wiphy: make wiphy work queue reentrant
PATCH v2 1/3 wiphy: make wiphy work queue reentrant
In some situations its convenient for the same work item to be inserted (rescheduled) while its in progress. FT for example does this now if a roam fails. The same ft_work item gets re-inserted which, currently, is not safe to do since the item is modified and removed once completed.
Currently FT reschedules the same work item in the 'do_work' callback but this patch actually allows for rescheduling as long as the item is currently running, even if do_work has returned and waiting for wiphy_radio_work_done.
A few new flags were added to the radio work item. One which is set when the item is running which prevents other items from being inserted ahead (previously done with priority=INT_MIN). This is needed since we cannot modify the priority anymore in case the item is being rescheduled.
The another flag is set when the item was rescheduled. Once the item is finished the 'rescheduled' flag is checked and the item will be removed and re-inserted into the queue if needed.
--- src/wiphy.c | 69 ++++++++++++++++++++++++++++++++++++++++++----------- src/wiphy.h | 2 ++ 2 files changed, 57 insertions(+), 14 deletions(-)
src/wiphy.c +55 -14
@@ -2570,6 +2570,17 @@ static void wiphy_reg_notify(struct l_genl_msg *msg, void *user_data)wiphy_dump_after_regdom (wiphy); } +static int insert_by_priority (const void *a,const void *b,void *user_data) +{ +const struct wiphy_radio_work_item *new = a; +const struct wiphy_radio_work_item *work = b; + +if (work->running || work->priority <= new->priority) +return 1 ; + +return -1 ; +} +static void wiphy_radio_work_next (struct wiphy *wiphy) {struct wiphy_radio_work_item *work;
@@ -2583,7 +2594,7 @@ static void wiphy_radio_work_next(struct wiphy *wiphy)* Ensures no other work item will get inserted before this one while * the work is being done. */ - work->priority =INT_MIN ; + work->running = true;l_debug ("Starting work item %u" , work->id);
@@ -2592,7 +2603,12 @@ static void wiphy_radio_work_next(struct wiphy *wiphy) wiphy->work_in_callback = false;if (done) { - work->id =0 ; +bool rescheduled = work->rescheduled; + + work->running = false; + +if (!rescheduled) + work->id =0 ;l_queue_remove (wiphy->work, work);
@@ -2600,26 +2616,49 @@ static void wiphy_radio_work_next(struct wiphy *wiphy)destroy_work (work); wiphy->work_in_callback = false; +/* +* If the item was rescheduled inside do_work() we can safely +* insert it here, otherwise destroy_work() could have freed it. +* The item could have been re-inserted inside destroy_work() +* but this is safe since the item was removed from the queue. +*/ +if (rescheduled) { + work->rescheduled = false; +l_queue_insert (wiphy->work, work, + insert_by_priority,NULL ); + } +wiphy_radio_work_next (wiphy); } } -static int insert_by_priority (const void *a,const void *b,void *user_data) -{ -const struct wiphy_radio_work_item *new = a; -const struct wiphy_radio_work_item *work = b; - -if (work->priority <= new->priority) -return 1 ; - -return -1 ; -} -uint32_t wiphy_radio_work_insert (struct wiphy *wiphy,struct wiphy_radio_work_item *item,int priority,const struct wiphy_radio_work_item_ops *ops) { +/* +* Handling the case of re-inserting the same work item that is in +* progress. A non-started work item should never be re-inserted +* into the queue. Keep the same ID, priority, and ops. If these somehow +* are different the caller should really be using a separate work item. +* Once the item is finished it will be removed and re-inserted based +* on the rescheduled flag. +*/ +if (item ==l_queue_peek_head (wiphy->work)) { +/* +* Shouldn't cause problems, but at least warn the caller they +* should really be using a separate item +*/ +L_WARN_ON (item->rescheduled || item->priority != priority || + ops != item->ops); + + item->rescheduled = true; +l_debug ("Rescheduled work item %u" , item->id); + +return item->id; + } + item->priority = priority; item->ops = ops; item->id = ++work_ids;
@@ -2656,6 +2695,7 @@ void wiphy_radio_work_done(struct wiphy *wiphy, uint32_t id)if (item->id == id) { next = true;l_queue_pop_head (wiphy->work); + item->running = false; }else item =l_queue_remove_if (wiphy->work, match_id,L_UINT_TO_PTR (id));
@@ -2664,7 +2704,8 @@ void wiphy_radio_work_done(struct wiphy *wiphy, uint32_t id)l_debug ("Work item %u done" , id); - item->id =0 ; +if (!item->rescheduled) + item->id =0 ; wiphy->work_in_callback = true;destroy_work (item);
src/wiphy.h +2 -0
@@ -45,6 +45,8 @@ struct wiphy_radio_work_item {uint32_t id;int priority;const struct wiphy_radio_work_item_ops *ops; + bool rescheduled : 1; + bool running : 1; };enum {
Signature · 1 line
2.25.1
PATCH v2 2/3 auto-t: modify PSK-roam test to use FT failure path
This adds another radio so IWD hits the FT failure path after authentication to the first BSS fails. This causes a wiphy work item to be rescheduled which previously was unsafe.
