Skip to contentSkip to conversation

OOM & [OT] util-linux-2.12e

DISCUSSION [email protected] 13 participants 44 messages Sep 19, 2004 – Sep 22, 2004

lore
Share & tools
lore
#0 lore raw

Just released (on ftp.win.tue.nl in /pub/linux-local/utils/util-linux) util-linux-2.12e.

The reason for this release were complaints that mount and umount OOM the kernel when the number of mounts is large.
And indeed - I tried with 30000 mounts and the OOM-killer killed everything in sight, including X's console, making X exit, killing all remaining processes.

The new versions have been polished a little bit so as not to waste too much memory, and now survive the 30000 mount/umount test for me. Further polishing is needed for the case of large numbers of mounts; when /etc/mtab is not a symlink to /proc/mounts then umount -a has quadratic behaviour (it updates mtab after each unmount) and that gets terribly slow.

About OOM: I am still of the opinion that the default state of the kernel must be one where OOM does not occur and malloc() tells us that we are out of memory. A system that suddenly decides to kill all processes is really very poor and unreliable.
Users can enable other behaviours if they don't care about reliability.

About mount: I wondered whether I should rewrite [u]mounts's handling of /etc/mtab so as to be a bit faster. But it seems a waste of time - /proc/mounts has many advantages: automatically up-to-date, correct also when namespaces are used, much faster. On the other hand, /etc/mtab contains mount options that are sometimes needed later.
If it were possible to store the mount options in the kernel, making them visible in /proc/mounts, then we could forget /etc/mtab altogether.

People have asked repeatedly for a way to mark lines in /etc/fstab so as to make clear that such lines are managed by some GUI or other external program. Labels like "kudzu".
In this release I added a comment convention for /etc/fstab: options can have a part starting with \; - that part is ignored by mount but can be used by other programs managing fstab.

If we would put the mount options in /proc/mounts, and introduced a comment convention (say, the part starting with \: is ignored by the kernel but can be used by programs reading /proc/mounts), then /etc/mtab can die. Comments? Better solutions?

About util-linux and stuff: I have maintained various packages for ten years or so - it may be time to pass things on to someone else. Write to [email protected] if you are interested in taking over or co-maintaining kbd or man or man-pages or util-linux.

Andries

#1 lore raw
Quoted 1 line · On Mon, Sep 20, [email protected] wrote:

then /etc/mtab can die. Comments? Better solutions?

Andries, /etc/mtab is obsolete since the day when /proc/self/mounts was introduced. So, kill it today from your mount binary! TODAY. ...

Then discuss what is still missing from /proc/self/mounts:

- the 'user' option for umount must be handled in some way
- loop mounts do not map to the real filename, users cant open the device node to run losetup on it. I think it will also get relative path names.
- fix all broken apps that still rely on mtab. like GNU df(1)

Signature · 3 lines
USB is for mice, FireWire is for men!

sUse lINUX ag, nÜRNBERG
#2 lore raw
Quoted 1 line · Olaf Hering

- fix all broken apps that still rely on mtab. like GNU df(1)

df does not rely on /etc/mtab. It relies on getmntent.

Andreas.

Signature · 4 lines
Andreas Schwab, SuSE Labs, [email protected]
SuSE Linux AG, Maxfeldstraße 5, 90409 Nürnberg, Germany
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."
#3 lore raw

Hi Andreas :)

* Andreas Schwab <[email protected]> dixit:

Quoted 2 lines
Quoted 1 line

- fix all broken apps that still rely on mtab. like GNU df(1)

df does not rely on /etc/mtab. It relies on getmntent.

Then my GNU df has any problem :???

$ df --version

df (coreutils) 5.2.1
Written by Torbjorn Granlund, David MacKenzie, and Paul Eggert.

Copyright (C) 2004 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

    $ mv /etc/mtab /etc/nomtab
    $ df
df: cannot read table of mounted filesystems: No such file or directory
    $ mv /etc/nomtab /etc/mtab
    $ df
Filesystem           1K-blocks      Used Available Use% Mounted on
/dev/root             78106388  58601124  19505264  76% /
tmpfs                    32768         0     32768   0% /dev/shm
/dev/backup           19530636  15707348   3823288  81% /var/backup

IMHO, it depeds on mtab. Maybe it is configurable thru
'./configure' at build time?

Raúl Núñez de Arenas Coronado

Signature · 2 lines
Linux Registered User 88736
http://www.pleyades.net & http://raul.pleyades.net/
#4 lore raw
Quoted 7 lines · DervishD

Hi Andreas :)

* Andreas Schwab <[email protected]> dixit:

Quoted 2 lines
Quoted 1 line

- fix all broken apps that still rely on mtab. like GNU df(1)

df does not rely on /etc/mtab. It relies on getmntent.

Then my GNU df has any problem :???

No, if any then getmntent.

Andreas.

