Skip to contentSkip to conversation

Bug#583435: rpcbind: Insecure handling of state files

DISCUSSION Aníbal Monsalve Salazar 3 participants 6 messages Jun 2, 2010 – Jun 3, 2010 5 messages not in the archive

Subsystems (2)
lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

#2 lore raw
Quoted 41 lines · Guillem Jover

Hi!

Quoted 10 lines · Guillem Jover

Package: rpcbind
Version: 0.2.0-4
Severity: serious
Tags: security

The rpcbind daemon, which runs as root, uses /tmp/portmap.xdr and /tmp/rpcbind.xdr for doing warm starts as what seems to be a way to preserve state between invokations. It parses (through libtirpc) and removes them on start. It creates them before exiting.

Quoted 4 lines

So first off, *any* user can craft those two files before the daemon has started for the first time, which the daemon will parse. This might be ok, depending on the checks done on parse, I'd still be very wary of letting a user be able to craft such files at will.

It seems to be doing no checks whatsoever. A simple test I performed at the time of filing this report, but didn't seem to have any obvious consequence, shows this which I noticed later on:

,---

gaara:~# /etc/init.d/rpcbind start
Starting rpcbind daemon....
gaara:~# ps axuOp|egrep '(^USER|[r]pcbind)'
USER       PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
root     23424  0.0  0.0  18768   704 ?        Ss   13:53   0:00 /sbin/rpcbind -w
gaara:~# /etc/init.d/rpcbind stop
Stopping rpcbind daemon....
gaara:~# dd if=/dev/urandom of=/tmp/rpcbind.xdr bs=1024 count=1
1+0 records in
1+0 records out
1024 bytes (1,0 kB) copied, 0,000861307 s, 1,2 MB/s
gaara:~# /etc/init.d/rpcbind start
Starting rpcbind daemon....
gaara:~# ps axuOp|egrep '(^USER|[r]pcbind)'
USER       PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
root     23440  0.0  0.0 4008972  772 ?        Ss   13:54   0:00 /sbin/rpcbind -w
`---
Quoted 7 lines

The first start is a normal clean invokation, the second one is using the crafted file. See how it has allocated almost 4 GiB. Disregard though, me running all this as root, a user would be able to craft those files as long as they were not already in /tmp.

thanks,
guillem

I'm sending this bug report to the linux-nfs mailing list.

The original bug report is at http://bugs.debian.org/583435

Thank you.

A message that is not in the archive (sent off-list or not yet received): 20100602112520.GA22639-q3oZDYqQg6zyUObV3Cmqeti2O/[email protected]

#4 lore raw
Quoted 4 lines · An=EDbal Monsalve Salazar
Quoted 3 lines · Guillem Jover

Hi!

On Thu, 2010-05-27 at 19:09:08 +0200, Guillem Jover wrote:

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

Package: rpcbind
Version: 0.2.0-4
Severity: serious
Tags: security

The rpcbind daemon, which runs as root, uses /tmp/portmap.xdr and /tmp/rpcbind.xdr for doing warm starts as what seems to be a way to preserve state between invokations. It parses (through libtirpc) an=

d

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

removes them on start. It creates them before exiting.

So first off, *any* user can craft those two files before the daemo=

n

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

has started for the first time, which the daemon will parse. This might be ok, depending on the checks done on parse, I'd still be ve=

ry

Quoted 3 lines
Quoted 3 lines
Quoted 1 line

wary of letting a user be able to craft such files at will.

It seems to be doing no checks whatsoever. A simple test I performed=

at

Quoted 8 lines
Quoted 8 lines

the time of filing this report, but didn't seem to have any obvious consequence, shows this which I noticed later on:

,---

gaara:~# /etc/init.d/rpcbind start
Starting rpcbind daemon....
gaara:~# ps axuOp|egrep '(^USER|[r]pcbind)'
USER       PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COM=

MAND

Quoted 1 line
Quoted 1 line

root 23424 0.0 0.0 18768 704 ? Ss 13:53 0:00 /sb=

in/rpcbind -w

Quoted 3 lines
Quoted 3 lines
gaara:~# /etc/init.d/rpcbind stop
Stopping rpcbind daemon....
gaara:~# dd if=3D/dev/urandom of=3D/tmp/rpcbind.xdr bs=3D1024 count=3D=

1

Quoted 7 lines
Quoted 7 lines

1+0 records in
1+0 records out
1024 bytes (1,0 kB) copied, 0,000861307 s, 1,2 MB/s

gaara:~# /etc/init.d/rpcbind start
Starting rpcbind daemon....
gaara:~# ps axuOp|egrep '(^USER|[r]pcbind)'
USER       PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COM=

MAND

Quoted 1 line
Quoted 1 line

root 23440 0.0 0.0 4008972 772 ? Ss 13:54 0:00 /sb=

in/rpcbind -w

Quoted 3 lines
Quoted 3 lines

`---

The first start is a normal clean invokation, the second one is usin=

g

Quoted 1 line
Quoted 1 line

the crafted file. See how it has allocated almost 4 GiB. Disregard t=

hough,

Quoted 1 line
Quoted 1 line

me running all this as root, a user would be able to craft those fil=

es as

Quoted 8 lines
Quoted 4 lines

long as they were not already in /tmp.

thanks,
guillem

I'm sending this bug report to the linux-nfs mailing list.

