1.\" Copyright (c) 1998 2.\" The Regents of the University of California. All rights reserved. 3.\" 4.\" Redistribution and use in source and binary forms, with or without 5.\" modification, are permitted provided that the following conditions 6.\" are met: 7.\" 1. Redistributions of source code must retain the above copyright 8.\" notice, this list of conditions and the following disclaimer. 9.\" 2. Redistributions in binary form must reproduce the above copyright 10.\" notice, this list of conditions and the following disclaimer in the 11.\" documentation and/or other materials provided with the distribution. 12.\" 3. Neither the name of the University nor the names of its contributors 13.\" may be used to endorse or promote products derived from this software 14.\" without specific prior written permission. 15.\" 16.\" THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 17.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19.\" ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 20.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26.\" SUCH DAMAGE. 27.\" 28.\" $FreeBSD$ 29.\" 30.Dd December 22, 1998 31.Dt BUF 9 32.Os 33.Sh NAME 34.Nm buf 35.Nd "kernel buffer I/O scheme used in FreeBSD VM system" 36.Sh DESCRIPTION 37The kernel implements a KVM abstraction of the buffer cache which allows it 38to map potentially disparate vm_page's into contiguous KVM for use by 39(mainly file system) devices and device I/O. 40This abstraction supports 41block sizes from DEV_BSIZE (usually 512) to upwards of several pages or more. 42It also supports a relatively primitive byte-granular valid range and dirty 43range currently hardcoded for use by NFS. 44The code implementing the 45VM Buffer abstraction is mostly concentrated in 46.Pa /usr/src/sys/kern/vfs_bio.c . 47.Pp 48One of the most important things to remember when dealing with buffer pointers 49(struct buf) is that the underlying pages are mapped directly from the buffer 50cache. 51No data copying occurs in the scheme proper, though some file systems 52such as UFS do have to copy a little when dealing with file fragments. 53The second most important thing to remember is that due to the underlying page 54mapping, the b_data base pointer in a buf is always *page* aligned, not 55*block* aligned. 56When you have a VM buffer representing some b_offset and 57b_size, the actual start of the buffer is (b_data + (b_offset & PAGE_MASK)) 58and not just b_data. 59Finally, the VM system's core buffer cache supports 60valid and dirty bits (m->valid, m->dirty) for pages in DEV_BSIZE chunks. 61Thus 62a platform with a hardware page size of 4096 bytes has 8 valid and 8 dirty 63bits. 64These bits are generally set and cleared in groups based on the device 65block size of the device backing the page. 66Complete page's worth are often 67referred to using the VM_PAGE_BITS_ALL bitmask (i.e., 0xFF if the hardware page 68size is 4096). 69.Pp 70VM buffers also keep track of a byte-granular dirty range and valid range. 71This feature is normally only used by the NFS subsystem. 72I am not sure why it 73is used at all, actually, since we have DEV_BSIZE valid/dirty granularity 74within the VM buffer. 75If a buffer dirty operation creates a 'hole', 76the dirty range will extend to cover the hole. 77If a buffer validation 78operation creates a 'hole' the byte-granular valid range is left alone and 79will not take into account the new extension. 80Thus the whole byte-granular 81abstraction is considered a bad hack and it would be nice if we could get rid 82of it completely. 83.Pp 84A VM buffer is capable of mapping the underlying VM cache pages into KVM in 85order to allow the kernel to directly manipulate the data associated with 86the (vnode,b_offset,b_size). 87The kernel typically unmaps VM buffers the moment 88they are no longer needed but often keeps the 'struct buf' structure 89instantiated and even bp->b_pages array instantiated despite having unmapped 90them from KVM. 91If a page making up a VM buffer is about to undergo I/O, the 92system typically unmaps it from KVM and replaces the page in the b_pages[] 93array with a place-marker called bogus_page. 94The place-marker forces any kernel 95subsystems referencing the associated struct buf to re-lookup the associated 96page. 97I believe the place-marker hack is used to allow sophisticated devices 98such as file system devices to remap underlying pages in order to deal with, 99for example, re-mapping a file fragment into a file block. 100.Pp 101VM buffers are used to track I/O operations within the kernel. 102Unfortunately, 103the I/O implementation is also somewhat of a hack because the kernel wants 104to clear the dirty bit on the underlying pages the moment it queues the I/O 105to the VFS device, not when the physical I/O is actually initiated. 106This 107can create confusion within file system devices that use delayed-writes because 108you wind up with pages marked clean that are actually still dirty. 109If not 110treated carefully, these pages could be thrown away! 111Indeed, a number of 112serious bugs related to this hack were not fixed until the 2.2.8/3.0 release. 113The kernel uses an instantiated VM buffer (i.e., struct buf) to place-mark pages 114in this special state. 115The buffer is typically flagged B_DELWRI. 116When a 117device no longer needs a buffer it typically flags it as B_RELBUF. 118Due to 119the underlying pages being marked clean, the B_DELWRI|B_RELBUF combination must 120be interpreted to mean that the buffer is still actually dirty and must be 121written to its backing store before it can actually be released. 122In the case 123where B_DELWRI is not set, the underlying dirty pages are still properly 124marked as dirty and the buffer can be completely freed without losing that 125clean/dirty state information. 126(XXX do we have to check other flags in 127regards to this situation ???) 128.Pp 129The kernel reserves a portion of its KVM space to hold VM Buffer's data 130maps. 131Even though this is virtual space (since the buffers are mapped 132from the buffer cache), we cannot make it arbitrarily large because 133instantiated VM Buffers (struct buf's) prevent their underlying pages in the 134buffer cache from being freed. 135This can complicate the life of the paging 136system. 137.\" .Sh SEE ALSO 138.\" .Xr <fillmein> 9 139.Sh HISTORY 140The 141.Nm 142manual page was originally written by 143.An Matthew Dillon 144and first appeared in 145.Fx 3.1 , 146December 1998. 147