Skip to contentSkip to conversation

Simulate medium behavior

RFC In review Daniel Wagner 2 participants 10 messages Apr 22, 2009 – May 14, 2009 Networking drivers (wireless)

lore
Share & tools
lore
#0 lore raw

RFC 1/1 Simulate medium behavior

Hi,

I'm still working improving the scan behavior of NM. Yes, I'm terrible slow. One thing which would help for this work is to simulate the wireless medium.

Below is my humble try to get a knob for enabling/disabling tx/rx. The variable name 'medium' is not really good, though I didn't come up with something better.

Writing a 0 to /debug/ieee80211/phy*/hwsim/medium enables rx/tx for the specific phy. Writing anything but 0 disables rx/tx. Not a very good API. It could be made something more realistic like reception quality is 80% associated some statistical dropping algorithm. Any thoughts?

Is this the right direction I'm heading to?

thanks,
daniel

---
 drivers/net/wireless/mac80211_hwsim.c |   32 +++++++++++++++++++++++++++++++-
 1 files changed, 31 insertions(+), 1 deletions(-)
drivers/net/wireless/mac80211_hwsim.c +31 -1
index d4fdc8b..6373226 100644
@@ -291,6 +291,9 @@ struct mac80211_hwsim_data {
 	bool ps_poll_pending;
 	struct dentry *debugfs;
 	struct dentry *debugfs_ps;
+
+	struct dentry *debugfs_medium;
+	unsigned int medium;
 };
 
 
@@ -399,6 +402,11 @@ static bool mac80211_hwsim_tx_frame(struct ieee80211_hw *hw,
 	rx_status.rate_idx = info->control.rates[0].idx;
 	/* TODO: simulate signal strength (and optional packet drop) */
 
+	if (data->medium != 0) {
+		/* drop packet, no ack received */
+		return false;
+	}
+
 	if (data->ps != PS_DISABLED)
 		hdr->frame_control |= cpu_to_le16(IEEE80211_FCTL_PM);
 
@@ -412,7 +420,8 @@ static bool mac80211_hwsim_tx_frame(struct ieee80211_hw *hw,
 
 		if (!data2->started || !data2->radio_enabled ||
 		    !hwsim_ps_rx_ok(data2, skb) ||
-		    data->channel->center_freq != data2->channel->center_freq)
+		    data->channel->center_freq != data2->channel->center_freq ||
+		    data2->medium != 0)
 			continue;
 
 		nskb = skb_copy(skb, GFP_ATOMIC);
@@ -725,6 +734,7 @@ static void mac80211_hwsim_free(void)
 	spin_unlock_bh(&hwsim_radio_lock);
 
 	list_for_each_entry(data, &tmplist, list) {
+		debugfs_remove(data->debugfs_medium);
 		debugfs_remove(data->debugfs_ps);
 		debugfs_remove(data->debugfs);
 		ieee80211_unregister_hw(data->hw);
@@ -876,6 +886,22 @@ static int hwsim_fops_ps_write(void *dat, u64 val)
 DEFINE_SIMPLE_ATTRIBUTE(hwsim_fops_ps, hwsim_fops_ps_read, hwsim_fops_ps_write,
 			"%llu\n");
 
+static int hwsim_fops_medium_read(void *dat, u64 *val)
+{
+	struct mac80211_hwsim_data *data = dat;
+	*val = data->medium;
+	return 0;
+}
+
+static int hwsim_fops_medium_write(void *dat, u64 val)
+{
+	struct mac80211_hwsim_data *data = dat;
+	data->medium = val;
+	return 0;
+}
+
+DEFINE_SIMPLE_ATTRIBUTE(hwsim_fops_medium, hwsim_fops_medium_read, hwsim_fops_medium_write,
+			"%llu\n");
 
 static int __init init_mac80211_hwsim(void)
 {
@@ -1106,6 +1132,10 @@ static int __init init_mac80211_hwsim(void)
 						       data->debugfs, data,
 						       &hwsim_fops_ps);
 
+		data->debugfs_medium = debugfs_create_file("medium", 0666,
+							   data->debugfs, data,
+							   &hwsim_fops_medium);
+
 		setup_timer(&data->beacon_timer, mac80211_hwsim_beacon,
 			    (unsigned long) hw);
 
Signature · 1 line
1.6.0.2.GIT
maintainer
#1 lore raw

Hi,

Quoted 7 lines

Writing a 0 to /debug/ieee80211/phy*/hwsim/medium enables rx/tx for the specific phy. Writing anything but 0 disables rx/tx. Not a very good API. It could be made something more realistic like reception quality is 80% associated some statistical dropping algorithm. Any thoughts?

