wireless network namespace work
Subsystems (3)
Johannes Berg
maintainercover wireless network namespace work
Hi,
This series implements most of the net namespace awareness for wireless (well, those bits that davem doesn't already have).
This depends on the following commits from net-next-2.6:
commit 30ffee8480c13fbcf8ab6c28e31f79dfff683117
Author: Johannes Berg <[email protected]>
Date: Fri Jul 10 09:51:35 2009 +0000
net: move and export get_net_ns_by_pid
commit 134e63756d5f3d0f7604dfcca847b09d1b14fd66
Author: Johannes Berg <[email protected]>
Date: Fri Jul 10 09:51:34 2009 +0000
genetlink: make netns aware
commit 11a28d373ed2539a110d56419457e2e7db221ac7
Author: Johannes Berg <[email protected]>
Date: Fri Jul 10 09:51:33 2009 +0000
net: make namespace iteration possible under RCU
commit 6c04bb18ddd633b7feac2c8fe2ae0bf61d20ca7a
Author: Johannes Berg <[email protected]>
Date: Fri Jul 10 09:51:32 2009 +0000
netlink: use call_rcu for netlink_change_ngroups
(well, it only depends on the first and second, but the second in turn depends on the other two)
Not sure -- can you cherry-pick these and things just work?
johannes
Johannes Berg
maintainerPATCH 1/3 mac80211: cooperate more with network namespaces
There are still two places in mac80211 that hardcode the initial net namespace (init_net). One of them is mandated by cfg80211 and will be removed by a separate patch, the other one is used for finding the network device of a pending packet via its ifindex.
Remove the latter use by keeping track of the device pointer itself, via the vif pointer, and avoid it going stale by dropping pending frames for a given interface when the interface is removed.
To keep track of the vif pointer for the correct interface, change the info->control.vif pointer's internal use to always be the correct vif, and only move it to the vif the driver expects (or NULL for monitor interfaces and injected packets) right before giving the packet to the driver.
--- net/mac80211/iface.c | 16 ++++++++++ net/mac80211/rx.c | 2 - net/mac80211/tx.c | 75 ++++++++++++++++++++------------------------------- 3 files changed, 46 insertions(+), 47 deletions(-)
net/mac80211/iface.c +15 -1
@@ -335,7 +335,10 @@ static int ieee80211_stop(struct net_devstruct ieee80211_local *local = sdata->local;struct ieee80211_if_init_conf conf;struct sta_info *sta; +unsigned long flags; +struct sk_buff *skb, *tmp;u32 hw_reconf_flags =0 ; +int i;/* * Stop TX on this interface first.
@@ -551,6 +554,18 @@ static int ieee80211_stop(struct net_devif (hw_reconf_flags)ieee80211_hw_config (local, hw_reconf_flags); +spin_lock_irqsave (&local->queue_stop_reason_lock, flags); +for (i =0 ; i <IEEE80211_MAX_QUEUES ; i++) { + skb_queue_walk_safe(&local->pending[i], skb, tmp) { +struct ieee80211_tx_info *info =IEEE80211_SKB_CB (skb); +if (info->control.vif == &sdata->vif) { +__skb_unlink (skb, &local->pending[i]); +dev_kfree_skb_irq (skb); + } + } + } +spin_unlock_irqrestore (&local->queue_stop_reason_lock, flags); +return 0 ; }
@@ -788,7 +803,6 @@ int ieee80211_if_add(struct ieee80211_lomemcpy (ndev->dev_addr, local->hw.wiphy->perm_addr,ETH_ALEN );SET_NETDEV_DEV (ndev,wiphy_dev (local->hw.wiphy)); - ndev->features |=NETIF_F_NETNS_LOCAL ;/* don't use IEEE80211_DEV_TO_SUB_IF because it checks too much */ sdata =netdev_priv (ndev);
net/mac80211/tx.c +30 -45
175 lines not on this page · raw message
net/mac80211/rx.c +1 -1
