I/O error using autofs "-hosts", "-noauto" special maps

["John D. Baker" <jdbaker%consolidated.net@localhost>]

Growing tired of my workstations wedging when my fileserver is busy and
some process decides to open a file on my 'amd'-managed home directory at
just the wrong time*, I decided to give the recently-added 'autofs' system
a try.

Ordinary direct and indirect maps (point to map file) seem to work fine.

I added:

  /home 	auto_home	-nosuid,intr,bg

to "/etc/auto_master" where "auto_home" contains:

  * 		servername:/path/to/home/&

and that works fine.


The special maps "-hosts" and "-noauto" fail with "I/O error".  When I
was first trying 'autofs', I just used the default "auto_master" which
only defined the special map

  /net 		-hosts 		-nobrowse,nosuid,intr

and this failed as described above when I try to access any machine
which is exporting filesystems via NFS.  Likewise, I converted an NFS
mount in "/etc/fstab" to "noauto" and tried to use the

  /- 		-noauto

special map entry and it likewise failed with "I/O" error when I tried
to access it.

(I've since re-commented the stock "/-" entry and made my own:

  /- 		auto_direct

with the relevant mount point, options, and location and that works
fine.

Any hints for troubleshooting?  The client 'dmesg' only shows:

  [  8126.457286] WARNING: autofs_trigger_one: request for /net/ completed with error 5

and a command accessing a path on a server via the "/net" mountpoint
result in, for example:

  $ ls /net/servername/path/on/server
  ls: /net/servername/path/on/server: Input/output error

I had a "/net" configured under 'amd' and that worked fine.  I've made
no configuration changes other than to no-longer run 'amd'.



* Since netbsd-7, when the fileserver is performing a long-running
  operation on a filesystem it exports and a client has that filesystem
  mounted via 'amd' and opens a file at just the wrong time, it gets stuck
  and all other NFS accesses--including to NFS mounts made in "/etc/fstab"
  hang.

  The most common way to trigger this is to have 'firefox' running on
  the client while the server is performing a 'cvs update' on the a
  'src' or 'pkgsrc' tree residing on the same exported filesystem that
  the client has mounted via 'amd'.

  Can also happen while paging through a slide-show of images in 'xv'
  while the fileserver is linking "gallium.*.so", or running 'fetchmail'
  on the client while the server is running the big 'find ...' operation
  in '/etc/daily', or having 'firefox' open during the '/etc/daily' run
  on the server.

  netbsd-6 and earlier had no issues like the above.

-- 
|/"\ John D. Baker, KN5UKS               NetBSD     Darwin/MacOS X
|\ / jdbaker[snail]consolidated[flyspeck]net  OpenBSD            FreeBSD
| X  No HTML/proprietary data in email.   BSD just sits there and works!
|/ \ GPGkeyID:  D703 4A7E 479F 63F8 D3F4  BD99 9572 8F23 E4AD 1645