Is this the right direction I'm heading to?

This looks like a very simple knob, just a little more work could get you much further -- have you seen this email/thread?

http://thread.gmane.org/gmane.linux.kernel.wireless.general/25001

johannes

#2 lore raw
Quoted 8 lines · Johannes Berg

Hi,

Quoted 6 lines

Writing a 0 to /debug/ieee80211/phy*/hwsim/medium enables rx/tx for the specific phy. Writing anything but 0 disables rx/tx. Not a very good API. It could be made something more realistic like reception quality is 80% associated some statistical dropping algorithm. Any thoughts?

Quoted 6 lines
Quoted 1 line

Is this the right direction I'm heading to?

This looks like a very simple knob, just a little more work could get you much further -- have you seen this email/thread?

http://thread.gmane.org/gmane.linux.kernel.wireless.general/25001

No, I haven't. Yep, my solution is really too simple. I like the virtual air grouping. I'll implement it.

thanks for the tip!
daniel

#3 lore raw
Quoted 11 lines · Daniel Wagner
Quoted 10 lines · Johannes Berg

Hi,

Quoted 7 lines

Writing a 0 to /debug/ieee80211/phy*/hwsim/medium enables rx/tx for the specific phy. Writing anything but 0 disables rx/tx. Not a very good API. It could be made something more realistic like reception quality is 80% associated some statistical dropping algorithm. Any thoughts?

Is this the right direction I'm heading to?

Quoted 7 lines
Quoted 4 lines

This looks like a very simple knob, just a little more work could get you much further -- have you seen this email/thread?

http://thread.gmane.org/gmane.linux.kernel.wireless.general/25001

No, I haven't. Yep, my solution is really too simple. I like the virtual air grouping. I'll implement it.

After a bit studying how to use configfs (and getting a bit lost) I found netconsole.c which uses configfs. So this version is a copy&past thing. Of course, this needs more work.

Is this what do did have in mind?

How it works:

# modprobe mac80211_hwsim
# mount -t configfs none /config

Make phy0 radio member of group 1:

# mkdir /config/mac80211_hwsim/phy0
# echo "phy0" > /config/mac80211_hwsim/phy0/dev_name
# echo 1 > /config/mac80211_hwsim/phy0/group
# echo 1 > /config/mac80211_hwsim/phy0/enable

I have only tested it very short, no problems found. Works as expected :)

cheers,
daniel

---
 drivers/net/wireless/Kconfig          |    9 +
 drivers/net/wireless/mac80211_hwsim.c |  324 ++++++++++++++++++++++++++++++++-
 2 files changed, 332 insertions(+), 1 deletions(-)
drivers/net/wireless/Kconfig +9 -0
index 2d8434f..5316675 100644
@@ -481,6 +481,15 @@ config MAC80211_HWSIM
 	  To compile this driver as a module, choose M here: the module will be
 	  called mac80211_hwsim.  If unsure, say N.
 
+config MAC80211_HWSIM_VGROUP
+	bool "enable virtual air grouping"
+	depends on MAC80211_HWSIM
+	default y
+	help
+	  This option enables virtual air grouping.
+
+	  If unsure, say N.
+
 config MWL8K
 	tristate "Marvell 88W8xxx PCI/PCIe Wireless support"
 	depends on MAC80211 && PCI && WLAN_80211 && EXPERIMENTAL
drivers/net/wireless/mac80211_hwsim.c +323 -1

360 lines not on this page · raw message

Signature · 1 line
1.6.2.4
maintainer
#4 lore raw

Hi Daniel,

Quoted 10 lines
Quoted 4 lines
Quoted 1 line

http://thread.gmane.org/gmane.linux.kernel.wireless.general/25001

No, I haven't. Yep, my solution is really too simple. I like the virtual air grouping. I'll implement it.

