Imagine a situation when a file became corrupted on a single device Btrfs. And in this case file contents are needed as is including corrupted bits.
As stated in man page for mount:
nodatasum Enable/disable data checksumming for newly created files.
So it won't disable checksum checks when reading files.
And as I guess all standard system calls involving filesystem for reading will fail at this file. Raw disk read is not an option because of likely fragmentation.
When I try to
cat such file, I get error:
cat: file: Input/output error
dmesg reports messages like these:
[631847.884641] BTRFS warning (device loop0): csum failed ino 257 off 0 csum 1280268577 expected csum 2391276770
Having information that corruption occurs at
off 0, so this is first 4096 bytes of file, indeed I'm able to retrieve non-corrupted blocks pretty easily:
dd if=file bs=4K skip=1
Btw, writing and appending data to non-corrupted blocks of corrupted file also succeeds. At least when there is only one corrupted block and it's not the last. At first it surprised me, but then I think probably it's a good feature, so that applications that must append critical data won't stuck on a corrupted file.
The question is how do I retrive data from corrupted block (or entire corrupted file)?