Signature · 4 lines
Andreas Schwab, SuSE Labs, [email protected]
SuSE Linux AG, Maxfeldstraße 5, 90409 Nürnberg, Germany
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."
#5 lore raw
Quoted 7 lines · Andreas Schwab
Quoted 4 lines · DervishD

Hi Andreas :)

* Andreas Schwab <[email protected]> dixit:

Quoted 8 lines
Quoted 5 lines
Quoted 3 lines
Quoted 1 line

- fix all broken apps that still rely on mtab. like GNU df(1)

df does not rely on /etc/mtab. It relies on getmntent.

Then my GNU df has any problem :???

No, if any then getmntent.

I don't get this. From "man getmntent" it seems that getmntent is just a parser for /etc/mtab, and that you must call "setmntent" with the filename you want to parse.

So if you do "setmntent("/etc/mtab",...)" you're explicitly saying that you want getmntent to use /etc/mtab. This is just a open/read in disguise.

Am I missing something?

Signature · 4 lines
Paulo Marques - www.grupopie.com

To err is human, but to really foul things up requires a computer.
Farmers' Almanac, 1978
#6 lore raw

Hi Paulo :)

* Paulo Marques <[email protected]> dixit:

Quoted 7 lines
Quoted 4 lines
Quoted 3 lines
Quoted 2 lines
Quoted 1 line

- fix all broken apps that still rely on mtab. like GNU df(1)

df does not rely on /etc/mtab. It relies on getmntent.

Then my GNU df has any problem :???

No, if any then getmntent.

I don't get this. From "man getmntent" it seems that getmntent is just a parser for /etc/mtab, and that you must call "setmntent" with the filename you want to parse.

From the code of coreutils 5.2.1, lib/mountlist.c, df uses the
default name for the mounted filesystems table. Under glibc, it uses _PATH_MOUNTED that is, effectively, "/etc/mtab". BTW, 'MOUNTED' is shown in glibc headers as a deprecated alias.

Quoted 4 lines

So if you do "setmntent("/etc/mtab",...)" you're explicitly saying that you want getmntent to use /etc/mtab. This is just a open/read in disguise.
Am I missing something?

Maybe: df (well, lib/mountlist.c) shouldn't try to 'detect' which
kernel are you running, if one with /etc/mtab or a newer one that only supports /proc/mounts (which needs procfs support, BTW, and that can be a problem too). Is glibc who should do all that, and try to open /etc/mtab and /proc/mounts, in that order. Being a userspace app, df should not mess with kernel interface changes. In the worst case, it *could* try to open /etc/mtab (with setmntent, I mean) and if that fails, open /proc/mounts if the operating system is Linux (which can be easily decided at './configure' time).

I'm really not sure about what should be fixed. IMHO, getmntent
should only try to open _PATH_MOUNTED. Any other desired behaviour should be set using setmntent. Just my 0,02 EUR.

Raúl Núñez de Arenas Coronado

Signature · 2 lines
Linux Registered User 88736
http://www.pleyades.net & http://raul.pleyades.net/
#7 lore raw
In reply to Paulo Marques
Quoted 14 lines · Paulo Marques
Quoted 4 lines · Andreas Schwab
Quoted 2 lines · DervishD

Hi Andreas :)

Quoted 8 lines
Quoted 7 lines

* Andreas Schwab <[email protected]> dixit:

Quoted 3 lines
Quoted 1 line

- fix all broken apps that still rely on mtab. like GNU df(1)

df does not rely on /etc/mtab. It relies on getmntent.

Then my GNU df has any problem :???

No, if any then getmntent.

Quoted 8 lines

I don't get this. From "man getmntent" it seems that getmntent is just a parser for /etc/mtab, and that you must call "setmntent" with the filename you want to parse.

So if you do "setmntent("/etc/mtab",...)" you're explicitly saying that you want getmntent to use /etc/mtab. This is just a open/read in disguise.

Am I missing something?

No. You are of course right, but the df sources don't reference /etc/mtab directly, but use _PATH_MOUNTED from <paths.h>.

Andreas.

Signature · 4 lines
Andreas Schwab, SuSE Labs, [email protected]
SuSE Linux AG, Maxfeldstraße 5, 90409 Nürnberg, Germany
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."
#8 lore raw
In reply to Andreas Schwab

Hi Andreas :)

* Andreas Schwab <[email protected]> dixit:

Quoted 4 lines
Quoted 3 lines
Quoted 2 lines
Quoted 1 line

- fix all broken apps that still rely on mtab. like GNU df(1)

df does not rely on /etc/mtab. It relies on getmntent.

Then my GNU df has any problem :???

No, if any then getmntent.

I've seen the code: my excuses. Is the libc which needs fixing,
and it probably have been fixed, I don't have the latest version.

Raúl Núñez de Arenas Coronado

Signature · 2 lines
Linux Registered User 88736
http://www.pleyades.net & http://raul.pleyades.net/
#9 lore raw
In reply to Olaf Hering

Hi,

Quoted 6 lines · Olaf Hering
Quoted 1 line · On Mon, Sep 20, [email protected] wrote:

then /etc/mtab can die. Comments? Better solutions?

