Skip to contentSkip to conversation

wpa supplicant/ipw3945, ESSID last char missing

DISCUSSION Norbert Preining 22 participants 100 messages Oct 2, 2006 – Oct 10, 2006 2 messages not in the archive

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

Dear all!

I have the following problem with wpa supplicant/ipw3945. First the versions:
kernel: 2.6.18-mm2 (self compiled)
ieee80211: 1.1.14
ipw3945: git source
ipw3945d: 1.7.19
wpa supplicant: 0.5.5 (Debian/unstable 0.5.5-1)

Config file of wpa_supplicant:
ctrl_interface=/var/run/wpa_supplicant
ctrl_interface_group=0
eapol_version=1
ap_scan=1

network={
        ssid="norbunet"
        key_mgmt=NONE
        auth_alg=SHARED
        wep_key0=HEXKEY1
        wep_key1=HEXKEY2
        wep_key2=HEXKEY3
        wep_key3=HEXKEY4
        wep_tx_keyidx=0
        priority=5
}

When I start ipw3945d and wpa_supplicant it does not connect. And the reason is that when typing
iwconfig eth2
(eth0 cable, eth1 not present!?!, eth2 ipw3945) I see that the ESSID is set to
"norbune"
instead of
"norbunet"

Calling
iwconfig eth2 essid "norbunet "
(mind the space at the end) immediately connects (even with encryption) and everything is working.

Do you have any idea what this might be related to?

The last kernel I tried which worked out of the box (well, with comnpiling ieee and ipw) was 2.6.18-rc4.

Best wishes

Norbert

-------------------------------------------------------------------------------
Dr. Norbert Preining <[email protected]> Università di Siena
Debian Developer <[email protected]> Debian TeX Group
gpg DSA: 0x09C5B094 fp: 14DF 2E6C 0307 BE6D AD76 A9C0 D2BF 4AA3 09C5 B094
-------------------------------------------------------------------------------
`I think you ought to know that I'm feeling very
depressed.'
`Life, don't talk to me about life.'
`Here I am, brain the size of a planet and they ask me to take you down to the bridge. Call that "job satisfaction"? 'Cos I don't.'
`I've got this terrible pain in all the diodes down my
left side.'
--- Guess who.
--- Douglas Adams, The Hitchhikers Guide to the Galaxy

#1 lore raw
Quoted 42 lines · Norbert Preining

Dear all!

I have the following problem with wpa supplicant/ipw3945. First the versions:

kernel: 2.6.18-mm2      (self compiled)
ieee80211:      1.1.14
ipw3945:        git source
ipw3945d:       1.7.19
wpa supplicant: 0.5.5   (Debian/unstable 0.5.5-1)

Config file of wpa_supplicant:
ctrl_interface=/var/run/wpa_supplicant
ctrl_interface_group=0
eapol_version=1
ap_scan=1

network={
        ssid="norbunet"
        key_mgmt=NONE
        auth_alg=SHARED
        wep_key0=HEXKEY1
        wep_key1=HEXKEY2
        wep_key2=HEXKEY3
        wep_key3=HEXKEY4
        wep_tx_keyidx=0
        priority=5
}

When I start ipw3945d and wpa_supplicant it does not connect. And the reason is that when typing
iwconfig eth2
(eth0 cable, eth1 not present!?!, eth2 ipw3945) I see that the ESSID is set to
"norbune"
instead of
"norbunet"

Calling
iwconfig eth2 essid "norbunet "
(mind the space at the end) immediately connects (even with encryption) and everything is working.

Quoted 4 lines

Do you have any idea what this might be related to?

The last kernel I tried which worked out of the box (well, with comnpiling ieee and ipw) was 2.6.18-rc4.

Hi Norbert,

we've been just through an email thread where it has been determined that wpa_supplicant 0.4.9 (I would assume that 0.5.5 is also okay) and wireless-tools from Jean's latest tarball are necessary to work with the recent wireless extensions v21 that have been merged in.

What wireless-tools are you using ?

Ciao,

--alessandro

"Well a man has two reasons for things that he does
the first one is pride and the second one is love all understandings must come by this way"

(Husker Du, 'She Floated Away')

#2 lore raw
Quoted 7 lines · Alessandro Suardi

we've been just through an email thread where it has been determined that wpa_supplicant 0.4.9 (I would assume that 0.5.5 is also okay) and wireless-tools from Jean's latest tarball are necessary to work with the recent wireless extensions v21 that have been merged in.

What wireless-tools are you using ?

wireless-tools from Debian/unstable, version 28-1, so I assume wireless v28. And the README tells the same. What would be the newest version?

Best wishes

Norbert

-------------------------------------------------------------------------------
Dr. Norbert Preining <[email protected]> Università di Siena
Debian Developer <[email protected]> Debian TeX Group
gpg DSA: 0x09C5B094 fp: 14DF 2E6C 0307 BE6D AD76 A9C0 D2BF 4AA3 09C5 B094
-------------------------------------------------------------------------------
FROLESWORTH (n.)
Measure. The minimum time it is necessary to spend frowning in deep concentration at each picture in an art gallery in order that everyone else doesn't think you've a complete moron.
--- Douglas Adams, The Meaning of Liff

#3 lore raw
Quoted 11 lines · Norbert Preining
Quoted 7 lines · Alessandro Suardi

we've been just through an email thread where it has been determined that wpa_supplicant 0.4.9 (I would assume that 0.5.5 is also okay) and wireless-tools from Jean's latest tarball are necessary to work with the recent wireless extensions v21 that have been merged in.

What wireless-tools are you using ?

wireless-tools from Debian/unstable, version 28-1, so I assume wireless v28. And the README tells the same. What would be the newest version?

http://www.hpl.hp.com/personal/Jean_Tourrilhes/Linux/wireless_tools.29.pre10.tar.gz

which, from Jean's page, has the following:

The main features of the latest beta is WE-21 support (long/short retry, power saving level, modulation), enhanced command line parser in iwconfig, scanning options, more WPA support and more footprint reduction tricks

Cheers,

--alessandro

"Well a man has two reasons for things that he does
the first one is pride and the second one is love all understandings must come by this way"

(Husker Du, 'She Floated Away')

#4 lore raw

Hi all!

Quoted 4 lines · Alessandro Suardi

The main features of the latest beta is WE-21 support (long/short retry, power saving level, modulation), enhanced command line parser in iwconfig, scanning options, more WPA support and more footprint reduction tricks

Bingo. I build the new 29-pre10 and everything is working.

Thanks for the tip!

Now for the Debian bug report.

Best wishes

Norbert

-------------------------------------------------------------------------------
Dr. Norbert Preining <[email protected]> Università di Siena
Debian Developer <[email protected]> Debian TeX Group
gpg DSA: 0x09C5B094 fp: 14DF 2E6C 0307 BE6D AD76 A9C0 D2BF 4AA3 09C5 B094
-------------------------------------------------------------------------------
LOCHRANZA (n.)
The long unaccomplished wail in the middle of a Scottish folk song where the pipes nip around the corner for a couple of drinks.
--- Douglas Adams, The Meaning of Liff

#5 lore raw
Quoted 6 lines · Norbert Preining
Quoted 4 lines

The main features of the latest beta is WE-21 support (long/short retry, power saving level, modulation), enhanced command line parser in iwconfig, scanning options, more WPA support and more footprint reduction tricks

Bingo. I build the new 29-pre10 and everything is working.

Sorry, that was over-optimistic. still same behaviour as with the Debian v28 version.

The last character is cut of from wpa_supplicant. I have to set the essid by hadn with
"real-essid "
mark the space at the end!

Best wishes

Norbert

-------------------------------------------------------------------------------
Dr. Norbert Preining <[email protected]> Università di Siena
Debian Developer <[email protected]> Debian TeX Group
gpg DSA: 0x09C5B094 fp: 14DF 2E6C 0307 BE6D AD76 A9C0 D2BF 4AA3 09C5 B094
-------------------------------------------------------------------------------
MELCOMBE REGIS (n.)
The name of the style of decoration used in cocktail lounges in mock Tudor hotels in Surrey.
--- Douglas Adams, The Meaning of Liff

#6 lore raw
Quoted 8 lines · Norbert Preining
Quoted 6 lines · Norbert Preining
Quoted 4 lines

The main features of the latest beta is WE-21 support (long/short retry, power saving level, modulation), enhanced command line parser in iwconfig, scanning options, more WPA support and more footprint reduction tricks

Bingo. I build the new 29-pre10 and everything is working.

Quoted 7 lines

Sorry, that was over-optimistic. still same behaviour as with the Debian v28 version.

The last character is cut of from wpa_supplicant. I have to set the essid by hadn with
"real-essid "
mark the space at the end!

You have a mismatch between your wireless-tools, your kernel, and/or wpa_supplicant. WE-21 uses the _real_ ssid length rather than the kludge of hacking off the last byte used previously. Please ensure that your tools, driver, and kernel are using WE-21.
'cat /proc/net/wireless' should tell you what your kernel is using. Getting the driver WE is a bit harder and you may have to look at the source.

Dan

Quoted 18 lines

Best wishes

Norbert

-------------------------------------------------------------------------------
Dr. Norbert Preining <[email protected]> Università di Siena
Debian Developer <[email protected]> Debian TeX Group
gpg DSA: 0x09C5B094 fp: 14DF 2E6C 0307 BE6D AD76 A9C0 D2BF 4AA3 09C5 B094
-------------------------------------------------------------------------------
MELCOMBE REGIS (n.)
The name of the style of decoration used in cocktail lounges in mock Tudor hotels in Surrey.
--- Douglas Adams, The Meaning of Liff
_______________________________________________
HostAP mailing list
[email protected]
http://lists.shmoo.com/mailman/listinfo/hostap

#7 lore raw
Quoted 17 lines · On Mon, 02 Oct 2006 12:58:24 -0400 Dan Williams <[email protected]> wrote:
Quoted 8 lines · Norbert Preining
Quoted 6 lines · Norbert Preining
Quoted 4 lines

The main features of the latest beta is WE-21 support (long/short retry, power saving level, modulation), enhanced command line parser in iwconfig, scanning options, more WPA support and more footprint reduction tricks

Bingo. I build the new 29-pre10 and everything is working.

Quoted 7 lines

Sorry, that was over-optimistic. still same behaviour as with the Debian v28 version.

The last character is cut of from wpa_supplicant. I have to set the essid by hadn with
"real-essid "
mark the space at the end!

Quoted 7 lines

You have a mismatch between your wireless-tools, your kernel, and/or wpa_supplicant. WE-21 uses the _real_ ssid length rather than the kludge of hacking off the last byte used previously. Please ensure that your tools, driver, and kernel are using WE-21.
'cat /proc/net/wireless' should tell you what your kernel is using. Getting the driver WE is a bit harder and you may have to look at the source.

Jean, John: the amount of trouble which this change is causing is quite high considering that we're not even at -rc1 yet. It's going to get worse.

It doesn't sound like it'll be too hard to arrange for the kernel to continue to work correctly with old userspace?

#8 lore raw
Quoted 10 lines · Andrew Morton

On Mon, 02 Oct 2006 12:58:24 -0400

Quoted 8 lines

You have a mismatch between your wireless-tools, your kernel, and/or wpa_supplicant. WE-21 uses the _real_ ssid length rather than the kludge of hacking off the last byte used previously. Please ensure that your tools, driver, and kernel are using WE-21.
'cat /proc/net/wireless' should tell you what your kernel is using. Getting the driver WE is a bit harder and you may have to look at the source.

Quoted 2 lines

Jean, John: the amount of trouble which this change is causing is quite high considering that we're not even at -rc1 yet. It's going to get worse.

	We have to split between the different issues we have seen.
	Tools issue (the wpa_supplicant problem). -> those can only be
fixed by people upgrading. Fortunately, there are not so many tools
affected, and new version of those tools were released last
April/May. As I said, most distro have those in the pipe.
	In-Kernel driver issues (the Orinoco driver problem). -> those
can be patched and fixed as we go along. I would not worry about
those.
	Out-of-kernel issues (the ipw3945 driver problem). -> those

drivers need to be updated. That's the problem of living outside the kernel. Very often those drivers are reactive with respect to kernel API changes, rather than pro-active, so there is not much we can do.

Quoted 2 lines

It doesn't sound like it'll be too hard to arrange for the kernel to continue to work correctly with old userspace?

Actually, it's impossible. New userspace can work across both
version, old userspace fails on new version.

The whole point of the -rc process is to find problems and the
scope of it, there is no way I can know everything. At this point, we can decide if WE-21 should go in 2.6.19 or wait for 2.6.20. But I know that most Linux-Wireless people such as Dan and Jouni have been waiting impatiently for those changes...

	Have fun...

	Jean
#9 lore raw
Quoted 30 lines · Jean Tourrilhes
Quoted 10 lines · Andrew Morton

On Mon, 02 Oct 2006 12:58:24 -0400

Quoted 8 lines

You have a mismatch between your wireless-tools, your kernel, and/or wpa_supplicant. WE-21 uses the _real_ ssid length rather than the kludge of hacking off the last byte used previously. Please ensure that your tools, driver, and kernel are using WE-21.
'cat /proc/net/wireless' should tell you what your kernel is using. Getting the driver WE is a bit harder and you may have to look at the source.

Quoted 2 lines

Jean, John: the amount of trouble which this change is causing is quite high considering that we're not even at -rc1 yet. It's going to get worse.

	We have to split between the different issues we have seen.
	Tools issue (the wpa_supplicant problem). -> those can only be
fixed by people upgrading. Fortunately, there are not so many tools
affected, and new version of those tools were released last
April/May. As I said, most distro have those in the pipe.
	In-Kernel driver issues (the Orinoco driver problem). -> those
can be patched and fixed as we go along. I would not worry about
those.
	Out-of-kernel issues (the ipw3945 driver problem). -> those

drivers need to be updated. That's the problem of living outside the kernel. Very often those drivers are reactive with respect to kernel API changes, rather than pro-active, so there is not much we can do.

Quoted 2 lines

It doesn't sound like it'll be too hard to arrange for the kernel to continue to work correctly with old userspace?

Quoted 8 lines

Actually, it's impossible. New userspace can work across both
version, old userspace fails on new version.

The whole point of the -rc process is to find problems and the
scope of it, there is no way I can know everything. At this point, we can decide if WE-21 should go in 2.6.19 or wait for 2.6.20. But I know that most Linux-Wireless people such as Dan and Jouni have been waiting impatiently for those changes...

Right; we need to get this settled and we need to figure out a way with Wireless Extensions to allow exact SSIDs to be sent and retrieved from the card/driver. How much cruft do we add to drivers and/or WE bits to bend over backwards and preserve compatibility with a lot of older bits? I can think of a few ways here to preserve backwards compat, but most involve adding bits to 3 places in each driver and a new flag to WE, which puts us right back in the same situation we're in right now; inconsistent drivers and poor semantics both inside and outside the kernel.

Maybe the answer here _is_ to bend over backwards to preserve compatibility, restore null-termination-and-increment-length bits in drivers and WE, and kludge all the user-space tools. Then just Do The Right Thing with nl80211/cfg80211 (whenever they come around) and fix stuff up in the nl80211 WE compat layer. I don't know. But nl80211 isn't here yet, and it's unclear when it will be.

Dan

Quoted 3 lines
	Have fun...

	Jean
#10 lore raw
Quoted 10 lines · Dan Williams

Right; we need to get this settled and we need to figure out a way with Wireless Extensions to allow exact SSIDs to be sent and retrieved from the card/driver. How much cruft do we add to drivers and/or WE bits to bend over backwards and preserve compatibility with a lot of older bits? I can think of a few ways here to preserve backwards compat, but most involve adding bits to 3 places in each driver and a new flag to WE, which puts us right back in the same situation we're in right now; inconsistent drivers and poor semantics both inside and outside the kernel.

