[gentoo-user] Weird "df" output

[gentoo-user] Weird "df" output

[waltdnes]

  I'll admit that my system setup is a bit unusual.  A long time ago, in
a place far away, hard drives were small, compared to today's standards.
The usual unix practice of multiple seprate partitions was not feasable
for me, but I did want to keep root on its own partition.  So I
compromised with a small / partition, with empty /home, /opt, /var,
/usr, and /tmp directories.  Their real equivalents are bind-mounted
from a much larger partition.  I just re-did my oldest machine.  It has
a primary partition 1, which covers the entire hard drive.  The /
partion on /dev/sda5 is approximately 500 megabytes (YES!).  There's a
3.8 gigabyte swap partion /dev/sda6, and the rest of the drive is
/dev/sda7.  Here's the relevant portion of /etc/fstab...

/dev/sda5               /           ext2     noatime,async      0 1
/dev/sda7               /home       ext3     noatime,async      0 1
/home/bindmounts/opt    /opt        auto     bind               0 0
/home/bindmounts/var    /var        auto     bind               0 0
/home/bindmounts/usr    /usr        auto     bind               0 0
/home/bindmounts/tmp    /tmp        auto     bind               0 0
/dev/sda6               none        swap     sw                 0 0

...and the output from "df"...

Filesystem     1K-blocks      Used Available Use% Mounted on
/dev/root         495944     49416    420928  11% /
devtmpfs           10240         0     10240   0% /dev
tmpfs             310080       356    309724   1% /run
shm              1550384         0   1550384   0% /dev/shm
cgroup_root        10240         0     10240   0% /sys/fs/cgroup
/dev/sda7      476205120 292365556 159643008  65% /opt

...showing /dev/sda7 mounted on /opt !?!?  mc (Midnight Commander) shows
152 of 454 gigabytes free on all of /home, /opt, /var, /usr, and /tmp,
which is correct, since they're all really bindmounts from /dev/sda7.
The / partition (/dev/sda5) has 411 of 484 megabytes free.  The machine
works OK, but the "df" output is a head-scratcher.  I've re-booted a
couple of times, with no change.

-- 
Walter Dnes <waltdnes@waltdnes.org>
I don't run "desktop environments"; I run useful applications

Re: [gentoo-user] Weird "df" output

[Jc García]

2015-11-25 16:10 GMT-06:00  <waltdnes@waltdnes.org>:

> /dev/sda7.  Here's the relevant portion of /etc/fstab...
...

> /home/bindmounts/opt    /opt        auto     bind               0 0

Why not use regular partiontions instand of bindmounts, you are just
doing weird stuff seems to me.

Re: [gentoo-user] Weird "df" output

[Rich Freeman]

On Wed, Nov 25, 2015 at 5:30 PM, Jc García <jyo.garcia@gmail.com> wrote:
> 2015-11-25 16:10 GMT-06:00  <waltdnes@waltdnes.org>:
>
>> /dev/sda7.  Here's the relevant portion of /etc/fstab...
> ...
>
>> /home/bindmounts/opt    /opt        auto     bind               0 0
>
> Why not use regular partiontions instand of bindmounts, you are just
> doing weird stuff seems to me.
>

Then all your free space gets fragmented across a dozen different
filesystems and you end up micromanaging everything.

I use one filesystem for root, and another filesystem for everything
else.  I mostly use symlinks, but at times Gentoo has been picky about
symlinks so in some cases I've used bind mounts instead.  When my root
was really tiny and almost everything was elsewhere I had a ton of
bind mounts.  Now that all the places portage touches are on root I
don't have to deal with that as much.

I think the df behavior above is somewhat normal when you have bind mounts.

