presto.4

来自「<B>Digital的Unix操作系统VAX 4.2源码</B>」· 4 代码 · 共 491 行 · 第 1/2 页

.TH presto 4
.SH Name
presto \- the Prestoserve pseudodevice driver
.SH Syntax
.nf
.ft B
#include <sys/presto.h>
#include <sys/prestoioctl.h>

pseudo-device	presto
.fi
.ft R
.SH Description
.NXR "presto pseudodevice driver"
The Prestoserve pseudodevice driver 
.PN presto 
caches synchronous writes in nonvolatile memory.  Prestoserve causes synchronous 
writes to be performed at memory speeds, rather than at disk speeds.
Synchronous writes that result in Prestoserve cache hits do not perform the
earlier physical disk writes, because only the last write is actually 
performed by Prestoserve.  Therefore, 50% to 65% of all the physical disk
write operations are avoided, because every sequential NFS write to 
a file also causes the inode and the indirect block to be synchronously written.
.PP
The
.PN presto 
driver is layered on other disk drivers, and
it intercepts the other drivers' I/O requests by replacing the entry points 
of the original driver in the bdevsw and cdevsw tables. The 
.PN presto
driver caches the intercepted synchronous write requests in 
the Prestoserve cache's nonvolatile memory.
When Prestoserve needs to perform actual I/O, it calls the original 
driver's entry points to perform the I/O.  A modified form of
Least Recently Used (LRU) replacement determines when the 
Prestoserve cache data needs to be written to the intended disks.
.PP
An accelerated disk device (one that has the 
.PN presto
pseudodevice driver layered on top of its driver),
uses the same major and minor devices that it used before it was accelerated.
.PP
The Prestoserve nonvolatile memory must be found at boot time in order 
for Prestoserve to perform its write-caching function.
In addition, Prestoserve must pass diagnostic tests, and
there must be sufficient backup battery power to guarantee a reasonable
amount of cache data stability (measured in days or weeks) in the event 
of a power or hardware failure.
See the appropriate appendix in the \fIGuide to Prestoserve\fP
for information on specific processors.
.PP
If the Prestoserve nonvolatile memory is not found or if there is not enough 
backup battery power, then the disks are not accelerated; however
they can be opened and used as usual.  In this case, the
.PN presto
driver simply passes all I/O requests directly through to the appropriate device.
.SS Operation
When Prestoserve is in the PRUP state, it caches all synchronous write 
requests for enabled file systems and block devices to the
.PN presto
driver in nonvolatile memory and writes the Prestoserve cache data 
asynchronously to the intended disks.  
.PP
When Prestoserve is in the
PRDOWN state, there is no data in the Prestoserve cache, no data 
is put into the Prestoserve cache, and all disk 
requests are passed directly to the real disk driver.
.PP
When Prestoserve is in the PRERROR
state, the data in the Prestoserve cache can not be written to the
intended disks because of a disk, system, or hardware error.  
.PP
When the system is shut down normally by using the
.PN reboot
system call from the 
.PN shutdown ,
.PN halt ,
or
.PN reboot 
command, the Prestoserve cache data is written to
the intended disks, and Prestoserve enters the PRDOWN state.
This allows you to fix any system or disk error or to upgrade or change
your system without losing the data in the Prestoserve cache or
corrupting your disks.
.PP
If your system was shut down without following normal shutdown
procedures, and you reboot the system, any data in the Prestoserve
cache is written to the intended disks, if possible.  If the data is 
successfully written to the intended disks (and if the nonvolatile 
memory and backup battery passed the diagnostic tests), Prestoserve 
enters the PRDOWN state.  If an error occurs, Prestoserve enters the 
PRERROR state.
.NT
Data can exist in the Prestoserve cache after you reboot the 
system only in the event of a previous power failure, disk device error,
or kernel crash that resulted from a software or hardware problem.
.NE
.PP
If an error from a disk device occurs
or if the backup battery power is insufficient, Prestoserve
writes the cache data to the intended disks, if possible, and enters the 
PRERROR state.  When Prestoserve is in the PRERROR
state,  new data that is written to a block not found in the Prestoserve
cache is passed directly to the real disk driver.
If new data is written to a block that is found in the cache,