@@ -1539,7 +1539,7 @@ ieee80211_rx_h_mesh_fwding(struct ieee80 info =IEEE80211_SKB_CB (fwd_skb);memset (info,0 ,sizeof (*info)); info->flags |=IEEE80211_TX_INTFL_NEED_TXPROCESSING ; - fwd_skb->iif = rx->dev->ifindex; + info->control.vif = &rx->sdata->vif;ieee80211_select_queue (local, fwd_skb);if (is_multicast_ether_addr (fwd_hdr->addr3)) memcpy(fwd_hdr->addr1, fwd_hdr->addr3,
Johannes Berg
maintainerPATCH 2/3 cfg80211: make aware of net namespaces
In reply to Johannes BergIn order to make cfg80211/nl80211 aware of network namespaces, we have to do the following things:
* del_virtual_intf method takes an interface index rather than a netdev pointer - simply change this
* nl80211 uses init_net a lot, it changes to use the sender's network namespace
* scan requests use the interface index, hold a netdev pointer and reference instead
* we want a wiphy and its associated virtual interfaces to be in one netns together, so
- we need to be able to change ns for a given interface, so export dev_change_net_namespace()
- for each virtual interface set the NETIF_F_NETNS_LOCAL flag, and clear that flag only when the wiphy changes ns, to disallow breaking this invariant
* when a network namespace goes away, we need to reparent the wiphy to init_net
* cfg80211 users that support creating virtual interfaces must create them in the wiphy's namespace, currently this affects only mac80211
The end result is that you can now switch an entire wiphy into a different network namespace with the new command
iw phy#<idx> set netns <pid>
and all virtual interfaces will follow (or the operation fails).
--- include/linux/nl80211.h | 9 ++ include/net/cfg80211.h | 40 ++++++++- net/core/dev.c | 1 net/mac80211/cfg.c | 14 --- net/wireless/core.c | 75 ++++++++++++++++- net/wireless/core.h | 5 - net/wireless/nl80211.c | 202 +++++++++++++++++++++++++++++++++--------------- net/wireless/scan.c | 22 +---- net/wireless/sme.c | 3 9 files changed, 272 insertions(+), 99 deletions(-)
include/net/cfg80211.h +36 -4
@@ -542,7 +542,7 @@ struct cfg80211_ssid {* @ie: optional information element(s) to add into Probe Request or %NULL * @ie_len: length of ie in octets * @wiphy: the wiphy this was for -* @ifidx: the interface index +* @dev: the interface */ struct cfg80211_scan_request {struct cfg80211_ssid *ssids;
@@ -554,7 +554,7 @@ struct cfg80211_scan_request {/* internal */ struct wiphy *wiphy; -int ifidx; +struct net_device *dev;bool aborted; };
@@ -845,7 +845,8 @@ struct cfg80211_bitrate_mask {* @resume: wiphy device needs to be resumed * * @add_virtual_intf: create a new virtual interface with the given name, -* must set the struct wireless_dev's iftype. +* must set the struct wireless_dev's iftype. Beware: You must create +* the new netdev in the wiphy's network namespace! * * @del_virtual_intf: remove the virtual interface determined by ifindex. *
@@ -937,7 +938,7 @@ struct cfg80211_ops {int (*add_virtual_intf)(struct wiphy *wiphy,char *name,enum nl80211_iftype type,u32 *flags,struct vif_params *params); -int (*del_virtual_intf)(struct wiphy *wiphy,int ifindex); +int (*del_virtual_intf)(struct wiphy *wiphy,struct net_device *dev); int (*change_virtual_intf)(struct wiphy *wiphy, struct net_device *dev, enum nl80211_iftype type, u32 *flags,
@@ -1088,6 +1089,9 @@ struct cfg80211_ops {* @frag_threshold: Fragmentation threshold (dot11FragmentationThreshold); * -1 = fragmentation disabled, only odd values >= 256 used * @rts_threshold: RTS threshold (dot11RTSThreshold); -1 = RTS/CTS disabled +* @net: the network namespace this wiphy currently lives in +* @netnsok: if set to false, do not allow changing the netns of this +* wiphy at all */ struct wiphy {/* assign these fields before you register the wiphy */
