[gentoo-amd64] Re: mount ext3 with ext4 driver

[Duncan]

Martin Herrman <martin@×××××××.nl> posted
40bb8d3b0905021016k6241308ek660351f5fa700f8a@××××××××××.com, excerpted
below, on  Sat, 02 May 2009 19:16:54 +0200:

> How stable do you think that the ext4 module actually is? (using recent
> kernels, e.g. 2.6.29)

Please kill the HTML posting.  Many users on these lists don't like it, 
believing a message worth reading is worth reading in plain text, and 
that HTML messages are therefore for spammers, crackers, AOLer types and 
others who obviously think HTML must be needed to make their message 
worth reading, since they couldn't stick with plain text.

As for the question, I'd hold off on ext4 for anything you care about and 
don't have backups for.  ext4 is still young enough that there were some 
very major controversy and issues debated on LKML for 2.6.29 and into 
2.6.30.  Thus, for something like /tmp, and the Gentoo tree that can 
simply be re-synced if there's issues, yes, but for /, /usr and /home, 
you're tempting fate unless you keep backups updated frequently enough 
that you're willing to lose what's not backed up.

It's also worth noting that as a result of the ensuing firestorm on LKML, 
ext3 has some changes in the upcoming 2.6.30 as well, with data=writeback 
improved and made the default (over the previous data=ordered), which 
should dramatically improve its performance in some cases, but at the 
cost of somewhat weakened reliability after a crash.  (It's possible some 
files may be truncated to zero-length, that would be recovered correctly 
under data=ordered, but the situation isn't as bad as it was before the 
changes.)

Thus, ext3 users who value their data may wish to do a bit more research 
before they upgrade to 2.6.30, and evaluate whether the improved new 
default data=writeback is sufficient for them, or whether they want to 
stick with the old default and specifically configure data=ordered.  
(Those using data=journal won't be affected, as that has always had to be 
specifically configured..)

Personally, while I'm using reiserfs here and it's still data=ordered by 
default thru 2.6.30-rc4 at least (and thus, shouldn't be changed 
until .31, if changed at all), there's no way I'd even consider 
data=writeback, improved or not.  I have too many bad memories of zeroed 
out files and /so/ looked forward to getting data=ordered on reiserfs, to 
even think about data=writeback.  It might indeed be faster, but it's 
something I'll be using.  I like my data, thank you very much!

As for ext4, it's not a question I'm dealing with (I'm waiting for btrfs 
to stabilize before switching off of resierfs, here), but given the 
recent ext4 issues that sparked the whole LKML flamefest, I'd not 
consider it for data I wanted to keep until at least 2.6.31, which of 
course hasn't started yet, so no one knows what it'll bring and whether 
ext4 will be basically done after .30, or whether there'll be more major 
changes in .31.

If you're interested in more, check the last three weeks' or so LWN 
(lwn.net) kernel pages.  There's links to the LKML threads, as well as to 
the Ubuntu bug that kicked it all off.  (Some proprietaryware NVidia 
driver folks were crashing frequently, and those who had chosen the new 
Ubuntu ext4 option noticed their gnome settings files were zeroed out 
after they rebooted.  That particular bug is papered over for 2.6.29 and 
generally fixed for 2.6.30, but there's still a smaller risk as long as 
you use the default data=writeback, the 2.6.30 ext3 default as well.  So 
as far as is known, with .30, ext4 SHOULD be as stable as ext3 in the 
same data= mode, but the fact that such things are still coming up means 
nobody's sure there aren't more such things to come up, thus, people who 
value their data will want to wait to at least 2.6.31 before trying ext4, 
and will want to be sure they specifically set data=ordered for ext3 
before switching to 2.6.30.)

-- 
Duncan - List replies preferred.   No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master."  Richard Stallman