There is no way old tools are going to set/process those extra
flags, so it does not work. And this won't fix out-of-tree drivers...

Quoted 3 lines

Maybe the answer here _is_ to bend over backwards to preserve compatibility, restore null-termination-and-increment-length bits in drivers and WE, and kludge all the user-space tools.

Let's not throw the baby with the bathwater.
This is *not* the first time I've done such incompatible
change in WE (size in set, pointer in scan, remove pointer in events). The other time, it went completely under the radar, you guys did not noticed anything (apart Jouni).
The difference is that this time I attempted to do it a bit
quicker than usual. Maybe I was too fast this time. But I've noticed that some distro have become slower to pick up changes than in the past, which is frustrating me.
To me, the solution is just to let a bit more time for the
userspace to propagate to the distro. And to educate users to pay attention to the incompatibility warnings displayed by the tools.

Quoted 1 line

Dan

	Have fun...

	Jean
#11 lore raw
In reply to Jean Tourrilhes
Quoted 5 lines · Jean Tourrilhes
Quoted 2 lines · Andrew Morton

It doesn't sound like it'll be too hard to arrange for the kernel to continue to work correctly with old userspace?

Quoted 9 lines

Actually, it's impossible. New userspace can work across both
version, old userspace fails on new version.

The whole point of the -rc process is to find problems and the
scope of it, there is no way I can know everything. At this point, we can decide if WE-21 should go in 2.6.19 or wait for 2.6.20. But I know that most Linux-Wireless people such as Dan and Jouni have been waiting impatiently for those changes...

It would be nice if need of a specific version of wireless tools was documented in Documentaion/Changes. It was a surprise for me when my wireless card stopped working.

Signature · 1 line
Dmitry
#12 lore raw
Quoted 9 lines · Dmitry Torokhov
Quoted 7 lines · Jean Tourrilhes

The whole point of the -rc process is to find problems and the
scope of it, there is no way I can know everything. At this point, we can decide if WE-21 should go in 2.6.19 or wait for 2.6.20. But I know that most Linux-Wireless people such as Dan and Jouni have been waiting impatiently for those changes...

Quoted 3 lines

It would be nice if need of a specific version of wireless tools was documented in Documentaion/Changes. It was a surprise for me when my wireless card stopped working.

The Wireless Tols themselves issue a nice warning telling you
about the version mismatch and the need to upgrade. This is even more powerful, as most people don't read the doc, but they run the tools.
Don't tell my you ignored the warning !

Quoted 1 line

Dmitry

Jean

#13 lore raw
Quoted 10 lines · Jean Tourrilhes
Quoted 9 lines · Dmitry Torokhov
Quoted 7 lines · Jean Tourrilhes

The whole point of the -rc process is to find problems and the
scope of it, there is no way I can know everything. At this point, we can decide if WE-21 should go in 2.6.19 or wait for 2.6.20. But I know that most Linux-Wireless people such as Dan and Jouni have been waiting impatiently for those changes...

Quoted 9 lines
Quoted 3 lines

It would be nice if need of a specific version of wireless tools was documented in Documentaion/Changes. It was a surprise for me when my wireless card stopped working.

The Wireless Tols themselves issue a nice warning telling you
about the version mismatch and the need to upgrade. This is even more powerful, as most people don't read the doc, but they run the tools.
Don't tell my you ignored the warning !

I think it was giving me that warning for the last couple of years... I think FC3 was shipped with version 17? "May not work" is different from "need release x.y.z" to work.

Signature · 1 line
Dmitry
#14 lore raw
In reply to Jean Tourrilhes
Quoted 27 lines · Jean Tourrilhes
Quoted 10 lines · Andrew Morton

On Mon, 02 Oct 2006 12:58:24 -0400

Quoted 8 lines

You have a mismatch between your wireless-tools, your kernel, and/or wpa_supplicant. WE-21 uses the _real_ ssid length rather than the kludge of hacking off the last byte used previously. Please ensure that your tools, driver, and kernel are using WE-21.
'cat /proc/net/wireless' should tell you what your kernel is using. Getting the driver WE is a bit harder and you may have to look at the source.

Quoted 2 lines

Jean, John: the amount of trouble which this change is causing is quite high considering that we're not even at -rc1 yet. It's going to get worse.

	We have to split between the different issues we have seen.
	Tools issue (the wpa_supplicant problem). -> those can only be
fixed by people upgrading. Fortunately, there are not so many tools
affected, and new version of those tools were released last
April/May. As I said, most distro have those in the pipe.
	In-Kernel driver issues (the Orinoco driver problem). -> those
can be patched and fixed as we go along. I would not worry about
those.
	Out-of-kernel issues (the ipw3945 driver problem). -> those

drivers need to be updated. That's the problem of living outside the kernel. Very often those drivers are reactive with respect to kernel API changes, rather than pro-active, so there is not much we can do.

Quoted 5 lines
Quoted 2 lines

It doesn't sound like it'll be too hard to arrange for the kernel to continue to work correctly with old userspace?

Actually, it's impossible. New userspace can work across both
version, old userspace fails on new version.

<rant>
Well, please tell me now what number of people actually _will_ upgrade?

And if they don't, will they use the -rc kernels? No, they won't, because of the apparent wireless breakage.

This way we loose quite a few testers and the entire development process is affected, and that's _only_ because you have decided it will be _convenient_ to change the ABI. However, such changes affect _everyone_ and in a wrong way, except for a few people who actually want the change. They cause more damage than they are worth, so they should be avoided at all reasonable cost.

It would be fair to introduce the change when distributions actually ship the userland tools capable of handling it, but not _now_.
</rant>

Greetings,
Rafael

Signature · 2 lines
You never change things by fighting the existing reality.
		R. Buckminster Fuller
#15 lore raw
Quoted 28 lines · Rafael J. Wysocki
Quoted 27 lines · Jean Tourrilhes
Quoted 10 lines · Andrew Morton

On Mon, 02 Oct 2006 12:58:24 -0400

Quoted 8 lines

You have a mismatch between your wireless-tools, your kernel, and/or wpa_supplicant. WE-21 uses the _real_ ssid length rather than the kludge of hacking off the last byte used previously. Please ensure that your tools, driver, and kernel are using WE-21.
'cat /proc/net/wireless' should tell you what your kernel is using. Getting the driver WE is a bit harder and you may have to look at the source.

Quoted 2 lines

Jean, John: the amount of trouble which this change is causing is quite high considering that we're not even at -rc1 yet. It's going to get worse.

	We have to split between the different issues we have seen.
	Tools issue (the wpa_supplicant problem). -> those can only be
fixed by people upgrading. Fortunately, there are not so many tools
affected, and new version of those tools were released last
April/May. As I said, most distro have those in the pipe.
	In-Kernel driver issues (the Orinoco driver problem). -> those
can be patched and fixed as we go along. I would not worry about
those.
	Out-of-kernel issues (the ipw3945 driver problem). -> those

drivers need to be updated. That's the problem of living outside the kernel. Very often those drivers are reactive with respect to kernel API changes, rather than pro-active, so there is not much we can do.

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

It doesn't sound like it'll be too hard to arrange for the kernel to continue to work correctly with old userspace?

Actually, it's impossible. New userspace can work across both
version, old userspace fails on new version.

<rant>
Well, please tell me now what number of people actually _will_ upgrade?

If you're using a distro, the distro maintainers should be making sure versions are compatible. If you don't, well, then you need to be making sure versions are compatible.

Quoted 10 lines

And if they don't, will they use the -rc kernels? No, they won't, because of the apparent wireless breakage.

This way we loose quite a few testers and the entire development process is affected, and that's _only_ because you have decided it will be _convenient_ to change the ABI. However, such changes affect _everyone_ and in a wrong way, except for a few people who actually want the change. They cause more damage than they are worth, so they should be avoided at all reasonable cost.

Quoted 2 lines

It would be fair to introduce the change when distributions actually ship the userland tools capable of handling it, but not _now_.

Distributions _are_ shipping those tools already. The problem is more with older distributions where, for example, the kernel gets upgraded but other stuff does not. If a kernel upgrade happens, then the distro needs to make sure userspace works with it. That's nothing new.

Dan

Quoted 6 lines

</rant>

Greetings,
Rafael

#16 lore raw
Quoted 4 lines · Dan Williams

Distributions _are_ shipping those tools already. The problem is more with older distributions where, for example, the kernel gets upgraded but other stuff does not. If a kernel upgrade happens, then the distro needs to make sure userspace works with it. That's nothing new.

Um, *which* distro's are shipping it already? RHEL4? SLES10? I thought we saw a note saying that even Debian **unstable** didn't have a new enough version of the wireless-tools....

- Ted

#17 lore raw
Quoted 9 lines · Theodore Tso
Quoted 4 lines · Dan Williams

Distributions _are_ shipping those tools already. The problem is more with older distributions where, for example, the kernel gets upgraded but other stuff does not. If a kernel upgrade happens, then the distro needs to make sure userspace works with it. That's nothing new.

Um, *which* distro's are shipping it already? RHEL4? SLES10? I thought we saw a note saying that even Debian **unstable** didn't have a new enough version of the wireless-tools....

I personally never said it was shipping already in all distro. Debian Testing has the proper version since last May (forget
about Stable, as usual). So, Debian is in fine shape, I would say...
Gentoo 2006.1 is obviously shipping with it.
FC6 has it since August. It's currently in RC.
I believe Mandriva 2007 has it, and it's also in RC. They make
it hard for non-club member to know what's happening.
SuSE I can't figure out.
Slackware has it in the dev version. 10.2 was one year ago, so
they are due for a new release, I guess.

Note that both rpmseek and rpmfind are obsolete, so it's
actually a pain trying to track down all that info.

Quoted 1 line

- Ted

	Have fun...

	Jean
#18 lore raw
Quoted 7 lines · Jean Tourrilhes (on Mon, 2 Oct 2006 14:58:12 -0700) wrote:
Quoted 6 lines · Theodore Tso
Quoted 4 lines · Dan Williams

Distributions _are_ shipping those tools already. The problem is more with older distributions where, for example, the kernel gets upgraded but other stuff does not. If a kernel upgrade happens, then the distro needs to make sure userspace works with it. That's nothing new.

Quoted 7 lines
Quoted 3 lines

Um, *which* distro's are shipping it already? RHEL4? SLES10? I thought we saw a note saying that even Debian **unstable** didn't have a new enough version of the wireless-tools....

I personally never said it was shipping already in all distro.
...
SuSE I can't figure out.

SLES9 SP3:
wireless-tools-27pre12-39.31 (WE_VERSION 16)
No wpa_supplicant

SLES10:
wireless-tools-28pre13-22.4 (WE_VERSION 19)
wpa_supplicant-0.4.8-14.8

SuSELinux 10.1:
wireless-tools-28pre13-20 (WE_VERSION 19)
wpa_supplicant-0.4.8-14

openSUSE-10.2-Alpha4:
wireless-tools-29pre10-3 (WE_VERSION 21)
wpa_supplicant-gui-0.4.8-17

Only openSUSE 10.2 (which is still in Alpha status) has WE_VERSION 21 support.

#19 lore raw
In reply to Theodore Tso
Quoted 9 lines · Theodore Tso
Quoted 4 lines · Dan Williams

Distributions _are_ shipping those tools already. The problem is more with older distributions where, for example, the kernel gets upgraded but other stuff does not. If a kernel upgrade happens, then the distro needs to make sure userspace works with it. That's nothing new.

Um, *which* distro's are shipping it already? RHEL4? SLES10? I thought we saw a note saying that even Debian **unstable** didn't have a new enough version of the wireless-tools....

That was my point initially. FC5-latest is apparently carrying tools which are "too old"... and I yum update twice or thrice a week. Not that *I* minded building packages from source, but this is likely a bit too much for a good slice of the userbase.

Ciao,

--alessandro

"Well a man has two reasons for things that he does
the first one is pride and the second one is love all understandings must come by this way"

(Husker Du, 'She Floated Away')

#20 lore raw
Quoted 7 lines · Alessandro Suardi
Quoted 6 lines · Theodore Tso
Quoted 4 lines · Dan Williams

Distributions _are_ shipping those tools already. The problem is more with older distributions where, for example, the kernel gets upgraded but other stuff does not. If a kernel upgrade happens, then the distro needs to make sure userspace works with it. That's nothing new.

Quoted 8 lines
Quoted 3 lines

Um, *which* distro's are shipping it already? RHEL4? SLES10? I thought we saw a note saying that even Debian **unstable** didn't have a new enough version of the wireless-tools....

That was my point initially. FC5-latest is apparently carrying
tools which are "too old"... and I yum update twice or thrice a week. Not that *I* minded building packages from source, but this is likely a bit too much for a good slice of the userbase.

And that's my fault. I was made aware of this issue last week and am currently testing out the newer wireless-tools with the intention of pushing them to FC5-updates. Had I actually been tracking the wireless-tools release cycle more closely, I would have pushed the wireless-tools-28 final version when it came out.

But, as far as I know (please correct me if I'm wrong), that 2.6.18 doesn't actually include WE-21! [1] Somebody is trying to run out-of-kernel ipw3945 drivers using a 2.6.18 kernel from FC5 that's WE-20 only, but the driver uses WE-21?

ipw3945 isn't in the upstream kernel, and Fedora certainly isn't going to go out of its way to make sure every piece of software under the sun that's not included in the distro works with it [2]. Distros try to make sure they are internally consistent, not globally consistent. If somebody uses out-of-kernel drivers, they take that responsibility. The solution to the problem is to make sure that ipw3945 gets in the kernel as quickly as possible, something I think we'd all like to have happen.

Dan

[1] /lib/modules/2.6.18-1.2708.fc6/build/include/linux/wireless.h defines WIRELESS_EXT = 20

[2] Yeah it would be nice; but go out of the way to have Skype w/ OSS, Nvidia binary drivers, ATI binary drivers, ndiswrapper, out-of-kernel-drivers? No.

Quoted 9 lines

Ciao,

--alessandro

"Well a man has two reasons for things that he does
the first one is pride and the second one is love all understandings must come by this way"

(Husker Du, 'She Floated Away')

#21 lore raw
Quoted 36 lines · Dan Williams
Quoted 7 lines · Alessandro Suardi
Quoted 6 lines · Theodore Tso
Quoted 4 lines · Dan Williams

Distributions _are_ shipping those tools already. The problem is more with older distributions where, for example, the kernel gets upgraded but other stuff does not. If a kernel upgrade happens, then the distro needs to make sure userspace works with it. That's nothing new.

Quoted 8 lines
Quoted 3 lines

Um, *which* distro's are shipping it already? RHEL4? SLES10? I thought we saw a note saying that even Debian **unstable** didn't have a new enough version of the wireless-tools....

That was my point initially. FC5-latest is apparently carrying
tools which are "too old"... and I yum update twice or thrice a week. Not that *I* minded building packages from source, but this is likely a bit too much for a good slice of the userbase.

And that's my fault. I was made aware of this issue last week and am currently testing out the newer wireless-tools with the intention of pushing them to FC5-updates. Had I actually been tracking the wireless-tools release cycle more closely, I would have pushed the wireless-tools-28 final version when it came out.

But, as far as I know (please correct me if I'm wrong), that 2.6.18 doesn't actually include WE-21! [1] Somebody is trying to run out-of-kernel ipw3945 drivers using a 2.6.18 kernel from FC5 that's WE-20 only, but the driver uses WE-21?

ipw3945 isn't in the upstream kernel, and Fedora certainly isn't going to go out of its way to make sure every piece of software under the sun that's not included in the distro works with it [2]. Distros try to make sure they are internally consistent, not globally consistent. If somebody uses out-of-kernel drivers, they take that responsibility. The solution to the problem is to make sure that ipw3945 gets in the kernel as quickly as possible, something I think we'd all like to have happen.

