Change to a dependency on libgssglue rather than libgssapi

From: Guillaume Rousse <Guillaume.Rousse@inria.fr>

Use the renamed library libgssglue rather than libgssapi.
Also bump the required version for librpcsecgss (to the one that
also requires libgssglue rather than libgssapi).

Signed-off-by: Kevin Coffman <kwc@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Update description of clientaddr= in nfs(5)

The description of clientaddr= in nfs(5) is out of date and omits some
other typical reasons for using it.  Expand its description.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: restore admin-specified clientaddr option

Revert an earlier change to make specifying the clientaddr= option illegal.

Jeff Layton pointed out that admins may want to specify the clientaddr=
option to advertise a different callback address when accessing an NFSv4
server through a NAT router.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Add a block comment before mount_error() and umount_error()

Add block comments before mount_error() and umount_error() to help
distinguish them from the other error reporting functions.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Remove get_my_ipv4addr() from utils/mount/stropts.c

The get_my_ipv4addr() function is no longer used, so remove it from
utils/mount/stropts.c.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Use get_client_addr() to build clientaddr= option

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Create a new API to find out client's address

Provide a version of clnt_ping() that discovers the client's address, but
doesn't do an RPC ping.  The in-kernel text-based mount code already does
a ping, so all we need here is address discovery.

As well, add a block comment in front of clnt_ping() that hopefully
elucidates the differences.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Refactor parse_devname and fill_ipv4_addr

We will eventually need the server address in both append_addr_opt()
and append_clientaddr_opt().  Call parse_devname() and fill_ipv4_addr()
from the top level functions.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Use dynamically allocated character buffers

Replace the use of static character arrays with buffers allocated via
xstrdup() in stropts.c.

I added a couple of extra length checks; not sure if these are really
necessary.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Remove get_my_ipv4addr() from append_addr_opt()

The result of the get_my_ipv4addr() function is not used in
append_addr_opt(), so remove the call and the ip_addr variable.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Don't allow the user to specify addr= or clientaddr=.

The current mount.nfs implementation doesn't allow users to specify their
own addr= or clientaddr= option.  The new string-based interface does
allow this, even though nfs(5) does not document 'addr=' and specifically
forbids adding 'clientaddr='.

Make the addition of either option by the user a permanent error.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Fix nfs4mount_s prototype.

The definition of nfs4mount_s in utils/mount/stropts.c doesn't match the
prototype declared in utils/mount/stropts.h.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Combine two verbose messages into one

A little less noise.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: getport() should avoid modifying passed-in arguments

Defensive coding: getport() shouldn't alter the passed-in server address,
but should treat it as read only.  Have it operate on a copy.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Replace fork() with daemon() for backgrounding mounts

Neil recommended this change to address a problem with background mount
processes handling signals properly during an init level change.  It is
also useful for preventing background mount processes from reporting
progress on the parent's tty, which is generally just annoying noise.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount: Small tidy up.

1/ remove confusing sentence from a comment.
2/ remove needless 'exit' - just fall through to common exit.

Signed-off-by: Neil Brown <neilb@suse.de>

nfsstat: Add -S/--since flag.

Read statistics from the file given with -S/--since and display
the difference between those and the current statistics.  Valid stat files
are those in the form of /proc/net/rpc/nfs, /proc/net/rpc/nfsd, or any
"pretty" output from nfsstat itself.  Statistics that are missing from a
"pretty" stat file are treated as zeroes.

Similar to Neil Brown's suggestion, one might use this in
conjunction with watch(1) like this:

$ watch "nfsstat --since /tmp/stats; nfsstat >/tmp/stats"

Signed-off-by: David M. Richter <richterd@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>

nfsstat: #define stat header labels.

Make #defines for the stat header labels, use for output, and add
to struct statinfo.  This will be used by a subsequent patch for
-S/--since.

Signed-off-by: David M. Richter <richterd@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>

nfsstat: Remove some pointer comparisons against global statinfos.

Pass-in an "is_srv" arg instead of doing pointer comparisons
against the static statinfo arrays themselves.

Signed-off-by: David M. Richter <richterd@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>

nfsstat: Obviate and remove copy_stats().

Rather than copying stats during --sleep, instead introduce a few
pointers and rearrange them to avoid the copy.  Will be useful with
--since, too.

