database { host: localhost; port: 3306; database: css; user: web; password: 12345; }

Edit-in-place. Explain the differences between a true edit-in-place, that changes the file itself, vs. one that replaces the file. Explain the advantages and disadvantages of each. Give at least one example of how to accomplish each method.

I would probably suggest reading Vim's :help on the 'backupcopy' option. If pressed: one is creating a new file + renaming on top of old file; the other is truncating the old file and then overwriting it with data (or overwriting and then truncating). The difference is (1) what happens if the program crashes in the middle of the write, and (2) what happens if some other program still has that file open. E.g. one method works for replacing executables that are currently being executed while the other fails with EBUSY. Another e.g. is crontab -e that wants the same file back and not a new one with the same filename.

Fully explain the standard base UNIX/Linux file permissions for at least non-ancient implementations thereof Include not only explaining SGID on directories, but how that varied historically going back at least to the preceding common implementations on that and how they varied/differed.

And this is where I would get stuck, because I don't know (and don't much care).

The rest of this I think I know, except for practical effect of dr--r--r-- directory permissions. You can ls but not stat/open the files inside?

Tell me about ssh certificates. Yes, ssh, not ssl, and not keys, but certificates.

All I know is that they exist and can be used to grant access without adding each key into authorized_keys.

rsync - two large files, same permissions, length, and mtime, but their content differs. If you use rsync to ensure that the 2nd of those files matches the first, do you have to use any non-default options for that to actually ensure that the file contents will get matched?

I'm pretty sure I do, because rsync has optimizations. The man page says the option is --checksum/-c.

Explain, atime, mtime, and ctime.

Last access (with digression about mount -o noatime/relatime), last modification (of file contents), last inode change (eg. chmod/chown). I remember doing experiments checking if opening a file for write/append access and writing zero bytes to it changes the mtime. (IIRC it doesn't.)

Bonus, for filesystems that support btime, explain that also.

Inode creation ("birth") time. When I last investigated it seemed a very non-standard thing with almost no POSIX APIs exposing it, requiring debugfs and such to see on ext2. I now see that even tools like ls can show birth times.

If one can do so, how can one set/change: atime? mtime?

/bin/touch, or the utimes() syscall.

ctime?

umm, chmod?

btime?

create a new file, move it on top of the old one?

Bonus: explain how to change the ctime of a file to a given arbitrary legitimate timestamp.

Ooh, is that possible? Without temporarily changing the system clock? Or fiddling with debugfs/banging bits on an unmounted filesystem?

Extra bonus: give at least two quite distinct ways to do that.

The above (changing system clock + debugfs).

Explain what eval does in shells that are (or can be) POSIX compliant (e.g. dash, bash, etc.). Give at least one example usage.

Evaluate its parameters as a shell command in the current shell.

eval "$(ssh-agent)"

Same question, except for exec.

Replace the current shell process with a new process running the specified command. All of my wrapper scripts that, idk, set extra environment variables (export MOZ_USE_WAYLAND=1), end with an exec /usr/bin/original-binary "$@".

Likewise on shells, explain exactly what is substituted in for $() or ``, be sure to be fully accurate regarding ending newline(s) or trailing empty lines or lines that only contain space characters.

Whee I would fail this. I almost never use $() without wrapping it in "", except when I know it will produce one word of output (like $(pidof process) when I know one and only one copy of it is running).

What if either of those are within " (double quote) characters?

The output is preserved exactly, I think.

What difference, if any, does that make, and in what contexts? Also explain the difference between $() and `` and why it's often preferable to use the former rather than the latter.

You can nest $()!

how can you create a file with a newline character in the name of the file?

I would try

$ touch "file
name"

and then rm -i ./file<tab> before it has a chance to mess things up.

To merely create a file, folks often give example using the touch command. In standard shells, how can one do that much more concisely, and without using any external command at all.

>> filename.txt

probably. I have used > file.txt to truncate files, but I've needed a replacement for touch. (Although > file.txt would also create, but I would fear accidentally overwriting an existing file if I mistype the filename.)

Some daemon process is running, you have it's PID. How do you determine what file(s), if any, it's using for stdin, stdout, and stderr, and without using the lsof command.

Good old ls -l /proc/$pid/fd.

for any block device, how can you determine its precise size, without reading it?

sfdisk -s /dev/thingy. (Only I see the manual now says it's deprecated and I should be using blockdev --getsz or blockdev --getsize64.)

I have also occasionally poked in /sys/class/block/* for this information.

two block devices under /dev have the same major and minor number. Are they the same device?

Yes.

Are they the same file?

Ehh. What is a 'file'? There are directory entries and there are inodes. Is a file an inode?

(Now I'm curious if one is allowed to hardlink device nodes. I don't see why not, TBH.)

They could be two names to the same inode, or they could be two separate inodes, or one could be a symlink to another.

For a given device under /dev, how can you locate all files / pathnames under /dev that refer to the same device?

Hm. find /dev -ls gives me what looks like major, minor device numbers in the size column. I could do something with grep and eyeballing. I don't see any options on matching on device numbers in find's man page.

I could write a Python script that uses os.walk() and os.stat() if I needed something automated and reliable.