nfs-utils.git
11 years agomount.nfs: Remove unused @salen parameter from nfs_ca_gai()
Chuck Lever [Tue, 14 Jul 2009 20:53:29 +0000 (16:53 -0400)]
mount.nfs: Remove unused @salen parameter from nfs_ca_gai()

Address compiler warning:

    network.c:1124: warning: unused parameter salen

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agomount.nfs: Fix some nfs_error() nits in network.c
Chuck Lever [Tue, 14 Jul 2009 20:51:27 +0000 (16:51 -0400)]
mount.nfs: Fix some nfs_error() nits in network.c

Fix a couple of nfs_error() call sites in utils/mount/network.c.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agomount.nfs: Remove unused parameter in try_mount()
Chuck Lever [Tue, 14 Jul 2009 20:50:00 +0000 (16:50 -0400)]
mount.nfs: Remove unused parameter in try_mount()

Address compiler warning:

    mount.c: At top level:
    mount.c:420: warning: unused parameter nomtab

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agomount.nfs: Use correct data type in discover_nfs_mount_data_version()
Chuck Lever [Tue, 14 Jul 2009 20:48:50 +0000 (16:48 -0400)]
mount.nfs: Use correct data type in discover_nfs_mount_data_version()

Address compiler warning:

    mount.c: In function discover_nfs_mount_data_version¿:
    mount.c:162: warning: comparison between signed and unsigned
    mount.c:164: warning: comparison between signed and unsigned
    mount.c:166: warning: comparison between signed and unsigned
    mount.c:168: warning: comparison between signed and unsigned
    mount.c:170: warning: comparison between signed and unsigned
    mount.c:178: warning: comparison between signed and unsigned

linux_version_code() and MAKE_VERSION() both return an unsigned int.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agosupport: Introduce sockaddr helpers to get and set IP port numbers
Chuck Lever [Tue, 14 Jul 2009 20:47:09 +0000 (16:47 -0400)]
support: Introduce sockaddr helpers to get and set IP port numbers

Introduce address family-agnostic functions that get and set IP port
numbers in socket addresses.  We can already replace a few similar
functions in the mount command, and a few more will come up with
statd and sm-notify.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agomount.nfs: Don't update extra_opts after text-based negotiation
Chuck Lever [Tue, 14 Jul 2009 20:45:07 +0000 (16:45 -0400)]
mount.nfs: Don't update extra_opts after text-based negotiation

The umount.nfs command will negotiate the mount options again, so all
that is needed in /etc/mnttab is the original set of options used for
the mount, plus the additional mandatory options like addr=''.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agomount.nfs: Clean up after restructuring version/protocol negotiation
Chuck Lever [Tue, 14 Jul 2009 20:39:17 +0000 (16:39 -0400)]
mount.nfs: Clean up after restructuring version/protocol negotiation

Fix up comments and function names to reflect the new version/protocol
negotiation scheme.  We can now remove a bunch of mount processing
that is specific to v2/v3, removing about 100 lines of logic from
stropts.c.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agomount.nfs: Clean up nfs_is_permanent_error()
Chuck Lever [Tue, 14 Jul 2009 20:37:02 +0000 (16:37 -0400)]
mount.nfs: Clean up nfs_is_permanent_error()

Clean up: Move nfs_is_permanent_error() closer to the functions that
call it, and update a documenting comment to reflect recent
restructuring in this area.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agomount.nfs: rearchitect mount version/protocol negotiation logic
Chuck Lever [Tue, 14 Jul 2009 20:35:26 +0000 (16:35 -0400)]
mount.nfs: rearchitect mount version/protocol negotiation logic

Text-based mounts try a mount operation first with default settings,
then negotiate via rpcbind queries and retry the mount, if the default
settings don't work.  This method introduces long delays in certain
common scenarios, and makes it difficult to tell when it is
appropriate to fail immediately or negotiate and retry.

To address these behavioral regressions, make text-based mounts
operate the same way that legacy mounts work.  Perform rpcbind queries
with short timeouts first, then use the results to determine
transport, version, and port number settings for the mount.

This allows the mount.nfs command to detect server settings, or
whether negotiation is even possible, quickly.  It also makes it
simple to determine when to fail vs. when to retry.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agomount.nfs: make nfs_options2pmap return errors
Chuck Lever [Tue, 14 Jul 2009 20:34:20 +0000 (16:34 -0400)]
mount.nfs: make nfs_options2pmap return errors

Up until now, nfs_options2pmap() has been passed mount options that
have already gone through the kernel's parser successfully.  So, it
never had to check for invalid mount option values.

However, we are about to pass it options that come right from the
user.  So nfs_options2pmap() will now need to report an error and
fail if it encounters a bogus value for any of the options it cares
about.

 =====

Note that nfs_options2pmap() will allow a bogus value for an option
if the same option is specified farther to the right with a useable
value.

For example, if a user specifies "proto=foo,...,tcp" then
nfs_options2pmap() uses "tcp" and ignores "proto=foo".

However, if the options are specified in the other order:
"tcp,...,proto=foo" then nfs_options2pmap() will fail.  This is a simple
and unambiguous extension of the "rightmost wins" rule.

Since mount.nfs strips out these options out and replaces them with
the rpcbind-negotiated options before invoking mount(2), the kernel
should never receive bogus values for these options from mount.nfs in
such cases.

This is probably slightly more flexible behavior than the legacy
mount implementation, but should be harmless.  All mount options
unrelated to pmap are ignored by nfs_options2pmap().

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agomount.nfs: force rpcbind queries if options aren't specified
Chuck Lever [Tue, 14 Jul 2009 20:31:15 +0000 (16:31 -0400)]
mount.nfs: force rpcbind queries if options aren't specified

nfs_options2pmap() fills in default values if the passed-in mount
options don't specify values.  This short-circuits the version, port,
and transport negotiation logic in nfs_probe_bothports().

Instead, nfs_options2pmap() should plant zeros in these pmap fields
to force nfs_probe_bothports() and nfs_advise_mount() to discover, via
rpcbind queries, what the server supports.

This fixes some scenarios where umount.nfs fails to connect to servers
that don't have all rpcbind ports open, in addition to fixing other
corner cases during mount.nfs version/protocol negotiation.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agomount.nfs: If port= specifies an unregistered port, retry, then fail
Chuck Lever [Tue, 14 Jul 2009 20:29:11 +0000 (16:29 -0400)]
mount.nfs: If port= specifies an unregistered port, retry, then fail

Suppose a port= option is specified on the mount command line, but not
enough other mount options are specified to avoid an rpcbind query to
discover the NFS service.

