fMRI Tutorials

Basic Linux Commands

7 min read

Most lab work — fMRI preprocessing, container runs, dataset transfer — happens on a remote Linux box (workstation or the LRZ cluster). This page is a survival kit: enough shell to navigate, edit configs, move data, manage permissions, and watch processes. Companion pages: tmux for persistent sessions, LRZ Cloud Computing for the cluster, Tips and Tricks for assorted gotchas.

Shells such as Bash, Zsh, and Tcsh provide command-line environments for Unix-like systems, each with slightly different syntax. Bash is the default on most servers; Zsh adds better completion and plugins; Tcsh shows up because AFNI uses it. Stick with Bash unless a tool tells you otherwise.

1 Files, directories, and navigation

Print Working Directory (pwd) shows the current directory.

pwd

List the files in a directory (ls); -l is long format, -h is human-readable sizes, -a includes hidden dotfiles.

ls -lah directory

To list files recursively, use ls -R.

Change directory (cd):

cd to/sub/folder
cd ~          # home
cd -          # previous directory
cd ..         # up one level

Copy (cp), move/rename (mv), remove (rm):

cp original duplicate
cp -r folder1 folder2     # recursive copy for directories
mv old_name new_name
rm file.txt
rm -r folder/             # recursive; be careful

Make a new directory (mkdir -p) or remove an empty one (rmdir):

mkdir -p $BIDS_DIR/derivatives/fmriprep
rmdir empty_folder

Show a text file in the console with cat, or only the head / tail:

cat readme.txt
head -n 5 readme.txt      # first 5 lines
tail -n 20 fmriprep.log   # last 20 lines
tail -f fmriprep.log      # follow live (e.g., watch a running job)

Selectively show columns from a delimited file using cut:

cut -f 2-5,8 -d , exp1.csv

Shows columns 2–5 and column 8 from exp1.csv (delimiter ,).

Symbolic links (ln -s) — useful for pointing into shared datasets without copying:

ln -s /data/shared/MyStudy $HOME/MyStudy

Inspecting BIDS layouts

tree -L 2 (if installed) shows a directory two levels deep — handy for sanity-checking a BIDS dataset:

tree -L 2 $BIDS_DIR

If tree isn’t available, use:

find $BIDS_DIR -maxdepth 2 -type d

Count all BOLD files across subjects:

find $BIDS_DIR -name "sub-*_bold.nii.gz" | wc -l

Size each subject folder (BIDS structure):

du -sh $BIDS_DIR/sub-*

2 Pipes, redirection, and wildcards

Store a command’s output in a file with > (overwrite) or >> (append):

head readme.txt > head.csv
echo "done" >> log.txt

Chain commands with the pipe symbol |. For example, show the last 5 trials of the first block (48 trials) from data.csv:

head -n 48 data.csv | tail -n 5

Wildcards in filenames:

  • * matches any string (e.g. sub-*_bold.nii.gz)
  • ? matches a single character (sub-0?.txt matches sub-01.txt but not sub-11.txt)
  • [...] matches any character inside the brackets (201[78].txt matches 2017.txt or 2018.txt)
  • {...} matches comma-separated patterns ({*.txt,*.csv})

Arithmetic expansion: $[] or $(()) evaluates the expression. echo $[3 + 2*4] outputs 11.

Variable expansion: build strings from variables.

a=sub
echo ${a}-01.txt    # outputs: sub-01.txt

Braces {} disambiguate when the variable name sits next to other characters: echo $ab.txt looks for the variable $ab, whereas echo ${a}b.txt outputs subb.txt.

3 Searching and text processing

grep — find lines

grep PATTERN file prints matching lines. Common flags:

  • -c: count matches instead of printing
  • -i: ignore case
  • -l: list filenames with matches
  • -n: include line numbers
  • -v: invert (lines that do not match)
  • -r: recursive into subdirectories
grep -rni "task-bisection" $BIDS_DIR

wc — counting

wc (word count) prints characters, words, and lines. -c, -w, -l select just one.

For example, count trials with curDur = 0.5 from Exp1.csv (column 8 is curDur):

cut -f 8 -d , Exp1.csv | grep -v curDur | grep 0.5 | wc -l

sort, uniq

sort orders lines; -r reverses, -n sorts numerically, -f is case-insensitive. uniq collapses adjacent duplicate lines, so always sort first. uniq -c prefixes each line with its count.