Signed-off-by: David M. Richter <richterd@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>

nfsstat: rename *_tmp variables to *_old.

To make a subsequent patch a little clearer, make it explicit that
current *_tmp variables are actually always/only the "old" set used in
comparisons.

Signed-off-by: David M. Richter <richterd@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>

Restrict -i option to root.

As it is still under development, don't expose it - just in case.

Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Enable mount.nfs to do text-based mount support

A new command line option, "-i", is added to mount.nfs to force the use of
the string interface for testing purposes.  "-s", "-t", and "-r" are
already taken or have legacy meaning so I picked "-i".

At some later point, when everyone is comfortable with the string mount
option parsing implementation, we will add a switch based on kernel
version, and remove the "-i" command line option.  For now, I am more
comfortable enabling it by hand instead.

Since this is a temporary arrangement, I'm leaving the option undocumented
in the mount.nfs man page.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Add new files for supporting string-ified mount options

Introduce support files which contain code that builds string mount
options and passes them to the kernel.  This is a pre-requisite for
actually enabling /sbin/mount.nfs to do text-based mounts.

This is only partially complete at the moment, but is presented so that
folks can start banging on the kernel mount option string parsing code.
There are clearly still parts that are not implemented quite yet, such
as bg and retry support, but it should be enough to get going.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Replace a "return 0; " that should be an "exit(0); "

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Treat another UDP/TCP pair of strings like the rest

See the error messages at the end of utils/mount/network.c:get_socket()

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

umount.nfs: Update umount.nfs(8)

Account for recent changes to the umount.nfs command.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Update the mount.nfs(8) man page

Account for recent changes to the mount.nfs command.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

nfs-utils: specify a create mode with open(...O_CREAT) call in xflock

The xflock function can create a file via open() with O_CREAT, but does
not specify the create mode when it does so. I think 0644 should be
appropriate given the current usage of this function.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Neil Brown <neilb@suse.de>

Remove ARCH-specific compile flags for ALPHA.

 -mno-fp-regs -ffixed-8

are used for compiling the kernel on alphas, and it seems they
were copied into nfs-utils long ago, even though they have no
relevance now.
As we now use floating point (just a little bit in nfsstat), remove
these pointless flags.

nfsstat: include -Z/--sleep in the manpage

Add a little blurb about -Z/--sleep to the nfsstat manpage.

Signed-off-by: David M. Richter <richterd@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>

nfsstat: rename --diff-stat to --sleep

As per Bruce Fields' suggestion, rename the clunky -D/--diff-stat
to the more-intuitive -Z/--sleep.

Signed-off-by: David M. Richter <richterd@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>

nfsstat: fix a bug in diff_stats()

Fix a bug in diff_stats() that causes false-positives in
has_stats(), which can result in a bunch of zeros being displayed instead
of suppressed as intended.

Signed-off-by: David M. Richter <richterd@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>

nfsstat: add a comment to has_stats()

Clarify what has_stats() is actually doing.

Signed-off-by: David M. Richter <richterd@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>

nfsstat: display 3 additional NFSv4 client op counters

Display three extra NFSv4 client counters that are already exposed
in /proc/net/rpc/nfs: GETACL, SETACL, and FS_LOCATIONS.  Won't cause
trouble on older kernels that might lack those counters.

Signed-off-by: David M. Richter <richterd@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>

fix regression with rmtab handling in newer nfs-utils code

commit dbfcba5b03481d49f3e78946fe8fd066e2025545
incorrectly added some mountlist_add calls.
Removed them again.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Fix background mounts

The new mount.nfs helper does not support background mounts.  Add support
for background mounts.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Refactor mount logic to allow support for "bg" option

Refactor main() a bit to make adding support for "bg" option easier.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: restore bg argument to nfsmount() and nfs4mount()

To enable background mounting again, restore the "bg" argument to
nfsmount() and nfs4mount() that was recently removed.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: get_socket() may clobber errno, but preserves .re_errno

get_socket() guarantees that rpc_createerr.cf_error.re_errno is set
correctly after an error, but it can wipe errno if it has to print an error
message.  Make sure that clnt_ping() checks the correct error code when
get_socket() returns.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Always close the socket at the end of getport()

These days, none of get_socket()'s callers pass an RPC_ANYSOCK on to the
RPC code.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Shorter timeout for TCP connects

The standard TCP connect timeout on Linux is 75 seconds, which can be
too long in some cases.  The timeout itself can be altered on a system-wide
basis, but we'd like mount to have it's own connect timeout that's tunable,
and defaults to a shorter value.

