SYSTRACE(1) BSD Reference Manual SYSTRACE(1)
systrace - generate and enforce system call policies
systrace [-AaCeitUuV] [-c user:group] [-d policydir] [-E logfile] [-f file] [-g gui] [-p pid] command ...
The systrace utility monitors and controls an application's access to the system by enforcing access policies for system calls. The systrace utili- ty might be used to trace an untrusted application's access to the sys- tem. Alternatively, it might be used to protect the system from software bugs (such as buffer overflows) by constraining a daemon's access to the system. Its privilege elevation feature can be used to obviate the need to run large, untrusted programs as root when only one or two system calls require root privilege. The access policy can be generated interactively or obtained from a poli- cy file. Interactive policy generation will be performed by the "notification user agent", normally xsystrace(1), unless text mode is specified via -t. When running in "automatic enforcement" mode, operations not covered by the policy raise an alarm and allow a user to refine the currently con- figured policy. The options are as follows: -A Automatically generate a policy that allows every operation the application executes. The created policy functions as a base that can be refined. -a Enables automatic enforcement of configured policies. An opera- tion not covered by policy is denied and logged via syslog(3), or to stderr if the -e flag is specified. -C Run systrace in cradle mode; currently, when multiple processes are started with systrace protection, each systrace starts its own UI (user interface) process. Cradle mode allows a user to attach all systrace processes to one UI. This may be useful, for example, in scenarios where systrace is being heavily used. If a cradle server is not running, one is launched. -c user:group Specifies the user and group that the monitored application should be executed with, which may be either non-negative in- tegers or names. This is useful in conjunction with privilege elevation and requires root privilege. -d policydir Specifies an alternative location for the user's directory from which policies are loaded and to which changed policies are stored. -E logfile Logs all policy violations or specifically logged system calls to logfile. -e Specifies to log to stderr instead of syslog(3). -f file The policies specified in file are added to the policies that systrace knows about. The dirname in the policy may contain an "*" to match any possible pathname. The wildcard is removed from the policy database the first time that a filename matches. -g gui Specifies an alternative location for the notification user in- terface. -i Inherits the policy after a call to execve(2). If this option is not specified, a new program will get its own policy. -p pid Specifies the pid of a process that systrace should attach to. The full path name of the corresponding binary has to be speci- fied as command. -t Uses text mode to ask for interactive policy generation. -U Ignore user-configured policies and use only global system poli- cies. -u Do not perform aliasing on system call names. Aliasing is en- abled by default to group similar system calls into a single compound name. For example, system calls that read from the filesystem like lstat() and access() are translated to fsread(). -V Prints the version number of systrace.
The policy is specified via the following grammar: filter = expression "then" action errorcode logcode expression = symbol | "not" expression | "(" expression ")" | expression "and" expression | expression "or" expression symbol = string typeoff "match" cmdstring | string typeoff "eq" cmdstring | string typeoff "neq" cmdstring | string typeoff "sub" cmdstring | string typeoff "nsub" cmdstring | string typeoff "inpath" cmdstring | string typeoff "re" cmdstring | "true" typeoff = /* empty */ | "[" number "]" action = "permit" | "deny" | "ask" errorcode = /* empty */ | "[" string "]" logcode = /* empty */ | "log" The cmdstring is an arbitrary string enclosed with quotation marks. The errorcode is used to return an errno(2) value to the system call when us- ing a deny action. The values "inherit" and "detach" have special mean- ings when used with a permit rule for the execve system call. When using "inherit," the current policy is inherited for the new binary. With "de- tach," systrace detaches from a process after successfully completing the execve system call. The ask action specifies that the user should be prompted for a decision every time that the rule matches. The filter operations have the following meaning: match Evaluates to true if file name globbing according to fnmatch(3) succeeds. eq Evaluates to true if the system call argument matches cmdstring exactly. neq This is the logical negation of eq. sub Performs a substring match on the system call argument. nsub This is the logical negation of sub. inpath Evaluates to true if the system call argument is a subpath of cmdstring. re Evaluates to true if the system call arguments matches the specified regular expression. By appending the log statement to a rule, a matching system call and its arguments are logged. This is useful, for example, to log all invocations of the execve system call. Policy entries may contain an appended predicate. Predicates have the following format: ", if" {"user", "group"} {"=", "!=", "<", ">" } {number, string} A rule is added to the configured policy only if its predicate evaluates to true. The environment variables $HOME, $USER and $CWD are substituted in rules. Comments, begun by an unquoted '#' character and continuing to the end of the line, are ignored.
With systrace it is possible to remove setuid or setgid binaries, and use the privilege elevation feature instead. Single system calls can be exe- cuted with higher privileges if specified by the policy. For example, native-bind: sockaddr eq "inet-[0.0.0.0]:22" then permit as root allows an unprivileged application to bind to a reserved port. Privilege elevation requires that the systrace process is executed as root. The following statements can be appended after the permit in a policy to elevate the privileges for the matching system call: as user as user:group as :group The effective uid and gid are elevated only for the duration of the sys- tem call, and are restored to the old values afterwards (except for the seteuid or setegid system calls).
/dev/systrace systrace device /etc/systrace global systrace policies $HOME/.etc/systrace user specified policies, one per binary, with slashes in the full pathname replaced by the underscore character.
An excerpt from a sample ls(1) policy might look as follows: Policy: /bin/ls, Emulation: native [...] native-fsread: filename eq "$HOME" then permit native-fchdir: permit [...] native-fsread: filename eq "/tmp" then permit native-stat: permit native-fsread: filename match "$HOME/*" then permit native-fsread: filename eq "/etc/pwd.db" then permit [...] native-fsread: filename eq "/etc" then deny[eperm], if group != wheel
systrace(4)
The systrace utility was developed by Niels Provos.
Applications that use clone()-like system calls to share the complete ad- dress space between processes may be able to replace system call argu- ments after they have been evaluated by systrace and escape policy en- forcement. MirBSD #10-current November 28, 2003 3