Avoid PTR lookups when possible
Subsystems (2)
PATCH 1/1 Avoid PTR lookups when possible
The attached patch adds a new command line switch to rpc.gssd to avoid PTR resolution when possible.
The current code *depends* on PTR resolution for GSSAPI authentication and this is *bad*.
It imposes an annoying, and unnecessary, constraint on the correctness of DNS resolution, which prevents mounts from working in networks where the PTR record cannot be easily controlled (for example networks where the forward name is reasonable while the PTR is set to some artificial name based on the IP address or so that is not the canonical name or where no PTR exist at all).
Depending on PTR resolution for GSSAPI is also very bad practice because it opens up DNS spoofing attacks where an attacker can try to redirect a user to the wrong server fooling mutual authentication, and induce a user to trust improper data or disclose (by copying on the impostor server) data that should be confidential.
Ideally people will always use the -N switch, however I added an actual switch to avoid breaking existing configuration where someone may intentionally or inadvertently be using multiple A names instead of using CNAMEs but relying on PTR records to find the "canonical name".
It would be probably nice to switch the behavior on by default in future.
Jeff in CC as he also implemented a similar switch for cifs.upcall (there it is called -t|--trust-dns, but in rpc.gsssd -t is already taken ...).
Simo.
Signature · 1 line
Simo Sorce * Red Hat, Inc * New York
Quoted 1 line
From
53dade1e5d22eee06a08ace2684257292514cb49Mon Sep 17 00:00:00 2001
From: Simo Sorce <[email protected]>
Date: Mon, 1 Apr 2013 21:00:55 -0400
Subject: [PATCH] Avoid forced reverse resolution for server name
A NFS client should be able to work properly even if the PTR record for the server is not set. there is no excuse to forcefully prevent that from working when it can.
So, use some heuristics to see if the server is a FQDN or an IP address. If it is an IP address or a shortname (no '.' in name) then do a PTR lookup, otherwise avoid it.
This is enabled by the new -N flag which is off by default for now, ideally we will turn this on by default after a transition period.
--- utils/gssd/gss_util.h | 2 ++ utils/gssd/gssd.c | 4 +++- utils/gssd/gssd.man | 7 ++++++- utils/gssd/gssd_proc.c | 31 +++++++++++++++++++++++++++---- 4 files changed, 38 insertions(+), 6 deletions(-)
utils/gssd/gss_util.h +2 -0
@@ -52,4 +52,6 @@ int gssd_check_mechs(void); gss_krb5_set_allowable_enctypes(min, cred, num, types)#endif +extern int avoid_ptr; +#endif /* _GSS_UTIL_H_ */
utils/gssd/gssd.c +3 -1
@@ -85,7 +85,7 @@ sig_hup(int signal)static void usage (char *progname) { -fprintf (stderr,"usage: %s [-f] [-l] [-M] [-n] [-v] [-r] [-p pipefsdir] [-k keytab] [-d ccachedir] [-t timeout] [-R preferred realm]\n" , +fprintf (stderr,"usage: %s [-f] [-l] [-M] [-n] [-v] [-r] [-p pipefsdir] [-k keytab] [-d ccachedir] [-t timeout] [-R preferred realm] [-N]\n" , progname);exit (1 ); }
@@ -150,6 +150,8 @@ main(int argc, char *argv[])errx (1 ,"Encryption type limits not supported by Kerberos libraries." );#endif break ; +case 'N' : + avoid_ptr =1 ;default :usage (argv[0 ]);break ;
utils/gssd/gssd.man +6 -1
@@ -8,7 +8,7 @@ rpc.gssd \- RPCSEC_GSS daemon .SH SYNOPSIS .B rpc.gssd -.RB [ \-fMnlvr ] +.RB [ \-fMnlvrN ] .RB [ \-k .IR keytab ] .RB [ \-p
@@ -266,6 +266,11 @@ new kernel contexts to be negotiated after seconds, which allows changing Kerberos tickets and identities frequently. The default is no explicit timeout, which means the kernel context will live the lifetime of the Kerberos service ticket used in its creation. +.TP +.B -N +Tries to avoid PTR lookups for resolving the server name, if the name used at +mount looks like a Fully Qualified Domain Name. This is import to avoid GSSAPI +issues when PTR records are set and to help avoid some kind of MITM attack. .SH SEE ALSO .BR rpc.svcgssd (8), .BR kerberos (1),
utils/gssd/gssd_proc.c +27 -4
@@ -67,6 +67,7 @@#include <errno.h> #include <gssapi/gssapi.h> #include <netdb.h> +#include <ctype.h> #include "gssd.h" #include "err_util.h"
@@ -107,6 +108,8 @@ struct pollfd * pollarray;unsigned long pollsize;/* the size of pollaray (in pollfd's) */ +int avoid_ptr =0 ; +/* * convert a presentation address string to a sockaddr_storage struct. Returns * true on success or false on failure.
@@ -165,12 +168,32 @@ addrstr_to_sockaddr(struct sockaddr *sa, const char *node, const char *port)* convert a sockaddr to a hostname */ static char * -sockaddr_to_hostname (const struct sockaddr *sa,const char *addr) +get_servername (const char *name,const struct sockaddr *sa,const char *addr) {socklen_t addrlen;int err;char *hostname;char hbuf[NI_MAXHOST ]; +const char *p; +int do_ptr_lookup =0 ; + +if (avoid_ptr) { +/* try to determine if this is FQDN, or an IP address with +* basic heuristics, if they fail try a PTR lookup */ + p =strrchr (name,'.' ); +if (!p ||strchr (name,':' )) + do_ptr_lookup =1 ;/* non-FQDN, or IPv6 */ +else { +for (++p; *p; p++) +if (!isdigit (*p)) +break ; +if (!*p) + do_ptr_lookup =1 ;/* IPv4 */ + } +if (!do_ptr_lookup) { +return strdup (name); + } + }switch (sa->sa_family) {case AF_INET :
@@ -208,7 +231,7 @@ read_service_info(char *info_file_name, char **servicename, char **servername, struct sockaddr *addr) {#define INFOBUFLEN 256char buf[INFOBUFLEN +1 ]; -static char dummy[128 ]; +static char server[128 ];int nbytes;static char service[128 ];static char address[128 ];
@@ -236,7 +259,7 @@ read_service_info(char *info_file_name, char **servicename, char **servername,"service: %127s %15s version %15s\n" "address: %127s\n" "protocol: %15s\n" , - dummy, + server, service, program, version, address, protoname);
@@ -258,7 +281,7 @@ read_service_info(char *info_file_name, char **servicename, char **servername,if (!addrstr_to_sockaddr (addr, address, port)) goto fail; - *servername =sockaddr_to_hostname (addr, address); + *servername =get_servername (server, addr, address);if (*servername ==NULL ) goto fail;
Signature · 1 line
1.8.1.4
Quoted 153 lines · On Tue, 02 Apr 2013 09:45:51 -0400 Simo Sorce <[email protected]> wrote:
From
53dade1e5d22eee06a08ace2684257292514cb49Mon Sep 17 00:00:00 2001
From: Simo Sorce <[email protected]>
Date: Mon, 1 Apr 2013 21:00:55 -0400
Subject: [PATCH] Avoid forced reverse resolution for server nameA NFS client should be able to work properly even if the PTR record for the server is not set. there is no excuse to forcefully prevent that from working when it can.
