cap – capabilities for setting the user id of processes


bind # dir

dir/caphash dir/capuse


This device enables a trusted process to create a capability that another process may then use to change its user id. The intent is to allow server processes, for example telnetd (see ipserv(8)), to change their user id after having proved to a trusted process, such as factotum(4), that they are indeed executing on behalf of a user. A trusted process is one running with the user id of the host owner (see /dev/hostowner in cons(3)).

A capability is a null terminated string consisting of the concatenation of an old user name, an “@”, a new user name, an “@”, and a string of randomly generated characters called the key. The trusted process enables the kernel to authenticate capabilities passed to it by writing to caphash a secure hash of the capability. The hash is 20 bytes long and generated by the following call:

	hmac_sha1(old_at_new, strlen(old_at_new), key, strlen(key),
		hash, nil);

The kernel maintains a list of hashes, freeing them after the corresponding capability is used or after a minute has passed since the write to caphash.

The trusted process may then pass the capability to any process running as the old user. That process may then use the capability to change identity to the new user. A process uses a capability by writing it to capuse. The kernel computes the same hash using the supplied capability and searches its list of hashes for a match. If one is found, the kernel sets the process’s user id to that in the capability.






Errors generated by reading and writing caphash and capuse can be obtained using errstr(2). A read of caphash with a length of less than 20 or a write to capuse that doesn’t contain two @ characters generates the error “read or write too small”. A write to capuse that has no matching hash generates the error “invalid capability”.