-- 
Rich

[gentoo-user] Re: Weird "df" output

[Jonathan Callen]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA512

On 11/25/2015 05:10 PM, waltdnes@waltdnes.org wrote:
>   I'll admit that my system setup is a bit unusual.  A long time ago, in
> a place far away, hard drives were small, compared to today's standards.
> The usual unix practice of multiple seprate partitions was not feasable
> for me, but I did want to keep root on its own partition.  So I
> compromised with a small / partition, with empty /home, /opt, /var,
> /usr, and /tmp directories.  Their real equivalents are bind-mounted
> from a much larger partition.  I just re-did my oldest machine.  It has
> a primary partition 1, which covers the entire hard drive.  The /
> partion on /dev/sda5 is approximately 500 megabytes (YES!).  There's a
> 3.8 gigabyte swap partion /dev/sda6, and the rest of the drive is
> /dev/sda7.  Here's the relevant portion of /etc/fstab...
>
> /dev/sda5               /           ext2     noatime,async      0 1
> /dev/sda7               /home       ext3     noatime,async      0 1
> /home/bindmounts/opt    /opt        auto     bind               0 0
> /home/bindmounts/var    /var        auto     bind               0 0
> /home/bindmounts/usr    /usr        auto     bind               0 0
> /home/bindmounts/tmp    /tmp        auto     bind               0 0
> /dev/sda6               none        swap     sw                 0 0
>
> ...and the output from "df"...
>
> Filesystem     1K-blocks      Used Available Use% Mounted on
> /dev/root         495944     49416    420928  11% /
> devtmpfs           10240         0     10240   0% /dev
> tmpfs             310080       356    309724   1% /run
> shm              1550384         0   1550384   0% /dev/shm
> cgroup_root        10240         0     10240   0% /sys/fs/cgroup
> /dev/sda7      476205120 292365556 159643008  65% /opt
>
> ...showing /dev/sda7 mounted on /opt !?!?  mc (Midnight Commander) shows
> 152 of 454 gigabytes free on all of /home, /opt, /var, /usr, and /tmp,
> which is correct, since they're all really bindmounts from /dev/sda7.
> The / partition (/dev/sda5) has 411 of 484 megabytes free.  The machine
> works OK, but the "df" output is a head-scratcher.  I've re-booted a
> couple of times, with no change.
>

df reads /proc/self/mountinfo to get the list of mount points. This file
will probably look something like this for your system (exact numbers,
order, and contents may vary):

1 0 4:0 / / rw,noatime,async - ext2 /dev/root rw
2 1 0:1 / /sys rw,nosuid,nodev,noexec,relatime - sysfs sysfs rw
3 1 0:2 / /proc rw,nosuid,nodev,noexec,relatime - proc proc rw
4 1 0:3 / /dev rw,nosuid - devtmpfs devtmpfs rw,size=10240k,mode=755
5 1 0:4 / /run rw,nosuid,nodev - tmpfs tmpfs rw,size=310080k,mode=755
5 4 0:5 / /dev/shm rw,nosuid,nodev - shm tmpfs rw
6 4 0:6 / /dev/pts rw,nosuid,noexec,relatime - devpts devpts rw,gid=5,mode=620,ptmxmode=000
7 2 0:7 / /sys/fs/cgroup rw,nosuid,nodev,noexec - tmpfs cgroup_root rw,size=1024k,mode=755
8 1 8:7 / /home rw,noatime,async - ext3 /dev/sda7 rw,data=ordered
9 1 8:7 /bindmounts/opt /opt rw,noatime,async - ext3 /dev/sda7 rw,data=ordered
10 1 8:7 /bindmounts/var /var rw,noatime,async - ext3 /dev/sda7 rw,data=ordered
11 1 8:7 /bindmounts/usr /usr rw,noatime,async - ext3 /dev/sda7 rw,data=ordered
12 1 8:7 /bindmounts/tmp /tmp rw,noatime,async - ext3 /dev/sda7 rw,data=ordered

Note that all the bind mounts show up with the exact same device name as
the original mount they were bound off of. In the interest of not
showing duplicate information, df will only show the mountpoint that has
the shortest path, using the first of those that have the same length
path. As "/opt" is shorter than "/home", that is the mountpoint df uses
as its display name.