@@ -1101,6 +1105,8 @@ struct wiphy {bool custom_regulatory;bool strict_regulatory; +bool netnsok; +enum cfg80211_signal_type signal_type;int bss_priv_size;
@@ -1139,9 +1145,35 @@ struct wiphy {/* dir in debugfs: ieee80211/<wiphyname> */ struct dentry *debugfsdir; +#ifdef CONFIG_NET_NS +/* the network namespace this phy lives in currently */ +struct net *_net; +#endif + char priv[0] __attribute__((__aligned__(NETDEV_ALIGN))); }; +#ifdef CONFIG_NET_NS +static inline struct net *wiphy_net (struct wiphy *wiphy) +{ +return wiphy->_net; +} + +static inline void wiphy_net_set (struct wiphy *wiphy,struct net *net) +{ + wiphy->_net = net; +} +#else +static inline struct net *wiphy_net (struct wiphy *wiphy) +{ +return &init_net; +} + +static inline void wiphy_net_set (struct wiphy *wiphy,struct net *net) +{ +} +#endif +/** * wiphy_priv - return priv from wiphy *
net/mac80211/cfg.c +2 -12
@@ -57,19 +57,9 @@ static int ieee80211_add_iface(struct wireturn 0 ; } -static int ieee80211_del_iface (struct wiphy *wiphy,int ifindex) +static int ieee80211_del_iface (struct wiphy *wiphy,struct net_device *dev) { -struct net_device *dev; -struct ieee80211_sub_if_data *sdata; - -/* we're under RTNL */ - dev =__dev_get_by_index (&init_net, ifindex); -if (!dev) -return -ENODEV ; - - sdata =IEEE80211_DEV_TO_SUB_IF (dev); - -ieee80211_if_remove (sdata); +ieee80211_if_remove (IEEE80211_DEV_TO_SUB_IF (dev));return 0 ; }
net/wireless/nl80211.c +141 -61
595 lines not on this page · raw message
net/wireless/core.c +70 -5
132 lines not on this page · raw message
net/wireless/scan.c +7 -15
@@ -32,9 +32,7 @@ void __cfg80211_scan_done(struct work_stmutex_lock (&rdev->mtx); request = rdev->scan_req; - dev =dev_get_by_index (&init_net, request->ifidx); -if (!dev) - goto out; + dev = request->dev;/* * This must be before sending the other events!
@@ -60,24 +58,16 @@ void __cfg80211_scan_done(struct work_stdev_put (dev); - out:cfg80211_unlock_rdev (rdev);kfree (request); }void cfg80211_scan_done (struct cfg80211_scan_request *request,bool aborted) { -struct net_device *dev =dev_get_by_index (&init_net, request->ifidx); -if (WARN_ON (!dev)) { -kfree (request); -return ; - } -WARN_ON (request !=wiphy_to_dev (request->wiphy)->scan_req); request->aborted = aborted;schedule_work (&wiphy_to_dev (request->wiphy)->scan_done_wk); -dev_put (dev); }EXPORT_SYMBOL (cfg80211_scan_done);
@@ -593,7 +583,7 @@ int cfg80211_wext_siwscan(struct net_devif (!netif_running (dev))return -ENETDOWN ; - rdev =cfg80211_get_dev_from_ifindex (dev->ifindex); + rdev =cfg80211_get_dev_from_ifindex (dev_net (dev), dev->ifindex);if (IS_ERR (rdev))return PTR_ERR (rdev);
@@ -618,7 +608,7 @@ int cfg80211_wext_siwscan(struct net_dev } creq->wiphy = wiphy; - creq->ifidx = dev->ifindex; + creq->dev = dev; creq->ssids = (void *)(creq +1 ); creq->channels = (void *)(creq->ssids +1 ); creq->n_channels = n_channels;
@@ -655,8 +645,10 @@ int cfg80211_wext_siwscan(struct net_devif (err) { rdev->scan_req =NULL ;kfree (creq); - }else + }else {nl80211_send_scan_start (rdev, dev); +dev_hold (dev); + } out:cfg80211_unlock_rdev (rdev);return err;
@@ -949,7 +941,7 @@ int cfg80211_wext_giwscan(struct net_devif (!netif_running (dev))return -ENETDOWN ; - rdev =cfg80211_get_dev_from_ifindex (dev->ifindex); + rdev =cfg80211_get_dev_from_ifindex (dev_net (dev), dev->ifindex);if (IS_ERR (rdev))return PTR_ERR (rdev);
include/linux/nl80211.h +9 -0