Andries, /etc/mtab is obsolete since the day when /proc/self/mounts was introduced. So, kill it today from your mount binary! TODAY. ...

How do you distinguish between manual and automatic loop device setup? How do you filter /proc/mounts for chroot environments?

bye, Roman

#10 lore raw
Quoted 4 lines · Roman Zippel

Hi,

On Mon, 20 Sep 2004, Olaf Hering wrote:

Quoted 8 lines
Quoted 6 lines · On Mon, Sep 20, [email protected] wrote:
Quoted 2 lines

then /etc/mtab can die. Comments? Better solutions?

Andries, /etc/mtab is obsolete since the day when /proc/self/mounts was introduced. So, kill it today from your mount binary! TODAY. ...

How do you distinguish between manual and automatic loop device setup?

-v

Quoted 1 line

How do you filter /proc/mounts for chroot environments?

you have a chroot enviroment without /proc mounted? Then just create /proc/mounts?

Signature · 3 lines
USB is for mice, FireWire is for men!

sUse lINUX ag, nÜRNBERG
#11 lore raw

Hi,

Quoted 3 lines · Olaf Hering
Quoted 1 line

How do you distinguish between manual and automatic loop device setup?

-v

???

Quoted 4 lines
Quoted 1 line

How do you filter /proc/mounts for chroot environments?

you have a chroot enviroment without /proc mounted? Then just create /proc/mounts?

$ cat /proc/mounts
rootfs / rootfs rw 0 0
/dev/root / ext3 rw 0 0
proc /proc proc rw 0 0
devpts /dev/pts devpts rw 0 0
/dev/hda5 /boot ext2 rw 0 0
/dev/hda11 /home ext3 rw 0 0
/dev/hda12 /test ext3 rw 0 0
none /test/proc proc rw 0 0
/dev/hda11 /test/home ext3 rw 0 0
/dev/hda6 / ext3 rw 0 0
none /proc proc rw 0 0
/dev/hda11 /home ext3 rw 0 0
$ mount
/dev/hda6 on / type ext3 (rw)
proc on /proc type proc (rw)
/dev/hda11 on /home type ext3 (rw)
$ ln -sf /proc/mounts /etc/mtab
$ df
Filesystem           1k-blocks      Used Available Use% Mounted on
rootfs                 2480352   1222760   1232392  50% /
/dev/root              2480352   1222760   1232392  50% /
df: `/dev/pts': No such file or directory
/dev/hda5              2480352   1222760   1232392  50% /boot
/dev/hda11            15874464  14187420   1647044  90% /home
df: `/test': No such file or directory
df: `/test/proc': No such file or directory
df: `/test/home': No such file or directory
/dev/hda6              2480352   1222760   1232392  50% /
/dev/hda11            15874464  14187420   1647044  90% /home

bye, Roman

#12 lore raw
Quoted 7 lines · Roman Zippel

Hi,

Quoted 3 lines · Olaf Hering
Quoted 1 line

How do you distinguish between manual and automatic loop device setup?

-v

What do you mean by auto vs. manual? I dont understand what you mean here.

Quoted 1 line

/dev/hda6 / ext3 rw 0 0

df should start with the last / entry. But this will work only if the chroot is on its own mount point, not if you chroot into a subdir of a mount point. It could just ignore the error.

Signature · 3 lines
USB is for mice, FireWire is for men!

sUse lINUX ag, nÜRNBERG
#13 lore raw

Hi,

Quoted 6 lines · Olaf Hering
Quoted 3 lines
Quoted 3 lines
Quoted 1 line

How do you distinguish between manual and automatic loop device setup?

-v

What do you mean by auto vs. manual? I dont understand what you mean here.

$ mount -oloop image /mnt

vs

$ losetup image /dev/loop0
$ mount /dev/loop0 /mnt

What should umount do, when called with /mnt?

Quoted 5 lines
Quoted 1 line

/dev/hda6 / ext3 rw 0 0

df should start with the last / entry. But this will work only if the chroot is on its own mount point, not if you chroot into a subdir of a mount point. It could just ignore the error.

Relying on any specifc ordering in /proc/mounts is broken.

bye, Roman

#14 lore raw
Quoted 11 lines · Roman Zippel

Hi,

Quoted 6 lines · Olaf Hering
Quoted 3 lines
Quoted 3 lines
Quoted 1 line

How do you distinguish between manual and automatic loop device setup?

-v

What do you mean by auto vs. manual? I dont understand what you mean here.

Quoted 8 lines
$ mount -oloop image /mnt

vs

$ losetup image /dev/loop0
$ mount /dev/loop0 /mnt

What should umount do, when called with /mnt?

I have /dev/loop0 in /proc/mounts, umount does nothing wrong here. What do you see?

Quoted 1 line

Relying on any specifc ordering in /proc/mounts is broken.

Thats most likely true. You could bind mount a mtab file. This specific case is tricky.