So, use some heuristics to see if the server is a FQDN or an IP address. If it is an IP address or a shortname (no '.' in name) then do a PTR lookup, otherwise avoid it.This is enabled by the new -N flag which is off by default for now, ideally we will turn this on by default after a transition period.
--- utils/gssd/gss_util.h | 2 ++ utils/gssd/gssd.c | 4 +++- utils/gssd/gssd.man | 7 ++++++- utils/gssd/gssd_proc.c | 31 +++++++++++++++++++++++++++---- 4 files changed, 38 insertions(+), 6 deletions(-)utils/gssd/gss_util.h +2
-0@@ -52,4 +52,6 @@ int gssd_check_mechs(void); gss_krb5_set_allowable_enctypes(min, cred, num, types)#endif +extern int avoid_ptr; +#endif /* _GSS_UTIL_H_ */ utils/gssd/gssd.c +3
-1@@ -85,7 +85,7 @@ sig_hup(int signal)static void usage (char *progname) { -fprintf (stderr,"usage: %s [-f] [-l] [-M] [-n] [-v] [-r] [-p pipefsdir] [-k keytab] [-d ccachedir] [-t timeout] [-R preferred realm]\n" , +fprintf (stderr,"usage: %s [-f] [-l] [-M] [-n] [-v] [-r] [-p pipefsdir] [-k keytab] [-d ccachedir] [-t timeout] [-R preferred realm] [-N]\n" , progname);exit (1 ); }@@ -150,6 +150,8 @@ main(int argc, char *argv[])errx (1 ,"Encryption type limits not supported by Kerberos libraries." );#endif break ; +case 'N' : + avoid_ptr =1 ;default :usage (argv[0 ]);break ;utils/gssd/gssd.man +6
-1@@ -8,7 +8,7 @@ rpc.gssd \- RPCSEC_GSS daemon .SH SYNOPSIS .B rpc.gssd -.RB [ \-fMnlvr ] +.RB [ \-fMnlvrN ] .RB [ \-k .IR keytab ] .RB [ \-p@@ -266,6 +266,11 @@ new kernel contexts to be negotiated after seconds, which allows changing Kerberos tickets and identities frequently. The default is no explicit timeout, which means the kernel context will live the lifetime of the Kerberos service ticket used in its creation. +.TP +.B -N +Tries to avoid PTR lookups for resolving the server name, if the name used at +mount looks like a Fully Qualified Domain Name. This is import to avoid GSSAPI +issues when PTR records are set and to help avoid some kind of MITM attack. .SH SEE ALSO .BR rpc.svcgssd (8), .BR kerberos (1),utils/gssd/gssd_proc.c +26
-3@@ -67,6 +67,7 @@#include <errno.h> #include <gssapi/gssapi.h> #include <netdb.h> +#include <ctype.h> #include "gssd.h" #include "err_util.h" @@ -107,6 +108,8 @@ struct pollfd * pollarray;unsigned long pollsize;/* the size of pollaray (in pollfd's) */ +int avoid_ptr =0 ; +/* * convert a presentation address string to a sockaddr_storage struct. Returns * true on success or false on failure. @@ -165,12 +168,32 @@ addrstr_to_sockaddr(struct sockaddr *sa, const char *node, const char *port)* convert a sockaddr to a hostname */ static char * -sockaddr_to_hostname (const struct sockaddr *sa,const char *addr) +get_servername (const char *name,const struct sockaddr *sa,const char *addr) {socklen_t addrlen;int err;char *hostname;char hbuf[NI_MAXHOST ]; +const char *p; +int do_ptr_lookup =0 ; + +if (avoid_ptr) { +/* try to determine if this is FQDN, or an IP address with +* basic heuristics, if they fail try a PTR lookup */ + p =strrchr (name,'.' ); +if (!p ||strchr (name,':' )) + do_ptr_lookup =1 ;/* non-FQDN, or IPv6 */ +else { +for (++p; *p; p++) +if (!isdigit (*p)) +break ; +if (!*p) + do_ptr_lookup =1 ;/* IPv4 */ + } +if (!do_ptr_lookup) { +return strdup (name); + } + }switch (sa->sa_family) {case AF_INET :@@ -208,7 +231,7 @@ read_service_info(char *info_file_name, char **servicename, char **servername, struct sockaddr *addr) {#define INFOBUFLEN 256char buf[INFOBUFLEN +1 ]; -static char dummy[128 ]; +static char server[128 ];int nbytes;static char service[128 ];static char address[128 ];@@ -236,7 +259,7 @@ read_service_info(char *info_file_name, char **servicename, char **servername,"service: %127s %15s version %15s\n" "address: %127s\n" "protocol: %15s\n" , - dummy, + server, service, program, version, address, protoname);
utils/gssd/gssd_proc.c read_service_info(char *info_file_name, char **servicename, char **servername, · line 287
Quoted 8 lines
utils/gssd/gssd_proc.c +1
-1@@ -258,7 +281,7 @@ read_service_info(char *info_file_name, char **servicename, char **servername,if (!addrstr_to_sockaddr (addr, address, port)) goto fail; - *servername =sockaddr_to_hostname (addr, address); + *servername =get_servername (server, addr, address);if (*servername ==NULL ) goto fail;
Ok, so with this we expect that users with "marginal" setups will break. At that point they'll need to hunt down the '-N' switch to gssd and turn it on (or fix their name resolution).
What I had in mind was a slightly more gentle transition:
Add a switch that can be on (new behavior), off (old behavior) or default. The default will start at "off" and then in a few releases, switch to "on".
In the meantime, if the switch is not set explicitly, throw a warning to syslog when the daemon starts that explains what the default is, and in what release it will change to the new behavior.
Then when the merge window for that release comes, you can reverse the default and remove the syslog warning. Eventually (in a year or two?) we could consider deprecating the new option altogether.
I do agree that it's a pain, but it at least gives users some warning and recourse for this sort of behavior change. It may also forstall some pleas for help on the mailing lists.
