmtinstall('Q', rravfmt);
	fmtinstall('H', sencodefmt);

	dnvars.oldest = maxage;
	dnvars.names = 0;
}

/*
 *  hash for a domain name
 */
static ulong
dnhash(char *name)
{
	ulong hash;
	uchar *val = (uchar*)name;

	for(hash = 0; *val; val++)
		hash = (hash*13) + tolower(*val)-'a';
	return hash % HTLEN;
}

/*
 *  lookup a symbol.  if enter is not zero and the name is
 *  not found, create it.
 */
DN*
dnlookup(char *name, int class, int enter)
{
	DN **l;
	DN *dp;

	l = &ht[dnhash(name)];
	lock(&dnlock);
	for(dp = *l; dp; dp = dp->next) {
		assert(dp->magic == DNmagic);
		if(dp->class == class && cistrcmp(dp->name, name) == 0){
			dp->referenced = now;
			unlock(&dnlock);
			return dp;
		}
		l = &dp->next;
	}
	if(enter == 0){
		unlock(&dnlock);
		return 0;
	}
	dnvars.names++;
	dp = emalloc(sizeof(*dp));
	dp->magic = DNmagic;
	dp->name = estrdup(name);
	assert(dp->name != 0);
	dp->class = class;
	dp->rr = 0;
	dp->next = 0;
	dp->referenced = now;
	*l = dp;
	unlock(&dnlock);

	return dp;
}

/*
 *  dump the cache
 */
void
dndump(char *file)
{
	DN *dp;
	int i, fd;
	RR *rp;

	fd = open(file, OWRITE|OTRUNC);
	if(fd < 0)
		return;
	lock(&dnlock);
	for(i = 0; i < HTLEN; i++){
		for(dp = ht[i]; dp; dp = dp->next){
			fprint(fd, "%s\n", dp->name);
			for(rp = dp->rr; rp; rp = rp->next)
				fprint(fd, "	%R %c%c %lud/%lud\n", rp, rp->auth?'A':'U',
					rp->db?'D':'N', rp->expire, rp->ttl);
		}
	}
	unlock(&dnlock);
	close(fd);
}

/*
 *  purge all records
 */
void
dnpurge(void)
{
	DN *dp;
	RR *rp, *srp;
	int i;

	lock(&dnlock);

	for(i = 0; i < HTLEN; i++)
		for(dp = ht[i]; dp; dp = dp->next){
			srp = rp = dp->rr;
			dp->rr = nil;
			for(; rp != nil; rp = rp->next)
				rp->cached = 0;
			rrfreelist(srp);
		}

	unlock(&dnlock);
}

/*
 *  check the age of resource records, free any that have timed out
 */
void
dnage(DN *dp)
{
	RR **l;
	RR *rp, *next;
	ulong diff;

	diff = now - dp->referenced;
	if(diff < Reserved)
		return;

	l = &dp->rr;
	for(rp = dp->rr; rp; rp = next){
		assert(rp->magic == RRmagic && rp->cached);
		next = rp->next;
		if(!rp->db)
		if(rp->expire < now || diff > dnvars.oldest){
			*l = next;
			rp->cached = 0;
			rrfree(rp);
			continue;
		}
		l = &rp->next;
	}
}

#define REF(x) if(x) x->refs++

/*
 *  our target is 4000 names cached, this should be larger on large servers
 */
#define TARGET 4000

/*
 *  periodicly sweep for old records and remove unreferenced domain names
 *
 *  only called when all other threads are locked out
 */
void
dnageall(int doit)
{
	DN *dp, **l;
	int i;
	RR *rp;
	static ulong nextage;

	if(dnvars.names < TARGET && now < nextage && !doit){
		dnvars.oldest = maxage;
		return;
	}

	if(dnvars.names > TARGET)
		dnvars.oldest /= 2;
	nextage = now + maxage;

	lock(&dnlock);

	/* time out all old entries (and set refs to 0) */
	for(i = 0; i < HTLEN; i++)
		for(dp = ht[i]; dp; dp = dp->next){
			dp->refs = 0;
			dnage(dp);
		}

