credentials API
===============

The credentials API provides an abstracted way of gathering username and
password credentials from the user (even though credentials in the wider
world can take many forms, in this document the word "credential" always
refers to a username and password pair).

This document describes two interfaces: the C API that the credential
subsystem provides to the rest of git, and the protocol that git uses to
communicate with system-specific "credential helpers". If you are
writing git code that wants to look up or prompt for credentials, see
the section "C API" below. If you want to write your own helper, see
the section on "Credential Helpers" below.

Typical setup
-------------

------------
+-----------------------+
| git code (C)          |--- to server requiring --->
|                       |        authentication
|.......................|
| C credential API      |--- prompt ---> User
+-----------------------+
        ^      |
        | pipe |
        |      v
+-----------------------+
| git credential helper |
+-----------------------+
------------

The git code (typically a remote-helper) will call the C API to obtain
credential data like a login/password pair (credential_fill). The
API will itself call a remote helper (e.g. "git credential-cache" or
"git credential-store") that may retrieve credential data from a
store. If the credential helper cannot find the information, the C API
will prompt the user. Then, the caller of the API takes care of
contacting the server, and does the actual authentication.

C API
-----

The credential C API is meant to be called by git code which needs to
acquire or store a credential. It is centered around an object
representing a single credential and provides three basic operations:
fill (acquire credentials by calling helpers and/or prompting the user),
approve (mark a credential as successfully used so that it can be stored
for later use), and reject (mark a credential as unsuccessful so that it
can be erased from any persistent storage).

Data Structures
~~~~~~~~~~~~~~~

`struct credential`::

        This struct represents a single username/password combination
        along with any associated context. All string fields should be
        heap-allocated (or NULL if they are not known or not applicable).
        The meaning of the individual context fields is the same as
        their counterparts in the helper protocol; see the section below
        for a description of each field.
+
The `helpers` member of the struct is a `string_list` of helpers.  Each
string specifies an external helper which will be run, in order, to
either acquire or store credentials. See the section on credential
helpers below.
+
This struct should always be initialized with `CREDENTIAL_INIT` or
`credential_init`.


Functions
~~~~~~~~~

`credential_init`::

        Initialize a credential structure, setting all fields to empty.

`credential_clear`::

        Free any resources associated with the credential structure,
        returning it to a pristine initialized state.

`credential_fill`::

        Instruct the credential subsystem to fill the username and
        password fields of the passed credential struct by first
        consulting helpers, then asking the user. After this function
        returns, the username and password fields of the credential are
        guaranteed to be non-NULL. If an error occurs, the function will
        die().

`credential_reject`::

        Inform the credential subsystem that the provided credentials
        have been rejected. This will cause the credential subsystem to
        notify any helpers of the rejection (which allows them, for
        example, to purge the invalid credentials from storage).  It
        will also free() the username and password fields of the
        credential and set them to NULL (readying the credential for
        another call to `credential_fill`). Any errors from helpers are
        ignored.

`credential_approve`::

        Inform the credential subsystem that the provided credentials
        were successfully used for authentication.  This will cause the
        credential subsystem to notify any helpers of the approval, so
        that they may store the result to be used again.  Any errors
        from helpers are ignored.

`credential_from_url`::

        Parse a URL into broken-down credential fields.