Signature · 1 line
Jeff Layton <[email protected]>
Simo Sorce wrote:
The attached patch adds a new command line switch to rpc.gssd to avoid PTR resolution when possible.
The current code *depends* on PTR resolution for GSSAPI authentication and this is *bad*.
It imposes an annoying, and unnecessary, constraint on the correctness of DNS resolution, which prevents mounts from working in networks where the PTR record cannot be easily controlled (for example networks where the forward name is reasonable while the PTR is set to some artificial name based on the IP address or so that is not the canonical name or where no PTR exist at all).
Depending on PTR resolution for GSSAPI is also very bad practice because it opens up DNS spoofing attacks where an attacker can try to redirect a user to the wrong server fooling mutual authentication, and induce a user to trust improper data or disclose (by copying on the impostor server) data that should be confidential.
What happens if it's a partially qualified domain name?
Wouldn't it be better to use something like inet_pton?
I agree that insisting on correct PTR records is a bad idea, but I don't understand your threat model. It shouldn't be possible for an attacker to do anything bad by redirecting the client to a spoof server. If it is possible, we've got bigger problems. How do you think that would work exactly?
Quoted 14 lines · Jim Rees
Simo Sorce wrote:
The attached patch adds a new command line switch to rpc.gssd to avoid PTR resolution when possible.
The current code *depends* on PTR resolution for GSSAPI authentication and this is *bad*.
It imposes an annoying, and unnecessary, constraint on the correctness of DNS resolution, which prevents mounts from working in networks where the PTR record cannot be easily controlled (for example networks where the forward name is reasonable while the PTR is set to some artificial name based on the IP address or so that is not the canonical name or where no PTR exist at all).
Quoted 7 lines
Depending on PTR resolution for GSSAPI is also very bad practice because it opens up DNS spoofing attacks where an attacker can try to redirect a user to the wrong server fooling mutual authentication, and induce a user to trust improper data or disclose (by copying on the impostor server) data that should be confidential.
What happens if it's a partially qualified domain name?
you use the "-N off" switch and rely on reverse resolution I guess. (about to send patch that fixes -N as Jeff asked).
Quoted 1 line
Wouldn't it be better to use something like
inet_pton?
I'll look.
Quoted 4 lines
I agree that insisting on correct PTR records is a bad idea, but I don't understand your threat model. It shouldn't be possible for an attacker to do anything bad by redirecting the client to a spoof server. If it is possible, we've got bigger problems. How do you think that would work exactly?
It's a MITM setting.
Suppose you have 2 servers:
secure.server.name where you store confidential documents public.server.name where everyone have access for reading (and you can write)
If you mount -t nfs secure.server.name/export and rely on PTR records and an attacker can spoof your PTR request to return public.server.name you will end up acquiring a ticket for public.server.name from the KDC. Then the attacker will redirect your mount to public.server.name all the while you think you have mounted secure.server.name
Now anything you copy there goes to a public server instead of your secure server.
Now imagine your secure server is a backup server which gets mounted by a cron job every night for backups of sensitive information.
Simo.
Signature · 1 line
Simo Sorce * Red Hat, Inc * New York
Simo Sorce wrote:
It's a MITM setting.
Suppose you have 2 servers:
secure.server.name where you store confidential documents public.server.name where everyone have access for reading (and you can write)
If you mount -t nfs secure.server.name/export and rely on PTR records and an attacker can spoof your PTR request to return public.server.name you will end up acquiring a ticket for public.server.name from the KDC. Then the attacker will redirect your mount to public.server.name all the while you think you have mounted secure.server.name
Now anything you copy there goes to a public server instead of your secure server.
Now imagine your secure server is a backup server which gets mounted by a cron job every night for backups of sensitive information.
I guess what you're saying is that the client might not know the full name of the secure server, and could be tricked into thinking it's public.server.name. That's not what I'd call mitm; kerberos is pretty much immune to mitm. It's not really a dns mitm either, it's a spoofed reply.
I would be in favor of requiring kerberized nfs clients to know the full name of the server they're trying to connect to. Kerberized nfs clients should not be relying on dns for any part of their security. But I guess we haven't been doing that, and to make that a requirement now would be a compatibility problem.
Quoted 19 lines · Jim Rees
Simo Sorce wrote:
It's a MITM setting.
Suppose you have 2 servers:
secure.server.name where you store confidential documents public.server.name where everyone have access for reading (and you can write)
If you mount -t nfs secure.server.name/export and rely on PTR records and an attacker can spoof your PTR request to return public.server.name you will end up acquiring a ticket for public.server.name from the KDC. Then the attacker will redirect your mount to public.server.name all the while you think you have mounted secure.server.name
Now anything you copy there goes to a public server instead of your secure server.
Quoted 7 lines
Now imagine your secure server is a backup server which gets mounted by a cron job every night for backups of sensitive information.
I guess what you're saying is that the client might not know the full name of the secure server, and could be tricked into thinking it's public.server.name. That's not what I'd call mitm; kerberos is pretty much immune to mitm. It's not really a dns mitm either, it's a spoofed reply.
No this is not what I am saying.
The client uses the fully qualified name at the mount command but due to the use of a PTR lookup it is fooled into acquiring a ticket for the wrong server which will then happily allow you to connect.
Currently:
The mount command tells the kernel both the server name 'secure.server.name' and it's IP address 1.2.3.4 however rpc.gssd currently completely ignored the server name passed in by the mount command (see the 'dummy' variable) and instead uses the ip address to do a reverse lookup.
PTR lookup of 1.2.3.4 --- MITM ---> public.server.com
Then it uses the retrieved name (public..) to construct the GSSAPI name it wants to connect to, the result is that the name [email protected] is constructed instead of the requested [email protected] that should be requested.
So the client does ask for the right server but can be easily tricked into connecting to another one.
Kerberos itself *is* correctly identifying the server, it's rpc.gssd that opened it'sself to being fooled before involving Kerberos.
Quoted 2 lines
I would be in favor of requiring kerberized nfs clients to know the full name of the server they're trying to connect to.
Yes, this would be my preference too.
Quoted 4 lines
Kerberized nfs clients
should not be relying on dns for any part of their security. But I guess we haven't been doing that, and to make that a requirement now would be a compatibility problem.
Yes which is why I am adding a switch, I strongly suggest distribution to use -N on in newer releases and allow -N off only for existing setups for compatibility reasons.
Simo.
Signature · 1 line
Simo Sorce * Red Hat, Inc * New York
Simo Sorce wrote:
No this is not what I am saying.
The client uses the fully qualified name at the mount command but due to the use of a PTR lookup it is fooled into acquiring a ticket for the wrong server which will then happily allow you to connect.
