On 2011-10-06 14:34:12, Paolo Bonzini wrote:
> I still see occasional errors:
>
> [33945.269075] ecryptfs_decrypt_page: Error attempting to read lower page; rc = [-4]
> [33945.269084] ecryptfs_readpage: Error decrypting page; rc = [-4]
>
> that are reported as "short read: Success" when doing a "git grep" on an
> encrypted repository. Redoing the grep fixes the problems for those
> files and may transfer it to others, until all of them are in the cache
> and the grep succeeds.
This is what the fix was intended to do. When eCryptfs is trying to
read from the lower filesystem and gets interrupted, pass the EINTR
error code onto userspace. However, eCryptfs doesn't need to write a log
message about it.
From your description, I can't tell if grep handled the EINTR error
correctly. If you can reproduce this, can you please strace grep and
attach the output so that I can take a better look at what grep is
seeing returned from system calls?
$ strace -o grep.strace grep [GREP ARGS]
Oh, and please listen to kirkland's request and give us `uname -a`
output.
On 2011-10-06 14:34:12, Paolo Bonzini wrote: decrypt_ page: Error attempting to read lower page; rc = [-4]
> I still see occasional errors:
>
> [33945.269075] ecryptfs_
> [33945.269084] ecryptfs_readpage: Error decrypting page; rc = [-4]
>
> that are reported as "short read: Success" when doing a "git grep" on an
> encrypted repository. Redoing the grep fixes the problems for those
> files and may transfer it to others, until all of them are in the cache
> and the grep succeeds.
This is what the fix was intended to do. When eCryptfs is trying to
read from the lower filesystem and gets interrupted, pass the EINTR
error code onto userspace. However, eCryptfs doesn't need to write a log
message about it.
From your description, I can't tell if grep handled the EINTR error
correctly. If you can reproduce this, can you please strace grep and
attach the output so that I can take a better look at what grep is
seeing returned from system calls?
$ strace -o grep.strace grep [GREP ARGS]
Oh, and please listen to kirkland's request and give us `uname -a`
output.