--- autotests/testPSK-roam/connection_test.py | 13 ++++--- autotests/testPSK-roam/failed_roam_test.py | 33 ++++++++++------ autotests/testPSK-roam/ft-psk-ccmp-3.conf | 39 +++++++++++++++++++ autotests/testPSK-roam/hw.conf | 3 +- .../testPSK-roam/roam_ap_disconnect_test.py | 4 +- 5 files changed, 74 insertions(+), 18 deletions(-) create mode 100644 autotests/testPSK-roam/ft-psk-ccmp-3.conf
v2:
* Added the hostapd configuration which was missing from the last patch set.
autotests/testPSK-roam/connection_test.py +8 -5
@@ -48,7 +48,7 @@ class Test(unittest.TestCase): self.rule2.enabled = True # Send 100 packets (to be dropped), should trigger beacon loss testutil.tx_packets(device.name, self.bss_hostapd[0].ifname, 100) - device.wait_for_event('packet-loss-roam', timeout=30) + device.wait_for_event('packet-loss-roam', timeout=60) else: device.roam(self.bss_hostapd[1].bssid)
@@ -173,19 +173,22 @@ class Test(unittest.TestCase): IWD.copy_to_storage('TestFT.psk') cls.bss_hostapd = [ HostapdCLI(config='ft-psk-ccmp-1.conf'), - HostapdCLI(config='ft-psk-ccmp-2.conf') ] + HostapdCLI(config='ft-psk-ccmp-2.conf'), + HostapdCLI(config='ft-psk-ccmp-3.conf') ] + cls.bss_hostapd[2].disable() + rad0 = hwsim.get_radio('rad0') - rad2 = hwsim.get_radio('rad2') + rad3 = hwsim.get_radio('rad3') cls.rule0 = hwsim.rules.create() - cls.rule0.source = rad2.addresses[0] + cls.rule0.source = rad3.addresses[0] cls.rule0.bidirectional = True cls.rule0.signal = -2000 cls.rule0.prefix = 'b0' cls.rule0.drop = True cls.rule1 = hwsim.rules.create() - cls.rule1.source = rad2.addresses[0] + cls.rule1.source = rad3.addresses[0] cls.rule1.prefix = '08' cls.rule1.drop = True
autotests/testPSK-roam/failed_roam_test.py +22 -11
@@ -62,21 +62,17 @@ class Test(unittest.TestCase): self.rule0.enabled = True - device.roam(self.bss_hostapd[1].bssid) - - # Roam should fail... - device.wait_for_event('ft-roam-failed') + # IWD should connect, then attempt a roam to BSS 1, which should fail... + device.wait_for_event('ft-roam-failed', timeout=60) # ... but IWD should remain connected self.assertTrue(device.state == DeviceState.connected) self.rule0.enabled = False - # Try again once more - device.roam(self.bss_hostapd[1].bssid) - - self.verify_roam(wd, device, self.bss_hostapd[0], self.bss_hostapd[1]) + # IWD should then try BSS 2, and succeed + self.verify_roam(wd, device, self.bss_hostapd[0], self.bss_hostapd[2]) - self.bss_hostapd[1].deauthenticate(device.address) + self.bss_hostapd[2].deauthenticate(device.address) condition = 'obj.state == DeviceState.disconnected' wd.wait_for_object_condition(device, condition)
@@ -88,6 +84,7 @@ class Test(unittest.TestCase): self.rule0.enabled = False self.rule1.enabled = False + self.rule2.enabled = False @classmethod def setUpClass(cls):
@@ -96,14 +93,16 @@ class Test(unittest.TestCase): IWD.copy_to_storage('TestFT.psk') cls.bss_hostapd = [ HostapdCLI(config='ft-psk-ccmp-1.conf'), - HostapdCLI(config='ft-psk-ccmp-2.conf') ] + HostapdCLI(config='ft-psk-ccmp-2.conf'), + HostapdCLI(config='ft-psk-ccmp-3.conf') ] cls.bss_hostapd[0].set_address('12:00:00:00:00:01') cls.bss_hostapd[1].set_address('12:00:00:00:00:02') + cls.bss_hostapd[2].set_address('12:00:00:00:00:03') # Drop Authenticate frames cls.rule0 = hwsim.rules.create() - cls.rule0.bidirectional = True + cls.rule0.source = hwsim.get_radio('rad1').addresses[0] cls.rule0.prefix = 'b0' cls.rule0.drop = True