@@ -262,6 +262,9 @@* reasons, for this the %NL80211_ATTR_DISCONNECTED_BY_AP and * %NL80211_ATTR_REASON_CODE attributes are used. * +* @NL80211_CMD_SET_WIPHY_NETNS: Set a wiphy's netns. Note that all devices +* associated with this wiphy must be down and will follow. +* * @NL80211_CMD_MAX: highest used command number * @__NL80211_CMD_AFTER_LAST: internal use */
@@ -336,6 +339,8 @@ enum nl80211_commands {NL80211_CMD_ROAM ,NL80211_CMD_DISCONNECT , + NL80211_CMD_SET_WIPHY_NETNS, +/* add new commands above here */ /* used to define NL80211_CMD_MAX below */
@@ -573,6 +578,8 @@ enum nl80211_commands {* and join_ibss(), key information is in a nested attribute each * with %NL80211_KEY_* sub-attributes * +* @NL80211_ATTR_PID: Process ID of a network namespace. +* * @NL80211_ATTR_MAX: highest attribute number currently defined * @__NL80211_ATTR_AFTER_LAST: internal use */
@@ -701,6 +708,8 @@ enum nl80211_attrs {NL80211_ATTR_KEY ,NL80211_ATTR_KEYS , +NL80211_ATTR_PID , +/* add attributes here, update the policy in nl80211.c */ __NL80211_ATTR_AFTER_LAST,
net/core/dev.c +1 -0
@@ -5344,6 +5344,7 @@ int dev_change_net_namespace(struct net_ out:return err; } +EXPORT_SYMBOL_GPL (dev_change_net_namespace); static int dev_cpu_callback(struct notifier_block *nfb, unsigned long action,
net/wireless/sme.c +2 -1
@@ -86,7 +86,7 @@ static int cfg80211_conn_scan(struct wir wdev->conn->params.ssid_len); request->ssids[0 ].ssid_len = wdev->conn->params.ssid_len; - request->ifidx = wdev->netdev->ifindex; + request->dev = wdev->netdev; request->wiphy = &rdev->wiphy; rdev->scan_req = request;
@@ -95,6 +95,7 @@ static int cfg80211_conn_scan(struct wirif (!err) { wdev->conn->state =CFG80211_CONN_SCANNING ;nl80211_send_scan_start (rdev, wdev->netdev); +dev_hold (wdev->netdev); }else { rdev->scan_req =NULL ;kfree (request);
net/wireless/core.h +4 -1
@@ -170,7 +170,10 @@ struct wiphy *wiphy_idx_to_wiphy(int wip/* identical to cfg80211_get_dev_from_info but only operate on ifindex */ extern struct cfg80211_registered_device * -cfg80211_get_dev_from_ifindex (int ifindex); +cfg80211_get_dev_from_ifindex (struct net *net,int ifindex); + +int cfg80211_switch_netns (struct cfg80211_registered_device *rdev, +struct net *net);static inline void cfg80211_lock_rdev (struct cfg80211_registered_device *rdev) {
Johannes Berg
maintainerPATCH 3/3 mac80211: allow using network namespaces
In reply to Johannes BergThis finally opens up the ability to put mac80211 devices into different network namespaces. As long as you don't have sysfs, that is.
---
Semantically, this depends on both
mac80211: cooperate more with network namespaces
and
cfg80211: make aware of net namespaces
but in the code only the latter dependency is obvious, so I'm sending it as a separate patch.
net/mac80211/iface.c | 1 + net/mac80211/main.c | 1 + 2 files changed, 2 insertions(+)
net/mac80211/main.c +1 -0
@@ -620,6 +620,7 @@ struct ieee80211_hw *ieee80211_alloc_hw(if (!wiphy)return NULL ; + wiphy->netnsok = true; wiphy->privid = mac80211_wiphy_privid;/* Yes, putting cfg80211_bss into ieee80211_bss is a hack */
net/mac80211/iface.c +1 -0
@@ -787,6 +787,7 @@ int ieee80211_if_add(struct ieee80211_lo name, ieee80211_if_setup);if (!ndev)return -ENOMEM ; +dev_net_set (ndev,wiphy_net (local->hw.wiphy)); ndev->needed_headroom = local->tx_headroom +4 *6 /* four MAC addresses */
Johannes Berg
maintainer[HACK] mac80211: don't set netdev parent pointer
In reply to Johannes BergTo allow playing with network namespaces even if you have sysfs configured... don't try this on your normal kernel, it will probably confuse userspace.
--- net/mac80211/iface.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
net/mac80211/iface.c +1 -1
@@ -803,7 +803,7 @@ int ieee80211_if_add(struct ieee80211_lo goto fail;memcpy (ndev->dev_addr, local->hw.wiphy->perm_addr,ETH_ALEN ); -SET_NETDEV_DEV (ndev,wiphy_dev (local->hw.wiphy)); +// SET_NETDEV_DEV(ndev, wiphy_dev(local->hw.wiphy)); /* don't use IEEE80211_DEV_TO_SUB_IF because it checks too much */ sdata =netdev_priv (ndev);
Re: [PATCH 1/3] mac80211: cooperate more with network namespaces
In reply to Johannes Bergnet/mac80211/tx.c void ieee80211_tx_pending(unsigned long · line 1887
Quoted 9 lines · Johannes Berg
net/mac80211/tx.c +5
-0@@ -1900,10 +1880,16 @@ void ieee80211_tx_pending(unsigned longwhile (!skb_queue_empty (&local->pending[i])) {struct sk_buff *skb =__skb_dequeue (&local->pending[i]); +struct ieee80211_tx_info *info =IEEE80211_SKB_CB (skb); +struct ieee80211_sub_if_data *sdata; + + sdata =vif_to_sdata (info->control.vif); +dev_hold (sdata->dev);
I'm getting a panic at this point if I run hostapd on ath9k and a Windows client tries to authenticate with a wrong WPA key.
Debugging shows that sdata->dev is NULL.
The current wireless-testing is affected. The revision preceding to this patch is not affected.
CONFIG_NET_NS is not enabled. The architecture is x86_64.
Signature · 2 lines
Regards, Pavel Roskin
Johannes Berg
maintainerRe: [PATCH 1/3] mac80211: cooperate more with network namespaces
Thanks Pavel,
Quoted 11 lines
net/mac80211/tx.c void ieee80211_tx_pending(unsigned long · line 1887
Quoted 7 lines
net/mac80211/tx.c +5
-0while (!skb_queue_empty (&local->pending[i])) {struct sk_buff *skb =__skb_dequeue (&local->pending[i]); +struct ieee80211_tx_info *info =IEEE80211_SKB_CB (skb); +struct ieee80211_sub_if_data *sdata; + + sdata =vif_to_sdata (info->control.vif); +dev_hold (sdata->dev);I'm getting a panic at this point if I run hostapd on ath9k and a Windows client tries to authenticate with a wrong WPA key.
Quoted 1 line
Debugging shows that
sdata->devis NULL.
I suspect 'sdata' is already pointing to something bogus, since it's the result of a container_of().
Can you try with the sanity checking patch I'm about to send?
johannes
Re: [PATCH 1/3] mac80211: cooperate more with network namespaces
Quoted 13 lines · Johannes Berg
Thanks Pavel,
Quoted 11 lines
net/mac80211/tx.c void ieee80211_tx_pending(unsigned long · line 1887
Quoted 7 lines
net/mac80211/tx.c +5
-0while (!skb_queue_empty (&local->pending[i])) {struct sk_buff *skb =__skb_dequeue (&local->pending[i]); +struct ieee80211_tx_info *info =IEEE80211_SKB_CB (skb); +struct ieee80211_sub_if_data *sdata; + + sdata =vif_to_sdata (info->control.vif); +dev_hold (sdata->dev);I'm getting a panic at this point if I run hostapd on ath9k and a Windows client tries to authenticate with a wrong WPA key.
Quoted 6 lines
Quoted 1 line
Debugging shows that
sdata->devis NULL.I suspect 'sdata' is already pointing to something bogus, since it's the result of a
container_of().Can you try with the sanity checking patch I'm about to send?
If you mean "verify info->control.vif is not NULL", it makes no difference for me. No BUG is triggered. info->control.vif is never NULL in ieee80211_tx_pending(), but sdata->dev is NULL the second time ieee80211_tx_pending() is called. Just skipping that case doesn't help, I still get a panic in that function after about 10 calls.
Signature · 2 lines
Regards, Pavel Roskin
Johannes Berg
maintainerRe: [PATCH 1/3] mac80211: cooperate more with network namespaces
Quoted 5 lines · Pavel Roskin
If you mean "verify
info->control.vifis not NULL", it makes no difference for me. No BUG is triggered.info->control.vifis never NULL inieee80211_tx_pending(), butsdata->devis NULL the second timeieee80211_tx_pending()is called. Just skipping that case doesn't help, I still get a panic in that function after about 10 calls.
