freebsd-skq/share/man/man9/alq.9
Lawrence Stewart 7d11e744c1 - Rework the underlying ALQ storage to be a circular buffer, which amongst other
things allows variable length messages to be easily supported.

- Extend KPI with alq_writen() and alq_getn() to support variable length
  messages, which is enabled at ALQ creation time depending on the
  arguments passed to alq_open(). Also add variants of alq_open() and
  alq_post() that accept a flags argument. The KPI is still fully
  backwards compatible and shouldn't require any change in ALQ consumers
  unless they wish to utilise the new features.

- Introduce the ALQ_NOACTIVATE and ALQ_ORDERED flags to allow ALQ consumers
  to have more control over IO scheduling and resource acquisition
  respectively.

- Strengthen invariants checking.

- Document ALQ changes in ALQ(9) man page.

Sponsored by:	FreeBSD Foundation
Reviewed by:	gnn, jeff, rpaulo, rwatson
MFC after:	1 month
2010-04-26 13:48:22 +00:00

442 lines
10 KiB
Groff

.\"
.\" Copyright (c) 2003 Hiten Pandya <hmp@FreeBSD.org>
.\" Copyright (c) 2009-2010 The FreeBSD Foundation
.\" All rights reserved.
.\"
.\" Portions of this software were developed at the Centre for Advanced
.\" Internet Architectures, Swinburne University of Technology, Melbourne,
.\" Australia by Lawrence Stewart under sponsorship from the FreeBSD
.\" Foundation.
.\"
.\" 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,
.\" without modification, immediately at the beginning of the file.
.\" 2. The name of the author may not be used to endorse or promote products
.\" derived from this software without specific prior written permission.
.\"
.\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR 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 AUTHOR 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.
.\"
.\" $FreeBSD$
.\"
.Dd April 26, 2010
.Dt ALQ 9
.Os
.Sh NAME
.Nm alq ,
.Nm alq_open_flags ,
.Nm alq_open ,
.Nm alq_writen ,
.Nm alq_write ,
.Nm alq_flush ,
.Nm alq_close ,
.Nm alq_getn ,
.Nm alq_get ,
.Nm alq_post_flags ,
.Nm alq_post
.Nd Asynchronous Logging Queues
.Sh SYNOPSIS
.In sys/alq.h
.Ft int
.Fo alq_open_flags
.Fa "struct alq **app"
.Fa "const char *file"
.Fa "struct ucred *cred"
.Fa "int cmode"
.Fa "int size"
.Fa "int flags"
.Fc
.Ft int
.Fo alq_open
.Fa "struct alq **app"
.Fa "const char *file"
.Fa "struct ucred *cred"
.Fa "int cmode"
.Fa "int size"
.Fa "int count"
.Fc
.Ft int
.Fn alq_writen "struct alq *alq" "void *data" "int len" "int flags"
.Ft int
.Fn alq_write "struct alq *alq" "void *data" "int flags"
.Ft void
.Fn alq_flush "struct alq *alq"
.Ft void
.Fn alq_close "struct alq *alq"
.Ft struct ale *
.Fn alq_getn "struct alq *alq" "int len" "int flags"
.Ft struct ale *
.Fn alq_get "struct alq *alq" "int flags"
.Ft void
.Fn alq_post_flags "struct alq *alq" "struct ale *ale" "int flags"
.Ft void
.Fn alq_post "struct alq *alq" "struct ale *ale"
.Sh DESCRIPTION
The
.Nm
facility provides an asynchronous fixed or variable length recording
mechanism, known as Asynchronous Logging Queues.
It can record to any
.Xr vnode 9 ,
thus providing the ability to journal logs to character
devices as well as regular files.
All functions accept a
.Vt "struct alq"
argument, which is an opaque type that maintains state information
for an Asynchronous Logging Queue.
The logging facility runs in a separate kernel thread, which services
all log entry requests.
.Pp
An
.Dq asynchronous log entry
is defined as
.Vt "struct ale" ,
which has the following members:
.Bd -literal -offset indent
struct ale {
intptr_t ae_bytesused; /* # bytes written to ALE. */
char *ae_data; /* Write ptr. */
int ae_pad; /* Unused, compat. */
};
.Ed
.Pp
An
.Nm
can be created in either fixed or variable length mode.
A variable length
.Nm
accommodates writes of varying length using
.Fn alq_writen
and
.Fn alq_getn .
A fixed length
.Nm
accommodates a fixed number of writes using
.Fn alq_write
and
.Fn alq_get ,
each of fixed size (set at queue creation time).
Fixed length mode is deprecated in favour of variable length mode.
.Sh FUNCTIONS
The
.Fn alq_open_flags
function creates a new variable length asynchronous logging queue.
The
.Fa file
argument is the name of the file to open for logging.
If the file does not yet exist,
.Fn alq_open
will attempt to create it.
The
.Fa cmode
argument will be passed to
.Fn vn_open
as the requested creation mode, to be used if the file will be created by
.Fn alq_open .
Consumers of this API may wish to pass
.Dv ALQ_DEFAULT_CMODE ,
a default creation mode suitable for most applications.
The
.Fa cred
argument specifies the credentials to use when opening and performing I/O on the file.
The
.Fa size
argument sets the size (in bytes) of the underlying queue.
The ALQ_ORDERED flag may be passed in via
.Fa flags
to indicate that the ordering of writer threads waiting for a busy
.Nm
to free up resources should be preserved.
.Pp
The deprecated
.Fn alq_open
function is implemented as a wrapper around
.Fn alq_open_flags
to provide backwards compatibility to consumers that have not been updated to
utilise the newer
.Fn alq_open_flags
function.
It passes all arguments through to
.Fn alq_open_flags
untouched except for
.Fa size
and
.Fa count ,
and sets
.Fa flags
to 0.
To create a variable length mode
.Nm ,
the
.Fa size
argument should be set to the size (in bytes) of the underlying queue and the
.Fa count
argument should be set to 0.
To create a fixed length mode
.Nm ,
the
.Fa size
argument should be set to the size (in bytes) of each write and the
.Fa count
argument should be set to the number of
.Fa size
byte chunks to reserve capacity for.
.Pp
The
.Fn alq_writen
function writes
.Fa len
bytes from
.Fa data
to the designated variable length mode queue
.Fa alq .
If
.Fn alq_writen
could not write the entry immediately and
.Dv ALQ_WAITOK
is set in
.Fa flags ,
the function will be allowed to
.Xr msleep_spin 9
with the
.Dq Li alqwnord
or
.Dq Li alqwnres
wait message.
A write will automatically schedule the queue
.Fa alq
to be flushed to disk.
This behaviour can be controlled by passing ALQ_NOACTIVATE via
.Fa flags
to indicate that the write should not schedule
.Fa alq
to be flushed to disk.
.Pp
The deprecated
.Fn alq_write
function is implemented as a wrapper around
.Fn alq_writen
to provide backwards compatibility to consumers that have not been updated to
utilise variable length mode queues.
The function will write
.Fa size
bytes of data (where
.Fa size
was specified at queue creation time) from the
.Fa data
buffer to the
.Fa alq .
Note that it is an error to call
.Fn alq_write
on a variable length mode queue.
.Pp
The
.Fn alq_flush
function is used for flushing
.Fa alq
to the log medium that was passed to
.Fn alq_open .
If
.Fa alq
has data to flush and is not already in the process of being flushed, the
function will block doing IO.
Otherwise, the function will return immediately.
.Pp
The
.Fn alq_close
function will close the asynchronous logging queue
.Fa alq
and flush all pending write requests to the log medium.
It will free all resources that were previously allocated.
.Pp
The
.Fn alq_getn
function returns an asynchronous log entry from
.Fa alq ,
