RUMPHIJACK(3) Library Functions Manual RUMPHIJACK(3)
rumphijack -- System call hijack library
used by ld.so(1)
The ld.so(1) runtime linker can be instructed to load rumphijack between
the main object and other libraries. This enables rumphijack to capture
and redirect system call requests to a rump kernel instead of the host
The behaviour of hijacked applications is affected by the following
If present, this variable specifies which system calls should be
hijacked. The string is parsed as a comma-separated list of
``name=value'' tuples. The possible lefthandside names are:
``path'' Pathname-based system calls are hijacked if the path
the system call is directed to resides under value.
In case of an absolute pathname argument, a literal
prefix comparison is made. In case of a relative
pathname, the current working direct is examined.
This also implies that neither ``..'' nor symbolic
links will cause the namespace to be switched.
``blanket'' A colon-separated list of rump path prefixes. This
acts almost like ``path'' with the difference that
the prefix does not get removed when passing the path
to the rump kernel. For example, if ``path'' is
/rump, accessing /rump/dev/bpf will cause /dev/bpf to
be accessed in the rump kernel. In contrast, if
``blanket'' contains /dev/bpf, accessing /dev/bpf
will cause an access to /dev/bpf in the rump kernel.
In case the current working directory is changed to a
blanketed directory, the current working directory
will still be reported with the rump prefix, if
available. Note, though, that some shells cache the
directory and may report something else. In case no
rump path prefix has been configured, the raw rump
directory is reported.
It is recommended to supply blanketed pathnames as
specific as possible, i.e. use /dev/bpf instead of
/dev unless necessary to do otherwise. Also, note
that the blanket prefix does not follow directory
borders. In other words, setting the blanket for
/dev/bpf means it is set for all pathnames with the
given prefix, not just ones in /dev.
``socket'' The specifier value contains a colon-separated list
of which protocol families should be hijacked. The
special value ``all'' can be specified as the first
element. It indicates that all protocol families
should be hijacked. Some can then be disabled by
prepending ``no'' to the name of the protocol family.
For example, ``inet:inet6'' specifies that only
PF_INET and PF_INET6 sockets should be hijacked,
while ``all:noinet'' specifies that all protocol
families except PF_INET should be hijacked.
``vfs'' The specifier value contains a colon-separated list
of which vfs-related system calls should be hijacked.
These differ from the pathname-based file system
syscalls in that there is no pathname to make the
selection based on. Current possible values are
``nfssvc'', ``getvfsstat'', and ``fhcalls''. They
indicate hijacking nfssvc(), getvfsstat(), and all
file handle calls, respectively. The file handle
calls include fhopen(), fhstat(), and fhstatvfs1().
It is also possible to use ``all'' and ``no'' in the
same fashion as with the socket hijack specifier.
``sysctl'' Direct the __sysctl() backend of the sysctl(3)
facility to the rump kernel. Acceptable values are
``yes'' and ``no'', meaning to call the rump or the
host kernel, respectively.
``fdoff'' Adjust the library's fd offset to the specified
value. All rump kernel descriptors have the offset
added to them before they are returned to the
application. This should be changed only if the
application defines a low non-default FD_SETSIZE for
select() or if it opens a very large number of file
descriptors. The default value is 128.
If the environment variable is unset, the default value
"path=/rump,socket=all:nolocal" is used. The rationale for this
is to have networked X clients work out-of-the-box: X clients use
local sockets to communicate with the server, so local sockets
must be used as a host service.
An empty string as a value means no calls are hijacked.
Change how rumpclient(3) attempts to reconnect to the server in
case the connection is lost. Acceptable values are:
``inftime'' retry indefinitely
``once'' retry once, when that connection fails, give up
``die'' call exit(3) if connection failure is detected
n Attempt reconnect for n seconds. The value 0 means
reconnection is not attempted. The value n must be a
See rumpclient(3) for more discussion.
Use an alternate TCP/IP stack for firefox with a persistent server
$ setenv RUMP_SERVER unix:///tmp/tcpip
$ setenv LD_PRELOAD /usr/lib/librumphijack.so
$ setenv RUMPHIJACK_RETRYCONNECT inftime
ld.so(1), rump_server(1), rump(3), rumpclient(3), rump_sp(7)
NetBSD 6.1.5 March 14, 2011 NetBSD 6.1.5