jim@aob.UUCP (02/23/87)
I have an Altos 2086 running XENIX 3.2f. I had a problem show up recently
that defies all my attempts to fix. What has happened is that a file name
in a directory has a null in the middle of the name. The file is actually
the name of a directory. The name used to be 'Doc', but is now 'D\0c'.
All the programs in the system now say 'D' is a bad file name. Since I
can't seem to write in the directory, I can't change the null to something
else or get rid of the 'c' after the null. Does anybody know how I can
get rid of this file name, or better yet, rename the file to a typable
name? Fsck doesn't find anything wrong with the file, and I can't unlink
the file because unlink can't find 'D\0c'.
--
Jim Anderson (612) 636-2869
Anderson O'Brien, Inc New mail:jim@aob.uucp
2575 N. Fairview Ave. Old mail:{rutgers,gatech,amdahl}!meccts!aob!jim
St. Paul, MN 55113 "Fireball... Let me see... How did that go?"pdb@sei.cmu.edu.UUCP (02/25/87)
As a last resort you could use 'clri' on the inode associated with that file (use 'ls -li' to find the inode numner). fsck will certainly allow you to clear the directory entry once it realizes the inode's been zeroed. --Pat.
ssl@ptsfa.UUCP (02/25/87)
In article <563@aw.sei.cmu.edu.sei.cmu.edu> pdb@sei.cmu.edu (Patrick Barron) writes: > >As a last resort you could use 'clri' on the inode associated with that >file (use 'ls -li' to find the inode numner). fsck will certainly allow >you to clear the directory entry once it realizes the inode's been zeroed. > >--Pat. But as a last to the last resort, if you don't really want to get rid of things, I'll try this before 'clri': 1. make sure there's only one directory and/or filename start with D 2. then 'mv D* goodname' Didn't try this with D\0a though, this might/might not work, but doesn't hurt give it a try before earsing everything. As long as D* expands into one name, mv will work. -- =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= Sam Lok San Francisco {ihnp4,pyramid,qantel}!ptsfa!ssl || To err is human, to really foul things Disclaimer!? WHAT disclaimer? || up requires super-user privilege!
chris@mimsy.UUCP (02/25/87)
In article <105@aob.UUCP> someone finds a directory entry with an embedded NUL, and has nothing handy to fix it. In article <563@aw.sei.cmu.edu.sei.cmu.edu> pdb@sei.cmu.edu (Patrick Barron) writes: >As a last resort you could use 'clri' on the inode associated with that >file (use 'ls -li' to find the inode numner). This will indeed make fsck clear the directory entry, but will also lose the contents of the file. If there is another link to the file, it can be copied first, but if not, there are other ways. The problem is in the directory itself, not in the file; so you can clri the directory inode, and fsck will put any orphaned files in the lost+found directory. This will lose the names of each of the inodes, but these can be saved first: # ls -li <bad_dir> > dir_list # clri <inode> # umount <drive> # fsck <drive> # mount <drive> # cd <lost+found_area> # mkdir <bad_dir> # sh <script> <bad_dir> < dir_list where `script' reads name/inode pairs out of a directory and `mv's the file `#<inode>' (six digits) to $2/<name>. Note that any files with other links will not be placed in lost+found, and must be re-linked `by hand' later (after tracking them down with ncheck). Also, make sure there is enough directory space in lost+found for all the files that will be orphaned, if necessary by augmenting the mklost+found shell script. There is still another way, easier yet but more dangerous and thrilling. You cannot edit a directory, but you can edit a file. And the only difference between the two is one byte. Using the source to clri, ---you say you have no source? Ah well. Perhaps if you ask nicely, your vendor will help. I would not count on it. Source is better. ---Using clri.c, or your magical knowledge of file system formats, write another program, `hacki'. Instead of zeroing the inode, toggle the S_IFDIR and S_IFREG bits. Run it once on the directory inode, and now it is a file. Use your favourite binary editor (Emacs works, but be sure it does not make a backup!), fix the embedded NUL character. Make sure the file size is exactly the same as before. Then run the hacki program once more. (I did this once on a 4.1BSD system. It worked quite well.) Of course, if you have source, you could alter fsck to check for embedded NULs in filenames. -- In-Real-Life: Chris Torek, Univ of MD Comp Sci Dept (+1 301 454 7690) UUCP: seismo!mimsy!chris ARPA/CSNet: chris@mimsy.umd.edu
ron@brl-sem.UUCP (02/26/87)
In article <105@aob.UUCP>, jim@aob.UUCP (Jim Anderson) writes: > name? Fsck doesn't find anything wrong with the file, and I can't unlink > the file because unlink can't find 'D\0c'. ls -i will give the inode number. Run clri on that inode and follow it up with FSCK. If you really want it back you can probably use dump and restore to fish it out. You could probably poke the block on the raw device, but this is likely to be more trouble than it is worth. -Ron
ron@brl-sem.UUCP (02/26/87)
In article <2136@ptsfa.UUCP>, ssl@ptsfa.UUCP (Sam Lok) writes: > But as a last to the last resort, if you don't really want to get rid > of things, I'll try this before 'clri': > 1. make sure there's only one directory and/or filename start > with D > 2. then 'mv D* goodname' THIS WILL NOT WORK. The problem is that the UNIX directory structure is exactly 14 characters long and some older versions of UNIX (version 7 which XENIX is based on, don't know if System V has the same problem) compare all the 14 characters to do a match. They expect all the unused characters to have nulls in them rather than be null terminated. There is NOTHING you can do through the UNIX file system that will reference this file. Think about it. D* would just expand to "D". Mv knows nothing special about "*", that is expanded by the shell and passed as an argument to mv. You still can't pass that to unlink or open because the null terminates the string. -Ron
guy@gorodish.UUCP (02/26/87)
>Of course, if you have source, you could alter fsck to check for >embedded NULs in filenames. And not to *generate* file names with embedded NULs. Some versions of "fsck" for the V7 file system (as used in most UNIX systems not using the 4.2BSD file system) would assume that any free directory slot in "lost+found" would have no non-NUL characters in the directory name portion of that slot, which is not the case. The V7 and System III versions did not zero out the entire file name; this is fixed in S5. Check the routine "mkentry".
