ext3 reliability requirements are not met

[alexander@…]

Pavel Machek published a patch that documents conditions under which the ext3 filesystem is reliable against power failures. These conditions aren't met by LFS systems.

Namely, the failed requirement is: "either write caching is disabled, or hw can do barriers and they are enabled". To disable write caching, one should do "hdparm -W0" on the device before remounting the filesystem read-write. To enable barriers (if the hardware supports them), use "barrier=1" mount option.

[alexander@…]

Forgot the URL of the discussion: ​http://lkml.org/lkml/2009/1/3/123

[alexander@…]

OTOH, Alan Cox said about distros in ​http://lkml.org/lkml/2009/1/4/70:

Generally they avoid setting -W0 because it ruins performance and can be very bad for disk lifetime.

So the bug is probably only about the missing "barrier=1" option.

[Matthew Burgess]

Alexander, I took a quick look at that thread, and the documentation states:

(Note that barriers are disabled by default, use "barrier=1" mount option after making sure hw can support them).

How does one determine if their hardware supports barriers or not? Just wondering what option we make the default, and therefore how best to document the alternative?

[alexander@…]

For anything dm-based (LVM, LUKS, ...): barriers are not supported. For SATA (i.e., /dev/sda2): if "hdparm -I /dev/sda | grep NCQ" returns non-empty output, then the hardware supports barriers. For other hardware types, I don't know.

[bdubbs@…]

Added an explanation for the barrier option in the fstab page. Fixed at revision 8881.