The fields in the file are as follows:

1: mount number
2: parent mount number
3: device major:minor mounted (0:x for filesystems not backed by a
block device)
4: path within filesystem mounted
5: mountpoint
6: per-mount options (may differ with multiple mounts of same
filesystem)
(any number of optional per-mount flags may follow, terminated with...)
7: the exact string "-"
8: filesystem type
9: filesystem device name (or string passed to mount(2) for virtual
filesystems)
10: per-device options (always the same for multiple mounts of same
device)
(any number of optional per-device flags may follow)

Note that on Linux, df now prefers to use /proc/self/mountinfo
*instead* of the old /etc/mtab file that (usually) gets written when
mount is called.

- -- 
Jonathan Callen
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2
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=MJKk
-----END PGP SIGNATURE-----

Re: [gentoo-user] Re: Weird "df" output

[waltdnes]

On Wed, Nov 25, 2015 at 07:15:44PM -0500, Jonathan Callen wrote

> Note that all the bind mounts show up with the exact same device name as
> the original mount they were bound off of. In the interest of not
> showing duplicate information, df will only show the mountpoint that has
> the shortest path, using the first of those that have the same length
> path. As "/opt" is shorter than "/home", that is the mountpoint df uses
> as its display name.

  Thanks for the info.  I'm not that knowledgable to be able to figure
out that it was a df "feature", rather than a problem with my system.

-- 
Walter Dnes <waltdnes@waltdnes.org>
I don't run "desktop environments"; I run useful applications

Re: [gentoo-user] Weird "df" output

[lee]

waltdnes@waltdnes.org writes:

>   I'll admit that my system setup is a bit unusual.  A long time ago, in
> a place far away, hard drives were small, compared to today's standards.
> The usual unix practice of multiple seprate partitions was not feasable
> for me, but I did want to keep root on its own partition.  So I

I remember partitioning a 60MB hard disk.  That was a really huge disk.
What's not feasible with partitioning a gigantic 500MB disk?

> compromised with a small / partition, with empty /home, /opt, /var,
> /usr, and /tmp directories.  Their real equivalents are bind-mounted
> from a much larger partition.

Why don't you just mount the large partition somewhere under /mnt and
create symlinks to the directories that are missing on the small
partition?

You might need /sbin and/or /bin on the small partition itself to be
able to mount anything at all.

Or, why don't you copy the system to the disk that has the large
partition and retire the 500MB disk?  That would reduce power
consumption and increase reliability by having less disks in use and by
making it more unlikely to mess up anything due to excessive
partitioning.


Not that I would do any of this.  Disks always come at least in pairs
because redundancy is required.  And since this is only one of your
machines, you could even run it diskless.

Re: [gentoo-user] Weird "df" output

[Mark David Dumlao]

[-- Attachment #1: Type: text/plain, Size: 1170 bytes --]

On Nov 26, 2015 08:30, "lee" <lee@yagibdah.de> wrote:
> waltdnes@waltdnes.org writes:
> > compromised with a small / partition, with empty /home, /opt, /var,
> > /usr, and /tmp directories.  Their real equivalents are bind-mounted
> > from a much larger partition.
>
> Why don't you just mount the large partition somewhere under /mnt and
> create symlinks to the directories that are missing on the small
> partition?

wrt space, that doesn't really change things.
wrt symlinks, some legacy tools, and regular unix tools have a completely
different behavior when traversing symlinks as opposed to regular
directories, which bindmounts emulate. although in practice i imagine it
wont affect him.

youre really just proposing a different way to do the same thing albeit his
approach is more stable.

> Or, why don't you copy the system to the disk that has the large
> partition and retire the 500MB disk?  That would reduce power
> consumption and increase reliability by having less disks in use and by
> making it more unlikely to mess up anything due to excessive
> partitioning.

its not 2 disks, its one disk and with partitions. at any rate his approach
is valid.

