New sentence, new line.

This page could use more markup.
This commit is contained in:
wiz 2009-04-11 16:55:04 +00:00
parent 7d9f3abb75
commit 25b7d5a452

View File

@ -1,4 +1,4 @@
.\" $NetBSD: evdns.3,v 1.2 2009/04/11 15:29:50 joerg Exp $
.\" $NetBSD: evdns.3,v 1.3 2009/04/11 16:55:04 wiz Exp $
.\"
.\" Copyright (c) 2006 Niels Provos <provos@citi.umich.edu>
.\" All rights reserved.
@ -85,40 +85,48 @@ Welcome, gentle reader
.Pp
Async DNS lookups are really a whole lot harder than they should be,
mostly stemming from the fact that the libc resolver has never been
very good at them. Before you use this library you should see if libc
very good at them.
Before you use this library you should see if libc
can do the job for you with the modern async call getaddrinfo_a
(see http://www.imperialviolet.org/page25.html#e498). Otherwise,
please continue.
(see http://www.imperialviolet.org/page25.html#e498).
Otherwise, please continue.
.Pp
This code is based on libevent and you must call event_init before
any of the APIs in this file. You must also seed the OpenSSL random
any of the APIs in this file.
You must also seed the OpenSSL random
source if you are using OpenSSL for ids (see below).
.Pp
This library is designed to be included and shipped with your source
code. You statically link with it. You should also test for the
existence of strtok_r and define HAVE_STRTOK_R if you have it.
code.
You statically link with it.
You should also test for the existence of strtok_r and define
HAVE_STRTOK_R if you have it.
.Pp
The DNS protocol requires a good source of id numbers and these
numbers should be unpredictable for spoofing reasons. There are
three methods for generating them here and you must define exactly
one of them. In increasing order of preference:
numbers should be unpredictable for spoofing reasons.
There are three methods for generating them here and you must define
exactly one of them.
In increasing order of preference:
.Pp
.Bl -tag -width "DNS_USE_GETTIMEOFDAY_FOR_ID" -compact -offset indent
.It DNS_USE_GETTIMEOFDAY_FOR_ID
Using the bottom 16 bits of the usec result from gettimeofday. This
is a pretty poor solution but should work anywhere.
Using the bottom 16 bits of the usec result from gettimeofday.
This is a pretty poor solution but should work anywhere.
.It DNS_USE_CPU_CLOCK_FOR_ID
Using the bottom 16 bits of the nsec result from the CPU's time
counter. This is better, but may not work everywhere. Requires
POSIX realtime support and you'll need to link against -lrt on
glibc systems at least.
counter.
This is better, but may not work everywhere.
Requires POSIX realtime support and you'll need to link against
-lrt on glibc systems at least.
.It DNS_USE_OPENSSL_FOR_ID
Uses the OpenSSL RAND_bytes call to generate the data. You must
have seeded the pool before making any calls to this library.
Uses the OpenSSL RAND_bytes call to generate the data.
You must have seeded the pool before making any calls to this
library.
.El
.Pp
The library keeps track of the state of nameservers and will avoid
them when they go down. Otherwise it will round robin between them.
them when they go down.
Otherwise it will round robin between them.
.Pp
Quick start guide:
.Bd -literal
@ -130,8 +138,8 @@ evdns_resolv_conf_parse(DNS_OPTIONS_ALL, "/etc/resolv.conf");
evdns_resolve("www.hostname.com", 0, callback, NULL);
.Ed
.Pp
When the lookup is complete the callback function is called. The
first argument will be one of the DNS_ERR_* defines in evdns.h.
When the lookup is complete the callback function is called.
The first argument will be one of the DNS_ERR_* defines in evdns.h.
Hopefully it will be DNS_ERR_NONE, in which case type will be
DNS_IPv4_A, count will be the number of IP addresses, ttl is the time
which the data can be cached for (in seconds), addresses will point
@ -141,19 +149,22 @@ evdns_resolve.
Searching:
.Pp
In order for this library to be a good replacement for glibc's resolver it
supports searching. This involves setting a list of default domains, in
which names will be queried for. The number of dots in the query name
determines the order in which this list is used.
supports searching.
This involves setting a list of default domains, in which names
will be queried for.
The number of dots in the query name determines the order in which
this list is used.
.Pp
Searching appears to be a single lookup from the point of view of the API,
although many DNS queries may be generated from a single call to
evdns_resolve. Searching can also drastically slow down the resolution
of names.
evdns_resolve.
Searching can also drastically slow down the resolution of names.
.Pp
To disable searching:
.Bl -enum -compact -offset indent
.It
Never set it up. If you never call
Never set it up.
If you never call
.Fn evdns_resolv_conf_parse ,
.Fn evdns_init ,
or
@ -175,9 +186,10 @@ pass the
flag.
.El
.Pp
The order of searches depends on the number of dots in the name. If the
number is greater than the ndots setting then the names is first tried
globally. Otherwise each search domain is appended in turn.
The order of searches depends on the number of dots in the name.
If the number is greater than the ndots setting then the names is
first tried globally.
Otherwise each search domain is appended in turn.
.Pp
The ndots setting can either be set from a resolv.conf, or by calling
evdns_search_ndots_set.
@ -189,24 +201,26 @@ For example, with ndots set to 1 (the default) and a search domain list of
.Pp
Query: www.abc
Order: www.abc., www.abc.myhome.net
.Sh API reference
.Ss API reference
.Bl -tag -width 0123456
.It Ft int Fn evdns_init
Initializes support for non-blocking name resolution by calling
.Fn evdns_resolv_conf_parse .
.It Ft int Fn evdns_nameserver_add "unsigned long int address"
Add a nameserver. The address should be an IP address in
network byte order. The type of address is chosen so that
it matches in_addr.s_addr.
Add a nameserver.
The address should be an IP address in network byte order.
The type of address is chosen so that it matches in_addr.s_addr.
Returns non-zero on error.
.It Ft int Fn evdns_nameserver_ip_add "const char *ip_as_string"
This wraps the above function by parsing a string as an IP
address and adds it as a nameserver.
Returns non-zero on error
.It Ft int Fn evdns_resolve "const char *name" "int flags" "evdns_callback_type callback" "void *ptr"
Resolve a name. The name parameter should be a DNS name.
The flags parameter should be 0, or DNS_QUERY_NO_SEARCH
which disables searching for this query. (see defn of
Resolve a name.
The name parameter should be a DNS name.
The flags parameter should be 0, or
.Dv DNS_QUERY_NO_SEARCH
which disables searching for this query (see defn of
searching above).
.Pp
The callback argument is a function which is called when
@ -223,12 +237,13 @@ Set the number of dots which, when found in a name, causes
the first query to be without any search domain.
.It Ft int Fn evdns_count_nameservers "void"
Return the number of configured nameservers (not necessarily the
number of running nameservers). This is useful for double-checking
whether our calls to the various nameserver configuration functions
have been successful.
number of running nameservers).
This is useful for double-checking whether our calls to the various
nameserver configuration functions have been successful.
.It Ft int Fn evdns_clear_nameservers_and_suspend "void"
Remove all currently configured nameservers, and suspend all pending
resolves. Resolves will not necessarily be re-attempted until
resolves.
Resolves will not necessarily be re-attempted until
.Fn evdns_resume
is called.
.It Ft int Fn evdns_resume "void"
@ -270,22 +285,27 @@ out of memory
short read from file
.El
.El
.Sh Internals
Requests are kept in two queues. The first is the inflight queue. In
this queue requests have an allocated transaction id and nameserver.
.Ss Internals
Requests are kept in two queues.
The first is the inflight queue.
In this queue requests have an allocated transaction id and
nameserver.
They will soon be transmitted if they haven't already been.
.Pp
The second is the waiting queue. The size of the inflight ring is
limited and all other requests wait in waiting queue for space. This
bounds the number of concurrent requests so that we don't flood the
nameserver. Several algorithms require a full walk of the inflight
The second is the waiting queue.
The size of the inflight ring is limited and all other requests
wait in waiting queue for space.
This bounds the number of concurrent requests so that we don't
flood the nameserver.
Several algorithms require a full walk of the inflight
queue and so bounding its size keeps thing going nicely under huge
(many thousands of requests) loads.
.Pp
If a nameserver loses too many requests it is considered down and we
try not to use it. After a while we send a probe to that nameserver
(a lookup for google.com) and, if it replies, we consider it working
again. If the nameserver fails a probe we wait longer to try again
try not to use it.
After a while we send a probe to that nameserver (a lookup for
google.com) and, if it replies, we consider it working again.
If the nameserver fails a probe we wait longer to try again
with the next probe.
.Sh SEE ALSO
.Xr event 3 ,