Under some circumstances, the rm command in Git-Bash deletes files that can't be deleted in explorer, cmd prompt, PowerShell, or using C++ standard library calls.
Why?
This is perplexing to me because I know there is no magic here and I assume that they are all using the same Win32 API.
For example, I have database snapshots that remain open and cannot be deleted using the other methods described, but are successfully deleted by Git-Bash rm:
Explorer delete: "The action cannot be completed because the file is open."
cmd: del <path> : "Access is denied"
PS: Remove-Item -Force -Path <path> : "Cannot remove item. Access to the path is denied."
C++ remove() : returns -1
C++ unlink() : returns -1
C++ _unlink() : returns -1
git-bash rm <path> : success
The above can be performed repeatedly on different files.
There are other locked files that git-bash rm deletes successfully as well (I have used it in the past, not recently and I don't have other specific examples).
However it doesn't always work: In a test application I opened a text file using fopen() and none of the methods, including Git-Bash rm, could successfully delete it.
So, how does Git-Bash rm work?
I was able to figure out how this works.
Interestingly, I used Git-Bash's strace utility on it's rm command.
It turns out that Git Bash uses CygWin, and the delete routine is found in the CygWin syscalls.cc file which tries to delete a file in a few different ways.
Eventually it tries to move the file to the Recycle Bin, where it deletes the locked file by opening it with the Windows Driver call NtOpenFile() with a FILE_DELETE_ON_CLOSE flag, then closing it using NtClose().
Not sure if it would be proper to copy CygWin's code into the response here, but details on all of the above can be found in the link provided.
Related
On the docker image debian:stretch-slim, couldn't delete a specific folder on a NFS drive, using rm -rf /folder-name as root (or rm-rf * after entering the folder-name).
Got the following error back:
rm: cannot remove 'test-ikmgfjhv/dev/.nfse47cf31c6b1dd52500000009': Device or resource busy
After a lot of searching, eventually got to the following link:
https://uisapp2.iu.edu/confluence-prd/pages/viewpage.action?pageId=123962105
Which Describes exactly why those files exist in NFS and how to handle them.
As I wasn't using the same machine the process runs on (another container), so in my case, I had to work around that and first make sure the process using the file is being killed on the first machine, then try to delete it on the second one, according to the project's needs.
It is possible that the .nfs file is attached to a process that is busy or running (like an open file, for example, a vim file).
For example, if the hidden file is .nfs000000000189806400000085, run this command to get the pid:
lsof .nfs000000000189806400000085
this will output the PID and other info related to that file
then kill the process:
kill - 9
Be aware that if the file was not saved you will lose the information.
While running any command if you get error like :
/home/mmandi/testcases/.nfs000000e75853 :device or resource busy.
Go to the directory where this file is being shown.
For e.g - In this case : /home/mmandi/testcases/
Do following :
# ls -la : This will display contents of the directory along with files starting with "."
Here it displays the .nfs000000e7585 file.
# lsof .nfs000000e7585
This will list down the PID
# Use Kill -9 PID.
I successfully compiled a MUD source code, and it says in the instructions to start up the server using
nohup ./startup &
although when I do this it gives me this error:
$ nohup: ignoring input and appending output to `nohup.out'
nohup: failed to run command `./startup': Permission denied
I have looked all over the internet to find the answer. A few of them said to put my cygwin directory in the root folder (I am using windows 7) and its directory is C:\cygwin
so thats not a problem.. Can anyone help me with this please??
Try chmod +x startup, maybe your startup file is not executable.
From "man nohup":
If the standard output is a terminal, all output written by the named
utility to its standard output shall be appended to the end of the
file nohup.out in the current directory. If nohup.out cannot be
created or opened for appending, the output shall be appended to the
end of the file nohup.out in the directory specified by the HOME
environment variable. If neither file can be created or opened for
appending, utility shall not be invoked. If a file is created, the
file's permission bits shall be set to S_IRUSR | S_IWUSR.
My guess is that since "sh -c" doesn't start a login shell, it is inheriting the environment of the invoking shell, including the HOME environment variable, and is trying to open it there. So I would check the permissions of both your current directory and $HOME. You can try to touch test.txt in current directory or $HOME to see if you can perform that command.
As staticx writes, check the permissions of the directory (and the user) - and the executable.
Instead of using nohup:
check if nohup is needed at all, try ./startup </dev/null >mud.out 2>mud.err &, then close the terminal window and check if it is running
or just run ./startup in a screen session and detach it (<ctrl>+<a>,<d>)
I am using C++ on Ubuntu. I have been using the command:
system("mkdir new_folder");
to make a new folder called new_folder. However, if that folder already exists, C++ outputs an error message (and continues to run afterwards).
Is there a way to stop the error message from printing out?
For this particular command use mkdir -p new_folder.
Generally, you want to fork your process and on one of the branches redirect stdout and stderr to /dev/null or similar then do exec to replace the process with the new one.
I’ve produced a C++ program in Eclipse running on Redhat, which compiles and runs fine through Eclipse.
I thought that to run it separately to Eclipse you use the build artifact which is in the directory set via the project’s properties.
However this executable doesn’t run (I know it’s an executable as I’ve set it to be an executable via the project’s properties and it shows up as such via the ls command and the file explorer).
When attempting to run it using the executable’s name, I get the error:
bash: <filename>: command not found
When attempting to run it as a bash file:
<filename>: <filename>: cannot execute binary file
And when running it with "./" before the file name, nothing happens. Nothing new appears in the running processes and the terminal just goes to the next line as though I’d just pressed enter with no command.
Any help?
You've more or less figure out the first error yourself. when you just run <filename> , it is not in your PATH environment variable, so you get "command not found". You have to give a full or relative path when to the program in order to run it, even if you're in the same directory as the program - you run it with ./<filename>
When you do run your program, it appears to just exit as soon as you start it - we can't help much with that without knowing what the program does or see some code.
You can do some debugging, e.g. after the program just exits run echo $? to see if it exited with a particular exit value, or run your program using the strace tool to see what it does (or do it the usual way, insert printf debugging, or debug it with gdb)
Hey! I am trying to get ant installed and actually already did following this instructions however, I get this error:
Buildfile: build.xml does not exist!
Build failed
which it says there I might get so I just tried executing the next command it says I should(since I'm under Windows it's this one):
build -Ddist.dir=<C:\Ant> dist
anyway I get "access denied" when hitting enter and I can't figure out why. I also tried
build install
and
build install-lite
but I always get that message =/ any ideas why? or what am I doing wrong?
Edit
Without the < > I get a:
'build' is not recognized as an
internal or external command, operable
program or batch file.
Edit2
Well, my ANT_HOME is in C:\Ant and I'm trying to run the command while placing myself on that folder, isn't that correct?
Ant is already installed correctly if you're getting this:
Buildfile: build.xml does not exist!
Build failed
The "build" commands you tried next are only for building Ant from source. You don't need to do this, since Ant is already installed.
In general, the "'XXXX' is not recognized..." means XXXX is not a command/executable, or that it's not on your PATH.
You are taking the < and > symbols too literally. Take them off.
Kind of like C:\Users\<Your ID> the symbols are placeholders.
Are you in the right directory? Do you have something else called build in your path that may be superceeding the build you are trying to call. Can you create a file in that directory?
PROMPT> copy con foo.txt
asdf
asdf
sdf
CONTROL-D
Not sure if this helps but the first question that comes to mind is "do you have appropriate permissions on the file"?