Quoted 8 lines

Dan

[1] /lib/modules/2.6.18-1.2708.fc6/build/include/linux/wireless.h defines WIRELESS_EXT = 20

[2] Yeah it would be nice; but go out of the way to have Skype w/ OSS, Nvidia binary drivers, ATI binary drivers, ndiswrapper, out-of-kernel-drivers? No.

While strongly agreeing with the above reasoning, I'd just point out that the two current reports were

mine -> FC5, in-kernel ipw2200, wpa_supplicant and wireless-tools "too old",
rebuild userspace from recent sources => all fine and dandy

Norbert's -> Debian unstable, out-of-kernel ipw3945, wpa_supplicant okay,
wireless-tools "too old", rebuild these from recent sources =>
still a few issues

so it was only myself running FC5, and with a WE-21 capable Torvalds kernel as I test at least 4 -git snapshots per week (and yes, 2.6.18 doesn't have WE-21,
which is the root of the reports - WE-21 went in in 2.6.18-git9).

Indeed the in-kernel driver appeared to be the easily solved case, as expected...

Thanks, ciao,

--alessandro

"Well a man has two reasons for things that he does
the first one is pride and the second one is love all understandings must come by this way"

(Husker Du, 'She Floated Away')

#22 lore raw
In reply to Dan Williams
Quoted 23 lines · Dan Williams
Quoted 7 lines · Alessandro Suardi
Quoted 6 lines · Theodore Tso
Quoted 4 lines · Dan Williams

Distributions _are_ shipping those tools already. The problem is more with older distributions where, for example, the kernel gets upgraded but other stuff does not. If a kernel upgrade happens, then the distro needs to make sure userspace works with it. That's nothing new.

Quoted 8 lines
Quoted 3 lines

Um, *which* distro's are shipping it already? RHEL4? SLES10? I thought we saw a note saying that even Debian **unstable** didn't have a new enough version of the wireless-tools....

That was my point initially. FC5-latest is apparently carrying
tools which are "too old"... and I yum update twice or thrice a week. Not that *I* minded building packages from source, but this is likely a bit too much for a good slice of the userbase.

And that's my fault. I was made aware of this issue last week and am currently testing out the newer wireless-tools with the intention of pushing them to FC5-updates. Had I actually been tracking the wireless-tools release cycle more closely, I would have pushed the wireless-tools-28 final version when it came out.

Quoted 4 lines

But, as far as I know (please correct me if I'm wrong), that 2.6.18 doesn't actually include WE-21! [1] Somebody is trying to run out-of-kernel ipw3945 drivers using a 2.6.18 kernel from FC5 that's WE-20 only, but the driver uses WE-21?

Obviously I was a bit optimisitic in accepting the notion that distros had already upgraded their userland bits to handle WE-21. I apologize.

As Dan points-out, it will be a while before distros (other than Fedora rawhide and equivalents) have 2.6.19 kernels for general users. For now, those experiencing this issue should be comfortable experiencing some breakage...?

So, is the window between now and the release of 2.6.19 big enough to give the distros time to get wireless-tools and wpa_supplicant into their update streams? Or do we need to go through the pain of reverting/delaying WE-21?

John

Signature · 2 lines
John W. Linville
[email protected]
#23 lore raw
Quoted 9 lines · John W. Linville

As Dan points-out, it will be a while before distros (other than Fedora rawhide and equivalents) have 2.6.19 kernels for general users. For now, those experiencing this issue should be comfortable experiencing some breakage...?

So, is the window between now and the release of 2.6.19 big enough to give the distros time to get wireless-tools and wpa_supplicant into their update streams? Or do we need to go through the pain of reverting/delaying WE-21?

There is a fundamental question hiding here, which is whether or not it is acceptable to break users who are running some large set of mainline distro's, such as RHEL 4, SLES/SLED 10, Ubuntu Dapper, et. al, but who want to upgrade to a newer 2.6 kernel?

Many users have moved to Ubuntu Dapper, or RHEL 4, or SLES/SLED 10 because they don't want to deal with a constantly changing/breaking GNOME/X world, where packages are constantly being updated and possibly breaking their desktop. Some of these users are in fact kernel developers, who want to live and test on the bleeding edge, but who don't want to deal with an unstable Desktop/X world, since that's not where their expertise lies. Other users are ones which have to use a mainstream distribution for one reason or another (maybe they have software that only works on RHEL 4), but are interested in testing bleeding edge kernels because they want to help contribute to testing and quality assurance. Is it acceptable to break them with ABI changes?

If the answer that it is acceptable to break the "slower moving" distro's, how much time do we need to allow to elapse before the "faster moving" distro's have accepted the necessary userspace bits? Is it 30 days? 60 days? 90 days? Or do we do it by distribution. If all of Debian testing, Ubuntu development, Fedora Core n and n-1, OpenSuse, Gentoo, has accepted the newer bits, is that enough time?

Clearly the wireless updates failed the second series of tests; but we haven't even decided, amongst kernel developers, under what circumstances is it permissible to break the first set of distro's. Clearly in the best of all worlds new interfaces are carefully documented, and no new interface is introduced without thinking very carefully about forwards and backwards compatibility. Unfortuately, the wireless ABI interface is a legacy interface which seems to be really broken in many different ways.

John, has the wireless community considered creating a new interface which *is* carefully designed, and supporting both the new and the legacy interface for 2-3 years until all of the mainstream distributions have had a chance to cycle? It would be hard, I know, but would it be harder than some of the alternatives, and would it be worth it?

Regards,

- Ted

#24 lore raw
Quoted 16 lines · Theodore Tso
Quoted 9 lines · John W. Linville

As Dan points-out, it will be a while before distros (other than Fedora rawhide and equivalents) have 2.6.19 kernels for general users. For now, those experiencing this issue should be comfortable experiencing some breakage...?

So, is the window between now and the release of 2.6.19 big enough to give the distros time to get wireless-tools and wpa_supplicant into their update streams? Or do we need to go through the pain of reverting/delaying WE-21?

There is a fundamental question hiding here, which is whether or not it is acceptable to break users who are running some large set of mainline distro's, such as RHEL 4, SLES/SLED 10, Ubuntu Dapper, et. al, but who want to upgrade to a newer 2.6 kernel?

Quoted 6 lines

Many users have moved to Ubuntu Dapper, or RHEL 4, or SLES/SLED 10 because they don't want to deal with a constantly changing/breaking GNOME/X world, where packages are constantly being updated and possibly breaking their desktop. Some of these users are in fact kernel developers, who want to live and test on the bleeding edge, but who don't want to deal with an unstable Desktop/X world, since that's

I'm certain these people already experience breakage when using new bits that haven't been settled into their desktop/distro of choice. It wasn't so long ago (2.6.10 - 2.6.13) that installing a new kernel would break the expectations of udev, HAL, and libsysfs while sysfs directory structure was getting laid out for stuff like power management, wireless devices, etc. If you're a core system developer, you've got to expect breakage somewhere.

Quoted 9 lines

not where their expertise lies. Other users are ones which have to use a mainstream distribution for one reason or another (maybe they have software that only works on RHEL 4), but are interested in testing bleeding edge kernels because they want to help contribute to testing and quality assurance. Is it acceptable to break them with ABI changes?

If the answer that it is acceptable to break the "slower moving" distro's, how much time do we need to allow to elapse before the

I'd point out here that one is not breaking the _distro_, as long as we assume that distros are internally consistent (which one of the major points of a distro!). What's getting broken is people who install/replace distro-provided software with their own bits. In the first case, the distro people are responsible to making sure that breakage does not occur, and that distro users are not affected. In the second case, that responsibility falls to the user who installed/replaced the distro-provided software, precisely because that software is no longer distro provided.

We've _got_ to accept that somebody installing their own stuff has _some_ responsibility to ensure compatibility of the random code they install. In a perfect world, distros never make a mistake. But usually a distro has a much broader and deeper set of expertise than any one person, and is at least peripherally aware of changes coming down the pike. One single person cannot hope to assume the responsibilities of many maintainers working by division of labor.

Obviously we don't try to break stuff unintentionally, or when the pain would be too severe, because we know better than most what's going on and it's Just Not Nice. But ultimately, whoever is installing the software bears the consequences of his/her actions, precisely because they pulled the trigger.

Quoted 14 lines

"faster moving" distro's have accepted the necessary userspace bits? Is it 30 days? 60 days? 90 days? Or do we do it by distribution. If all of Debian testing, Ubuntu development, Fedora Core n and n-1, OpenSuse, Gentoo, has accepted the newer bits, is that enough time?

Clearly the wireless updates failed the second series of tests; but we haven't even decided, amongst kernel developers, under what circumstances is it permissible to break the first set of distro's. Clearly in the best of all worlds new interfaces are carefully documented, and no new interface is introduced without thinking very carefully about forwards and backwards compatibility. Unfortuately, the wireless ABI interface is a legacy interface which seems to be really broken in many different ways.

Quoted 3 lines

John, has the wireless community considered creating a new interface which *is* carefully designed, and supporting both the new and the legacy interface for 2-3 years until all of the mainstream

Yes, nl80211/cfg80211 (sent to netdev@ last week) is the likely successor. Please, if you have suggestions for ABI/API-proofability, review the patch and post suggestions! We all know a carefully designed is not just about the code, but about the semantics, structures, etc as well. It would be quite valuable to have everyone's input to make sure it's as future-proof as possible.

Dan

Quoted 7 lines

distributions have had a chance to cycle? It would be hard, I know, but would it be harder than some of the alternatives, and would it be worth it?

Regards,

- Ted

#25 lore raw
Quoted 6 lines · Dan Williams
Quoted 3 lines · Theodore Tso

John, has the wireless community considered creating a new interface which *is* carefully designed, and supporting both the new and the legacy interface for 2-3 years until all of the mainstream

Quoted 6 lines

Yes, nl80211/cfg80211 (sent to netdev@ last week) is the likely successor. Please, if you have suggestions for ABI/API-proofability, review the patch and post suggestions! We all know a carefully designed is not just about the code, but about the semantics, structures, etc as well. It would be quite valuable to have everyone's input to make sure it's as future-proof as possible.

Dan is quick on his keyboard today! :-)

As Dan points-out, there is serious work underway on the next generation wireless management A[BP]I. That work is (perhaps unfortunately) closely linked to the new wireless stack work currently underway.

John

P.S. Ted, thanks for your calm and respectful communications!

Signature · 2 lines
John W. Linville
[email protected]
#26 lore raw
In reply to Dan Williams
Quoted 9 lines · Dan Williams

I'd point out here that one is not breaking the _distro_, as long as we assume that distros are internally consistent (which one of the major points of a distro!). What's getting broken is people who install/replace distro-provided software with their own bits. In the first case, the distro people are responsible to making sure that breakage does not occur, and that distro users are not affected. In the second case, that responsibility falls to the user who installed/replaced the distro-provided software, precisely because that software is no longer distro provided.

I'm using short-hand here. When I say breaking the _distro_, what I mean is that we are breaking the ability for users of that distro to test development kernels, which is generally considered a good thing by the kernel development community.

One could make the case that telling them that they have to download something relatively trivial, like wireless tools, is less of a big deal than something huge and very painful to build and replace, like glibc (with udev, hal, et. al, probably falling somewhere in between these two extremes), but up until now, the goal that kernel development has made is that we add new interfaces, but we try extremely hard not to break existing interfaces without a lot of notice.

Consider that that the normal, MINIMUM waiting period for *removing* an interface or functionality is six months. I would argue that this means that we should be waiting that long before making backwards-incompatible changes to an interface should be at least that long.

Quoted 6 lines

Yes, nl80211/cfg80211 (sent to netdev@ last week) is the likely successor. Please, if you have suggestions for ABI/API-proofability, review the patch and post suggestions! We all know a carefully designed is not just about the code, but about the semantics, structures, etc as well. It would be quite valuable to have everyone's input to make sure it's as future-proof as possible.

As a suggestion, has someone written up the a formal interface specification? If so, posting that for review along with the code might be a good idea.

Regards,

- Ted

#27 lore raw
Quoted 6 lines · Theodore Tso

Consider that that the normal, MINIMUM waiting period for *removing* an interface or functionality is six months. I would argue that this means that we should be waiting that long before making backwards-incompatible changes to an interface should be at least that long.

Wireless Tools with the new API were released in
April. wpa_supplicant with the new API was release May 5th. That makes it 6 months.
Wireless Tools v28 was included in Gentoo April 11th and in
Debian testing May 27th. wpa_supplicant 0.4.9 was included in Gentoo May 27th and Debian testing June 10th.

Please consider the facts before assuming.

Jean

#28 lore raw
In reply to Theodore Tso
Quoted 7 lines · Theodore Tso

John, has the wireless community considered creating a new interface which *is* carefully designed, and supporting both the new and the legacy interface for 2-3 years until all of the mainstream distributions have had a chance to cycle? It would be hard, I know, but would it be harder than some of the alternatives, and would it be worth it?

No API is guaranteed to be stable forever. And I think the
overall track record of stability for the Wireless Extensions is pretty good.
On top of that, the tools themselves *WARNS YOU* when there is
an API incompatibility, giving the user the change to correct the issue. There is not many API having such feature...

In my mind, the whole point of the GIT and RC process is to
test changes before setting them in stone, so that we have the time to correct our mistakes. I definitely think the process is working properly.
If you think I jumped the gun, consider that both FC and
Mandriva which have the right userspace bits are in RC phase (FC6-test3 and 2007-rc2), and therefore will ship about the same time 2.6.19 hits final. And Debian with the right userspace bits is supposed to be released in december (no comment).

Quoted 3 lines

Regards,

- Ted

	Have fun...

	Jean
#29 lore raw
In reply to Theodore Tso
Quoted 10 lines · Theodore Tso

There is a fundamental question hiding here, which is whether or not it is acceptable to break users who are running some large set of mainline distro's, such as RHEL 4, SLES/SLED 10, Ubuntu Dapper, et. al, but who want to upgrade to a newer 2.6 kernel?

Many users have moved to Ubuntu Dapper, or RHEL 4, or SLES/SLED 10 because they don't want to deal with a constantly changing/breaking GNOME/X world, where packages are constantly being updated and possibly breaking their desktop.

In the past, I personally tried to upgrade Red-Hat Workstation
4 with a pristine 2.6 kernel. This was far from trivial, as Red-Hat did compile their kernel with some weird options/patches, and userspace (libc) were expecting those.
On the other hand, I've been personally running the latest
2.6.X kernels on Debian stable for as long as 2.6.X was available. And, things *do* break, in the past I had trouble with module tools, I can't run devfs or udev, Pcmcia is on the verge of breaking, etc...

In other words, running a bleeding edge kernel with a
super-stable distro has never been for the casual user. And, I wonder what's the wisdom of it for the casual user, has he certainly can't use the advanced features of the new kernel unless he updates his userspace.
My main box is Debian stable with a 2.4.X kernel. For that
box, I don't see the point of going to the latest 2.6.X kernel, it would give me more trouble than benefits.

Just for kicks. Today, a new Slackware was released. And guess
what, it has Wireless Tools 28 ;-)

Jean

#30 lore raw
Quoted 3 lines · jt

Just for kicks. Today, a new Slackware was released. And guess
what, it has Wireless Tools 28 ;-)

It must be one of those day. Mandriva 2007 was just released
today, and guess what, it has Wireless Tools v28 and wpa_supplicant 0.5.5.

Jean

#31 lore raw
In reply to Jean Tourrilhes
Quoted 31 lines · Jean Tourrilhes
Quoted 10 lines · Theodore Tso

There is a fundamental question hiding here, which is whether or not it is acceptable to break users who are running some large set of mainline distro's, such as RHEL 4, SLES/SLED 10, Ubuntu Dapper, et. al, but who want to upgrade to a newer 2.6 kernel?