The original bug report is at http://bugs.debian.org/583435

Would /var/run (or a subdirectory of it) be a better choice than /tmp ?

#5 lore raw

Hi!

Quoted 10 lines · Chuck Lever
Quoted 9 lines · On 06/ 2/10 07:25 AM, An=C3=ADbal Monsalve Salazar wrote:
Quoted 8 lines · Guillem Jover
Quoted 7 lines · Guillem Jover

Package: rpcbind
Version: 0.2.0-4
Severity: serious
Tags: security

The rpcbind daemon, which runs as root, uses /tmp/portmap.xdr and /tmp/rpcbind.xdr for doing warm starts as what seems to be a way =

to

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

preserve state between invokations. It parses (through libtirpc) =

and

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

removes them on start. It creates them before exiting.

So first off, *any* user can craft those two files before the dae=

mon

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

has started for the first time, which the daemon will parse. This might be ok, depending on the checks done on parse, I'd still be =

very

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 1 line

wary of letting a user be able to craft such files at will.

It seems to be doing no checks whatsoever. A simple test I performe=

d at

Quoted 4 lines
Quoted 4 lines
Quoted 2 lines

the time of filing this report, but didn't seem to have any obvious consequence, shows this which I noticed later on:

The original bug report is at http://bugs.debian.org/583435

I'm adding here part of the initial mail that I trimmed when replying to myself:

,---
The second problem is that those files get created by the daemon on shutdown, and they *do* follow symlinks. So a user can drop two symlinks
there while the daemon is running and overwrite any file on the file system on shutdown.

The fix would consist of passing to configure something like =E2=80=9C--with-statedir=3D/var/cache/rpcbind=E2=80=9D, and make sure the=
daemon creates
such directory if missing on exit in src/warmstart.c:write_struct(), which it does not seem to be doing currently.

In addition it would be wise to notify upstream to change the default statedir to something else than /tmp.
`---

Quoted 1 line

Would /var/run (or a subdirectory of it) be a better choice than /tmp ?

/var/run might not be preserved across reboots, but regardless of that I think /var/cache is a better fit, it's internal state, but it's used to speed up start up time, and can be removed w/o ill effects.

regards,
guillem

A message that is not in the archive (sent off-list or not yet received): 20100603202743.GA6643-v62vTE6/wQGgM1MOaoewpti2O/[email protected]

#7 lore raw
Quoted 13 lines · Guillem Jover

Hi!

Quoted 10 lines · Chuck Lever
Quoted 9 lines · On 06/ 2/10 07:25 AM, An=C3=ADbal Monsalve Salazar wrote:
Quoted 8 lines · Guillem Jover
Quoted 7 lines · Guillem Jover

Package: rpcbind
Version: 0.2.0-4
Severity: serious
Tags: security

The rpcbind daemon, which runs as root, uses /tmp/portmap.xdr and /tmp/rpcbind.xdr for doing warm starts as what seems to be a way =

to

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

preserve state between invokations. It parses (through libtirpc) =

and

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

removes them on start. It creates them before exiting.

So first off, *any* user can craft those two files before the dae=

mon

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

has started for the first time, which the daemon will parse. This might be ok, depending on the checks done on parse, I'd still be =

very

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 1 line

wary of letting a user be able to craft such files at will.

It seems to be doing no checks whatsoever. A simple test I perform=

ed at

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

the time of filing this report, but didn't seem to have any obviou=

s

Quoted 14 lines
Quoted 3 lines
Quoted 3 lines
Quoted 1 line

consequence, shows this which I noticed later on:

The original bug report is at http://bugs.debian.org/583435

I'm adding here part of the initial mail that I trimmed when replying to myself:

,---
The second problem is that those files get created by the daemon on shutdown, and they *do* follow symlinks. So a user can drop two symlinks
there while the daemon is running and overwrite any file on the file system on shutdown.

Quoted 2 lines

The fix would consist of passing to configure something like =E2=80=9C--with-statedir=3D/var/cache/rpcbind=E2=80=9D, and make sure=

the daemon creates

Quoted 6 lines

such directory if missing on exit in src/warmstart.c:write_struct(), which it does not seem to be doing currently.

In addition it would be wise to notify upstream to change the default statedir to something else than /tmp.
`---

Agree changing the upstream default is a good idea.

Generally, that kind of directory is created as part of installation=20 (like, by rpm --install) rather than by the daemon itself.

Quoted 1 line
Quoted 1 line

Would /var/run (or a subdirectory of it) be a better choice than /tm=

p ?