After a bit studying how to use configfs (and getting a bit lost) I found netconsole.c which uses configfs. So this version is a copy&past thing. Of course, this needs more work.

Is this what do did have in mind?

Very cool. We had a patch at some point to add configfs support to hwsim but it seems it got lost :(

But I don't think your configfs support is optimal since you need to create those dirs for phy0 and put the name in etc. even when phy0 already exists. I'm sorry to have wasted your time with the configfs suggestion, but I think with how hwsim _currently_ behaves, without configfs support for creating new radios etc., putting this into debugfs would be nicer, since debugfs already has a dir for this from cfg80211/mac80211...

drivers/net/wireless/Kconfig config MAC80211_HWSIM · line 491

Quoted 8 lines
drivers/net/wireless/Kconfig +8 -0
+config MAC80211_HWSIM_VGROUP
+	bool "enable virtual air grouping"
+	depends on MAC80211_HWSIM
+	default y
+	help
+	  This option enables virtual air grouping.
+
+	  If unsure, say N.

I also wouldn't make it configurable, just make it default to group == 1 so that it behaves as it was.

johannes

#5 lore raw

Hi Johannes

Quoted 11 lines · Johannes Berg

Hi Daniel,

Quoted 9 lines
Quoted 4 lines
Quoted 1 line

http://thread.gmane.org/gmane.linux.kernel.wireless.general/25001

No, I haven't. Yep, my solution is really too simple. I like the virtual air grouping. I'll implement it.

After a bit studying how to use configfs (and getting a bit lost) I found netconsole.c which uses configfs. So this version is a copy&past thing. Of course, this needs more work.

Quoted 8 lines
Quoted 1 line

Is this what do did have in mind?

Very cool. We had a patch at some point to add configfs support to hwsim but it seems it got lost :(

But I don't think your configfs support is optimal since you need to create those dirs for phy0 and put the name in etc. even when phy0 already exists.

Yes, that was also what I thought. If I understood it correctly that is exactly the main difference between configfs and debugfs. In configfs the userspace has the owner ship of the 'objects' (mkdir, rmdir) whereas in debugfs the kernel controls the lifetime of the 'objects'.

Quoted 5 lines

I'm sorry to have wasted your time with the configfs suggestion, but I think with how hwsim _currently_ behaves, without configfs support for creating new radios etc., putting this into debugfs would be nicer, since debugfs already has a dir for this from cfg80211/mac80211...

No problem. I was able to learn something :) Ok, I'll cook up the next version.

Quoted 11 lines

drivers/net/wireless/Kconfig config MAC80211_HWSIM · line 491

Quoted 8 lines
drivers/net/wireless/Kconfig +8 -0
+config MAC80211_HWSIM_VGROUP
+	bool "enable virtual air grouping"
+	depends on MAC80211_HWSIM
+	default y
+	help
+	  This option enables virtual air grouping.
+
+	  If unsure, say N.

I also wouldn't make it configurable, just make it default to group == 1 so that it behaves as it was.

Good point.

Thanks for your comments!

cheers,
daniel

#6 lore raw
In reply to Johannes Berg

Hi Johannes,

Quoted 7 lines · Johannes Berg

But I don't think your configfs support is optimal since you need to create those dirs for phy0 and put the name in etc. even when phy0 already exists. I'm sorry to have wasted your time with the configfs suggestion, but I think with how hwsim _currently_ behaves, without configfs support for creating new radios etc., putting this into debugfs would be nicer, since debugfs already has a dir for this from cfg80211/mac80211...

Here we go (slow as usual :)

Subject: [PATCH] mac80211_hwsim: Group radios

Each bit in /debug/ieee80211/phy*/hwsim/group represents one group. By default all radios belong to the same group "1". A radio can belong to several groups.