Currently:
The mount command tells the kernel both the server name 'secure.server.name' and it's IP address 1.2.3.4 however rpc.gssd currently completely ignored the server name passed in by the mount command (see the 'dummy' variable) and instead uses the ip address to do a reverse lookup.
PTR lookup of 1.2.3.4 --- MITM ---> public.server.com
Then it uses the retrieved name (public..) to construct the GSSAPI name it wants to connect to, the result is that the name [email protected] is constructed instead of the requested [email protected] that should be requested.
So the client does ask for the right server but can be easily tricked into connecting to another one.
Kerberos itself *is* correctly identifying the server, it's rpc.gssd that opened it'sself to being fooled before involving Kerberos.
Ok, I think I understand now. I suggest adding some of that text into the commit log. And stop using the term "mitm". A mitm attack is used to convince both ends of a connection that they are talking to each other. DNS is not a mutually authenticated exchange.
This discussion is making me nervous. It makes me think no one has thought through the use of kerberos to authenticate nfs connections on linux.
Quoted 21 lines · Jim Rees
Simo Sorce wrote:
No this is not what I am saying.
The client uses the fully qualified name at the mount command but due to the use of a PTR lookup it is fooled into acquiring a ticket for the wrong server which will then happily allow you to connect.Currently:
The mount command tells the kernel both the server name 'secure.server.name' and it's IP address 1.2.3.4 however rpc.gssd currently completely ignored the server name passed in by the mount command (see the 'dummy' variable) and instead uses the ip address to do a reverse lookup.PTR lookup of 1.2.3.4 --- MITM ---> public.server.com
Then it uses the retrieved name (public..) to construct the GSSAPI name it wants to connect to, the result is that the name [email protected] is constructed instead of the requested [email protected] that should be requested.
Quoted 8 lines
So the client does ask for the right server but can be easily tricked into connecting to another one.
Kerberos itself *is* correctly identifying the server, it's rpc.gssd that opened it'sself to being fooled before involving Kerberos.
Ok, I think I understand now. I suggest adding some of that text into the commit log.
To be honest this is a well known problem in the kerberos community, I will see if I can find a public text that references this problem and add a link to it in the commit log ?
It's a bit of a long explanation otherwise ... maybe I can point at this thread in the archives ?
Quoted 3 lines
And stop using the term "mitm". A mitm attack is used to
convince both ends of a connection that they are talking to each other. DNS is not a mutually authenticated exchange.
Well it is still a sort of Man in the Middle, as you also have to redirect communications (nfsv4 uses TCP) for it to be effective, it is just not exploiting a crypto issue.
Quoted 2 lines
This discussion is making me nervous. It makes me think no one has thought through the use of kerberos to authenticate nfs connections on linux.
I think this is just an oversight, this 'bug' has been present since the original patches have been merged in.
Simo.
Signature · 1 line
Simo Sorce * Red Hat, Inc * New York
Quoted 3 lines · Simo Sorce
And stop using the term "mitm". A mitm attack is used to
convince both ends of a connection that they are talking to each other. DNS is not a mutually authenticated exchange.
Well it is still a sort of Man in the Middle, as you also have to redirect communications (nfsv4 uses TCP) for it to be effective, it is just not exploiting a crypto issue.
Now you've lost me again. I thought we were discussing dns. What does nfs have to do with it?
Quoted 6 lines · Jim Rees
Quoted 3 lines · Simo Sorce
And stop using the term "mitm". A mitm attack is used to
convince both ends of a connection that they are talking to each other. DNS is not a mutually authenticated exchange.
Quoted 6 lines
Well it is still a sort of Man in the Middle, as you also have to redirect communications (nfsv4 uses TCP) for it to be effective, it is just not exploiting a crypto issue.
Now you've lost me again. I thought we were discussing dns. What does nfs have to do with it?
It's complicated, but if you re-read the scenario I wrote and think how the rpcgss communication happens you should see it.
Simo.
Signature · 1 line
Simo Sorce * Red Hat, Inc * New York
Quoted 38 lines · Simo Sorce
Quoted 21 lines · Jim Rees
Simo Sorce wrote:
No this is not what I am saying.
The client uses the fully qualified name at the mount command but due to the use of a PTR lookup it is fooled into acquiring a ticket for the wrong server which will then happily allow you to connect.Currently:
The mount command tells the kernel both the server name 'secure.server.name' and it's IP address 1.2.3.4 however rpc.gssd currently completely ignored the server name passed in by the mount command (see the 'dummy' variable) and instead uses the ip address to do a reverse lookup.PTR lookup of 1.2.3.4 --- MITM ---> public.server.com
Then it uses the retrieved name (public..) to construct the GSSAPI name it wants to connect to, the result is that the name [email protected] is constructed instead of the requested [email protected] that should be requested.
Quoted 8 lines
So the client does ask for the right server but can be easily tricked into connecting to another one.
Kerberos itself *is* correctly identifying the server, it's rpc.gssd that opened it'sself to being fooled before involving Kerberos.
Ok, I think I understand now. I suggest adding some of that text into the commit log.
To be honest this is a well known problem in the kerberos community, I will see if I can find a public text that references this problem and add a link to it in the commit log ?
It's a bit of a long explanation otherwise ... maybe I can point at this thread in the archives ?
Quoted 7 lines
Quoted 3 lines
And stop using the term "mitm". A mitm attack is used to
convince both ends of a connection that they are talking to each other. DNS is not a mutually authenticated exchange.Well it is still a sort of Man in the Middle, as you also have to redirect communications (nfsv4 uses TCP) for it to be effective, it is just not exploiting a crypto issue.
Like Jim I don't understand how that fits the definition of a "man in the middle" attack.
I also wonder whether "GSSAPI name" is specialized jargon. And I don't understand what you mean by "tries to avoid" reverse lookups--does it use them or not? How about:
With this option, gssd expects the name given on the mount command line to be exactly the name that the server authenticates as. This option is strongly recommended.
Without this option, gssd will instead do a reverse lookup on the server's IP address to decide what name it should use. This is vulnerable to attacks on DNS, and is fragile in cases where reverse DNS (PTR) records are not set up correctly.
--b.
J. Bruce Fields wrote:
Like Jim I don't understand how that fits the definition of a "man in the middle" attack.
If Bruce is having trouble too, I think the explanation needs to be improved. But I think the latest patch set drops the talk about a vulnerability, and I'm fine with that.
Quoted 8 lines · Jim Rees
J. Bruce Fields wrote:
Like Jim I don't understand how that fits the definition of a "man in the middle" attack.
If Bruce is having trouble too, I think the explanation needs to be improved. But I think the latest patch set drops the talk about a vulnerability, and I'm fine with that.
Yeah I figured that just concentrating on the functional aspect of being able to work on networks with missing PTR records is sufficient to explain why this patchset is good :-)
However let me give a try at explaining in a different way using the same example I gave before but from a different angle.
