.TH SEGMENT 3 
.SH NAME
segment \- long lived memory segments
.SH SYNOPSIS
.nf
.B bind '#g' /mnt/segment

.BI #g/ segn
.BI #g/ segn /ctl
.BI #g/ segn /data
.BI #g/ segn /free
 ...
.fi
.SH DESCRIPTION
.PP
The
.I segment
device provides a 2-level file system representing
long-lived sharable segments that processes may
.IR segattach (2).
The name of the directory is the
.I class
argument to
.IR segattach .
.PP
New segments are created under the top level
using
.B create
(see
.IR open (2)).
The
.B DMDIR
bit must be set in the permissions.
.IR Remove (2)'ing
the directory makes the segment no longer
available for
.IR segattach .
However, the segment will continue to exist until all
processes using it either exit or
.I segdetach
it.
.PP
Within each segment directory are three files,
.BR data ,
.BR ctl ,
and
.BR free .
Reading and writing
.B data
affects the contents of the segment.
Reading and writing 
.B ctl
retrieves and sets the segment's properties.
The
.B free
file is used to manage deallocation of buffers for zero-copy.
.PP
There are only three control messages, which sets the segment's
virtual address and length in bytes:
.EX
	addr \fIaddress length\fP
	umsg \fIaddress length\fP
	kmsg \fIaddress length\fP
.EE
The first one creates a shared segment. The second one creates a shared segment
for use with zero-copy. The third one creates a shared segment for use with
zero copy, identified as the one for use by the kernel. See below for an explanation.
.I Address
is automatically rounded down to a page boundary and
.I length
is rounded up to end the segment at a page boundary.
If the address given is zero, the system picks a unique address
(system-wide) for the segment.
The segment will reside at the same virtual address in
all processes sharing it.
When the segment
is attached using
.IR segattach,
the address and length arguments are ignored in the call;
they are defined only by the
control message.
Once the address and length are set, they cannot be reset.
.PP
Reading the control file
returns a message of the same format with the segment's actual
start address and length.
.PP
Opening
.B data
or reading
.B ctl
before setting the virtual address yields the error
``segment not yet allocated''.
.PP
The permissions check when
.IR segattach ing
is equivalent to the one performed when opening
.B data
with mode ORDWR.
.PP
Segments for zero copy (either
.I umsg
or
.I kmsg
segments) make use of the
.B free
file to control data buffers for zero-copy. Each process exchanging zero-copy
buffers must attach at least to a
.I kmsg
segment, also known as a KZIO (Kernel zero-copy I/O) segment. It might also attach
to one or more
.I umsg
segments, also known as ZIO (zero-copy I/O) segments. ZIO system calls assume
that addresses contained in ZIO (or KZIO) segments may be exchanged for I/O without
requiring data copies. This is not so for addresses referring to any other segment.
.PP
Allocation of buffers within a ZIO segment is handled by the process that created
the segment. Other processes, and the kernel, are expected only to use such buffers,
but they do not allocate or deallocate buffers directly. Allocation of buffers within a KZIO
segment is handled only the kernel. User processes are expected to use such buffers, but
not to allocate or deallocate them.
.PP
When a buffer is no longer in use, the
.B free
file can be used to write its virtual address. As a result, the kernel will notify
any process reading such file of the address for the, now unused, data buffer. If the
address refers to a KZIO segment, the kernel handles deallocation on its own.
.SH EXAMPLE
.PP
Create a one megabyte segment at address 0x10000000:
.EX
	% bind '#g' /mnt/segment
	% mkdir /mnt/segment/example
	% echo 'va 0x10000000 0x100000' > /mnt/segment/example/ctl
.EE
.PP
Put the string ``hi mom'' at the start of the segment:
.EX
	% echo -n hi mom > /mnt/segment/example/data
.EE
.PP
Attach the segment to a process:
.EX
{
	ulong va;

	va = segattach(0, "example", 0, 0);
}
.EE
.SH "SEE ALSO
.IR segattach (2)
.SH SOURCE
.B /sys/src/9/port/devsegment.c