make_file() could ignore ENODEV errors (from afs filesystem)

[Dave Dykstra]

I think it would be better to fix the bug of rsync doing a readlink on
excluded files.  What version are you using?  There was a bug related to
that a version or two back.  What command line are you using?  Somebody
reported a similar problem recently with an infrequently used option, I
think it was -R but I can't remember for sure.

- Dave


On Wed, Jun 19, 2002 at 03:23:07PM +0100, Kevin J Walters wrote:
> >>>>> "d" == Dave Dykstra <dwd at bell-labs.com> writes:
> 
> d> On Tue, Jun 18, 2002 at 11:43:56AM +0100, Kevin J Walters wrote:
> >> 
> >> Hi,
> >> 
> >> Would it be acceptable to patch make_file() in flist.c to ignore any
> >> ENODEV's and not mark them as io_error=1?
> >> 
> >> In afs, a mountpoint to a volume which does not exist gives an
> >> ENODEV. As such, it can't be copied, so should (or could) just be
> >> ignored?
> 
> d> Is this a normal situation in afs (I've never used it)?  Can you use
> d> --exclude to skip it?
> 
> It's not unheard of. The mount operation is a user level command, so
> any user has the potential to create a mountpoint to a volume which
> may not exist (ENODEV) or may not be reachable (ETIMEDOUT or just
> blocks).
> 
> I am using --exclude '.asd/*', this is what comes out,
> 
> building file list ... readlink .asd/A: No such device
> readlink .asd/B: No such device
> readlink .asd/C: No such device
> done
> IO error encountered - skipping file deletion
> wrote 2507 bytes  read 20 bytes  561.56 bytes/sec
> total size is 172798  speedup is 68.38
> rsync error: partial transfer (code 23) at main.c(578)
> 
> so i think the allegation of a 'partial transfer' is not correct
> because i did exclude those directories and it did tranfer the
> included stuff correctly. --exclude '/*' works but it's a bit too
> severe :)

> The output doesn't bother me, it's just that exit 23 which is
> preventing me checking for real errors.
> 
> Could io_error only be set if !check_exclude_file(f, fname, &st) and
> maybe leave the rprintf but with a WARNING?
> 
> Why is the check_exclude_file() not performed before the
> readlink_stat()? This would mean the stat was never attempted (saving
> a system call) for an excluded thing (including things that error)?
> 
> d> Do you know if there is precedent in other standard open source tools
> d> to ignore ENODEV?  In general I don't think it's a good idea to ignore
> d> errors.  It might mean something else in a different situation.
> 
> Don't know of any precedent. The only error i've seen out of a
> filesystem is sunos/solaris ufs EIO when a disk is broken. Not sure
> what error a soft NFS mount gives.
> 
> 
> regards
> 
> |<evin
> 
> -- 
> Kevin J Walters                      Morgan Stanley
> kjw at ms.com, kjw at acm.org              20 Cabot Square
> Tel: 020 7425 7886                   Canary Wharf
> Fax: 020 7677 8504                   London E14 4QW
> 
> -- 
> To unsubscribe or change options: http://lists.samba.org/mailman/listinfo/rsync
> Before posting, read: http://www.tuxedo.org/~esr/faqs/smart-questions.html