@@ -113,6 +112,18 @@ class Test(unittest.TestCase): cls.rule1.prefix = 'd0' cls.rule1.drop = True + # Causes IWD to immediately roam away from BSS 0 + cls.rule2 = hwsim.rules.create() + cls.rule2.source = hwsim.get_radio('rad0').addresses[0] + cls.rule2.signal = -8000 + cls.rule2.enabled = True + + # Causes IWD to first prefer BSS 1 to roam, then BSS 2. + cls.rule3 = hwsim.rules.create() + cls.rule3.source = hwsim.get_radio('rad2').addresses[0] + cls.rule3.signal = -7000 + cls.rule3.enabled = True + @classmethod def tearDownClass(cls): IWD.clear_storage()
autotests/testPSK-roam/ft-psk-ccmp-3.conf new file +39 -0
@@ -0,0 +1,39 @@ +hw_mode=g +channel=2 +ssid=TestFT +utf8_ssid=1 +ctrl_interface=/var/run/hostapd + +r1_key_holder=120000000002 +nas_identifier=dummy2 + +wpa=2 +# Can support WPA-PSK and FT-PSK (space separated list) and/or EAP at the same +# time but we want to force FT +wpa_key_mgmt=FT-PSK +wpa_pairwise=CCMP +wpa_passphrase=EasilyGuessedPassword +ieee80211w=1 +rsn_preauth=1 +rsn_preauth_interfaces=lo +disable_pmksa_caching=0 +# Allow PMK cache to be shared opportunistically among configured interfaces +# and BSSes (i.e., all configurations within a single hostapd process). +okc=1 +mobility_domain=1234 +reassociation_deadline=60000 +r0kh=12:00:00:00:00:01 dummy1 000102030405060708090a0b0c0d0e0f +r0kh=12:00:00:00:00:02 dummy2 000102030405060708090a0b0c0d0e0f +r1kh=12:00:00:00:00:01 00:00:00:00:00:01 000102030405060708090a0b0c0d0e0f +r1kh=12:00:00:00:00:02 00:00:00:00:00:02 000102030405060708090a0b0c0d0e0f +# Push mode only needed for 8021x, not PSK mode since msk already known +pmk_r1_push=0 +# Allow locally generated FT response so we don't have to configure push/pull +# between BSSes running as separate hostapd processes as in the test-runner +# case. Only works with FT-PSK, otherwise brctl needs to be installed and +# CONFIG_BRIDGE enabled in the kernel. +ft_psk_generate_local=1 +ft_over_ds=0 +ap_table_expiration_time=36000 +ap_table_max_size=10 +rrm_neighbor_report=1
autotests/testPSK-roam/hw.conf +2 -1
@@ -1,8 +1,9 @@ [SETUP] -num_radios=3 +num_radios=4 start_iwd=0 hwsim_medium=yes [HOSTAPD] rad0=ft-psk-ccmp-1.conf rad1=ft-psk-ccmp-2.conf +rad2=ft-psk-ccmp-3.conf
autotests/testPSK-roam/roam_ap_disconnect_test.py +3 -1
@@ -70,8 +70,10 @@ class Test(unittest.TestCase): IWD.copy_to_storage('TestFT.psk') cls.bss_hostapd = [ HostapdCLI(config='ft-psk-ccmp-1.conf'), - HostapdCLI(config='ft-psk-ccmp-2.conf') ] + HostapdCLI(config='ft-psk-ccmp-2.conf'), + HostapdCLI(config='ft-psk-ccmp-3.conf') ] cls.bss_hostapd[1].disable() + cls.bss_hostapd[2].disable() cls.bss_hostapd[0].set_value('ocv', '0') cls.bss_hostapd[0].set_value('ieee80211w', '0')
Signature · 1 line
2.25.1
PATCH v2 3/3 auto-t: increase timeout in testPSK-roam
In reply to James PrestwoodThis was causing random failures and increasing the timeout seems to be a lot more reliable.