If the NFS service isn't registered on [100003, 3, "tcp", port] (even
if the server is listening on the specified port), the legacy mount.nfs
command fails immediately with:

 mount.nfs: mount to NFS server 'server' failed: RPC Error: Success

What's more, this mount request should succeeded if an NFS service is
registered on the specified port for another version and/or protocol.

So instead, let's retry the rpcbind query with the other versions and
transport protocols to be absolutely sure that port won't work with
either version or transport.  Then, if all fails, report:

 mount.nfs: mount to NFS server 'server' failed:
RPC Error: Program not registered

This change also affects text-based mounts that require negotiation
by the mount.nfs command.

Note that if the mount options specify all four pmap parameters for
NFS, the rpcbind query for the NFS service is skipped entirely.  The
mount command then hangs and times out later if NFS service is not
listening on the requested tuple.  This is unchanged from previous
behavior.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agogetport: Convert TCP connection refused to RPC_CANTRECV
Chuck Lever [Tue, 14 Jul 2009 20:27:54 +0000 (16:27 -0400)]
getport: Convert TCP connection refused to RPC_CANTRECV

In a similar vein to the timeout logic we just restored, a refused
TCP connection should be mapped to an equivalent UDP error code:
RPC_CANTRECV.

This is new behavior for TCP connections; the legacy mount command
appears to have simply failed immediately if a TCP connection was
refused during an rpcbind query.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agogetport: Restore historical TCP connect timeout error code
Chuck Lever [Tue, 14 Jul 2009 20:26:42 +0000 (16:26 -0400)]
getport: Restore historical TCP connect timeout error code

The latest versions of mount.nfs appear not to fall back to
UDP if TCP isn't available on the server.

Our new nfs_getport() implementation is missing a bit of logic
from the original mount getport() implementation.  Without it,
nfs_probe_port() sees a TCP connect timeout as a permanent error,
so it fails immediately instead of attempting to try again with
UDP.

Similar changes for our new ping API (see the old clnt_ping()
function, which is still in utils/mount/network.c).

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agomount.nfs: Add more debugging output around nfs_getport()
Chuck Lever [Tue, 14 Jul 2009 20:25:43 +0000 (16:25 -0400)]
mount.nfs: Add more debugging output around nfs_getport()

So we can see how rpcbind queries are failing during mount processing,
add some debugging messages (enabled with "mount.nfs -v") around the
nfs_getport() calls.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agogetport: Clear shared error fields before trying rpcbind queries
Chuck Lever [Tue, 14 Jul 2009 20:24:11 +0000 (16:24 -0400)]
getport: Clear shared error fields before trying rpcbind queries

Some RPC errors set fields in rpc_createerr.cf_error in addition
to cf_stat.  Be sure to clear _all_ error fields in rpc_createerr
each time through the rpcbind API.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agogetport: RPC_PROGNOTREGISTERED is a permanent error
Chuck Lever [Tue, 14 Jul 2009 20:21:01 +0000 (16:21 -0400)]
getport: RPC_PROGNOTREGISTERED is a permanent error

rpcbind returns RPC_PROGNOTREGISTERED if it knows for certain that an
RPC program is not supported for a given transport.  This is a
permanent and authoritative error, so the library's rpcbind query API
should never retry the query -- it will only get the same answer.

A similar change was submitted for libtirpc.  Unlike rpcb_getaddr(3t),
mount.nfs's rpcbind client only retries once (with RPCB3PROC_GETADDR),
but an extra TCP socket in this case would leave another port in
TIME_WAIT.  It's infrequent enough, but might as well get rid of it.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agosupport: Set proper retransmit timeout for datagram transports
Chuck Lever [Tue, 14 Jul 2009 20:18:37 +0000 (16:18 -0400)]
support: Set proper retransmit timeout for datagram transports

Instead of setting the total timeout and the retransmit timeout to the
same value for datagram transports, use a 1 second retransmit timeout,
so we actually get a retransmit or two before failing.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agosupport: Don't return RPC_UNKNOWNHOST from rpc_socket.c
Chuck Lever [Tue, 14 Jul 2009 20:17:28 +0000 (16:17 -0400)]
support: Don't return RPC_UNKNOWNHOST from rpc_socket.c

RPC_UNKNOWNHOST means a hostname isn't known -- basically it's
EAI_NONAME from getaddrinfo(3).  Since the functions in rpc_socket.c
don't take a hostname argument, RPC_UNKNOWNHOST is not an appropriate
return code from these functions.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agosupport: Use HAVE_LIBTIRPC to switch in bindresvport_sa(3t)
Chuck Lever [Tue, 14 Jul 2009 20:16:46 +0000 (16:16 -0400)]
support: Use HAVE_LIBTIRPC to switch in bindresvport_sa(3t)

commit 383a026d99624c88c0e802103ef4c4865db8eb71, which fixed an
earlier commit, is still not quite correct.

bindresvport_sa(3t) is available whenever libtirpc is linked.
There's no need to use IPV6_SUPPORTED here.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoNew versions of libtool add extra aclocal scripts
Chuck Lever [Tue, 14 Jul 2009 20:14:19 +0000 (16:14 -0400)]
New versions of libtool add extra aclocal scripts

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agogetport: Remove unneeded @salen arguments
Chuck Lever [Tue, 14 Jul 2009 20:13:39 +0000 (16:13 -0400)]
getport: Remove unneeded @salen arguments

Clean up: Now that getnameinfo(3) is no longer used, the @salen
argument to nfs_sockaddr2universal() is no longer needed.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agogetport: replace getnameinfo(NI_NUMERICHOST) with inet_ntop(3)
Chuck Lever [Tue, 14 Jul 2009 20:12:23 +0000 (16:12 -0400)]
getport: replace getnameinfo(NI_NUMERICHOST) with inet_ntop(3)

getnameinfo(3) with the NI_NUMERICHOST flag is used in
support/nfs/getport.c to convert socket addresses to universal address
strings.

Older versions of glibc do not have getnameinfo(3), however.  In order
for nfs-utils to build on older systems we switch in legacy code via
HAVE_GETNAMEINFO and use inet_ntoa(3).

A problem with this is that we have to double our test matrix to be
sure that both versions of these routines build and operate correctly.
Another minor problem is that inet_ntoa(3) is officially deprecated.

So let's always use a single implementation based on inet_ntop(3).
Universal address strings do not support link-local / scope IDs, so we
don't lose any functionality by using inet_ntop(3) here.

This means we open code a bit of logic that is available in most
modern versions of glibc, but in return we can use exactly the same
code for all builds (on systems with getnameinfo(3) and without).