Example
~~~~~~~

The example below shows how the functions of the credential API could be
used to login to a fictitious "foo" service on a remote host:

-----------------------------------------------------------------------
int foo_login(struct foo_connection *f)
{
        int status;
        /*
         * Create a credential with some context; we don't yet know the
         * username or password.
         */

        struct credential c = CREDENTIAL_INIT;
        c.protocol = xstrdup("foo");
        c.host = xstrdup(f->hostname);

        /*
         * Fill in the username and password fields by contacting
         * helpers and/or asking the user. The function will die if it
         * fails.
         */
        credential_fill(&c);

        /*
         * Otherwise, we have a username and password. Try to use it.
         */
        status = send_foo_login(f, c.username, c.password);
        switch (status) {
        case FOO_OK:
                /* It worked. Store the credential for later use. */
                credential_accept(&c);
                break;
        case FOO_BAD_LOGIN:
                /* Erase the credential from storage so we don't try it
                 * again. */
                credential_reject(&c);
                break;
        default:
                /*
                 * Some other error occured. We don't know if the
                 * credential is good or bad, so report nothing to the
                 * credential subsystem.
                 */
        }

        /* Free any associated resources. */
        credential_clear(&c);

        return status;
}
-----------------------------------------------------------------------


Credential Helpers
------------------

Credential helpers are programs executed by git to fetch or save
credentials from and to long-term storage (where "long-term" is simply
longer than a single git process; e.g., credentials may be stored
in-memory for a few minutes, or indefinitely on disk).

Each helper is specified by a single string. The string is transformed
by git into a command to be executed using these rules:

  1. If the helper string begins with "!", it is considered a shell
     snippet, and everything after the "!" becomes the command.

  2. Otherwise, if the helper string begins with an absolute path, the
     verbatim helper string becomes the command.

  3. Otherwise, the string "git credential-" is prepended to the helper
     string, and the result becomes the command.

The resulting command then has an "operation" argument appended to it
(see below for details), and the result is executed by the shell.

Here are some example specifications:

----------------------------------------------------
# run "git credential-foo"
foo

# same as above, but pass an argument to the helper
foo --bar=baz

# the arguments are parsed by the shell, so use shell
# quoting if necessary
foo --bar="whitespace arg"

# you can also use an absolute path, which will not use the git wrapper
/path/to/my/helper --with-arguments

# or you can specify your own shell snippet
!f() { echo "password=`cat $HOME/.secret`"; }; f
----------------------------------------------------

Generally speaking, rule (3) above is the simplest for users to specify.
Authors of credential helpers should make an effort to assist their
users by naming their program "git-credential-$NAME", and putting it in
the $PATH or $GIT_EXEC_PATH during installation, which will allow a user
to enable it with `git config credential.helper $NAME`.

When a helper is executed, it will have one "operation" argument
appended to its command line, which is one of:

`get`::

        Return a matching credential, if any exists.

`store`::

        Store the credential, if applicable to the helper.

`erase`::

        Remove a matching credential, if any, from the helper's storage.

The details of the credential will be provided on the helper's stdin
stream. The credential is split into a set of named attributes.
Attributes are provided to the helper, one per line. Each attribute is
specified by a key-value pair, separated by an `=` (equals) sign,
followed by a newline. The key may contain any bytes except `=`,
newline, or NUL. The value may contain any bytes except newline or NUL.
In both cases, all bytes are treated as-is (i.e., there is no quoting,
and one cannot transmit a value with newline or NUL in it). The list of
attributes is terminated by a blank line or end-of-file.

Git will send the following attributes (but may not send all of
them for a given credential; for example, a `host` attribute makes no
sense when dealing with a non-network protocol):

`protocol`::

        The protocol over which the credential will be used (e.g.,
        `https`).

`host`::

        The remote hostname for a network credential.

`path`::

        The path with which the credential will be used. E.g., for
        accessing a remote https repository, this will be the
        repository's path on the server.

`username`::

        The credential's username, if we already have one (e.g., from a
        URL, from the user, or from a previously run helper).

`password`::

        The credential's password, if we are asking it to be stored.

For a `get` operation, the helper should produce a list of attributes
on stdout in the same format. A helper is free to produce a subset, or
even no values at all if it has nothing useful to provide. Any provided
attributes will overwrite those already known about by git.

For a `store` or `erase` operation, the helper's output is ignored.
If it fails to perform the requested operation, it may complain to
stderr to inform the user. If it does not support the requested
operation (e.g., a read-only store), it should silently ignore the
request.

If a helper receives any other operation, it should silently ignore the
request. This leaves room for future operations to be added (older
helpers will just ignore the new requests).