|
Документ взят из кэша поисковой машины. Адрес
оригинального документа
: http://rtm-cs.sinp.msu.ru/manual/howto/mini/Automount.html
Дата изменения: Thu May 4 20:16:33 2000 Дата индексирования: Mon Oct 1 21:25:06 2012 Кодировка: |
don@sabotage.org
/auto/grumblesmurf is mounted and kill -SIGUSR1 won't unmount it.
---timeout instead of --timeout
Automouting is the process where mounting (and unmounting) of certain filesystems is done automatically by a daemon. If the filesystem is unmounted, and a user attempts to access it, it will be automatically (re)mounted. This is especially useful in large networked environments and for crossmounting filesystems between a few machines (especially ones which are not always online). It may also be very useful for removable devices, or a few other uses, such as easy switching between a forced-on ascii conversion mount of a dos filesystem and a forced-off ascii conversion mount of the same dos fs.
There are two types of automounters in linux; AMD and autofs. AMD is the automount daemon, and supposedly works like the SunOS AMD. It is implemented in user space, meaning it's not part of the kernel. It's not necessary for the kernel to understand automounting if you NFS mount to the local host, through the AMD daemon, which routes all automount filesystem traffic through the NFS system. Autofs is a newer system assisted by the kernel, meaning that the kernel's filesystem code knows where the automount mount points are on an otherwise normal underlying fs, and the automount program takes it from there. Only autofs will be described in this mini-howto.
Because autofs is implemented in kernel-space, your kernel must have support compiled in. In 2.0.xx it is an experimental option, but appears to be quite stable. In 2.2.xx it is a normal option.
The automount program and its configuration files are also necessary; using the rpms (from RedHat, as part of the install) is a great way to go. The automount program should be started by an rc script under the /etc/rc.d/init.d directory . The rpm installs this, but you will need to make sure it gets started, either by linking it from your rc?.d directory, using Redhat's control-panel, or on another distribution by getting the thing started anyway you care to. Non-rpm distros will have to do whatever's applicable to their system. And don't look too hard at what the rc script does; if you're reading this howto you probably don't want to know.
Installing the RPM's will get you to this point easily enough, but here's the part you might not be sure about if you haven't done this before.
There are two files in /etc, one called auto.master and one called auto.misc.
My auto.master looks like this:
/auto /etc/auto.misc --timeout 60
The first entry is not the mount point. It's where the set of mount points (found in the second entry) are going to be. The third option says that the mounted filesystems can try to unmount themselves 60 seconds after use. They can't unmount if being used, of course.
Auto.misc is a "map file". The map file can have any name; this one is named auto.misc because it originally controlled /misc. Multiple map files can be defined in auto.master. My auto.misc looks like this:
kernel -ro,soft,intr ftp.kernel.org:/pub/linux cd -fstype=iso9660,ro :/dev/cdrom zip -fstype=auto :/dev/hdd4 floppy -fstype=vfat :/dev/fd0
The first column (the "key") is the mount point. In this case it would be /auto/floppy or whatever. The middle set are the options; read the mount manpage for details on this. And the last column specifies where the fs comes from. The "kernel" entry is supposed to be an NFS mount. The : on all the other lines means its a local device.
Some of you may be eyeing that 60 second timeout and thinking, that's a long
time to wait to eject a floppy.. Maybe I'll just sync the disks and pop
it out mounted and nobody will notice. Let me suggest saner alternatives.
First of all, you can change the timeout. But that could be a little
inefficient; telling the system to unmount stuff after only 15 seconds or
whatever. Depending on your setup, you may be able to simply run the umount
command as a normal user. But there is actually a way to ask the automount program to umount. If
you send (with the program kill) the signal SIGUSR1 to the automount process,
it will unmount everything it can. But before people start making unmount
buttons on their window managers, there's a little problem.
