intro – introduction to Plan 9 from User Space|
Plan 9 is a distributed computing environment built at Bell Labs
starting in the late 1980s. The system can be obtained from Bell
Labs at http://9p.io/plan9 and runs on PCs and a variety of other
platforms. Plan 9 became a convenient platform for experimenting
with new ideas, applications, and services.
Plan 9 from User Space provides many of the ideas, applications,
and services from Plan 9 on Unix-like systems. It runs on FreeBSD
(x86, x86-64), Linux (x86, x86-64, PowerPC and ARM), Mac OS X
(x86, x86-64, and PowerPC), NetBSD (x86 and PowerPC), OpenBSD
(x86 and PowerPC), Dragonfly BSD (x86-64), and SunOS (x86-64 and
Many of the familiar Unix commands, for example cat(1), ls(1),
and wc(1), are present, but in their Plan 9 forms: cat takes no
options, ls does not columnate its output when printing to a terminal,
and wc counts UTF characters. In some cases, the differences are
quite noticeable: grep(1) and sed(1) expect Plan 9 regular expressions
which are closest to what Unix calls extended regular expressions.
Because of these differences, it is not recommended to put $PLAN9/bin
before the usual system bin directories in your search path. Instead,
put it at the end of your path and use the 9(1) script when you
want to invoke the Plan 9 version of a traditional Unix command.
Occasionally the Plan 9 programs have been changed to adapt to
Unix. Mk(1) now allows mkfiles to choose their own shell, and
rc(1) has a ulimit builtin and manages $PATH.
Many of the graphical programs from Plan 9 are present, including
sam(1) and acme(1). An X11 window manager rio(1) mimics Plan 9’s
window system, with command windows implemented by the external
program 9term(1). Following the style of X Windows, these programs
run in new windows rather than the one in which they are invoked.
take a −W option to specify the size and placement of the new
window. The argument is one of widthxheight, widthxheight@xmin,xmax,
The plumber(4) helps to connect the various Plan 9 programs together,
and fittings like web(1) connect it to external programs such
as web browsers; one can click on a URL in acme and see the page
load in Firefox.
Plan 9 from User Space expects its own directory tree, conventionally
/usr/local/plan9. When programs need to access files in the tree,
they expect the $PLAN9 environment variable to contain the name
of the root of the tree. See install(1) for details about installation.
User-level file servers
This cannot be done directly on Unix. Instead the servers listen
for 9P connections on Unix domain sockets; clients connect to
these sockets and speak 9P directly using the 9pclient(3) library.
Intro(4) tells more of the story. The effect is not as clean as
on Plan 9, but it gets the job done and still provides a uniform
and easy-to-understand mechanism.
The 9p(1) client can be used in shell scripts or by hand to carry
out simple interactions with servers. Netfiles(1) is an experimental
client for acme.
In Plan 9, user-level file servers present file trees via the
Plan 9 file protocol, 9P. Processes can mount arbitrary file servers
and customize their own name spaces. These facilities are used
to connect programs. Clients interact with file servers by reading
and writing files.
Some programs rely on large databases that would be cumbersome
to include in every release. Scripts are provided that download
these databases separately. These databases can be downloaded
separately. See $PLAN9/dict/README and $PLAN9/sky/README.
The only way to write multithreaded programs is to use the thread(3)
library. Rfork(3) exists but is not as capable as on Plan 9. There
are many unfortunate by necessary preprocessor diversions to make
Plan 9 and Unix libraries coexist. See intro(3) for details.
The debuggers acid(1) and db(1) and the debugging library mach(3)
are works in progress. They are platform-independent, so that
x86 Linux core dumps can be inspected on PowerPC Mac OS X machines,
but they are also fairly incomplete. The x86 target is the most
mature; initial PowerPC support exists; and other targets are
debuggers can only inspect, not manipulate, target processes.
Support for operating system threads and for 64-bit architectures
needs to be rethought. On x86 Linux systems, acid and db can be
relied upon to produce reasonable stack traces (often in cases
when GNU gdb cannot) and dump data structures, but that it is
the extent to which they have
been developed and exercised.
The shell scripts 9c and 9l (see 9c(1)) provide a simple interface
to the underlying system compiler and linker, similar to the 2c
and 2l families on Plan 9. 9c compiles source files, and 9l links
object files into executables. When using Plan 9 libraries, 9l
infers the correct set of libraries from the object files, so
that no −l options are needed.
Of the more recent additions to Plan 9, factotum(4), secstore(1),
and secstored(1), vac(1), vacfs(4), and venti(8) are all ported.
A backup system providing a dump file system built atop Venti
is in progress; see vbackup(8).
The vast majority of the familiar Plan 9 programs have been ported,
including the Unicode-aware troff(1).
Porting to new systems
There are other smaller system dependencies, such as the terminal
handling code in 9term(1) and the implementation of getcallerpc(3),
but these are usually simple and are not on the critical path
for getting the system up and running.
Porting the tree to new operating systems or architectures should
be straightforward, as system-specific code has been kept to a
minimum. The largest pieces of system-specific code are <u.h>, which
must include the right system files and set up the right integer
type definitions, and libthread, which must implement spin locks,
operating system thread
creation, and context switching routines. Portable implementations
of these using <pthread.h> and <ucontext.h> already exist. If your
system supports them, you may not need to write any system specific
code at all.
The rest of this manual describes Plan 9 from User Space. Many
of the man pages have been brought from Plan 9, but they have
been updated, and others have been written from scratch.
The manual pages are in a Unix style tree, with names like $PLAN9/man/man1/cat.1
instead of Plan 9’s simpler $PLAN9/man/1/cat, so that the Unix
man(1) utility can handle it. Some systems, for example Debian
Linux, deduce the man page locations from the search path, so
that adding $PLAN9/bin to your path is sufficient to cause
$PLAN9/man to be consulted for manual pages using the system man.
On other systems, or to look at manual pages with the same name
as a system page, invoke the Plan 9 man directly, as in 9 man
The manual sections follow the Unix numbering conventions, not
the Plan 9 ones.
Section (1) describes general publicly accessible commands.
Section (3) describes C library functions.
Section (4) describes user-level file servers.
Section (7) describes file formats and protocols. (On Unix, section
(5) is technically for file formats but seems now to be used for
describing specific files.)
Section (8) describes commands used for system administration.
Section (9p) describes the Plan 9 file protocol 9P.
These pages describe parts of the system that are new or different
from Plan 9 from Bell Labs:|
9(1), 9c(1), 9p(1), 9term(1), acidtypes in acid(1), dial(1), git(1),
label(1), the MKSHELL variable in mk(1), namespace(1), netfiles(1),
page(1), psfonts(1), rio(1), web(1), wintext(1)|
intro(3), 9pclient(3), the unix network in dial(3), exits(3),
get9root(3), getns(3), notify(3), post9pservice(3), rfork(3),
searchpath(3), sendfd(3), udpread(3), venti(3), wait(3), wctl(3)
intro(4), 9pserve(4), import(4),
In Plan 9, a program’s exit status is an arbitrary text string,
while on Unix it is an integer. Section (1) of this manual describes
commands as though they exit with string statuses. In fact, exiting
with an empty status corresponds to exiting with status 0, and
exiting with any non-empty string corresponds to exiting with
status 1. See exits(3).