---
 drivers/net/wireless/mac80211_hwsim.c |   30 +++++++++++++++++++++++++++++-
 1 files changed, 29 insertions(+), 1 deletions(-)
drivers/net/wireless/mac80211_hwsim.c +29 -1
index 61a4ad7..741f2e2 100644
@@ -291,6 +291,9 @@ struct mac80211_hwsim_data {
 	bool ps_poll_pending;
 	struct dentry *debugfs;
 	struct dentry *debugfs_ps;
+
+	u64 group;
+	struct dentry *debugfs_group;
 };
 
 
@@ -412,7 +415,8 @@ static bool mac80211_hwsim_tx_frame(struct ieee80211_hw *hw,
 
 		if (!data2->started || !data2->radio_enabled ||
 		    !hwsim_ps_rx_ok(data2, skb) ||
-		    data->channel->center_freq != data2->channel->center_freq)
+		    data->channel->center_freq != data2->channel->center_freq ||
+		    !(data->group & data2->group))
 			continue;
 
 		nskb = skb_copy(skb, GFP_ATOMIC);
@@ -720,6 +724,7 @@ static void mac80211_hwsim_free(void)
 	spin_unlock_bh(&hwsim_radio_lock);
 
 	list_for_each_entry(data, &tmplist, list) {
+		debugfs_remove(data->debugfs_group);
 		debugfs_remove(data->debugfs_ps);
 		debugfs_remove(data->debugfs);
 		ieee80211_unregister_hw(data->hw);
@@ -872,6 +877,24 @@ DEFINE_SIMPLE_ATTRIBUTE(hwsim_fops_ps, hwsim_fops_ps_read, hwsim_fops_ps_write,
 			"%llu\n");
 
 
+static int hwsim_fops_group_read(void *dat, u64 *val)
+{
+	struct mac80211_hwsim_data *data = dat;
+	*val = data->group;
+	return 0;
+}
+
+static int hwsim_fops_group_write(void *dat, u64 val)
+{
+	struct mac80211_hwsim_data *data = dat;
+	data->group = val;
+	return 0;
+}
+
+DEFINE_SIMPLE_ATTRIBUTE(hwsim_fops_group,
+			hwsim_fops_group_read, hwsim_fops_group_write,
+			"%llu\n");
+
 static int __init init_mac80211_hwsim(void)
 {
 	int i, err = 0;
@@ -976,6 +999,8 @@ static int __init init_mac80211_hwsim(void)
 
 			hw->wiphy->bands[band] = sband;
 		}
+		/* By default all radios are belonging to the first group */
+		data->group = 1;
 
 		/* Work to be done prior to ieee80211_register_hw() */
 		switch (regtest) {
@@ -1100,6 +1125,9 @@ static int __init init_mac80211_hwsim(void)
 		data->debugfs_ps = debugfs_create_file("ps", 0666,
 						       data->debugfs, data,
 						       &hwsim_fops_ps);
+		data->debugfs_group = debugfs_create_file("group", 0666,
+							data->debugfs, data,
+							&hwsim_fops_group);
 
 		setup_timer(&data->beacon_timer, mac80211_hwsim_beacon,
 			    (unsigned long) hw);
Signature · 1 line
1.6.2.4
maintainer
#7 lore raw

On Thu, 2009-05-14 at 20:45 +0200, Daniel Wagner wrote:

Looks good to me, couple of comments below.

Quoted 7 lines
drivers/net/wireless/mac80211_hwsim.c +7 -0
+static int hwsim_fops_group_read(void *dat, u64 *val)
+{
+	struct mac80211_hwsim_data *data = dat;
+	*val = data->group;
+	return 0;
+}
+

drivers/net/wireless/mac80211_hwsim.c DEFINE_SIMPLE_ATTRIBUTE(hwsim_fops_ps, hwsim_fops_ps_read, hwsim_fops_ps_write, · line 896

Quoted 10 lines
drivers/net/wireless/mac80211_hwsim.c +10 -0
+static int hwsim_fops_group_write(void *dat, u64 val)
+{
+	struct mac80211_hwsim_data *data = dat;
+	data->group = val;
+	return 0;
+}
+
+DEFINE_SIMPLE_ATTRIBUTE(hwsim_fops_group,
+			hwsim_fops_group_read, hwsim_fops_group_write,
+			"%llu\n");