Signature · 3 lines
USB is for mice, FireWire is for men!

sUse lINUX ag, nÜRNBERG
#15 lore raw

Hi,

Quoted 10 lines · Olaf Hering
Quoted 8 lines
$ mount -oloop image /mnt

vs

$ losetup image /dev/loop0
$ mount /dev/loop0 /mnt

What should umount do, when called with /mnt?

I have /dev/loop0 in /proc/mounts, umount does nothing wrong here.

What exactly is that "nothing wrong"?

Quoted 4 lines
Quoted 1 line

Relying on any specifc ordering in /proc/mounts is broken.

Thats most likely true. You could bind mount a mtab file. This specific case is tricky.

This is a rather common case and IMO it can expect a clean solution.

bye, Roman

#16 lore raw
Quoted 8 lines · Roman Zippel

Hi,

Quoted 4 lines · Olaf Hering
Quoted 4 lines
$ mount -oloop image /mnt

vs

Quoted 8 lines
Quoted 6 lines
Quoted 4 lines
$ losetup image /dev/loop0
$ mount /dev/loop0 /mnt

What should umount do, when called with /mnt?

I have /dev/loop0 in /proc/mounts, umount does nothing wrong here.

What exactly is that "nothing wrong"?

It umounts /mnt for me.

Signature · 3 lines
USB is for mice, FireWire is for men!

sUse lINUX ag, nÜRNBERG
#17 lore raw

Hi,

Quoted 7 lines · Olaf Hering
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
$ mount -oloop image /mnt

vs

$ losetup image /dev/loop0
$ mount /dev/loop0 /mnt
Quoted 7 lines
Quoted 5 lines
Quoted 3 lines
Quoted 1 line

What should umount do, when called with /mnt?

I have /dev/loop0 in /proc/mounts, umount does nothing wrong here.

What exactly is that "nothing wrong"?

It umounts /mnt for me.

What happens to /dev/loop0?

bye, Roman

#18 lore raw
Quoted 1 line · Roman Zippel

What happens to /dev/loop0?

I dont know, whats supposed to happen? losetup -d?

Signature · 3 lines
USB is for mice, FireWire is for men!

sUse lINUX ag, nÜRNBERG
#19 lore raw

Hi,

Quoted 3 lines · Olaf Hering
Quoted 1 line

What happens to /dev/loop0?

I dont know, whats supposed to happen? losetup -d?

Yes, depending on how it was mounted, but that information isn't in /proc/mounts.
(BTW how difficult was it to find this out yourself? Have you even tried it?)

bye, Roman

#20 lore raw
Quoted 4 lines · Roman Zippel

Hi,

On Mon, 20 Sep 2004, Olaf Hering wrote:

Quoted 8 lines
Quoted 3 lines
Quoted 1 line

What happens to /dev/loop0?

I dont know, whats supposed to happen? losetup -d?

Yes, depending on how it was mounted, but that information isn't in /proc/mounts.
(BTW how difficult was it to find this out yourself? Have you even tried it?)

Sure I have tried it. I wonder why umount doesnt do the losetup? I have never looked at the umount sources, nor have I played with all the possible ways of loop mount.

Signature · 3 lines
USB is for mice, FireWire is for men!

sUse lINUX ag, nÜRNBERG
#21 lore raw
Quoted 1 line · Olaf Hering

Sure I have tried it. I wonder why umount doesnt do the losetup?

umount should only call losetup if mount did it too.

Andreas.

Signature · 4 lines
Andreas Schwab, SuSE Labs, [email protected]
SuSE Linux AG, Maxfeldstraße 5, 90409 Nürnberg, Germany
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."
#22 lore raw
Quoted 5 lines · Andreas Schwab
Quoted 1 line · Olaf Hering

Sure I have tried it. I wonder why umount doesnt do the losetup?

umount should only call losetup if mount did it too.

we could reuse -n for that.

Signature · 3 lines
USB is for mice, FireWire is for men!

sUse lINUX ag, nÜRNBERG
#23 lore raw
In reply to Roman Zippel
Quoted 11 lines · Roman Zippel
Quoted 2 lines

What do you mean by auto vs. manual? I dont understand what you mean here.

$ mount -oloop image /mnt

vs

$ losetup image /dev/loop0
$ mount /dev/loop0 /mnt

What should umount do, when called with /mnt?

Does the kernel crash and burn if you pass the filesystem an option it does not know about on a mount? If not then just have mount pass all the options it gets to the kernel, the fs weeds out what it likes and the full thing gets stored for use in /proc/self/mounts.

That would mean the above would have loop stored for the first and not for the second and so umount would know what to do with each case.

(from another bit of this thread) This would also mean that the user keyword can be stored too.

Signature · 1 line
    Red herrings strewn hither and yon.
#24 lore raw
Quoted 2 lines · CaT

Does the kernel crash and burn if you pass the filesystem an option it does not know about on a mount?

It will refuse to mount.

Andreas.

Signature · 4 lines
Andreas Schwab, SuSE Labs, [email protected]
SuSE Linux AG, Maxfeldstraße 5, 90409 Nürnberg, Germany
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."
#25 lore raw
In reply to Roman Zippel
Quoted 7 lines · Roman Zippel
Quoted 4 lines
Quoted 1 line

then /etc/mtab can die. Comments? Better solutions?