An additional benefit is we can avoid using NI_MAXHOST for character
buffers that live on the stack: it's 1025 bytes.  Instead,
INET6_ADDRSTRLEN is used, which is just 46 bytes, plus an additional
eight bytes for the port information.  We add beefier buffer overflow
detection logic as well.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agogetport: Remove AI_ADDRCONFIG from nfs_gp_loopback_address()
Chuck Lever [Tue, 14 Jul 2009 20:11:08 +0000 (16:11 -0400)]
getport: Remove AI_ADDRCONFIG from nfs_gp_loopback_address()

AI_ADDRCONFIG was used ostensibly to figure out if the local system
had IPv6 available when generating a loopback address.

A legacy version of nfs_gp_loopback_address() was created to handle
ANYADDR address generation for old versions of glibc where
AI_ADDRCONFIG doesn't exist.  This means we have to be careful to
test both the normal and legacy versions when committing changes in
this area.

But it turns out that even contemporary versions of glibc ignore
AI_ADDRCONFIG when the hostname string is NULL.  getaddrinfo(3)
always returns an AF_INET and an AF_INET6 loopback address in this
case, no matter how the system is configured.

Change nfs_gp_loopback_address() to have one version that simply looks
up "localhost" instead of doing anything fancy.  If "localhost" is an
IPv6 address, we'll use that.  Otherwise, it should nearly always be
an AF_INET loopback address.

This eliminates the need for AI_ADDRCONFIG, and removes the duplicate
version of nfs_gp_loopback_address().  Note that callers never used
the port number in the returned socket address, so get rid of the
"sunrpc" service string too.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agogetport: RPCB_GETADDR's r_addr should contain rpcbind port, not zero
Chuck Lever [Tue, 14 Jul 2009 20:09:03 +0000 (16:09 -0400)]
getport: RPCB_GETADDR's r_addr should contain rpcbind port, not zero

Similar to a change made to the kernel's rpcbind client.  See
kernel commit 143b6c4008a7928de7e139c3a77a90e4cad8db2c.

The r_addr argument of RPCB_GETADDR procedures contains the
rpcbind server's address and port number.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agogetport: RPCB_GETADDR r_owner should be an empty string
Chuck Lever [Tue, 14 Jul 2009 20:07:35 +0000 (16:07 -0400)]
getport: RPCB_GETADDR r_owner should be an empty string

Some servers reject RPCB_GETADDR requests with a non-empty r_owner
field.  "RPC: Server can't decode arguments"

An empty string is already used by libtirpc and the kernel
for RPCB_GETADDR requests.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agomydaemon: remove closeall() calls from mydaemon()
Steve Dickson [Mon, 29 Jun 2009 14:44:20 +0000 (10:44 -0400)]
mydaemon: remove closeall() calls from mydaemon()

idmapd and svcgssd have a mydaemon() routine that uses closeall() to
close file descriptors. Unfortunately, they aren't using it correctly
and it ends up closing the pipe that the child process uses to talk to
its parent.

Fix this by not using closeall() in this routine and instead, just close
the file descriptors that we know need to be closed. If /dev/null can't
be opened for some reason, then just have the child exit with a non-zero
error.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoThe closeall function is broken in such a way that it almost never nfs-utils-1-2-1-rc1
Steve Dickson [Mon, 22 Jun 2009 14:05:44 +0000 (10:05 -0400)]
The closeall function is broken in such a way that it almost never
closes any file descriptors. It's calling strtol on the text
representation of the file descriptor, and then checking to see if the
value of *endptr is not '\0' before trying to close the file. This check
is wrong.

When strtol returns an endptr that points to a NULL byte, that indicates
that the conversion was completely successful. I believe this check
should instead be requiring that endptr is pointing to '\0' before
closing the fd.

Also, fix up the function to check for conversion errors from strtol. If
one occurs, just skip the close on that entry.

Finally, as Trond pointed out, it's unlikely that readdir will return a
blank string in d_name but that situation wouldn't be detected by the
current code. This patch adds such a check and skips the close if it
occurs.

Reported-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoMake --enable-tirpc the default. If --enable-tirpc wasn't explicitly
Steve Dickson [Mon, 22 Jun 2009 13:49:17 +0000 (09:49 -0400)]
Make --enable-tirpc the default. If --enable-tirpc wasn't explicitly
specified, but TIRPC libs or headers aren't present then just throw a
warning and disable it. If it was explicitly specified, then throw an
error and exit if they aren't present.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoSet the verbosity level in both the librpcsecgss and
Steve Dickson [Mon, 22 Jun 2009 13:17:27 +0000 (09:17 -0400)]
Set the verbosity level in both the librpcsecgss and
libnfsidmapd libraries when verbosity level is set
by the '-v' flag it on either daemon.

Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoRetry export if getfh fails.
NeilBrown [Wed, 3 Jun 2009 19:48:08 +0000 (15:48 -0400)]
Retry export if getfh fails.

mountd tries to avoid telling the kernel to export something
when the kernel already knows to do that.
However sometimes (exportfs -r) the kernel can be told
to forget something without mountd realising.
So if mountd finds that it cannot get a  valid filehandle,
make sure it really has been exported to the kernel.

This only applies if the nfsd filesystem is not mounted.

Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoRelease 1.2.0
Steve Dickson [Tue, 2 Jun 2009 15:20:50 +0000 (11:20 -0400)]
Release 1.2.0

Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoReplace the Sun RPC license with the BSD license, nfs-utils-1-2-0
Tom spot Callaway [Tue, 2 Jun 2009 14:43:05 +0000 (10:43 -0400)]
Replace the Sun RPC license with the BSD license,
with the explicit permission of Sun Microsystems

Signed-off-by: Tom "spot" Callaway <tcallawa@redhat.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoumount.nfs: Harden umount.nfs error reporting nfs-utils-1-1-7-rc1
Chuck Lever [Mon, 18 May 2009 17:29:38 +0000 (13:29 -0400)]
umount.nfs: Harden umount.nfs error reporting

Add additional error reporting to nfs_advise_umount().
These messages can be displayed if the "-v" option
is specified with umount.nfs.  Normally these
messages do not appear.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agonfs-utils: Fix IPv6 support in support/nfs/rpc_socket.c
Chuck Lever [Mon, 18 May 2009 17:27:25 +0000 (13:27 -0400)]
nfs-utils: Fix IPv6 support in support/nfs/rpc_socket.c

Use the correct #ifdef variable to enable IPv6 support for privileged
RPC clients.  Without this fix, unmounting an IPv6 NFSv2/v3 server
fails.