Many users have moved to Ubuntu Dapper, or RHEL 4, or SLES/SLED 10 because they don't want to deal with a constantly changing/breaking GNOME/X world, where packages are constantly being updated and possibly breaking their desktop.

In the past, I personally tried to upgrade Red-Hat Workstation
4 with a pristine 2.6 kernel. This was far from trivial, as Red-Hat did compile their kernel with some weird options/patches, and userspace (libc) were expecting those.
On the other hand, I've been personally running the latest
2.6.X kernels on Debian stable for as long as 2.6.X was available. And, things *do* break, in the past I had trouble with module tools, I can't run devfs or udev, Pcmcia is on the verge of breaking, etc...

In other words, running a bleeding edge kernel with a
super-stable distro has never been for the casual user. And, I wonder what's the wisdom of it for the casual user, has he certainly can't use the advanced features of the new kernel unless he updates his userspace.
My main box is Debian stable with a 2.4.X kernel. For that
box, I don't see the point of going to the latest 2.6.X kernel, it would give me more trouble than benefits.

Quoted 2 lines

Just for kicks. Today, a new Slackware was released. And guess
what, it has Wireless Tools 28 ;-)

I'm going to push wireless-tools-28 final for FC5-updates this week too.

Dan

Quoted 2 lines

Jean

#32 lore raw
In reply to Jean Tourrilhes
Quoted 4 lines · Jean Tourrilhes

In the past, I personally tried to upgrade Red-Hat Workstation
4 with a pristine 2.6 kernel. This was far from trivial, as Red-Hat did compile their kernel with some weird options/patches, and userspace (libc) were expecting those.

I (well, me and my team) am supporting a customer using a RHEL 4 userspace and a 2.6.16 kernel with Ingo's real-time patches. We just used Red Hat config file as the basis, and it wasn't that hard. There were some initrd breakages, which I've complained about in the past, but the goal is that this sort of thing is supposed to work! (And for the most part, it does).

Quoted 5 lines

On the other hand, I've been personally running the latest
2.6.X kernels on Debian stable for as long as 2.6.X was available. And, things *do* break, in the past I had trouble with module tools, I can't run devfs or udev, Pcmcia is on the verge of breaking, etc...

I'm currently using the latest 2.6 kernel with Ubuntu 6.06 (their stable release), and to date, I haven't had any problems.

Of course, that may be about to change, given that Ubuntu is shipping with wireless-tools version "27+28pre13-1ub", which I assume is a version between 27 and 28. Do you know off-hand whether this is is WE-21 capable?

- Ted

#33 lore raw
Quoted 8 lines · Theodore Tso

I'm currently using the latest 2.6 kernel with Ubuntu 6.06 (their stable release), and to date, I haven't had any problems.

Of course, that may be about to change, given that Ubuntu is shipping with wireless-tools version "27+28pre13-1ub", which I assume is a version between 27 and 28. Do you know off-hand whether this is is WE-21 capable?

No it's not.

Quoted 1 line

- Ted

Jean

#34 lore raw
In reply to Dan Williams
Quoted 5 lines · Dan Williams

But, as far as I know (please correct me if I'm wrong), that 2.6.18 doesn't actually include WE-21! [1] Somebody is trying to run out-of-kernel ipw3945 drivers using a 2.6.18 kernel from FC5 that's WE-20 only, but the driver uses WE-21?

	No, It was out-of-kernel ipw3945 + 2.6.18-git9 kernel. The

kernel was using WE-21 and the driver was using WE-20.

Jean

#35 lore raw
In reply to Theodore Tso
Quoted 9 lines · Theodore Tso
Quoted 4 lines · Dan Williams

Distributions _are_ shipping those tools already. The problem is more with older distributions where, for example, the kernel gets upgraded but other stuff does not. If a kernel upgrade happens, then the distro needs to make sure userspace works with it. That's nothing new.

Um, *which* distro's are shipping it already? RHEL4? SLES10? I thought we saw a note saying that even Debian **unstable** didn't have a new enough version of the wireless-tools....

Debian is currently in (pre-) freeze mode, and new upstream versions of core packages are uploaded very very carefully. To see whats going on with the wireless-tools package, see [1, 2].

[3] shows that 29pre10 was uploaded to experimental on Thu, 4 May 2006

So you are right, debian unstable doesn't ship the latest version, because that would have the potential to make problems with the release of debian 4.0 (aka etch). The updated package however is there. If this package should go to etch, because 2.6.18 is likely to be the kernel etch ships, then both the maintainer and the release team needs to be convinced about that.

[1] http://packages.debian.org/wireless-tools
[2] http://packages.qa.debian.org/wireless-tools
[3] http://packages.qa.debian.org/w/wireless-tools/news/20060504T210628Z.html

Signature · 2 lines
Gruesse/greetings,
Reinhard Tartler, KeyID 945348A4
#36 lore raw
In reply to Andrew Morton
Quoted 6 lines · Andrew Morton

Jean, John: the amount of trouble which this change is causing is quite high considering that we're not even at -rc1 yet. It's going to get worse.

It doesn't sound like it'll be too hard to arrange for the kernel to continue to work correctly with old userspace?

Actually, I have the perfect solution. We ship a new version
of the Wireless Tools and wpa_supplicant alongside the kernel, and install those when the user install the new kernel. That way, we make sure they always use the correct version of userspace with the kernel.
Regards,

Jean

#37 lore raw
Quoted 5 lines · Jean Tourrilhes
Quoted 4 lines · Andrew Morton

Jean, John: the amount of trouble which this change is causing is quite high considering that we're not even at -rc1 yet. It's going to get worse.

Quoted 7 lines
Quoted 2 lines

It doesn't sound like it'll be too hard to arrange for the kernel to continue to work correctly with old userspace?

Actually, I have the perfect solution. We ship a new version
of the Wireless Tools and wpa_supplicant alongside the kernel, and install those when the user install the new kernel. That way, we make sure they always use the correct version of userspace with the kernel.

Like udev :) And libsysfs :)

Quoted 4 lines

Regards,

Jean

#38 lore raw
Quoted 8 lines · Dan Williams
Quoted 5 lines · Jean Tourrilhes

Actually, I have the perfect solution. We ship a new version
of the Wireless Tools and wpa_supplicant alongside the kernel, and install those when the user install the new kernel. That way, we make sure they always use the correct version of userspace with the kernel.

Like udev :) And libsysfs :)

Wait, wait, this starts to look too much like *BSD ;-)

Jean

#39 lore raw
In reply to Alessandro Suardi
Quoted 7 lines · Alessandro Suardi

we've been just through an email thread where it has been determined that wpa_supplicant 0.4.9 (I would assume that 0.5.5 is also okay) and wireless-tools from Jean's latest tarball are necessary to work with the recent wireless extensions v21 that have been merged in.

What wireless-tools are you using ?

This must be considered a kernel bug - it's not allowed to break userspace compatibility in a stable series.

Lee

#40 lore raw
Quoted 11 lines · Lee Revell
Quoted 7 lines · Alessandro Suardi

we've been just through an email thread where it has been determined that wpa_supplicant 0.4.9 (I would assume that 0.5.5 is also okay) and wireless-tools from Jean's latest tarball are necessary to work with the recent wireless extensions v21 that have been merged in.

What wireless-tools are you using ?

This must be considered a kernel bug - it's not allowed to break userspace compatibility in a stable series.

But there is no development series. The closest thing we have is the merge window after each release -- which is exactly when this issue revealed itself.

Wireless in general (and the wireless extensions api in particular) is a bit of a 'whipping boy' in the Linux world. OK, we suck. Everyone wants to display their wisdom by telling us how much we suck! We know all about it...

We have sucked, and we continue to suck -- and we are working on it. But, we are not going to be able to whip-up this omelette without breaking a few eggs. If we can't do that during the merge windows, WHEN CAN WE DO IT?

John

Signature · 2 lines
John W. Linville
[email protected]
#41 lore raw

John,

Has someone documented somewhere what all of the constraints
are? I have gathered from various messages that part of the problem is that different drivers and different userspace tools have a different idea of what the structures are and what various size fields mean, and that trying to coordinate changes between the kernel, multiple userspace tools, and some very popular out-of-tree drivers (some of which have been kept out of the kernel for issues that I don't agree with, but which the GPL purists go balistic over), that you feel that the problem is over-constrained.

Is there a document or an e-mail message that in one place
describes what the current situation is, why it sucks, what the changes are, and what eggs are getting broken and why it's better than where we are today?

Thanks, regards,

- Ted

#42 lore raw
In reply to John W. Linville
Quoted 22 lines · John W. Linville
Quoted 11 lines · Lee Revell
Quoted 7 lines · Alessandro Suardi

we've been just through an email thread where it has been determined that wpa_supplicant 0.4.9 (I would assume that 0.5.5 is also okay) and wireless-tools from Jean's latest tarball are necessary to work with the recent wireless extensions v21 that have been merged in.

What wireless-tools are you using ?

This must be considered a kernel bug - it's not allowed to break userspace compatibility in a stable series.

But there is no development series. The closest thing we have is the merge window after each release -- which is exactly when this issue revealed itself.

Wireless in general (and the wireless extensions api in particular) is a bit of a 'whipping boy' in the Linux world. OK, we suck. Everyone wants to display their wisdom by telling us how much we suck! We know all about it...

Quoted 4 lines

We have sucked, and we continue to suck -- and we are working on it. But, we are not going to be able to whip-up this omelette without breaking a few eggs. If we can't do that during the merge windows, WHEN CAN WE DO IT?

This is a question for Linus, it's his rule...

Lee

#43 lore raw
Quoted 7 lines · Lee Revell
Quoted 5 lines

We have sucked, and we continue to suck -- and we are working on it. But, we are not going to be able to whip-up this omelette without breaking a few eggs. If we can't do that during the merge windows, WHEN CAN WE DO IT?

This is a question for Linus, it's his rule...

In general, the answer to "when can we break user space" is very simple.

Never.

It's just not acceptable. We maintain old interfaces for years (and in some cases, well over a decade by now), simply because the pain from not doing so is horrendous, and it makes debugging impossible. You get into situations where users need to upgrade to tools that don't work with older kernels, and can thus not downgrade, etc etc.

That said, system tools have been given some dispensation. Not a lot, and sometimes they take more liberties than they are given - udev in particular had too many problems, and some people who maintained it seemed to think that they had the "right" to break things as much as they did. I think that got cleared up, but the point is: system tools sometimes cannot avoid some issues, but damn it, breaking them "just because it fixes things" is NOT ACCEPTABLE.

So even if it's a case of "..but it fixes a problem", that in itself is not a valid reason for breaking a tool that used to work. It really _has_ to be: "It fixes a serious problem, and we tried hard as _hell_ to avoid any breakage at all, and it wasn't possible, but this is a one-time event".

Quite frankly, the parts of the discussion I have seen does not at all imply to me that people tried as hard as possible. For example, I saw Jean say that "The wireless tools themselves told about the version mismatch", as if that had ANY RELEVANCE AT ALL!

If the kernel knows about the version number, it should make sure that the interfaces are honored. Yes, it often means some ugly code at the interface layer, but dammit, that's the job of the kernel. It's the whole reason why we have some of the abstraction layers we do in the kernel. The original reason the readdir() function got a "filldir_t" callback was simply the fact that we needed to support two different formats for it. The reason we have the whole internal "struct kstat" is exactly the same.

We don't do this version skew dance. If we need to break something, it had better be some damn substantial reasons, and even then we're generally better off supporting _both_ interfaces for a while (perhaps using a version code), and then marking the old one deprecated.

Ugly? Yes, often. But _users_ do matter more. Really.

Linus

#44 lore raw
Quoted 4 lines · Linus Torvalds

We don't do this version skew dance. If we need to break something, it had better be some damn substantial reasons, and even then we're generally better off supporting _both_ interfaces for a while (perhaps using a version code), and then marking the old one deprecated.

FWIW, this clean-up is not intended to break older binaries. It is intended to standardize driver implementations of the WE API. Breakage is (merely!) a side-effect...

The overall purpose of the WE-21 patch was to continue disambiguating how drivers implement the WE API. This is intended to (hopefully) avoid strange, hidden bugs lurking out there in driver/tool interaction, especially for tools other than wireless-tools (e.g. NetworkManager, wpa_supplicant, etc). In this case, it looks like maybe some older versions of these tools were effectively exploting the strange, hidden bugs... :-(

It seems there were a few genuine bugs which crept into the WE-21 implementation. Jean has posted fixes for those today. It looks like those patches get things working again when combined with updated tools.

Today's news seems to indicate that at least the major distros are already shipping the updated tools, or on the verge of shipping them. The window of breakage for most users looks like it will be fairly small, no matter what action taken.

The more we can clean-up the WE API, the easier it will be to implement the cfg80211 WE compatibility layer intended for wireless-dev. I think WE-21 makes things better in that respect.

Finally, I already scaled-back Jean's original WE-21 patch. I only anticipate minor bug fixes for WE from now on, with nl80211/cfg80211 as the heir-apparent.

With all that said, I'd prefer to keep the existing WE-21 patches and add Jean's fixes ASAP. Is this acceptable? If not, I'll submit the reversions to Jeff ASAP.

Suggestions?

John

Signature · 2 lines
John W. Linville
[email protected]
#45 lore raw
Quoted 11 lines · John W. Linville

The more we can clean-up the WE API, the easier it will be to implement the cfg80211 WE compatibility layer intended for wireless-dev. I think WE-21 makes things better in that respect.

Finally, I already scaled-back Jean's original WE-21 patch. I only anticipate minor bug fixes for WE from now on, with nl80211/cfg80211 as the heir-apparent.

With all that said, I'd prefer to keep the existing WE-21 patches and add Jean's fixes ASAP. Is this acceptable? If not, I'll submit the reversions to Jeff ASAP.

I for one don't want to change the userspace ABI for this... If I had realized the userspace ABI was changing (<- my fault), I wouldn't have merged the changes in the first place.

Jeff

#46 lore raw
Quoted 13 lines · Jeff Garzik
Quoted 11 lines · John W. Linville

The more we can clean-up the WE API, the easier it will be to implement the cfg80211 WE compatibility layer intended for wireless-dev. I think WE-21 makes things better in that respect.

Finally, I already scaled-back Jean's original WE-21 patch. I only anticipate minor bug fixes for WE from now on, with nl80211/cfg80211 as the heir-apparent.

With all that said, I'd prefer to keep the existing WE-21 patches and add Jean's fixes ASAP. Is this acceptable? If not, I'll submit the reversions to Jeff ASAP.

Quoted 5 lines

I for one don't want to change the userspace ABI for this... If I had realized the userspace ABI was changing (<- my fault), I wouldn't have merged the changes in the first place.

Jeff

Jeff,

This was discussed publically on this mailing list last
January.

		http://marc.theaimsgroup.com/?t=113710086000009&r=1&w=2
	I made clear at that time that I did not like this change

because the userspace ABI would change, but I was overruled by a wide consensus. So, don't blame me.
This change was rediscussed and reconfirmed at the Wireless
Summit. I only tried to make such change smooth, but it was not my idea.

Now it's too late, those changes have propagated to userspace
tools, and are now shipping in some actual release of some distro. So, what are we going to say to Mandriva 2007 and FC6 users, to revert back to an *older* version of the tools ?
Because userspace has already been updated, we have only two
options, merge it now, or in 2.6.20.

Regards,

Jean

#47 lore raw
Quoted 6 lines · Jean Tourrilhes

Now it's too late, those changes have propagated to userspace
tools, and are now shipping in some actual release of some distro. So, what are we going to say to Mandriva 2007 and FC6 users, to revert back to an *older* version of the tools ?
Because userspace has already been updated, we have only two
options, merge it now, or in 2.6.20.

If the choice is between the ABI used by a bunch of distros in production, and the ABI used by two re-release distros, I think the choice is obvious...