Andries, /etc/mtab is obsolete since the day when /proc/self/mounts was introduced. So, kill it today from your mount binary! TODAY. ...

How do you distinguish between manual and automatic loop device setup? How do you filter /proc/mounts for chroot environments?

My apologies if something like this is already in the source. Just throwing out an idea to the wolves. ;)

Quoted 1 line

From /proc/self/mounts I take it that that's meant to be a list of

mountpoints the requesting process sees. So... why not have instances of mountlists. The default is what the kernel starts with and each process inherits. The chroot creates a 2nd instance so that any processes spawned from that one see the mountlist from that instance.

Each process would, obviously, inherit the mountlist instance from its parent.

The hassle happens when you want to deal with the process that began an instance dieing without taking its children out with it. One way to deal with it would be to introduce a counter, but that would require locking to keep it serialised.

Signature · 1 line
    Red herrings strewn hither and yon.
#26 lore raw
In reply to Olaf Hering

Hi Olaf :)

* Olaf Hering <[email protected]> dixit:

Quoted 3 lines
Quoted 1 line

then /etc/mtab can die. Comments? Better solutions?

Andries, /etc/mtab is obsolete since the day when /proc/self/mounts was introduced. So, kill it today from your mount binary! TODAY. ...

Bad idea... ;))) I upgraded my 'mount' yesterday. I was using a
mount from Debian, from 1998 more or less, that worked flawlessly except for the '--bind' feature and things like those. I used /etc/mtab as a symlink to /proc/mounts, and all worked OK except for the double root entry and the need to manually call losetup to delete unused /dev/loop entries.

But after the upgrade I no longer could umount a filesystem that
I mounted as 'user', because the device is a symlink and the 'user' option is not stored in /proc/mounts. So my problems were:

Quoted 5 lines

- the 'user' option for umount must be handled in some way
- loop mounts do not map to the real filename, users cant open the device node to run losetup on it. I think it will also get relative path names.
- fix all broken apps that still rely on mtab. like GNU df(1)

Exactly!!! I will add the double root filesystem entry (one from
the device, the other one being 'rootfs / rootfs rw 0 0'). Until all these problems are solved, is not a good idea to remove /etc/mtab :(

I have another issue with a chroot environment I have, which has
its own copy of proc and devpts mounted. Those appear in /proc/mounts twice, but not in /etc/mtab (they appear in the /etc/mtab copy of the chroot env).

Raúl Núñez de Arenas Coronado

Signature · 2 lines
Linux Registered User 88736
http://www.pleyades.net & http://raul.pleyades.net/
#27 lore raw
Quoted 20 lines · DervishD

Hi Olaf :)

* Olaf Hering <[email protected]> dixit:

Quoted 7 lines
Quoted 3 lines

then /etc/mtab can die. Comments? Better solutions?

Andries, /etc/mtab is obsolete since the day when /proc/self/mounts was introduced. So, kill it today from your mount binary! TODAY. ...

Bad idea... ;))) I upgraded my 'mount' yesterday. I was using a
mount from Debian, from 1998 more or less, that worked flawlessly except for the '--bind' feature and things like those. I used /etc/mtab as a symlink to /proc/mounts, and all worked OK except for the double root entry and the need to manually call losetup to delete unused /dev/loop entries.

Quoted 5 lines

But after the upgrade I no longer could umount a filesystem that
I mounted as 'user', because the device is a symlink and the 'user' option is not stored in /proc/mounts. So my problems were:

Using a mtab that is a link to /proc/mounts fails with quota too. Quta tools read /etc/mtab looking for "usrquota" and or "grpquota" mount options. These appear in a normal /etc/mtab but not in /proc/mounts,

so the tools gets the mistaken impression that no fs actually use quotas.

Helge Hafting

#28 lore raw
Quoted 3 lines · Helge Hafting

Using a mtab that is a link to /proc/mounts fails with quota too. Quta tools read /etc/mtab looking for "usrquota" and or "grpquota" mount options. These appear in a normal /etc/mtab but not in /proc/mounts,

I have never played with quota. But: does the kernel or a userland tool if quota is active for a mount point? smells like a kernel bug.

Signature · 3 lines
USB is for mice, FireWire is for men!

sUse lINUX ag, nÜRNBERG
#29 lore raw
Quoted 4 lines · Olaf Hering

On Mon, Sep 20, Helge Hafting wrote:

Quoted 10 lines
Quoted 5 lines

Using a mtab that is a link to /proc/mounts fails with quota too. Quta tools read /etc/mtab looking for "usrquota" and or "grpquota" mount options. These appear in a normal /etc/mtab but not in /proc/mounts,

I have never played with quota. But: does the kernel or a userland tool if quota is active for a mount point? smells like a kernel bug.

The kernel must know that quota is in use, or it'd be unable to refuse the syscalls when someone tries to go over his quota.