Introduced by commit 8c94296bc84f3a204f2061c0391a1d2350e4f37e.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoWhen compiling nfs-utils-1.1.6, I get this error:
Robert Schwebel [Mon, 18 May 2009 15:36:14 +0000 (11:36 -0400)]
When compiling nfs-utils-1.1.6, I get this error:

arm-iwmmx-linux-gnueabi-gcc -DHAVE_CONFIG_H -I. -I../../support/include
-isystem
/home/rsc/svn/oselas/bsp/OSELAS.BSP-Pengutronix-AllYes-trunk/platform-phyCORE-PXA270.PCM990/sysroot-target/include
-isystem
/home/rsc/svn/oselas/bsp/OSELAS.BSP-Pengutronix-AllYes-trunk/platform-phyCORE-PXA270.PCM990/sysroot-target/usr/include
-D_GNU_SOURCE -Wall -Wstrict-prototypes  -pipe -g -O2 -MT sm_inter_svc.o
-MD -MP -MF .deps/sm_inter_svc.Tpo -c -o sm_inter_svc.o sm_inter_svc.c
sm_inter_svc.c:10:39: error: sys/ttycom.h: No such file or directory

Use sys/ioctl.h instead.

Signed-off-by: Robert Schwebel <r.schwebel@pengutronix.de>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agomount: remove legacy version of nfs_name_to_address()
Chuck Lever [Mon, 18 May 2009 15:17:49 +0000 (11:17 -0400)]
mount: remove legacy version of nfs_name_to_address()

Currently we have two separate copies of nfs_name_to_address() since
some older glibc's don't define AI_ADDRCONFIG.  This means extra
work to build- and run-test both functions when code is changed in
this area.

It is also the case that gethostbyname(3) is deprecated, and should
not be used in new code.

Remove the legacy code in favor of always using getaddrinfo(3).

We can also get rid of nfs_name_to_address()'s @family argument as
well.

Note also this addresses a bug in nfsumount.c -- it was calling
nfs_name_to_address() with AF_UNSPEC unconditionally, even if the
legacy version of nfs_name_to_address(), which doesn't support
AF_UNSPEC, was in use.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agosm-notify: Failed DNS lookups should be retried
Chuck Lever [Mon, 18 May 2009 15:08:53 +0000 (11:08 -0400)]
sm-notify: Failed DNS lookups should be retried

Currently, if getaddrinfo(3) fails when trying to resolve a hostname,
sm-notify gives up immediately on that host.  If sm-notify is started
before network service is available on a system, that means it quits
without notifying anyone.  Or, if DNS service isn't available due to
a network partition or because the DNS server crashed, sm-notify will
simply remove all of its callback files and exit.

Really, sm-notify should try harder.  We know that the hostnames
passed in to notify_host() have already been vetted by statd, which
won't monitor a hostname that it can't resolve.  So it's likely that
any DNS failure we meet here is a temporary condition.  If it isn't,
then sm-notify will stop trying to notify that host in 15 minutes
anyway.

[ The host's file is left in /var/lib/nfs/sm.bak in this case, but
  sm.bak is not read again until the next time sm-notify runs. ]

sm-notify already has retry logic for handling RPC timeouts.  We can
co-opt that to drive DNS resolution retries.

We also add AI_ADDRCONFIG because on systems whose network startup is
handled by NetworkManager, there appears to be a bug that causes
processes that started calling getaddinfo(3) before the network came
up to continue getting EAI_AGAIN even after the network is fully
operating.

As I understand it, legacy glibc (before AI_ADDRCONFIG was exposed in
headers) sets AI_ADDRCONFIG by default, although I haven't checked
this.  In any event, pre-glibc-2.2 systems probably won't run
NetworkManager anyway, so this may not be much of a problem for them.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agosm-notify: Don't orphan addrinfo structs
Chuck Lever [Mon, 18 May 2009 15:03:54 +0000 (11:03 -0400)]
sm-notify: Don't orphan addrinfo structs

sm-notify orphans an addrinfo struct in its address list rotation
logic if only a single result was returned from getaddrinfo(3).

For each host, the first time through notify_host(), we want to
send a PMAP_GETPORT request.  ->ai is NULL, and retries is set to 100,
forcing a DNS lookup and an address rotation.  If only a single
addrinfo struct is returned, the rotation logic causes a NULL to be
planted in ->ai, copied from the ai_next field of the returned result.

This means that the second time through notify_host() (to perform the
actual SM_NOTIFY call) we do a second DNS lookup, since ->ai is NULL.
The result of the first lookup has been orphaned, and extra network
traffic is generated.

This scenario is actually fairly common.  Since we pass

  .ai_protocol = IPPROTO_UDP,

  to getaddrinfo(3), for most hosts, which have a single forward and
  reverse pointer in the DNS database, we get back a single addrinfo
  struct as a result.

  To address this problem, only perform the address list rotation if
  there is more than one element on the list returned by getaddrinfo(3).

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoutils/nfsd: enable nfs minorvers4 by default
Benny Halevy [Mon, 18 May 2009 14:50:11 +0000 (10:50 -0400)]
utils/nfsd: enable nfs minorvers4 by default

Enable support for the maximum minor version (4.1 at the moment)
by default.  It can be disabled using the -N command line
option.

Signed-off-by: Benny Halevy <bhalevy@panasas.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoExtend -N command line option syntax to accept
Benny Halevy [Mon, 18 May 2009 14:47:51 +0000 (10:47 -0400)]
Extend -N command line option syntax to accept
<version>.<minorversion> to disable support
for <minorversion>. Only 4.1 is currently supported.

Signed-off-by: Benny Halevy <bhalevy@panasas.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoutils/nfsd: add support for minorvers4
Benny Halevy [Mon, 4 May 2009 15:44:49 +0000 (11:44 -0400)]
utils/nfsd: add support for minorvers4

minorvers4 can be used to either enable or disable nfsv4.x.

If minorvers4 is a positive integer n, in the allowed range (only
minorversion 1 is supported for now), the string "+4.n" is appended
to the versions string written onto /proc/fs/nfsd/versions.

Correspondingly, if minorver4 is a negative integer -n, the string
"-4.n" is written.

With the default value, minorvers4==0, the minor version
setting is not changed.

Note that unlike the protocol versions 2, 3, or 4.  The minor version
setting controls the *maximum* minor version nfsd supports.  Particular
minor version cannot be controlled on their own.  With only minor
version 1 supported at the moment the difference doesn't matter,
but for future minor versions greater than 1, enabling minor
version X will enable support for all minor versions 1 through X.
Disabling minor version X will disable support for minor
versions X and up, enabling 1 through X-1.

Signed-off-by: Benny Halevy <bhalevy@panasas.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoHost aliases need to be checked when netgroups is used in exports.
Steve Dickson [Wed, 29 Apr 2009 16:38:02 +0000 (12:38 -0400)]
Host aliases need to be checked when netgroups is used in exports.

Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoRelease 1.1.6 nfs-utils-1-1-6
Steve Dickson [Mon, 20 Apr 2009 17:32:50 +0000 (13:32 -0400)]
Release 1.1.6

Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoumount.nfs: Fix return value of nfs_mount_protocol() nfs-utils-1-1-6-rc7
Chuck Lever [Sat, 18 Apr 2009 13:45:46 +0000 (09:45 -0400)]
umount.nfs: Fix return value of nfs_mount_protocol()

Fix a copy-paste error introduced in nfs_mount_protocol().  It should
return an IPPROTO_ number, not an NFS version number.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoumount.nfs: Use a privileged port when sending UMNT requests
Chuck Lever [Sat, 18 Apr 2009 13:44:56 +0000 (09:44 -0400)]
umount.nfs: Use a privileged port when sending UMNT requests

Turns out we do actually need to use a privileged port for UMNT.  The
Linux rpc.mountd complains if an ephemeral source port is used:

  Apr 17 15:52:19 ingres mountd[2061]: refused unmount request from
    192.168.0.59 for /export (/export): illegal port 60932

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agosupport: Provide an API for creating a privileged RPC client
Chuck Lever [Sat, 18 Apr 2009 13:43:58 +0000 (09:43 -0400)]
support: Provide an API for creating a privileged RPC client

We needed to guarantee that some RPC programs, such as PMAP, got an
unprivileged port, to prevent exhausting the local privileged port
space sending RPC requests that don't need such privileges.
nfs_get_rpcclient() provides that feature.

However, some RPC programs, such as MNT and UMNT, require a privileged
port.  So, let's provide an additional API for this that also supports
IPv6 and setting a destination port.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoutils/nfsd: fix -N optarg error printout
Benny Halevy [Wed, 15 Apr 2009 18:16:08 +0000 (14:16 -0400)]
utils/nfsd: fix -N optarg error printout

as currently printed c is the version number, not a string char,
therefore is should be printed as %d not %c.  That said, just print
optarg as %s since it might be non-numeric.

Signed-off-by: Benny Halevy <bhalevy@panasas.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agogetport.c: fix non-standard C
Chuck Lever [Wed, 15 Apr 2009 16:52:48 +0000 (12:52 -0400)]
getport.c: fix non-standard C

Squelch a compiler warning in getport.c:

  getport.c:65: warning: ¿static¿ is not at beginning of declaration

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agonfs-utils: reverse order of librpcsecgss and libgssglue checks
Chuck Lever [Wed, 15 Apr 2009 16:38:40 +0000 (12:38 -0400)]
nfs-utils: reverse order of librpcsecgss and libgssglue checks

The check that validates the version of librpcsecgss also needs to
have libgssglue installed.  Without libgssglue, ./configure complains
that it can't find rpcsecgss, even though it's installed.

It also turns out that the error message generated by pkg-config is
more complete than the one we have in aclocal/rpcsec_vers.m4, so just
let those PKG_CHECK_MODULES m4 macros use the default error message.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agonfs-utils: add IPv6 code to gssd nfs-utils-1-1-6-rc6
Jeff Layton [Wed, 15 Apr 2009 16:04:06 +0000 (12:04 -0400)]
nfs-utils: add IPv6 code to gssd

All of the pieces to handle IPv6 are now in place. Add IPv6-specific
code wrapped in the proper #ifdef's so that IPv6 support works when
it's enabled at build-time.

Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agonfs-utils: switch gssd to use standard function for getting an RPC client
Jeff Layton [Wed, 15 Apr 2009 16:01:46 +0000 (12:01 -0400)]
nfs-utils: switch gssd to use standard function for getting an RPC client

We already have a common function for setting up an RPC client. That
function uses the tirpc API when tirpc is enabled and is also already
IPv6 enabled. Switch gssd to use it.

Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agonfs-utils: query for remote port using rpcbind instead of getaddrinfo
Jeff Layton [Wed, 15 Apr 2009 15:36:07 +0000 (11:36 -0400)]
nfs-utils: query for remote port using rpcbind instead of getaddrinfo

We already have the server's address from the upcall, so we don't really
need to look it up again, and querying the local services DB for the
port that the remote server is listening on is just plain wrong.

Use rpcbind to set the port for the program and version that we were
given in the upcall. The exception here is NFSv4. Since NFSv4 mounts
are supposed to use a well-defined port then skip the rpcbind query
for that and just set the port to the standard one (2049).

Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agonfs-utils: store the address given in the upcall for later use
Jeff Layton [Wed, 15 Apr 2009 15:34:51 +0000 (11:34 -0400)]
nfs-utils: store the address given in the upcall for later use

The current upcall could be more efficient. We first convert the address
to a hostname, and then later when we set up the RPC client, we do a
hostname lookup to convert it back to an address.

Begin to change this by keeping the address in the clnt_info that we get
out of the upcall. Since a sockaddr has a port field, we can also
eliminate the port from the clnt_info.

Finally, switch to getnameinfo() instead of gethostbyaddr(). We'll need
to use that call anyway when we add support for IPv6.

Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agonfs-utils: make getnameinfo() required for --enable-gss
Jeff Layton [Wed, 15 Apr 2009 14:29:04 +0000 (10:29 -0400)]
nfs-utils: make getnameinfo() required for --enable-gss

Systems that are so old that they don't have getnameinfo() in glibc are
probably also running kernels that are so old that they don't support
gssapi upcalls anyway.

Make --enable-gss dependent on the presence of the getnameinfo()
function.  This allows us to reduce some conditional compilation.

Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoThe --list option does not work on server stats. nfs-utils-1-1-6-rc5
Steve Dickson [Wed, 8 Apr 2009 14:26:26 +0000 (10:26 -0400)]
The --list option does not work on server stats.

The print_stats_list() routine was using the client's
stats to decide whether to display any stats. This did
not work when there was only server stats.

This patch breaks up print_stats_list into two different
routines allowing both server and clients stats to be
listed.

Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoThe server stats were not being updated with the
Steve Dickson [Wed, 8 Apr 2009 13:48:58 +0000 (09:48 -0400)]
The server stats were not being updated with the
-Z options causing the stats to be incorrect.

Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoEliminate the displaying zero stats when the explicit protocol
Steve Dickson [Wed, 8 Apr 2009 13:28:22 +0000 (09:28 -0400)]
Eliminate the displaying zero stats when the explicit protocol
is specified (-2, -3, -4) the -Z and or --list options.

When a particular protocol is specified and either
the -Z or --list options are used, zeros or blank lines
are echoed to the screen when there is not any NFS traffic.
This cause any useful data to be scroll off the screen.

With this patch only non-zero stats will be shown, which
makes the output of these options more condensed and
in turn more useful.

Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agonfsstat: Add --list flag
Kevin Constantine [Sat, 4 Apr 2009 11:18:26 +0000 (07:18 -0400)]
nfsstat: Add --list flag

nfsstat.c: Adds the --list flag to print information in a list format
   instead of the standard multi-column format
nfsstat.man: Updates the manpage to include the --list flag.