	/* mark all referenced domain names */
	for(i = 0; i < HTLEN; i++)
		for(dp = ht[i]; dp; dp = dp->next)
			for(rp = dp->rr; rp; rp = rp->next){
				REF(rp->owner);
				if(rp->negative){
					REF(rp->negsoaowner);
					continue;
				}
				switch(rp->type){
				case Thinfo:
					REF(rp->cpu);
					REF(rp->os);
					break;
				case Ttxt:
					break;
				case Tcname:
				case Tmb:
				case Tmd:
				case Tmf:
				case Tns:
					REF(rp->host);
					break;
				case Tmg:
				case Tmr:
					REF(rp->mb);
					break;
				case Tminfo:
					REF(rp->rmb);
					REF(rp->mb);
					break;
				case Trp:
					REF(rp->rmb);
					REF(rp->rp);
					break;
				case Tmx:
					REF(rp->host);
					break;
				case Ta:
				case Taaaa:
					REF(rp->ip);
					break;
				case Tptr:
					REF(rp->ptr);
					break;
				case Tsoa:
					REF(rp->host);
					REF(rp->rmb);
					break;
				}
			}

	/* sweep and remove unreferenced domain names */
	for(i = 0; i < HTLEN; i++){
		l = &ht[i];
		for(dp = *l; dp; dp = *l){
			if(dp->rr == 0 && dp->refs == 0){
				assert(dp->magic == DNmagic);
				*l = dp->next;
				if(dp->name)
					free(dp->name);
				dp->magic = ~dp->magic;
				dnvars.names--;
				free(dp);
				continue;
			}
			l = &dp->next;
		}
	}

	unlock(&dnlock);
}

/*
 *  timeout all database records (used when rereading db)
 */
void
dnagedb(void)
{
	DN *dp;
	int i;
	RR *rp;
	static ulong nextage;

	lock(&dnlock);

	/* time out all database entries */
	for(i = 0; i < HTLEN; i++)
		for(dp = ht[i]; dp; dp = dp->next)
			for(rp = dp->rr; rp; rp = rp->next)
				if(rp->db)
					rp->expire = 0;

	unlock(&dnlock);
}

/*
 *  mark all local db records about my area as authoritative, time out any others
 */
void
dnauthdb(void)
{
	DN *dp;
	int i;
	Area *area;
	RR *rp;
	static ulong nextage;

	lock(&dnlock);

	/* time out all database entries */
	for(i = 0; i < HTLEN; i++)
		for(dp = ht[i]; dp; dp = dp->next){
			area = inmyarea(dp->name);
			for(rp = dp->rr; rp; rp = rp->next)
				if(rp->db){
					if(area){
						if(rp->ttl < area->soarr->soa->minttl)
							rp->ttl = area->soarr->soa->minttl;
						rp->auth = 1;
					}
					if(rp->expire == 0){
						rp->db = 0;
						dp->referenced = now - Reserved - 1;
					}
				}
		}

	unlock(&dnlock);
}

/*
 *  keep track of other processes to know if we can
 *  garbage collect.  block while garbage collecting.
 */
int
getactivity(Request *req, int recursive)
{
	int rv;

	if(traceactivity) syslog(0, "dns", "get %d by %d from %p", dnvars.active, getpid(), getcallerpc(&req));
	lock(&dnvars);
	/*
	 * can't block here if we're already holding one
	 * of the dnvars.active (recursive).  will deadlock.
	 */
	while(!recursive && dnvars.mutex){
		unlock(&dnvars);
		sleep(200);
		lock(&dnvars);
	}
	rv = ++dnvars.active;
	now = time(0);
	req->id = ++dnvars.id;
	unlock(&dnvars);

	return rv;
}
void
putactivity(int recursive)
{
	static ulong lastclean;

	if(traceactivity) syslog(0, "dns", "put %d by %d", dnvars.active, getpid());
	lock(&dnvars);
	dnvars.active--;
	assert(dnvars.active >= 0); /* "dnvars.active %d", dnvars.active */;