From "man mount":
grpquota / noquota / quota / usrquota
These options are accepted but ignored. (However, quota
utili‐
ties may react to such strings in /etc/fstab.)

quota utilities indeed react to such strings in /etc/mtab too.

Qutas aren't actually enabled when the mount options are used, they are enabled when the "quotaon" tool runs. I guess it uses some special syscall or ioctl to really turn quota on.

Doing it at mount time instead, byt actually using those options, seems saner to me. But I guess they had their reasons. . .

Helge Hafting

#30 lore raw
Quoted 11 lines · Helge Hafting
Quoted 9 lines · Olaf Hering
Quoted 4 lines · Helge Hafting

Using a mtab that is a link to /proc/mounts fails with quota too. Quta tools read /etc/mtab looking for "usrquota" and or "grpquota" mount options. These appear in a normal /etc/mtab but not in /proc/mounts,

I have never played with quota. But: does the kernel or a userland tool if quota is active for a mount point? smells like a kernel bug.

- to make the quota active (enable it), the mount option is required

- to display an enabled quota as mount option, the quota on that 'mount point' has to be enabled

chicken egg thing, eh?

besides that, not every mountpoint can support quota and quota should (must) not be enabled at mount time because before the quota is enabled, the quota hash has to be initialized to the current usage ...

Quoted 2 lines

The kernel must know that quota is in use, or it'd be unable to refuse the syscalls when someone tries to go over his quota.

yep, after 'enabling' quota on a specific filesystem the kernel knows which quota is enabled ...

Quoted 11 lines

From "man mount":
grpquota / noquota / quota / usrquota
These options are accepted but ignored. (However, quota
utili???
ties may react to such strings in /etc/fstab.)

quota utilities indeed react to such strings in /etc/mtab too.

Qutas aren't actually enabled when the mount options are used, they are enabled when the "quotaon" tool runs. I guess it uses some special syscall or ioctl to really turn quota on.

yes, and basically the 'mount options' are not stored by the kernel in any way, so it would be simple to remove that check in the quota tools and replace them by something else ...

Quoted 2 lines

Doing it at mount time instead, byt actually using those options, seems saner to me. But I guess they had their reasons. . .

yes, quota calculation, see above ...

HTH,
Herbert

Quoted 8 lines

Helge Hafting

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#31 lore raw
Quoted 21 lines · Herbert Poetzl
Quoted 13 lines · Helge Hafting
Quoted 9 lines · Olaf Hering
Quoted 5 lines · Helge Hafting

Using a mtab that is a link to /proc/mounts fails with quota too. Quta tools read /etc/mtab looking for "usrquota" and or "grpquota" mount options. These appear in a normal /etc/mtab but not in /proc/mounts,

Quoted 4 lines
Quoted 4 lines

I have never played with quota. But: does the kernel or a userland tool if quota is active for a mount point? smells like a kernel bug.

Quoted 9 lines

- to make the quota active (enable it), the mount option is required

- to display an enabled quota as mount option, the quota on that 'mount point' has to be enabled

chicken egg thing, eh?

Chicken egg by design perhaps, but I can't see
why it have to be that way.

Quoted 6 lines

besides that, not every mountpoint can support quota and quota should (must) not be enabled at mount time because before the quota is enabled, the quota hash has to be initialized to the current usage ...

A better way:
Enable the quota at mount time. If the system isn't ready for that, i.e. the quota files aren't created/updated - then refuse the mount. (Alternatively, mount without quota and log a complaint.) The administrator can then mount without quota, run checkquota, and mount /fs -o remount,usrquota to turn quota on.

In other words:
quotaon becomes mount -o remount,usrquota
quotaoff becomes mount -o remount,nousrquota
(And/or grpquota of course)

Does it have to be any more complicated than that?

Quoted 8 lines
Quoted 4 lines

Doing it at mount time instead, byt actually using those options, seems saner to me. But I guess they had their reasons. . .

yes, quota calculation, see above ...

What I don't get is why we have to mount with quota options that aren't acutally used, and then turn quota on. Why not mount without quota, and then remount with quota options when enabling quota for the first time?

The common case should be a fs that was shut down cleanly and was mounted with quota the last time it was used. So it should be able to mount directly with quota on, because all the on-disk quota information is valid and up to date.

Helge Hafting

#32 lore raw
Quoted 41 lines · Helge Hafting
Quoted 14 lines · Herbert Poetzl
Quoted 11 lines · Helge Hafting
Quoted 9 lines · Olaf Hering
Quoted 5 lines · Helge Hafting

Using a mtab that is a link to /proc/mounts fails with quota too. Quta tools read /etc/mtab looking for "usrquota" and or "grpquota" mount options. These appear in a normal /etc/mtab but not in /proc/mounts,

I have never played with quota. But: does the kernel or a userland tool if quota is active for a mount point? smells like a kernel bug.

Quoted 8 lines

- to make the quota active (enable it), the mount option is required

- to display an enabled quota as mount option, the quota on that 'mount point' has to be enabled

chicken egg thing, eh?

Chicken egg by design perhaps, but I can't see
why it have to be that way.