Signed-off-by: Kevin Constantine <kevin.constantine@disneyanimation.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoKeep the interval output quite when there is no NFS traffic.
Steve Dickson [Sat, 4 Apr 2009 10:46:09 +0000 (06:46 -0400)]
Keep the interval output quite when there is no NFS traffic.

The  'nfsstat -Z5' command continually outputs the following
when there is no NFS traffic.

Client rpc stats:
calls      retrans    authrefrsh
0          0          0

This patch adds code that will keep the interval output
quite so real results will not be scrolled of the screen

Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agonfsstat: Print diff stats every N seconds
Kevin Constantine [Sat, 4 Apr 2009 10:29:01 +0000 (06:29 -0400)]
nfsstat: Print diff stats every N seconds

nfsstat.c: Implements an optional "interval" argument to --sleep
nfsstat.man: Explains the use of --sleep[interval]

Reviewed-By: Greg Banks <gnb@fmeh.org>
Signed-off-by: Kevin Constantine <kevin.constantine@disneyanimation.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agocacheio: return any original error from qword_eol nfs-utils-1-1-6-rc4
Kevin Coffman [Fri, 3 Apr 2009 19:20:09 +0000 (15:20 -0400)]
cacheio: return any original error from qword_eol

If the initial fflush() fails in qword_eol, log the failure
and return the indication of the original failure, not the
successful cover-up.

Signed-off-by: Kevin Coffman <kwc@citi.umich.edu>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agosvcgssd: check the return code from qword_eol() and log failures
Kevin Coffman [Fri, 3 Apr 2009 19:18:16 +0000 (15:18 -0400)]
svcgssd: check the return code from qword_eol() and log failures

If qword_eol() fails while writing the context information, log
an indication of the failure.

This addresses at least one cause of the intermittent, and
previously undiagnosed, problem of the server returning
GSS_S_NO_CONTEXT when a context was seemingly successfully
created and sent down to the kernel.  In my case there was a
mis-match between kernel and user-land configuration resulting in
the proper kernel module not being loaded. Therefore the write
of the context failed, but was not logged by svcgssd.  When the
kernel goes to find the resulting context, it was really not
there and correctly returned GSS_S_NO_CONTEXT to the client.

Signed-off-by: Kevin Coffman <kwc@citi.umich.edu>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoMountd should use separate lockfiles
Ben Myers [Fri, 3 Apr 2009 19:13:10 +0000 (15:13 -0400)]
Mountd should use separate lockfiles

Mountd keeps file descriptors used for locks separate from
those used for io and seems to assume that the lock will
only be released on close of the file descriptor that was used
with fcntl.  Actually the lock is released when any file
descriptor for that file is closed.  When setexportent() is called
after xflock() he closes and reopens the io file descriptor and defeats the
lock.

This patch fixes that by using a separate file for locking, cleaning
them up when finished.

Signed-off-by: Ben Myers <bpm@sgi.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agonfs-utils: fix AC_CHECK_FUNC calls in configure.ac
Chuck Lever [Fri, 3 Apr 2009 19:03:15 +0000 (15:03 -0400)]
nfs-utils: fix AC_CHECK_FUNC calls in configure.ac

AC_CHECK_FUNC and AC_CHECK_FUNCS take 3 args. Any ones beyond that are
ignored. In several places, we're passing the "action-if-not-found" in
as the 4th arg so it's being ignored.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoIn recent Fedora builds, the '-D _FORTIFY_SOURCE=2' compile nfs-utils-1-1-6-rc3
Steve Dickson [Mon, 23 Mar 2009 21:13:01 +0000 (17:13 -0400)]
In recent Fedora builds, the '-D _FORTIFY_SOURCE=2' compile
flag has been set. This cause warnings to be generated when
return values from reads/writes (and other calls) are not
checked. The patch address those warnings.

Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoRegardless though, it's dangerous to keep invalid pointers around like
Jeff Layton [Mon, 23 Mar 2009 12:13:01 +0000 (08:13 -0400)]
Regardless though, it's dangerous to keep invalid pointers around like
this. Later code changes may make it more likely for this problem to
occur.

Also eliminate some unneeded NULL pointer checks before freeing memory.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agogssd: free buffer allocated by gssd_k5_err_msg
Jeff Layton [Mon, 23 Mar 2009 12:12:37 +0000 (08:12 -0400)]
gssd: free buffer allocated by gssd_k5_err_msg

There's no way for the caller of gssd_k5_err_msg to know whether to free
the string it returns. It can call krb5_get_error_message which returns
a string that must be freed via krb5_free_error_string. The other ways
that it can return a string require that the memory not be freed.

Deal with this by copying the string to a new buffer in all cases. Then
we can properly free the string allocated by krb5_get_error_message.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agogssd: NULL-terminate buffer after read in read_service_info (try #2)
Jeff Layton [Mon, 23 Mar 2009 12:12:14 +0000 (08:12 -0400)]
gssd: NULL-terminate buffer after read in read_service_info (try #2)

Valgrind complains that we're passing an unintialized buffer to sscanf
here. The main problem seems to be that we're not ensuring that the
buffer is NULL terminated before we pass it off.

This is the second version of this patch, the first one did not increase
the buffer allocation by 1 which could have led to clobbering the next
byte on the stack if nbytes == INFOBUFLEN.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agogssd: initialize fakeseed in prepare_krb5_rfc1964_buffer
Jeff Layton [Mon, 23 Mar 2009 12:11:41 +0000 (08:11 -0400)]
gssd: initialize fakeseed in prepare_krb5_rfc1964_buffer

This causes a compiler warning and also means that we're stuffing
the buffer with uninitialized junk from the stack. Other places in
this code initialize "fakeseed" to 0. Do the same here.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agonfs-utils: clean up handling of libgssglue in gssd Makefile nfs-utils-1-1-6-rc2
Chuck Lever [Mon, 16 Mar 2009 17:42:44 +0000 (13:42 -0400)]
nfs-utils: clean up handling of libgssglue in gssd Makefile

From: Jeff Layton <jlayton@redhat.com>

Make the pkgconfig check for libgssglue conditional on tirpc being
enabled. When it's disabled, the pkgconfig check for librpcsecgss will
pull in the gssglue lib and include dir automatically.

Also, make sure we include GSSGLUE_CFLAGS and the GSSGLUE_LIBS to the
appropriate places in utils/gssd/Makefile.am so that we pick up
the gssglue libs when tirpc is enabled.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agonfs-utils: Include legacy or TI-RPC headers, not both
Chuck Lever [Mon, 16 Mar 2009 17:40:47 +0000 (13:40 -0400)]
nfs-utils: Include legacy or TI-RPC headers, not both

Data type incompatibilities between the legacy RPC headers and the
TI-RPC headers mean we can't use libtirpc with code that was compiled
against the legacy RPC headers.  The definition of rpcprog_t for
example is "unsigned long" in the legacy library, but it's "uint32_t"
for TI-RPC.  On 32-bit systems, these types happen to have the same
width, but on 64-bit systems they don't, making more complex data
structures that use these types in fields ABI incompatible.