The automount process is run by root, and it will only accept signals from root. Half of the reason you're probably doing automounting is so you can mount an unmount *without* being root. It would be easy to make a suid-root C program which does the dirty deed. However, by using sudo it is possible to allow users to send the proper kill signal. The only problem is that sudo will not let you use ` to process subcommands, which you would have to do to find the current PID. You should have a program called killall, which will let you do this: (thanks for the suggestions)
ALL ALL=NOPASSWD:/usr/bin/killall -USR1 automountOtherwise, you would have to allow your users to send -SIGUSR1 to all processes. That has various effects on programs; it will recycle some window managers, but kills xemacs. So here's hoping there's no buffer overruns in killall...
If automount is setup properly, whatever mount point you're looking for will be there if you try and use it, even though you don't see it when not in use. If you're browsing the directory with a graphical tool, you may need to type in the name manually; most programs will try what you give it, and the drive will be mounted before it notices. Unfortunately not being able to choose from the available invisible mount points is probably the major drawback of autofs. If it really bugs you, edit the configuration files. (Hint, the ones that end in .c for "configuration")
One workaround several people have tried is to create symbolic links to where automount will create something once it's mounted. This will likely prevent the program from complaining a directory doesn't exist (if the mount works, that is) but careless directory listings will cause filesystems to be mounted.
The df command. mount with no options will do the same, plus show the options its mounted with.
This is not a problem with automount. As of this writing, the "auto" fs type does not attempt a vfat mount before it successfully mounts an msdos filesystem. VFAT is the Win95 and WinNT long filenames crammed into a FAT/MSDOS filesystem.
According to one of the authors of mount, since mount is only a wrapper around a system call which must specify the filesystem type, it's still the responsibility of the user to come up with the fs type. Having mount take a list of filesystems to try in order, rather than the current "heuristic" is under consideration. Some users have simply not compiled msdos into the kernel; this prevents it from being tested prior to vfat. This will work for most people; a few actually need msdos fs and it caused me quite some frustration to not have a module handy when I actually needed it.
I'm sure that if anyone wants to go to the effort of finding the owner(s) of the mount program, your comments would be welcome. So unless you don't compile msdos in, for now this means that you can't mount vfat unless you give up the ability to autodetect all other fs's. Hopefully it will be configurable someday. In the mean time, feel free to create multiple mount points with different fs types specified.
/auto/grumblesmurf is mounted and kill -SIGUSR1 won't unmount it.
It's being used by something. Root probably can't manually unmount it either. If you're the one who caused it to be mounted (i.e. it can't be someone else using it) look around for a shell that might be in that directory. If there are none, look for something else (particularly something that might have gone though that directory like a directory browser) that might have left an invisible foot in the door so to speak. If you've given up looking, try using the fuser program.
Oooh. Well, out of a statistical sample of only one person, none of the results were positive. You have been warned. If you want /grumblesmurf, then I suggest a symbolic link. Much safer.
Not as far as I know. Try using one map file, with specific options for individual entries.
---timeout instead of --timeout
Uh. Ok, I've made a note about it. Another solution to "timeout not working" problems would be to add a -t time option to the autofs script.
Check the man page for mount for some of the options, such as setting the uid=value or umask=value options. One option that appears to be missing for FAT filesystems is mode=value. Sorry. Check in with the people who do mounting.
Not me. I didn't have anything to do with it. I just wanted to bring everyone's attention to what a great job had been done with autofs, and how easy it is to use. Compared to the original perpetrators of AMD (Hint, they sell an overpriced unice with prehistoric versions of free tools) the autofs is very well documented and the implementors have my sincere thanks. Everything is stamped copyright Transmeta so sorry I can't provide a credits list, but I would bet Peter Anvin is responsible for quite a bit of it. Peter also held a session on autofs at linuxworldexpo on March 3, 1999.
There's a autofs tutorial at http://www.linuxhq.com/lg/issue24/nielsen.html. See also am-utils at http://www.cs.columbia.edu/~ezk/am-utils
(Thanks for these URLs)