Quoted 6 lines

besides that, not every mountpoint can support quota and quota should (must) not be enabled at mount time because before the quota is enabled, the quota hash has to be initialized to the current usage ...

A better way:
Enable the quota at mount time. If the system isn't ready for that, i.e. the quota files aren't created/updated - then refuse the mount. (Alternatively, mount without quota and log a complaint.) The administrator can then mount without quota, run checkquota, and mount /fs -o remount,usrquota to turn quota on.

Quoted 6 lines

In other words:
quotaon becomes mount -o remount,usrquota
quotaoff becomes mount -o remount,nousrquota
(And/or grpquota of course)

Does it have to be any more complicated than that?

probably not, but I guess you have to discuss
this with the quota maintainer ... ;)

best,
Herbert

Quoted 18 lines
Quoted 5 lines
Quoted 2 lines

Doing it at mount time instead, byt actually using those options, seems saner to me. But I guess they had their reasons. . .

yes, quota calculation, see above ...

What I don't get is why we have to mount with quota options that aren't acutally used, and then turn quota on. Why not mount without quota, and then remount with quota options when enabling quota for the first time?

The common case should be a fs that was shut down cleanly and was mounted with quota the last time it was used. So it should be able to mount directly with quota on, because all the on-disk quota information is valid and up to date.

Helge Hafting

#33 lore raw
In reply to DervishD

Salut,

Quoted 6 lines · DervishD

Bad idea... ;))) I upgraded my 'mount' yesterday. I was using a
mount from Debian, from 1998 more or less, that worked flawlessly except for the '--bind' feature and things like those. I used /etc/mtab as a symlink to /proc/mounts, and all worked OK except for the double root entry and the need to manually call losetup to delete unused /dev/loop entries.

I keep a mvmount and bindmount program on my farm for that exact purpose. :)

Tonnerre

#34 lore raw
In reply to Olaf Hering

Hi Olaf ;),

Quoted 4 lines

Andries, /etc/mtab is obsolete since the day when /proc/self/mounts was introduced. So, kill it today from your mount binary! TODAY. ...

Then discuss what is still missing from /proc/self/mounts:

I agree with your approach Olaf. Better _fix_ first mount to remove totally mtab support, and after put in /proc/self/mount the additional info that previously only existed in /etc/mtab. Or i'm afraid that it will never be removed...

Heck, i should learn myself some kernel programming and add the new options to /proc/self/mounts. I'm really tired of having an /etc/mtab file in my system...

bye

Signature · 4 lines
David Gómez

"The question of whether computers can think is just like the question of
whether submarines can swim." -- Edsger W. Dijkstra
#35 lore raw
In reply to [email protected]

Hi Andries :)

* [email protected] <[email protected]> dixit:

Quoted 4 lines

If we would put the mount options in /proc/mounts, and introduced a comment convention (say, the part starting with \: is ignored by the kernel but can be used by programs reading /proc/mounts), then /etc/mtab can die. Comments? Better solutions?

If you add a comment convention to /proc/mounts so you can use it
as a substitute for /etc/mtab, you will probably break the apps that use /etc/mtab. I was wondering, then... does the kernel *read* /proc/mounts contents? If the answer is no, then you can add all syntactic noise you want to /proc/mounts, exporting options needed for userspace programs, with no problem. You can make /proc/mounts to look like /etc/mtab. That will solve most of the problems.

If the kernel needs to read /proc/mounts, then you have a
problem: you will need /etc/mtab as long as you have to use loop devices, user mounts, etc.

Raúl Núñez de Arenas Coronado

Signature · 2 lines
Linux Registered User 88736
http://www.pleyades.net & http://raul.pleyades.net/
#36 lore raw
Quoted 1 line · DervishD

does the kernel *read* /proc/mounts contents?

/proc/mounts is kernel-generated on the fly (it's alive only during the read() call).

Xav

#37 lore raw

Hi Xavier :)

* Xavier Bestel <[email protected]> dixit:

Quoted 3 lines
Quoted 1 line

does the kernel *read* /proc/mounts contents?

/proc/mounts is kernel-generated on the fly (it's alive only during the read() call).

Then you can cripple it with any extra contents you want, am I
wrong? The kernel won't mind...

Raúl Núñez de Arenas Coronado

Signature · 2 lines
Linux Registered User 88736
http://www.pleyades.net & http://raul.pleyades.net/
#38 lore raw
Quoted 9 lines · DervishD

Hi Xavier :)

* Xavier Bestel <[email protected]> dixit:

Quoted 3 lines
Quoted 1 line

does the kernel *read* /proc/mounts contents?

/proc/mounts is kernel-generated on the fly (it's alive only during the read() call).

Then you can cripple it with any extra contents you want, am I
wrong? The kernel won't mind...

Sure.

Xav

#39 lore raw
Quoted 4 lines · Xavier Bestel
Quoted 3 lines · DervishD

Hi Xavier :)

* Xavier Bestel <[email protected]> dixit:

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
Quoted 1 line

does the kernel *read* /proc/mounts contents?