- Assume user Alice has very important documents that are backed up daily to a server named secure.server.name where Alice can write to and nobody can read from.
- Assume an attacker Eve, that is in the condition of intercepting and modifying all network traffic from Alice.
- Assume there exist another server called public.server.name where Alice can write to and Eve can read from.
Eve wants to fool Alice in mounting /export from public.server.name instead or secure.server.name so she can copy Alice's documents away.
So Eve, manipulates traffic and when Alice's machine tries to talk to secure.server.name, Eve redirects all the traffic to public.server.name instead. (Can be done by a simple DNS poisoning attack that gives back to Alice computer the public.server.name IP address, or by actual DNAT on Alice's computer packets).
Now, normally, if Krb5 authentication is used the mount would fail, because Alice would try to authenticate to public.server.com using a ticket she obtained to talk to secure.server.com (remember that Alice thinks she is mounting the secure.server.name share).
Here the reverse DNS resolution performed by Alice's computer's rpc.gssd daemon is what allows Eve to successfully redirect Alice's mount instead. Because at the time Alice's computer needs to grab a ticket for the 'target' server, Eve fakes the DNS replies and tells Alice's computer that the destination name for the secure.server.name's IP address is 'public.server.name'.
So with current behavior Alice's computer asks the KDC for a ticket for 'public.server.name' instead of 'secure.server.name'.
Now Krb5 authentication does not fail anymore because Alice does have the 'right' ticket for the server it is connecting to.
Alice authentication against the 'substituted' (by Eve, through packet redirection) server therefore succeeds, and her computer is fooled to think it really mounted secure.server.name when it really has mounted public.server.name, the cron job for the backup starts and copies all files to public.server.name, where Eve can find and access them.
Call it MITM, call it something else, it's an attack, and we can prevent it.
HTH,
Simo.
Signature · 1 line
Simo Sorce * Red Hat, Inc * New York
Quoted 6 lines · Simo Sorce
Quoted 5 lines · Jim Rees
J. Bruce Fields wrote:
Like Jim I don't understand how that fits the definition of a "man in the middle" attack.
Quoted 7 lines
Quoted 3 lines
If Bruce is having trouble too, I think the explanation needs to be improved. But I think the latest patch set drops the talk about a vulnerability, and I'm fine with that.
Yeah I figured that just concentrating on the functional aspect of being able to work on networks with missing PTR records is sufficient to explain why this patchset is good :-)
Definitely, you're fine. I was just nitpicking language, the problem's clear enough:
Quoted 35 lines
However let me give a try at explaining in a different way using the same example I gave before but from a different angle.
- Assume user Alice has very important documents that are backed up daily to a server named secure.server.name where Alice can write to and nobody can read from.
- Assume an attacker Eve, that is in the condition of intercepting and modifying all network traffic from Alice.
- Assume there exist another server called public.server.name where Alice can write to and Eve can read from.Eve wants to fool Alice in mounting /export from public.server.name instead or secure.server.name so she can copy Alice's documents away.
So Eve, manipulates traffic and when Alice's machine tries to talk to secure.server.name, Eve redirects all the traffic to public.server.name instead. (Can be done by a simple DNS poisoning attack that gives back to Alice computer the public.server.name IP address, or by actual DNAT on Alice's computer packets).
Now, normally, if Krb5 authentication is used the mount would fail, because Alice would try to authenticate to public.server.com using a ticket she obtained to talk to secure.server.com (remember that Alice thinks she is mounting the secure.server.name share).Here the reverse DNS resolution performed by Alice's computer's rpc.gssd daemon is what allows Eve to successfully redirect Alice's mount instead. Because at the time Alice's computer needs to grab a ticket for the 'target' server, Eve fakes the DNS replies and tells Alice's computer that the destination name for the secure.server.name's IP address is 'public.server.name'.
So with current behavior Alice's computer asks the KDC for a ticket for 'public.server.name' instead of 'secure.server.name'.
Quoted 8 lines
Now Krb5 authentication does not fail anymore because Alice does have the 'right' ticket for the server it is connecting to.
Alice authentication against the 'substituted' (by Eve, through packet redirection) server therefore succeeds, and her computer is fooled to think it really mounted secure.server.name when it really has mounted public.server.name, the cron job for the backup starts and copies all files to public.server.name, where Eve can find and access them.
Argh, sorry Simo, I didn't mean to make you write all that out.
Quoted 1 line
Call it MITM, call it something else,
This was literally my only complaint. The above doesn't sound to me like what I'd normally call a "man in the middle" attack.
--b.
Quoted 56 lines · J. Bruce Fields
Quoted 6 lines · Simo Sorce
Quoted 5 lines · Jim Rees
J. Bruce Fields wrote:
Like Jim I don't understand how that fits the definition of a "man in the middle" attack.
Quoted 7 lines
Quoted 3 lines
If Bruce is having trouble too, I think the explanation needs to be improved. But I think the latest patch set drops the talk about a vulnerability, and I'm fine with that.
Yeah I figured that just concentrating on the functional aspect of being able to work on networks with missing PTR records is sufficient to explain why this patchset is good :-)
Definitely, you're fine. I was just nitpicking language, the problem's clear enough:
Quoted 38 lines
However let me give a try at explaining in a different way using the same example I gave before but from a different angle.
- Assume user Alice has very important documents that are backed up daily to a server named secure.server.name where Alice can write to and nobody can read from.
- Assume an attacker Eve, that is in the condition of intercepting and modifying all network traffic from Alice.
- Assume there exist another server called public.server.name where Alice can write to and Eve can read from.Eve wants to fool Alice in mounting /export from public.server.name instead or secure.server.name so she can copy Alice's documents away.
So Eve, manipulates traffic and when Alice's machine tries to talk to secure.server.name, Eve redirects all the traffic to public.server.name instead. (Can be done by a simple DNS poisoning attack that gives back to Alice computer the public.server.name IP address, or by actual DNAT on Alice's computer packets).
Now, normally, if Krb5 authentication is used the mount would fail, because Alice would try to authenticate to public.server.com using a ticket she obtained to talk to secure.server.com (remember that Alice thinks she is mounting the secure.server.name share).Here the reverse DNS resolution performed by Alice's computer's rpc.gssd daemon is what allows Eve to successfully redirect Alice's mount instead. Because at the time Alice's computer needs to grab a ticket for the 'target' server, Eve fakes the DNS replies and tells Alice's computer that the destination name for the secure.server.name's IP address is 'public.server.name'.
So with current behavior Alice's computer asks the KDC for a ticket for 'public.server.name' instead of 'secure.server.name'.