Adopt a new strategy to deal with this issue.  When --enable-tirpc is
set, append "-I/usr/include/tirpc" to the compilation steps.  This
should cause the compiler to grab the tirpc/ headers instead of the
legacy headers.  Now, for TI-RPC builds, the TI-RPC legacy functions
and the TI-RPC headers will be used.  On legacy systems, the legacy
headers and legacy glibc RPC implementation will be used.

A new ./configure option is introduced to allow system integrators to
use TI-RPC headers in some other location than /usr/include/tirpc.
/usr/include/tirpc remains the default setting for this new option.

The gssd implementation presents a few challenges, but it turns out
the gssglue library is similar to the auth_gss pieces of TI-RPC.  To
avoid similar header incompatibility issues, gssd now uses libtirpc
instead of libgssglue if --enable-tirpc is specified.  There may be
other issues to tackle with gssd, but for now, we just make sure it
builds with --enable-tirpc.

Note also: svc_getcaller() is a macro in both cases that points to
a sockaddr field in the svc_req structure.  The legacy version points
to a sockaddr_in type field, but the TI-RPC version points to a
sockaddr_in6 type field.

rpc.mountd unconditionally casts the result of svc_getcaller() to a
sockaddr_in *.  This should be OK for TI-RPC as well, since rpc.mountd
still uses legacy RPC calls (provided by glibc, or emulated by TI-RPC)
to set up its listeners, and therefore rpc.mountd callers will always
be from AF_INET addresses for now.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoconfigure: move IPv6 feature checks into aclocal
Chuck Lever [Mon, 16 Mar 2009 17:37:09 +0000 (13:37 -0400)]
configure: move IPv6 feature checks into aclocal

Clean up: for consistency with other local feature checks, move IPv6
feature checks into aclocal/

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agonfs-utils: don't need extra libs to do AC_CHECK_LIBS for librpcsecgss
Chuck Lever [Mon, 16 Mar 2009 17:32:49 +0000 (13:32 -0400)]
nfs-utils: don't need extra libs to do AC_CHECK_LIBS for librpcsecgss

From: Jeff Layton <jlayton@redhat.com>

The conftest should work without these extra libs being included.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoconfigure: Move rpcsecgss checking into aclocal
Chuck Lever [Mon, 16 Mar 2009 17:31:27 +0000 (13:31 -0400)]
configure: Move rpcsecgss checking into aclocal

Clean up: Introduce two more aclocal scripts for handling rpcsecgss
dependency checking.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoconfigure: pull common nfsidmap and event checks into aclocal/
Chuck Lever [Mon, 16 Mar 2009 17:30:26 +0000 (13:30 -0400)]
configure: pull common nfsidmap and event checks into aclocal/

Clean up: Create an aclocal script for the nfsidmap library and
headers checks used for both --enable-gss and --enable-nfsv4.
Move libevent checks out too.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agonfs-utils: replace function-specific switches with HAVE_LIBTIRPC
Chuck Lever [Mon, 16 Mar 2009 17:29:18 +0000 (13:29 -0400)]
nfs-utils: replace function-specific switches with HAVE_LIBTIRPC

Instead of switching in TI-RPC-specific logic with a function-specific
switch like HAVE_CLNT_VG_CREATE, let's use the more generic
HAVE_LIBTIRPC macro everywhere.

This simplifies ./configure (always a good thing), and makes it more
clear in the source code exactly what the extra conditionally compiled
code is for.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoconfigure: move TI-RPC checks into aclocal
Chuck Lever [Mon, 16 Mar 2009 17:28:42 +0000 (13:28 -0400)]
configure: move TI-RPC checks into aclocal

Define an aclocal test for TI-RPC headers and library, and move the
TI-RPC checks earlier in our configure script so other feature checks
can use the availability of TI-RPC to decide what to do.

Since bindresvport_sa is required just for IPv6 support, move that
check to the IPv6 feature tests.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoconfigure: add defensive quoting in some function checks
Chuck Lever [Mon, 16 Mar 2009 17:26:34 +0000 (13:26 -0400)]
configure: add defensive quoting in some function checks

Clean up: Add proper m4 quoting in macros that check for the presence
of some functions in configure.ac.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoAdded back the some logging variables which are no nfs-utils-1-1-5-rc1 nfs-utils-1-1-6-rc1
Steve Dickson [Sun, 8 Mar 2009 14:10:25 +0000 (10:10 -0400)]
Added back the some logging variables which are no
longer used but, unfortunately, they are extern-ed by
public headers files which are not under the control
of this package.

Spotted-by: Juergen Daubert <jue@jue.li>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoRelease 1.1.5 nfs-utils-1-1-5
Steve Dickson [Thu, 5 Mar 2009 11:42:56 +0000 (06:42 -0500)]
Release 1.1.5

Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agogssd: include gssapi/gssapi.h before write_bytes.h
Chuck Lever [Thu, 5 Mar 2009 11:23:25 +0000 (06:23 -0500)]
gssd: include gssapi/gssapi.h before write_bytes.h

In gssd/context_lucid.c, ensure that gss_buffer_desc and gss_OID_desc
are defined before write_bytes.h, which uses these definitions, is
included.  With TI-RPC, these definitions are not provided by
rpc/rpc.h.

It appears that <gssapi/gssapi_krb5.h> already includes krb5.h and
gssapi.h (on my system, anyway) so let's drop those includes.

Ideally write_bytes.h itself should include the needed headers, but
some source files that use Heimdal include a different, Heimdal-
compatible, header to get these definitions.

Pointed-out-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agogssd: Use same style for including config.h that rest of nfs-utils uses
Chuck Lever [Thu, 5 Mar 2009 11:22:31 +0000 (06:22 -0500)]
gssd: Use same style for including config.h that rest of nfs-utils uses

Clean up.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agomount.nfs: squelch compiler warning for TI-RPC builds
Chuck Lever [Thu, 5 Mar 2009 11:21:34 +0000 (06:21 -0500)]
mount.nfs: squelch compiler warning for TI-RPC builds

The printf format string in nfs_pp_debug() assumes the @program and
@version arguments are unsigned long, because the legacy RPC headers
define both rpcprog_t and rpcvers_t as unsigned long types.

However, the TI-RPC headers define both types as uint32_t, which
requires a different printf format type.  If we replace the legacy
headers with TI-RPC headers, this type mismatch generates compiler
warnings that are nothing but noise.