matt@oddjob.UUCP (02/26/87)
In article <5587@mimsy.UUCP> chris@mimsy.UUCP (Chris Torek) writes:
) In article <105@aob.UUCP> someone finds a directory entry with an
) embedded NUL, and has nothing handy to fix it.
)
) There is still another way, easier yet but more dangerous and
) thrilling. You cannot edit a directory, but you can edit a file.
) And the only difference between the two is one byte. ...
Hoowie! Is that complicated. Sounds fun though. I just
unmount the partition, get out a calculator, and adb the raw
disk.
Time to write "kmem.el" for emacs....
Mattdf@nud.UUCP (02/27/87)
In article <105@aob.UUCP> jim@aob.UUCP (Jim Anderson) writes: >I have an Altos 2086 running XENIX 3.2f. I had a problem show up recently >that defies all my attempts to fix. What has happened is that a file name >in a directory has a null in the middle of the name. The file is actually >the name of a directory. The name used to be 'Doc', but is now 'D\0c'. I don't know how the null got into your directory entry in the first place, but to fix it you will have to use fsdb or a similar tool to change the name in the directory entry through a mechanism outside the bounds of the normal file access system calls. The kernel routine namei (which associates filenames to inodes) can be blamed for making recovery difficult. It takes the null terminated file name that you pass it, adds additional nulls to the end to make it a total of 14 bytes in length, then compares *all* 14 bytes with the name in the directory entry. This makes it impossible to match a directory entry with a null embedded in the name. (Of course, the null *can't* possibly get there in the first place :-)) Namei works this way on System V up to and including System V.3. -Dale --- Dale Farnsworth (NO7K) seismo!noao!mcdsun!nud!df (602)438-5739 ihnp4!mot!nud!df
rlk@chinet.UUCP (02/27/87)
I've used the following to correct directory entries with unprintable
characters. I don't know if it will work with the NULL.
ls -i D* to get the inode number
find . -inum number -exec mv {} new_name \;
You might want to verify that the inode number is unique if you have
multiple filesystems, if they are subdirs of `.'
