memgrep is a grep for /proc/pid/mem. It's licensed under the MIT license.

The binaries are also bound by the Hyperscan license due to static linking. Well, would be if the static linking worked. For the time being you need to sudo apt install libhyperscan5.

Why won't regular grep do? You can try it, but it will just immediately fail with an I/O error. This is because you can only read certain parts of the file and the parts you can read are listed in the file /proc/pid/maps.

So, basically, this tool combines parsing the maps file and grepping those regions for given process ids, or for all processes, while doing it in parallel with the CPUs you have.

Note that the value kernel parameter kernel.yama.ptrace_scope can affect memgrep abilities, even among your own processes. So for best results you should choose between setting that value to 0, or using sudo to run the binary as root.

% memgrep --help

memgrep 1.2.1
Erkki Seppälä <[email protected]>
Process address space grepping tool

USAGE:
    memgrep [FLAGS] [OPTIONS] [--] [regex]

FLAGS:
    -a, --all             Grep all processes
    -c, --count           Show only the number of non-zero matches
        --include-self    Include also this process in the results (implied by --pids)
    -l, --list            Show list the processes, not the matches
    -o, --show-content    Show the contents of the match (useful when using wildcards in regex)
    -h, --help            Prints help information
    -V, --version         Prints version information

OPTIONS:
    -p, --pid <pid>...         Process id to grep

You need to provide either -p pid or -a; you must provide exactly one pcre regex. For case-insensitive matching you can prefix your regex with (?i) (using the one of the supported option modifiers).

Note that due to used command line parser it is currently impossible to provide non-utf8 parameters, as there is no function to provide patterns from file or as e.g. hex data. This means your searches will try match the UTF8 byte representation of your patterns. The searches themselves are implemented without regard for UTF8; as the tool is scanning memory directly I chose it makes no sense to support any particular encoding.

Why not?

While this application might have few "real" use cases, it can be fun for discovering if some processes contain e.g. the string "hello world" with

memgrep -a -r 'hello world'

Similarly you can use it to discover how many processes are aware of your password.. Preferably avoid trying that in a multi-user system (due to password being visible in ps) and avoid putting that command to your command history; with e.g. zsh you can achieve this with setopt histignorespace and then prefixing the command with a space.

Perhaps you can also use it for recovering deleted regions from your text editor; chances are the previously removed content is still recoverable from the memory. The tool will output the byte ranges you can use with e.g. gdb:

% gdb -p 1438910 -batch-silent -ex 'dump memory contents.txt 0x7f262c6a5c6d 0x7f262c6a5c72'

Note that the value kernel parameter kernel.yama.ptrace_scope can affect your ability to run that command.

Fun stuff to try: sudo memgrep -a -o '.{20}backdoor.{20}'.

sudo apt-get install libhyperscan-dev

cargo build. You can also download a binary for Linux/x86_64 from the GitHub releases page. With the scanning work implemented in the HyperScan library, the speed of operation is similar between release and debug builds.

..or you can just directly install it with cargo:

cargo install --git https://github.com/eras/memgrep