bin/70600: fsck throws files away when it can't grow lost+found

[Charles Sprickman]

>Number:         70600
>Category:       bin
>Synopsis:       fsck throws files away when it can't grow lost+found
>Confidential:   no
>Severity:       non-critical
>Priority:       medium
>Responsible:    freebsd-bugs
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Wed Aug 18 02:20:21 GMT 2004
>Closed-Date:
>Last-Modified:
>Originator:     Charles Sprickman
>Release:        FreeBSD 4.9
>Organization:
Spork, LLC
>Environment:
FreeBSD skunkworks.spork.us 4.9-RELEASE-p4 FreeBSD 4.9-RELEASE-p4 #6: Mon Mar 29 00:49:50 EST 2004
>Description:
When fsck is run on a well-trashed partition that has a large number of files/directories (for example a mail server using Maildirs, a news server, large office file server), there seems to be some internal limit on the number of directories and/or files that fsck can create within lost+found.  This results in many possibly recoverable files being deleted.
>How-To-Repeat:
Trash a filesystem badly. :)  Then run fsck on it.  At some point, fsck will start spitting out the error "SORRY. NO SPACE IN lost+found DIRECTORY".  Since one is probably running with "-y" due to the large number of prompts a bad fs will have, from that point on everything fsck tries to "reconnect" into lost+found gets trashed.  You have lost possibly good data at that point.
>Fix:
I don't really have any C skills, but on a box where I'm attempting recovery off of a dd image of a trashed fs, I've modified "dir.c" like so to make fsck just exit when it can no longer add files to lost+found:

if (makeentry(lfdir, orphan, (name ? name : tempname)) == 0) {
                pfatal("SORRY. NO SPACE IN lost+found DIRECTORY");
                printf("exiting...\n\n");
                exit (99);
                return (0);
        }

I then mount the fs r/w, copy off everything in lost+found, rm -rf lost+found, unmount, and then run fsck again.  I keep doing this until it's finished.  

I imagine someone more familiar with fsck could have it do something else at that point; like create a "lost+found1" directory and keep incrementing.  That's my best guess as to how to approach this.

Thanks,

Charles
>Release-Note:
>Audit-Trail:
>Unformatted: