34
votes

How about sharing your favorite lessons learned moments?

locked by Michael Mrozek Sep 16 '11 at 12:17

This question exists because it has historical significance, but it is not considered a good, on-topic question for this site so please do not use it as evidence that you can ask similar questions here. This question and its answers are frozen and cannot be changed. See the help center for guidance on writing a good question.

Read more about locked posts here.

  • Funny how many of the answers involve the rm command. Or sad. – Barry Brown Aug 31 '11 at 17:09

53 Answers 53

2
votes

Debian dist-upgrade to unstable (or was it testing?) on a remote production server.

Ignored warning about libc switching to nptl threads.

Not sure where installation actually failed, but I was left with one root console on a dial-up line with every single app exploding. Just one running ssh and bash.

Had a lot of fun of recovering it. Uploaded statically linked dpkg, rolled back libc, built a custom kernel with RAID support. I think that took around 3 hours with my dial-up line. When I finally rebooted it, fsck took like 10 minutes. Quite painful 10 minutes I should say.

1
vote

I wanted to delete a file and its backup copy (file and file~) on a SuSE system I rarely used and didn't notice that command completion was configured differently. I typed rm fi<tab> and expected the cursor to be after the last letter. The system ignored ~ files and inserted a space after the file name.

So I wanted to type

rm file*

and I did

rm file *

Now I am always checking what the suggested completion is.

0
votes

I did chown -R /usr. Somehow, I thought it was a good idea. Subsequently, a puppet script did some bad stuff, and somehow I lost sudo rights.

Our qualified sysadmin wasn't able to recover my sudo rights, and I had to re-image the machine.

1
vote

Installing Debian (with net-install) on a computer and realise after rebooting that I had somehow skipped the part where one is supposed to choose which packages to install.

Sure, a non-graphical system with basically only pwd, ls and cd is working just fine ;).

  • huh? Debian net-install gives you more than enough. Most servers I install don't get much more during the basic install. Who needs a graphical system anyway :) – blauwblaatje Jul 28 '11 at 7:41
1
vote

1st ex employee: How do I do (something trivial)?
2nd ex employee: sudo rm -rf /
1st ex employee: Haha ok
...
1st ex employee (Having forgotten that he'd just sudoed something else, so it didn't ask for his password again) : oh F**K!!!

1
vote

The first time I installed GNU Linux on my desktop, I installed Debian, no help, I only installed the basic system, no GUI. And I was like:

"OMFG OMFG WHAT DO I HAVE TO DO NOW?... I will need to go back to Windows"

But then I remembered how to install packages (first time in GNU Linux, only a jose@debian:$ output and only read some things from Debian) and the name of an IRC Client: IRSSI and the name: GNOME and then I installed them... since that day, I install IRSSI and GNOME in every machine...

It feels... good experimenting with your computer xD

1
vote

Many web apps like magento or eZ Publish have a var/cache/ folder and a way to clear the cache that is faster than going in the app backend is to do this:

rm -rf var/cache

After doing this a couple of times, it is scientifically proven that you always end up:

  • Either doing this in the root of your server; or
  • Adding a / before var

Lesson learned: create an alias for this command with an absolute path in it.

2
votes

First one

One time, I had defined an alias to help me clean temp files:

alias clean_dir_tmp="find /tmp -maxdepth 1 -user **** | grep *** | xargs rm -rf"

and of course one day, when I had forget what the alias was doing I mistyped:

$ clean_dir_tmp *

I think I lost a good couple of seconds before I realized what was happening... :(

Second one

I was working as usual with my Mac laptop and my Ubuntu desktop. When I plugged the external HD of the Mac (HFS+ filesystem) in Ubuntu, I noticed that the owner was ?????. My UID was different on my Mac and on my desktop so since I wanted a nicer output with ls -l, I decided to change that.

So I modified the UID on the mac and launched a big chown -R * on the HD from Ubuntu. The only thing that I didn't know was that the HFS+ driver for Linux was not stable. To this day, I haven't been able to mount this HD on Linux or my Mac again...

  • I've been burned by aliases, too, so nowadays I use bash functions instead, which are a much better-designed feature. – Neil Mayhew Jul 5 '11 at 5:20
1
vote

Twenty minutes ago, I was painstakingly recreating a complex directory structure from files I had on my other PC. I decided to run du to see if it was near completion. Given the size, I knew it would take few minutes, so went to get some coffee.

On my return I noticed, to my severe dismay, that instead of running

du -hs /path/to/important/folder

I had absent-mindedly run

rm -r  /path/to/important/folder
1
vote

Back in the day you had to do erase the first 512 bytes of a partition to properly format FAT drives from Linux. This is done using the dd command.

dd if=/dev/zero of=/dev/hda1 bs=512 count=1

Except the FAT partition was /dev/hdb1

I didn't realize what had happened until after I rebooted. Luckily I was able to recover it by re-installing Lilo, or something.

1
vote

Trying to get the Xwindows driver for my Nvidia card working when Fedora initially released the Nouveau driver. I had downloaded the Nvidia source to compile and install myself as I had many times in the past, but this release, I could just not get it to work. There were quite a few steps to find in the Fedora Forums to completely disable the Nouveau driver, and get the Nvidia driver working. Quite painful to say the least.

  • Nvidia makes my brain scream. I only ever use intel GPUs these days. – Eli Frey Aug 15 '10 at 4:16
2
votes

I don't keep track of errors on my own boxes, but from the last 15 years here are my two work fatalities:

1995: Standard 'rm -rf as root' failure on a single box (not my design!) acting as a combined NIS master server + sole DNS server for the company + primary SMTP/POP3 server for the company. It was SunOS 4.1.3_U1 as I recall. Unsurprisingly, like the witness to a horiffic crime, I do not remember the ensuing 48 hours.

1998: Ran newfs (SunOS) on a production AFS (now OpenAFS) file server instead of the replacement we were standing up. Spent the afternoon and night restoring from tape.

2
votes

I mapped the CapsLock to ESC on the entire system. When I did it, the Capslock was on.

A reboot removed the permanent state of CapsLock. It was mapped to ESC

It wasn't really painful but I felt stupid when I realized what I'd done!

3
votes

On a Debian installation back in 1999. 14 floppy disks for the basic installation.

I tried to get xfree86 working. But X didn't start. I had to find out the settings of my graphics card (memory, horizontal and vertical refresh rate), which was completely undocumented. It turned out superprobe found out the correct amount of internal graphics card memory (1024 kB).

But it took me nearly a week to find out that the resolution setting (1024x768) didn't work. I had to switch it to 640x480 until the graphics card finally worked (at 1024x768... buuuuuug....).

I tried to get the serial port mouse to work on COM1. So I tried to get the mouse to work. Reading a book (back then I had no usable high-speed internet), I tried with

/dev/ttys0

And it didn't work and didn't work. It took me nearly another week to find out this was because I needed to type the S is uppercase, not lowercase...

/dev/ttyS0

It was about then when I finally realized what 'case-sensitive' really means.

3
votes

I had two drives installed at one point and had the root filesystem of the second drive mounted in a directory within /mnt. I was in that directory and tried to delete var but ended up typing rm -rf /var instead. Some instinct seemed to kick in that said var must be preceded with a slash!

When I realised what I'd done I immediately hit Ctrl-C but it was too late. My rpm database had long since left the building. I spent ages getting everything back to normal.

Now for the painful part.

I go back into that directory in /mnt to resume what I'd been doing. What do I type? Well, let's just say that instinct kicked in again.

At least I was able to restore the system a lot quicker the second time ;)

  • 1
    I take it this was before you found out about chroot(1)? – Kevin M Oct 9 '10 at 22:13
6
votes

I was a lab assistant for a Linux class. One of the students called me over because she could no longer su - because she was getting permission denied. OK, she's misremembered/mistyped the password. Reboot into single-user mode and reset. What?! su STILL doesn't work?! It MUST bow to my will! So I reboot into single user mode to find out what she did. I realized that she ran chmod -R 777 /var/www/html/drupal-6.19 /

Note the space between the directory name and the final slash.

After a few minutes of "I really don't want to have her reinstall, so what is this doing and how.", I managed to find that /bin/su now had file permissions of 777. That can also be read as file permissions of 0777, which removes the setuid bit from /bin/su. A quick chmod u+s /bin/su and I was a hero.

  • 5
    That was the only thing you needed to change to unbreak her system? – Michael Mrozek Oct 10 '10 at 0:10
  • 1
    I'm sure it was still broken, but maybe it worked well enough for the rest of the class. – Stefan Lasiewski Oct 10 '10 at 3:53
  • 1
    It did work well enough for the rest of the class; the school evidently has a requirement that all student drives be reinstalled every quarter. And she wasn't going on with Linux after this quarter was over, so I didn't bother fixing it the rest of the way. – Kevin M Oct 10 '10 at 12:43
7
votes

While in my 2nd year of studying computer science we were given a homework assignment to write a program in C that would spawn a number of subprocesses with fork and make them communicate with pipes in a "circle" and figure out which one should be the "leader".

We were still quite noobs back then and most of peple didn't have any Linux machines, so we worked on our accounts on our faculty's main server (which was hosting official site and staff accounts and sites as well). Most of the people wrote forkbombs at some stage of trying to do the homework. Over half of my group got to the abusers file. That was the highest load on that server in a looong time :)

  • that was silly of them... they shouldn't have put you on the same machine as webservers and stuff. (My school has the same thing, but we have severs dedicated to remote shells so the yearly forkbombs don't really get too out of hand) – Spudd86 Feb 9 '11 at 3:26
  • Microsoft got the faculty to get into partnership and windows and .net were much promoted then already. Nobody expected a bunch of noobs on the server :) – naugtur Feb 10 '11 at 17:21
16
votes

I remember trying to send a SysRq key sequence to a remote machine...

...but it was captured by the local one.

18
votes
rm -f * ~

and

rm -rf ${DIR}/

when DIR was not set!

  • 4
    You mean ${DIR}? Because $(DIR) would try to execute DIR command. – Denilson Sá Maia Nov 13 '10 at 22:49
18
votes

vi and Caps-Lock vs. /etc/passwd

  1. Connect to an old Solaris box using an old serial terminal that doesn't refresh the screen correctly.
  2. su -
  3. vi /etc/passwd. There is no vipw, and "we're just making minor edits" anyways.
  4. Hit Caps-Lock key and don't notice.
  5. Hit j a couple times to scroll down. Ignore the fact that you actually just typed J ("Join"), which combines this line with the next line. The serial terminal screen was not refreshing correctly, so you didn't see that you just combined the first 5 lines into one Loooooong line, thereby corrupting the first 5 users ('root', 'daemon', etc).
  6. Finish your OTHER intended edits to the file, way down at the bottom.
  7. Save file.
  8. Log out.

I did this once. Amazingly, the system remained functional for months. Cronjobs ran fine, no errors stood out in the logfiles.

We didn't notice this problem until we rebooted the system months later and couldn't log in at the console. ps showed a bunch of jobs owned by UID '0' not by user 'root'.

You could not log in as root, nor run su or su -, and there was no sudo on this box. There was no floppy drive, the CD-ROM was busted and no USB ports (so no external CD-ROM). Single user mode did not work, because you need to type in the password for root, and that comes from /etc/passwd.

  • 2
    Surely "J" is the "Join" (as in join lines together) command? – dr-jan Aug 19 '10 at 18:07
  • 2
    Wait a minute, I don't need to type the password to boot into single user mode. In fact that is how I reset the root password if I lose it. Or was it different in the old time? – phunehehe Nov 20 '10 at 16:05
  • On Solaris, you need to know the root password to enter single user mode. I had to do this 2 weeks ago actually. – Stefan Lasiewski Nov 22 '10 at 5:38
  • @phunehehe I think your thinking of the init=/bin/sh trick. That does indeed not require the root password. However runlevel 1 does still require root pass – Patrick Feb 7 '11 at 16:16
21
votes

I wiped the partition table of my main drive by accident, thinking I was working on another drive.

With scrollback, careful use of df, memory, and luck I was able to recreate it exactly, rewrite it, reboot, and hope... And it worked.

  • 6
    A friend did this, and I helped him restore the table. Unfortunately, we didn't have the previous values for the partition sizes and offsets, so we used a bash loop with dd reading the first 4k block of every cylinder piped into file - to find the superblock and thus the start of the filesystem. This was on a live CD and there wasn't enough RAM to do everything we needed to do (which included installing a package or two) so we piped into a process running in ssh on another machine. – Neil Mayhew Sep 12 '10 at 23:37
  • 2
    Ouch. That's why I always use sfdisk -O to back up the partition table, always. FYI: cgsecurity.org/wiki/TestDisk can automate what @Neil Mayhew's did. – ephemient Nov 17 '10 at 18:25
  • 2
    I was there too, testdisk saved my box – phunehehe Nov 20 '10 at 16:12
  • These days, we have gpart, which looks for things that might resemble filesystems and constructs a partition table from that. – Simon Richter Mar 10 '11 at 17:54
29
votes

I meant well, I really did. Trying to chmod recursively a directory and ended up swapping ./ with /.

As root of course, because only with root can true pain (and thus enlightenment) be achieved.

  • 23
    +1 for quotable "Only with root can true pain (and thus enlightenment) be achieved." – Stefan Lasiewski Aug 27 '10 at 22:49
33
votes

Surprised nobody else has mentioned this one yet:

rm -rf .*

(While attempting to remove all hidden files and subdirectories, completely forgetting that it will recurse into . and ..)

  • 3
    While that has burned me in the past, many versions of rm won't do that now. I tried on Darwin and got the error rm: "." and ".." may not be removed. – Stefan Lasiewski Aug 24 '10 at 3:43
  • 7
    I do that all the time, on all kinds of platforms. Works as intended on OpenBSD, NetBSD, Linux, and OpenSolaris. – polemon Aug 29 '10 at 13:47
  • If you want to match all dotfiles except for . and .., use .[^.]*. (Well, this will actually miss all files starting with .., but usually there is only one.) – Sven Marnach Jun 4 '11 at 23:38
  • @Sven: Another way is .??*, which I find easier to type. This one won't match two-letter dot files like .a, but those are unusual too. I search the config files in my home directory with grep -r .??*, for example. – Neil Mayhew Jul 5 '11 at 3:38
37
votes

When I first started working as a user consultant for the university I was attending, I was given limited sudo rights to help students who had lost/forgotten their passwords. sudo passwd <username> was my new friend. An hour after my orientation, my curiosity got the better of me and I typed in sudo passwd and stared in horror at the prompt for a new password. I was a bit scared to ^C my way out of it, thinking (erroniously, it turns out) that I might leave the account in question in a transient state, so I entered a password and immediately walked upstairs to the hallowed 2nd floor domain of the campus SuperUser and asked if he would like to know the root password of the main system.

  • 4
    heh, ownage! Confused Deputy :P – Spudd86 Aug 17 '10 at 19:33
  • 30
    you could have entered a mismatch password at the 2nd confirmation prompt, so nothing would have been affected then and passwd would exit. – Wadih M. Sep 1 '10 at 14:53
  • 1
    @Wadih: passwd behaves funny when run as root. For example, when failing the typo check, it asks again. – György Andrasek Mar 6 '11 at 3:57
  • You could have just changed it back to what it was before. – HandyGandy Mar 10 '11 at 5:24
  • 1
    @HandyGandy he didn't know the original root password. He just had sudo rights to run passwd. – gnud Mar 10 '11 at 14:28
68
votes

I was curious if chmod 000 / would work.