Jeff

#48 lore raw
Quoted 8 lines · Jeff Garzik
Quoted 6 lines · Jean Tourrilhes

Now it's too late, those changes have propagated to userspace
tools, and are now shipping in some actual release of some distro. So, what are we going to say to Mandriva 2007 and FC6 users, to revert back to an *older* version of the tools ?
Because userspace has already been updated, we have only two
options, merge it now, or in 2.6.20.

Quoted 5 lines

If the choice is between the ABI used by a bunch of distros in production, and the ABI used by two re-release distros, I think the choice is obvious...

Jeff

Which means 2.6.20. Case closed.

Jean

#49 lore raw
In reply to Jeff Garzik
Quoted 8 lines · Jeff Garzik
Quoted 6 lines · Jean Tourrilhes

Now it's too late, those changes have propagated to userspace
tools, and are now shipping in some actual release of some distro. So, what are we going to say to Mandriva 2007 and FC6 users, to revert back to an *older* version of the tools ?
Because userspace has already been updated, we have only two
options, merge it now, or in 2.6.20.

Quoted 5 lines

If the choice is between the ABI used by a bunch of distros in production, and the ABI used by two re-release distros, I think the choice is obvious...

Jeff

And remember that it's not *terminally* broken, the user can
still use the old tools with WE-21, he just need to add a dummy character at the end of the ESSID string.
It's probably annoying, but it's not the end of the world.

Jean

#50 lore raw
In reply to Jeff Garzik
Quoted 11 lines · Jeff Garzik
Quoted 6 lines · Jean Tourrilhes

Now it's too late, those changes have propagated to userspace
tools, and are now shipping in some actual release of some distro. So, what are we going to say to Mandriva 2007 and FC6 users, to revert back to an *older* version of the tools ?
Because userspace has already been updated, we have only two
options, merge it now, or in 2.6.20.

If the choice is between the ABI used by a bunch of distros in production, and the ABI used by two re-release distros, I think the choice is obvious...

I (grudgingly?) agree...

Unfortunately, I don't see any way to "fix" WE-21 without similarly breaking wireless-tools 29 and other "WE-21 aware" apps. And since I'll bet that the various WE-aware apps have checks like "if WE > 20" for managing ESSID length settings, we may have painted ourselves into a korner (sic).

I.E. With "WE-21 aware" tools already in the wild, it isn't now clear to me how WE can evolve any further than WE-20. Unless the ESSID length changes in WE-21 is somehow deemed acceptable for 2.6.20 or some later kernel version, I don't see how WE can continue to change.

Wireless developers, it's time to get d80211 ready to merge...or figure-out how to get nl80211 on the current stack...hmmm...

John

---

The following changes since commit 8ccb3dcd1f8e80e8702642e1de26541b52f6bb7c:
Linus Torvalds:
x86: Fix booting with "no387 nofxsr"

are found in the git repository at:

git://git.kernel.org/pub/scm/linux/kernel/git/linville/wireless-2.6.git we21-revert

John W. Linville:
wireless: revert WE-21 patches

 drivers/net/wireless/airo.c                 |   19 ++++----
 drivers/net/wireless/atmel.c                |   18 ++++---
 drivers/net/wireless/bcm43xx/bcm43xx_wx.c   |    2 -
 drivers/net/wireless/hostap/hostap_ioctl.c  |   10 ++--
 drivers/net/wireless/ipw2100.c              |   14 +++---
 drivers/net/wireless/ipw2200.c              |   16 ++++--
 drivers/net/wireless/orinoco.c              |   10 ++--
 drivers/net/wireless/prism54/isl_ioctl.c    |   16 +++---
 drivers/net/wireless/ray_cs.c               |    2 -
 drivers/net/wireless/wl3501_cs.c            |    6 +-
 drivers/net/wireless/zd1201.c               |    2 -
 drivers/net/wireless/zd1211rw/zd_netdev.c   |    2 -
 include/linux/netdevice.h                   |    1 
 include/linux/wireless.h                    |   24 ++--------
 net/core/net-sysfs.c                        |    5 ++
 net/core/wireless.c                         |   67 ++++++++++-----------------
 net/ieee80211/softmac/ieee80211softmac_wx.c |    8 ++-
 17 files changed, 98 insertions(+), 124 deletions(-)
drivers/net/wireless/airo.c +9 -10
index 39d0934..1864b07 100644
@@ -5883,7 +5883,7 @@ static int airo_set_essid(struct net_dev
 		int	index = (dwrq->flags & IW_ENCODE_INDEX) - 1;
 
 		/* Check the size of the string */
-		if(dwrq->length > IW_ESSID_MAX_SIZE) {
+		if(dwrq->length > IW_ESSID_MAX_SIZE+1) {
 			return -E2BIG ;
 		}
 		/* Check if index is valid */
@@ -5895,7 +5895,7 @@ static int airo_set_essid(struct net_dev
 		memset(SSID_rid.ssids[index].ssid, 0,
 		       sizeof(SSID_rid.ssids[index].ssid));
 		memcpy(SSID_rid.ssids[index].ssid, extra, dwrq->length);
-		SSID_rid.ssids[index].len = dwrq->length;
+		SSID_rid.ssids[index].len = dwrq->length - 1;
 	}
 	SSID_rid.len = sizeof(SSID_rid);
 	/* Write it to the card */
@@ -6005,7 +6005,7 @@ static int airo_set_nick(struct net_devi
 	struct airo_info *local = dev->priv;
 
 	/* Check the size of the string */
-	if(dwrq->length > 16) {
+	if(dwrq->length > 16 + 1) {
 		return -E2BIG;
 	}
 	readConfigRid(local, 1);
@@ -6030,7 +6030,7 @@ static int airo_get_nick(struct net_devi
 	readConfigRid(local, 1);
 	strncpy(extra, local->config.nodeName, 16);
 	extra[16] = '\0';
-	dwrq->length = strlen(extra);
+	dwrq->length = strlen(extra) + 1;
 
 	return 0;
 }
@@ -6782,9 +6782,9 @@ static int airo_set_retry(struct net_dev
 	}
 	readConfigRid(local, 1);
 	if(vwrq->flags & IW_RETRY_LIMIT) {
-		if(vwrq->flags & IW_RETRY_LONG)
+		if(vwrq->flags & IW_RETRY_MAX)
 			local->config.longRetryLimit = vwrq->value;
-		else if (vwrq->flags & IW_RETRY_SHORT)
+		else if (vwrq->flags & IW_RETRY_MIN)
 			local->config.shortRetryLimit = vwrq->value;
 		else {
 			/* No modifier : set both */
@@ -6820,14 +6820,14 @@ static int airo_get_retry(struct net_dev
 	if((vwrq->flags & IW_RETRY_TYPE) == IW_RETRY_LIFETIME) {
 		vwrq->flags = IW_RETRY_LIFETIME;
 		vwrq->value = (int)local->config.txLifetime * 1024;
-	} else if((vwrq->flags & IW_RETRY_LONG)) {
-		vwrq->flags = IW_RETRY_LIMIT | IW_RETRY_LONG;
+	} else if((vwrq->flags & IW_RETRY_MAX)) {
+		vwrq->flags = IW_RETRY_LIMIT | IW_RETRY_MAX;
 		vwrq->value = (int)local->config.longRetryLimit;
 	} else {
 		vwrq->flags = IW_RETRY_LIMIT;
 		vwrq->value = (int)local->config.shortRetryLimit;
 		if((int)local->config.shortRetryLimit != (int)local->config.longRetryLimit)
-			vwrq->flags |= IW_RETRY_SHORT;
+			vwrq->flags |= IW_RETRY_MIN;
 	}
 
 	return 0;
@@ -7005,7 +7005,6 @@ static int airo_set_power(struct net_dev
 			local->config.rmode |= RXMODE_BC_MC_ADDR;
 			set_bit (FLAG_COMMIT, &local->flags);
 		case IW_POWER_ON:
-			/* This is broken, fixme ;-) */
 			break;
 		default:
 			return -EINVAL;
drivers/net/wireless/atmel.c +9 -9
index 0fc267d..995c7be 100644
@@ -1656,13 +1656,13 @@ static int atmel_set_essid(struct net_de
 		priv->connect_to_any_BSS = 0;
 
 		/* Check the size of the string */
-		if (dwrq->length > MAX_SSID_LENGTH)
+		if (dwrq->length > MAX_SSID_LENGTH + 1)
 			 return -E2BIG;
 		if (index != 0)
 			return -EINVAL;
 
-		memcpy(priv->new_SSID, extra, dwrq->length);
-		priv->new_SSID_size = dwrq->length;
+		memcpy(priv->new_SSID, extra, dwrq->length - 1);
+		priv->new_SSID_size = dwrq->length - 1;
 	}
 
 	return -EINPROGRESS;
@@ -2120,9 +2120,9 @@ static int atmel_set_retry(struct net_de
 	struct atmel_private *priv = netdev_priv(dev);
 
 	if (!vwrq->disabled && (vwrq->flags & IW_RETRY_LIMIT)) {
-		if (vwrq->flags & IW_RETRY_LONG)
+		if (vwrq->flags & IW_RETRY_MAX)
 			priv->long_retry = vwrq->value;
-		else if (vwrq->flags & IW_RETRY_SHORT)
+		else if (vwrq->flags & IW_RETRY_MIN)
 			priv->short_retry = vwrq->value;
 		else {
 			/* No modifier : set both */
@@ -2144,15 +2144,15 @@ static int atmel_get_retry(struct net_de
 
 	vwrq->disabled = 0;      /* Can't be disabled */
 
-	/* Note : by default, display the short retry number */
-	if (vwrq->flags & IW_RETRY_LONG) {
-		vwrq->flags = IW_RETRY_LIMIT | IW_RETRY_LONG;
+	/* Note : by default, display the min retry number */
+	if (vwrq->flags & IW_RETRY_MAX) {
+		vwrq->flags = IW_RETRY_LIMIT | IW_RETRY_MAX;
 		vwrq->value = priv->long_retry;
 	} else {
 		vwrq->flags = IW_RETRY_LIMIT;
 		vwrq->value = priv->short_retry;
 		if (priv->long_retry != priv->short_retry)
-			vwrq->flags |= IW_RETRY_SHORT;
+			vwrq->flags |= IW_RETRY_MIN;
 	}
 
 	return 0;
drivers/net/wireless/bcm43xx/bcm43xx_wx.c +1 -1
index 9b7b15c..888077f 100644
@@ -334,7 +334,7 @@ static int bcm43xx_wx_get_nick(struct ne
 	size_t len;
 
 	mutex_lock(&bcm->mutex);
-	len = strlen(bcm->nick);
+	len = strlen(bcm->nick) + 1;
 	memcpy(extra, bcm->nick, len);
 	data->data.length = (__u16)len;
 	data->data.flags = 1;
drivers/net/wireless/hostap/hostap_ioctl.c +5 -5
index d061fb3..7a49785 100644
@@ -1412,9 +1412,9 @@ #if 0
 	/* what could be done, if firmware would support this.. */
 
 	if (rrq->flags & IW_RETRY_LIMIT) {
-		if (rrq->flags & IW_RETRY_LONG)
+		if (rrq->flags & IW_RETRY_MAX)
 			HFA384X_RID_LONGRETRYLIMIT = rrq->value;
-		else if (rrq->flags & IW_RETRY_SHORT)
+		else if (rrq->flags & IW_RETRY_MIN)
 			HFA384X_RID_SHORTRETRYLIMIT = rrq->value;
 		else {
 			HFA384X_RID_LONGRETRYLIMIT = rrq->value;
@@ -1468,14 +1468,14 @@ static int prism2_ioctl_giwretry(struct 
 				rrq->value = le16_to_cpu(altretry);
 			else
 				rrq->value = local->manual_retry_count;
-		} else if ((rrq->flags & IW_RETRY_LONG)) {
-			rrq->flags = IW_RETRY_LIMIT | IW_RETRY_LONG;
+		} else if ((rrq->flags & IW_RETRY_MAX)) {
+			rrq->flags = IW_RETRY_LIMIT | IW_RETRY_MAX;
 			rrq->value = longretry;
 		} else {
 			rrq->flags = IW_RETRY_LIMIT;
 			rrq->value = shortretry;
 			if (shortretry != longretry)
-				rrq->flags |= IW_RETRY_SHORT;
+				rrq->flags |= IW_RETRY_MIN;
 		}
 	}
 	return 0;
drivers/net/wireless/ipw2100.c +7 -7
index 599e2fe..3e01c3d 100644
@@ -6972,7 +6972,7 @@ static int ipw2100_wx_set_essid(struct n
 	}
 
 	if (wrqu->essid.flags && wrqu->essid.length) {
-		length = wrqu->essid.length;
+		length = wrqu->essid.length - 1;
 		essid = extra;
 	}
 
@@ -7065,7 +7065,7 @@ static int ipw2100_wx_get_nick(struct ne
 
 	struct ipw2100_priv *priv = ieee80211_priv(dev);
 
-	wrqu->data.length = strlen(priv->nick);
+	wrqu->data.length = strlen(priv->nick) + 1;
 	memcpy(extra, priv->nick, wrqu->data.length);
 	wrqu->data.flags = 1;	/* active */
 
@@ -7357,14 +7357,14 @@ static int ipw2100_wx_set_retry(struct n
 		goto done;
 	}
 
-	if (wrqu->retry.flags & IW_RETRY_SHORT) {
+	if (wrqu->retry.flags & IW_RETRY_MIN) {
 		err = ipw2100_set_short_retry(priv, wrqu->retry.value);
 		IPW_DEBUG_WX("SET Short Retry Limit -> %d \n",
 			     wrqu->retry.value);
 		goto done;
 	}
 
-	if (wrqu->retry.flags & IW_RETRY_LONG) {
+	if (wrqu->retry.flags & IW_RETRY_MAX) {
 		err = ipw2100_set_long_retry(priv, wrqu->retry.value);
 		IPW_DEBUG_WX("SET Long Retry Limit -> %d \n",
 			     wrqu->retry.value);
@@ -7397,14 +7397,14 @@ static int ipw2100_wx_get_retry(struct n
 	if ((wrqu->retry.flags & IW_RETRY_TYPE) == IW_RETRY_LIFETIME)
 		return -EINVAL;
 
-	if (wrqu->retry.flags & IW_RETRY_LONG) {
-		wrqu->retry.flags = IW_RETRY_LIMIT | IW_RETRY_LONG;
+	if (wrqu->retry.flags & IW_RETRY_MAX) {
+		wrqu->retry.flags = IW_RETRY_LIMIT | IW_RETRY_MAX;
 		wrqu->retry.value = priv->long_retry_limit;
 	} else {
 		wrqu->retry.flags =
 		    (priv->short_retry_limit !=
 		     priv->long_retry_limit) ?
-		    IW_RETRY_LIMIT | IW_RETRY_SHORT : IW_RETRY_LIMIT;
+		    IW_RETRY_LIMIT | IW_RETRY_MIN : IW_RETRY_LIMIT;
 
 		wrqu->retry.value = priv->short_retry_limit;
 	}
drivers/net/wireless/ipw2200.c +9 -7
index 5685d7b..7358664 100644
@@ -8875,6 +8875,8 @@ static int ipw_wx_set_essid(struct net_d
         }
 
 	length = min((int)wrqu->essid.length, IW_ESSID_MAX_SIZE);
+	if (!extra[length - 1])
+		length--;
 
 	priv->config |= CFG_STATIC_ESSID;
 