Now Krb5 authentication does not fail anymore because Alice does have the 'right' ticket for the server it is connecting to.
Quoted 7 lines
Quoted 5 lines
Alice authentication against the 'substituted' (by Eve, through packet redirection) server therefore succeeds, and her computer is fooled to think it really mounted secure.server.name when it really has mounted public.server.name, the cron job for the backup starts and copies all files to public.server.name, where Eve can find and access them.
Argh, sorry Simo, I didn't mean to make you write all that out.
It's ok, I am planning to make it into a blog post, so next time someone asks about this particular issue with PTR records and GSSAPI I can point them there. So it is not worthless, I call it a draft :-)
Quoted 4 lines
Quoted 1 line
Call it MITM, call it something else,
This was literally my only complaint. The above doesn't sound to me like what I'd normally call a "man in the middle" attack.
Yup, ok.
So if anyone can ack the actual patches (or tell me if there is anything else to change I'd love to move forward, I have other patches in the pipeline :-)
Simo.
Signature · 1 line
Simo Sorce * Red Hat, Inc * New York
Quoted 5 lines · Simo Sorce
Yup, ok.
So if anyone can ack the actual patches (or tell me if there is anything else to change I'd love to move forward, I have other patches in the pipeline :-)
Look fine to me; I didn't have any objection outside documentation.
I've lost track of which are the latest patches, honestly--hopefully Steve's been keeping track, it's up to him to merge them.
--b.
Quoted 23 lines · Simo Sorce
The attached patch adds a new command line switch to rpc.gssd to avoid PTR resolution when possible.
The current code *depends* on PTR resolution for GSSAPI authentication and this is *bad*.
It imposes an annoying, and unnecessary, constraint on the correctness of DNS resolution, which prevents mounts from working in networks where the PTR record cannot be easily controlled (for example networks where the forward name is reasonable while the PTR is set to some artificial name based on the IP address or so that is not the canonical name or where no PTR exist at all).Depending on PTR resolution for GSSAPI is also very bad practice because it opens up DNS spoofing attacks where an attacker can try to redirect a user to the wrong server fooling mutual authentication, and induce a user to trust improper data or disclose (by copying on the impostor server) data that should be confidential.
Ideally people will always use the -N switch, however I added an actual switch to avoid breaking existing configuration where someone may intentionally or inadvertently be using multiple A names instead of using CNAMEs but relying on PTR records to find the "canonical name".
Quoted 7 lines
It would be probably nice to switch the behavior on by default in future.
Jeff in CC as he also implemented a similar switch for cifs.upcall (there it is called -t|--trust-dns, but in rpc.gsssd -t is already taken ...).
A nits..
Quoted 4 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Please do not post patches as attachments. In-lining them in the email makes it easier to review...
From
53dade1e5d22eee06a08ace2684257292514cb49Mon Sep 17 00:00:00 2001
From: Simo Sorce <[email protected]>
Date: Mon, 1 Apr 2013 21:00:55 -0400
Subject: [PATCH] Avoid forced reverse resolution for server name
A NFS client should be able to work properly even if the PTR record for the server is not set. there is no excuse to forcefully prevent that from working when it can.
So, use some heuristics to see if the server is a FQDN or an IP address. If it is an IP address or a shortname (no '.' in name) then do a PTR lookup, otherwise avoid it.
This is enabled by the new -N flag which is off by default for now, ideally we will turn this on by default after a transition period.
--- utils/gssd/gss_util.h | 2 ++ utils/gssd/gssd.c | 4 +++- utils/gssd/gssd.man | 7 ++++++- utils/gssd/gssd_proc.c | 31 +++++++++++++++++++++++++++---- 4 files changed, 38 insertions(+), 6 deletions(-)
utils/gssd/gss_util.h +2 -0
@@ -52,4 +52,6 @@ int gssd_check_mechs(void); gss_krb5_set_allowable_enctypes(min, cred, num, types)#endif +extern int avoid_ptr; +#endif /* _GSS_UTIL_H_ */
utils/gssd/gssd.c +1 -1
@@ -85,7 +85,7 @@ sig_hup(int signal)static void usage (char *progname) { - fprintf(stderr, "usage: %s [-f] [-l] [-M] [-n] [-v] [-r] [-p pipefsdir] [-k keytab] [-d ccachedir] [-t timeout] [-R preferred realm]\n", + fprintf(stderr, "usage: %s [-f] [-l] [-M] [-n] [-v] [-r] [-p pipefsdir] [-k keytab] [-d ccachedir] [-t timeout] [-R preferred realm] [-N]\n",
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Could please move the [-N] in front of [-n]... At lease that that point in the list things are in alphabetical order
progname);
exit(1);
}
@@ -150,6 +150,8 @@ main(int argc, char *argv[])
errx(1, "Encryption type limits not supported by Kerberos libraries.");
#endif
break;
+ case 'N':
+ avoid_ptr = 1;
default:
usage(argv[0]);
break;
utils/gssd/gssd.man +1 -1
@@ -8,7 +8,7 @@ rpc.gssd \- RPCSEC_GSS daemon .SH SYNOPSIS .B rpc.gssd -.RB [ \-fMnlvr ] +.RB [ \-fMnlvrN ]
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Same as above....
.RB [ \-k
.IR keytab ]
.RB [ \-p
@@ -266,6 +266,11 @@ new kernel contexts to be negotiated after
seconds, which allows changing Kerberos tickets and identities frequently. The default is no explicit timeout, which means the kernel context will live the lifetime of the Kerberos service ticket used in its creation.
+.TP
+.B -N
+Tries to avoid PTR lookups for resolving the server name, if the name used at +mount looks like a Fully Qualified Domain Name. This is import to avoid GSSAPI +issues when PTR records are set and to help avoid some kind of MITM attack.
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
I'm sorry this does not make much sense. Can we use more meaningful terms like DNS lookups or DNS PTR lookups... something not so condensed
Quoted 2 lines
Also, could you please talk about who will be needing this new flag and how will they know they needed?
.SH SEE ALSO
.BR rpc.svcgssd (8),
.BR kerberos (1),
utils/gssd/gssd_proc.c +24 -1
@@ -67,6 +67,7 @@#include <errno.h> #include <gssapi/gssapi.h> #include <netdb.h> +#include <ctype.h> #include "gssd.h" #include "err_util.h"
@@ -107,6 +108,8 @@ struct pollfd * pollarray;unsigned long pollsize;/* the size of pollaray (in pollfd's) */ +int avoid_ptr =0 ; +/* * convert a presentation address string to a sockaddr_storage struct. Returns * true on success or false on failure.
