freebsd-nq/contrib/amd/BUGS

	    LIST OF KNOWN BUGS IN AM-UTILS OR OPERATING SYSTEMS

Note: report am-utils bugs via Bugzilla to https://bugzilla.am-utils.org/ or
by email to the am-utils mailing list (see www.am-utils.org).


(1) mips-sgi-irix*

[1A] known to have flaky NFS V.3 and TCP.  Amd tends to hang or spin
infinitely after a few hours or days of use.  Users must install recommended
patches from vendor.  Patches help, but not all the time.  Otherwise avoid
using NFS V.3 and TCP on these systems, by setting

	/defaults opts:=vers=2,proto=udp

[1B] yp_all() leaks a file descriptor.  Eventually amd runs out of file
descriptors and hangs.  Am-utils circumvents this by using its own version
of yp_all which uses udp and iterates over NIS maps.  The latter isn't as
reliable as yp_all() which uses TCP, but it is better than hanging.

(I have some reports that older version of hpux-9, with older libc, also
leak file descriptors.)

[1C] SGI's MIPSpro C compiler on IRIX 6 has the unfortunate habit of
creating code specificially for the machine it runs on.  The ABI and ISA
used depend very much on the OS version and compiler release used.  This
means that the resulting amd binary won't run on machines different from
the build host, particularly older ones.  Older versions of am-utils
enforced the O32 ABI when compiling with cc to work around this, but this
ABI is deprecated in favor of the N32 ABI now, so we use -n32 -mips3 to
ensure that the binaries run on every host capable of running IRIX 6 at
all.  If this is not appropriate for you, configure with something like
CC='cc -64' instead to get the desired ABI and ISA.

(2) alpha-unknown-linux-gnu (RedHat Linux 4.2)

hasmntopt(mnt, opt) can go into an infinite loop if opt is any substring
of mnt->mnt_opts.  Redhat 5.0 does not have this libc bug.  Here is an
example program:

#include <stdio.h>
#include <mntent.h>
main()
{
  struct mntent mnt;
  char *cp;
  mnt.mnt_opts = "intr,rw,port=1023,timeo=8,foo=br,retrans=110,indirect,map=/usr/local/AMD/etc/amd.proj,boo";
  cp = hasmntopt(&mnt, "ro");
  printf("cp = %s\n", cp);
  exit(0);
}

It is possible that sufficiently newer version of libc for RH4.2 fix this
problem.


(3) mips-dec-ultrix4.3

Rainer Orth <ro@TechFak.Uni-Bielefeld.DE> reports

[3A] One needs the Kernel Config Files (UDTBIN430) subset installed to
compile am-utils, otherwise essential header files (net/if.h, net/route.h,
rpcsvc/mount.h, rpcsvc/yp_prot.h, rpcsvc/ypclnt.h, sys/proc.h) are
missing.

[3B] It's probably impossible to build am-utils with DEC C on Ultrix V4.3.
This compiler is pseudo-ANSI only.  Maybe the new ANSI C compiler in V4.3A
and beyond will do.  I successfully used gcc 2.8.1.

[3C] You need to build against a recent libhesiod (I used 3.0.2) and
libresolv/lib44bsd (I used BIND 4.9.5-P1).  The resolver routines in
libc seem to cause random memory corruption.  It is necessary to specify
LIBS=-l44bsd.  lib44bsd is a helper library of libresolv used to supply
functions like strdup which are missing on the host system.  This isn't
currently autoconfiscated.

[3D] You need to configure with CONFIG_SHELL=/bin/sh5 /bin/sh5 buildall;
/bin/sh cannot handle the shell functions used in buildall and is both
buggy and slow.

[3E] At least the gcc 2.7.0 fixincludes-mangled <sys/utsname.h> needs a
forward declaration of struct utsname to avoid lots of gcc warnings:

RCS file: RCS/utsname.h,v
retrieving revision 1.1
diff -u -r1.1 utsname.h
--- utsname.h   1995/06/19 13:07:01     1.1
+++ utsname.h   1998/01/27 12:34:26
@@ -59,6 +59,7 @@
 #ifdef KERNEL
 #include "../h/limits.h"
 #else /* user mode */
+struct utsname;
 extern int     uname _PARAMS((struct utsname *));
 #endif
 #define __SYS_NMLN 32


(4) powerpc-ibm-aix4.2.1.0

[4A] "Randall S. Winchester" <rsw AT Glue.umd.edu> reports that for amd to
start, you need to kill and restart rpc.mountd and possibly also make sure
that nfsd is running.  Normally these are not required.