@@ -8951,7 +8953,7 @@ static int ipw_wx_get_nick(struct net_de
 	struct ipw_priv *priv = ieee80211_priv(dev);
 	IPW_DEBUG_WX("Getting nick\n");
 	mutex_lock(&priv->mutex);
-	wrqu->data.length = strlen(priv->nick);
+	wrqu->data.length = strlen(priv->nick) + 1;
 	memcpy(extra, priv->nick, wrqu->data.length);
 	wrqu->data.flags = 1;	/* active */
 	mutex_unlock(&priv->mutex);
@@ -9274,9 +9276,9 @@ static int ipw_wx_set_retry(struct net_d
 		return -EINVAL;
 
 	mutex_lock(&priv->mutex);
-	if (wrqu->retry.flags & IW_RETRY_SHORT)
+	if (wrqu->retry.flags & IW_RETRY_MIN)
 		priv->short_retry_limit = (u8) wrqu->retry.value;
-	else if (wrqu->retry.flags & IW_RETRY_LONG)
+	else if (wrqu->retry.flags & IW_RETRY_MAX)
 		priv->long_retry_limit = (u8) wrqu->retry.value;
 	else {
 		priv->short_retry_limit = (u8) wrqu->retry.value;
@@ -9305,11 +9307,11 @@ static int ipw_wx_get_retry(struct net_d
 		return -EINVAL;
 	}
 
-	if (wrqu->retry.flags & IW_RETRY_LONG) {
-		wrqu->retry.flags = IW_RETRY_LIMIT | IW_RETRY_LONG;
+	if (wrqu->retry.flags & IW_RETRY_MAX) {
+		wrqu->retry.flags = IW_RETRY_LIMIT | IW_RETRY_MAX;
 		wrqu->retry.value = priv->long_retry_limit;
-	} else if (wrqu->retry.flags & IW_RETRY_SHORT) {
-		wrqu->retry.flags = IW_RETRY_LIMIT | IW_RETRY_SHORT;
+	} else if (wrqu->retry.flags & IW_RETRY_MIN) {
+		wrqu->retry.flags = IW_RETRY_LIMIT | IW_RETRY_MIN;
 		wrqu->retry.value = priv->short_retry_limit;
 	} else {
 		wrqu->retry.flags = IW_RETRY_LIMIT;
drivers/net/wireless/orinoco.c +5 -5
index 9e19a96..1840b69 100644
@@ -3037,7 +3037,7 @@ static int orinoco_ioctl_getessid(struct
 	}
 
 	erq->flags = 1;
-	erq->length = strlen(essidbuf);
+	erq->length = strlen(essidbuf) + 1;
 
 	return 0;
 }
@@ -3078,7 +3078,7 @@ static int orinoco_ioctl_getnick(struct 
 	memcpy(nickbuf, priv->nick, IW_ESSID_MAX_SIZE+1);
 	orinoco_unlock(priv, &flags);
 
-	nrq->length = strlen(nickbuf);
+	nrq->length = strlen(nickbuf)+1;
 
 	return 0;
 }
@@ -3575,14 +3575,14 @@ static int orinoco_ioctl_getretry(struct
 		rrq->value = lifetime * 1000;	/* ??? */
 	} else {
 		/* By default, display the min number */
-		if ((rrq->flags & IW_RETRY_LONG)) {
-			rrq->flags = IW_RETRY_LIMIT | IW_RETRY_LONG;
+		if ((rrq->flags & IW_RETRY_MAX)) {
+			rrq->flags = IW_RETRY_LIMIT | IW_RETRY_MAX;
 			rrq->value = long_limit;
 		} else {
 			rrq->flags = IW_RETRY_LIMIT;
 			rrq->value = short_limit;
 			if(short_limit != long_limit)
-				rrq->flags |= IW_RETRY_SHORT;
+				rrq->flags |= IW_RETRY_MIN;
 		}
 	}
 
drivers/net/wireless/prism54/isl_ioctl.c +8 -8
index 286325c..c09fbf7 100644
@@ -742,9 +742,9 @@ prism54_set_essid(struct net_device *nde
 
 	/* Check if we were asked for `any' */
 	if (dwrq->flags && dwrq->length) {
-		if (dwrq->length > 32)
+		if (dwrq->length > min(33, IW_ESSID_MAX_SIZE + 1))
 			return -E2BIG;
-		essid.length = dwrq->length;
+		essid.length = dwrq->length - 1;
 		memcpy(essid.octets, extra, dwrq->length);
 	} else
 		essid.length = 0;
@@ -814,7 +814,7 @@ prism54_get_nick(struct net_device *ndev
 	dwrq->length = 0;
 
 	down_read(&priv->mib_sem);
-	dwrq->length = strlen(priv->nickname);
+	dwrq->length = strlen(priv->nickname) + 1;
 	memcpy(extra, priv->nickname, dwrq->length);
 	up_read(&priv->mib_sem);
 
@@ -992,9 +992,9 @@ prism54_set_retry(struct net_device *nde
 		return -EINVAL;
 
 	if (vwrq->flags & IW_RETRY_LIMIT) {
-		if (vwrq->flags & IW_RETRY_SHORT)
+		if (vwrq->flags & IW_RETRY_MIN)
 			slimit = vwrq->value;
-		else if (vwrq->flags & IW_RETRY_LONG)
+		else if (vwrq->flags & IW_RETRY_MAX)
 			llimit = vwrq->value;
 		else {
 			/* we are asked to set both */
@@ -1035,18 +1035,18 @@ prism54_get_retry(struct net_device *nde
 		    mgt_get_request(priv, DOT11_OID_MAXTXLIFETIME, 0, NULL, &r);
 		vwrq->value = r.u * 1024;
 		vwrq->flags = IW_RETRY_LIFETIME;
-	} else if ((vwrq->flags & IW_RETRY_LONG)) {
+	} else if ((vwrq->flags & IW_RETRY_MAX)) {
 		/* we are asked for the long retry limit */
 		rvalue |=
 		    mgt_get_request(priv, DOT11_OID_LONGRETRIES, 0, NULL, &r);
 		vwrq->value = r.u;
-		vwrq->flags = IW_RETRY_LIMIT | IW_RETRY_LONG;
+		vwrq->flags = IW_RETRY_LIMIT | IW_RETRY_MAX;
 	} else {
 		/* default. get the  short retry limit */
 		rvalue |=
 		    mgt_get_request(priv, DOT11_OID_SHORTRETRIES, 0, NULL, &r);
 		vwrq->value = r.u;
-		vwrq->flags = IW_RETRY_LIMIT | IW_RETRY_SHORT;
+		vwrq->flags = IW_RETRY_LIMIT | IW_RETRY_MIN;
 	}
 
 	return rvalue;
drivers/net/wireless/ray_cs.c +1 -1
index e82548e..4574290 100644
@@ -1173,7 +1173,7 @@ static int ray_set_essid(struct net_devi
 		return -EOPNOTSUPP;
 	} else {
 		/* Check the size of the string */
-		if(dwrq->length > IW_ESSID_MAX_SIZE) {
+		if(dwrq->length > IW_ESSID_MAX_SIZE + 1) {
 			return -E2BIG;
 		}
 
drivers/net/wireless/wl3501_cs.c +3 -3
index e3ae5f6..e0d294c 100644
@@ -1802,15 +1802,15 @@ static int wl3501_get_retry(struct net_d
 				      &retry, sizeof(retry));
 	if (rc)
 		goto out;
-	if (wrqu->retry.flags & IW_RETRY_LONG) {
-		wrqu->retry.flags = IW_RETRY_LIMIT | IW_RETRY_LONG;
+	if (wrqu->retry.flags & IW_RETRY_MAX) {
+		wrqu->retry.flags = IW_RETRY_LIMIT | IW_RETRY_MAX;
 		goto set_value;
 	}
 	rc = wl3501_get_mib_value(this, WL3501_MIB_ATTR_SHORT_RETRY_LIMIT,
 				  &retry, sizeof(retry));
 	if (rc)
 		goto out;
-	wrqu->retry.flags = IW_RETRY_LIMIT | IW_RETRY_SHORT;
+	wrqu->retry.flags = IW_RETRY_LIMIT | IW_RETRY_MIN;
 set_value:
 	wrqu->retry.value = retry;
 	wrqu->retry.disabled = 0;
drivers/net/wireless/zd1201.c +1 -1
index 80af9a9..f50ec10 100644
@@ -1218,7 +1218,7 @@ static int zd1201_set_essid(struct net_d
 		return -EINVAL;
 	if (data->length < 1)
 		data->length = 1;
-	zd->essidlen = data->length;
+	zd->essidlen = data->length-1;
 	memset(zd->essid, 0, IW_ESSID_MAX_SIZE+1);
 	memcpy(zd->essid, essid, data->length);
 	return zd1201_join(zd, zd->essid, zd->essidlen);
drivers/net/wireless/zd1211rw/zd_netdev.c +1 -1
index af3a7b3..440ef24 100644
@@ -82,7 +82,7 @@ static int iw_get_nick(struct net_device
 		       union iwreq_data *req, char *extra)
 {
 	strcpy(extra, "zd1211");
-	req->data.length = strlen(extra);
+	req->data.length = strlen(extra) + 1;
 	req->data.flags = 1;
 	return 0;
 }
include/linux/netdevice.h +1 -0
index 9264139..f159c72 100644
@@ -334,6 +334,7 @@ #define NETIF_F_ALL_CSUM	(NETIF_F_IP_CSU
 
 
 	struct net_device_stats* (*get_stats)(struct net_device *dev);
+	struct iw_statistics*	(*get_wireless_stats)(struct net_device *dev);
 
 	/* List of functions to handle Wireless Extensions (instead of ioctl).
 	 * See <net/iw_handler.h> for details. Jean II */
include/linux/wireless.h +4 -20
index a50a013..1358856 100644
@@ -1,7 +1,7 @@
 /*
  * This file define a set of standard wireless extensions
  *
- * Version :	21	14.3.06
+ * Version :	20	17.2.06
  *
  * Authors :	Jean Tourrilhes - HPL - <[email protected]>
  * Copyright (c) 1997-2006 Jean Tourrilhes, All Rights Reserved.
@@ -69,14 +69,9 @@ #define _LINUX_WIRELESS_H
 
 /***************************** INCLUDES *****************************/
 
-/* This header is used in user-space, therefore need to be sanitised
- * for that purpose. Those includes are usually not compatible with glibc.
- * To know which includes to use in user-space, check iwlib.h. */
-#ifdef __KERNEL__
 #include <linux/types.h>		/* for "caddr_t" et al		*/
 #include <linux/socket.h>		/* for "struct sockaddr" et al	*/
 #include <linux/if.h>			/* for IFNAMSIZ and co... */
-#endif	/* __KERNEL__ */
 
 /***************************** VERSION *****************************/
 /*
@@ -85,7 +80,7 @@ #endif	/* __KERNEL__ */
  * (there is some stuff that will be added in the future...)
  * I just plan to increment with each new version.
  */
-#define WIRELESS_EXT	21
+#define WIRELESS_EXT	20
 
 /*
  * Changes :
@@ -213,14 +208,6 @@ #define WIRELESS_EXT	21
  * V19 to V20
  * ----------
  *	- RtNetlink requests support (SET/GET)
- *
- * V20 to V21
- * ----------
- *	- Remove (struct net_device *)->get_wireless_stats()
- *	- Change length in ESSID and NICK to strlen() instead of strlen()+1
- *	- Add IW_RETRY_SHORT/IW_RETRY_LONG retry modifiers
- *	- Power/Retry relative values no longer * 100000
- *	- Add explicit flag to tell stats are in 802.11k RCPI : IW_QUAL_RCPI
  */
 
 /**************************** CONSTANTS ****************************/
@@ -461,7 +448,6 @@ #define IW_QUAL_DBM		0x08	/* Level + Noi
 #define IW_QUAL_QUAL_INVALID	0x10	/* Driver doesn't provide value */
 #define IW_QUAL_LEVEL_INVALID	0x20
 #define IW_QUAL_NOISE_INVALID	0x40
-#define IW_QUAL_RCPI		0x80	/* Level + Noise are 802.11k RCPI */
 #define IW_QUAL_ALL_INVALID	0x70
 
 /* Frequency flags */
@@ -514,12 +500,10 @@ #define IW_RETRY_ON		0x0000	/* No detail
 #define IW_RETRY_TYPE		0xF000	/* Type of parameter */
 #define IW_RETRY_LIMIT		0x1000	/* Maximum number of retries*/
 #define IW_RETRY_LIFETIME	0x2000	/* Maximum duration of retries in us */
-#define IW_RETRY_MODIFIER	0x00FF	/* Modify a parameter */
+#define IW_RETRY_MODIFIER	0x000F	/* Modify a parameter */
 #define IW_RETRY_MIN		0x0001	/* Value is a minimum  */
 #define IW_RETRY_MAX		0x0002	/* Value is a maximum */
 #define IW_RETRY_RELATIVE	0x0004	/* Value is not in seconds/ms/us */
-#define IW_RETRY_SHORT		0x0010	/* Value is for short packets  */
-#define IW_RETRY_LONG		0x0020	/* Value is for long packets */
 
 /* Scanning request flags */
 #define IW_SCAN_DEFAULT		0x0000	/* Default scan of the driver */
@@ -1033,7 +1017,7 @@ struct	iw_range
 	/* Note : this frequency list doesn't need to fit channel numbers,
 	 * because each entry contain its channel index */
 
-	__u32		enc_capa;	/* IW_ENC_CAPA_* bit field */
+	__u32		enc_capa; /* IW_ENC_CAPA_* bit field */
 };
 
 /*
net/core/net-sysfs.c +4 -1
index f47f319..1347276 100644
@@ -344,6 +344,8 @@ static ssize_t wireless_show(struct clas
 		if(dev->wireless_handlers &&
 		   dev->wireless_handlers->get_wireless_stats)
 			iw = dev->wireless_handlers->get_wireless_stats(dev);
+		else if (dev->get_wireless_stats)
+			iw = dev->get_wireless_stats(dev);
 		if (iw != NULL)
 			ret = (*format)(iw, buf);
 	}
@@ -463,7 +465,8 @@ int netdev_register_sysfs(struct net_dev
 		*groups++ = &netstat_group;
 
 #ifdef WIRELESS_EXT
-	if (net->wireless_handlers && net->wireless_handlers->get_wireless_stats)
+	if (net->get_wireless_stats
+	    || (net->wireless_handlers && net->wireless_handlers->get_wireless_stats))
 		*groups++ = &wireless_group;
 #endif
 
net/core/wireless.c +26 -41

125 lines not on this page · raw message

net/ieee80211/softmac/ieee80211softmac_wx.c +4 -4
index 2aa779d..75320b6 100644
@@ -80,10 +80,10 @@ ieee80211softmac_wx_set_essid(struct net
 	 * If it's our network, ignore the change, we're already doing it!
 	 */
 	if((sm->associnfo.associating || sm->associated) &&
-	   (data->essid.flags && data->essid.length)) {
+	   (data->essid.flags && data->essid.length && extra)) {
 		/* Get the associating network */
 		n = ieee80211softmac_get_network_by_bssid(sm, sm->associnfo.bssid);
-		if(n && n->essid.len == data->essid.length &&
+		if(n && n->essid.len == (data->essid.length - 1) &&
 		   !memcmp(n->essid.data, extra, n->essid.len)) {
 			dprintk(KERN_INFO PFX "Already associating or associated to "MAC_FMT"\n",
 				MAC_ARG(sm->associnfo.bssid));
@@ -109,8 +109,8 @@ ieee80211softmac_wx_set_essid(struct net
 	sm->associnfo.static_essid = 0;
 	sm->associnfo.assoc_wait = 0;
 
-	if (data->essid.flags && data->essid.length) {
-		length = min((int)data->essid.length, IW_ESSID_MAX_SIZE);
+	if (data->essid.flags && data->essid.length && extra /*required?*/) {
+		length = min(data->essid.length - 1, IW_ESSID_MAX_SIZE);
 		if (length) {
 			memcpy(sm->associnfo.req_essid.data, extra, length);
 			sm->associnfo.static_essid = 1;
Signature · 2 lines
John W. Linville
[email protected]
#51 lore raw
Quoted 5 lines · John W. Linville

Unfortunately, I don't see any way to "fix" WE-21 without similarly breaking wireless-tools 29 and other "WE-21 aware" apps. And since I'll bet that the various WE-aware apps have checks like "if WE > 20" for managing ESSID length settings, we may have painted ourselves into a korner (sic).

