Linux – Second field in /etc/shadow

As far as I know, all unix variants have an /etc/passwd file with the traditional layout, for the sake of applications that read the file directly. Each line in the file contains colon-separated records which correspond to struct passwd fields:

1. user name (login)
2. encrypted password (if present)
3. user id (number, in decimal)
4. principal group id (number, in decimal)
5. Gecos field. This field is not used by system programs except for display purposes. It is normally a comma-separated list of fields, the first three being full name, office number and telephone number.
6. home directory
7. login shell

One thing that varies between systems is how much liberty you can take with the syntax. For example, GNU libc (i.e. Linux) ignores lines that begin with #: they are comments. GNU libc also ignores whitespace at the beginning of a line, so they can be indented. An invalid line might cause programs to stop processing the file or to skip to the next line.

Most modern systems no longer store an encrypted password in the second field. The content of that field is not a reliable indication of whether the user has a password set (and even if you found that out, this is not a reliable indication of whether the user can log in, because there are many other authentication methods such as SSH keys, one-time passwords, biometrics, smartcards, …).

When passwords aren't in /etc/passwd, where they are is system-dependent. The Rosetta Stone for Unix mentions many unix variants.

• Solaris uses /etc/shadow, and this has been copied by others including Linux. Linux and Solaris shadow files have the same format; I don't know if the other systems that have a file called /etc/shadow use the same format.
• BSD systems have /etc/master.passwd, and additionally have database files for faster access, updated by pwd_mkdb.

Remember that /etc/passwd hasn't been guaranteed to contain the full list of users for a couple of decades: users can come from other databases such as NIS (YP) or LDAP. As a system administrator, avoid edit the /etc/passwd file directly; use vipw instead, if your system provides it (and if it doesn't, consult your manuals to see what method is recommended to modify the user database).

What I wrote above goes for groups, too. The fields in /etc/group are struct group members: group name, password (largely unused), numerical group id, and a comma-separated list of user names (the users who have this group as a secondary group). Linux has a /etc/gshadow file, but this is rarely used, as group authentication is not widely practiced.

NP in the password field of /etc/shadow indicates that that the account cannot be logged into with a password but can be logged into with other authentication methods, such as su down from root or cron jobs. NP means that password authentication will always fail, but other login methods may succeed. You can set an account in this state with passwd -N. This differs from *LK* (reported as LK by passwd -s), which disables all logins to the account regardless of the authentication method.

Confusingly, when passwd -s sees NP in /etc/shadow, it reports NL, whereas NP in the passwd -l report indicates that the account is open to all winds: users will be authenticated without even getting a password prompt (this is indicated by an empty password field in /etc/shadow).

UP is a documented code in the passwd -s output on Solaris 11 (not on Solaris 11 Express). It means that “this account has not yet been activated by the administrator and cannot be used.” If I understand the documentation correctly, its effect is similar to NP; the intent is that the system administrator will run passwd later to set a password (i.e. it's the first stage in the process where the admin creates the account for a future user, then later has the user type a password when they first come on-site). The documentation doesn't indicate whether passwd -s reports UP when it finds that in /etc/shadow; while this is plausible, the confusion around NP invites caution.

Usually, anything in the password field of /etc/shadow (or other password database) that isn't an empty string is treated as a hashed password, and leads to a denied authentication if it doesn't match any of the valid hashed password formats. This is the case with normal password authentication on OpenSolaris, I can't speak for other versions but would be somewhat surprised if this wasn't the case.

Note that if there are several entries for the same user, I think only the first one is taken into account. (At least that's the case under Linux, and I have no reason to believe that Solaris would be different in this respect.)