The get_socket() function is a utility function that does TCP connects for
getport, clnt_ping, and other functions.  Add logic there to use a
non-blocking connect() and select() in order to time out a connect
operation that's taking too long.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Remove a redundant port assignment

getport() always fills in its port number before calling GETPORT.  No need
for the caller to do this.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: umount needs to pick up transport option from /etc/mtab

Now that umount's default transport protocol has become more flexible, it
will need to detect both proto=udp and proto=tcp in /etc/mtab.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: fix printf's normally hidden behind NFS_MOUNT_DEBUG

After I enabled NFS_MOUNT_DEBUG the compiler started spitting out warnings.
Fix up commas, output formatting, and double-wide character support.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Support double-wide characters in printed strings

Previous NLS changes missed a spot or two.  This patch tries to get
most of them, but probably misses a few more.

In errors.c:mount_errors() I've removed a period at the end of the error
messages; this is consistent with other error messages I've examined.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

nfsstat: add -D/--diff-stat

Add -D/--diff-stat: instead of immediately displaying total
collected NFS stats and exiting, nfsstat will take a snapshot of current
statistics and pause until the user hits ^C, at which point it takes a
second snapshot and then prints out the difference of the two; i.e.,
only the statistics collected during the pause.

Signed-off-by: David M. Richter <richterd@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>

nfsstat: just declare -D/--diff-stat's variables

Add the temp vars -D/--diff-stat will use.  (just using a separate
patch to make the next one easier to read)

Signed-off-by: David M. Richter <richterd@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>

nfsstat: use macros to declare struct statinfo arrays

Use macros to build the arrays of struct statinfos.  This will
make adding the extra set of temporary variables needed by -D/--diff-stat
much easier.

Signed-off-by: David M. Richter <richterd@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>

nfsstat: standardize naming of opcount arrays

Standardized Naming 2: Make the client and server "call counts"
arrays' variable names reflect/include the actual text labels from the
/proc files -- e.g., "cltv2info" becomes "cltproc2info".  A subsequent
patch will rely on this naming scheme.

Signed-off-by: David M. Richter <richterd@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>

nfsstat: standardize naming of server variables

Standardized Naming 1: Instead of having a variety of different
server-related variable name prefixes ("srv", "svr", "svc", "SVC"), set
them all to "srv" or "SRV".  A subsequent patch will rely on this naming
scheme.

Signed-off-by: David M. Richter <richterd@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>

nfsstat: add get_stats()

Share some of the existing statistics-gathering code by folding it
into a function, get_stats().

Signed-off-by: David M. Richter <richterd@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>

nfsstat: add has_stats()

To help readability, add has_stats() and use it when deciding
whether to print.

Signed-off-by: David M. Richter <richterd@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>

Fix more 'prototype' warnings and other warnings.

Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: clean up a couple of printfs in fstab.c

Clean up for consistent use.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

libnfs.a: eliminate another dependency on a global variable

The file support/nfs/fstab.c, which is linked into libnfs.a, depends on the
global variable "verbose."  This variable is defined and used only in the
mount command, and the functions in fstab.c are used only by the mount
command.

Move fstab.c and support/include/fstab.h to utils/mount.  This file
placement is also consistent with at least one other mount helper,
mount.ocfs2.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

libnfs.a: eliminate conn.c and conn.h

conn.[ch] are now no longer needed.  Clean them out and delete them.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: use nfs_error instead of fprintf in get_socket()

Use nfs_error() where appropriate.  I used "goto" here to reduce string
splitting and indenting past the point of readability.  Gee, it would be
nice if C had proper exception handling...

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

libnfs.a: move get_socket() function to utils/mount/network.c

Now we can address the real problem: that get_socket() depends on the
global variable "verbose" which is only available in the mount command.