Count trials per sample duration:

cut -d , -f 8 Exp1.csv | grep -v curDur | sort | uniq -c

sed — find/replace

sed (stream editor) does substitutions:

sed 's/unix/linux/' aText.txt      # first match per line
sed 's/unix/linux/2' aText.txt     # 2nd match per line
sed 's/unix/linux/g' aText.txt     # all matches
sed 's/unix/linux/3g' aText.txt    # from 3rd match to end of line

The delimiter / can be | (handy when replacing paths).

awk — column-aware tool

awk treats each line as columns. One-liner: print column 2 where column 3 equals 0.5:

awk -F, '$3 == 0.5 {print $2}' Exp1.csv

Task

Suppose you want to rename all filenames containing _run to _task-bisection_run. Think about pipes.

4 SSH, file transfer, and remote work

SSH basics

ssh user@host.lrz.de

Generate a key pair (Ed25519 is the modern default):

ssh-keygen -t ed25519 -C "your.email@lmu.de"

This creates ~/.ssh/id_ed25519 (private — never share) and ~/.ssh/id_ed25519.pub (public — copy to the server). Add the public key to the server’s ~/.ssh/authorized_keys, or use:

ssh-copy-id user@host

~/.ssh/config — your friend

Set up short aliases in ~/.ssh/config:

Host lrz
    HostName login.lrz.de
    User di12abc
    IdentityFile ~/.ssh/id_ed25519

Host workstation
    HostName 10.0.1.42
    User mystudent
    IdentityFile ~/.ssh/id_ed25519

Now ssh lrz is enough. See also LRZ Cloud Computing for cluster-specific setup.

scp vs rsync

scp is fine for a single file. For datasets — and especially anything that might be interrupted — use rsync: it resumes, skips unchanged files, and shows progress.

# single file
scp results.csv user@host:/path/to/
 
# directory, resumable, with progress
rsync -avz --progress local/directory user@host:/path/to/destination
 
# pull from remote
rsync -avz --progress user@host:/path/to/directory local/directory

Flags: -a archive (preserves permissions, timestamps), -v verbose, -z compress in transit. Add --dry-run first to preview.

wget vs curl

Both download files. wget is simpler for “grab this URL into a file”; curl is more flexible (REST APIs, custom headers, uploads).

wget https://example.org/dataset.tar.gz
curl -L -o dataset.tar.gz https://example.org/dataset.tar.gz

screen and tmux

Long jobs over SSH die when your connection drops — unless you run them inside a terminal multiplexer. screen is the old standby; the lab uses tmux (see tmux).

5 Permissions and ownership

Linux permissions cover read (r), write (w), execute (x) for three classes: user, group, other. ls -l shows them as -rwxr-xr--.

chmod 600 ~/.ssh/id_ed25519           # private key: only you can read/write
chmod 755 my_script.sh                # owner rwx, group/others rx
chmod -R u+rwX,g+rX,o-rwx /data/private

Change ownership / group:

chown newuser file.txt
chgrp -R labgroup /path/to/shared    # recursively assign group

Group ownership matters on shared servers: set the group of a shared dataset to your lab group so collaborators can read it.

6 Disk space and process monitoring

Disk usage

df -h                          # free space per mount, human-readable
du -sh $BIDS_DIR               # total size of a folder
du -sh */ | sort -hr           # subdirs of cwd, largest first

Processes

top                            # live process list (everywhere)
htop                           # nicer interactive top (if installed)
ps aux | grep fmriprep         # find a specific process
kill <PID>                     # ask process to quit
kill -9 <PID>                  # force kill (last resort)

GPU

nvidia-smi                     # GPU usage, memory, processes
watch -n 2 nvidia-smi          # refresh every 2 s

7 Archiving and compression

Create a gzipped tarball, and extract one:

tar czvf archive.tar.gz folder/      # c=create, z=gzip, v=verbose, f=file
tar xzvf archive.tar.gz              # x=extract
tar tzvf archive.tar.gz              # t=list contents

For single files: gzip file / gunzip file.gz.

8 Editors — minimum survival

You’ll edit .bashrc, ~/.ssh/config, job scripts. Two options on every server:

nano (easy)

nano ~/.bashrc

Bottom of the screen shows the shortcuts. Ctrl-O save, Ctrl-X exit, Ctrl-W search.

vim (powerful)

vim ~/.bashrc

Modal: starts in normal mode. Press i to enter insert mode, type, then Esc back to normal. Save and quit with :wq, quit without saving :q!. That’s enough to survive — invest more time later if you stick with it.

9 Where to go next

Graph View

Backlinks