[4B] "Stefan Vogel" <vogel AT physik.unizh.ch> reports that if your amq
executable dump core unexpectedly, then it may be a bug in gcc 2.7.x.
Upgrade to gcc 2.8.x or use IBM's xlC compiler.

[C] Do not link amd with libnsl.  It is buggy and causes amd to core dump
in strlen inside strdup inside svc_register().


(5) *-linux-rh51 (RedHat Linux 5.1)

There's a UDP file descriptor leak in libnsl in RedHat Linux 5.1.  This
library part of glibc2.  Am-utils currently declares redhat 5.1 systems as
having a "broken yp_all" and using an internal, slower, leak-free version.
The leak is known to the glibc maintainers and a fix from them is due soon,
but it is not yet in the glibc-2.0.7-19 RPM.


(6) rs6000-ibm-aix4.1.x

A bug in libc results in an amq binary that doesn't work; amq -v dumps core
in xdr_string.  There is no known fix (source code or vendor patch) at this
time.  (Please let us know if you have a fix; see www.am-utils.org.)


(7) *-aix4.3.2.0

The plock() function will pre-reserve all of the memory up to the maximum
listed in the ulimit.  If the ulimit is infinite, plock() will try to take
all of the system's memory, and fail with ENOMEM (Not Enough Space).
Normally ulimit may be set to a few gigs of max memory usage, but even that
is too much; Amd doesn't need more than a few megs of resident memory size
(depending on the particular usage, number of maps, etc.)  Solution: lower
your ulimit before starting amd.  This can be done inside the ctl-amd
script, but be careful not to limit it too low.  Alternatively, don't use
plock on aix-4.3: set it to plock=no in amd.conf (which is the default if
you do nothing).


(8) *-linux (systems using glibc 2.1, such as RedHat-6.x)

There's a UDP file descriptor leak in the NIS routines in glibc, especially
those that do yp_bind.  Until this is bug fixed, do not set nis_domain in
amd.conf, but let the system pick up the default domain name as set by your
system.  That would avoid using the buggy yp_bind routines in libc.


(9) *-linux (SuSE systems using unfsd)

The user-level nfsd (2.2beta44) on older SuSE Linux systems (and possibly
others) dies with a SEGV when amd tries to contact it for access to a volume
that does not exist, or one for which there is no permission to mount.


(10) *-*-hpux11

If you're using NFSv3, you must install HP patches PHNE_20344 and
PHNE_20371.  If you don't, and you try to use amd with NFSv3 over TCP, your
kernel will panic.


(11) *-linux* (any system using a 2.2.18+ kernel)

The Linux kernels don't support Amd's direct mounts very well, leading to
erratic behavior: shares that don't get remounted after the first timeout,
inability to restart Amd because its mount points cannot be unmounted, etc.
There are some kernel patches on the am-utils Web site, which solve these
problems.  See http://www.am-utils.org/patches/.

Later 2.4.x kernels completely disallow the hack amd was using for direct
mounts, so another solution will have to be found.

Note: the above is for the old-style amd mount_type = nfs. The autofs mounts
don't support direct mounts at all (due to lack of kernel support).

(12) *-aix5.1.0.0 and *-hpux9*

/bin/sh is broken and fails to run the configure script properly. You need
to use /bin/ksh instead. The buildall script will do it for you; if for some
reason you need to run configure directly, run it using 'ksh configure'
instead of just 'configure'.

[12A] *-aix5.2.*

Apparently there is an NFS client side bug in vmount() which causes amd to
hang when it starts (and tries to NFS-mount itself).  According to IBM
engineers, this has to do with partial support code for IPv6: the NFS kernel
code doesn't appear to recognize the sin_family of the amd vmount(),
although amd does the right thing.  The bug doesn't appear to be in 5.1 or
4.3.3.  A fix from IBM is available, APAR number IY41417.

A binary built on 4.3.3 will not work on 5.2, because the kernel ABIs have
changed.

[12C] *-aix*

It is important that you install bos.net.nfs.adt before configuring and
building am-utils.  If you don't, you will get compile-time or
configure-time errors, especially when configure tries to find AIX's
definition of struct nfs_args.

(13) *-linux and *-darwin6.0

Certain linux kernels (2.4.18+ are fine, 2.4.10- are probably bad, those in
between have not been tested) have a bug which causes them to reconnect
broken NFS/TCP connections using unprivileged ports (greater than 1024),
unlike the initial connections which do originate from privileged
ports.  This can upset quite a few NFS servers and causes accesses to the
mounted shares to fail with "Operation not permitted" (EPERM).