initialised to point at a buffer capable of receiving
.Fa len
bytes of data.
This function leaves
.Fa alq
in a locked state, until a subsequent
.Fn alq_post
or
.Fn alq_post_flags
call is made.
If
.Fn alq_getn
could not obtain
.Fa len
bytes of buffer immediately and
.Dv ALQ_WAITOK
is set in
.Fa flags ,
the function will be allowed to
.Xr msleep_spin 9
with the
.Dq Li alqgnord
or
.Dq Li alqgnres
wait message.
The caller can choose to write less than
.Fa len
bytes of data to the returned asynchronous log entry by setting the entry's
ae_bytesused field to the number of bytes actually written.
This must be done prior to calling
.Fn alq_post .
.Pp
The deprecated
.Fn alq_get
function is implemented as a wrapper around
.Fn alq_getn
to provide backwards compatibility to consumers that have not been updated to
utilise variable length mode queues.
The asynchronous log entry returned will be initialised to point at a buffer
capable of receiving
.Fa size
bytes of data (where
.Fa size
was specified at queue creation time).
Note that it is an error to call
.Fn alq_get
on a variable length mode queue.
.Pp
The
.Fn alq_post_flags
function schedules the asynchronous log entry
.Fa ale
(obtained from
.Fn alq_getn
or
.Fn alq_get )
for writing to
.Fa alq .
The ALQ_NOACTIVATE flag may be passed in via
.Fa flags
to indicate that the queue should not be immediately scheduled to be flushed to
disk.
This function leaves
.Fa alq
in an unlocked state.
.Pp
The
.Fn alq_post
function is implemented as a wrapper around
.Fn alq_post_flags
to provide backwards compatibility to consumers that have not been updated to
utilise the newer
.Fn alq_post_flags
function.
It simply passes all arguments through to
.Fn alq_post_flags
untouched, and sets
.Fa flags
to 0.
.Sh IMPLEMENTATION NOTES
The
.Fn alq_writen
and
.Fn alq_write
functions both perform a
.Xr bcopy 3
from the supplied
.Fa data
buffer into the underlying
.Nm
buffer.
Performance critical code paths may wish to consider using
.Fn alq_getn
(variable length queues) or
.Fn alq_get
(fixed length queues) to avoid the extra memory copy. Note that a queue
remains locked between calls to
.Fn alq_getn
or
.Fn alq_get
and
.Fn alq_post
or
.Fn alq_post_flags ,
so this method of writing to a queue is unsuitable for situations where the
time between calls may be substantial.
.Sh LOCKING
Each asynchronous logging queue is protected by a spin mutex.
.Pp
Functions
.Fn alq_flush
and
.Fn alq_open
may attempt to acquire an internal sleep mutex, and should
consequently not be used in contexts where sleeping is
not allowed.
.Sh RETURN VALUES
The
.Fn alq_open
function returns one of the error codes listed in
.Xr open 2 ,
if it fails to open
.Fa file ,
or else it returns 0.
.Pp
The
.Fn alq_writen
and
.Fn alq_write
functions return
.Er EWOULDBLOCK
if
.Dv ALQ_NOWAIT
was set in
.Fa flags
and either the queue is full or the system is shutting down.
.Pp
The
.Fn alq_getn
and
.Fn alq_get
functions return
.Dv NULL
if
.Dv ALQ_NOWAIT
was set in
.Fa flags
and either the queue is full or the system is shutting down.
.Pp
NOTE: invalid arguments to non-void functions will result in
undefined behaviour.
.Sh SEE ALSO
.Xr kproc 9 ,
.Xr ktr 9 ,
.Xr msleep_spin 9 ,
.Xr syslog 3 ,
.Xr vnode 9
.Sh HISTORY
The
Asynchronous Logging Queues (ALQ) facility first appeared in
.Fx 5.0 .
.Sh AUTHORS
.An -nosplit
The
.Nm
facility was written by
.An Jeffrey Roberson Aq jeff@FreeBSD.org
and extended by
.An Lawrence Stewart Aq lstewart@freebsd.org .
.Pp
This manual page was written by
.An Hiten Pandya Aq hmp@FreeBSD.org
and revised by
.An Lawrence Stewart Aq lstewart@freebsd.org .