[-- Attachment #2: Type: text/html, Size: 1483 bytes --]

Re: [gentoo-user] Weird "df" output

[lee]

Mark David Dumlao <madumlao@gmail.com> writes:

> On Nov 26, 2015 08:30, "lee" <lee@yagibdah.de> wrote:
>> waltdnes@waltdnes.org writes:
>> > compromised with a small / partition, with empty /home, /opt, /var,
>> > /usr, and /tmp directories.  Their real equivalents are bind-mounted
>> > from a much larger partition.
>>
>> Why don't you just mount the large partition somewhere under /mnt and
>> create symlinks to the directories that are missing on the small
>> partition?
>
> wrt space, that doesn't really change things.
> wrt symlinks, some legacy tools, and regular unix tools have a completely
> different behavior when traversing symlinks as opposed to regular
> directories, which bindmounts emulate. although in practice i imagine it
> wont affect him.

Which tools come to mind?

> youre really just proposing a different way to do the same thing albeit his
> approach is more stable.

Yes, there are many possibilities here.

>> Or, why don't you copy the system to the disk that has the large
>> partition and retire the 500MB disk?  That would reduce power
>> consumption and increase reliability by having less disks in use and by
>> making it more unlikely to mess up anything due to excessive
>> partitioning.
>
> its not 2 disks, its one disk and with partitions.

He said that he "has a primary partition 1, which covers the entire hard
drive" and "a small / partition".  That made me think that he has two
disks.

> at any rate his approach is valid.

Re: [gentoo-user] Weird "df" output

[waltdnes]

On Thu, Nov 26, 2015 at 11:57:20AM +0100, lee wrote

> He said that he "has a primary partition 1, which covers the entire
> hard drive" and "a small / partition".  That made me think that he
> has two disks.

  Primary partitions are numbered 1 through 4 and logical partitions are
numbered 5 and up.  The "primary partition" is the entire physical disk.
5 is the small /, 6 is swap, and 7 is the rest of the disk.

-- 
Walter Dnes <waltdnes@waltdnes.org>
I don't run "desktop environments"; I run useful applications

Re: [gentoo-user] Weird "df" output

[Neil Bothwick]

[-- Attachment #1: Type: text/plain, Size: 640 bytes --]

On Thu, 26 Nov 2015 06:27:39 -0500, waltdnes@waltdnes.org wrote:

> > He said that he "has a primary partition 1, which covers the entire
> > hard drive" and "a small / partition".  That made me think that he
> > has two disks.  
> 
>   Primary partitions are numbered 1 through 4 and logical partitions are
> numbered 5 and up.  The "primary partition" is the entire physical disk.
> 5 is the small /, 6 is swap, and 7 is the rest of the disk.

Why are you using logical partitions at all when three primary partitions
will do the job?


-- 
Neil Bothwick

Did you know that eskimos have 17 different words for linguist?

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 181 bytes --]

Re: [gentoo-user] Weird "df" output

[Peter Humphrey]

On Thursday 26 November 2015 06:27:39 waltdnes@waltdnes.org wrote:
> On Thu, Nov 26, 2015 at 11:57:20AM +0100, lee wrote
> 
> > He said that he "has a primary partition 1, which covers the entire
> > hard drive" and "a small / partition".  That made me think that he
> > has two disks.
> 
>   Primary partitions are numbered 1 through 4 and logical partitions are
> numbered 5 and up.  The "primary partition" is the entire physical disk.

In which case it's an extended partition; calling it a primary is 
confusing[1], even though an extended partition is a kind of primary 
partition.

> 5 is the small /, 6 is swap, and 7 is the rest of the disk.