The apps are based on a pre-release kernel, which everyone knows could change, precisely for reasons like this. Sounds like somebody took a risk, and lost...

Jeff

#52 lore raw
Quoted 7 lines · Jeff Garzik
Quoted 5 lines · John W. Linville

Unfortunately, I don't see any way to "fix" WE-21 without similarly breaking wireless-tools 29 and other "WE-21 aware" apps. And since I'll bet that the various WE-aware apps have checks like "if WE > 20" for managing ESSID length settings, we may have painted ourselves into a korner (sic).

Quoted 5 lines

The apps are based on a pre-release kernel, which everyone knows could change, precisely for reasons like this. Sounds like somebody took a risk, and lost...

Jeff

Jeff,

Let's not make a mountain of this molehill. If you want to use
old versions of Wireless Tools and wpa_supplicant with WE-21, what you need is just to add a dummy character at the end of your ESSID. And everything will be fine.

Also, there is no other way to update cleanly a kernel API
than to push userspace first. I think I took way more care in term of smoothing over the API transition than any other kernel subsystem, so I don't know what could have been done better. I don't remember this level of flamewar when those other subsystems did change their userspace APIs.
I try to be constructive about all this, so let's find a way
forward without loosing perspective.

Regards,

Jean

#53 lore raw
Quoted 4 lines · Jean Tourrilhes

Let's not make a mountain of this molehill. If you want to use
old versions of Wireless Tools and wpa_supplicant with WE-21, what you need is just to add a dummy character at the end of your ESSID. And everything will be fine.

FALSE. Old apps are by definition already in place. "all you need to do..." is an impossible condition to satisfy.

Quoted 6 lines

Also, there is no other way to update cleanly a kernel API
than to push userspace first. I think I took way more care in term of smoothing over the API transition than any other kernel subsystem, so I don't know what could have been done better. I don't remember this level of flamewar when those other subsystems did change their userspace APIs.

Userspace APIs can change, as long as they remain backwards compatible.

Just follow LKML to see all the flack people get, when userspace APIs change in an incompatible way. Or heck, just look at the changelog for the patches we revert, when such brokenness is detected.

Finally, until an API is actually in a kernel release, it has the potential to change. That's just a fact of Linux development. I certainly understand trying to get stuff out ahead of a kernel release, but you must understand the negative consequences of doing so, when something goes wrong. Like it did here.

Jeff

#54 lore raw
In reply to Jean Tourrilhes
Quoted 4 lines · Jean Tourrilhes

Let's not make a mountain of this molehill. If you want to use
old versions of Wireless Tools and wpa_supplicant with WE-21, what you need is just to add a dummy character at the end of your ESSID. And everything will be fine.

Nope, everything won't be fine with WPA-PSK which is using SSID as part of the key derivation. In other words, you cannot add a dummy character in the end of the SSID in wpa_supplicant configuration for a WPA-PSK network and expect everything to work.

Signature · 1 line
Jouni Malinen                                            PGP id EFC895FA
#55 lore raw
In reply to John W. Linville
Quoted 3 lines · John W. Linville

I.E. With "WE-21 aware" tools already in the wild, it isn't now clear to me how WE can evolve any further than WE-20.

Well, if you get a WE-22 out soon enough, the situation will be one where people who are fast at updating will have a fixed version quickly, and the ones that aren't quick at updating will never have even seen the broken case.

And without any actual release kernel actually having had the issue, we should be pretty well off - the only people who actually saw semantics change were people who build their own kernels etc, and those people aren't the problem cases.

The users that you need to care about are the ones that upgrade rather seldom and/or need to maintain a stable setup for other reasons (eg it's not at all unheard of to have a common base release, but then have certain machines on the network with more modern kernels because they have new hardware that requires a modern kernel for support, for example: that's the situation where you may want to have a much older common user space, even if you have a new kernel).

Kernel developers tend to be happy to upgrade their user programs, and don't generally see it as a big problem. The people for whom it is a problem are elsewhere :)

Linus

#56 lore raw
Quoted 11 lines · Linus Torvalds
Quoted 3 lines · John W. Linville

I.E. With "WE-21 aware" tools already in the wild, it isn't now clear to me how WE can evolve any further than WE-20.

Well, if you get a WE-22 out soon enough, the situation will be one where people who are fast at updating will have a fixed version quickly, and the ones that aren't quick at updating will never have even seen the broken case.

Wrong.
Slackware has just released with WE-21 aware tools. Users of
this version of Slackware will never see anything else than WE-21 aware tools. And if there is a distro and users which are conservative, this is Slackware.
Same deal for Mandriva 2007.
Regards,

Jean

#57 lore raw
Quoted 6 lines · Jean Tourrilhes

Wrong.
Slackware has just released with WE-21 aware tools. Users of
this version of Slackware will never see anything else than WE-21 aware tools. And if there is a distro and users which are conservative, this is Slackware.
Same deal for Mandriva 2007.

And those distros have the standard option of doing what you always do when you release based on an unreleased API...

Jeff

#58 lore raw
In reply to Linus Torvalds
Quoted 11 lines · Linus Torvalds
Quoted 3 lines · John W. Linville

I.E. With "WE-21 aware" tools already in the wild, it isn't now clear to me how WE can evolve any further than WE-20.

Well, if you get a WE-22 out soon enough, the situation will be one where people who are fast at updating will have a fixed version quickly, and the ones that aren't quick at updating will never have even seen the broken case.

Linus,

You can't froze kernel userspace API forever. That is simply
not workable, it will lead to stagnation and obsolescence. This is especially unfair because some other kernel userspace API are allow to change whenever their maintainers feels like.

Just to give you an example why we sometime need to
change. The first two generations of 802.11 hardware were using the ESSID as a C-string (no NUL char), so the API was also using a C-string (no NUL char). New 802.11 hardware do accept NUL in the ESSID, therefore the API need to evolve away from C-string to offer this new feature to userspace. Especially that new WPA standard may use that in the future (cf. Jouni's e-mail).

In the past, kernel userspace API changes were done during the
devel series, but we don't have this option anymore. What I would like people to discuss is what are the best practice to perform kernel userspace API changes in 2.6.X.
I personally thought that I went beyond the usual practice by
waiting 6 months, auditing all userspace and making sure the bits had propagated to distro. And let's not forget that the tools warn users about API mismatch.
If you feel we need 1 more month, that's perfectly ok with all
of us. If your target is that some specific distros ship with compatible userspace, I can personally monitor that and report. You may want to be a bit more explicit in your standards, that would help all of us doing a better job.

	Have fun...

	Jean
#59 lore raw
Quoted 13 lines · Jean Tourrilhes

You can't froze kernel userspace API forever. That is simply
not workable, it will lead to stagnation and obsolescence. This is especially unfair because some other kernel userspace API are allow to change whenever their maintainers feels like.

Just to give you an example why we sometime need to
change. The first two generations of 802.11 hardware were using the ESSID as a C-string (no NUL char), so the API was also using a C-string (no NUL char). New 802.11 hardware do accept NUL in the ESSID, therefore the API need to evolve away from C-string to offer this new feature to userspace. Especially that new WPA standard may use that in the future (cf. Jouni's e-mail).

Quoted 4 lines

In the past, kernel userspace API changes were done during the
devel series, but we don't have this option anymore. What I would like people to discuss is what are the best practice to perform kernel userspace API changes in 2.6.X.

You can _add_ to the userspace API, because that obviously does not break old programs.

You just should not introduce changes which break old programs.

Binaries built in the 1990's still work under Linux, you know...

API changes require new ioctl/sub-ioctl numbers, so that new programs can take advantage of new capabilities while old programs continue to work.

Jeff

#60 lore raw
Quoted 2 lines · Jeff Garzik

Binaries built in the 1990's still work under Linux, you know...

Completely untrue for system utils. I've got plenty of counter
examples.

Jean

#61 lore raw
In reply to Jean Tourrilhes
Quoted 3 lines · Jean Tourrilhes

You can't froze kernel userspace API forever. That is simply
not workable

Stop arguing this way.

It's not what we have ever done. We've _extended_ the API. But we don't break old ones.

I don't even see why you argue. Even the people directly involved with this thing seem to say that it should have some simple translation layer and do the internal format thing. We've had major subsystem that do that, and I don't see why you think wireless is so different, and so special in this respect.

The whole _point_ of a kernel is to act as a abstraction layer and resource management between user programs and hardware/outside world. That's why kernels _exist_. Breaking user-land API's is thus by definition something totally idiotic.

If you need to break something, you create a new interface, and try to translate between the two, and maybe you deprecate the old one so that it can be removed once it's not in use any more. If you can't see that this is how a kernel should work, you're missing the point of having a kernel in the first place.

Also, I don't want to hear about how this makes things harder and more complicated. The fact is, we're programmers, and we should care about the _users_. If we don't, we're just masturbating. There's a whole other side to this "create software" than just the "me, me, me" side, and if you lose sight of that side, that's a really bad thing.

Linus

#62 lore raw
Quoted 11 lines · Linus Torvalds
Quoted 3 lines · Jean Tourrilhes

You can't froze kernel userspace API forever. That is simply
not workable

Stop arguing this way.

It's not what we have ever done. We've _extended_ the API. But we don't break old ones.

Old APIs get deprecated, and people are forced to the new API,
which is exactly the same as far as userspace is concerned. This transition is exactly the same as what you propose, both kernel API coexist for some time, except it happens in userspace instead of in kernel, which is an implementation detail.
So, my question is when can I remove the old ESSID API.

Quoted 5 lines

I don't even see why you argue. Even the people directly involved with this thing seem to say that it should have some simple translation layer and do the internal format thing. We've had major subsystem that do that, and I don't see why you think wireless is so different, and so special in this respect.

The Wireless people (Jouni, Dan) decided to change the
*userspace* API. We could translate the new *userspace* API to the old kernel API, but I don't see the point.

Quoted 3 lines

If you need to break something, you create a new interface, and try to translate between the two, and maybe you deprecate the old one so that it can be removed once it's not in use any more.

That's exactly what it hinges on. What is your criteria for
removing the old ESSID API. My understanding was 6 months.

Jean

#63 lore raw
Quoted 16 lines · Jean Tourrilhes
Quoted 7 lines · Linus Torvalds
Quoted 2 lines · Jean Tourrilhes

You can't froze kernel userspace API forever. That is simply
not workable

Stop arguing this way.

It's not what we have ever done. We've _extended_ the API. But we don't break old ones.

Old APIs get deprecated, and people are forced to the new API,
which is exactly the same as far as userspace is concerned. This transition is exactly the same as what you propose, both kernel API coexist for some time, except it happens in userspace instead of in kernel, which is an implementation detail.
So, my question is when can I remove the old ESSID API.

Quoted 5 lines
Quoted 1 line

I don't even see why you argue. Even the people directly involved with this thing seem to say that it should have some simple translation layer and do the internal format thing. We've had major subsystem that do that, and I don't see why you think wireless is so different, and so special in this respect.

The Wireless people (Jouni, Dan) decided to change the
*userspace* API. We could translate the new *userspace* API to the old kernel API, but I don't see the point.

Kernel API and userspace API are two vastly different things. We change the kernel API all the time. We _don't_ change the userspace API, except when "change" is defined as an additional to an existing API.

Quoted 4 lines
Quoted 1 line

If you need to break something, you create a new interface, and try to translate between the two, and maybe you deprecate the old one so that it can be removed once it's not in use any more.

That's exactly what it hinges on. What is your criteria for
removing the old ESSID API. My understanding was 6 months.

There is no reason why the old ESSID API cannot live on for years and years. Just like stat(2) version 1 doesn't support modern time granularity, old ESSID API won't support the full range of modern ESSIDs. So what? That's what a new API -- living alongside the old one -- is for.

Jeff

#64 lore raw
In reply to Jean Tourrilhes
Quoted 10 lines · Jean Tourrilhes
Quoted 3 lines

It's not what we have ever done. We've _extended_ the API. But we don't break old ones.

Old APIs get deprecated, and people are forced to the new API,
which is exactly the same as far as userspace is concerned. This transition is exactly the same as what you propose, both kernel API coexist for some time, except it happens in userspace instead of in kernel, which is an implementation detail.
So, my question is when can I remove the old ESSID API.

That isn't the question here.

The current situation seems to be designed to add the new one and removing the old one as a single step. THAT IS BROKEN.

The new one and the old one needs to work at the same time, exactly so that there's a transition mechanism.

That's the part you seem to now have understood. There should be no "flag day" when people have to switch over.

Quoted 3 lines

The Wireless people (Jouni, Dan) decided to change the
*userspace* API. We could translate the new *userspace* API to the old kernel API, but I don't see the point.

You do not indeed see the point.

The point is, we can switch internal kernel ABI's - new or old - at any point. But user-level ABI's should never require a one-way update.

Quoted 2 lines

That's exactly what it hinges on. What is your criteria for
removing the old ESSID API. My understanding was 6 months.

But we didn't have 6 months of the new API, did we? People complained.

The person you merged through explicitly said that if he had realized what you did, he wouldn't have merged.

That should tell you something. Why are you ignoring this?

Linus

#65 lore raw
Quoted 3 lines · Linus Torvalds

The current situation seems to be designed to add the new one and removing the old one as a single step. THAT IS BROKEN.

This is not the case, I don't know where you got this
idea. This is a *two* step process with a 6 month interval between the two steps. This was clearly detailed earlier in this thread.

Quoted 2 lines

The new one and the old one needs to work at the same time, exactly so that there's a transition mechanism.

Yes, this is precisely what we have been doing, the two APIs
have been working at the same time for more than 6 months.

Quoted 4 lines
Quoted 2 lines

That's exactly what it hinges on. What is your criteria for
removing the old ESSID API. My understanding was 6 months.

But we didn't have 6 months of the new API, did we?

Yes, we had more of 6 months of the new API. Please check the
facts : included April 11th in Gentoo.

Quoted 1 line

People complained.

Yes, maybe 6 months was two short. That's why I say that we
should give it one or two more months. Maybe we need FC6 to be released.

Quoted 2 lines

The person you merged through explicitly said that if he had realized what you did, he wouldn't have merged.

I did not merge through Jeff.

Quoted 1 line

Linus

Jean

#66 lore raw
Quoted 3 lines · Jean Tourrilhes

Yes, this is precisely what we have been doing, the two APIs
have been working at the same time for more than 6 months.

In the kernel for any particular driver?

Or just in user-land?

There's a big difference.

Linus

#67 lore raw

In other words, if we really have had the code to handle both interfaces in the kernel, I vote for just reverting the patch that "fixed" it to just one.

But I suspect that's not what you're really saying. I think you're saying is that we've had two different interfaces for _different_ chips, and that some user-space tools have supported both. And since clearly the distros haven't updated to those tools yet (or this wouldn't be an issue), we still want to avoid a flag-day, and wait until they have done so.

Linus

#68 lore raw
In reply to Linus Torvalds
Quoted 4 lines · Linus Torvalds
Quoted 1 line · Jean Tourrilhes
Quoted 8 lines
Quoted 2 lines

Yes, this is precisely what we have been doing, the two APIs
have been working at the same time for more than 6 months.

In the kernel for any particular driver?

Or just in user-land?

There's a big difference.

Both actually.
I've slightly simplified the situation, because you may not
care for all the details, but if you want to know the gory details...
It's all started with some kernel drivers changing the way
they handle ESSID. That was last January. So, in the kernel, you had half the drivers doing the old way (NUL terminated), and half doing the new way (not NUL terminated).
I immediately asked to revert to the old way. All the Wireless
people were against me, and this mish-mash of API was kept (it was public on netdev). At this point, nobody cared that userspace API was broken and I was left cleaning up the mess.
Of course, some part of the Wireless Tools did broke on the
new API (I had bug reports), so I had to push new version of Wireless Tools. And I took this opportunity to finish moving over the API to the new way.