--
---
UUCP: ..!ihnp4!chinet!uklpl!rlk || MCIMail: rklappal || Compuserve: 74106,1021
..!ihnp4!ihu1h!rlk
---simon@its63b.UUCP (02/27/87)
In article <5587@mimsy.UUCP> chris@mimsy.UUCP (Chris Torek) writes: >In article <105@aob.UUCP> someone finds a directory entry with an >embedded NUL, and has nothing handy to fix it. > >There is still another way, easier yet but more dangerous and >thrilling. You cannot edit a directory, but you can edit a file. >And the only difference between the two is one byte. > >Write another program, `hacki'. >Instead of zeroing the inode, toggle the S_IFDIR and S_IFREG bits. >Run it once on the directory inode, and now it is a file. Use your >favourite binary editor (Emacs works, but be sure it does not make >a backup!), fix the embedded NUL character. Make sure the file size >is exactly the same as before. Then run the hacki program once >more. > >Of course, if you have source, you could alter fsck to check for >embedded NULs in filenames. Another "entity" you can edit is a disk - using a nice powerful editor like adb :-) So, just find the i-number for the directory, then: power-up your adb, find the inode-entry for the directory, from which you can find the data-block list (luckily there're hardly ever any indirect blocks in a directory, unless its very big), find the bad bytes, and change them to something sensible. Why write new code when you do anything you need using existing utilities :-) -- ----------------------------- Simon Brown @@@\ Department of Computer Science /-/@ \ University of Edinburgh, Scotland, UK. / /@ \@@ / /@ \-\ UUCP: seismo!mcvax!ukc!{its63b,cstvax}!simon / /@/-/ \ \ JANET: simon@uk.ac.ed.{its63b,cstvax} / /@/ / /-/ \ \ ARPA: simon@{its63b,cstvax}.ed.ac.uk /-/@/-/ /-/ \ \ ------------------------------ ~~~~~~~~~~~~~~~~~~~~ "Life's like that, you know"
spf@clyde.UUCP (02/27/87)
In article <5587@mimsy.UUCP> chris@mimsy.UUCP (Chris Torek) writes: >In article <105@aob.UUCP> someone finds a directory entry with an >embedded NUL, and has nothing handy to fix it. I have a public-domain c program call rmfile which is interactive and lets you either delete or rename files (including those with garbage characters in their names). I've never paid attention to how it works, but I've used it and it DOES work. It's about 246 lines of c source, and compiled on my PC6300 under XENIX V with no trouble. If there's interest, I'll post it to comp.sources or something. Let me know if you want it. Steve
terryl@tekcrl.UUCP (02/28/87)
In article <1667@oddjob.UChicago.EDU> matt@oddjob.uchicago.edu (Yes, *THAT* Matt Crawford) writes: +In article <5587@mimsy.UUCP> chris@mimsy.UUCP (Chris Torek) writes: +) In article <105@aob.UUCP> someone finds a directory entry with an +) embedded NUL, and has nothing handy to fix it. +) +) There is still another way, easier yet but more dangerous and +) thrilling. You cannot edit a directory, but you can edit a file. +) And the only difference between the two is one byte. ... + +Hoowie! Is that complicated. Sounds fun though. I just +unmount the partition, get out a calculator, and adb the raw +disk. Any wizard worth his paycheck would take it one step further: adb the block partition(don't have to unmount it first) and then use adb to modify the ACTUAL directory block where the entry with the embedded NULL in it. It's doable (I've done it on a V7 file system; I don't know if I'd want to try it on a 4.2 file system). You wanna talk about complicated????
berger@datacube.UUCP (02/28/87)
I've got a similar problem on a System V.2 3b2/310. The root
directory looks like something wrote binary data into the directory
file. Here is an excerpt (there are more garbage listings as well as
normal file listings) from an ls -li:
total 2353
593 -rwxrwxrwx 1 root other 14336 Feb 19 15:10 &^B6SZj^B4^\^CAR^C1
391 -rwxrwxrwx 1 root other 14336 Feb 19 15:08 L^C^TP^AF^V^D^A^X^^
582 -rwxrwxrwx 1 root other 14336 Feb 19 15:09 L^D^^P$^V^Cb*$^D^D^
585 -rwxrwxrwx 1 root other 14336 Feb 19 15:09 R^Apj3 ^C^_R^Apjw
397 -rwxrwxrwx 1 root other 14336 Feb 19 15:08 S^B^^^L^AQS^B ^L^B"
(The data has bee filtered thru see to show the binary data)
The file system seems ok, it fsck's ok and such. When I try to rm one of
these, it says that the file is non-existant and everything runs ok.
Besides dumping and restoring the file system, is there any way to clean this
up? Would it be worth using fsdb (file system debugger)?
It makes me nervous that there are a variety of inodes listed but they all have
the same file size. Could the inode listing be noise as well?
Bob Berger
Datacube Inc. Systems / Software Group 4 Dearborn Rd. Peabody, Ma 01960
VOICE: 617-535-6644; FAX: (617) 535-5643; TWX: (710) 347-0125
UUCP: ihnp4!datacube!berger
{seismo,cbosgd,cuae2,mit-eddie}!mirror!datacube!bergerblm@cxsea.UUCP (02/28/87)
In article <105@aob.UUCP> jim@aob.UUCP (Jim Anderson) writes: |I have an Altos 2086 running XENIX 3.2f. I had a problem show up recently |that defies all my attempts to fix. What has happened is that a file name |in a directory has a null in the middle of the name. I'm not familiar with XENIX, but System V has a utility called fsdb, that let's you do strange and wonderous things to a file system, one of which is to modify the name field of a directory entry. fsdb has to be run on an unmounted file system, probably as superuser (if the permissions are set up the way they should be on the file systems device). In my manual, the second to last example is setting a file name. If XENIX doesn't have fsdb, I guess I can't help you. clri/fsck is probably the only way to go. -- Brian L. Matthews "Facts are impotent against ...{mnetor,uw-beaver!ssc-vax}!cxsea!blm loud and frequent assertion." +1 206 251 6811 Computer X Inc. - a division of Motorola New Enterprises
dave@onfcanim.UUCP (02/28/87)
>) There is still another way, easier yet but more dangerous and >) thrilling. You cannot edit a directory, but you can edit a file. >) And the only difference between the two is one byte. ... > >Hoowie! Is that complicated. Sounds fun though. I just >unmount the partition, get out a calculator, and adb the raw >disk. It's most easily done as a three-step process: 1) unmount the disk, figure out where that inode lives on the disk, adb disk and patch the inode's "format" word to make it into a plain file. 2) mount the disk (while single user!), and edit the directory with adb, emacs, whatever 3) unmount the disk, patch the inode back into a directory, fsck the filesystem unless you have a great deal of confidence in yourself, then bring the machine back up. This minimizes the possibility of clobbering something other than the directory you want to alter.
