Chuck Lever [Fri, 28 Sep 2007 20:36:35 +0000 (16:36 -0400)]
text-based mount.nfs: parse option strings into lists
Adapt a parsing trick used by Python.
Parse mount option strings into an abstract data type so we don't have to
copy and/or tokenize the whole option string multiple times while trying
to manipulate the mount options. Then, just before calling the mount(2)
system call, convert the object back into a C string.
One major advantage of this approach is that we can copy the final version
of the mount options into /etc/mtab when we're done, instead of copying in
the original mount options that the user specified. Any fallback from NFS
v3 to NFS v2 or TCP to UDP that was done by mount.nfs will be reflected in
/etc/mtab.
This patch adds methods for creating and manipulating mount option data
objects.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Neil Brown <neilb@suse.de>
Chuck Lever [Fri, 28 Sep 2007 20:36:30 +0000 (16:36 -0400)]
mount.nfs: add new string tokenizer facility
To quote the strtok(3) man page: "Avoid using these functions."
OK. We've created our own. The main reason for this is that strtok(3)
doesn't handle quoted delimiters at all. We need to handle this:
context="foo,bar"
where 'context' is a single mount option that sets a token string that
possibly uses the same delimiter that the mount command uses to separate
options (that is, a comma).
Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Neil Brown <neilb@suse.de>
Jeff Layton [Thu, 27 Sep 2007 10:54:04 +0000 (06:54 -0400)]
rpc.mountd: add new mode for handling netgroup-heavy configurations
If a host is a member of a large number of netgroups, it becomes easily
possible for client_compose to generate a m_hostname string that
overflows the maximum string length allowed by the kernel caches.
This patch adds a new mode for mountd where it will map IP address to IP
address in the auth.unix.ip cache. When this enabled, mountd doesn't
bother using client_compose to build the m_hostname string. It just
populates it with the dotted-quad ip address. When mountd handles a
mount request, it then has an IP address and a path. It then calls
client_check to check the host against export entries where the path has
already matched.
Since we don't bother looking up netgroups which have no relation to the
mount, this can be a big performance gain in netgroup-heavy
configurations. The downside is that every host has a corresponding
entry in the nfsd.export and nfsd.fh caches as well as the auth.unix.ip
cache.
The new behavior is automatically enabled if the length of all of the
concatenated netgroup names in the export table is longer than half
NFSCLNT_IDMAX. The rationale for this logic is that this should allow
for a host to be a member of a long list of netgroups while still
allowing for other matches.
Signed-off-by: Jeff Layton <jlayton@redhat.com> Acked-by: Steve Dickson <steved@redhat.com> Signed-off-by: Neil Brown <neilb@suse.de>
Jeff Layton [Thu, 27 Sep 2007 10:53:58 +0000 (06:53 -0400)]
rpc.mountd: create client_resolve and change client_compose to take a hostent arg
This moves the resolution of IP address to hostent into a helper function
and has other functions call it. Having client_compose take a hostent arg
allows us to avoid an extra hostname lookup in the auth_authenticate
codepath as well. Instead of redoing this lookup in client_compose, we can
simply reuse the hostent that was already generated in auth_authenticate.
Signed-off-by: Jeff Layton <jlayton@redhat.com> Acked-by: Steve Dickson <steved@redhat.com> Signed-off-by: Neil Brown <neilb@suse.de>
Jeff Layton [Thu, 27 Sep 2007 10:53:53 +0000 (06:53 -0400)]
rpc.mountd: make exportent->e_hostname a dynamically-allocated string
This makes the e_hostname field of the exportent into a pointer to a
dynamically allocated string. This is necessary since this is field is
often filled out from the m_hostname. This too adds a few
micro-optimizations as we can avoid copying the string in some places
and simply pass a pointer to the original string instead.
Signed-off-by: Jeff Layton <jlayton@redhat.com> Acked-by: Steve Dickson <steved@redhat.com> Signed-off-by: Neil Brown <neilb@suse.de>
Jeff Layton [Thu, 27 Sep 2007 10:53:48 +0000 (06:53 -0400)]
rpc.mountd: Change nfs_client->m_hostname to be a dynamically-allocated string
Change nfs_client->m_hostname to be dynamically allocated rather than a
fixed length array of size NFSCLNT_IDMAX. This also adds a bit of
micro-optimization in a few places since it reduces the amount of string
copying that needs to be done.
Signed-off-by: Jeff Layton <jlayton@redhat.com> Acked-by: Steve Dickson <steved@redhat.com> Signed-off-by: Neil Brown <neilb@suse.de>
Jeff Layton [Thu, 27 Sep 2007 10:53:43 +0000 (06:53 -0400)]
rpc.mountd: fix memory leak and error handling in nfsd_fh
nfsd_fh() uses strdup for creating found_path and doesn't check the
return value. It also doesn't free this memory when the function
returns. Check the return value of strdup and return immediately
if it's NULL. Also, free found_path on exit.