1.  As witness Lee's reading.

-- 
Rgds
Peter

Re: [gentoo-user] Weird "df" output

[lee]

waltdnes@waltdnes.org writes:

> On Thu, Nov 26, 2015 at 11:57:20AM +0100, lee wrote
>
>> He said that he "has a primary partition 1, which covers the entire
>> hard drive" and "a small / partition".  That made me think that he
>> has two disks.
>
>   Primary partitions are numbered 1 through 4 and logical partitions are
> numbered 5 and up.  The "primary partition" is the entire physical disk.

Hm, I don't consider extended partitions as primary ones but as extended
ones.  When I need more than four partitions, I create three primary
ones, an extended one and logical ones within the extended one.  Why
would I do that any other way?

You cannot have a primary partition that covers the entire disk and then
some.

Re: [gentoo-user] Weird "df" output

[waltdnes]

On Sat, Nov 28, 2015 at 12:13:29AM +0100, lee wrote
> waltdnes@waltdnes.org writes:
> 
> > On Thu, Nov 26, 2015 at 11:57:20AM +0100, lee wrote
> >
> >> He said that he "has a primary partition 1, which covers the entire
> >> hard drive" and "a small / partition".  That made me think that he
> >> has two disks.
> >
> >   Primary partitions are numbered 1 through 4 and logical partitions are
> > numbered 5 and up.  The "primary partition" is the entire physical disk.
> 
> Hm, I don't consider extended partitions as primary ones but as extended
> ones.  When I need more than four partitions, I create three primary
> ones, an extended one and logical ones within the extended one.  Why
> would I do that any other way?
> 
> You cannot have a primary partition that covers the entire disk and then
> some.

  OK, Primary and/or Extended partitions are numbered 1-to-4.  Logical
partitions within extended partitions are numbered 5 and up.

-- 
Walter Dnes <waltdnes@waltdnes.org>
I don't run "desktop environments"; I run useful applications

Re: [gentoo-user] Weird "df" output

[Neil Bothwick]

[-- Attachment #1: Type: text/plain, Size: 880 bytes --]

On Sat, 28 Nov 2015 07:27:18 -0500, waltdnes@waltdnes.org wrote:

> > Hm, I don't consider extended partitions as primary ones but as
> > extended ones.  When I need more than four partitions, I create three
> > primary ones, an extended one and logical ones within the extended
> > one.  Why would I do that any other way?
> > 
> > You cannot have a primary partition that covers the entire disk and
> > then some.  
> 
>   OK, Primary and/or Extended partitions are numbered 1-to-4.  Logical
> partitions within extended partitions are numbered 5 and up.

And an extended partition is a special type of primary partition.

Or you can avoid all this legacy idiocy and confusion by using GPT, which
lets you have a sensible number of partitions without fragile kludges.


-- 
Neil Bothwick

Shell to DOS... Shell to DOS... DOS, do you copy? Shell to DOS...

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 181 bytes --]

Re: [gentoo-user] Weird "df" output

[waltdnes]

On Sat, Nov 28, 2015 at 12:35:34PM +0000, Neil Bothwick wrote
> On Sat, 28 Nov 2015 07:27:18 -0500, waltdnes@waltdnes.org wrote:
> 
> >   OK, Primary and/or Extended partitions are numbered 1-to-4.  Logical
> > partitions within extended partitions are numbered 5 and up.
> 
> And an extended partition is a special type of primary partition.
> 
> Or you can avoid all this legacy idiocy and confusion by using GPT, which
> lets you have a sensible number of partitions without fragile kludges.

  The entire point of my bindmounting procedure (remember the original
post?) is to minimize the number of partitions without strangling root
if a logger goes wild.  I generally run with 3 partitions, including
swap.

-- 
Walter Dnes <waltdnes@waltdnes.org>
I don't run "desktop environments"; I run useful applications

Re: [gentoo-user] Weird "df" output

[Neil Bothwick]

[-- Attachment #1: Type: text/plain, Size: 916 bytes --]

On Sat, 28 Nov 2015 18:52:23 -0500, waltdnes@waltdnes.org wrote:

> > And an extended partition is a special type of primary partition.
> > 
> > Or you can avoid all this legacy idiocy and confusion by using GPT,
> > which lets you have a sensible number of partitions without fragile
> > kludges.  
> 
>   The entire point of my bindmounting procedure (remember the original
> post?) is to minimize the number of partitions without strangling root
> if a logger goes wild.  I generally run with 3 partitions, including
> swap.