Weird.
So where does it panic if you skip with sdata->dev == NULL?
I don't see how sdata->dev can possibly be NULL. Hmm. Ok, I think we may not be recycling things correctly. Is it possible that this is a frame that was rejected by the device, or maybe a frame that was attempted to transmit, but then got into software retransmit?
johannes
Johannes Berg
maintainerRe: [PATCH 1/3] mac80211: cooperate more with network namespaces
Quoted 8 lines · Johannes Berg
Quoted 5 lines · Pavel Roskin
If you mean "verify
info->control.vifis not NULL", it makes no difference for me. No BUG is triggered.info->control.vifis never NULL inieee80211_tx_pending(), butsdata->devis NULL the second timeieee80211_tx_pending()is called. Just skipping that case doesn't help, I still get a panic in that function after about 10 calls.
Quoted 7 lines
Weird.
So where does it panic if you skip withsdata->dev== NULL?I don't see how
sdata->devcan possibly be NULL. Hmm. Ok, I think we may not be recycling things correctly. Is it possible that this is a frame that was rejected by the device, or maybe a frame that was attempted to transmit, but then got into software retransmit?
I have to leave now for an hour or two, but I'll review that code after I return. It's possible that this never happened before because skb->iif wasn't touched by drivers, but skb->cb is.
Actually, you're using ath5k, right? So we can see what info->control.vif points to after ath5k has used the skb, because apparently it's a valid pointer at least, just not a real vif pointer any more.
johannes
Re: [PATCH 1/3] mac80211: cooperate more with network namespaces
Quoted 4 lines · Johannes Berg
Actually, you're using ath5k, right? So we can see what
info->control.vifpoints to after ath5k has used the skb, because apparently it's a valid pointer at least, just not a real vif pointer any more.
I'm using ath9k.
Here's something interesting I have found. sdata->dev is valid if info->flags is 0x14091 and not valid if it's 0x12091.
0x4000 is IEEE80211_TX_INTFL_NEED_TXPROCESSING and 0x2000 is IEEE80211_TX_INTFL_RCALGO.
This patch prevents any panics. Moreover, I get a connection with the correct PSK.
net/mac80211/tx.c +3 -0
@@ -1889,6 +1889,9 @@ void ieee80211_tx_pending(unsigned long data)struct ieee80211_tx_info *info =IEEE80211_SKB_CB (skb);struct ieee80211_sub_if_data *sdata; +if (!(info->flags &IEEE80211_TX_INTFL_NEED_TXPROCESSING )) +continue ; +if (WARN_ON (!info->control.vif)) {kfree_skb (skb);continue ;
Signature · 2 lines
Regards, Pavel Roskin
Re: [PATCH 1/3] mac80211: cooperate more with network namespaces
net/mac80211/tx.c void ieee80211_tx_pending(unsigned long data) · line 1893
Quoted 2 lines · Pavel Roskin
net/mac80211/tx.c +2
-0+if (!(info->flags &IEEE80211_TX_INTFL_NEED_TXPROCESSING )) +continue ;
P.S. ieee80211_tx_pending_skb() actually checks IEEE80211_TX_INTFL_NEED_TXPROCESSING and doesn't use sdata in that case.
Perhaps we could use skb->dev instead of sdata->dev in that case.
Also, the existence of ieee80211_tx_pending_skb() doesn't seem justified in its present form. It's a piece of ieee80211_tx_pending() that calculates the same variables and does the checks that the caller should be doing. I would just integrate it into ieee80211_tx_pending(). If the result is too complex, we could split it in a more logical way.
Signature · 2 lines
Regards, Pavel Roskin
Johannes Berg
maintainerRe: [PATCH 1/3] mac80211: cooperate more with network namespaces
In reply to Pavel RoskinQuoted 9 lines · Pavel Roskin
Quoted 4 lines · Johannes Berg
Actually, you're using ath5k, right? So we can see what
info->control.vifpoints to after ath5k has used the skb, because apparently it's a valid pointer at least, just not a real vif pointer any more.I'm using ath9k.