Signed-off-by: Jeff Layton <jlayton@redhat.com> Acked-by: Steve Dickson <steved@redhat.com> Signed-off-by: Neil Brown <neilb@suse.de>
Chuck Lever [Mon, 24 Sep 2007 15:29:31 +0000 (11:29 -0400)]
umount.nfs: umount doesn't recognize a busy file system
umount.nfs shouldn't remove a busy file system from /etc/mtab, and should
report and return an error. I also added an extra "goto" to make the flow
of control more clear, and to reduce the chance that a future change in
this logic will break it.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Neil Brown <neilb@suse.de>
Chuck Lever [Mon, 24 Sep 2007 15:29:15 +0000 (11:29 -0400)]
mount.nfs: Add error messages for errors reported by text-based mount(2)
The text-based mount(2) system call API can return some additional errors
that we would like to report correctly to our users. These should be safe
to use with the legacy mount(2) ABI as well.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Neil Brown <neilb@suse.de>
Chuck Lever [Mon, 24 Sep 2007 15:29:10 +0000 (11:29 -0400)]
text-based mount.nfs: Add text-based error reporting function
The mount_errors() function prints an error based on what just happened in
the user-space RPC library. This is meaningless for text-based mounts,
since they don't use the RPC library for most things.
Add a new error printing function that the text-based logic can use to
report an error.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Neil Brown <neilb@suse.de>
Chuck Lever [Mon, 24 Sep 2007 15:29:05 +0000 (11:29 -0400)]
mount.nfs: rename mount_errors()
The function mount_errors() actually reports RPC errors generated by the
user-land RPC library. We're about to add a similar function for reporting
system call errors via errno, so rename mount_errors() to be more specific
about what it does.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Neil Brown <neilb@suse.de>
Chuck Lever [Mon, 24 Sep 2007 15:28:33 +0000 (11:28 -0400)]
mount.nfs: add EX_SUCCESS exit code
We've had some recent trouble, especially in the umount code, that appears
to be due to functions returning a 1 or a 0 return code when they should be
returning a mount exit code (such as EX_FAIL) or a 0.
To help clearly distinguish these two classes of functions, define an
EX_SUCCESS exit code, which is equal to zero.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Neil Brown <neilb@suse.de>
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>
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>
Chuck Lever [Fri, 24 Aug 2007 17:11:38 +0000 (13:11 -0400)]
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>
Chuck Lever [Fri, 24 Aug 2007 17:11:32 +0000 (13:11 -0400)]
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>
Chuck Lever [Fri, 24 Aug 2007 17:11:16 +0000 (13:11 -0400)]
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>
Chuck Lever [Tue, 21 Aug 2007 16:19:08 +0000 (12:19 -0400)]
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>
david m. richter [Mon, 13 Aug 2007 19:16:14 +0000 (15:16 -0400)]
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:
Chuck Lever [Fri, 10 Aug 2007 22:20:01 +0000 (18:20 -0400)]
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>
Chuck Lever [Fri, 10 Aug 2007 22:19:55 +0000 (18:19 -0400)]
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>
Jeff Layton [Tue, 7 Aug 2007 12:01:08 +0000 (08:01 -0400)]
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>
Neil Brown [Thu, 9 Aug 2007 01:23:24 +0000 (11:23 +1000)]
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.
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: 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>
Chuck Lever [Fri, 3 Aug 2007 17:24:00 +0000 (13:24 -0400)]
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>
Chuck Lever [Fri, 3 Aug 2007 17:23:49 +0000 (13:23 -0400)]
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>
david m. richter [Tue, 31 Jul 2007 21:39:49 +0000 (17:39 -0400)]
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>
david m. richter [Tue, 31 Jul 2007 21:39:23 +0000 (17:39 -0400)]
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>
david m. richter [Tue, 31 Jul 2007 21:39:16 +0000 (17:39 -0400)]
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>
Chuck Lever [Sat, 28 Jul 2007 21:51:00 +0000 (17:51 -0400)]
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>
Chuck Lever [Sat, 28 Jul 2007 21:50:50 +0000 (17:50 -0400)]
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>
Chuck Lever [Sat, 28 Jul 2007 21:50:40 +0000 (17:50 -0400)]
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>
Chuck Lever [Sat, 28 Jul 2007 21:50:30 +0000 (17:50 -0400)]
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>
Chuck Lever [Sat, 28 Jul 2007 21:50:19 +0000 (17:50 -0400)]
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>
Chuck Lever [Sat, 28 Jul 2007 21:50:04 +0000 (17:50 -0400)]
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>
Chuck Lever [Sat, 28 Jul 2007 21:49:43 +0000 (17:49 -0400)]
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>
Chuck Lever [Sat, 28 Jul 2007 21:49:28 +0000 (17:49 -0400)]
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>
projects / nfs-utils.git / log