Werner Braun posted on Tue, 18 Apr 2017 14:41:49 +0200 as excerpted: > Hi, > > i have a WD WD40EZRX with strange beaviour off btrfs check vs. btrfs > scrub > > running btrfs check --check-data-csum returns no errors on the disk > > running btrfs scrub on the disk finds tons of errors
A dev could confirm this, but AFAIK from the comments on the patches I've seen going by, btrfs check --check-data-csum only checks that there's a valid copy available; that is, it stops as soon as it finds a valid copy, and if that's the first one, it won't check the second that's available in dup or raid1/10 mode. A scrub of the full filesystem (as opposed to a single device, for those with more than one) will however check both copies and fix the second copy from the first, if necessary. Of course this only applies to blocks that /have/ a second copy, that is, those in chunks that are raid1, raid10, or dup. On a default single device btrfs, this will be metadata chunks in dup mode, not data chunks in single mode. I think the low-memory-mode check may work differently than normal mode check in this regard, and as I said, this is based on the comments on patches going by, so it's possible the newest versions have changed this, but I'm not sure. Meanwhile, it's worth noting that btrfs scrub calls in-kernel code to do the scrub, while check does everything in userspace, and the comments in the patches suggest the code has diverged, so it's not entirely surprising that the results differ. Of course as Hugo mentions, scrub is done with the filesystem mounted, as well, and it's possible dio bypasses the normal buffered-write locking that prevents blocks from changing out from under scrub as it's doing its thing, allowing current direct-write access to screw things up in- flight. Check's userspace access is done with the filesystem unmounted, so that shouldn't be possible there, unless something's writing directly to the device itself, not thru the filesystem. -- 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 -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