@@ -165,12 +168,32 @@ addrstr_to_sockaddr(struct sockaddr *sa, const char *node, const char *port)* convert a sockaddr to a hostname */ static char * -sockaddr_to_hostname (const struct sockaddr *sa,const char *addr) +get_servername (const char *name,const struct sockaddr *sa,const char *addr) {socklen_t addrlen;int err;char *hostname;char hbuf[NI_MAXHOST ]; +const char *p; +int do_ptr_lookup =0 ; + +if (avoid_ptr) { +/* try to determine if this is FQDN, or an IP address with +* basic heuristics, if they fail try a PTR lookup */ + p =strrchr (name,'.' ); +if (!p ||strchr (name,':' )) + do_ptr_lookup =1 ;/* non-FQDN, or IPv6 */ +else { +for (++p; *p; p++) +if (!isdigit (*p)) +break ; +if (!*p) + do_ptr_lookup =1 ;/* IPv4 */ + } +if (!do_ptr_lookup) { +return strdup (name); + } + }
Is this "reinventing the wheel"? ;-) Don't we already have a why of determining a FQDN? Maybe not...
steved.
switch (sa->sa_family) {
case AF_INET:
@@ -208,7 +231,7 @@ read_service_info(char *info_file_name, char **servicename, char **servername,
struct sockaddr *addr) {
#define INFOBUFLEN 256
char buf[INFOBUFLEN + 1];
- static char dummy[128];
+ static char server[128];
int nbytes;
static char service[128];
static char address[128];
@@ -236,7 +259,7 @@ read_service_info(char *info_file_name, char **servicename, char **servername,
"service: %127s %15s version %15s\n"
"address: %127s\n"
"protocol: %15s\n",
- dummy,
+ server,
service, program, version,
address,
protoname);
@@ -258,7 +281,7 @@ read_service_info(char *info_file_name, char **servicename, char **servername,
if (!addrstr_to_sockaddr(addr, address, port))
goto fail;
- *servername = sockaddr_to_hostname(addr, address);
+ *servername = get_servername(server, addr, address);
if (*servername == NULL)
goto fail;
Signature · 1 line
1.8.1.4
Quoted 179 lines · Steve Dickson
Quoted 23 lines · Simo Sorce
The attached patch adds a new command line switch to rpc.gssd to avoid PTR resolution when possible.
The current code *depends* on PTR resolution for GSSAPI authentication and this is *bad*.
It imposes an annoying, and unnecessary, constraint on the correctness of DNS resolution, which prevents mounts from working in networks where the PTR record cannot be easily controlled (for example networks where the forward name is reasonable while the PTR is set to some artificial name based on the IP address or so that is not the canonical name or where no PTR exist at all).Depending on PTR resolution for GSSAPI is also very bad practice because it opens up DNS spoofing attacks where an attacker can try to redirect a user to the wrong server fooling mutual authentication, and induce a user to trust improper data or disclose (by copying on the impostor server) data that should be confidential.
Ideally people will always use the -N switch, however I added an actual switch to avoid breaking existing configuration where someone may intentionally or inadvertently be using multiple A names instead of using CNAMEs but relying on PTR records to find the "canonical name".
Quoted 7 lines
It would be probably nice to switch the behavior on by default in future.
Jeff in CC as he also implemented a similar switch for cifs.upcall (there it is called -t|--trust-dns, but in rpc.gsssd -t is already taken ...).
A nits..
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Please do not post patches as attachments. In-lining them in the email makes it easier to review...
From
53dade1e5d22eee06a08ace2684257292514cb49Mon Sep 17 00:00:00 2001
From: Simo Sorce <[email protected]>
Date: Mon, 1 Apr 2013 21:00:55 -0400
Subject: [PATCH] Avoid forced reverse resolution for server nameA NFS client should be able to work properly even if the PTR record for the server is not set. there is no excuse to forcefully prevent that from working when it can.
So, use some heuristics to see if the server is a FQDN or an IP address. If it is an IP address or a shortname (no '.' in name) then do a PTR lookup, otherwise avoid it.This is enabled by the new -N flag which is off by default for now, ideally we will turn this on by default after a transition period.
---utils/gssd/gss_util.h | 2 ++ utils/gssd/gssd.c | 4 +++- utils/gssd/gssd.man | 7 ++++++- utils/gssd/gssd_proc.c | 31 +++++++++++++++++++++++++++---- 4 files changed, 38 insertions(+), 6 deletions(-)utils/gssd/gss_util.h +2
-0@@ -52,4 +52,6 @@ int gssd_check_mechs(void); gss_krb5_set_allowable_enctypes(min, cred, num, types)#endif +extern int avoid_ptr; +#endif /* _GSS_UTIL_H_ */ utils/gssd/gssd.c +0
-0@@ -85,7 +85,7 @@ sig_hup(int signal)static void
usage(char *progname)
{utils/gssd/gssd.c +1
-1- fprintf(stderr, "usage: %s [-f] [-l] [-M] [-n] [-v] [-r] [-p pipefsdir] [-k keytab] [-d ccachedir] [-t timeout] [-R preferred realm]\n", + fprintf(stderr, "usage: %s [-f] [-l] [-M] [-n] [-v] [-r] [-p pipefsdir] [-k keytab] [-d ccachedir] [-t timeout] [-R preferred realm] [-N]\n",Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Could please move the [-N] in front of [-n]... At lease that that point in the list things are in alphabetical order
progname);exit(1);
}utils/gssd/gssd.c +2
-0@@ -150,6 +150,8 @@ main(int argc, char *argv[])errx (1 ,"Encryption type limits not supported by Kerberos libraries." );#endif break ; +case 'N' : + avoid_ptr =1 ;default :usage (argv[0 ]);break ;utils/gssd/gssd.man +0
-0@@ -8,7 +8,7 @@rpc.gssd \-
RPCSEC_GSSdaemon
.SH SYNOPSIS
.B rpc.gssd
-.RB [ \-fMnlvr ]
+.RB [ \-fMnlvrN ]Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Same as above....
.RB [ \-k
.IR keytab ]
.RB [ \-putils/gssd/gssd.man +0
-0@@ -266,6 +266,11 @@ new kernel contexts to be negotiated afterseconds, which allows changing Kerberos tickets and identities frequently. The default is no explicit timeout, which means the kernel context will live the lifetime of the Kerberos service ticket used in its creation. +.TP
+.B -Nutils/gssd/gssd.man +3
-0+Tries to avoid PTR lookups for resolving the server name, if the name used at +mount looks like a Fully Qualified Domain Name. This is import to avoid GSSAPI +issues when PTR records are set and to help avoid some kind of MITM attack.Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
I'm sorry this does not make much sense. Can we use more meaningful terms like DNS lookups or DNS PTR lookups... something not so condensed
Quoted 2 lines
Also, could you please talk about who will be needing this new flag and how will they know they needed?