Move get_socket() into utils/mount/network.c, and make it static.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

libnfs.a: move clnt_ping() to utils/mount

Continue clean up of mount functionality in libnfs.a by moving clnt_ping()
to utils/mount/network.c.  Note that socklen_t is an unsigned int... the
i386 gcc compiler threw a signedness warning about the 3rd argument of
getsockname().

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

libnfs.a: move more mount-only functions out of libnfs.a

Continue clean-up with nfsvers_to_mnt() and mntvers_to_nfs().

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

libnfs.a: move mnt_{open, close}clnt calls to utils/mount/network.c

It turns out that get_socket() accesses a global variable, "verbose," that
is only available in the mount command; yet it's in libnfs.a.  This creates
an undocumented API dependency that will bite someone someday.  This
mount-specific functionality doesn't really belong in libnfs.a anyway.

The simplest way to resolve this is to move all of the functions in
support/nfs/conn.c into utils/mount.  network.c seems like the logical
place to put these.  An added benefit is we eventually get to make
get_socket() static.

Let's start with the mnt_{open,close}clnt functions.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: clean up getport() function

Mostly comment clarification.  Also replace some naked undocumented
integers with macros, and make getport() static.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: fix hang when getport() uses TCP against unavailable servers

If get_socket() can't get us an open TCP socket, we know the server is
down, so make  getport() exit early instead of hanging.  This logic is
copied from clnt_ping().

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

umount.nfs: eliminate a nearly empty header file.

Clean-up: move nfsumount() global declaration to nfs_mount.h, and remove
nfsumount.h.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

umount.nfs: Pass return codes from do_nfs_umount back

Do we really care about these return codes?

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

umount.nfs: Make do_nfs_mount use conventional EX_ style return codes

do_nfs_mount() should return EX_ style return codes and not 1 or 0 in order
to distinguish between usage errors and other problems (such as EX_FILEIO
or RPC errors).

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

umount.nfs: make _nfsumount() static

_nfsumount() is never called outside of nfsumount.c.  Also give it a more
conventional name.

Note that it's return code is ignored.  That will be addressed in a
subsequent patch.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

umount.nfs: make del_mtab() a static function

Clean-up: del_mtab() isn't used outside of nfsumount.c.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

Make sure nfs_call_umount's callers are handling its return code correctly

umount.nfs is treating nfs_call_umount's return code like a standard mount
return code (EX_SOMETHING) when its really an RPC return code.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

umount.nfs: move nfs_call_umount to network.c

nfs_call_umount() is shared by nfsmount.c and nfsumount.c, and manages a
network function (building the RPC umount call to the server's MNT daemon).
So move it to network.c with other network-related functions.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

umount.nfs: make umount_usage() static

No good reason to export umount_usage.  Make it static.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

umount.nfs: make umount.nfs use UDP by default

mount.nfs currently uses UDP by default when calling a MNT server.  Make
umount.nfs do the same.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

umount.nfs: some umount.nfs error return codes are confusing

umount.nfs should return the standard EX_ mount return codes.  At some
point in the past, it was returning 0 for failure and 1 for success, and
some of these have been changed and some haven't.  See if we can rectify
this.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: fix more nits with error messages

Catch-all.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

Add -Wstrict-prototypes to compiler args, and fix warnings caused.

Use __fpurge to ensure single-line writes to cache files

On a recent Debian/Sid machine, I saw libc retrying stdio writes that
returned write errors.  The result is that if an export downcall returns
an error (which it can in normal operation, since it currently
(incorrectly) returns -ENOENT on any negative downcall), then subsequent
downcalls will write multiple lines (including the original line that
received the error).

The result is that the server fails to respond to any rpc call that
refers to an unexported mount point (such as a readdir of a directory
containing such a mountpoint), so client commands hang.

I don't know whether this libc behavior is correct or expected, but it
seems safest to add the __fpurge() (suggested by Neil) to ensure data is
thrown away.

Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>

fix broken compile due to missing pseudoflavors.h

Commit e9b0bed761bc77ba046f53be2ec324c4b61357ff should also have added
this file.

Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: White space clean up

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Use nfs_error in nfsmount.c, nfsumount.c, and nfs4mount.c

Clean up. - fprintf becomes nfs_error

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Error handling clean-up

o  Use nfs_error( _() ) instead of fprintf(stderr,
o  Use the mount return code macros instead of bare integers
o  Free mount_point after it has been canonicalized

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: When -v is set and the mount succeeds, print the mount info

Add a bit of logic that appears to be in other mount helpers.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

Remove the running_bg parameter

Clean up: when nfsmount and nfs4mount are called, running_bg is always set
to zero.  Remove the parameter from the call.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Change *flags to flags

It's not necessary to treat the *flags parameter to nfsmount and nfs4mount
as an output parameter.  Nothing is passed back.  Replace it with a normal
call-by-value.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: No need to return nfs_mount_data structs

Refactor mount processing slightly to remove an output parameter and an
unnecessary type cast.  The mount syscall is now made from inside
nfs_mount or nfs4mount, rather than in common code after those are called.

Code review suggests that EX_BG was never returned by mount.nfs because the
logic I just replaced was always returning EX_FAIL.  The new logic should
properly return EX_BG when appropriate.

However, it is unclear whether /bin/mount handles backgrounding the mount
request, or whether mount.nfs should.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Move MS_DUMMY and friends to header file

We move the definitions of MS_USER and friends to our local copy of
mount_constants.h.  These will need to be available in nfsmount.c and
nfs4mount.c when we move the mount system call out of mount.c.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Move start_statd into nfs_mount

Move start_startd into network.c, and move the call inside of
nfs_mount, instead of immediately after - conceptually a better place
for it.

Also fix a minor nit:  Since the mount actually fails if start_statd
doesn't work, cause mount.nfs to exit with a status of EX_FAIL.

Still need to do something about "running_bg" in nfsmount.c:nfsmount().

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Move network functions into a common source module

Separate network oriented functions from filesystem oriented
functions, for general cleanliness.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Eliminate some C macros that use implicit arguments

Get rid of nfs_probelist, mnt_probelist, and proto_probelist in order to
remove the use of HAVE_RELIABLE_TCP, MAX_NFSPROT, and MAX_MNTPROT.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Always preset nfs_mount_version

nfs_mount_version is a global integer that is set based on a guess
about which nfs_mount_data version is appropriate for the kernel we're
running on.

Make it always available and have the correct value before calling mount
and unmount so they don't have to worry about setting it themselves.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Create a common source module for reporting mount errors

Clean up, and pre-requisite for subsequent fixes.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: clean-up add_mtab

Clean up add_mtab(), and make /sbin/mount.nfs[4] return a proper error if
it fails.

Also include an unbalanced unlock_mtab() noticed by Steve D.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: /bin/mount already handles --bind & friends

Clean-up: remove logic to handle --bind and other such command-line
options from mount.nfs[4].

These options are already handled in /bin/mount, and the logic for handling
them in the NFS helper is currently disabled.  Other helpers such as
mount.ocfs2 appear not to support --bind (ie. they rely on /bin/mount to
do it).

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Remove support for "-t" option

/bin/mount will never pass "-t" to a mount helper, since it passes the
fs-type in the name of the program it is executing.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

mount.nfs: Simplify generation of progname

Use basename() instead of our own majick.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

rpcdebug: Update 'rpcdebug' tool with recently added debugging flags

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>

Fix umount.nfs exit status

Hi,

As per a bug report from a user:

mount.c seems to assume that nfsumount() uses standard C true/false
return values, and inverts them for the exit status (where 0 is
traditionally considered success). However, nfsumount() consistently
seems to use 0 for success, and thus a success gets returned as exit
status 1 and a failure as exit status 0. This confuses at least
the GNOME drive manager applet, and probably others as well.

Signed-off-by: Steinar H. Gunderson <sesse@debian.org>
Signed-off-by: Neil Brown <neilb@suse.de>

Added missing unlock_mtab() call in the add_mtab()
routine.

Signed-off-by: Steve Dickson <steved@redhat.com>

The wrong bit field is being passed to NFSCTL_TCPISSET()
during one of the sanity checks in rpc.nfsd.

Signed-off-by: Steve Dickson <steved@redhat.com>

document the sec= option

Document the sec= option in the exports man page.

Not done: it would be nice to have an example or two here (and not just
in the final "EXAMPLE" section, though that would be nice too).  I was
just too lazy to figure out the formatting.

Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>