Quoted 2 lines

/var/run might not be preserved across reboots, but regardless of tha=

t I

Quoted 2 lines

think /var/cache is a better fit, it's internal state, but it's used to speed up start up time, and can be removed w/o ill effects.

No, it's not intended to speed start up.

The cache files aren't really supposed to be retained over a reboot.=20 After a system restart, all of the RPC services will restart and=20 register themselves again. If just rpcbind restarts, all that=20 registration state is lost, so that's the point of saving it in a file.

I don't have a preference wrt /var/run or /var/cache.

#8 lore raw
Quoted 9 lines · Chuck Lever
Quoted 8 lines · Guillem Jover

The second problem is that those files get created by the daemon on shutdown, and they *do* follow symlinks. So a user can drop two symlinks
there while the daemon is running and overwrite any file on the file system on shutdown.

The fix would consist of passing to configure something like =E2=80=9C--with-statedir=3D/var/cache/rpcbind=E2=80=9D, and make sur=

e the daemon creates

Quoted 4 lines
Quoted 4 lines

such directory if missing on exit in src/warmstart.c:write_struct(), which it does not seem to be doing currently.

In addition it would be wise to notify upstream to change the defaul=

t

Quoted 6 lines
Quoted 1 line

statedir to something else than /tmp.

=20
Agree changing the upstream default is a good idea.
=20
Generally, that kind of directory is created as part of installation (like, by rpm --install) rather than by the daemon itself.

At least for /var/run I think it's common for systems to mount it as tmpfs, so the directories might not be there on boot. But those can always be created by the init script (or equivalent), it might be a problem if run from inetd though.

Quoted 1 line
Quoted 1 line
Quoted 1 line

Would /var/run (or a subdirectory of it) be a better choice than /t=

mp ?

Quoted 2 lines
Quoted 2 lines

/var/run might not be preserved across reboots, but regardless of th=

at I

Quoted 10 lines
Quoted 2 lines

think /var/cache is a better fit, it's internal state, but it's used to speed up start up time, and can be removed w/o ill effects.

=20
No, it's not intended to speed start up.
=20
The cache files aren't really supposed to be retained over a reboot. After a system restart, all of the RPC services will restart and register themselves again. If just rpcbind restarts, all that registration state is lost, so that's the point of saving it in a file.

Ah, yeah that makes more sense! More so given the configure option, I should have written "AFAIS" or something like that. :)

Quoted 1 line

I don't have a preference wrt /var/run or /var/cache.

So given that this is actually run time state, /var/run seems more appropriate, indeed.

regards,
guillem

A message that is not in the archive (sent off-list or not yet received): 20100603210707.GA7377-v62vTE6/wQGgM1MOaoewpti2O/[email protected]

#10 lore raw
Quoted 6 lines · Guillem Jover
Quoted 5 lines · Chuck Lever
Quoted 4 lines · Guillem Jover

The second problem is that those files get created by the daemon on shutdown, and they *do* follow symlinks. So a user can drop two symlinks
there while the daemon is running and overwrite any file on the fil=

e

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

system on shutdown.

The fix would consist of passing to configure something like =E2=80=9C--with-statedir=3D/var/cache/rpcbind=E2=80=9D, and make su=

re the daemon creates

Quoted 1 line
Quoted 1 line
Quoted 1 line

such directory if missing on exit in src/warmstart.c:write_struct()=

,

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

which it does not seem to be doing currently.

In addition it would be wise to notify upstream to change the defau=

lt

Quoted 9 lines
Quoted 6 lines
Quoted 1 line

statedir to something else than /tmp.

Agree changing the upstream default is a good idea.

Generally, that kind of directory is created as part of installation (like, by rpm --install) rather than by the daemon itself.

At least for /var/run I think it's common for systems to mount it as tmpfs, so the directories might not be there on boot. But those ca=

n

Quoted 2 lines

always be created by the init script (or equivalent), it might be a problem if run from inetd though.

Sure, that makes sense. Having the daemon create the directory also=20 means there are fewer ways distributors can get this wrong.

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

Would /var/run (or a subdirectory of it) be a better choice than /=

tmp ?

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

/var/run might not be preserved across reboots, but regardless of t=

hat I

Quoted 1 line
Quoted 1 line
Quoted 1 line

think /var/cache is a better fit, it's internal state, but it's use=

d

Quoted 17 lines
Quoted 9 lines
Quoted 1 line

to speed up start up time, and can be removed w/o ill effects.

No, it's not intended to speed start up.

The cache files aren't really supposed to be retained over a reboot. After a system restart, all of the RPC services will restart and register themselves again. If just rpcbind restarts, all that registration state is lost, so that's the point of saving it in a file.

Ah, yeah that makes more sense! More so given the configure option, I should have written "AFAIS" or something like that. :)

Quoted 1 line

I don't have a preference wrt /var/run or /var/cache.

So given that this is actually run time state, /var/run seems more appropriate, indeed.