Hi,

I've a basic question about the following:

I'm aware of when linux kernel boot it mounts the root filesystem as read-only in order to perform fs check and such. Then it remounts itself as read-write (rw).

As far as I understand (at least for Read Hat/CentOS based distros) that is done into rc.sysint script (using "mount -n -o remount,rw") upon sbin/init runs it once at boot time.

On the other hand there exist /etc/fstab having entries for filesystem to mount automatically including root filesystem (filesystem mounted as /)

Now the question is: when in the boot process /etc/fstab is actually read and referenced filesystems mounted ?

Thanks

what init stage is the fstab file read
Google says,

https://www.debian.org/doc/manuals/d...e/ch03.en.html

http://www.linuxjournal.com/article/3016

https://www.safaribooksonline.com/li...9/ch04s02.html

http://etutorials.org/Linux+systems/...nitialization/

I'm working on a custom Red Hat based system using Upstart init (see below)
At system boot Upstart init runs /etc/rc.d/rc.sysinit shell script (Upstart init is sysvinit compatible). First lines in that script create mountpoints for /proc and such...
Now my question is: As far as I know the 'real' root filesystem is mounted at boot time as read-only (ro) (in order to do a fschk for instance ) thus I guess mkdir should get an error when trying to create those directory...right ?

Then rc.sysinit mounts the root filesystem as rw
So my second question is: why mount explicitly the root filesystem in rw when the same option exist in the related entry in /etc/fstab ? We could simply do a mount -a to mount all filesytems referenced in /etc/fstab including remounting the root filesystem in rw, or not ?

If Red Hat does things in the same order as LFS, then / gets mounted three times. The first mount is read-write. This is when the dynamic filesystem mount points get checked and created if necessary. In most cases they will exist already, but the -p option prevents an error being reported. Then / is remounted read-only for fsck. Then it gets remounted rw. LFS does that in a later script.

I assume / has to be mounted explicitly before "mount -a" is run so that mounts can be logged.

Here with dynamic filesystems do you refer to proc, sys and tmpfs for instance ?

Excuse me I'm a beginner...but why -p option in mkdir would prevent an error being reported ?

Sorry, what do you mean here ?

Yes.
Didn't you read the mkdir man page? It's the obvious place to get this sort of information about how a command behaves. It says there that when the -p option is used, it prevents an error when the directory already exists.
The system logs are in /var/log, and that is usually on the root partition. So if you want the mounting of other partitions to be recorded in the logs, the root partition must already be mounted read-write.

Moved: This thread is more suitable in <Red Hat> and has been moved accordingly to help your thread/question get the exposure it deserves.

Just to be sure I've understand it correctly...from my understanding when you mount a filesystem (using command "mount" even with options) the command itself does not write to log files.

Instead into init rc scripts (shell scripts under /etc/init.d/rc.d/<>) using for instance the function "action" (defined in /etc/init.d/functions) messages about operations (mounting a filesystem in this case) are written to log files. So, as you pointed out, in order to record in the logs the mounting of other partitions, the root partition itself (where usually /var/log lives) must already be mounted read-write. Does it make sense ?

That depends on how Red Hat does its logs. I haven't used a Red Hat distro for years. If they log directly to /var/log/messages or /var/log/syslog, then yes, that requires / to be mounted read-write. But some distros create their boot log on /run, which is usually a tmpfs (effectively a ramdisk), and only copy it over to /var/log on completion of the boot process. In that case / could be mounted read-only and it would still work.

This thread was moved into the Red Hat section precisely so that people more experienced with this distro family could chip in and answer your questions. I think I have come to the end of my knowledge.

Prior to RHEL 7, /etc/mtab was an ordinary file maintained by the mount command. That command has a "-n" (--no-mtab) option to mount without writing to /etc/mtab, necessary when /etc is on a read-only filesystem.

Starting with RHEL7, /etc/mtab is a symlink to /proc/self/mounts and is not maintained by the mount command.

Sure....also on my system it's a symlink
As said before for LFS, can you confirm that also for Red Hat the 'real' root filesystem / (just to be clear no the initial initrd or iniramfs mounted on /) gets mounted three times ?

Any help ?

Which RHEL/CentOS version? A lot of things changed in version 7.

Can you consider RHEL 7.2 or just Centos 6.0 (running on another PC).

Thanks

For RHEL/CentOS 6, it's all in the rc.sysinit file that you mentioned. The reason that /etc/fstab is not consulted for remounting root read/write is because there is a large section of that file devoted to the case of the "readonlyroot" option in the kernel command line, and read/write is assumed without that, regardless of what is in /etc/fstab. Special filesystems critical to the boot sequence (/proc, /sys, ...) get mounted early by the fstab that is in the initrd prior to mounting the real root filesystem. All other non-network filesystems are mounted by a "mount -a" command at about line 525 in rc.sysinit, and that mount command uses /etc/fstab.

For RHEL/CentOS 7, you'll have to talk to a systemd fanboi expert.