.\"	$NetBSD: altq.9,v 1.8 2002/05/28 11:41:45 wiz Exp $
.\"	$OpenBSD: altq.9,v 1.4 2001/07/12 12:41:42 itojun Exp $
.\"
.\" Copyright (C) 2004 Max Laier. All rights reserved.
.\" Copyright (C) 2001
.\" Sony Computer Science Laboratories Inc.  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.
.\"
.\" THIS SOFTWARE IS PROVIDED BY SONY CSL 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 SONY CSL 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 August 25, 2004
.Dt ALTQ 9
.Os
.\"
.Sh NAME
.Nm ALTQ
.Nd kernel interfaces for manipulating output queues on network interfaces
.Sh SYNOPSIS
.In sys/types.h
.In sys/socket.h
.In net/if.h
.In net/if_var.h
.\"
.Ss Enqueue macros
.Fn IFQ_ENQUEUE "struct ifaltq *ifq" "struct mbuf *m" "int error"
.Fn IFQ_HANDOFF "struct ifnet *ifp" "struct mbuf *m" "int error"
.Fo IFQ_HANDOFF_ADJ
.Fa "struct ifnet *ifp" "struct mbuf *m" "int adjust" "int error"
.Fc
.\"
.Ss Dequeue macros
.Fn IFQ_DEQUEUE "struct ifaltq *ifq" "struct mbuf *m"
.Fn IFQ_POLL_NOLOCK "struct ifaltq *ifq" "struct mbuf *m"
.Fn IFQ_PURGE "struct ifaltq *ifq"
.Fn IFQ_IS_EMPTY "struct ifaltq *ifq"
.\"
.Ss Driver managed dequeue macros
.Fn IFQ_DRV_DEQUEUE "struct ifaltq *ifq" "struct mbuf *m"
.Fn IFQ_DRV_PREPEND "struct ifaltq *ifq" "struct mbuf *m"
.Fn IFQ_DRV_PURGE "struct ifaltq *ifq"
.Fn IFQ_DRV_IS_EMPTY "struct ifaltq *ifq"
.\"
.Ss General setup macros
.Fn IFQ_SET_MAXLEN "struct ifaltq *ifq" "int len"
.Fn IFQ_INC_LEN "struct ifaltq *ifq"
.Fn IFQ_DEC_LEN "struct ifaltq *ifq"
.Fn IFQ_INC_DROPS "struct ifaltq *ifq"
.Fn IFQ_SET_READY "struct ifaltq *ifq"
.Sh DESCRIPTION
The
.Nm
system is a framework to manage queuing disciplines on network
interfaces.
.Nm
introduces new macros to manipulate output queues.
The output queue macros are used to abstract queue operations and not to
touch the internal fields of the output queue structure.
The macros are independent from the
.Nm
implementation, and compatible with the traditional
.Vt ifqueue
macros for ease of transition.
.Pp
.Fn IFQ_ENQUEUE ,
.Fn IFQ_HANDOFF
and
.Fn IFQ_HANDOFF_ADJ
enqueue a packet
.Fa m
to the queue
.Fa ifq .
The underlying queuing discipline may discard the packet.
The
.Fa error
argument is set to 0 on success, or
.Er ENOBUFS
if the packet is discarded.
The packet pointed to by
.Fa m
will be freed by the device driver on success, or by the queuing discipline on
failure, so the caller should not touch
.Fa m
after enqueuing.
.Fn IFQ_HANDOFF
and
.Fn IFQ_HANDOFF_ADJ
combine the enqueue operation with statistic generation and call
.Fn if_start
upon successful enqueue to initiate the actual send.
.Pp
.Fn IFQ_DEQUEUE
dequeues a packet from the queue.
The dequeued packet is returned in
.Fa m ,
or
.Fa m
is set to
.Dv NULL
if no packet is dequeued.
The caller must always check
.Fa m
since a non-empty queue could return
.Dv NULL
under rate-limiting.
.Pp
.Fn IFQ_POLL_NOLOCK
returns the next packet without removing it from the queue.
The caller must hold the queue mutex when calling
.Fn IFQ_POLL_NOLOCK
in order to guarantee that a subsequent call to
.Fn IFQ_DEQUEUE_NOLOCK
dequeues the same packet.
.Pp
.Fn IFQ_*_NOLOCK
variants (if available) always assume that the caller holds the queue mutex.
They can be grabbed with
.Fn IFQ_LOCK
and released with
.Fn IFQ_UNLOCK .
.Pp
.Fn IFQ_PURGE
discards all the packets in the queue.
The purge operation is needed since a non-work conserving queue cannot be
emptied by a dequeue loop.
.Pp
.Fn IFQ_IS_EMPTY
can be used to check if the queue is empty.
Note that
.Fn IFQ_DEQUEUE
could still return
.Dv NULL
if the queuing discipline is non-work conserving.
.Pp
.Fn IFQ_DRV_DEQUEUE
moves up to
.Fa ifq->ifq_drv_maxlen
packets from the queue to the
.Dq "driver managed"
queue and returns the first one via
.Fa m .
As for
.Fn IFQ_DEQUEUE ,
.Fa m
can be
.Dv NULL
even for a non-empty queue.
Subsequent calls to
.Fn IFQ_DRV_DEQUEUE
pass the packets from the
.Dq "driver managed"
queue without obtaining the queue mutex.
It is the responsibility of the caller to protect against concurrent access.
Enabling
.Nm
for a given queue sets
.Va ifq_drv_maxlen
to 0 as the
.Dq "bulk dequeue"
performed by
.Fn IFQ_DRV_DEQUEUE
for higher values of
.Va ifq_drv_maxlen
is adverse to
.Nm ALTQ Ns 's
internal timing.
Note that a driver must not mix
.Fn IFQ_DRV_*
macros with the default dequeue macros as the default macros do not look at the
.Dq "driver managed"
queue which might lead to an mbuf leak.
.Pp
.Fn IFQ_DRV_PREPEND
prepends
.Fa m
to the
.Dq "driver managed"
queue from where it will be obtained with the next call to
.Fn IFQ_DRV_DEQUEUE .
.Pp
.Fn IFQ_DRV_PURGE
flushes all packets in the
.Dq "driver managed"
queue and calls to
.Fn IFQ_PURGE
afterwards.
.Pp
.Fn IFQ_DRV_IS_EMPTY
checks for packets in the
.Dq "driver managed"
part of the queue.
If it is empty, it forwards to
.Fn IFQ_IS_EMPTY .
.Pp
.Fn IFQ_SET_MAXLEN
sets the queue length limit to the default FIFO queue.
The
.Va ifq_drv_maxlen
member of the
.Vt ifaltq
structure controls the length limit of the
.Dq "driver managed"
queue.
.Pp
.Fn IFQ_INC_LEN
and
.Fn IFQ_DEC_LEN
increment or decrement the current queue length in packets.
This is mostly for internal purposes.
.Pp
.Fn IFQ_INC_DROPS
increments the drop counter and is identical to
.Fn IF_DROP .
It is defined for naming consistency only.
.Pp
.Fn IFQ_SET_READY
sets a flag to indicate that a driver was converted to use the new macros.
.Nm
can be enabled only on interfaces with this flag.
.Sh COMPATIBILITY
.Ss Vt ifaltq Ss structure
In order to keep compatibility with the existing code, the new
output queue structure
.Vt ifaltq
has the same fields.
The traditional
.Fn IF_*
macros and the code directly referencing the fields within
.Va if_snd
still work with
.Vt ifaltq .
.Bd -literal
            ##old-style##                           ##new-style##
                                       |
 struct ifqueue {                      | struct ifaltq {
    struct mbuf *ifq_head;             |    struct mbuf *ifq_head;
    struct mbuf *ifq_tail;             |    struct mbuf *ifq_tail;
    int          ifq_len;              |    int          ifq_len;
    int          ifq_maxlen;           |    int          ifq_maxlen;
    int          ifq_drops;            |    int          ifq_drops;
 };                                    |    /* driver queue fields */
                                       |    ......
                                       |    /* altq related fields */
                                       |    ......
                                       | };
                                       |
.Ed
The new structure replaces
.Vt "struct ifqueue"
in
.Vt "struct ifnet" .
.Bd -literal
            ##old-style##                           ##new-style##
                                       |
 struct ifnet {                        | struct ifnet {
     ....                              |     ....
                                       |
     struct ifqueue if_snd;            |     struct ifaltq if_snd;
                                       |
     ....                              |     ....
 };                                    | };
                                       |
.Ed
The (simplified) new
.Fn IFQ_*
macros look like:
.Bd -literal
	#define IFQ_DEQUEUE(ifq, m)			\e
		if (ALTQ_IS_ENABLED((ifq))		\e
			ALTQ_DEQUEUE((ifq), (m));	\e
		else					\e
			IF_DEQUEUE((ifq), (m));
.Ed
.Ss Enqueue operation
The semantics of the enqueue operation is changed.
In the new style,
enqueue and packet drop are combined since they cannot be easily
separated in many queuing disciplines.
The new enqueue operation corresponds to the following macro that is
written with the old macros.
.Bd -literal
#define	IFQ_ENQUEUE(ifq, m, error)                      \e
do {                                                    \e
        if (IF_QFULL((ifq))) {                          \e
                m_freem((m));                           \e
                (error) = ENOBUFS;                      \e
                IF_DROP(ifq);                           \e
        } else {                                        \e
                IF_ENQUEUE((ifq), (m));                 \e
                (error) = 0;                            \e
        }                                               \e
} while (0)
.Ed
.Pp
.Fn IFQ_ENQUEUE
does the following:
.Pp
.Bl -hyphen -compact
.It
queue a packet,
.It
drop (and free) a packet if the enqueue operation fails.
.El
.Pp
If the enqueue operation fails,
.Fa error
is set to
.Er ENOBUFS .
The
.Fa m
mbuf is freed by the queuing discipline.
The caller should not touch mbuf after calling
.Fn IFQ_ENQUEUE
so that the caller may need to copy
.Va m_pkthdr.len
or
.Va m_flags
field beforehand for statistics.
.Fn IFQ_HANDOFF
and
.Fn IFQ_HANDOFF_ADJ
can be used if only default interface statistics and an immediate call to
.Fn if_start
are desired.
The caller should not use
.Fn senderr
since mbuf was already freed.
.Pp
The new style
.Fn if_output
looks as follows:
.Bd -literal
            ##old-style##                           ##new-style##
                                       |
 int                                   | int
 ether_output(ifp, m0, dst, rt0)       | ether_output(ifp, m0, dst, rt0)
 {                                     | {
     ......                            |     ......
                                       |
                                       |     mflags = m->m_flags;
                                       |     len = m->m_pkthdr.len;
     s = splimp();                     |     s = splimp();
     if (IF_QFULL(&ifp->if_snd)) {     |     IFQ_ENQUEUE(&ifp->if_snd, m,
                                       |                 error);
         IF_DROP(&ifp->if_snd);        |     if (error != 0) {
         splx(s);                      |         splx(s);
         senderr(ENOBUFS);             |         return (error);
     }                                 |     }
     IF_ENQUEUE(&ifp->if_snd, m);      |
     ifp->if_obytes +=                 |     ifp->if_obytes += len;
                    m->m_pkthdr.len;   |
     if (m->m_flags & M_MCAST)         |     if (mflags & M_MCAST)
         ifp->if_omcasts++;            |         ifp->if_omcasts++;
                                       |
     if ((ifp->if_flags & IFF_OACTIVE) |     if ((ifp->if_flags & IFF_OACTIVE)
         == 0)                         |         == 0)
         (*ifp->if_start)(ifp);        |         (*ifp->if_start)(ifp);
     splx(s);                          |     splx(s);
     return (error);                   |     return (error);
                                       |
 bad:                                  | bad:
     if (m)                            |     if (m)
         m_freem(m);                   |         m_freem(m);
     return (error);                   |     return (error);
 }                                     | }
                                       |
.Ed
.Sh HOW TO CONVERT THE EXISTING DRIVERS
First, make sure the corresponding
.Fn if_output
is already converted to the new style.