Prestoserve replaces the existing block and attempts to 
write the block to the real disk driver to determine 
if the error condition on that block still exists.
If the write is successful and if all the  Prestoserve cache 
data can be written to the intended disks, Prestoserve leaves the
PRERROR state.
.PP
The Prestoserve cache never discards data without being explicitly 
told to do so by using a PRRESET 
ioctl command.  This can be done by using the 
.PN "presto -R"
command.  This command should only be used when there is a fatal disk error
and when the data is not important.  
.SS ioctl Commands
The
.PN presto
pseudodevice driver does not intercept ioctl
commands;  they go directly to the actual disk driver.  The following ioctl 
commands can be performed on the Prestoserve control device
.PN /dev/pr0 .
Some ioctl 
commands affect all of Prestoserve operation,
while others only affect a particular accelerated file system.
.TP 
PRGETSTATUS
The argument to ioctl
is a pointer to a
.B "presto_status" 
structure, which contains battery status information, Prestoserve
state, current and maximum nonvolatile memory sizes, and various Prestoserve
statistics.
.TP
PRSETSTATE
The argument to ioctl is a pointer to an
.BR int .
The
.B int
can be either PRUP
to enable Prestoserve or PRDOWN
to disable Prestoserve.  When a system reboots, Prestoserve is in the
PRDOWN state and must be explicitly enabled by an ioctl.
You enable Prestoserve by using the 
.PN "presto \-u" 
command, which can be placed in the 
.PN /etc/rc.local
file.   When Prestoserve goes from the PRDOWN
state to the PRUP
state, the Prestoserve I/O statistics are reset.  When Prestoserve
goes from the PRUP state to the PRDOWN state, all the Prestoserve
buffers are written to the intended disks, and the buffers are invalidated.
.TP
PRSETMEMSZ
The argument to ioctl is a pointer to an
.BR int .
This
.B int
is the size in bytes of the Prestoserve nonvolatile memory to be used.
This size cannot be larger than the maximum size reported in the
.B presto_status
structure.
.TP
PRRESET
The argument to ioctl is ignored.  Like the PRSETSTATE
ioctl, PRRESET
sets the Prestoserve state to PRDOWN,
but it also reinitializes all of nonvolatile Prestoserve memory.  If 
Prestoserve was in the PRERROR
state and some Prestoserve buffers could not be written
to the intended disks because of disk I/O errors, the data in the buffers is lost.
This is the only method you can use to force Prestoserve 
to discard data that cannot be written to disk, and it can be accomplished by
using the 
.PN "presto -R"
command.
.TP
PRFLUSH
The argument to ioctl is ignored.
All the data in the Prestoserve buffers is written to the intended 
disks, but the buffers are 
not invalidated. This command can be used by a daemon that flushes the 
cache periodically to minimize the risk to data in the event of a
catastrophic failure.  The cache data can be flushed to the intended
disks by using the 
.PN "presto -F"
command.
.TP
PRGETUPRTAB
The argument to ioctl is a pointer to a
.BR "struct uprtab" .
On input, the field
.B upt_bmajordev
specifies the block device major number of the device whose
.B "struct uprtab"
should be returned.  The field
.B upt_bmajordev
is set to 
.B NODEV ,
which is defined in the header file 
.PN <sys/param.h> ,
if the requested device does not exist or if it is not 
accelerated.  The
.B "struct uprtab"
contains a 
.B upt_enabled 
field that is a bit vector indexed by a minor device number and that 
indicates whether the minor device number has Prestoserve
caching enabled.
.TP
PRNEXTUPRTAB
The argument to ioctl is a pointer to a
.BR "struct uprtab" .
This ioctl returns the
.B "struct uprtab"
for the accelerated device with the
smallest block device major number that is greater than the
.B bmajordev
field of the
.B "struct uprtab"
argument.  This allows each accelerated device's
.B "struct uprtab"
to be retrieved sequentially by specifying the previous device's block device 
major number. To get the first accelerated device's
.BR "struct uprtab" ,
set the
.B upt_bmajordev
field to
.B NODEV .  
Use the same
.B "struct uprtab"
that was returned on the previous call for the next call.
When the
.B upt_bmajordev
field of the argument
.B "struct uprtab"
is greater than or equal to the last accelerated
device's major block device number, the
.B "struct uprtab"
that is returned has the
.B upt_bmajordev
field set to
.BR NODEV .