guy@gorodish.UUCP (03/01/87)
>>In article <105@aob.UUCP> someone finds a directory entry with an >>embedded NUL, and has nothing handy to fix it. > >I have a public-domain c program call rmfile which is interactive >and lets you either delete or rename files (including those with >garbage characters in their names). I've never paid attention to >how it works, but I've used it and it DOES work. Unless this program does so by opening the raw disk and writing directly on the directories, it *cannot* work if you try to use it to handle files with embedded NULs in their names. Since the file names passed as arguments to UNIX system calls are C strings, which *by definition* cannot have embedded NULs, no UNIX system call can possibly refer to such a file (unless you throw in a system call that takes character arrays representing containing NON-NUL-terminated strings as arguments).
kww@cbosgd.UUCP (03/02/87)
In article <563@aw.sei.cmu.edu.sei.cmu.edu> pdb@sei.cmu.edu (Patrick Barron) writes: > >As a last resort you could use 'clri' on the inode associated with that >file (use 'ls -li' to find the inode numner). fsck will certainly allow >you to clear the directory entry once it realizes the inode's been zeroed. > >--Pat. This was in response to renaming a directory that had an embbeded null character in it, something like "D\0c" I believe. What about using 'fsdb' (file system debugger) and run it on the raw file system. All you need to do is find the directory block of the parent directory, and then find the "D\oc" entry and change it to "Doc". With fsdb(1M) it is not too hard since it sorta understands UNIX file systems. If you don't have fsdb (e.g., like BSD systems maybe), adb(1) will also do the trick when run on the raw device corresponding to the file system, although with 'adb', you will have to compute disk offsets, etc. yourself. Oh yeah, to be on the safe side, run /etc/umount on the file system BEFORE you start to ``play''. (Especially if you have to use adb.) Of course if its on root device, you can't do this, but then you ought to do it in single user mode, after having sync(1M)'ed it a few times to make sure everything has been flushed to the disk. Its really not as hard as you think it might be, especially if you have fsdb available. Of course, all this discussion presumes that you have SU priviledges, or writtable special files (E-gad!, talk about security holes!) Good luck!
mjy@galbp.UUCP (Mike Yoffee) (03/05/87)
> the file because unlink can't find 'D\0c'.
....you could port 'fsdb' and get rid of it using
that utility - provided you have access to 'fsdb' or it's source.rbj@icst-cmr.arpa (Root Boy Jim) (03/06/87)
I like Chris Torek's idea on this one: change the directory to a regular file, edit the `file', change it back to a directory, and voila! So I propose ioctls to convert the inode of any open file to any known type (consider the ramifications of converting a text file to a symbolic link). This would also handle a problem I've always wanted to solve: making a link to a given numbered inode. Yow! Have I lost my mind? Is this heresy? (Root Boy) Jim "Just Say Yes" Cottrell <rbj@icst-cmr.arpa> Why did Paul Simon name his album after Elvis Presley's house?
berger@datacube.UUCP (03/08/87)
Too all those who sent helpfull suggestions, in particular the use
of rm -ri .
Yep, the rm -ri . worked great. I felt a bit silly after posting the question.
I was [re]reading the note in mod.announce.newuser that lists the most
asked net questions (and their answers) and this was mentioned there!
I guess its always rereading stuff like that even if you've been on the net
awhile !:=)
Thanks again,
Bob Berger
Datacube Inc. Systems / Software Group 4 Dearborn Rd. Peabody, Ma 01960
VOICE: 617-535-6644; FAX: (617) 535-5643; TWX: (710) 347-0125
UUCP: ihnp4!datacube!berger
{seismo,cbosgd,cuae2,mit-eddie}!mirror!datacube!bergermangler@cit-vax.UUCP (03/12/87)
In article <4768@brl-adm.ARPA>, rbj@icst-cmr.arpa (Root Boy Jim) writes: > So I propose ioctls to convert the inode of any open file to > any known type Rather than an ioctl(), mknod ought to do this when given an existing file. Don Speck speck@vlsi.caltech.edu {seismo,rutgers,ames}!cit-vax!speck