Booting from a specific HDD with UEFI

dbishop · Tux's lil' helper Joined: 08 Dec 2007 Posts: 107

I am trying to get one of my Gentoo systems to boot from a very specific SATA disk (GPT and UEFI), when the type and number of other disks can vary (SAS, SATA, some raid, others not). For the most part it works. But there are some corner cases I am dealing with that result in total failure. here is a rundown of some pertinent details:

1) The machine has both SATA and SAS interfaces (in this case the SAS is an LSI HBA, SATA are the usual SATA3 controller combos).

2) The boot disk is SATA3 (GPT)

3) There are other disks, some SATA, some SAS-based, BUT these are NOT permanently mounted and what is present at boot varies quite a lot. <-- This is the issue

The system will boot correctly (in minutes, that is, as opposed to hours) if there are no other SATA or SAS disks plugged in. However, if these other disks are plugged in, the system gets lost in hours-long fsck's -- these file systems range from 4TB to 16TB, some raid, some not, and there are a number of file system varians in play too -- reiserfs, ext2, ext4 with 16K blocks, ntfs just to name a few.

To make matters worse, this particular system is headless, and access is only via ssh, so once it begins fsck'ing there is no choice but to hard-boot the machine ensuring the other disks areother disks unplugged. (The SAS disks are in an external JBOD box so I just power it down, and the other SATA disks are in hot-swap trays, so I pop them out for boot, then plug everything back in and power on the JBOD box post boot. Sometimes I forget -- or can't because it's remote.

I have set the dump/p[ass to 0 0 in fstab, but I have been reading that tune2fs might be able to do the trick, but it seems that it won't work if the devices move around (as the do with these disk systems, they get hotplugged and moved across systems all the time).

Is tune2fs the solution? Will it work even if the disks are not always there or get hotplugged, or their /dev/ node changes with each boot?

I am grateful for any help.

NeddySeagoon · Posted: Thu May 21, 2015 5:20 pm Post subject:

dbishop,

There are several issues here and its not clear which one is the root of your problem.

There is the HDD detection order before the kernel is loaded and the HDD detection order by the kernel.
There is not much to do about first one. You can make a small raid1 /boot partition on all the drives, then any drive can provide the boot files
You also need to install the boot loader on them all.

After the kernel is loaded, you can pass root=UUID=<filesystem_UUID> which needs an initrd, or you can use root=PARTUUID=<root_partition_UUID> which the kernel can handle unaided.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

dbishop · Tux's lil' helper Joined: 08 Dec 2007 Posts: 107

Hah -- don't know if you recall, but this is the same exact machine you helped me with before -- I ended up giving it network interfaces neddy1 and neddy2. You're doubly invested now :-)

UEFI and GPT have pretty much gotten the correct boot disk to get identified well enough. Now the problem is that if there are other disks -- particularly large ones -- they tend to want to get fsck'd. This can take *hours*.

It is a very dynamic situation. The only disk guaranteed to be there at boot is the boot disk. All others are optional. So There may be one or two or three SATA data disks there, there may be one or two SAS data disks or RAIDs.

How can I reliably make sure that fsck is only ever run on the boot disk?

NeddySeagoon · Posted: Thu May 21, 2015 6:41 pm Post subject:

dbishop,

For the extX family of fs read

dbishop · Tux's lil' helper Joined: 08 Dec 2007 Posts: 107

We are discussing data-only disks -- they are not system disks at all. It's fine that the main boot disk gets checked (but all the partitions are reiserfs and I have never has any real problems with reiserfs over literally hundreds of disks and thousands of partitions. But fsck-away on those. Quick. Painless. Fine.

I don't care if the data disks get fsck'd, that's not the point, I just want it to happen after the system gets booted. These disks aren't even automounted.

The funny thing is that when they are added to the system post-boot, fsck's are never run. Same system, same disks.

I'll read the man page just to see what can be done.

As always, thanks for your help.

dbishop · Tux's lil' helper Joined: 08 Dec 2007 Posts: 107

Adding to the irony of all this is the fact that the file systems that I want to bypass checking use a 16K block size. I don't even think fsck works with it. check this out:

krinn · Watchman Joined: 02 May 2003 Posts: 7470

Read this: https://forums.gentoo.org/viewtopic-t-1007788.html

1/ find and plug your sata boot disk on the lowest (first) sata port, and just never replace it so this port will always kept your boot disk
2/ of course set the sata controller driver as buildin your kernel
3/ any other controllers -> set them as module (so any disks plug on them can only take a drive name next to the ones attach to the controller that is buildin the kernel ; a disk on it could never get sda name)
4/ mount your disk in your fstab using label, partuid or anything that can ident the disk to mount ie: