madwifi/net80211/ieee80211_output.c

2415 lines
69 KiB
C
Raw Permalink Normal View History

/*-
* Copyright (c) 2001 Atsushi Onoe
* Copyright (c) 2002-2005 Sam Leffler, Errno Consulting
* 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. The name of the author may not be used to endorse or promote products
* derived from this software without specific prior written permission.
*
* Alternatively, this software may be distributed under the terms of the
* GNU General Public License ("GPL") version 2 as published by the Free
* Software Foundation.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``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 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.
*
* $Id$
*/
#ifndef EXPORT_SYMTAB
#define EXPORT_SYMTAB
#endif
/*
* IEEE 802.11 output handling.
*/
#if !defined(AUTOCONF_INCLUDED) && !defined(CONFIG_LOCALVERSION)
#include <linux/config.h>
#endif
#include <linux/version.h>
#include <linux/module.h>
#include <linux/skbuff.h>
#include <linux/netdevice.h>
#include <linux/if_vlan.h>
#include <linux/ip.h> /* XXX for TOS */
#include "if_llc.h"
#include "if_ethersubr.h"
#include "if_media.h"
#include <net80211/ieee80211_var.h>
#include <net80211/ieee80211_monitor.h>
#include <net80211/if_athproto.h>
#ifdef IEEE80211_DEBUG
/*
* Decide if an outbound management frame should be
* printed when debugging is enabled. This filters some
* of the less interesting frames that come frequently
* (e.g. beacons).
*/
static __inline int
doprint(struct ieee80211vap *vap, int subtype)
{
if (subtype == IEEE80211_FC0_SUBTYPE_PROBE_RESP)
return (vap->iv_opmode == IEEE80211_M_IBSS);
return 1;
}
#endif
/*
* Determine the priority based on VLAN and/or IP TOS. Priority is
* written into the skb->priority field. On success, returns 0. Failure
* due to bad or mis-matched vlan tag is indicated by non-zero return.
*/
static int
ieee80211_classify(struct ieee80211_node *ni, struct sk_buff *skb)
{
struct ieee80211vap *vap = ni->ni_vap;
struct ether_header *eh = (struct ether_header *)skb->data;
int v_wme_ac = 0, d_wme_ac = 0;
/* default priority */
skb->priority = WME_AC_BE;
if (!(ni->ni_flags & IEEE80211_NODE_QOS))
return 0;
/*
* If node has a vlan tag then all traffic
* to it must have a matching vlan id.
*/
if (ni->ni_vlan != 0 && vlan_tx_tag_present(skb)) {
u_int32_t tag=0;
int v_pri;
if (vap->iv_vlgrp == NULL) {
IEEE80211_NODE_STAT(ni, tx_novlantag);
ni->ni_stats.ns_tx_novlantag++;
return 1;
}
if (((tag = vlan_tx_tag_get(skb)) & VLAN_VID_MASK) !=
(ni->ni_vlan & VLAN_VID_MASK)) {
IEEE80211_NODE_STAT(ni, tx_vlanmismatch);
ni->ni_stats.ns_tx_vlanmismatch++;
return 1;
}
if (ni->ni_flags & IEEE80211_NODE_QOS) {
v_pri = (tag >> VLAN_PRI_SHIFT) & VLAN_PRI_MASK;
switch (v_pri) {
case 1:
case 2: /* Background (BK) */
v_wme_ac = WME_AC_BK;
break;
case 0:
case 3: /* Best Effort (BE) */
v_wme_ac = WME_AC_BE;
break;
case 4:
case 5: /* Video (VI) */
v_wme_ac = WME_AC_VI;
break;
case 6:
case 7: /* Voice (VO) */
v_wme_ac = WME_AC_VO;
break;
}
}
}
if (eh->ether_type == __constant_htons(ETHERTYPE_IP)) {
const struct iphdr *ip = (struct iphdr *)
(skb->data + sizeof (struct ether_header));
/*
* IP frame, map the TOS field.
*
* XXX: fill out these mappings???
*/
switch (ip->tos) {
case 0x08: /* Background */
case 0x20:
d_wme_ac = WME_AC_BK;
break;
case 0x28: /* Video */
case 0xa0:
d_wme_ac = WME_AC_VI;
break;
case 0x30: /* Voice */
case 0xe0:
case 0x88: /* XXX UPSD */
case 0xb8:
d_wme_ac = WME_AC_VO;
break;
default: /* All others */
d_wme_ac = WME_AC_BE;
break;
}
} else {
d_wme_ac = WME_AC_BE;
}
skb->priority = d_wme_ac;
if (v_wme_ac > d_wme_ac)
skb->priority = v_wme_ac;
/* Applying ACM policy */
if (vap->iv_opmode == IEEE80211_M_STA) {
struct ieee80211com *ic = ni->ni_ic;
while (skb->priority != WME_AC_BK &&
ic->ic_wme.wme_wmeBssChanParams.cap_wmeParams[skb->priority].wmep_acm) {
switch (skb->priority) {
case WME_AC_BE:
skb->priority = WME_AC_BK;
break;
case WME_AC_VI:
skb->priority = WME_AC_BE;
break;
case WME_AC_VO:
skb->priority = WME_AC_VI;
break;
default:
skb->priority = WME_AC_BK;
break;
}
}
}
return 0;
}
/*
* Context: process context (BHs disabled)
* It must return either NETDEV_TX_OK or NETDEV_TX_BUSY
*/
int
ieee80211_hardstart(struct sk_buff *skb, struct net_device *dev)
{
struct ieee80211vap *vap = netdev_priv(dev);
struct ieee80211com *ic = vap->iv_ic;
struct net_device *parent = ic->ic_dev;
struct ieee80211_node *ni = NULL;
struct ether_header *eh;
/* Reset the SKB of new frames reaching this layer BEFORE
This patch augments the current reference counting code with: * Counters for total outstanding instances for each resource type (skb, ath_node and ath_buf) * One pair of acquisition/release functions per resource type in unlocked and one in locked * Adds some more _debug versions of functions in the call chain that acquire/release resources so that the original func/line in the driver as well as the func/line that affected the resource use can be shown in the trace. Intermediate stack frames aren't necessary to trace the leaks. * Changes naming convention for "lock-required" functions to suffix _locked for the versions that expect locking, to be consistent with some other places in the code. * Consolidate debug messages to the helper functions that actually affect the reference count or acquire/release a resource * Additional sanity checks and leak detection (esp for detecting node ref leaks through skb) * skb references are nulled out by the new sbk unref/free function. I've tested these changes extensively and found lots of cases where we didn't get enough node references when cloning skbuff, and where the kernel drops packets due to performance issues and leaks our node references. With these changes and the tracing enabled I have verified that: * TX BUF: tx buffers always go down to zero when the tx queue is done, and you can watch tx queue usage ratio go up and down again as the driver is working. There are no leaks here at the moment, although there *are* some in the madwifi-dfs branch during CAC at the moment. * skbuff leaks in all the common flows are fixed. We were leaking node references in a lot of places where kernel was dropping skb's due to congestion and we were failing to increment node references when cloning skbuffs. These are now detected, as are skbuffs that are reaped by the kernel while still holding a node reference. * the ath_node count works correctly and on an idle system we get about 5 references per station table node, with 2 node instances per VAP. One for the bss and one for the node in the station table, I believe. The ath_node count goes up and down but always lands back at the stable number based on the vaps you have configured and the number of actual stations in the station table. The point here is that it's pretty constant what you will see over time, despite excessive node creation/release in our code during input (esp input_all). Thank god for the slab allocator. git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@2902 0192ed92-7a03-0410-a25b-9323aeb14dbd
2007-11-21 23:14:11 +03:00
* we invoke ieee80211_skb_track. */
memset(SKB_CB(skb), 0, sizeof(struct ieee80211_cb));
/* If an SKB is passed in directly from the kernel,
* we take responsibility for the reference. */
This patch augments the current reference counting code with: * Counters for total outstanding instances for each resource type (skb, ath_node and ath_buf) * One pair of acquisition/release functions per resource type in unlocked and one in locked * Adds some more _debug versions of functions in the call chain that acquire/release resources so that the original func/line in the driver as well as the func/line that affected the resource use can be shown in the trace. Intermediate stack frames aren't necessary to trace the leaks. * Changes naming convention for "lock-required" functions to suffix _locked for the versions that expect locking, to be consistent with some other places in the code. * Consolidate debug messages to the helper functions that actually affect the reference count or acquire/release a resource * Additional sanity checks and leak detection (esp for detecting node ref leaks through skb) * skb references are nulled out by the new sbk unref/free function. I've tested these changes extensively and found lots of cases where we didn't get enough node references when cloning skbuff, and where the kernel drops packets due to performance issues and leaks our node references. With these changes and the tracing enabled I have verified that: * TX BUF: tx buffers always go down to zero when the tx queue is done, and you can watch tx queue usage ratio go up and down again as the driver is working. There are no leaks here at the moment, although there *are* some in the madwifi-dfs branch during CAC at the moment. * skbuff leaks in all the common flows are fixed. We were leaking node references in a lot of places where kernel was dropping skb's due to congestion and we were failing to increment node references when cloning skbuffs. These are now detected, as are skbuffs that are reaped by the kernel while still holding a node reference. * the ath_node count works correctly and on an idle system we get about 5 references per station table node, with 2 node instances per VAP. One for the bss and one for the node in the station table, I believe. The ath_node count goes up and down but always lands back at the stable number based on the vaps you have configured and the number of actual stations in the station table. The point here is that it's pretty constant what you will see over time, despite excessive node creation/release in our code during input (esp input_all). Thank god for the slab allocator. git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@2902 0192ed92-7a03-0410-a25b-9323aeb14dbd
2007-11-21 23:14:11 +03:00
ieee80211_skb_track(skb);
/* NB: parent must be up and running. */
if ((parent->flags & (IFF_RUNNING|IFF_UP)) != (IFF_RUNNING|IFF_UP))
goto bad;
/* No data frames go out unless we're running. */
if (vap->iv_state != IEEE80211_S_RUN) {
IEEE80211_DPRINTF(vap, IEEE80211_MSG_OUTPUT,
"%s: ignore data packet, state %u\n",
__func__, vap->iv_state);
#if 0
vap->iv_stats.ist_tx_discard++;
#endif
goto bad;
}
if (vap->iv_opmode == IEEE80211_M_MONITOR) {
ieee80211_monitor_encap(vap, skb);
ieee80211_parent_queue_xmit(skb);
return NETDEV_TX_OK;
}
/* If we have detected a radar on the current channel, or another node
* told us to stop transmitting, we no longer transmit. Note : we
* still allow a monitor interface to transmit */
if (IEEE80211_IS_CHAN_RADAR(ic->ic_curchan) &&
(ic->ic_flags & IEEE80211_F_DOTH)) {
IEEE80211_DPRINTF(vap, IEEE80211_MSG_DOTH,
"%s: dropping data since we are under radar\n",
__func__);
goto bad;
}
/* Cancel any running BG scan */
ieee80211_cancel_scan(vap);
/* Find the node for the destination so we can do
* things like power save. */
eh = (struct ether_header *)skb->data;
if (vap->iv_opmode == IEEE80211_M_WDS)
ni = ieee80211_find_txnode(vap, vap->wds_mac);
else
ni = ieee80211_find_txnode(vap, eh->ether_dhost);
if (ni == NULL) {
/* NB: ieee80211_find_txnode does stat+msg */
goto bad;
}
/* Calculate priority so drivers can find the TX queue. */
if (ieee80211_classify(ni, skb)) {
IEEE80211_NOTE(vap, IEEE80211_MSG_OUTPUT, ni,
"%s: discard, classification failure", __func__);
goto bad;
}
SKB_NI(skb) = ieee80211_ref_node(ni);
/* Power-save checks. */
if (WME_UAPSD_AC_CAN_TRIGGER(skb->priority, ni)) {
/* U-APSD power save queue */
/* XXXAPSD: assuming triggerable means deliverable */
M_FLAG_SET(skb, M_UAPSD);
} else if ((ni->ni_flags & IEEE80211_NODE_PWR_MGT)) {
/* Station in power save mode; stick the frame
* on the STA's power save queue and continue.
* We'll get the frame back when the time is right. */
ieee80211_unref_node(&ni);
return ieee80211_pwrsave(skb);
}
dev->trans_start = jiffies;
#ifdef ATH_SUPERG_XR
/* Broadcast/multicast packets need to be sent on XR vap in addition to
* normal vap. */
if (vap->iv_xrvap && (ni == vap->iv_bss) &&
vap->iv_xrvap->iv_sta_assoc) {
struct sk_buff *skb1 = skb_copy(skb, GFP_ATOMIC);
if (skb1) {
memset(SKB_CB(skb1), 0, sizeof(struct ieee80211_cb));
This patch augments the current reference counting code with: * Counters for total outstanding instances for each resource type (skb, ath_node and ath_buf) * One pair of acquisition/release functions per resource type in unlocked and one in locked * Adds some more _debug versions of functions in the call chain that acquire/release resources so that the original func/line in the driver as well as the func/line that affected the resource use can be shown in the trace. Intermediate stack frames aren't necessary to trace the leaks. * Changes naming convention for "lock-required" functions to suffix _locked for the versions that expect locking, to be consistent with some other places in the code. * Consolidate debug messages to the helper functions that actually affect the reference count or acquire/release a resource * Additional sanity checks and leak detection (esp for detecting node ref leaks through skb) * skb references are nulled out by the new sbk unref/free function. I've tested these changes extensively and found lots of cases where we didn't get enough node references when cloning skbuff, and where the kernel drops packets due to performance issues and leaks our node references. With these changes and the tracing enabled I have verified that: * TX BUF: tx buffers always go down to zero when the tx queue is done, and you can watch tx queue usage ratio go up and down again as the driver is working. There are no leaks here at the moment, although there *are* some in the madwifi-dfs branch during CAC at the moment. * skbuff leaks in all the common flows are fixed. We were leaking node references in a lot of places where kernel was dropping skb's due to congestion and we were failing to increment node references when cloning skbuffs. These are now detected, as are skbuffs that are reaped by the kernel while still holding a node reference. * the ath_node count works correctly and on an idle system we get about 5 references per station table node, with 2 node instances per VAP. One for the bss and one for the node in the station table, I believe. The ath_node count goes up and down but always lands back at the stable number based on the vaps you have configured and the number of actual stations in the station table. The point here is that it's pretty constant what you will see over time, despite excessive node creation/release in our code during input (esp input_all). Thank god for the slab allocator. git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@2902 0192ed92-7a03-0410-a25b-9323aeb14dbd
2007-11-21 23:14:11 +03:00
#ifdef IEEE80211_DEBUG_REFCNT
M_FLAG_SET(skb1, M_SKB_TRACKED);
This patch augments the current reference counting code with: * Counters for total outstanding instances for each resource type (skb, ath_node and ath_buf) * One pair of acquisition/release functions per resource type in unlocked and one in locked * Adds some more _debug versions of functions in the call chain that acquire/release resources so that the original func/line in the driver as well as the func/line that affected the resource use can be shown in the trace. Intermediate stack frames aren't necessary to trace the leaks. * Changes naming convention for "lock-required" functions to suffix _locked for the versions that expect locking, to be consistent with some other places in the code. * Consolidate debug messages to the helper functions that actually affect the reference count or acquire/release a resource * Additional sanity checks and leak detection (esp for detecting node ref leaks through skb) * skb references are nulled out by the new sbk unref/free function. I've tested these changes extensively and found lots of cases where we didn't get enough node references when cloning skbuff, and where the kernel drops packets due to performance issues and leaks our node references. With these changes and the tracing enabled I have verified that: * TX BUF: tx buffers always go down to zero when the tx queue is done, and you can watch tx queue usage ratio go up and down again as the driver is working. There are no leaks here at the moment, although there *are* some in the madwifi-dfs branch during CAC at the moment. * skbuff leaks in all the common flows are fixed. We were leaking node references in a lot of places where kernel was dropping skb's due to congestion and we were failing to increment node references when cloning skbuffs. These are now detected, as are skbuffs that are reaped by the kernel while still holding a node reference. * the ath_node count works correctly and on an idle system we get about 5 references per station table node, with 2 node instances per VAP. One for the bss and one for the node in the station table, I believe. The ath_node count goes up and down but always lands back at the stable number based on the vaps you have configured and the number of actual stations in the station table. The point here is that it's pretty constant what you will see over time, despite excessive node creation/release in our code during input (esp input_all). Thank god for the slab allocator. git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@2902 0192ed92-7a03-0410-a25b-9323aeb14dbd
2007-11-21 23:14:11 +03:00
#endif /* #ifdef IEEE80211_DEBUG_REFCNT */
SKB_NI(skb1) = ieee80211_find_txnode(vap->iv_xrvap,
eh->ether_dhost);
/* Ignore this return code. */
ieee80211_parent_queue_xmit(skb1);
}
}
#endif
ieee80211_unref_node(&ni);
ieee80211_parent_queue_xmit(skb);
return NETDEV_TX_OK;
bad:
if (skb != NULL)
This patch augments the current reference counting code with: * Counters for total outstanding instances for each resource type (skb, ath_node and ath_buf) * One pair of acquisition/release functions per resource type in unlocked and one in locked * Adds some more _debug versions of functions in the call chain that acquire/release resources so that the original func/line in the driver as well as the func/line that affected the resource use can be shown in the trace. Intermediate stack frames aren't necessary to trace the leaks. * Changes naming convention for "lock-required" functions to suffix _locked for the versions that expect locking, to be consistent with some other places in the code. * Consolidate debug messages to the helper functions that actually affect the reference count or acquire/release a resource * Additional sanity checks and leak detection (esp for detecting node ref leaks through skb) * skb references are nulled out by the new sbk unref/free function. I've tested these changes extensively and found lots of cases where we didn't get enough node references when cloning skbuff, and where the kernel drops packets due to performance issues and leaks our node references. With these changes and the tracing enabled I have verified that: * TX BUF: tx buffers always go down to zero when the tx queue is done, and you can watch tx queue usage ratio go up and down again as the driver is working. There are no leaks here at the moment, although there *are* some in the madwifi-dfs branch during CAC at the moment. * skbuff leaks in all the common flows are fixed. We were leaking node references in a lot of places where kernel was dropping skb's due to congestion and we were failing to increment node references when cloning skbuffs. These are now detected, as are skbuffs that are reaped by the kernel while still holding a node reference. * the ath_node count works correctly and on an idle system we get about 5 references per station table node, with 2 node instances per VAP. One for the bss and one for the node in the station table, I believe. The ath_node count goes up and down but always lands back at the stable number based on the vaps you have configured and the number of actual stations in the station table. The point here is that it's pretty constant what you will see over time, despite excessive node creation/release in our code during input (esp input_all). Thank god for the slab allocator. git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@2902 0192ed92-7a03-0410-a25b-9323aeb14dbd
2007-11-21 23:14:11 +03:00
ieee80211_dev_kfree_skb(&skb);
if (ni != NULL)
ieee80211_unref_node(&ni);
return NETDEV_TX_OK;
}
/*
* SKB is consumed in all cases.
*/
void ieee80211_parent_queue_xmit(struct sk_buff *skb) {
struct ieee80211vap *vap = netdev_priv(skb->dev);
vap->iv_devstats.tx_packets++;
vap->iv_devstats.tx_bytes += skb->len;
vap->iv_ic->ic_lastdata = jiffies;
/* Dispatch the packet to the parent device */
skb->dev = vap->iv_ic->ic_dev;
if (dev_queue_xmit(skb) == NET_XMIT_DROP)
vap->iv_devstats.tx_dropped++;
}
/*
* Set the direction field and address fields of an outgoing
* non-QoS frame. Note this should be called early on in
* constructing a frame as it sets i_fc[1]; other bits can
* then be or'd in.
*/
static void
ieee80211_send_setup(struct ieee80211vap *vap,
struct ieee80211_node *ni,
struct ieee80211_frame *wh,
int type,
const u_int8_t sa[IEEE80211_ADDR_LEN],
const u_int8_t da[IEEE80211_ADDR_LEN],
const u_int8_t bssid[IEEE80211_ADDR_LEN])
{
#define WH4(wh) ((struct ieee80211_frame_addr4 *)wh)
wh->i_fc[0] = IEEE80211_FC0_VERSION_0 | type;
if ((type & IEEE80211_FC0_TYPE_MASK) == IEEE80211_FC0_TYPE_DATA) {
switch (vap->iv_opmode) {
case IEEE80211_M_STA:
wh->i_fc[1] = IEEE80211_FC1_DIR_TODS;
IEEE80211_ADDR_COPY(wh->i_addr1, bssid);
IEEE80211_ADDR_COPY(wh->i_addr2, sa);
IEEE80211_ADDR_COPY(wh->i_addr3, da);
break;
case IEEE80211_M_IBSS:
case IEEE80211_M_AHDEMO:
wh->i_fc[1] = IEEE80211_FC1_DIR_NODS;
IEEE80211_ADDR_COPY(wh->i_addr1, da);
IEEE80211_ADDR_COPY(wh->i_addr2, sa);
IEEE80211_ADDR_COPY(wh->i_addr3, bssid);
break;
case IEEE80211_M_HOSTAP:
wh->i_fc[1] = IEEE80211_FC1_DIR_FROMDS;
IEEE80211_ADDR_COPY(wh->i_addr1, da);
IEEE80211_ADDR_COPY(wh->i_addr2, bssid);
IEEE80211_ADDR_COPY(wh->i_addr3, sa);
break;
case IEEE80211_M_WDS:
wh->i_fc[1] = IEEE80211_FC1_DIR_DSTODS;
/* XXX cheat, bssid holds RA */
IEEE80211_ADDR_COPY(wh->i_addr1, bssid);
IEEE80211_ADDR_COPY(wh->i_addr2, vap->iv_myaddr);
IEEE80211_ADDR_COPY(wh->i_addr3, da);
IEEE80211_ADDR_COPY(WH4(wh)->i_addr4, sa);
break;
case IEEE80211_M_MONITOR: /* NB: to quiet compiler */
break;
}
} else {
wh->i_fc[1] = IEEE80211_FC1_DIR_NODS;
IEEE80211_ADDR_COPY(wh->i_addr1, da);
IEEE80211_ADDR_COPY(wh->i_addr2, sa);
IEEE80211_ADDR_COPY(wh->i_addr3, bssid);
}
wh->i_dur = 0;
/* NB: use non-QoS tid */
*(__le16 *)&wh->i_seq[0] =
htole16(ni->ni_txseqs[0] << IEEE80211_SEQ_SEQ_SHIFT);
ni->ni_txseqs[0]++;
#undef WH4
}
/*
* Send a management frame to the specified node. The node pointer
* must have a reference as the pointer will be passed to the driver
* and potentially held for a long time. If the frame is successfully
* dispatched to the driver, then it is responsible for freeing the
* reference (and potentially freeing up any associated storage).
*/
static void
ieee80211_mgmt_output(struct ieee80211_node *ni, struct sk_buff *skb, int type,
const u_int8_t da[IEEE80211_ADDR_LEN])
{
struct ieee80211vap *vap = ni->ni_vap;
struct ieee80211com *ic = ni->ni_ic;
struct ieee80211_frame *wh;
KASSERT(ni != NULL, ("null node"));
SKB_NI(skb) = ni;
wh = (struct ieee80211_frame *)
skb_push(skb, sizeof(struct ieee80211_frame));
ieee80211_send_setup(vap, ni, wh,
IEEE80211_FC0_TYPE_MGT | type,
vap->iv_myaddr, da, vap->iv_bssid);
/* XXX power management */
if ((SKB_CB(skb)->flags & M_LINK0) != 0 && ni->ni_challenge != NULL) {
SKB_CB(skb)->flags &= ~M_LINK0;
IEEE80211_NOTE_MAC(vap, IEEE80211_MSG_AUTH, wh->i_addr1,
"encrypting frame (%s)", __func__);
wh->i_fc[1] |= IEEE80211_FC1_PROT;
}
if (IEEE80211_VAP_IS_SLEEPING(ni->ni_vap))
wh->i_fc[1] |= IEEE80211_FC1_PWR_MGT;
#ifdef IEEE80211_DEBUG
/* avoid printing too many frames */
if ((ieee80211_msg_debug(vap) && doprint(vap, type)) ||
ieee80211_msg_dumppkts(vap)) {
printk(KERN_DEBUG "[" MAC_FMT "] send %s on channel %u\n",
MAC_ADDR(wh->i_addr1),
ieee80211_mgt_subtype_name[
(type & IEEE80211_FC0_SUBTYPE_MASK) >>
IEEE80211_FC0_SUBTYPE_SHIFT],
ieee80211_chan2ieee(ic, ic->ic_curchan));
}
#endif
IEEE80211_NODE_STAT(ni, tx_mgmt);
(void) ic->ic_mgtstart(ic, skb);
}
/* Send a null data frame to the specified node.
*
* NB: the caller provides us with our own node reference this must not be
* leaked; this is necessary to deal with a race condition when
* probing for inactive stations. */
int
ieee80211_send_nulldata(struct ieee80211_node *ni)
{
struct ieee80211vap *vap = ni->ni_vap;
struct ieee80211com *ic = ni->ni_ic;
struct sk_buff *skb;
struct ieee80211_frame *wh;
u_int8_t *frm;
skb = ieee80211_getmgtframe(&frm, 0);
if (skb == NULL) {
/* XXX debug msg */
vap->iv_stats.is_tx_nobuf++;
ieee80211_unref_node(&ni);
return -ENOMEM;
}
wh = (struct ieee80211_frame *)
skb_push(skb, sizeof(struct ieee80211_frame));
ieee80211_send_setup(vap, ni, wh,
IEEE80211_FC0_TYPE_DATA | IEEE80211_FC0_SUBTYPE_NULL,
vap->iv_myaddr, ni->ni_macaddr, vap->iv_bssid);
/* NB: power management bit is never sent by an AP */
if ((IEEE80211_VAP_IS_SLEEPING(ni->ni_vap)) &&
vap->iv_opmode != IEEE80211_M_HOSTAP &&
vap->iv_opmode != IEEE80211_M_WDS)
wh->i_fc[1] |= IEEE80211_FC1_PWR_MGT;
IEEE80211_NODE_STAT(ni, tx_data);
IEEE80211_DPRINTF(vap, IEEE80211_MSG_DEBUG | IEEE80211_MSG_DUMPPKTS,
"[" MAC_FMT "] send null data frame on channel %u, pwr mgt %s\n",
MAC_ADDR(ni->ni_macaddr),
ieee80211_chan2ieee(ic, ic->ic_curchan),
wh->i_fc[1] & IEEE80211_FC1_PWR_MGT ? "ena" : "dis");
/* XXX assign some priority; this probably is wrong */
skb->priority = WME_AC_BE;
SKB_NI(skb) = PASS_NODE(ni);
(void) ic->ic_mgtstart(ic, skb); /* cheat */
return 0;
}
/*
* NB: unlike ieee80211_send_nulldata(), the node refcnt is
* bumped within this function.
*/
int
ieee80211_send_qosnulldata(struct ieee80211_node *ni, int ac)
{
struct ieee80211vap *vap = ni->ni_vap;
struct ieee80211com *ic = ni->ni_ic;
struct sk_buff *skb;
struct ieee80211_qosframe *qwh;
u_int8_t *frm;
int tid;
skb = ieee80211_getmgtframe(&frm, 2);
if (skb == NULL) {
/* XXX debug msg */
vap->iv_stats.is_tx_nobuf++;
return -ENOMEM;
}
SKB_NI(skb) = ieee80211_ref_node(ni);
skb->priority = ac;
qwh = (struct ieee80211_qosframe *)skb_push(skb, sizeof(struct ieee80211_qosframe));
qwh = (struct ieee80211_qosframe *)skb->data;
ieee80211_send_setup(vap, ni, (struct ieee80211_frame *)qwh,
IEEE80211_FC0_TYPE_DATA,
vap->iv_myaddr, /* SA */
ni->ni_macaddr, /* DA */
vap->iv_bssid);
qwh->i_fc[0] = IEEE80211_FC0_VERSION_0 | IEEE80211_FC0_TYPE_DATA |
IEEE80211_FC0_SUBTYPE_QOS_NULL;
if (IEEE80211_VAP_IS_SLEEPING(ni->ni_vap))
qwh->i_fc[1] |= IEEE80211_FC1_PWR_MGT;
/* map from access class/queue to 11e header priority value */
tid = WME_AC_TO_TID(ac);
qwh->i_qos[0] = tid & IEEE80211_QOS_TID;
if (ic->ic_wme.wme_wmeChanParams.cap_wmeParams[ac].wmep_noackPolicy)
qwh->i_qos[0] |= (1 << IEEE80211_QOS_ACKPOLICY_S) & IEEE80211_QOS_ACKPOLICY;
qwh->i_qos[1] = 0;
IEEE80211_NODE_STAT(ni, tx_data);
if (WME_UAPSD_AC_CAN_TRIGGER(skb->priority, ni)) {
/* U-APSD power save queue */
/* XXXAPSD: assuming triggerable means deliverable */
M_FLAG_SET(skb, M_UAPSD);
}
(void) ic->ic_mgtstart(ic, skb); /* cheat */
return 0;
}
EXPORT_SYMBOL(ieee80211_send_qosnulldata);
/*
* Ensure there is sufficient headroom and tailroom to
* encapsulate the 802.11 data frame. If room isn't
* already there, reallocate so there is enough space.
* Drivers and cipher modules assume we have done the
* necessary work and fail rudely if they don't find
* the space they need.
*/
static struct sk_buff *
ieee80211_skbhdr_adjust(struct ieee80211vap *vap, int hdrsize,
struct ieee80211_key *key, struct sk_buff *skb, int ismulticast)
{
/* XXX pre-calculate per node? */
int need_headroom = LLC_SNAPFRAMELEN + hdrsize + IEEE80211_ADDR_LEN;
int need_tailroom = 0;
#ifdef ATH_SUPERG_FF
int isff = ATH_FF_MAGIC_PRESENT(skb);
int inter_headroom = sizeof(struct ether_header) + LLC_SNAPFRAMELEN + ATH_FF_MAX_HDR_PAD;
struct sk_buff *skb2 = NULL;
if (isff) {
need_headroom += sizeof(struct athl2p_tunnel_hdr) + ATH_FF_MAX_HDR_PAD +
inter_headroom;
skb2 = skb->next;
}
#endif
if (key != NULL) {
const struct ieee80211_cipher *cip = key->wk_cipher;
/*
* Adjust for crypto needs. When hardware crypto is
* being used we assume the hardware/driver will deal
* with any padding (on the fly, without needing to
* expand the frame contents). When software crypto
* is used we need to ensure room is available at the
* front and back and also for any per-MSDU additions.
*/
/* XXX belongs in crypto code? */
need_headroom += cip->ic_header;
/* XXX pre-calculate per key */
if (key->wk_flags & IEEE80211_KEY_SWCRYPT)
need_tailroom += cip->ic_trailer;
/*
** If tx frag is needed and cipher is TKIP,
** then allocate the additional tailroom for SW MIC computation.
*/
if (skb->len > vap->iv_fragthreshold &&
!ismulticast &&
cip->ic_cipher == IEEE80211_CIPHER_TKIP)
need_tailroom += cip->ic_miclen;
else
if (key->wk_flags & IEEE80211_KEY_SWMIC)
need_tailroom += cip->ic_miclen;
}
if (skb_shared(skb)) {
/* Take our own reference to the node in the clone */
ieee80211_ref_node(SKB_NI(skb));
/* Unshare the node, decrementing users in the old skb */
skb = skb_unshare(skb, GFP_ATOMIC);
}
#ifdef ATH_SUPERG_FF
if (isff) {
if (skb == NULL) {
IEEE80211_DPRINTF(vap, IEEE80211_MSG_OUTPUT,
"%s: cannot unshare for encapsulation\n",
__func__);
vap->iv_stats.is_tx_nobuf++;
This patch augments the current reference counting code with: * Counters for total outstanding instances for each resource type (skb, ath_node and ath_buf) * One pair of acquisition/release functions per resource type in unlocked and one in locked * Adds some more _debug versions of functions in the call chain that acquire/release resources so that the original func/line in the driver as well as the func/line that affected the resource use can be shown in the trace. Intermediate stack frames aren't necessary to trace the leaks. * Changes naming convention for "lock-required" functions to suffix _locked for the versions that expect locking, to be consistent with some other places in the code. * Consolidate debug messages to the helper functions that actually affect the reference count or acquire/release a resource * Additional sanity checks and leak detection (esp for detecting node ref leaks through skb) * skb references are nulled out by the new sbk unref/free function. I've tested these changes extensively and found lots of cases where we didn't get enough node references when cloning skbuff, and where the kernel drops packets due to performance issues and leaks our node references. With these changes and the tracing enabled I have verified that: * TX BUF: tx buffers always go down to zero when the tx queue is done, and you can watch tx queue usage ratio go up and down again as the driver is working. There are no leaks here at the moment, although there *are* some in the madwifi-dfs branch during CAC at the moment. * skbuff leaks in all the common flows are fixed. We were leaking node references in a lot of places where kernel was dropping skb's due to congestion and we were failing to increment node references when cloning skbuffs. These are now detected, as are skbuffs that are reaped by the kernel while still holding a node reference. * the ath_node count works correctly and on an idle system we get about 5 references per station table node, with 2 node instances per VAP. One for the bss and one for the node in the station table, I believe. The ath_node count goes up and down but always lands back at the stable number based on the vaps you have configured and the number of actual stations in the station table. The point here is that it's pretty constant what you will see over time, despite excessive node creation/release in our code during input (esp input_all). Thank god for the slab allocator. git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@2902 0192ed92-7a03-0410-a25b-9323aeb14dbd
2007-11-21 23:14:11 +03:00
ieee80211_dev_kfree_skb(&skb2);
return NULL;
}
/* first skb header */
if (skb_headroom(skb) < need_headroom) {
struct sk_buff *tmp = skb;
skb = skb_realloc_headroom(skb, need_headroom);
if (skb == NULL) {
IEEE80211_DPRINTF(vap, IEEE80211_MSG_OUTPUT,
"%s: cannot expand storage (head1)\n",
__func__);
vap->iv_stats.is_tx_nobuf++;
This patch augments the current reference counting code with: * Counters for total outstanding instances for each resource type (skb, ath_node and ath_buf) * One pair of acquisition/release functions per resource type in unlocked and one in locked * Adds some more _debug versions of functions in the call chain that acquire/release resources so that the original func/line in the driver as well as the func/line that affected the resource use can be shown in the trace. Intermediate stack frames aren't necessary to trace the leaks. * Changes naming convention for "lock-required" functions to suffix _locked for the versions that expect locking, to be consistent with some other places in the code. * Consolidate debug messages to the helper functions that actually affect the reference count or acquire/release a resource * Additional sanity checks and leak detection (esp for detecting node ref leaks through skb) * skb references are nulled out by the new sbk unref/free function. I've tested these changes extensively and found lots of cases where we didn't get enough node references when cloning skbuff, and where the kernel drops packets due to performance issues and leaks our node references. With these changes and the tracing enabled I have verified that: * TX BUF: tx buffers always go down to zero when the tx queue is done, and you can watch tx queue usage ratio go up and down again as the driver is working. There are no leaks here at the moment, although there *are* some in the madwifi-dfs branch during CAC at the moment. * skbuff leaks in all the common flows are fixed. We were leaking node references in a lot of places where kernel was dropping skb's due to congestion and we were failing to increment node references when cloning skbuffs. These are now detected, as are skbuffs that are reaped by the kernel while still holding a node reference. * the ath_node count works correctly and on an idle system we get about 5 references per station table node, with 2 node instances per VAP. One for the bss and one for the node in the station table, I believe. The ath_node count goes up and down but always lands back at the stable number based on the vaps you have configured and the number of actual stations in the station table. The point here is that it's pretty constant what you will see over time, despite excessive node creation/release in our code during input (esp input_all). Thank god for the slab allocator. git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@2902 0192ed92-7a03-0410-a25b-9323aeb14dbd
2007-11-21 23:14:11 +03:00
ieee80211_dev_kfree_skb(&skb2);
return NULL;
} else
ieee80211_skb_copy_noderef(tmp, skb);
ieee80211_dev_kfree_skb(&tmp);
/* NB: cb[] area was copied, but not next ptr. must do that
* prior to return on success. */
}
/* second skb with header and tail adjustments possible */
if (skb_tailroom(skb2) < need_tailroom) {
int n = 0;
if (inter_headroom > skb_headroom(skb2))
n = inter_headroom - skb_headroom(skb2);
if (pskb_expand_head(skb2, n,
need_tailroom - skb_tailroom(skb2), GFP_ATOMIC)) {
This patch augments the current reference counting code with: * Counters for total outstanding instances for each resource type (skb, ath_node and ath_buf) * One pair of acquisition/release functions per resource type in unlocked and one in locked * Adds some more _debug versions of functions in the call chain that acquire/release resources so that the original func/line in the driver as well as the func/line that affected the resource use can be shown in the trace. Intermediate stack frames aren't necessary to trace the leaks. * Changes naming convention for "lock-required" functions to suffix _locked for the versions that expect locking, to be consistent with some other places in the code. * Consolidate debug messages to the helper functions that actually affect the reference count or acquire/release a resource * Additional sanity checks and leak detection (esp for detecting node ref leaks through skb) * skb references are nulled out by the new sbk unref/free function. I've tested these changes extensively and found lots of cases where we didn't get enough node references when cloning skbuff, and where the kernel drops packets due to performance issues and leaks our node references. With these changes and the tracing enabled I have verified that: * TX BUF: tx buffers always go down to zero when the tx queue is done, and you can watch tx queue usage ratio go up and down again as the driver is working. There are no leaks here at the moment, although there *are* some in the madwifi-dfs branch during CAC at the moment. * skbuff leaks in all the common flows are fixed. We were leaking node references in a lot of places where kernel was dropping skb's due to congestion and we were failing to increment node references when cloning skbuffs. These are now detected, as are skbuffs that are reaped by the kernel while still holding a node reference. * the ath_node count works correctly and on an idle system we get about 5 references per station table node, with 2 node instances per VAP. One for the bss and one for the node in the station table, I believe. The ath_node count goes up and down but always lands back at the stable number based on the vaps you have configured and the number of actual stations in the station table. The point here is that it's pretty constant what you will see over time, despite excessive node creation/release in our code during input (esp input_all). Thank god for the slab allocator. git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@2902 0192ed92-7a03-0410-a25b-9323aeb14dbd
2007-11-21 23:14:11 +03:00
ieee80211_dev_kfree_skb(&skb2);
IEEE80211_DPRINTF(vap, IEEE80211_MSG_OUTPUT,
"%s: cannot expand storage (tail2)\n",
__func__);
vap->iv_stats.is_tx_nobuf++;
/* this shouldn't happen, but don't send first ff either */
This patch augments the current reference counting code with: * Counters for total outstanding instances for each resource type (skb, ath_node and ath_buf) * One pair of acquisition/release functions per resource type in unlocked and one in locked * Adds some more _debug versions of functions in the call chain that acquire/release resources so that the original func/line in the driver as well as the func/line that affected the resource use can be shown in the trace. Intermediate stack frames aren't necessary to trace the leaks. * Changes naming convention for "lock-required" functions to suffix _locked for the versions that expect locking, to be consistent with some other places in the code. * Consolidate debug messages to the helper functions that actually affect the reference count or acquire/release a resource * Additional sanity checks and leak detection (esp for detecting node ref leaks through skb) * skb references are nulled out by the new sbk unref/free function. I've tested these changes extensively and found lots of cases where we didn't get enough node references when cloning skbuff, and where the kernel drops packets due to performance issues and leaks our node references. With these changes and the tracing enabled I have verified that: * TX BUF: tx buffers always go down to zero when the tx queue is done, and you can watch tx queue usage ratio go up and down again as the driver is working. There are no leaks here at the moment, although there *are* some in the madwifi-dfs branch during CAC at the moment. * skbuff leaks in all the common flows are fixed. We were leaking node references in a lot of places where kernel was dropping skb's due to congestion and we were failing to increment node references when cloning skbuffs. These are now detected, as are skbuffs that are reaped by the kernel while still holding a node reference. * the ath_node count works correctly and on an idle system we get about 5 references per station table node, with 2 node instances per VAP. One for the bss and one for the node in the station table, I believe. The ath_node count goes up and down but always lands back at the stable number based on the vaps you have configured and the number of actual stations in the station table. The point here is that it's pretty constant what you will see over time, despite excessive node creation/release in our code during input (esp input_all). Thank god for the slab allocator. git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@2902 0192ed92-7a03-0410-a25b-9323aeb14dbd
2007-11-21 23:14:11 +03:00
ieee80211_dev_kfree_skb(&skb);
}
} else if (skb_headroom(skb2) < inter_headroom) {
struct sk_buff *tmp = skb2;
skb2 = skb_realloc_headroom(skb2, inter_headroom);
if (skb2 == NULL) {
IEEE80211_DPRINTF(vap, IEEE80211_MSG_OUTPUT,
"%s: cannot expand storage (head2)\n",
__func__);
vap->iv_stats.is_tx_nobuf++;
/* this shouldn't happen, but don't send first ff either */
This patch augments the current reference counting code with: * Counters for total outstanding instances for each resource type (skb, ath_node and ath_buf) * One pair of acquisition/release functions per resource type in unlocked and one in locked * Adds some more _debug versions of functions in the call chain that acquire/release resources so that the original func/line in the driver as well as the func/line that affected the resource use can be shown in the trace. Intermediate stack frames aren't necessary to trace the leaks. * Changes naming convention for "lock-required" functions to suffix _locked for the versions that expect locking, to be consistent with some other places in the code. * Consolidate debug messages to the helper functions that actually affect the reference count or acquire/release a resource * Additional sanity checks and leak detection (esp for detecting node ref leaks through skb) * skb references are nulled out by the new sbk unref/free function. I've tested these changes extensively and found lots of cases where we didn't get enough node references when cloning skbuff, and where the kernel drops packets due to performance issues and leaks our node references. With these changes and the tracing enabled I have verified that: * TX BUF: tx buffers always go down to zero when the tx queue is done, and you can watch tx queue usage ratio go up and down again as the driver is working. There are no leaks here at the moment, although there *are* some in the madwifi-dfs branch during CAC at the moment. * skbuff leaks in all the common flows are fixed. We were leaking node references in a lot of places where kernel was dropping skb's due to congestion and we were failing to increment node references when cloning skbuffs. These are now detected, as are skbuffs that are reaped by the kernel while still holding a node reference. * the ath_node count works correctly and on an idle system we get about 5 references per station table node, with 2 node instances per VAP. One for the bss and one for the node in the station table, I believe. The ath_node count goes up and down but always lands back at the stable number based on the vaps you have configured and the number of actual stations in the station table. The point here is that it's pretty constant what you will see over time, despite excessive node creation/release in our code during input (esp input_all). Thank god for the slab allocator. git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@2902 0192ed92-7a03-0410-a25b-9323aeb14dbd
2007-11-21 23:14:11 +03:00
ieee80211_dev_kfree_skb(&skb);
skb = NULL;
} else
ieee80211_skb_copy_noderef(tmp, skb);
ieee80211_dev_kfree_skb(&tmp);
}
if (skb) {
skb->next = skb2;
}
return skb;
}
#endif /* ATH_SUPERG_FF */
if (skb == NULL) {
IEEE80211_DPRINTF(vap, IEEE80211_MSG_OUTPUT,
"%s: cannot unshare for encapsulation\n", __func__);
vap->iv_stats.is_tx_nobuf++;
} else if (skb_tailroom(skb) < need_tailroom) {
int n = 0;
if (need_headroom > skb_headroom(skb))
n = need_headroom - skb_headroom(skb);
if (pskb_expand_head(skb, n, need_tailroom -
skb_tailroom(skb), GFP_ATOMIC)) {
IEEE80211_DPRINTF(vap, IEEE80211_MSG_OUTPUT,
"%s: cannot expand storage (tail)\n", __func__);
vap->iv_stats.is_tx_nobuf++;
This patch augments the current reference counting code with: * Counters for total outstanding instances for each resource type (skb, ath_node and ath_buf) * One pair of acquisition/release functions per resource type in unlocked and one in locked * Adds some more _debug versions of functions in the call chain that acquire/release resources so that the original func/line in the driver as well as the func/line that affected the resource use can be shown in the trace. Intermediate stack frames aren't necessary to trace the leaks. * Changes naming convention for "lock-required" functions to suffix _locked for the versions that expect locking, to be consistent with some other places in the code. * Consolidate debug messages to the helper functions that actually affect the reference count or acquire/release a resource * Additional sanity checks and leak detection (esp for detecting node ref leaks through skb) * skb references are nulled out by the new sbk unref/free function. I've tested these changes extensively and found lots of cases where we didn't get enough node references when cloning skbuff, and where the kernel drops packets due to performance issues and leaks our node references. With these changes and the tracing enabled I have verified that: * TX BUF: tx buffers always go down to zero when the tx queue is done, and you can watch tx queue usage ratio go up and down again as the driver is working. There are no leaks here at the moment, although there *are* some in the madwifi-dfs branch during CAC at the moment. * skbuff leaks in all the common flows are fixed. We were leaking node references in a lot of places where kernel was dropping skb's due to congestion and we were failing to increment node references when cloning skbuffs. These are now detected, as are skbuffs that are reaped by the kernel while still holding a node reference. * the ath_node count works correctly and on an idle system we get about 5 references per station table node, with 2 node instances per VAP. One for the bss and one for the node in the station table, I believe. The ath_node count goes up and down but always lands back at the stable number based on the vaps you have configured and the number of actual stations in the station table. The point here is that it's pretty constant what you will see over time, despite excessive node creation/release in our code during input (esp input_all). Thank god for the slab allocator. git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@2902 0192ed92-7a03-0410-a25b-9323aeb14dbd
2007-11-21 23:14:11 +03:00
ieee80211_dev_kfree_skb(&skb);
}
} else if (skb_headroom(skb) < need_headroom) {
struct sk_buff *tmp = skb;
skb = skb_realloc_headroom(skb, need_headroom);
/* Increment reference count after copy */
if (skb == NULL) {
IEEE80211_DPRINTF(vap, IEEE80211_MSG_OUTPUT,
"%s: cannot expand storage (head)\n", __func__);
vap->iv_stats.is_tx_nobuf++;
} else
ieee80211_skb_copy_noderef(tmp, skb);
ieee80211_dev_kfree_skb(&tmp);
}
return skb;
}
#define KEY_UNDEFINED(k) ((k).wk_cipher == &ieee80211_cipher_none)
/*
* Return the transmit key to use in sending a unicast frame.
* If a unicast key is set we use that. When no unicast key is set
* we fall back to the default transmit key.
*/
static __inline struct ieee80211_key *
ieee80211_crypto_getucastkey(struct ieee80211vap *vap, struct ieee80211_node *ni)
{
if (KEY_UNDEFINED(ni->ni_ucastkey)) {
if (vap->iv_def_txkey == IEEE80211_KEYIX_NONE ||
KEY_UNDEFINED(vap->iv_nw_keys[vap->iv_def_txkey]))
return NULL;
return &vap->iv_nw_keys[vap->iv_def_txkey];
} else
return &ni->ni_ucastkey;
}
/*
* Return the transmit key to use in sending a multicast frame.
* Multicast traffic always uses the group key which is installed as
* the default tx key.
*/
static __inline struct ieee80211_key *
ieee80211_crypto_getmcastkey(struct ieee80211vap *vap, struct ieee80211_node *ni)
{
if (vap->iv_def_txkey == IEEE80211_KEYIX_NONE ||
KEY_UNDEFINED(vap->iv_nw_keys[vap->iv_def_txkey]))
return NULL;
return &vap->iv_nw_keys[vap->iv_def_txkey];
}
/*
* Encapsulate an outbound data frame. The mbuf chain is updated and
* a reference to the destination node is returned. If an error is
* encountered NULL is returned and the node reference will also be NULL.
*
* NB: The caller is responsible for freeing a returned node reference.
* The convention is ic_bss is not reference counted; the caller must
* maintain that.
*/
struct sk_buff *
ieee80211_encap(struct ieee80211_node *ni, struct sk_buff *skb, int *framecnt)
{
#define WH4(wh) ((struct ieee80211_frame_addr4 *)wh)
struct ieee80211vap *vap = ni->ni_vap;
struct ieee80211com *ic = ni->ni_ic;
struct ether_header eh;
struct ieee80211_frame *wh, *twh;
struct ieee80211_key *key;
struct llc *llc;
int hdrsize, datalen, addqos;
int hdrsize_nopad;
struct sk_buff *framelist = NULL;
struct sk_buff *tskb;
int fragcnt = 1;
int pdusize = 0;
int ismulticast = 0;
int use4addr = 0;
#ifdef ATH_SUPERG_FF
struct sk_buff *skb2 = NULL;
struct ether_header eh2;
int isff = ATH_FF_MAGIC_PRESENT(skb);
if (isff) {
#if 0
IEEE80211_DPRINTF(vap, IEEE80211_MSG_SUPG,
"%s: handling fast-frame skb (%p)\n", __func__, skb);
#endif
skb2 = skb->next;
if (skb2 == NULL) {
IEEE80211_DPRINTF(vap, IEEE80211_MSG_SUPG,
"%s: fast-frame error, only 1 skb\n", __func__);
goto bad;
}
memcpy(&eh2, skb2->data, sizeof(struct ether_header));
skb_pull(skb2, sizeof(struct ether_header));
}
#endif
memcpy(&eh, skb->data, sizeof(struct ether_header));
skb_pull(skb, sizeof(struct ether_header));
/*
* Ensure space for additional headers. First identify
* transmit key to use in calculating any buffer adjustments
* required. This is also used below to do privacy
* encapsulation work. Then calculate the 802.11 header
* size and any padding required by the driver.
*
* Note key may be NULL if we fall back to the default
* transmit key and that is not set. In that case the
* buffer may not be expanded as needed by the cipher
* routines, but they will/should discard it.
*/
if (vap->iv_flags & IEEE80211_F_PRIVACY) {
if (vap->iv_opmode == IEEE80211_M_STA ||
!IEEE80211_IS_MULTICAST(eh.ether_dhost))
key = ieee80211_crypto_getucastkey(vap, ni);
else
key = ieee80211_crypto_getmcastkey(vap, ni);
if ((key == NULL) && (eh.ether_type != htons(ETHERTYPE_PAE))) {
IEEE80211_NOTE_MAC(vap, IEEE80211_MSG_CRYPTO,
eh.ether_dhost,
"no default transmit key (%s) deftxkey %u",
__func__, vap->iv_def_txkey);
vap->iv_stats.is_tx_nodefkey++;
goto bad;
}
} else
key = NULL;
addqos = (ni->ni_flags & IEEE80211_NODE_QOS) &&
(eh.ether_type != htons(ETHERTYPE_PAE));
if (addqos)
hdrsize = sizeof(struct ieee80211_qosframe);
else
hdrsize = sizeof(struct ieee80211_frame);
switch (vap->iv_opmode) {
case IEEE80211_M_IBSS:
case IEEE80211_M_AHDEMO:
ismulticast = IEEE80211_IS_MULTICAST(eh.ether_dhost);
break;
case IEEE80211_M_WDS:
use4addr = 1;
ismulticast = IEEE80211_IS_MULTICAST(ni->ni_macaddr);
break;
case IEEE80211_M_HOSTAP:
if (!IEEE80211_IS_MULTICAST(eh.ether_dhost) &&
!IEEE80211_ADDR_EQ(eh.ether_dhost, ni->ni_macaddr)) {
use4addr = 1;
ismulticast = IEEE80211_IS_MULTICAST(ni->ni_macaddr);
} else
ismulticast = IEEE80211_IS_MULTICAST(eh.ether_dhost);
break;
case IEEE80211_M_STA:
if ((vap->iv_flags_ext & IEEE80211_FEXT_WDS) &&
!IEEE80211_ADDR_EQ(eh.ether_shost, vap->iv_myaddr)) {
use4addr = 1;
ismulticast = IEEE80211_IS_MULTICAST(ni->ni_macaddr);
/* Add a WDS entry to the station VAP */
if (IEEE80211_IS_MULTICAST(eh.ether_dhost)) {
struct ieee80211_node_table *nt = &ic->ic_sta;
struct ieee80211_node *ni_wds
= ieee80211_find_wds_node(nt, eh.ether_shost);
if (ni_wds)
ieee80211_unref_node(&ni_wds);
else
ieee80211_add_wds_addr(nt, ni, eh.ether_shost, 0);
}
} else
ismulticast = IEEE80211_IS_MULTICAST(vap->iv_bssid);
break;
default:
break;
}
if (use4addr)
hdrsize += IEEE80211_ADDR_LEN;
hdrsize_nopad = hdrsize;
if (ic->ic_flags & IEEE80211_F_DATAPAD)
hdrsize = roundup(hdrsize, sizeof(u_int32_t));
skb = ieee80211_skbhdr_adjust(vap, hdrsize, key, skb, ismulticast);
if (skb == NULL) {
/* NB: ieee80211_skbhdr_adjust handles msgs+statistics */
goto bad;
}
#ifdef ATH_SUPERG_FF
if (isff) {
struct ether_header *eh_inter;
struct athl2p_tunnel_hdr *ffhdr;
u_int16_t payload = skb->len + LLC_SNAPFRAMELEN;
int padded_len = payload + LLC_SNAPFRAMELEN + sizeof(struct ether_header);
/* in case header adjustments altered skb2 */
skb2 = skb->next;
if (skb2 == NULL) {
IEEE80211_DPRINTF(vap, IEEE80211_MSG_SUPG,
"%s: skb (%p) hdr adjust dropped 2nd skb\n",
__func__, skb);
goto bad;
}
/*
* add first skb tunnel hdrs
*/
llc = (struct llc *)skb_push(skb, LLC_SNAPFRAMELEN);
llc->llc_dsap = llc->llc_ssap = LLC_SNAP_LSAP;
llc->llc_control = LLC_UI;
llc->llc_snap.org_code[0] = 0;
llc->llc_snap.org_code[1] = 0;
llc->llc_snap.org_code[2] = 0;
llc->llc_snap.ether_type = eh.ether_type;
eh_inter = (struct ether_header *)skb_push(skb, sizeof(struct ether_header));
memcpy(eh_inter, &eh, sizeof(struct ether_header) - sizeof eh.ether_type);
eh_inter->ether_type = htons(payload);
/* overall ff encap header */
/* XXX: the offset of 2, below, should be computed. but... it will not
* practically ever change.
*/
ffhdr = (struct athl2p_tunnel_hdr *)skb_push(skb, sizeof(struct athl2p_tunnel_hdr) + 2);
memset(ffhdr, 0, sizeof(struct athl2p_tunnel_hdr) + 2);
/*
* add second skb tunnel hdrs
*/
payload = skb2->len + LLC_SNAPFRAMELEN;
llc = (struct llc *)skb_push(skb2, LLC_SNAPFRAMELEN);
if (llc == NULL) {
IEEE80211_DPRINTF(vap, IEEE80211_MSG_SUPG,
"%s: failed to push llc for 2nd skb (%p)\n",
__func__, skb);
return NULL;
}
llc->llc_dsap = llc->llc_ssap = LLC_SNAP_LSAP;
llc->llc_control = LLC_UI;
llc->llc_snap.org_code[0] = 0;
llc->llc_snap.org_code[1] = 0;
llc->llc_snap.org_code[2] = 0;
llc->llc_snap.ether_type = eh2.ether_type;
eh_inter = (struct ether_header *)skb_push(skb2, sizeof(struct ether_header));
if (eh_inter == NULL) {
IEEE80211_DPRINTF(vap, IEEE80211_MSG_SUPG,
"%s: failed to push eth hdr for 2nd skb (%p)\n",
__func__, skb);
return NULL;
}
memcpy(eh_inter, &eh2, sizeof(struct ether_header) - sizeof eh2.ether_type);
eh_inter->ether_type = htons(payload);
/* variable length pad */
skb_push(skb2, roundup(padded_len, 4) - padded_len);
}
#endif
llc = (struct llc *)skb_push(skb, LLC_SNAPFRAMELEN);
llc->llc_dsap = llc->llc_ssap = LLC_SNAP_LSAP;
llc->llc_control = LLC_UI;
#ifndef ATH_SUPERG_FF
llc->llc_snap.org_code[0] = 0;
llc->llc_snap.org_code[1] = 0;
llc->llc_snap.org_code[2] = 0;
llc->llc_snap.ether_type = eh.ether_type;
#else /* ATH_SUPERG_FF */
if (isff) {
llc->llc_snap.org_code[0] = ATH_SNAP_ORGCODE_0;
llc->llc_snap.org_code[1] = ATH_SNAP_ORGCODE_1;
llc->llc_snap.org_code[2] = ATH_SNAP_ORGCODE_2;
llc->llc_snap.ether_type = htons(ATH_ETH_TYPE);
} else {
llc->llc_snap.org_code[0] = 0;
llc->llc_snap.org_code[1] = 0;
llc->llc_snap.org_code[2] = 0;
llc->llc_snap.ether_type = eh.ether_type;
}
#endif /* ATH_SUPERG_FF */
datalen = skb->len; /* NB: w/o 802.11 header */
wh = (struct ieee80211_frame *)skb_push(skb, hdrsize);
wh->i_fc[0] = IEEE80211_FC0_VERSION_0 | IEEE80211_FC0_TYPE_DATA;
wh->i_dur = 0;
if (use4addr) {
wh->i_fc[1] = IEEE80211_FC1_DIR_DSTODS;
IEEE80211_ADDR_COPY(wh->i_addr1, ni->ni_macaddr);
IEEE80211_ADDR_COPY(wh->i_addr2, vap->iv_myaddr);
IEEE80211_ADDR_COPY(wh->i_addr3, eh.ether_dhost);
IEEE80211_ADDR_COPY(WH4(wh)->i_addr4, eh.ether_shost);
} else {
switch (vap->iv_opmode) {
case IEEE80211_M_IBSS:
case IEEE80211_M_AHDEMO:
wh->i_fc[1] = IEEE80211_FC1_DIR_NODS;
IEEE80211_ADDR_COPY(wh->i_addr1, eh.ether_dhost);
IEEE80211_ADDR_COPY(wh->i_addr2, eh.ether_shost);
/*
* NB: always use the bssid from iv_bssid as the
* neighbor's may be stale after an ibss merge
*/
IEEE80211_ADDR_COPY(wh->i_addr3, vap->iv_bssid);
break;
case IEEE80211_M_STA:
wh->i_fc[1] = IEEE80211_FC1_DIR_TODS;
IEEE80211_ADDR_COPY(wh->i_addr1, vap->iv_bssid);
IEEE80211_ADDR_COPY(wh->i_addr2, eh.ether_shost);
IEEE80211_ADDR_COPY(wh->i_addr3, eh.ether_dhost);
break;
case IEEE80211_M_HOSTAP:
wh->i_fc[1] = IEEE80211_FC1_DIR_FROMDS;
IEEE80211_ADDR_COPY(wh->i_addr1, eh.ether_dhost);
IEEE80211_ADDR_COPY(wh->i_addr2, vap->iv_bssid);
IEEE80211_ADDR_COPY(wh->i_addr3, eh.ether_shost);
if (M_PWR_SAV_GET(skb)) {
if (IEEE80211_NODE_SAVEQ_QLEN(ni)) {
wh->i_fc[1] |= IEEE80211_FC1_MORE_DATA;
M_PWR_SAV_CLR(skb);
}
}
break;
case IEEE80211_M_WDS:
wh->i_fc[1] = IEEE80211_FC1_DIR_DSTODS;
IEEE80211_ADDR_COPY(wh->i_addr1, ni->ni_macaddr);
IEEE80211_ADDR_COPY(wh->i_addr2, vap->iv_myaddr);
IEEE80211_ADDR_COPY(wh->i_addr3, eh.ether_dhost);
IEEE80211_ADDR_COPY(WH4(wh)->i_addr4, eh.ether_shost);
break;
case IEEE80211_M_MONITOR:
goto bad;
}
}
if (IEEE80211_VAP_IS_SLEEPING(vap))
wh->i_fc[1] |= IEEE80211_FC1_PWR_MGT;
if (addqos) {
struct ieee80211_qosframe *qwh =
(struct ieee80211_qosframe *)wh;
u_int8_t *qos;
int tid;
qos = &qwh->i_qos[0];
if (use4addr)
qos += IEEE80211_ADDR_LEN;
/* map from access class/queue to 11e header priority value */
tid = WME_AC_TO_TID(skb->priority);
qos[0] = tid & IEEE80211_QOS_TID;
if (ic->ic_wme.wme_wmeChanParams.cap_wmeParams[skb->priority].wmep_noackPolicy)
qos[0] |= (1 << IEEE80211_QOS_ACKPOLICY_S) & IEEE80211_QOS_ACKPOLICY;
qos[1] = 0;
qwh->i_fc[0] |= IEEE80211_FC0_SUBTYPE_QOS;
*(__le16 *)&wh->i_seq[0] =
htole16(ni->ni_txseqs[tid] << IEEE80211_SEQ_SEQ_SHIFT);
ni->ni_txseqs[tid]++;
} else {
*(__le16 *)wh->i_seq =
htole16(ni->ni_txseqs[0] << IEEE80211_SEQ_SEQ_SHIFT);
ni->ni_txseqs[0]++;
}
/* Is transmit fragmentation needed? */
if (skb->len > vap->iv_fragthreshold &&
!IEEE80211_IS_MULTICAST(wh->i_addr1)) {
int pktlen, skbcnt, tailsize, ciphdrsize;
struct ieee80211_cipher *cip;
pktlen = skb->len;
ciphdrsize = 0;
tailsize = IEEE80211_CRC_LEN;
if (key != NULL) {
cip = (struct ieee80211_cipher *)key->wk_cipher;
ciphdrsize = cip->ic_header;
tailsize += (cip->ic_trailer + cip->ic_miclen);
/* Add the 8 bytes MIC length. */
if (cip->ic_cipher == IEEE80211_CIPHER_TKIP)
pktlen += IEEE80211_WEP_MICLEN;
}
pdusize = vap->iv_fragthreshold - (hdrsize_nopad + ciphdrsize);
fragcnt = *framecnt =
((pktlen - hdrsize_nopad) / pdusize) +
(((pktlen - hdrsize_nopad) % pdusize == 0) ? 0 : 1);
/*
* Allocate sk_buff for each subsequent fragment; First fragment
* reuses input skb.
*/
for (skbcnt = 1; skbcnt < fragcnt; skbcnt++) {
This patch augments the current reference counting code with: * Counters for total outstanding instances for each resource type (skb, ath_node and ath_buf) * One pair of acquisition/release functions per resource type in unlocked and one in locked * Adds some more _debug versions of functions in the call chain that acquire/release resources so that the original func/line in the driver as well as the func/line that affected the resource use can be shown in the trace. Intermediate stack frames aren't necessary to trace the leaks. * Changes naming convention for "lock-required" functions to suffix _locked for the versions that expect locking, to be consistent with some other places in the code. * Consolidate debug messages to the helper functions that actually affect the reference count or acquire/release a resource * Additional sanity checks and leak detection (esp for detecting node ref leaks through skb) * skb references are nulled out by the new sbk unref/free function. I've tested these changes extensively and found lots of cases where we didn't get enough node references when cloning skbuff, and where the kernel drops packets due to performance issues and leaks our node references. With these changes and the tracing enabled I have verified that: * TX BUF: tx buffers always go down to zero when the tx queue is done, and you can watch tx queue usage ratio go up and down again as the driver is working. There are no leaks here at the moment, although there *are* some in the madwifi-dfs branch during CAC at the moment. * skbuff leaks in all the common flows are fixed. We were leaking node references in a lot of places where kernel was dropping skb's due to congestion and we were failing to increment node references when cloning skbuffs. These are now detected, as are skbuffs that are reaped by the kernel while still holding a node reference. * the ath_node count works correctly and on an idle system we get about 5 references per station table node, with 2 node instances per VAP. One for the bss and one for the node in the station table, I believe. The ath_node count goes up and down but always lands back at the stable number based on the vaps you have configured and the number of actual stations in the station table. The point here is that it's pretty constant what you will see over time, despite excessive node creation/release in our code during input (esp input_all). Thank god for the slab allocator. git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@2902 0192ed92-7a03-0410-a25b-9323aeb14dbd
2007-11-21 23:14:11 +03:00
tskb = ieee80211_dev_alloc_skb(hdrsize + ciphdrsize + pdusize + tailsize);
if (tskb == NULL)
break;
tskb->next = framelist;
framelist = tskb;
}
if (skbcnt != fragcnt)
goto bad;
}
else
*framecnt = fragcnt;
if (key != NULL) {
/*
* IEEE 802.1X: send EAPOL frames always in the clear.
* WPA/WPA2: encrypt EAPOL keys when pairwise keys are set.
*/
if (eh.ether_type != __constant_htons(ETHERTYPE_PAE) ||
((vap->iv_flags & IEEE80211_F_WPA) &&
(vap->iv_opmode == IEEE80211_M_STA ?
!KEY_UNDEFINED(*key) : !KEY_UNDEFINED(ni->ni_ucastkey)))) {
int force_swmic = (fragcnt > 1) ? 1 : 0;
wh->i_fc[1] |= IEEE80211_FC1_PROT;
if (!ieee80211_crypto_enmic(vap, key, skb, force_swmic)) {
IEEE80211_NOTE_MAC(vap, IEEE80211_MSG_OUTPUT,
eh.ether_dhost,
"%s", "enmic failed, discard frame");
vap->iv_stats.is_crypto_enmicfail++;
goto bad;
}
}
}
if (fragcnt > 1) {
int fragnum, offset, pdulen;
void *pdu;
fragnum = 0;
wh = twh = (struct ieee80211_frame *)skb->data;
/*
** Setup WLAN headers as fragment headers
*/
wh->i_fc[1] |= IEEE80211_FC1_MORE_FRAG;
*(__le16 *)&wh->i_seq[0] |=
htole16((fragnum & IEEE80211_SEQ_FRAG_MASK) <<
IEEE80211_SEQ_FRAG_SHIFT);
fragnum++;
offset = hdrsize + pdusize;
datalen = (skb->len - hdrsize) - pdusize;
IEEE80211_NODE_STAT(ni, tx_data);
IEEE80211_NODE_STAT_ADD(ni, tx_bytes, pdusize);
for (tskb = framelist; tskb != NULL; tskb = tskb->next) {
/*
* Copy WLAN header into each frag header skb
*/
twh = (struct ieee80211_frame *)skb_put(tskb, hdrsize);
memcpy((void *)twh, (void *)wh, hdrsize);
*(__le16 *)&twh->i_seq[0] |=
htole16((fragnum & IEEE80211_SEQ_FRAG_MASK) <<
IEEE80211_SEQ_FRAG_SHIFT);
fragnum++;
if (pdusize <= datalen)
pdulen = pdusize;
else
pdulen = datalen;
/*
* Copy fragment payload from input skb.
* Doing copies isn't intuitive from
* a performance perspective, however,
* for this case, it is believed to be
* more efficient than cloning skbs.
*/
pdu = skb_put(tskb, pdulen);
memcpy(pdu, (void *)skb->data + offset, pdulen);
offset += pdusize;
datalen -= pdusize;
IEEE80211_NODE_STAT(ni, tx_data);
IEEE80211_NODE_STAT_ADD(ni, tx_bytes, pdulen);
}
twh->i_fc[1] &= ~IEEE80211_FC1_MORE_FRAG;
skb_trim(skb, hdrsize + pdusize);
skb->next = framelist;
} else {
IEEE80211_NODE_STAT(ni, tx_data);
IEEE80211_NODE_STAT_ADD(ni, tx_bytes, datalen);
#ifdef ATH_SUPERG_FF
/* Account for a second skb in the same packet when FF is on */
if (skb->next) {
datalen = skb->next->len;
IEEE80211_NODE_STAT(ni, tx_data);
IEEE80211_NODE_STAT_ADD(ni, tx_bytes, datalen);
}
#endif
}
return skb;
bad:
if (framelist != NULL) {
This patch augments the current reference counting code with: * Counters for total outstanding instances for each resource type (skb, ath_node and ath_buf) * One pair of acquisition/release functions per resource type in unlocked and one in locked * Adds some more _debug versions of functions in the call chain that acquire/release resources so that the original func/line in the driver as well as the func/line that affected the resource use can be shown in the trace. Intermediate stack frames aren't necessary to trace the leaks. * Changes naming convention for "lock-required" functions to suffix _locked for the versions that expect locking, to be consistent with some other places in the code. * Consolidate debug messages to the helper functions that actually affect the reference count or acquire/release a resource * Additional sanity checks and leak detection (esp for detecting node ref leaks through skb) * skb references are nulled out by the new sbk unref/free function. I've tested these changes extensively and found lots of cases where we didn't get enough node references when cloning skbuff, and where the kernel drops packets due to performance issues and leaks our node references. With these changes and the tracing enabled I have verified that: * TX BUF: tx buffers always go down to zero when the tx queue is done, and you can watch tx queue usage ratio go up and down again as the driver is working. There are no leaks here at the moment, although there *are* some in the madwifi-dfs branch during CAC at the moment. * skbuff leaks in all the common flows are fixed. We were leaking node references in a lot of places where kernel was dropping skb's due to congestion and we were failing to increment node references when cloning skbuffs. These are now detected, as are skbuffs that are reaped by the kernel while still holding a node reference. * the ath_node count works correctly and on an idle system we get about 5 references per station table node, with 2 node instances per VAP. One for the bss and one for the node in the station table, I believe. The ath_node count goes up and down but always lands back at the stable number based on the vaps you have configured and the number of actual stations in the station table. The point here is that it's pretty constant what you will see over time, despite excessive node creation/release in our code during input (esp input_all). Thank god for the slab allocator. git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@2902 0192ed92-7a03-0410-a25b-9323aeb14dbd
2007-11-21 23:14:11 +03:00
ieee80211_dev_kfree_skb_list(&framelist);
}
if (skb != NULL) {
This patch augments the current reference counting code with: * Counters for total outstanding instances for each resource type (skb, ath_node and ath_buf) * One pair of acquisition/release functions per resource type in unlocked and one in locked * Adds some more _debug versions of functions in the call chain that acquire/release resources so that the original func/line in the driver as well as the func/line that affected the resource use can be shown in the trace. Intermediate stack frames aren't necessary to trace the leaks. * Changes naming convention for "lock-required" functions to suffix _locked for the versions that expect locking, to be consistent with some other places in the code. * Consolidate debug messages to the helper functions that actually affect the reference count or acquire/release a resource * Additional sanity checks and leak detection (esp for detecting node ref leaks through skb) * skb references are nulled out by the new sbk unref/free function. I've tested these changes extensively and found lots of cases where we didn't get enough node references when cloning skbuff, and where the kernel drops packets due to performance issues and leaks our node references. With these changes and the tracing enabled I have verified that: * TX BUF: tx buffers always go down to zero when the tx queue is done, and you can watch tx queue usage ratio go up and down again as the driver is working. There are no leaks here at the moment, although there *are* some in the madwifi-dfs branch during CAC at the moment. * skbuff leaks in all the common flows are fixed. We were leaking node references in a lot of places where kernel was dropping skb's due to congestion and we were failing to increment node references when cloning skbuffs. These are now detected, as are skbuffs that are reaped by the kernel while still holding a node reference. * the ath_node count works correctly and on an idle system we get about 5 references per station table node, with 2 node instances per VAP. One for the bss and one for the node in the station table, I believe. The ath_node count goes up and down but always lands back at the stable number based on the vaps you have configured and the number of actual stations in the station table. The point here is that it's pretty constant what you will see over time, despite excessive node creation/release in our code during input (esp input_all). Thank god for the slab allocator. git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@2902 0192ed92-7a03-0410-a25b-9323aeb14dbd
2007-11-21 23:14:11 +03:00
ieee80211_dev_kfree_skb_list(&skb);
}
return NULL;
#undef WH4
}
EXPORT_SYMBOL(ieee80211_encap);
#undef KEY_UNDEFINED
/*
* Add a supported rates element id to a frame.
*/
u_int8_t *
ieee80211_add_rates(u_int8_t *frm, const struct ieee80211_rateset *rs)
{
int nrates;
*frm++ = IEEE80211_ELEMID_RATES;
nrates = rs->rs_nrates;
if (nrates > IEEE80211_RATE_SIZE)
nrates = IEEE80211_RATE_SIZE;
*frm++ = nrates;
memcpy(frm, rs->rs_rates, nrates);
return frm + nrates;
}
/*
* Add an extended supported rates element id to a frame.
*/
u_int8_t *
ieee80211_add_xrates(u_int8_t *frm, const struct ieee80211_rateset *rs)
{
/*
* Add an extended supported rates element if operating in 11g mode.
*/
if (rs->rs_nrates > IEEE80211_RATE_SIZE) {
int nrates = rs->rs_nrates - IEEE80211_RATE_SIZE;
*frm++ = IEEE80211_ELEMID_XRATES;
*frm++ = nrates;
memcpy(frm, rs->rs_rates + IEEE80211_RATE_SIZE, nrates);
frm += nrates;
}
return frm;
}
/*
* Add an ssid elemet to a frame.
*/
static u_int8_t *
ieee80211_add_ssid(u_int8_t *frm, const u_int8_t *ssid, u_int len)
{
*frm++ = IEEE80211_ELEMID_SSID;
*frm++ = len;
memcpy(frm, ssid, len);
return frm + len;
}
/*
* Add an erp element to a frame.
*/
u_int8_t *
ieee80211_add_erp(u_int8_t *frm, struct ieee80211com *ic)
{
u_int8_t erp;
*frm++ = IEEE80211_ELEMID_ERP;
*frm++ = 1;
erp = 0;
if (ic->ic_nonerpsta != 0)
erp |= IEEE80211_ERP_NON_ERP_PRESENT;
if (ic->ic_flags & IEEE80211_F_USEPROT)
erp |= IEEE80211_ERP_USE_PROTECTION;
if ((ic->ic_flags & IEEE80211_F_USEBARKER) || (ic->ic_nonerpsta > 0))
erp |= IEEE80211_ERP_LONG_PREAMBLE;
*frm++ = erp;
return frm;
}
/*
* Add a country information element to a frame.
*/
u_int8_t *
ieee80211_add_country(u_int8_t *frm, struct ieee80211com *ic)
{
/* add country code */
memcpy(frm, (u_int8_t *)&ic->ic_country_ie,
ic->ic_country_ie.country_len + 2);
frm += ic->ic_country_ie.country_len + 2;
return frm;
}
/*
* Add Power Constraint information element
*/
u_int8_t *
ieee80211_add_pwrcnstr(u_int8_t *frm, struct ieee80211com *ic)
{
struct ieee80211_ie_pwrcnstr *ie =
(struct ieee80211_ie_pwrcnstr *)frm;
ie->pc_id = IEEE80211_ELEMID_PWRCNSTR;
ie->pc_len = 1;
ie->pc_lpc = IEEE80211_PWRCONSTRAINT_VAL(ic);
frm += sizeof(*ie);
return frm;
}
/*
* Add Power Capability information element
*/
static u_int8_t *
ieee80211_add_pwrcap(u_int8_t *frm, struct ieee80211com *ic)
{
struct ieee80211_ie_pwrcap *ie =
(struct ieee80211_ie_pwrcap *)frm;
ie->pc_id = IEEE80211_ELEMID_PWRCAP;
ie->pc_len = 2;
ie->pc_mintxpow = ic->ic_bsschan->ic_minpower;
ie->pc_maxtxpow = ic->ic_bsschan->ic_maxpower;
frm += sizeof(*ie);
return frm;
}
/*
* Add Supported Channels information element
*/
static u_int8_t *
ieee80211_add_suppchan(u_int8_t *frm, struct ieee80211com *ic)
{
memcpy(frm, (u_int8_t *)&ic->ic_sc_ie,
ic->ic_sc_ie.sc_len + 2);
frm += ic->ic_sc_ie.sc_len + 2;
return frm;
}
static u_int8_t *
ieee80211_setup_wpa_ie(struct ieee80211vap *vap, u_int8_t *ie)
{
#define WPA_OUI_BYTES 0x00, 0x50, 0xf2
#define ADDSHORT(frm, v) do { \
frm[0] = (v) & 0xff; \
frm[1] = (v) >> 8; \
frm += 2; \
} while (0)
#define ADDSELECTOR(frm, sel) do { \
memcpy(frm, sel, 4); \
frm += 4; \
} while (0)
static const u_int8_t oui[4] = { WPA_OUI_BYTES, WPA_OUI_TYPE };
static const u_int8_t cipher_suite[][4] = {
{ WPA_OUI_BYTES, WPA_CSE_WEP40 }, /* NB: 40-bit */
{ WPA_OUI_BYTES, WPA_CSE_TKIP },
{ 0x00, 0x00, 0x00, 0x00 }, /* XXX WRAP */
{ WPA_OUI_BYTES, WPA_CSE_CCMP },
{ 0x00, 0x00, 0x00, 0x00 }, /* XXX CKIP */
{ WPA_OUI_BYTES, WPA_CSE_NULL },
};
static const u_int8_t wep104_suite[4] =
{ WPA_OUI_BYTES, WPA_CSE_WEP104 };
static const u_int8_t key_mgt_unspec[4] =
{ WPA_OUI_BYTES, WPA_ASE_8021X_UNSPEC };
static const u_int8_t key_mgt_psk[4] =
{ WPA_OUI_BYTES, WPA_ASE_8021X_PSK };
const struct ieee80211_rsnparms *rsn = &vap->iv_bss->ni_rsn;
u_int8_t *frm = ie;
u_int8_t *selcnt;
*frm++ = IEEE80211_ELEMID_VENDOR;
*frm++ = 0; /* length filled in below */
memcpy(frm, oui, sizeof(oui)); /* WPA OUI */
frm += sizeof(oui);
ADDSHORT(frm, WPA_VERSION);
/* XXX filter out CKIP */
/* multicast cipher */
if (rsn->rsn_mcastcipher == IEEE80211_CIPHER_WEP &&
rsn->rsn_mcastkeylen >= 13)
ADDSELECTOR(frm, wep104_suite);
else
ADDSELECTOR(frm, cipher_suite[rsn->rsn_mcastcipher]);
/* unicast cipher list */
selcnt = frm;
ADDSHORT(frm, 0); /* selector count */
if (rsn->rsn_ucastcipherset & (1 << IEEE80211_CIPHER_AES_CCM)) {
selcnt[0]++;
ADDSELECTOR(frm, cipher_suite[IEEE80211_CIPHER_AES_CCM]);
}
if (rsn->rsn_ucastcipherset & (1 << IEEE80211_CIPHER_TKIP)) {
selcnt[0]++;
ADDSELECTOR(frm, cipher_suite[IEEE80211_CIPHER_TKIP]);
}
/* authenticator selector list */
selcnt = frm;
ADDSHORT(frm, 0); /* selector count */
if (rsn->rsn_keymgmtset & WPA_ASE_8021X_UNSPEC) {
selcnt[0]++;
ADDSELECTOR(frm, key_mgt_unspec);
}
if (rsn->rsn_keymgmtset & WPA_ASE_8021X_PSK) {
selcnt[0]++;
ADDSELECTOR(frm, key_mgt_psk);
}
/* optional capabilities */
if ((rsn->rsn_caps != 0) && (rsn->rsn_caps != RSN_CAP_PREAUTH))
ADDSHORT(frm, rsn->rsn_caps);
/* calculate element length */
ie[1] = frm - ie - 2;
KASSERT(ie[1] + 2 <= sizeof(struct ieee80211_ie_wpa),
("WPA IE too big, %u > %u",
ie[1] + 2, (int)sizeof(struct ieee80211_ie_wpa)));
return frm;
#undef ADDSHORT
#undef ADDSELECTOR
#undef WPA_OUI_BYTES
}
static u_int8_t *
ieee80211_setup_rsn_ie(struct ieee80211vap *vap, u_int8_t *ie)
{
#define RSN_OUI_BYTES 0x00, 0x0f, 0xac
#define ADDSHORT(frm, v) do { \
frm[0] = (v) & 0xff; \
frm[1] = (v) >> 8; \
frm += 2; \
} while (0)
#define ADDSELECTOR(frm, sel) do { \
memcpy(frm, sel, 4); \
frm += 4; \
} while (0)
static const u_int8_t cipher_suite[][4] = {
{ RSN_OUI_BYTES, RSN_CSE_WEP40 }, /* NB: 40-bit */
{ RSN_OUI_BYTES, RSN_CSE_TKIP },
{ RSN_OUI_BYTES, RSN_CSE_WRAP },
{ RSN_OUI_BYTES, RSN_CSE_CCMP },
{ 0x00, 0x00, 0x00, 0x00 }, /* XXX CKIP */
{ RSN_OUI_BYTES, RSN_CSE_NULL },
};
static const u_int8_t wep104_suite[4] =
{ RSN_OUI_BYTES, RSN_CSE_WEP104 };
static const u_int8_t key_mgt_unspec[4] =
{ RSN_OUI_BYTES, RSN_ASE_8021X_UNSPEC };
static const u_int8_t key_mgt_psk[4] =
{ RSN_OUI_BYTES, RSN_ASE_8021X_PSK };
const struct ieee80211_rsnparms *rsn = &vap->iv_bss->ni_rsn;
u_int8_t *frm = ie;
u_int8_t *selcnt;
*frm++ = IEEE80211_ELEMID_RSN;
*frm++ = 0; /* length filled in below */
ADDSHORT(frm, RSN_VERSION);
/* XXX filter out CKIP */
/* multicast cipher */
if (rsn->rsn_mcastcipher == IEEE80211_CIPHER_WEP &&
rsn->rsn_mcastkeylen >= 13)
ADDSELECTOR(frm, wep104_suite);
else
ADDSELECTOR(frm, cipher_suite[rsn->rsn_mcastcipher]);
/* unicast cipher list */
selcnt = frm;
ADDSHORT(frm, 0); /* selector count */
if (rsn->rsn_ucastcipherset & (1 << IEEE80211_CIPHER_AES_CCM)) {
selcnt[0]++;
ADDSELECTOR(frm, cipher_suite[IEEE80211_CIPHER_AES_CCM]);
}
if (rsn->rsn_ucastcipherset & (1 << IEEE80211_CIPHER_TKIP)) {
selcnt[0]++;
ADDSELECTOR(frm, cipher_suite[IEEE80211_CIPHER_TKIP]);
}
/* authenticator selector list */
selcnt = frm;
ADDSHORT(frm, 0); /* selector count */
if (rsn->rsn_keymgmtset & WPA_ASE_8021X_UNSPEC) {
selcnt[0]++;
ADDSELECTOR(frm, key_mgt_unspec);
}
if (rsn->rsn_keymgmtset & WPA_ASE_8021X_PSK) {
selcnt[0]++;
ADDSELECTOR(frm, key_mgt_psk);
}
/* capabilities */
ADDSHORT(frm, rsn->rsn_caps);
/* XXX PMKID */
/* calculate element length */
ie[1] = frm - ie - 2;
KASSERT(ie[1] + 2 <= sizeof(struct ieee80211_ie_wpa),
("RSN IE too big, %u > %u",
ie[1] + 2, (int)sizeof(struct ieee80211_ie_wpa)));
return frm;
#undef ADDSELECTOR
#undef ADDSHORT
#undef RSN_OUI_BYTES
}
/*
* Add a WPA/RSN element to a frame.
*/
u_int8_t *
ieee80211_add_wpa(u_int8_t *frm, struct ieee80211vap *vap)
{
KASSERT(vap->iv_flags & IEEE80211_F_WPA, ("no WPA/RSN!"));
if (vap->iv_flags & IEEE80211_F_WPA2)
frm = ieee80211_setup_rsn_ie(vap, frm);
if (vap->iv_flags & IEEE80211_F_WPA1)
frm = ieee80211_setup_wpa_ie(vap, frm);
return frm;
}
#define WME_OUI_BYTES 0x00, 0x50, 0xf2
/*
* Add a WME Info element to a frame.
*/
static u_int8_t *
ieee80211_add_wme(u_int8_t *frm, struct ieee80211_node *ni)
{
static const u_int8_t oui[4] = { WME_OUI_BYTES, WME_OUI_TYPE };
struct ieee80211_ie_wme *ie = (struct ieee80211_ie_wme *)frm;
struct ieee80211_wme_state *wme = &ni->ni_ic->ic_wme;
struct ieee80211vap *vap = ni->ni_vap;
*frm++ = IEEE80211_ELEMID_VENDOR;
*frm++ = 0; /* length filled in below */
memcpy(frm, oui, sizeof(oui)); /* WME OUI */
frm += sizeof(oui);
*frm++ = WME_INFO_OUI_SUBTYPE; /* OUI subtype */
*frm++ = WME_VERSION; /* protocol version */
/* QoS Info field depends on operating mode */
switch (vap->iv_opmode) {
case IEEE80211_M_HOSTAP:
*frm = wme->wme_bssChanParams.cap_info_count;
if (IEEE80211_VAP_UAPSD_ENABLED(vap))
*frm |= WME_CAPINFO_UAPSD_EN;
frm++;
break;
case IEEE80211_M_STA:
*frm++ = vap->iv_uapsdinfo;
break;
default:
*frm++ = 0;
}
ie->wme_len = frm - &ie->wme_oui[0];
return frm;
}
/*
* Add a WME Parameter element to a frame.
*/
u_int8_t *
ieee80211_add_wme_param(u_int8_t *frm, struct ieee80211_wme_state *wme,
int uapsd_enable)
{
#define SM(_v, _f) (((_v) << _f##_S) & _f)
#define ADDSHORT(frm, v) do { \
frm[0] = (v) & 0xff; \
frm[1] = (v) >> 8; \
frm += 2; \
} while (0)
static const u_int8_t oui[4] = { WME_OUI_BYTES, WME_OUI_TYPE };
struct ieee80211_wme_param *ie = (struct ieee80211_wme_param *)frm;
int i;
*frm++ = IEEE80211_ELEMID_VENDOR;
*frm++ = 0; /* length filled in below */
memcpy(frm, oui, sizeof(oui)); /* WME OUI */
frm += sizeof(oui);
*frm++ = WME_PARAM_OUI_SUBTYPE; /* OUI subtype */
*frm++ = WME_VERSION; /* protocol version */
*frm = wme->wme_bssChanParams.cap_info_count;
if (uapsd_enable)
*frm |= WME_CAPINFO_UAPSD_EN;
frm++;
*frm++ = 0; /* reserved field */
for (i = 0; i < WME_NUM_AC; i++) {
const struct wmeParams *ac =
&wme->wme_bssChanParams.cap_wmeParams[i];
*frm++ = SM(i, WME_PARAM_ACI) |
SM(ac->wmep_acm, WME_PARAM_ACM) |
SM(ac->wmep_aifsn, WME_PARAM_AIFSN);
*frm++ = SM(ac->wmep_logcwmax, WME_PARAM_LOGCWMAX) |
SM(ac->wmep_logcwmin, WME_PARAM_LOGCWMIN);
ADDSHORT(frm, ac->wmep_txopLimit);
}
ie->param_len = frm - &ie->param_oui[0];
return frm;
#undef ADDSHORT
}
#undef WME_OUI_BYTES
/*
* Add an Atheros Advanaced Capability element to a frame
*/
u_int8_t *
ieee80211_add_athAdvCap(u_int8_t *frm, u_int8_t capability, u_int16_t defaultKey)
{
static const u_int8_t oui[6] = {(ATH_OUI & 0xff), ((ATH_OUI >>8) & 0xff),
((ATH_OUI >> 16) & 0xff), ATH_OUI_TYPE,
ATH_OUI_SUBTYPE, ATH_OUI_VERSION};
struct ieee80211_ie_athAdvCap *ie = (struct ieee80211_ie_athAdvCap *)frm;
*frm++ = IEEE80211_ELEMID_VENDOR;
*frm++ = 0; /* Length filled in below */
memcpy(frm, oui, sizeof(oui)); /* Atheros OUI, type, subtype, and version for adv capabilities */
frm += sizeof(oui);
*frm++ = capability;
/* Setup default key index in little endian byte order */
*frm++ = (defaultKey & 0xff);
*frm++ = ((defaultKey >> 8) & 0xff);
ie->athAdvCap_len = frm - &ie->athAdvCap_oui[0];
return frm;
}
/*
* Add XR IE element to a frame
*/
#ifdef ATH_SUPERG_XR
u_int8_t *
ieee80211_add_xr_param(u_int8_t *frm, struct ieee80211vap *vap)
{
static const u_int8_t oui[6] = {(ATH_OUI & 0xff), ((ATH_OUI >>8) & 0xff),
((ATH_OUI >> 16) & 0xff), ATH_OUI_TYPE_XR,
ATH_OUI_SUBTYPE_XR, ATH_OUI_VER_XR};
struct ieee80211_xr_param *ie = (struct ieee80211_xr_param *)frm;
*frm++ = IEEE80211_ELEMID_VENDOR;
*frm++ = 0; /* Length filled in below */
memcpy(frm, oui, sizeof(oui)); /* Atheros OUI, type, subtype, and version for adv capabilities */
frm += sizeof(oui);
*frm++ = 0; /* XR info */
/* copy the BSSIDs */
if (vap->iv_flags & IEEE80211_F_XR) {
IEEE80211_ADDR_COPY(frm, vap->iv_xrvap->iv_bssid); /* Base BSSID */
frm += IEEE80211_ADDR_LEN;
IEEE80211_ADDR_COPY(frm, vap->iv_bssid); /* XR BSSID */
frm += IEEE80211_ADDR_LEN;
*(__le16 *)frm = htole16(vap->iv_bss->ni_intval); /* XR beacon interval */
frm += 2;
*frm++ = vap->iv_xrvap->iv_ath_cap; /* Base mode capability */
*frm++ = vap->iv_ath_cap; /* XR mode capability */
} else {
IEEE80211_ADDR_COPY(frm, vap->iv_bssid);
frm += IEEE80211_ADDR_LEN;
IEEE80211_ADDR_COPY(frm, vap->iv_xrvap->iv_bssid);
frm += IEEE80211_ADDR_LEN;
*(__le16 *)frm = htole16(vap->iv_bss->ni_intval);
frm += 2;
*frm++ = vap->iv_ath_cap;
*frm++ = vap->iv_xrvap->iv_ath_cap;
}
ie->param_len = frm - &ie->param_oui[0];
return frm;
}
#endif
/*
* Send a probe request frame with the specified ssid
* and any optional information element data.
*/
int
ieee80211_send_probereq(struct ieee80211_node *ni,
const u_int8_t sa[IEEE80211_ADDR_LEN],
const u_int8_t da[IEEE80211_ADDR_LEN],
const u_int8_t bssid[IEEE80211_ADDR_LEN],
const u_int8_t *ssid, size_t ssidlen,
const void *optie, size_t optielen)
{
struct ieee80211vap *vap = ni->ni_vap;
struct ieee80211com *ic = ni->ni_ic;
enum ieee80211_phymode mode;
struct ieee80211_frame *wh;
struct sk_buff *skb;
u_int8_t *frm;
/*
* prreq frame format
* [tlv] ssid
* [tlv] supported rates
* [tlv] extended supported rates
* [tlv] user-specified IEs
*/
skb = ieee80211_getmgtframe(&frm, 2 + IEEE80211_NWID_LEN +
2 + IEEE80211_RATE_SIZE +
2 + (IEEE80211_RATE_MAXSIZE - IEEE80211_RATE_SIZE) +
(optie != NULL ? optielen : 0) +
vap->app_ie[IEEE80211_APPIE_FRAME_PROBE_REQ].length);
if (skb == NULL) {
vap->iv_stats.is_tx_nobuf++;
return -ENOMEM;
}
frm = ieee80211_add_ssid(frm, ssid, ssidlen);
mode = ieee80211_chan2mode(ic->ic_curchan);
frm = ieee80211_add_rates(frm, &ic->ic_sup_rates[mode]);
frm = ieee80211_add_xrates(frm, &ic->ic_sup_rates[mode]);
if (optie != NULL) {
memcpy(frm, optie, optielen);
frm += optielen;
}
if (vap->app_ie[IEEE80211_APPIE_FRAME_PROBE_REQ].ie) {
memcpy(frm, vap->app_ie[IEEE80211_APPIE_FRAME_PROBE_REQ].ie,
vap->app_ie[IEEE80211_APPIE_FRAME_PROBE_REQ].length);
frm += vap->app_ie[IEEE80211_APPIE_FRAME_PROBE_REQ].length;
}
skb_trim(skb, frm - skb->data);
SKB_NI(skb) = ieee80211_ref_node(ni);
wh = (struct ieee80211_frame *)
skb_push(skb, sizeof(struct ieee80211_frame));
ieee80211_send_setup(vap, ni, wh,
IEEE80211_FC0_TYPE_MGT | IEEE80211_FC0_SUBTYPE_PROBE_REQ,
sa, da, bssid);
/* XXX power management? */
IEEE80211_NODE_STAT(ni, tx_probereq);
IEEE80211_NODE_STAT(ni, tx_mgmt);
IEEE80211_DPRINTF(vap, IEEE80211_MSG_DEBUG | IEEE80211_MSG_DUMPPKTS,
"[" MAC_FMT "] send probe req on channel %u\n",
MAC_ADDR(wh->i_addr1),
ieee80211_chan2ieee(ic, ic->ic_curchan));
(void)ic->ic_mgtstart(ic, skb);
return 0;
}
/* Start a new Channel Switch process. It will first check if there is already
* one Channel Switch process running and if so, will determine which one will
* run. This function must be the only function setting IEEE80211_F_CHANSWITCH
* in ic_flags.
*
* is_beacon_frame : true if the csa_count comes from a beacon frame we just
* received. */
void
ieee80211_start_new_csa(struct ieee80211vap *vap,
u_int8_t csa_mode,
struct ieee80211_channel *csa_chan,
u_int8_t csa_count,
int is_beacon_frame)
{
struct ieee80211com *ic = vap->iv_ic;
u_int32_t now_tu, nexttbtt, expires_tu;
unsigned long now, expires;
/* 802.11h 7.3.2.20 : A value of 1 indicates that the switch will
* occur immediately before the next TBTT. A value of 0 indicates that
* the switch will occur at any time after the frame containing the
* element is transmitted. */
now_tu = IEEE80211_TSF_TO_TU(vap->iv_get_tsf(vap));
now = jiffies;
if (csa_count == 0) {
expires_tu = now_tu;
expires = now;
IEEE80211_DPRINTF(vap, IEEE80211_MSG_DOTH,
"%s: now:%d count:%d => expires:%d\n",
__func__, now, csa_count, expires);
} else {
/* csa_count includes the current frame if it is a beacon
* frame. */
if (is_beacon_frame)
csa_count --;
/* Compute the closest nexttbtt, next time a beacon for this
* VAP will be sent. */
nexttbtt = vap->iv_get_nexttbtt(vap);
/* Compute ic_csa_expires_tu = nexttbtt + csa_count *
* ni_intval. */
expires_tu = nexttbtt + csa_count * vap->iv_bss->ni_intval;
IEEE80211_DPRINTF(vap, IEEE80211_MSG_DOTH,
"%s: now_tu:%ld nexttbtt_tu:%ld "
"=> expires_tu:%ld\n",
__func__, now_tu, nexttbtt,
expires_tu);
/* Convert to jiffies, including a margin. */
expires = now +
IEEE80211_TU_TO_JIFFIES(expires_tu - now_tu - 10) - 1;
IEEE80211_DPRINTF(vap, IEEE80211_MSG_DOTH,
"%s: now:%d count:%d => expires:%d\n",
__func__, now, csa_count, expires);
}
/* If we have a CS in progress, we ignore this new CSA IE if the
* channel switch time is later than the current one. */
if ((ic->ic_flags & IEEE80211_F_CHANSWITCH) &&
(expires_tu > ic->ic_csa_expires_tu)) {
/* We do not ignore csa_mode if it says we must stop sending
* right now. */
if (ic->ic_csa_mode == IEEE80211_CSA_CAN_STOP_TX &&
csa_mode == IEEE80211_CSA_MUST_STOP_TX) {
IEEE80211_DPRINTF(vap, IEEE80211_MSG_DOTH,
"%s: Updating CSA mode\n",
__func__);
ic->ic_csa_mode = csa_mode;
}
IEEE80211_DPRINTF(vap, IEEE80211_MSG_DOTH,
"%s: Ignored CSA IE since a sooner "
"channel switch is scheduled\n", __func__);
return;
}
ic->ic_csa_mode = csa_mode;
ic->ic_csa_chan = csa_chan;
ic->ic_csa_expires_tu = expires_tu;
mod_timer(&ic->ic_csa_timer, expires);
ic->ic_flags |= IEEE80211_F_CHANSWITCH;
}
/* Send a broadcast CSA frame, announcing the new channel. References are from
* IEEE 802.11h-2003. CSA frame format is an "Action" frame (Type: 00, Subtype:
* 1101, see 7.1.3.1.2)
*
* [1] Category : 0, Spectrum Management, 7.3.1.11
* [1] Action : 4, Channel Switch Announcement, 7.4.1 and 7.4.1.5
* [1] Element ID : 37, Channel Switch Announcement, 7.3.2
* [1] Length : 3, 7.3.2.20
* [1] Channel Switch Mode : 1, stop transmission immediately
* [1] New Channel Number
* [1] Channel Switch Count in TBTT : 0, immediate channel switch
*
* csa_mode : IEEE80211_CSA_MANDATORY / IEEE80211_CSA_ADVISORY
* csa_chan : new IEEE channel number
* csa_tbtt : TBTT until Channel Switch happens */
void
ieee80211_send_csa_frame(struct ieee80211vap *vap,
u_int8_t csa_mode,
u_int8_t csa_chan,
u_int8_t csa_count)
{
struct ieee80211_node *ni = vap->iv_bss;
struct ieee80211com *ic = ni->ni_ic;
struct sk_buff *skb;
const int frm_len = 7;
u_int8_t *frm;
IEEE80211_DPRINTF(vap, IEEE80211_MSG_DOTH,
"%s: Sending action frame with CSA IE: %u/%u/%u\n",
__func__, csa_mode, csa_chan, csa_count);
skb = ieee80211_getmgtframe(&frm, frm_len);
if (skb == NULL) {
IEEE80211_NOTE(vap, IEEE80211_MSG_ANY, ni,
"%s: cannot get buf; size %u", __func__, frm_len);
vap->iv_stats.is_tx_nobuf++;
return ;
}
*frm++ = IEEE80211_ACTION_SPECTRUM_MANAGEMENT; /* Category */
*frm++ = IEEE80211_ACTION_S_CHANSWITCHANN; /* Spectrum Management */
*frm++ = IEEE80211_ELEMID_CHANSWITCHANN;
*frm++ = 3;
*frm++ = csa_mode;
*frm++ = csa_chan;
*frm++ = csa_count;
ieee80211_mgmt_output(ieee80211_ref_node(ni), skb,
IEEE80211_FC0_SUBTYPE_ACTION,
ic->ic_dev->broadcast);
}
/*
* Send a management frame. The node is for the destination (or ic_bss
* when in station mode). Nodes other than ic_bss have their reference
* count bumped to reflect our use for an indeterminate time.
*/
int
ieee80211_send_mgmt(struct ieee80211_node *ni, int type, int arg)
{
#define senderr(_x, _v) do { vap->iv_stats._v++; ret = _x; goto bad; } while (0)
struct ieee80211vap *vap = ni->ni_vap;
struct ieee80211com *ic = ni->ni_ic;
struct sk_buff *skb;
u_int8_t *frm;
int frm_len;
u_int16_t capinfo;
ieee80211_keyix_t def_keyindex;
int has_challenge, is_shared_key, ret, timer, status;
KASSERT(ni != NULL, ("null node"));
timer = 0;
switch (type) {
case IEEE80211_FC0_SUBTYPE_PROBE_RESP:
/*
* probe response frame format
* [8] time stamp
* [2] beacon interval
* [2] capability information
* [tlv] ssid
* [tlv] supported rates
* [7] FH/DS parameter set
* [tlv] IBSS parameter set
* [tlv] country code
* [3] power constraint
* [3] extended rate phy (ERP)
* [tlv] extended supported rates
* [tlv] WME parameters
* [tlv] WPA/RSN parameters
* [tlv] Atheros Advanced Capabilities
* [tlv] AtherosXR parameters
*/
frm_len = 8
+ sizeof(u_int16_t)
+ sizeof(u_int16_t)
+ 2 + IEEE80211_NWID_LEN
+ 2 + IEEE80211_RATE_SIZE
+ 7 /* max(7,3) */
/* XXX allocate max size */
+ 2 + ic->ic_country_ie.country_len
+ 3
+ 3
+ 2 + (IEEE80211_RATE_MAXSIZE - IEEE80211_RATE_SIZE)
+ sizeof(struct ieee80211_wme_param)
/* XXX !WPA1+WPA2 fits w/o a cluster */
+ (vap->iv_flags & IEEE80211_F_WPA ?
2 * sizeof(struct ieee80211_ie_wpa) : 0)
+ sizeof(struct ieee80211_ie_athAdvCap)
+ vap->app_ie[IEEE80211_APPIE_FRAME_PROBE_RESP].length;
#ifdef ATH_SUPERG_XR
if (vap->iv_ath_cap & IEEE80211_ATHC_XR)
frm_len += sizeof(struct ieee80211_xr_param);
#endif
skb = ieee80211_getmgtframe(&frm, frm_len);
if (skb == NULL)
senderr(ENOMEM, is_tx_nobuf);
/* timestamp should be filled later */
memset(frm, 0, 8);
frm += 8;
/* beacon interval */
*(__le16 *)frm = htole16(vap->iv_bss->ni_intval);
frm += 2;
/* cap. info */
if (vap->iv_opmode == IEEE80211_M_IBSS)
capinfo = IEEE80211_CAPINFO_IBSS;
else
capinfo = IEEE80211_CAPINFO_ESS;
if (vap->iv_flags & IEEE80211_F_PRIVACY)
capinfo |= IEEE80211_CAPINFO_PRIVACY;
if ((ic->ic_flags & IEEE80211_F_SHPREAMBLE) &&
IEEE80211_IS_CHAN_2GHZ(ic->ic_curchan))
capinfo |= IEEE80211_CAPINFO_SHORT_PREAMBLE;
if (ic->ic_flags & IEEE80211_F_SHSLOT)
capinfo |= IEEE80211_CAPINFO_SHORT_SLOTTIME;
*(__le16 *)frm = htole16(capinfo);
frm += 2;
/* ssid */
frm = ieee80211_add_ssid(frm, vap->iv_bss->ni_essid,
vap->iv_bss->ni_esslen);
/* supported rates */
frm = ieee80211_add_rates(frm, &ni->ni_rates);
/* XXX: FH/DS parameter set, correct ? */
if (ic->ic_phytype == IEEE80211_T_FH) {
*frm++ = IEEE80211_ELEMID_FHPARMS;
*frm++ = 5;
*frm++ = ni->ni_fhdwell & 0x00ff;
*frm++ = (ni->ni_fhdwell >> 8) & 0x00ff;
*frm++ = IEEE80211_FH_CHANSET(
ieee80211_chan2ieee(ic, ic->ic_curchan));
*frm++ = IEEE80211_FH_CHANPAT(
ieee80211_chan2ieee(ic, ic->ic_curchan));
*frm++ = ni->ni_fhindex;
} else {
*frm++ = IEEE80211_ELEMID_DSPARMS;
*frm++ = 1;
*frm++ = ieee80211_chan2ieee(ic, ic->ic_curchan);
}
if (vap->iv_opmode == IEEE80211_M_IBSS) {
*frm++ = IEEE80211_ELEMID_IBSSPARMS;
*frm++ = 2;
*frm++ = 0;
*frm++ = 0; /* TODO: ATIM window */
}
/* country code */
if ((ic->ic_flags & IEEE80211_F_DOTH) ||
(ic->ic_flags_ext & IEEE80211_FEXT_COUNTRYIE))
frm = ieee80211_add_country(frm, ic);
/* power constraint */
if (ic->ic_flags & IEEE80211_F_DOTH)
frm = ieee80211_add_pwrcnstr(frm, ic);
/* ERP */
if (IEEE80211_IS_CHAN_ANYG(ic->ic_curchan))
frm = ieee80211_add_erp(frm, ic);
/* Ext. Supp. Rates */
frm = ieee80211_add_xrates(frm, &ni->ni_rates);
/* WME */
if (vap->iv_flags & IEEE80211_F_WME)
frm = ieee80211_add_wme_param(frm, &ic->ic_wme,
IEEE80211_VAP_UAPSD_ENABLED(vap));
/* WPA */
if (vap->iv_flags & IEEE80211_F_WPA)
frm = ieee80211_add_wpa(frm, vap);
/* AthAdvCaps */
if (vap->iv_bss && vap->iv_bss->ni_ath_flags)
frm = ieee80211_add_athAdvCap(frm, vap->iv_bss->ni_ath_flags,
vap->iv_bss->ni_ath_defkeyindex);
#ifdef ATH_SUPERG_XR
/* XR params */
if (vap->iv_xrvap && vap->iv_ath_cap & IEEE80211_ATHC_XR)
frm = ieee80211_add_xr_param(frm, vap);
#endif
if (vap->app_ie[IEEE80211_APPIE_FRAME_PROBE_RESP].ie) {
memcpy(frm, vap->app_ie[IEEE80211_APPIE_FRAME_PROBE_RESP].ie,
vap->app_ie[IEEE80211_APPIE_FRAME_PROBE_RESP].length);
frm += vap->app_ie[IEEE80211_APPIE_FRAME_PROBE_RESP].length;
}
skb_trim(skb, frm - skb->data);
break;
case IEEE80211_FC0_SUBTYPE_AUTH:
status = arg >> 16;
arg &= 0xffff;
has_challenge = ((arg == IEEE80211_AUTH_SHARED_CHALLENGE ||
arg == IEEE80211_AUTH_SHARED_RESPONSE) &&
ni->ni_challenge != NULL);
/*
* Deduce whether we're doing open authentication or
* shared key authentication. We do the latter if
* we're in the middle of a shared key authentication
* handshake or if we're initiating an authentication
* request and configured to use shared key.
*/
is_shared_key = has_challenge ||
arg >= IEEE80211_AUTH_SHARED_RESPONSE ||
(arg == IEEE80211_AUTH_SHARED_REQUEST &&
vap->iv_bss->ni_authmode == IEEE80211_AUTH_SHARED);
skb = ieee80211_getmgtframe(&frm,
3 * sizeof(u_int16_t)
+ (has_challenge && status == IEEE80211_STATUS_SUCCESS ?
sizeof(u_int16_t)+IEEE80211_CHALLENGE_LEN : 0));
if (skb == NULL)
senderr(ENOMEM, is_tx_nobuf);
((__le16 *)frm)[0] =
(is_shared_key) ? htole16(IEEE80211_AUTH_ALG_SHARED)
: htole16(IEEE80211_AUTH_ALG_OPEN);
((__le16 *)frm)[1] = htole16(arg); /* sequence number */
((__le16 *)frm)[2] = htole16(status); /* status */
if (has_challenge && status == IEEE80211_STATUS_SUCCESS) {
((__le16 *)frm)[3] =
htole16((IEEE80211_CHALLENGE_LEN << 8) |
IEEE80211_ELEMID_CHALLENGE);
memcpy(&((__le16 *)frm)[4], ni->ni_challenge,
IEEE80211_CHALLENGE_LEN);
if (arg == IEEE80211_AUTH_SHARED_RESPONSE) {
IEEE80211_NOTE(vap, IEEE80211_MSG_AUTH, ni,
"request encrypt frame (%s)", __func__);
SKB_CB(skb)->flags |= M_LINK0; /* WEP-encrypt, please */
}
}
/* XXX not right for shared key */
if (status == IEEE80211_STATUS_SUCCESS)
IEEE80211_NODE_STAT(ni, tx_auth);
else
IEEE80211_NODE_STAT(ni, tx_auth_fail);
if (vap->iv_opmode == IEEE80211_M_STA)
timer = IEEE80211_TRANS_WAIT;
break;
case IEEE80211_FC0_SUBTYPE_DEAUTH:
IEEE80211_NOTE(vap, IEEE80211_MSG_AUTH, ni,
"send station deauthenticate (reason %d)", arg);
skb = ieee80211_getmgtframe(&frm, sizeof(u_int16_t));
if (skb == NULL)
senderr(ENOMEM, is_tx_nobuf);
*(__le16 *)frm = htole16(arg); /* reason */
IEEE80211_NODE_STAT(ni, tx_deauth);
IEEE80211_NODE_STAT_SET(ni, tx_deauth_code, arg);
ieee80211_node_unauthorize(ni); /* port closed */
break;
case IEEE80211_FC0_SUBTYPE_ASSOC_REQ:
case IEEE80211_FC0_SUBTYPE_REASSOC_REQ:
/*
* asreq frame format
* [2] capability information
* [2] listen interval
* [6*] current AP address (reassoc only)
* [tlv] ssid
* [tlv] supported rates
* [4] power capability (802.11h)
This patch augments the current reference counting code with: * Counters for total outstanding instances for each resource type (skb, ath_node and ath_buf) * One pair of acquisition/release functions per resource type in unlocked and one in locked * Adds some more _debug versions of functions in the call chain that acquire/release resources so that the original func/line in the driver as well as the func/line that affected the resource use can be shown in the trace. Intermediate stack frames aren't necessary to trace the leaks. * Changes naming convention for "lock-required" functions to suffix _locked for the versions that expect locking, to be consistent with some other places in the code. * Consolidate debug messages to the helper functions that actually affect the reference count or acquire/release a resource * Additional sanity checks and leak detection (esp for detecting node ref leaks through skb) * skb references are nulled out by the new sbk unref/free function. I've tested these changes extensively and found lots of cases where we didn't get enough node references when cloning skbuff, and where the kernel drops packets due to performance issues and leaks our node references. With these changes and the tracing enabled I have verified that: * TX BUF: tx buffers always go down to zero when the tx queue is done, and you can watch tx queue usage ratio go up and down again as the driver is working. There are no leaks here at the moment, although there *are* some in the madwifi-dfs branch during CAC at the moment. * skbuff leaks in all the common flows are fixed. We were leaking node references in a lot of places where kernel was dropping skb's due to congestion and we were failing to increment node references when cloning skbuffs. These are now detected, as are skbuffs that are reaped by the kernel while still holding a node reference. * the ath_node count works correctly and on an idle system we get about 5 references per station table node, with 2 node instances per VAP. One for the bss and one for the node in the station table, I believe. The ath_node count goes up and down but always lands back at the stable number based on the vaps you have configured and the number of actual stations in the station table. The point here is that it's pretty constant what you will see over time, despite excessive node creation/release in our code during input (esp input_all). Thank god for the slab allocator. git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@2902 0192ed92-7a03-0410-a25b-9323aeb14dbd
2007-11-21 23:14:11 +03:00
* [tlv] supported channels element (802.11h)
* [tlv] extended supported rates
* [tlv] WME [if enabled and AP capable]
* [tlv] Atheros advanced capabilities
* [tlv] user-specified IEs
*/
skb = ieee80211_getmgtframe(&frm,
sizeof(u_int16_t) +
sizeof(u_int16_t) +
IEEE80211_ADDR_LEN +
2 + IEEE80211_NWID_LEN +
2 + IEEE80211_RATE_SIZE +
4 + (2 + ic->ic_sc_ie.sc_len) +
2 + (IEEE80211_RATE_MAXSIZE - IEEE80211_RATE_SIZE) +
sizeof(struct ieee80211_ie_wme) +
sizeof(struct ieee80211_ie_athAdvCap) +
(vap->iv_opt_ie != NULL ? vap->iv_opt_ie_len : 0) +
vap->app_ie[IEEE80211_APPIE_FRAME_ASSOC_REQ].length);
if (skb == NULL)
senderr(ENOMEM, is_tx_nobuf);
capinfo = 0;
if (vap->iv_opmode == IEEE80211_M_IBSS)
capinfo |= IEEE80211_CAPINFO_IBSS;
else /* IEEE80211_M_STA */
capinfo |= IEEE80211_CAPINFO_ESS;
if (vap->iv_flags & IEEE80211_F_PRIVACY)
capinfo |= IEEE80211_CAPINFO_PRIVACY;
/*
* NB: Some 11a APs reject the request when
* short premable is set.
*/
/* Capability information */
if ((ic->ic_flags & IEEE80211_F_SHPREAMBLE) &&
IEEE80211_IS_CHAN_2GHZ(ic->ic_curchan))
capinfo |= IEEE80211_CAPINFO_SHORT_PREAMBLE;
if ((ni->ni_capinfo & IEEE80211_CAPINFO_SHORT_SLOTTIME) &&
(ic->ic_caps & IEEE80211_C_SHSLOT))
capinfo |= IEEE80211_CAPINFO_SHORT_SLOTTIME;
*(__le16 *)frm = htole16(capinfo);
frm += 2;
/* listen interval */
*(__le16 *)frm = htole16(ic->ic_lintval);
frm += 2;
/* Current AP address */
if (type == IEEE80211_FC0_SUBTYPE_REASSOC_REQ) {
IEEE80211_ADDR_COPY(frm, vap->iv_bssid);
frm += IEEE80211_ADDR_LEN;
}
/* ssid */
frm = ieee80211_add_ssid(frm, ni->ni_essid, ni->ni_esslen);
/* supported rates */
frm = ieee80211_add_rates(frm, &ni->ni_rates);
/* power capability/supported channels */
if (ic->ic_flags & IEEE80211_F_DOTH) {
frm = ieee80211_add_pwrcap(frm, ic);
frm = ieee80211_add_suppchan(frm, ic);
}
/* ext. supp. rates */
frm = ieee80211_add_xrates(frm, &ni->ni_rates);
/* wme */
if ((vap->iv_flags & IEEE80211_F_WME) && ni->ni_wme_ie != NULL)
frm = ieee80211_add_wme(frm, ni);
/* ath adv. cap */
if (ni->ni_ath_flags & vap->iv_ath_cap) {
IEEE80211_NOTE(vap, IEEE80211_MSG_ASSOC, ni,
"Adding ath adv cap ie: ni_ath_flags = %02x, "
"iv_ath_cap = %02x", ni->ni_ath_flags,
vap->iv_ath_cap);
/* Setup default key index for static wep case */
def_keyindex = IEEE80211_INVAL_DEFKEY;
if (((vap->iv_flags & IEEE80211_F_WPA) == 0) &&
(ni->ni_authmode != IEEE80211_AUTH_8021X) &&
(vap->iv_def_txkey != IEEE80211_KEYIX_NONE))
def_keyindex = vap->iv_def_txkey;
frm = ieee80211_add_athAdvCap(frm,
ni->ni_ath_flags & vap->iv_ath_cap,
def_keyindex);
}
/* User-spec */
if (vap->iv_opt_ie != NULL) {
memcpy(frm, vap->iv_opt_ie, vap->iv_opt_ie_len);
frm += vap->iv_opt_ie_len;
}
if (vap->app_ie[IEEE80211_APPIE_FRAME_ASSOC_REQ].ie) {
memcpy(frm, vap->app_ie[IEEE80211_APPIE_FRAME_ASSOC_REQ].ie,
vap->app_ie[IEEE80211_APPIE_FRAME_ASSOC_REQ].length);
frm += vap->app_ie[IEEE80211_APPIE_FRAME_ASSOC_REQ].length;
}
skb_trim(skb, frm - skb->data);
timer = IEEE80211_TRANS_WAIT;
break;
case IEEE80211_FC0_SUBTYPE_ASSOC_RESP:
case IEEE80211_FC0_SUBTYPE_REASSOC_RESP:
/*
* asreq frame format
* [2] capability information
* [2] status
* [2] association ID
* [tlv] supported rates
* [tlv] extended supported rates
* [tlv] WME (if enabled and STA enabled)
* [tlv] Atheros Advanced Capabilities
*/
skb = ieee80211_getmgtframe(&frm,
3 * sizeof(u_int16_t) +
2 + IEEE80211_RATE_SIZE +
2 + (IEEE80211_RATE_MAXSIZE - IEEE80211_RATE_SIZE) +
sizeof(struct ieee80211_wme_param) +
(vap->iv_ath_cap ? sizeof(struct ieee80211_ie_athAdvCap):0) +
vap->app_ie[IEEE80211_APPIE_FRAME_ASSOC_RESP].length);
if (skb == NULL)
senderr(ENOMEM, is_tx_nobuf);
/* Capability Information */
capinfo = IEEE80211_CAPINFO_ESS;
if (vap->iv_flags & IEEE80211_F_PRIVACY)
capinfo |= IEEE80211_CAPINFO_PRIVACY;
if ((ic->ic_flags & IEEE80211_F_SHPREAMBLE) &&
IEEE80211_IS_CHAN_2GHZ(ic->ic_curchan))
capinfo |= IEEE80211_CAPINFO_SHORT_PREAMBLE;
if (ic->ic_flags & IEEE80211_F_SHSLOT)
capinfo |= IEEE80211_CAPINFO_SHORT_SLOTTIME;
*(__le16 *)frm = htole16(capinfo);
frm += 2;
/* status */
*(__le16 *)frm = htole16(arg);
frm += 2;
/* Assoc ID */
if (arg == IEEE80211_STATUS_SUCCESS) {
*(__le16 *)frm = htole16(ni->ni_associd);
IEEE80211_NODE_STAT(ni, tx_assoc);
} else
IEEE80211_NODE_STAT(ni, tx_assoc_fail);
frm += 2;
/* supported rates */
frm = ieee80211_add_rates(frm, &ni->ni_rates);
/* ext. suppo. rates */
frm = ieee80211_add_xrates(frm, &ni->ni_rates);
/* wme */
if ((vap->iv_flags & IEEE80211_F_WME) && ni->ni_wme_ie != NULL)
frm = ieee80211_add_wme_param(frm, &ic->ic_wme,
IEEE80211_VAP_UAPSD_ENABLED(vap));
/* athAdvCap */
if (vap->iv_ath_cap)
frm = ieee80211_add_athAdvCap(frm,
vap->iv_ath_cap & ni->ni_ath_flags,
ni->ni_ath_defkeyindex);
if (vap->app_ie[IEEE80211_APPIE_FRAME_ASSOC_RESP].ie) {
memcpy(frm, vap->app_ie[IEEE80211_APPIE_FRAME_ASSOC_RESP].ie,
vap->app_ie[IEEE80211_APPIE_FRAME_ASSOC_RESP].length);
frm += vap->app_ie[IEEE80211_APPIE_FRAME_ASSOC_RESP].length;
}
skb_trim(skb, frm - skb->data);
break;
case IEEE80211_FC0_SUBTYPE_DISASSOC:
IEEE80211_NOTE(vap, IEEE80211_MSG_ASSOC, ni,
"send station disassociate (reason %d)", arg);
skb = ieee80211_getmgtframe(&frm, sizeof(u_int16_t));
if (skb == NULL)
senderr(ENOMEM, is_tx_nobuf);
*(__le16 *)frm = htole16(arg); /* reason */
IEEE80211_NODE_STAT(ni, tx_disassoc);
IEEE80211_NODE_STAT_SET(ni, tx_disassoc_code, arg);
break;
default:
IEEE80211_NOTE(vap, IEEE80211_MSG_ANY, ni,
"invalid mgmt frame type %u", type);
senderr(EINVAL, is_tx_unknownmgt);
/* NOTREACHED */
}
ieee80211_mgmt_output(ieee80211_ref_node(ni), skb, type,
ni->ni_macaddr);
if (timer)
mod_timer(&vap->iv_mgtsend, jiffies + timer * HZ);
return 0;
bad:
return ret;
#undef senderr
}
/*
* Send PS-POLL from to bss. Should only be called when as STA.
*/
void
ieee80211_send_pspoll(struct ieee80211_node *ni)
{
struct ieee80211vap *vap = ni->ni_vap;
struct ieee80211com *ic = ni->ni_ic;
struct sk_buff *skb;
struct ieee80211_ctlframe_addr2 *wh;
This patch augments the current reference counting code with: * Counters for total outstanding instances for each resource type (skb, ath_node and ath_buf) * One pair of acquisition/release functions per resource type in unlocked and one in locked * Adds some more _debug versions of functions in the call chain that acquire/release resources so that the original func/line in the driver as well as the func/line that affected the resource use can be shown in the trace. Intermediate stack frames aren't necessary to trace the leaks. * Changes naming convention for "lock-required" functions to suffix _locked for the versions that expect locking, to be consistent with some other places in the code. * Consolidate debug messages to the helper functions that actually affect the reference count or acquire/release a resource * Additional sanity checks and leak detection (esp for detecting node ref leaks through skb) * skb references are nulled out by the new sbk unref/free function. I've tested these changes extensively and found lots of cases where we didn't get enough node references when cloning skbuff, and where the kernel drops packets due to performance issues and leaks our node references. With these changes and the tracing enabled I have verified that: * TX BUF: tx buffers always go down to zero when the tx queue is done, and you can watch tx queue usage ratio go up and down again as the driver is working. There are no leaks here at the moment, although there *are* some in the madwifi-dfs branch during CAC at the moment. * skbuff leaks in all the common flows are fixed. We were leaking node references in a lot of places where kernel was dropping skb's due to congestion and we were failing to increment node references when cloning skbuffs. These are now detected, as are skbuffs that are reaped by the kernel while still holding a node reference. * the ath_node count works correctly and on an idle system we get about 5 references per station table node, with 2 node instances per VAP. One for the bss and one for the node in the station table, I believe. The ath_node count goes up and down but always lands back at the stable number based on the vaps you have configured and the number of actual stations in the station table. The point here is that it's pretty constant what you will see over time, despite excessive node creation/release in our code during input (esp input_all). Thank god for the slab allocator. git-svn-id: http://madwifi-project.org/svn/madwifi/trunk@2902 0192ed92-7a03-0410-a25b-9323aeb14dbd
2007-11-21 23:14:11 +03:00
skb = ieee80211_dev_alloc_skb(sizeof(struct ieee80211_ctlframe_addr2));
if (skb == NULL) return;
SKB_NI(skb) = ieee80211_ref_node(ni);
skb->priority = WME_AC_VO;
wh = (struct ieee80211_ctlframe_addr2 *)skb_put(skb, sizeof(struct ieee80211_ctlframe_addr2));
wh->i_aidordur = htole16(0xc000 | IEEE80211_NODE_AID(ni));
IEEE80211_ADDR_COPY(wh->i_addr1, vap->iv_bssid);
IEEE80211_ADDR_COPY(wh->i_addr2, vap->iv_myaddr);
wh->i_fc[0] = 0;
wh->i_fc[1] = 0;
wh->i_fc[0] = IEEE80211_FC0_VERSION_0 | IEEE80211_FC0_TYPE_CTL |
IEEE80211_FC0_SUBTYPE_PS_POLL;
if (IEEE80211_VAP_IS_SLEEPING(ni->ni_vap))
wh->i_fc[1] |= IEEE80211_FC1_PWR_MGT;
(void) ic->ic_mgtstart(ic, skb); /* cheat */
}
#ifdef ATH_SUPERG_XR
/*
* constructs and returns a contention free frames.
* currently used for Group poll in XR mode.
*/
struct sk_buff *
ieee80211_getcfframe(struct ieee80211vap *vap, int type)
{
u_int8_t *frm;
struct sk_buff *skb;
struct ieee80211_frame *wh;
struct ieee80211com *ic = vap->iv_ic;
skb = ieee80211_getmgtframe(&frm, 0);
if (skb == NULL)
return NULL;
wh = (struct ieee80211_frame *)
skb_push(skb, sizeof(struct ieee80211_frame));
if (type == IEEE80211_FC0_SUBTYPE_CFPOLL) {
wh->i_fc[1] = IEEE80211_FC1_DIR_FROMDS;
wh->i_fc[0] = IEEE80211_FC0_VERSION_0 | IEEE80211_FC0_TYPE_DATA | type;
wh->i_dur = htole16(0x8000);
} else if (type == IEEE80211_FC0_SUBTYPE_CF_END) {
wh->i_fc[1] = IEEE80211_FC1_DIR_NODS;
wh->i_fc[0] = IEEE80211_FC0_VERSION_0 | IEEE80211_FC0_TYPE_CTL | type;
wh->i_dur = 0;
}
IEEE80211_ADDR_COPY(wh->i_addr1, ic->ic_dev->broadcast);
IEEE80211_ADDR_COPY(wh->i_addr2, vap->iv_myaddr);
IEEE80211_ADDR_COPY(wh->i_addr3, vap->iv_bssid);
return skb;
}
EXPORT_SYMBOL(ieee80211_getcfframe);
#endif