/proc/mounts is kernel-generated on the fly (it's alive only during the read() call).

Then you can cripple it with any extra contents you want, am I
wrong? The kernel won't mind...

Sure.

well, yes and no, actually all those funny 'mount' options provided by the kernel 'on the fly' are just bits which get set (or cleared) after 'parsing' the mount options at mount time ...

this means:

- bad mount option, no mount
- ignored mount option, no flag
- known mount option, flag is set (or cleared)

of course it would be possible to 'extend' the kernel by some 'important' flags which are _not_ used by the kernel itself ...

HTH,
Herbert

Quoted 7 lines

Xav

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#40 lore raw
In reply to [email protected]
Quoted 6 lines · [email protected] wrote:

People have asked repeatedly for a way to mark lines in /etc/fstab so as to make clear that such lines are managed by some GUI or other external program. Labels like "kudzu".
In this release I added a comment convention for /etc/fstab: options can have a part starting with \; - that part is ignored by mount but can be used by other programs managing fstab.

How about implementing /etc/fstab option parsing code that is compatible with existing libc /etc/fstab parsing code:

defaults,noauto,comment=kudzu,rw

                ^^^^^^^^^^^^^
Signature · 1 line
Jari Ruusu  1024R/3A220F51 5B 4B F9 BB D3 3F 52 E9  DB 1D EB E3 24 0E A9 DD
#41 lore raw
Quoted 5 lines · Jari Ruusu

How about implementing /etc/fstab option parsing code that is compatible with existing libc /etc/fstab parsing code:

defaults,noauto,comment=kudzu,rw

                ^^^^^^^^^^^^^

Is there such libc parsing code? Can you tell me which libc? Which file? Invoked for what function calls?

Andries

#42 lore raw
Quoted 9 lines · Andries Brouwer
Quoted 5 lines · Jari Ruusu

How about implementing /etc/fstab option parsing code that is compatible with existing libc /etc/fstab parsing code:

defaults,noauto,comment=kudzu,rw

                ^^^^^^^^^^^^^

Is there such libc parsing code? Can you tell me which libc? Which file? Invoked for what function calls?

man setmntent

SYNOPSIS

       #include <stdio.h>
       #include <mntent.h>
       FILE *setmntent(const char *filename, const char *type);
       struct mntent *getmntent(FILE *filep);
       int addmntent(FILE *filep, const struct mntent *mnt);
       int endmntent(FILE *filep);
       char *hasmntopt(const struct mntent *mnt, const char *opt);

Mount is not the only piece of code that parses fstab. Even swapon and swapoff programs that are part of util-linux were broken by this change. The 'comment=fubar' mount option requires two line change to mount.c, and most of all, does not break any existing fstab parsing code.

Your fstab options comment change means that all code that parses fstab needs to be modified to understand the new comment separator sequence. If they are not modified, they will mis-parse the comment separator sequence and mis-parse options beyond the comment separator sequence.

Not directly related to above, but you need to release new version of util-linux soon anyway. You intruduced this type of gems to util-linux-2.12e

--- util-linux-2.12d/mount/lomount.c Sun Jul 11 20:26:41 2004
+++ util-linux-2.12e/mount/lomount.c Fri Sep 17 01:28:58 2004

+ memset(&loopinfo64, 0, sizeof(loopinfo64));

close (fd);
---------------^^
+
+ if (i) {
+ ioctl (fd, LOOP_CLR_FD, 0);
-----------------------^^
+ return 1;
+ }
+
if (verbose > 1)
printf(_("set_loop(%s,%s,%llu): success\n"),
device, file, offset);
return 0;
-
- fail:
- (void) ioctl (fd, LOOP_CLR_FD, 0);
- close (fd);
- return 1;
}

Signature · 1 line
Jari Ruusu  1024R/3A220F51 5B 4B F9 BB D3 3F 52 E9  DB 1D EB E3 24 0E A9 DD
#43 lore raw
Quoted 10 lines · Jari Ruusu
Quoted 8 lines
Quoted 5 lines

How about implementing /etc/fstab option parsing code that is compatible with existing libc /etc/fstab parsing code:

defaults,noauto,comment=kudzu,rw

                ^^^^^^^^^^^^^

Is there such libc parsing code? Can you tell me which libc? Which file? Invoked for what function calls?

man setmntent

I thought you were pointing me at a comment convention. But as far as I can see there is none today.

If on the other hand the above is the suggestion for a comment convention, then I like it, but it has the disadvantage that it takes part of the namespace of filesystem mount options. My version did not do that.

Quoted 2 lines

Not directly related to above, but you need to release new version of util-linux soon anyway.

OK - done.

Review

Quoted 6 lines

You intruduced this type of gems to util-linux-2.12e

(quoted diff) +3 -0
 	close (fd);
+
+	if (i) {
+		ioctl (fd, LOOP_CLR_FD, 0);

A merge error. Thanks! Fixed.

Andries

[PS - you write "type of", vaguely suggesting that you could point out more flaws. If you can, please do.]