We are about to provide a switch at ./configure time to allow the use
of either the legacy RPC headers or the TI-RPC headers, so we need
a printf format that works in both cases.

To squelch the compiler warnings that occur when using the TI-RPC
headers, cast both arguments in the fprintf statement to the widest of
the two types ("unsigned long" or "uint32_t").

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agonfs-utils: Provide type-checked version of svc_getcaller()
Chuck Lever [Thu, 5 Mar 2009 11:20:40 +0000 (06:20 -0500)]
nfs-utils: Provide type-checked version of svc_getcaller()

TI-RPC's version of the svc_getcaller() macro points to a sockaddr_in6,
not a sockaddr_in, though for AF_INET callers, an AF_INET address
resides there.  To squelch compiler warnings when the TI-RPC version of
the svc_req structure is used, add inline helpers with appropriate
type casting.

Note that tcp_wrappers support only AF_INET addresses for now.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoconfigure: Remove CPPFLAGS substitution
Chuck Lever [Thu, 5 Mar 2009 11:16:04 +0000 (06:16 -0500)]
configure: Remove CPPFLAGS substitution

At least on my systems, the AM_CPPFLAGS substitution at the end of
configure.ac is not needed.  It adds an extra copy of
"-I../../support/includes" to each compile step.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agonfs-utils: Replace getservbyname(3)
Chuck Lever [Thu, 5 Mar 2009 11:13:33 +0000 (06:13 -0500)]
nfs-utils: Replace getservbyname(3)

The getservbyname(3) function is not re-entrant, and anyway,
the man page says it is obsolete.  Replace it with a call
to getaddrinfo(3).

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoUpdate example in exports(5) man page
Valerie Aurora Henson [Wed, 4 Mar 2009 22:53:41 +0000 (17:53 -0500)]
Update example in exports(5) man page

Signed-off-by: Valerie Aurora Henson <vaurora@redhat.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoHave sizeof using the correct value in nfs_getlocalport()
Ben Greear [Wed, 4 Mar 2009 22:07:55 +0000 (17:07 -0500)]
Have sizeof using the correct value in nfs_getlocalport()

Signed-off-by: Ben Greear <greearb@candelatech.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agonfs-utils: Remove fprintf() call from support/nfs/getport.c
Chuck Lever [Wed, 4 Mar 2009 21:58:47 +0000 (16:58 -0500)]
nfs-utils: Remove fprintf() call from support/nfs/getport.c

getport.c is effectively a library, so it should not be emitting
messages via printf.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoshowmount command: remove legacy RPC logic
Chuck Lever [Wed, 4 Mar 2009 21:31:10 +0000 (16:31 -0500)]
showmount command: remove legacy RPC logic

Turns out both legacy RPC and TI-RPC have a clnt_create(3) API.  So
there's
really no need to keep the old open-coded transport logic around.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agonfs(5): document new [no]resvport option
Chuck Lever [Wed, 4 Mar 2009 21:29:29 +0000 (16:29 -0500)]
nfs(5): document new [no]resvport option

Kernel 2.6.25 deprecates intr/nointr.  Reflect this change in nfs(5).

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agonfs(5): document new [no]resvport option
Chuck Lever [Wed, 4 Mar 2009 21:27:36 +0000 (16:27 -0500)]
nfs(5): document new [no]resvport option

Kernel 2.6.28 adds a new mount option: [no]resvport.  Document the new
option in the nfs(5) man page.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoThe legacy version of nfs_name_to_address() returned an incorrect
Chuck Lever [Wed, 4 Mar 2009 21:26:48 +0000 (16:26 -0500)]
The legacy version of nfs_name_to_address() returned an incorrect
value for the size of the returned address.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoconfigure: fix AC_CACHE_VAL warnings on Fedora 10
Chuck Lever [Wed, 4 Mar 2009 20:54:39 +0000 (15:54 -0500)]
configure: fix AC_CACHE_VAL warnings on Fedora 10

Autoconf 2.63 (and maybe earlier releases) complains about the cache
variable name used in aclocal/libblkid.m4:

configure.ac:217: warning: AC_CACHE_VAL(libblkid_is_recent, ...):
suspicious cache-id, must contain _cv_ to be cached
../../lib/autoconf/general.m4:1974: AC_CACHE_VAL is expanded from...
aclocal/libblkid.m4:2: AC_BLKID_VERS is expanded from...
configure.ac:217: the top level

This addresses
  https://bugzilla.redhat.com/attachment.cgi?bugid=481386 .

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoGeneral clean up. Removed unused routines. Reworked syslog
Steve Dickson [Wed, 4 Mar 2009 20:36:27 +0000 (15:36 -0500)]
General clean up. Removed unused routines. Reworked syslog
message to (hopefully) make it more sensible. Move
"#ifdef HAVE_LIBWRAP" around so nothing will be defined
when tcp wrapper is not configured.

Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoConverted good_client() to correctly use the tcp wrapper
Steve Dickson [Wed, 4 Mar 2009 20:24:58 +0000 (15:24 -0500)]
Converted good_client() to correctly use the tcp wrapper
interface and added a note to the mountd man page saying
hostnames will be ignored when they can not be looked up.

Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoOnly hash on IP address and Program number. Including the Procedure
Steve Dickson [Wed, 4 Mar 2009 20:22:37 +0000 (15:22 -0500)]
Only hash on IP address and Program number. Including the Procedure
number only creates needles extra hash entries.

Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agoExportfs and rpc.mountd optimalization
Tomas Richter [Wed, 18 Feb 2009 18:33:27 +0000 (13:33 -0500)]
Exportfs and rpc.mountd optimalization

There were some problems with exportfs and rpc.mountd for long export
lists - see https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=76643
I do optimalization as my bachelors thesis (Facuulty of informatics,
Masaryk's university Brno, Czech Republic), under lead of Yenya
Kasprzak.

Both exportfs and rpc.mount build linked list of exports (shared
functions in export.c). Every time they are inserting new export into
list, they search for same export in list.
I replaced linked list by hash table and functions export_add and
export_lookup by functions hash_export_add and hash_export_lookup
(export.c).

Because some other functions required exportlist as linked list, hash
table has some implementation modification im comparison with ordinary
hash table. It also keeps exports in linked list  and has pointer to
head of the list. So there's no need of implementation function
<for_all_in_hash_table>.

Signed-off-by: Tomas Richter <krik3t@gmail.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
11 years agorpc.idmapd should tell the user more verbosely if Dnotify
Lukas Hejtmanek [Wed, 18 Feb 2009 12:48:11 +0000 (07:48 -0500)]
rpc.idmapd should tell the user more verbosely if Dnotify
support is missing in kernel.

Signed-off-by: Lukas Hejtmanek <xhejtman@ics.muni.cz>
Signed-off-by: Steve Dickson <steved@redhat.com>