I think %llx would work better for this? At least I can do bitops better in hex :)

drivers/net/wireless/mac80211_hwsim.c static int __init init_mac80211_hwsim(void) · line 1130

Quoted 6 lines
drivers/net/wireless/mac80211_hwsim.c +3 -0
 		data->debugfs_ps = debugfs_create_file("ps", 0666,
 						       data->debugfs, data,
 						       &hwsim_fops_ps);
+		data->debugfs_group = debugfs_create_file("group", 0666,
+							data->debugfs, data,
+							&hwsim_fops_group);

Good thing this is a test module only... debugfs files with user access permissions are a security risk. (a user can open the file and leave the fd open, if root then rmmods future reads/writes to the fd by the user crash the kernel)

Anyway looks good, please submit properly with [PATCH] etc :)

johannes

#8 lore raw
Quoted 16 lines · Johannes Berg

On Thu, 2009-05-14 at 20:45 +0200, Daniel Wagner wrote:

Looks good to me, couple of comments below.

Quoted 12 lines
drivers/net/wireless/mac80211_hwsim.c +12 -0
+static int hwsim_fops_group_read(void *dat, u64 *val)
+{
+	struct mac80211_hwsim_data *data = dat;
+	*val = data->group;
+	return 0;
+}
+
+static int hwsim_fops_group_write(void *dat, u64 val)
+{
+	struct mac80211_hwsim_data *data = dat;
+	data->group = val;
+	return 0;
Quoted 8 lines

drivers/net/wireless/mac80211_hwsim.c DEFINE_SIMPLE_ATTRIBUTE(hwsim_fops_ps, hwsim_fops_ps_read, hwsim_fops_ps_write, · line 896

Quoted 5 lines
drivers/net/wireless/mac80211_hwsim.c +5 -0
+}
+
+DEFINE_SIMPLE_ATTRIBUTE(hwsim_fops_group,
+			hwsim_fops_group_read, hwsim_fops_group_write,
+			"%llu\n");

I think %llx would work better for this? At least I can do bitops better in hex :)

Yeah, I have to improve my copy&paste-foo :)

Quoted 12 lines

drivers/net/wireless/mac80211_hwsim.c static int __init init_mac80211_hwsim(void) · line 1130

Quoted 7 lines
drivers/net/wireless/mac80211_hwsim.c +3 -0
 		data->debugfs_ps = debugfs_create_file("ps", 0666,
 						       data->debugfs, data,
 						       &hwsim_fops_ps);
+		data->debugfs_group = debugfs_create_file("group", 0666,
+							data->debugfs, data,
+							&hwsim_fops_group);

Good thing this is a test module only... debugfs files with user access permissions are a security risk. (a user can open the file and leave the fd open, if root then rmmods future reads/writes to the fd by the user crash the kernel)

There isn't much we can do about that, right?

Quoted 1 line

Anyway looks good, please submit properly with [PATCH] etc :)

Done.

Thanks for reviewing!

cheers,
daniel

maintainer
#9 lore raw
Quoted 12 lines · Daniel Wagner
Quoted 11 lines

drivers/net/wireless/mac80211_hwsim.c static int __init init_mac80211_hwsim(void) · line 1130

Quoted 6 lines
drivers/net/wireless/mac80211_hwsim.c +3 -0
 		data->debugfs_ps = debugfs_create_file("ps", 0666,
 						       data->debugfs, data,
 						       &hwsim_fops_ps);
+		data->debugfs_group = debugfs_create_file("group", 0666,
+							data->debugfs, data,
+							&hwsim_fops_group);

Good thing this is a test module only... debugfs files with user access permissions are a security risk. (a user can open the file and leave the fd open, if root then rmmods future reads/writes to the fd by the user crash the kernel)

Quoted 1 line

There isn't much we can do about that, right?

I generally just use 0600 to not allow regular users to open the files. Maybe at some point, with the revoke() discussion, we'll actually get a solution.

johannes