	/*
	 *  clean out old entries and check for new db periodicly
	 *  can't block here if being called to let go a "recursive" lock
	 *  or we'll deadlock waiting for ourselves to give up the dnvars.active.
	 */
	if(recursive || dnvars.mutex || (needrefresh == 0 && dnvars.active > 0)){
		unlock(&dnvars);
		return;
	}

	/* wait till we're alone */
	dnvars.mutex = 1;
	while(dnvars.active > 0){
		unlock(&dnvars);
		sleep(100);
		lock(&dnvars);
	}
	unlock(&dnvars);

	db2cache(needrefresh);
	dnageall(0);

	/* let others back in */
	lastclean = now;
	needrefresh = 0;
	dnvars.mutex = 0;
}

/*
 *  Attach a single resource record to a domain name.
 *	- Avoid duplicates with already present RR's
 *	- Chain all RR's of the same type adjacent to one another
 *	- chain authoritative RR's ahead of non-authoritative ones
 */
static void
rrattach1(RR *new, int auth)
{
	RR **l;
	RR *rp;
	DN *dp;

	assert(new->magic == RRmagic && !new->cached);

	if(!new->db)
		new->expire = new->ttl;
	else
		new->expire = now + Year;
	dp = new->owner;
	assert(dp->magic == DNmagic);
	new->auth |= auth;
	new->next = 0;

	/*
	 *  find first rr of the right type
	 */
	l = &dp->rr;
	for(rp = *l; rp; rp = *l){
		assert(rp->magic == RRmagic && rp->cached);
		if(rp->type == new->type)
			break;
		l = &rp->next;
	}

	/*
	 *  negative entries replace positive entries
	 *  positive entries replace negative entries
	 *  newer entries replace older entries with the same fields
	 */
	for(rp = *l; rp; rp = *l){
		assert(rp->magic == RRmagic && rp->cached);
		if(rp->type != new->type)
			break;

		if(rp->db == new->db && rp->auth == new->auth){
			/* negative drives out positive and vice versa */
			if(rp->negative != new->negative){
				*l = rp->next;
				rp->cached = 0;
				rrfree(rp);
				continue;
			}

			/* all things equal, pick the newer one */
			if(rp->arg0 == new->arg0 && rp->arg1 == new->arg1){
				/* new drives out old */
				if(new->ttl > rp->ttl || new->expire > rp->expire){
					*l = rp->next;
					rp->cached = 0;
					rrfree(rp);
					continue;
				} else {
					rrfree(new);
					return;
				}
			}

			/*  Hack for pointer records.  This makes sure
			 *  the ordering in the list reflects the ordering
			 *  received or read from the database
			 */
			if(rp->type == Tptr){
				if(!rp->negative && !new->negative
				&& rp->ptr->ordinal > new->ptr->ordinal)
					break;
			}
		}
		l = &rp->next;
	}

	/*
	 *  add to chain
	 */
	new->cached = 1;
	new->next = *l;
	*l = new;
}

/*
 *  Attach a list of resource records to a domain name.
 *	- Avoid duplicates with already present RR's
 *	- Chain all RR's of the same type adjacent to one another
 *	- chain authoritative RR's ahead of non-authoritative ones
 *	- remove any expired RR's
 */
void
rrattach(RR *rp, int auth)
{
	RR *next;

	lock(&dnlock);
	for(; rp; rp = next){
		next = rp->next;
		rp->next = 0;

		/* avoid any outside spoofing */
		if(cachedb && !rp->db && inmyarea(rp->owner->name))
			rrfree(rp);
		else
			rrattach1(rp, auth);
	}
	unlock(&dnlock);
}

/*
 *  allocate a resource record of a given type
 */
RR*
rralloc(int type)
{
	RR *rp;

	rp = emalloc(sizeof(*rp));
	rp->magic = RRmagic;
	rp->pc = getcallerpc(&type);
	rp->type = type;
	switch(type){
	case Tsoa:
		rp->soa = emalloc(sizeof(*rp->soa));
		rp->soa->slaves = nil;
		break;
	case Tkey:
		rp->key = emalloc(sizeof(*rp->key));
		break;
	case Tcert:
		rp->cert = emalloc(sizeof(*rp->cert));
		break;
	case Tsig:
		rp->sig = emalloc(sizeof(*rp->sig));
		break;
	case Tnull:
		rp->null = emalloc(sizeof(*rp->null));
		break;
	}
	rp->ttl = 0;
	rp->expire = 0;
	rp->next = 0;
	return rp;
}