The darwin (MacOS X) kernel defaults to using unprivileged ports, but that
can be changed by setting the resvport mount flag (which amd sets by
default).  Nonetheless, if a TCP connection breaks, under certain unclear
circumstances the kernel might "forget" about that flag and start using
unprivileged ports, causing the same EPERM error above.

(14) Solaris

The line "%option" in *.l files may cause Solaris /usr/ccs/bin/lex to abort
with the error "missing translation value."  This is a bug in Solaris lex.

Moreover, both Solaris yacc and lex produce code that does not pass strict
compilation such as "gcc -Wall -Werror".

Use GNU Flex and Bison instead.  You can download ready-made binaries from
www.sunfreeware.com.  Note, however, that sometimes the binaries on
sunfreeware.com don't seem to work, often because they are built against an
older revision of Solaris or build tools.  In that case, build a fresh
version of GNU flex and/or bison from the latest stable sources.  See
http://www.gnu.org/software/flex/ and http://www.gnu.org/software/bison/.

(15) Solaris 8 + patch 10899[34]-xx (18 <= xx < 25) or patch 11260[56]-xx

With this patch, Sun updated the autofs kernel module and automountd
userspace daemon from version 3 to version 4.  They also updated the
/usr/include/rpcsvc/autofs_prot.x file, but forgot to regenerate the
autofs_prot.h file.  Thus, when amd is compiled, it uses the old header and
thinks it should use autofs version 3, when in fact the kernel now supports
(and expects) only version 4.

The workaround is to run 'rpcgen -C -h /usr/include/rpcsvc/autofs_prot.x >
/usr/include/rpcsvc/autofs_prot.h' and completely reconfigure and rebuild
am-utils (removing config.cache before running configure).

The problem is fixed in patch revisions 10899[34]-25 and up.


(16) Linux kernel 2.4+ and lofs mounts

Lofs mounts are not supported by the linux kernel, at all, but since 2.4.0
the kernel supports a similar type of mount called a bind mount.  Its
semantics are closer to those of a hardlink than to those of lofs, and one
of the results is that bind mounts ignore any mount options paseed to them.

Amd uses bind mounts internally to emulate lofs mounts, which means that
lofs mounts on linux will effectively ignore their mount parameters and
inherit whatever options the original filesystem mounted upon had.


(17) autoconf 2.57

If you see configure warnings of the following kind:

configure: WARNING: sys/proc.h: present but cannot be compiled
configure: WARNING: sys/proc.h: check for missing prerequisite headers?
configure: WARNING: sys/proc.h: proceeding with the preprocessor's result
configure: WARNING:     ## ------------------------------------ ##
configure: WARNING:     ## Report this to bug-autoconf@gnu.org. ##
configure: WARNING:     ## ------------------------------------ ##

please ignore them.  They are not real errors, and neither
bug-autoconf@gnu.org nor the am-utils maintainers are interested in hearing
about them.  Autoconf simply tries to do more than we need and attempts to
compile each header in isolation, which fails for many system headers.
That's ok, because we only need to know if a header file exists -- we know
how to use it properly ourselves.

While autoconf does offer a way to specify other files to be included with
the tested header, in order to avoid these warnings, using it would enlarge
the resulting configure script by an order of magnitude, and for no real
gain.  Configure is big enough as it is, we don't need any more useless
baggage in it.

(18) NetBSD 2.0.2, FreeBSD 5.4, OpenBSD 3.7, and quite possibly most other
     BSDs and other OSs (as of September 2005)

Some BSD kernels don't have a way to turn off the NFS attribute cache.  They
don't have a 'noac' mount flag, and setting various cache timeout fields in
struct nfs_args doesn't turn off the attribute cache; instead, it sets the
attribute cache timeout to some internal hard-coded default (usually
anywhere from 5-30 seconds).  If Amd cannot turn off the NFS attribute
cache, under heavy Amd usage, users could get ESTALE errors from automounted
symlinks, or find that those symlinks point to the wrong place.  One
workaround which would minimize this effect is to set auto_attrcache=1 in
your amd.conf, but it doesn't eliminate the problem!  The best solutions are
(1) to use Amd in Autofs mode, if it's supported in your OS, and (2) talk to
your OS vendor to support a true "noac" flag.  See README.attrcache for more
details.

Erez & the am-utils team.