.SH SEE ALSO
.BR rpc.svcgssd (8),
.BR kerberos (1),utils/gssd/gssd_proc.c +1
-0@@ -67,6 +67,7 @@#include <errno.h> #include <gssapi/gssapi.h> #include <netdb.h> +#include <ctype.h> #include "gssd.h" #include "err_util.h" @@ -107,6 +108,8 @@ struct pollfd * pollarray;unsigned long pollsize; /* the size of pollaray (in pollfd's) */
utils/gssd/gssd_proc.c +2
-0+int avoid_ptr =0 ; +/*
* convert a presentation address string to asockaddr_storagestruct. Returns
* true on success or false on failure.utils/gssd/gssd_proc.c +0
-0@@ -165,12 +168,32 @@ addrstr_to_sockaddr(struct sockaddr *sa, const char *node, const char *port)* convert a sockaddr to a hostname */ static char *
utils/gssd/gssd_proc.c +1
-1-sockaddr_to_hostname(const struct sockaddr *sa, const char *addr) +get_servername(const char *name, const struct sockaddr *sa, const char *addr){ socklen_t addrlen; int err; char *hostname; char hbuf[NI_MAXHOST];utils/gssd/gssd_proc.c +20
-0+const char *p; +int do_ptr_lookup =0 ; + +if (avoid_ptr) { +/* try to determine if this is FQDN, or an IP address with +* basic heuristics, if they fail try a PTR lookup */ + p =strrchr (name,'.' ); +if (!p ||strchr (name,':' )) + do_ptr_lookup =1 ;/* non-FQDN, or IPv6 */ +else { +for (++p; *p; p++) +if (!isdigit (*p)) +break ; +if (!*p) + do_ptr_lookup =1 ;/* IPv4 */ + } +if (!do_ptr_lookup) { +return strdup (name); + } + }
Quoted 2 lines
Is this "reinventing the wheel"? ;-) Don't we already have a why of determining a FQDN? Maybe not...
Yes, we do. Have a look in support/export/hostname.c for tools and helpers for dealing with IP addresses properly.
Quoted 43 lines
steved.
switch (sa->sa_family) { case AF_INET:utils/gssd/gssd_proc.c +3
-3@@ -208,7 +231,7 @@ read_service_info(char *info_file_name, char **servicename, char **servername, struct sockaddr *addr) {#define INFOBUFLEN 256char buf[INFOBUFLEN +1 ]; -static char dummy[128 ]; +static char server[128 ];int nbytes;static char service[128 ];static char address[128 ];@@ -236,7 +259,7 @@ read_service_info(char *info_file_name, char **servicename, char **servername,"service: %127s %15s version %15s\n" "address: %127s\n" "protocol: %15s\n" , - dummy, + server, service, program, version, address, protoname);@@ -258,7 +281,7 @@ read_service_info(char *info_file_name, char **servicename, char **servername,if (!addrstr_to_sockaddr (addr, address, port)) goto fail; - *servername =sockaddr_to_hostname (addr, address); + *servername =get_servername (server, addr, address);if (*servername ==NULL ) goto fail;Signature · 8 lines
1.8.1.4 -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
Quoted 7 lines · Chuck Lever
Quoted 1 line
Is this "reinventing the wheel"? ;-) Don't we already have a why of
determining a FQDN?
Quoted 1 line
Maybe not...
Yes, we do. Have a look in
support/export/hostname.cfor tools and helpers for dealing with IP addresses properly.
The only function I see that may fit the bill is host_addrinfo() which calls getaddrinfo() which I avoided because gssapi will alraedy do that.
Maybe it is ok to only check by negatives ?
IE check if the string is an ipv4 or an ipv6 address and if not just pass the string in as is ?
Simo.
Signature · 1 line
Simo Sorce * Red Hat, Inc * New York
Quoted 10 lines · Simo Sorce
Quoted 7 lines · Chuck Lever
Quoted 1 line
Is this "reinventing the wheel"? ;-) Don't we already have a why of
determining a FQDN?
Quoted 1 line
Maybe not...
Yes, we do. Have a look in
support/export/hostname.cfor tools and helpers for dealing with IP addresses properly.The only function I see that may fit the bill is
host_addrinfo()which callsgetaddrinfo()which I avoided because gssapi will alraedy do that.
That looks like the right way to do it. It may be reasonable to assume our resolver caches results, so a second call may not be much of a latency hit.
But AI_CANONNAME may rely on the target host having a PTR record.
Quoted 3 lines
Maybe it is ok to only check by negatives ?
IE check if the string is an ipv4 or an ipv6 address and if not just pass the string in as is ?
IIRC, the pton() function fails if your string does not contain a presentation address. But that allows unqualified domain labels, doesn't it?
--
Chuck Lever
chuck[dot]lever[at]oracle[dot]com
Quoted 10 lines · Chuck Lever
Quoted 8 lines · Simo Sorce
Quoted 7 lines · Chuck Lever
Quoted 1 line
Is this "reinventing the wheel"? ;-) Don't we already have a why of
determining a FQDN?
Quoted 1 line
Maybe not...
Yes, we do. Have a look in
support/export/hostname.cfor tools and helpers for dealing with IP addresses properly.
Quoted 6 lines
Quoted 2 lines
The only function I see that may fit the bill is
host_addrinfo()which callsgetaddrinfo()which I avoided because gssapi will alraedy do that.That looks like the right way to do it. It may be reasonable to assume our resolver caches results, so a second call may not be much of a latency hit.
But
AI_CANONNAMEmay rely on the target host having a PTR record.
No, I already filed bugs against glibc in the past where it happened by mistake that a PTR request was done with some flag combinations. Some distributions have patches to fix that as getaddrinfo should exclusively deal with CNAME -> A name "canonicalization" and nothing else. Using a PTR request is a bug.
However I already addressed the fact I cannot use any function in that file from gssd_proc.c so I simply used inet_pton directly, works fine and according to comments in host_pton it is also required even if getaddrinfo was used, and sufficient for my purposes.
Quoted 5 lines
Quoted 3 lines
Maybe it is ok to only check by negatives ?
IE check if the string is an ipv4 or an ipv6 address and if not just pass the string in as is ?IIRC, the
pton()function fails if your string does not contain a presentation address.
This is fine.
Quoted 1 line
But that allows unqualified domain labels, doesn't it?
GSSAPI will call getaddrinfo() anyway so if a shortname is used it will be canonicalized as long as the local resolver can resolve it into a fqdn.
If it can't though luck, either use a fqdn on the mount command or do not enable this feature (which is why it is off by default and selectable in the last patchset).
Simo.
Signature · 1 line
Simo Sorce * Red Hat, Inc * New York