So if you only need 3 partitions, why bother with an extended one in the
first place?

I'd still use GPT, it's just so much nicer.

Or if you want to be properly bleeding edge, use btrfs and do away with
all that bind mounting stuff, and get even more confusing df output...


-- 
Neil Bothwick

Everything should be made as simple as possible, but no simpler.

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 181 bytes --]

Re: [gentoo-user] Weird "df" output

[Rich Freeman]

On Thu, Nov 26, 2015 at 5:57 AM, lee <lee@yagibdah.de> wrote:
> Mark David Dumlao <madumlao@gmail.com> writes:
>> wrt symlinks, some legacy tools, and regular unix tools have a completely
>> different behavior when traversing symlinks as opposed to regular
>> directories, which bindmounts emulate. although in practice i imagine it
>> wont affect him.
>
> Which tools come to mind?
>

As I mentioned earlier, I believe portage doesn't like it if /usr or
/var are symlinks.  But, I'm going from memory - it has been a while
since I had to bind-mount them.

-- 
Rich

Re: [gentoo-user] Weird "df" output

[Tom H]

On Thu, Nov 26, 2015 at 12:10 AM,  <waltdnes@waltdnes.org> wrote:
>
> I'll admit that my system setup is a bit unusual. A long time ago, in
> a place far away, hard drives were small, compared to today's standards.
> The usual unix practice of multiple seprate partitions was not feasable
> for me, but I did want to keep root on its own partition. So I
> compromised with a small / partition, with empty /home, /opt, /var,
> /usr, and /tmp directories. Their real equivalents are bind-mounted
> from a much larger partition. I just re-did my oldest machine. It has
> a primary partition 1, which covers the entire hard drive. The /
> partion on /dev/sda5 is approximately 500 megabytes (YES!). There's a
> 3.8 gigabyte swap partion /dev/sda6, and the rest of the drive is
> /dev/sda7. Here's the relevant portion of /etc/fstab...
>
> /dev/sda5               /           ext2     noatime,async      0 1
> /dev/sda7               /home       ext3     noatime,async      0 1
> /home/bindmounts/opt    /opt        auto     bind               0 0
> /home/bindmounts/var    /var        auto     bind               0 0
> /home/bindmounts/usr    /usr        auto     bind               0 0
> /home/bindmounts/tmp    /tmp        auto     bind               0 0
> /dev/sda6               none        swap     sw                 0 0
>
> ...and the output from "df"...
>
> Filesystem     1K-blocks      Used Available Use% Mounted on
> /dev/root         495944     49416    420928  11% /
> devtmpfs           10240         0     10240   0% /dev
> tmpfs             310080       356    309724   1% /run
> shm              1550384         0   1550384   0% /dev/shm
> cgroup_root        10240         0     10240   0% /sys/fs/cgroup
> /dev/sda7      476205120 292365556 159643008  65% /opt
>
> ...showing /dev/sda7 mounted on /opt !?!? mc (Midnight Commander) shows
> 152 of 454 gigabytes free on all of /home, /opt, /var, /usr, and /tmp,
> which is correct, since they're all really bindmounts from /dev/sda7.
> The / partition (/dev/sda5) has 411 of 484 megabytes free. The machine
> works OK, but the "df" output is a head-scratcher. I've re-booted a
> couple of times, with no change.

If the same filesystem's mounted two or more times, then df shows the
shortest mountpoint.

In your case, /home /opt /tmp /usr /var are mounts of /dev/sda7 so
/opt is shown, because it's the first mountpoint lexically-speaking.

If you use "-a", then /home will be displayed too.