man/man9/buf.9

.\" Copyright (c) 1998
.\"	The Regents of the University of California.  All rights reserved.
.\"
.\" Redistribution and use in source and binary forms, with or without
.\" modification, are permitted provided that the following conditions
.\" are met:
.\" 1. Redistributions of source code must retain the above copyright
.\"    notice, this list of conditions and the following disclaimer.
.\" 2. Redistributions in binary form must reproduce the above copyright
.\"    notice, this list of conditions and the following disclaimer in the
.\"    documentation and/or other materials provided with the distribution.
.\" 3. All advertising materials mentioning features or use of this software
.\"    must display the following acknowledgement:
.\"	This product includes software developed by the University of
.\"	California, Berkeley and its contributors.
.\" 4. Neither the name of the University nor the names of its contributors
.\"    may be used to endorse or promote products derived from this software
.\"    without specific prior written permission.
.\"
.\" THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
.\" ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
.\" SUCH DAMAGE.
.\"
.\" $FreeBSD: src/share/man/man9/buf.9,v 1.5.2.5 2001/12/17 11:30:18 ru Exp $
.\" $DragonFly: src/share/man/man9/buf.9,v 1.2 2003/06/17 04:37:01 dillon Exp $
.\"
.Dd December 22, 1998
.Dt BUF 9
.Os
.Sh NAME
.Nm buf
.Nd "kernel buffer I/O scheme used in FreeBSD VM system"
.Sh DESCRIPTION
The kernel implements a KVM abstraction of the buffer cache which allows it
to map potentially disparate vm_page's into contiguous KVM for use by
(mainly filesystem) devices and device I/O.  This abstraction supports
block sizes from DEV_BSIZE (usually 512) to upwards of several pages or more.
It also supports a relatively primitive byte-granular valid range and dirty
range currently hardcoded for use by NFS.  The code implementing the
VM Buffer abstraction is mostly concentrated in
.Pa /usr/src/sys/kern/vfs_bio.c .
.Pp
One of the most important things to remember when dealing with buffer pointers
(struct buf) is that the underlying pages are mapped directly from the buffer
cache.  No data copying occurs in the scheme proper, though some filesystems
such as UFS do have to copy a little when dealing with file fragments.  The
second most important thing to remember is that due to the underlying page
mapping, the b_data base pointer in a buf is always *page* aligned, not
*block* aligned.  When you have a VM buffer representing some b_offset and
b_size, the actual start of the buffer is (b_data + (b_offset & PAGE_MASK))
and not just b_data.  Finally, the VM system's core buffer cache supports
valid and dirty bits (m->valid, m->dirty) for pages in DEV_BSIZE chunks.  Thus
a platform with a hardware page size of 4096 bytes has 8 valid and 8 dirty
bits.  These bits are generally set and cleared in groups based on the device
block size of the device backing the page.  Complete page's worth are often
referred to using the VM_PAGE_BITS_ALL bitmask (i.e. 0xFF if the hardware page
size is 4096).
.Pp
VM buffers also keep track of a byte-granular dirty range and valid range.
This feature is normally only used by the NFS subsystem.  I'm not sure why it
is used at all, actually, since we have DEV_BSIZE valid/dirty granularity
within the VM buffer.  If a buffer dirty operation creates a 'hole',
the dirty range will extend to cover the hole.  If a buffer validation
operation creates a 'hole' the byte-granular valid range is left alone and
will not take into account the new extension.  Thus the whole byte-granular
abstraction is considered a bad hack and it would be nice if we could get rid
of it completely.
.Pp
A VM buffer is capable of mapping the underlying VM cache pages into KVM in
order to allow the kernel to directly manipulate the data associated with
the (vnode,b_offset,b_size).  The kernel typically unmaps VM buffers the moment
they are no longer needed but often keeps the 'struct buf' structure
instantiated and even bp->b_pages array instantiated despite having unmapped
them from KVM.  If a page making up a VM buffer is about to undergo I/O, the
system typically unmaps it from KVM and replaces the page in the b_pages[]
array with a placemarker called bogus_page.  The placemarker forces any kernel
subsystems referencing the associated struct buf to re-lookup the associated
page.  I believe the placemarker hack is used to allow sophisticated devices
such as filesystem devices to remap underlying pages in order to deal with,
for example, remapping a file fragment into a file block.
.Pp
VM buffers are used to track I/O operations within the kernel.  Unfortunately,
the I/O implementation is also somewhat of a hack because the kernel wants
to clear the dirty bit on the underlying pages the moment it queues the I/O
to the VFS device, not when the physical I/O is actually initiated.  This
can create confusion within filesystem devices that use delayed-writes because
you wind up with pages marked clean that are actually still dirty.  If not
treated carefully, these pages could be thrown away!  Indeed, a number of
serious bugs related to this hack were not fixed until the 2.2.8/3.0 release.
The kernel uses an instantiated VM buffer (i.e. struct buf) to placemark pages
in this special state.  The buffer is typically flagged B_DELWRI.  When a
device no longer needs a buffer it typically flags it as B_RELBUF.  Due to
the underlying pages being marked clean, the B_DELWRI|B_RELBUF combination must
be interpreted to mean that the buffer is still actually dirty and must be
written to its backing store before it can actually be released.  In the case
where B_DELWRI is not set, the underlying dirty pages are still properly
marked as dirty and the buffer can be completely freed without losing that
clean/dirty state information.  ( XXX do we have to check other flags in
regards to this situation ??? ).
.Pp
The kernel reserves a portion of its KVM space to hold VM Buffer's data
maps.  Even though this is virtual space (since the buffers are mapped
from the buffer cache), we cannot make it arbitrarily large because
instantiated VM Buffers (struct buf's) prevent their underlying pages in the
buffer cache from being freed.  This can complicate the life of the paging
system.
.Pp
.\" .Sh SEE ALSO
.\" .Xr <fillmein> 9
.Sh HISTORY
The
.Nm
manual page was originally written by
.An Matthew Dillon
and first appeared in
.Fx 3.1 ,
December 1998.