Slashdot Mirror
Data Loss Bug In OS X 10.5 Leopard
An anonymous reader writes "Leopard's Finder has a glaring bug in its directory-moving code, leading to horrendous data loss if a destination volume disappears while a move operation is in progress. This author first came across it when Samba crashed while he was moving a directory from his desktop over to a Samba mount on his FreeBSD server."
I lost a huge amount of data being MOVED (not copied) from one volume into a virtual volume DMG file. Lost and gone forever, lots of important files. What happened? I simply bumped the laptop Mac Book Pro during the move... zap... gone forever. The DMG file was blank! Yes, complely zero bytes except for a bit of header non-file data. It sucks bad.
No offense meant here, but normal move/copy operations are traditionally highly destructive events on MacOS anyway. For instance there is absolutely no simple way to merge two folders contents together on the mac. If you drag a folder called "Documents" into your home directory and click on "OK", the Mac OS will happily delete your entire documents folder. I was reminded of this enormous frustration while recovering from some multi-volume backups recently, having to resort to an obscure OS X commandline tool 'pax' and Leopard's newfound support of hardlinks to make some simple file copies play nice and not unnecessarily consume 3 times the disk space they should have.
For all of the flack the Windows file copy interface gets, it is both safer and more flexible than trying to use the Finder: an interface that makes file management so stupefying it becomes impossible.
Why don't you just use rsync?
Give me Classic Slashdot or give me death!
Oh yeah, rsync. Is that one still broken on 10.5? Apple's build of rsync on 10.4 consistently choked on a few files when my dad started using it on his Mac Pro.
Typically if you are moving within the same logical device the file pointer is moved and no copying need take place.
When moving to another device your code reads and writes, within a loop and traps exceptions (such as the device suddenly vanished, where the OS should raise an exception and your application traps it.) A wide variety of errors could occur while moving and in the event any of them happen the user should be notified in an appropriate manner and the original data not deleted.
I've written a number of applications which moved files or data between databases and it's fundamental your application is on the watch for any problems. Not to have an exception raised or to trap any and all, well, that's simply an inexcusable lapse.
This sort of thing is extremely critical if you happend to be defragmenting a disk drive. Long before Macs and PCs we had to defrag our mainframe drives and the applications which did the work were quite careful. Often the best practice, if you had the resource of a second drive, was to simply defrag to a new drive then re-assign the new dist as the original.
A feeling of having made the same mistake before: Deja Foobar
NFS write failure on Linux 2.4, check your data is gone.
Uhh no. In linux mv's always make sure the data moved then delete the original file (as far back as 2.4). You can test this by dd'ing a large file (use if=/dev/urandom), then run a md5sum on it, then do a mv within the same drive, over nfs, over samba/cifs, to floppy (not sure what happens here because linux caches writes to floppies until umount), to usb drive, whatever. While the mv is in operation just pull the plug on the system (target or source system). Your old file will still be there!
Um... no. REALLY no. Please don't do that.
The syntax for "||" is:
If command 1 fails, do command 2 - otherwise exit (where you used "command1 || command2").
In this case, your command will either copy all the files from $from to $to or delete all files at $from.
What you probably meant is cp $from $to && rm -r $from, which only performs the second command if the first succeeded. This solution is far from perfect for reasons mentioned by other posters, but it's still significantly safer and more useful than yours...
There have been resource fork patches for some time, but somewhat unreliable. Version 3.0 is going to support resource forks (and other types of extended attributes) out of the box. The setup we have been using extracts resource forks to a separate file on the mac and then backs them up alongside the original files. The restore process just performs the reverse. It will be nice once we can switch to rsync 3.0 and get rid of that step.