.Pp
Look for
.Va if_snd
in the driver.
Probably, you need to make changes to the lines that include
.Va if_snd .
.Ss Empty check operation
If the code checks
.Va ifq_head
to see whether the queue is empty or not, use
.Fn IFQ_IS_EMPTY .
.Bd -literal
            ##old-style##                           ##new-style##
                                       |
 if (ifp->if_snd.ifq_head != NULL)     | if (!IFQ_IS_EMPTY(&ifp->if_snd))
                                       |
.Ed
.Fn IFQ_IS_EMPTY
only checks if there is any packet stored in the queue.
Note that even when
.Fn IFQ_IS_EMPTY
is
.Dv FALSE ,
.Fn IFQ_DEQUEUE
could still return
.Dv NULL
if the queue is under rate-limiting.
.Ss Dequeue operation
Replace
.Fn IF_DEQUEUE
by
.Fn IFQ_DEQUEUE .
Always check whether the dequeued mbuf is
.Dv NULL
or not.
Note that even when
.Fn IFQ_IS_EMPTY
is
.Dv FALSE ,
.Fn IFQ_DEQUEUE
could return
.Dv NULL
due to rate-limiting.
.Bd -literal
            ##old-style##                           ##new-style##
                                       |
 IF_DEQUEUE(&ifp->if_snd, m);          | IFQ_DEQUEUE(&ifp->if_snd, m);
                                       | if (m == NULL)
                                       |     return;
                                       |
.Ed
A driver is supposed to call
.Fn if_start
from transmission complete interrupts in order to trigger the next dequeue.
.Ss Poll-and-dequeue operation
If the code polls the packet at the head of the queue and actually uses
the packet before dequeuing it, use
.Fn IFQ_POLL_NOLOCK
and
.Fn IFQ_DEQUEUE_NOLOCK .
.Bd -literal
            ##old-style##                           ##new-style##
                                       |
                                       | IFQ_LOCK(&ifp->if_snd);
 m = ifp->if_snd.ifq_head;             | IFQ_POLL_NOLOCK(&ifp->if_snd, m);
 if (m != NULL) {                      | if (m != NULL) {
                                       |
     /* use m to get resources */      |     /* use m to get resources */
     if (something goes wrong)         |     if (something goes wrong)
                                       |         IFQ_UNLOCK(&ifp->if_snd);
         return;                       |         return;
                                       |
     IF_DEQUEUE(&ifp->if_snd, m);      |     IFQ_DEQUEUE_NOLOCK(&ifp->if_snd, m);
                                       |     IFQ_UNLOCK(&ifp->if_snd);
                                       |
     /* kick the hardware */           |     /* kick the hardware */
 }                                     | }
                                       |
.Ed
It is guaranteed that
.Fn IFQ_DEQUEUE_NOLOCK
under the same lock as a previous
.Fn IFQ_POLL_NOLOCK
returns the same packet.
Note that they need to be guarded by
.Fn IFQ_LOCK .
.Ss Eliminating Fn IF_PREPEND
If the code uses
.Fn IF_PREPEND ,
you have to eliminate it unless you can use a
.Dq "driver managed"
queue which allows the use of
.Fn IFQ_DRV_PREPEND
as a substitute.
A common usage of
.Fn IF_PREPEND
is to cancel the previous dequeue operation.
You have to convert the logic into poll-and-dequeue.
.Bd -literal
            ##old-style##                           ##new-style##
                                       |
                                       | IFQ_LOCK(&ifp->if_snd);
 IF_DEQUEUE(&ifp->if_snd, m);          | IFQ_POLL_NOLOCK(&ifp->if_snd, m);
 if (m != NULL) {                      | if (m != NULL) {
                                       |
     if (something_goes_wrong) {       |     if (something_goes_wrong) {
         IF_PREPEND(&ifp->if_snd, m);  |         IFQ_UNLOCK(&ifp->if_snd);
         return;                       |         return;
     }                                 |     }
                                       |
                                       |     /* at this point, the driver
                                       |      * is committed to send this
                                       |      * packet.
                                       |      */
                                       |     IFQ_DEQUEUE_NOLOCK(&ifp->if_snd, m);
                                       |     IFQ_UNLOCK(&ifp->if_snd);
                                       |
     /* kick the hardware */           |     /* kick the hardware */
 }                                     | }
                                       |
.Ed
.Ss Purge operation
Use
.Fn IFQ_PURGE
to empty the queue.
Note that a non-work conserving queue cannot be emptied by a dequeue loop.
.Bd -literal
            ##old-style##                           ##new-style##
                                       |
 while (ifp->if_snd.ifq_head != NULL) {|  IFQ_PURGE(&ifp->if_snd);
     IF_DEQUEUE(&ifp->if_snd, m);      |
     m_freem(m);                       |
 }                                     |
                                       |
.Ed
.Ss Conversion using a driver managed queue
Convert
.Fn IF_*
macros to their equivalent
.Fn IFQ_DRV_*
and employ
.Fn IFQ_DRV_IS_EMPTY
where appropriate.
.Bd -literal
            ##old-style##                           ##new-style##
                                       |
 if (ifp->if_snd.ifq_head != NULL)     | if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
                                       |
.Ed
Make sure that calls to
.Fn IFQ_DRV_DEQUEUE ,
.Fn IFQ_DRV_PREPEND
and
.Fn IFQ_DRV_PURGE
are protected with a mutex of some kind.
.Ss Attach routine
Use
.Fn IFQ_SET_MAXLEN
to set
.Va ifq_maxlen
to
.Fa len .
Initialize
.Va ifq_drv_maxlen
with a sensible value if you plan to use the
.Fn IFQ_DRV_*
macros.
Add
.Fn IFQ_SET_READY
to show this driver is converted to the new style.
(This is used to distinguish new-style drivers.)
.Bd -literal
            ##old-style##                           ##new-style##
                                       |
 ifp->if_snd.ifq_maxlen = qsize;       | IFQ_SET_MAXLEN(&ifp->if_snd, qsize);
                                       | ifp->if_snd.ifq_drv_maxlen = qsize;
                                       | IFQ_SET_READY(&ifp->if_snd);
 if_attach(ifp);                       | if_attach(ifp);
                                       |
.Ed
.Ss Other issues
The new macros for statistics:
.Bd -literal
            ##old-style##                           ##new-style##
                                       |
 IF_DROP(&ifp->if_snd);                | IFQ_INC_DROPS(&ifp->if_snd);
                                       |
 ifp->if_snd.ifq_len++;                | IFQ_INC_LEN(&ifp->if_snd);
                                       |
 ifp->if_snd.ifq_len--;                | IFQ_DEC_LEN(&ifp->if_snd);
                                       |
.Ed
.Sh QUEUING DISCIPLINES
Queuing disciplines need to maintain
.Fa ifq_len
(used by
.Fn IFQ_IS_EMPTY ) .
Queuing disciplines also need to guarantee that the same mbuf is returned if
.Fn IFQ_DEQUEUE
is called immediately after
.Fn IFQ_POLL .
.Sh SEE ALSO
.Xr pf 4 ,
.Xr pf.conf 5 ,
.Xr pfctl 8
.Sh HISTORY
The
.Nm
system first appeared in March 1997.