--- autotests/testPSK-roam/roam_ap_disconnect_test.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
autotests/testPSK-roam/roam_ap_disconnect_test.py +1 -1
@@ -33,7 +33,7 @@ class Test(unittest.TestCase): # Since both BSS's have low signal, the roam should fail and trigger # another roam scan. device.wait_for_event('roam-scan-triggered', timeout=30) - device.wait_for_event('no-roam-candidates', timeout=30) + device.wait_for_event('no-roam-candidates', timeout=60) # Hostapd sends disconnect self.bss_hostapd[0].disable()
Signature · 1 line
2.25.1
Hi James,
Quoted 11 lines · James Prestwood
In some situations its convenient for the same work item to be inserted (rescheduled) while its in progress. FT for example does this now if a roam fails. The same
ft_workitem gets re-inserted which, currently, is not safe to do since the item is modified and removed once completed.Currently FT reschedules the same work item in the '
do_work' callback but this patch actually allows for rescheduling as long as the item is currently running, even ifdo_workhas returned and waiting forwiphy_radio_work_done.
Quoted 5 lines
A few new flags were added to the radio work item. One which is set when the item is running which prevents other items from being inserted ahead (previously done with priority=INT_MIN). This is needed since we cannot modify the priority anymore in case the item is being rescheduled.
This seems way too complicated.
Quoted 9 lines
The another flag is set when the item was rescheduled. Once the item is finished the 'rescheduled' flag is checked and the item will be removed and re-inserted into the queue if needed.
--- src/wiphy.c | 69 ++++++++++++++++++++++++++++++++++++++++++----------- src/wiphy.h | 2 ++ 2 files changed, 57 insertions(+), 14 deletions(-)
<snip>
src/wiphy.c static void wiphy_radio_work_next(struct wiphy *wiphy) · line 2600
Quoted 9 lines
src/wiphy.c +1
-1@@ -2583,7 +2594,7 @@ static void wiphy_radio_work_next(struct wiphy *wiphy)* Ensures no other work item will get inserted before this one while * the work is being done. */ - work->priority =INT_MIN ; + work->running = true;l_debug ("Starting work item %u" , work->id);
Can't you simply take note of the current item id, set the priority, take it off the queue here, prior to starting.
Review static void wiphy_radio_work_next(struct wiphy *wiphy) · line 2604
Quoted 3 lines
(quoted diff) +0
-0@@ -2592,7 +2603,12 @@ static void wiphy_radio_work_next(struct wiphy *wiphy) wiphy->work_in_callback = false;
If done and work id changed, there's nothing to do.
If done, then proceed as now.
If not done, reinsert the work to the head.
src/wiphy.c static void wiphy_radio_work_next(struct wiphy *wiphy) · line 2616
Quoted 13 lines
src/wiphy.c +6
-1if (done) { - work->id =0 ; +bool rescheduled = work->rescheduled; + + work->running = false; + +if (!rescheduled) + work->id =0 ;l_queue_remove (wiphy->work, work);@@ -2600,26 +2616,49 @@ static void wiphy_radio_work_next(struct wiphy *wiphy)destroy_work (work);
I'm really not sure this makes sense. You are calling destroy on an item you just potentially reinserted. I think a reinserted item should not trigger the destroy. The caller has full control over the item anyway, so they can take care of this on their end. In fact, I'd still think that the caller should use a different API for this to be absolutely clear, even if it is just wiphy_radio_work_insert() with some extra checks.
Regards,
-Denis