/*
 *  free a resource record and any related structs
 */
void
rrfree(RR *rp)
{
	DN *dp;
	RR *nrp;
	Txt *t;

	assert(rp->magic = RRmagic);
	assert(!rp->cached);

	dp = rp->owner;
	if(dp){
		assert(dp->magic == DNmagic);
		for(nrp = dp->rr; nrp; nrp = nrp->next)
			assert(nrp != rp); /* "rrfree of live rr" */;
	}

	switch(rp->type){
	case Tsoa:
		freeserverlist(rp->soa->slaves);
		free(rp->soa);
		break;
	case Tkey:
		free(rp->key->data);
		free(rp->key);
		break;
	case Tcert:
		free(rp->cert->data);
		free(rp->cert);
		break;
	case Tsig:
		free(rp->sig->data);
		free(rp->sig);
		break;
	case Tnull:
		free(rp->null->data);
		free(rp->null);
		break;
	case Ttxt:
		while(rp->txt != nil){
			t = rp->txt;
			rp->txt = t->next;
			free(t->p);
			free(t);
		}
		break;
	}

	rp->magic = ~rp->magic;
	free(rp);
}

/*
 *  free a list of resource records and any related structs
 */
void
rrfreelist(RR *rp)
{
	RR *next;

	for(; rp; rp = next){
		next = rp->next;
		rrfree(rp);
	}
}

extern RR**
rrcopy(RR *rp, RR **last)
{
	RR *nrp;
	SOA *soa;
	Key *key;
	Cert *cert;
	Sig *sig;
	Null *null;
	Txt *t, *nt, **l;

	nrp = rralloc(rp->type);
	switch(rp->type){
	case Ttxt:
		*nrp = *rp;
		l = &nrp->txt;
		*l = nil;
		for(t = rp->txt; t != nil; t = t->next){
			nt = emalloc(sizeof(*nt));
			nt->p = estrdup(t->p);
			nt->next = nil;
			*l = nt;
			l = &nt->next;
		}
		break;
	case Tsoa:
		soa = nrp->soa;
		*nrp = *rp;
		nrp->soa = soa;
		*nrp->soa = *rp->soa;
		nrp->soa->slaves = copyserverlist(rp->soa->slaves);
		break;
	case Tkey:
		key = nrp->key;
		*nrp = *rp;
		nrp->key = key;
		*key = *rp->key;
		key->data = emalloc(key->dlen);
		memmove(key->data, rp->key->data, rp->key->dlen);
		break;
	case Tsig:
		sig = nrp->sig;
		*nrp = *rp;
		nrp->sig = sig;
		*sig = *rp->sig;
		sig->data = emalloc(sig->dlen);
		memmove(sig->data, rp->sig->data, rp->sig->dlen);
		break;
	case Tcert:
		cert = nrp->cert;
		*nrp = *rp;
		nrp->cert = cert;
		*cert = *rp->cert;
		cert->data = emalloc(cert->dlen);
		memmove(cert->data, rp->cert->data, rp->cert->dlen);
		break;
	case Tnull:
		null = nrp->null;
		*nrp = *rp;
		nrp->null = null;
		*null = *rp->null;
		null->data = emalloc(null->dlen);
		memmove(null->data, rp->null->data, rp->null->dlen);
		break;
	default:
		*nrp = *rp;
		break;
	}
	nrp->cached = 0;
	nrp->next = 0;
	*last = nrp;
	return &nrp->next;
}

/*
 *  lookup a resource record of a particular type and
 *  class attached to a domain name.  Return copies.
 *
 *  Priority ordering is:
 *	db authoritative
 *	not timed out network authoritative
 *	not timed out network unauthoritative
 *	unauthoritative db
 *
 *  if flag NOneg is set, don't return negative cached entries.
 *  return nothing instead.