Quoted 8 lines
Here's something interesting I have found.
sdata->devis valid ifinfo->flagsis0x14091and not valid if it's0x12091.
0x4000isIEEE80211_TX_INTFL_NEED_TXPROCESSINGand0x2000isIEEE80211_TX_INTFL_RCALGO.This patch prevents any panics. Moreover, I get a connection with the correct PSK.
Ok. So the question is where that skb is coming from that's causing the issues (your patch leaks it btw)
johannes
Johannes Berg
maintainerRe: [PATCH 1/3] mac80211: cooperate more with network namespaces
In reply to Pavel RoskinQuoted 8 lines · Pavel Roskin
Review
Quoted 2 lines · Pavel Roskin
(quoted diff) +2
-0+ if (!(info->flags & IEEE80211_TX_INTFL_NEED_TXPROCESSING)) + continue;
P.S.ieee80211_tx_pending_skb()actually checksIEEE80211_TX_INTFL_NEED_TXPROCESSINGand doesn't use sdata in that case.Perhaps we could use
skb->devinstead ofsdata->devin that case.
Nah, the bug is far more far-reaching, please see and try the patch below.
Quoted 5 lines
Also, the existence of
ieee80211_tx_pending_skb()doesn't seem justified in its present form. It's a piece ofieee80211_tx_pending()that calculates the same variables and does the checks that the caller should be doing. I would just integrate it intoieee80211_tx_pending(). If the result is too complex, we could split it in a more logical way.
Well... go ahead and try to clean it up -- I don't think it's possible due to the locking in ieee80211_tx_pending(), you definitely don't want to have a function that must be called with a lock held and then drops it in the middle only to re-acquire it. Just inlining it makes the whole thing unreadable imo.
Anyway, please let me know if the patch below fixes things for you. Again, one of these things where I'm sure there's an issue here, but I'm not sure it's the one you're seeing.
johannes
--- net/mac80211/main.c | 26 ++++++++++++++++++++++++++ 1 file changed, 26 insertions(+)
net/mac80211/main.c +26 -0
@@ -310,6 +310,31 @@ static void ieee80211_handle_filtered_fr {struct ieee80211_tx_info *info =IEEE80211_SKB_CB (skb); +/* +* XXX: This is temporary! +* +* The problem here is that when we get here, the driver will +* quilte likely have pretty much overwritten info->control by +* using info->driver_data or info->rate_driver_data. Thus, +* when passing out the frame to the driver again, we would be +* passing completely bogus data since the driver would then +* expect a properly filled info->control. In mac80211 itself +* the same problem occurs, since we need info->control.vif +* internally. +* +* To fix this, we should send the frame through TX processing +* again. However, it's not that simple, since the frame will +* have been software-encrypted (if applicable) already, and +* encrypting it again doesn't do much good. So to properly do +* that, we not only have to skip the actual 'raw' encryption +* (key selection etc. still has to be done!) but also the +* sequence number assignment since that impacts the crypto +* encapsulation, of course. +* +* Hence, for now, fix the bug by just dropping the frame. +*/ + goto drop; + sta->tx_filtered_count++;/*
@@ -363,6 +388,7 @@ static void ieee80211_handle_filtered_frreturn ; } + drop:#ifdef CONFIG_MAC80211_VERBOSE_DEBUG if (net_ratelimit ()) printk(KERN_DEBUG "%s: dropped TX filtered frame, "
Re: [PATCH 1/3] mac80211: cooperate more with network namespaces
Quoted 2 lines · Johannes Berg
Nah, the bug is far more far-reaching, please see and try the patch below.
Yes, it's working! There are no crashes with correct and incorrect PSK.
I'm not very interested in making code look nicer if the real problem is elsewhere. I just assumed that the problem was where the panic happened.
Signature · 2 lines
Regards, Pavel Roskin
Johannes Berg
maintainerRe: [PATCH 1/3] mac80211: cooperate more with network namespaces
Quoted 6 lines · Pavel Roskin
Quoted 2 lines · Johannes Berg
Nah, the bug is far more far-reaching, please see and try the patch below.
Yes, it's working! There are no crashes with correct and incorrect PSK.
Good to know, thanks.
Quoted 3 lines
I'm not very interested in making code look nicer if the real problem is elsewhere. I just assumed that the problem was where the panic happened.
Right, just wanted to point it out anyway.
johannes