Well, flawlessly. A few minutes later I was searching for a rescue CD.

  • 11
    Wow I never even considered that until you mentioned it. Good one. – Stefan Lasiewski Oct 11 '10 at 17:57
  • What happens if you do it on a regular file ? – Luc M Nov 16 '10 at 15:08
  • @Luc I guess you just need to chmod it back – phunehehe Nov 20 '10 at 16:07
  • I usually do such 'experiments' in my Arch VM. If it bombs I just roll it back. – André Paramés Jan 20 '11 at 23:47
  • I made this mistake once. It was highly embarrassing, but a valuable lesson. – Charles Sprayberry Jul 12 '11 at 3:38
7
votes

This happened to me last year. I was removing some files from the server using a temporary variable:

rm -rf ${prefix}*

Guess what? The variable $prefix was not defined!
You can imagine the disaster... it resulted in some very critical files deleted.

I almost broke the Control-C and ran to the CPU to remove the network cable!!

Hahaha I'm sure someone had already done this...

30
votes

Had a friend run :() { :|:&}; : on a remote server where we didn't have console access to. Couldn't reboot it, completely frozen, production server.

Broken down (by request) to make it a bit more readable.

:() # Define ':' as a function. Every time we say ':' execute the following code block
{ # Start of code block
    : # Call ':' again. 
    | # Pipe output to...
    : # Another ':' 
    & # Disown process. 
    # All on one line this would read :|:&, 
} # End of code block
; # End definition of ':' as a function
: # Call ':'

It might be easier to look at it as

bomb() { bomb|bomb& }; bomb
  • 1
    Would You be so kind to explain what it does? I can't figure it out and I don't want to try it ;) – naugtur Aug 19 '10 at 8:54
  • 23
    Look's like a fork bomb. Don't try this unless you really know what you're doing. – Zaid Aug 19 '10 at 9:29
  • 5
    Yup. It's a forkbomb. And a nice one :) It seems to fork bash interpreter as it endlessly tries to parse it. If You skip the last : it does nothing. And it doesn't use memory at all, just forks a lot. [Yes, I did try it:)]. Effects can be blocked with a quota on number of processes per user. – naugtur Aug 19 '10 at 10:16
  • 8
  • 4
    +1 for the above comment. It's far easier to understand than it looks. – Umang Aug 19 '10 at 11:54
5
votes

A company that I used to work for had its product running on SCO. I was doing some debugging about applications getting very slow on our demo server and at the same time there was a bunch of customers being given a demo/lecture about upcoming new features.

So, I ran the application that used to get stuck, did my stuff on it to verify the root cause, but since it was still "stuck", I tried to kill it:

pkill -9 mytestapplication

What I did learn was that pkill doesn't do exactly the same on SCO as it does on linux =)

... It basically kills everything the user has access to, and with root... that's everything =)

  • 1
    Could you please describe what it does on SCO? I have no idea, and I couldn't find it easily on google. – Denilson Sá Maia Nov 13 '10 at 22:55
  • 1
    it basicly kills everything the user has access to, and with root .. its everything =) – rasjani Nov 18 '10 at 12:09
5
votes
git init
git clean -f

This does not remove the repository. This removes everything that is not in the repository.

After trying to get rid of the existing repo and then start source control again (on the completed first version of a project), these two commands nuked my entire code.

  • git rebase (with squashing) and git filter-branch (for removing files) are your friends :-) By this I mean that there are safer ways to remove cruft from your repo history and end up with a relatively clean project. I usually make a backup copy of the current branch before extensive rebasing. – Neil Mayhew Jul 5 '11 at 5:14
1
vote

I was once trying to zero out a USB thumbdrive using dd.

Needless to say, when tty1 started spitting out ReiserFS errors from my root partition, I had to reinstall...

Not the answer you're looking for? Browse other questions tagged or ask your own question.