Sometime breaking userspace APIs is perfectly OK, while
sometimes it's not. You just have to make sure that Linus does not hear about it, I guess ;-)

Quoted 1 line

Linus

	Have fun...

	Jean
#69 lore raw
Quoted 4 lines · Jean Tourrilhes

Sometime breaking userspace APIs is perfectly OK, while
sometimes it's not. You just have to make sure that Linus does not hear about it, I guess ;-)

I see the smiley, and I think you're trying to be funny and clever, but the thing is, I actually think that's _true_.

It's perfectly fine to break ABI's if nobody ever complains loudly enough that other developers notice.

So yes, we could actually even make it a real hard rule:

"Breaking ABI's is fine. As long as you can hide the breakage so well
that nobody complains loudly enough that anybody ever notices".

The very fact that this turned into a discussion is a sign that the ABI breakage wasn't handled well enough. Usually, when we do something, nobody ever even notices.

(For an example of such a ABI breakage: I changed ptrace() to not allow ptracing another thread in the same thread group about a year ago, because it turned out that it was a serious local DoS problem. In the 12 months since, I think we had two people who ever actually noticed, and both of them actually caused some discussion about ways to perhaps unbreak it.)

Linus

#70 lore raw
Quoted 4 lines · Linus Torvalds

The very fact that this turned into a discussion is a sign that the ABI breakage wasn't handled well enough. Usually, when we do something, nobody ever even notices.

There was the grand total of *ONE* user who was personally
impacted by the userspace API change (the two other, one was hit by a bug, now fixed, one was hit because of kernel API change + external driver). And I immediately proposed to postpone the change to a later time.
The people who contributed most to this tread were had not
experienced any breakage. I don't know what conclusion to reach from that, and I don't understand why it took such proportions.

Guessing the right time to deprecate an old API can be a trial
and error process. So, we just have to wait for the release of FC6...

Jean

#71 lore raw
Quoted 6 lines · Jean Tourrilhes

There was the grand total of *ONE* user who was personally
impacted by the userspace API change (the two other, one was hit by a bug, now fixed, one was hit because of kernel API change + external driver). And I immediately proposed to postpone the change to a later time.

Heh, ok. I'm personally not able to really judge any wireless-extensions stuff, so I have to go by how noisy the discussion becomes ;)

I'll leave this to you guys to sort out, I just did want to pipe up (since I was asked to) that as far as _I_ am concerned, user-space interfaces really are very important. Whether this is one of the really important ones or not, I leave to the people involved to figure out, but I really don't want to have developers dismissing the issue.

(I think some people used to, and I think we've gotten better at it. But maybe I'm just being overly optimistic again ;)

Linus

#72 lore raw
In reply to Jean Tourrilhes
Quoted 11 lines · Jean Tourrilhes
Quoted 4 lines · Linus Torvalds

The very fact that this turned into a discussion is a sign that the ABI breakage wasn't handled well enough. Usually, when we do something, nobody ever even notices.

There was the grand total of *ONE* user who was personally
impacted by the userspace API change (the two other, one was hit by a bug, now fixed, one was hit because of kernel API change + external driver). And I immediately proposed to postpone the change to a later time.

And said user, being me, is currently running with upgraded userspace without any issues (counting upgrading userspace as a non-issue).

I originally logged my report as I do for other things that break or look different in new snapshots, in order to provide early feedback to the kernel developers - I guess it's the actual point of having snapshots from kernel.org...

Thanks, ciao,

--alessandro

"Well a man has two reasons for things that he does
the first one is pride and the second one is love all understandings must come by this way"

(Husker Du, 'She Floated Away')

#73 lore raw
Quoted 14 lines · Alessandro Suardi
Quoted 4 lines · Jean Tourrilhes
Quoted 3 lines · Linus Torvalds

The very fact that this turned into a discussion is a sign that the ABI breakage wasn't handled well enough. Usually, when we do something,

Quoted 8 lines

nobody

Quoted 1 line

ever even notices.

There was the grand total of *ONE* user who was personally
impacted by the userspace API change (the two other, one was hit by a bug, now fixed, one was hit because of kernel API change + external driver). And I immediately proposed to postpone the change to a later time.

Quoted 7 lines

And said user, being me, is currently running with upgraded userspace without any issues (counting upgrading userspace as a non-issue).

I originally logged my report as I do for other things that break or look different in new snapshots, in order to provide early feedback to the kernel developers - I guess it's the actual point of having snapshots from kernel.org...

Precisely. We are not omniscient. Based on your feedback, I
decided to postpone WE-21.
Your feedback was useful and appreciated. I will never blame
the messenger.

Quoted 3 lines

Thanks, ciao,

--alessandro

	Ciao...

	Jean
#74 lore raw
In reply to Jean Tourrilhes

Hi!

Quoted 9 lines
Quoted 3 lines

The very fact that this turned into a discussion is a sign that the ABI breakage wasn't handled well enough. Usually, when we do something, nobody ever even notices.

There was the grand total of *ONE* user who was personally
impacted by the userspace API change (the two other, one was hit by a bug, now fixed, one was hit because of kernel API change + external driver). And I immediately proposed to postpone the change to a later time.

The fact that all the wireless tools print...

Warning: Driver for device eth1 has been compiled with version 20 of Wireless Extension, while this program supports up to version 17.

...shows that wireless is not as backwards compatible as usual kernel ABIs are. I'm afraid that wireless extensions can not be helped (designed with back-compatibility at wrong place), but can we make sure new netlink ABI does not have similar problems?
Pavel

Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
#75 lore raw
In reply to Jean Tourrilhes
Quoted 6 lines · Jean Tourrilhes
Quoted 2 lines · Linus Torvalds

The person you merged through explicitly said that if he had realized what you did, he wouldn't have merged.

I did not merge through Jeff.

You did, just indirectly. I was directly upstream from you.

For the record, I did not fully comprehend that we would be breaking existing tools (and their users) -- I certainly should have, but I did not. I apologize both to you for being part of this scenario I inadvertantly allowed to unfold and to the users who experienced the resulting breakage.

This was the second time I took patches for extending WE, and I have received nothing but grief from either set of patches. Even had things gone smoothly, WE was already hated near universally. WE has survived based on being "good enough" for a long time. But I think it is safe to say that if WE were not already in the kernel, it would have little chance of making it in today.

All the legitimate options for extending WE now amount to forking a new API. But work is already underway on a WE replacement. I think the best option is to invest in that replacement, and a compatibility layer to support older WE-aware applications. Please see the nl80211 and cfg80211 currently on the netdev list.

I do not intend or expect to take any more WE enhancment patches. Only bug fixes to WE will be accepted from now onward.

Jean, I thank you for your long-running contributions. I hope this will not discourage you from further participation.

Thanks,

John

Signature · 2 lines
John W. Linville
[email protected]
#76 lore raw
In reply to Jean Tourrilhes
Quoted 2 lines · Jean Tourrilhes

That's exactly what it hinges on. What is your criteria for
removing the old ESSID API. My understanding was 6 months.

Just to make it clear. The current practice is that 6 months is the _minimum_, after an extensive discussion on LKML and an understanding that costs of supporting both the old and the new interface outweighs the cost and pain to the user community of removing the old interface. Then after there is an agrement on how long the deprecation window will be (and sometimes it is negotiated to be longer than six months), it is documented in

/usr/src/linux/Documentation/feature-removal-schedule.txt

But it's never been the case that after six months, we can remove a feature just because we feel like it, or it's slightly more convenient to programmers at the cost of imposing pain on users, or at the cost of discouraging users from testing bleeding edge kernels.

As others have pointed out, we have maintained old stat system call interfaces for over a ***decade***.

So perhaps that's something to keep in mind as we start considering with the next generation wireless interface looks like, and whether it is sufficiently well defined and has enough forwards and backwards compatibility, both at the driver level and at the userspace level, so that we can avoid these sorts of problems going forward.

Regards,

- Ted

P.S. Because of all of these changing interfaces, I *still* haven't been able to get wpa_supplicant working with LEAP so I can get wireless access to in IBM offices using my ipw3945 driver. I've tried, and failed. Sigh, I guess I'm not smart enough....

#77 lore raw
Quoted 5 lines · Theodore Tso

P.S. Because of all of these changing interfaces, I *still* haven't been able to get wpa_supplicant working with LEAP so I can get wireless access to in IBM offices using my ipw3945 driver. I've tried, and failed. Sigh, I guess I'm not smart enough....

Which changing interface are you talking about ? I'm afraid
that WPA is a complex business and the ipw3945 driver is still beta. Jouni and the people on the hostap mailing list should be able to help.
Regards,

Jean

#78 lore raw

Re: LEAP (was: wpa supplicant/ipw3945, ESSID last char missing)

In reply to Theodore Tso
Quoted 4 lines · Theodore Tso

P.S. Because of all of these changing interfaces, I *still* haven't been able to get wpa_supplicant working with LEAP so I can get wireless access to in IBM offices using my ipw3945 driver. I've tried, and failed. Sigh, I guess I'm not smart enough....

This is getting somewhat off topic to the main thread, but anyway, LEAP is quite an odd beast as far as EAP methods are concerned and the way it is implemented in Cisco APs makes it even worse.. LEAP can mean so many different things that it is difficult to give any generic answer to how to do this. Just about any other wireless security configuration would be easier to explain.. ;-)

Feel free to write me more details on the configuration used in the network and I can try to figure out how that would need to be configured. I would need to know whether LEAP is being used with IEEE 802.1X (dynamic WEP keys; key_mgmt=IEEE8021X in wpa_supplicant)) or with WPA (key_mgmt=WPA-EAP in wpa_supplicant). In addition, it would be useful to know whether the APs are configured to require Cisco prorietary "Network EAP" authentication algorithm (auth_alg=LEAP in wpa_supplicant) or not. Many of the drivers do not support that at all.. I don't know whether ipw3945 does since I have not tested this myself and do not remember having heard of a clear report on this being used.

And just hope that the APs do not require Cisco proprietary CKIP or CMIC encryption algorithms which are most likely not supported by ipw3945 (or most Linux drivers for that matter)..

Signature · 1 line
Jouni Malinen                                            PGP id EFC895FA
#79 lore raw
In reply to John W. Linville

We are ready and looking forward to d80211 stack but When it will be in the Kernel?

Quoted 793 lines · John W. Linville
Quoted 11 lines · Jeff Garzik
Quoted 7 lines · Jean Tourrilhes

Now it's too late, those changes have propagated to userspace
tools, and are now shipping in some actual release of some distro. So, what are we going to say to Mandriva 2007 and FC6 users, to revert back to an *older* version of the tools ?
Because userspace has already been updated, we have only two
options, merge it now, or in 2.6.20.

If the choice is between the ABI used by a bunch of distros in production, and the ABI used by two re-release distros, I think the choice is obvious...

I (grudgingly?) agree...

Unfortunately, I don't see any way to "fix" WE-21 without similarly breaking wireless-tools 29 and other "WE-21 aware" apps. And since I'll bet that the various WE-aware apps have checks like "if WE > 20" for managing ESSID length settings, we may have painted ourselves into a korner (sic).

I.E. With "WE-21 aware" tools already in the wild, it isn't now clear to me how WE can evolve any further than WE-20. Unless the ESSID length changes in WE-21 is somehow deemed acceptable for 2.6.20 or some later kernel version, I don't see how WE can continue to change.

Wireless developers, it's time to get d80211 ready to merge...or figure-out how to get nl80211 on the current stack...hmmm...

John

---

The following changes since commit 8ccb3dcd1f8e80e8702642e1de26541b52f6bb7c:
Linus Torvalds:
x86: Fix booting with "no387 nofxsr"

are found in the git repository at:

git://git.kernel.org/pub/scm/linux/kernel/git/linville/wireless-2.6.git we21-revert

John W. Linville:
wireless: revert WE-21 patches

 drivers/net/wireless/airo.c                 |   19 ++++----
 drivers/net/wireless/atmel.c                |   18 ++++---
 drivers/net/wireless/bcm43xx/bcm43xx_wx.c   |    2 -
 drivers/net/wireless/hostap/hostap_ioctl.c  |   10 ++--
 drivers/net/wireless/ipw2100.c              |   14 +++---
 drivers/net/wireless/ipw2200.c              |   16 ++++--
 drivers/net/wireless/orinoco.c              |   10 ++--
 drivers/net/wireless/prism54/isl_ioctl.c    |   16 +++---
 drivers/net/wireless/ray_cs.c               |    2 -
 drivers/net/wireless/wl3501_cs.c            |    6 +-
 drivers/net/wireless/zd1201.c               |    2 -
 drivers/net/wireless/zd1211rw/zd_netdev.c   |    2 -
 include/linux/netdevice.h                   |    1 
 include/linux/wireless.h                    |   24 ++--------
 net/core/net-sysfs.c                        |    5 ++
 net/core/wireless.c                         |   67 ++++++++++-----------------
 net/ieee80211/softmac/ieee80211softmac_wx.c |    8 ++-
 17 files changed, 98 insertions(+), 124 deletions(-)
drivers/net/wireless/airo.c +9 -10
index 39d0934..1864b07 100644
@@ -5883,7 +5883,7 @@ static int airo_set_essid(struct net_dev
 		int	index = (dwrq->flags & IW_ENCODE_INDEX) - 1;
 
 		/* Check the size of the string */
-		if(dwrq->length > IW_ESSID_MAX_SIZE) {
+		if(dwrq->length > IW_ESSID_MAX_SIZE+1) {
 			return -E2BIG ;
 		}
 		/* Check if index is valid */
@@ -5895,7 +5895,7 @@ static int airo_set_essid(struct net_dev
 		memset(SSID_rid.ssids[index].ssid, 0,
 		       sizeof(SSID_rid.ssids[index].ssid));
 		memcpy(SSID_rid.ssids[index].ssid, extra, dwrq->length);
-		SSID_rid.ssids[index].len = dwrq->length;
+		SSID_rid.ssids[index].len = dwrq->length - 1;
 	}
 	SSID_rid.len = sizeof(SSID_rid);
 	/* Write it to the card */
@@ -6005,7 +6005,7 @@ static int airo_set_nick(struct net_devi
 	struct airo_info *local = dev->priv;
 
 	/* Check the size of the string */
-	if(dwrq->length > 16) {
+	if(dwrq->length > 16 + 1) {
 		return -E2BIG;
 	}
 	readConfigRid(local, 1);
@@ -6030,7 +6030,7 @@ static int airo_get_nick(struct net_devi
 	readConfigRid(local, 1);
 	strncpy(extra, local->config.nodeName, 16);
 	extra[16] = '\0';
-	dwrq->length = strlen(extra);
+	dwrq->length = strlen(extra) + 1;
 
 	return 0;
 }
@@ -6782,9 +6782,9 @@ static int airo_set_retry(struct net_dev
 	}
 	readConfigRid(local, 1);
 	if(vwrq->flags & IW_RETRY_LIMIT) {
-		if(vwrq->flags & IW_RETRY_LONG)
+		if(vwrq->flags & IW_RETRY_MAX)
 			local->config.longRetryLimit = vwrq->value;
-		else if (vwrq->flags & IW_RETRY_SHORT)
+		else if (vwrq->flags & IW_RETRY_MIN)
 			local->config.shortRetryLimit = vwrq->value;
 		else {
 			/* No modifier : set both */
@@ -6820,14 +6820,14 @@ static int airo_get_retry(struct net_dev
 	if((vwrq->flags & IW_RETRY_TYPE) == IW_RETRY_LIFETIME) {
 		vwrq->flags = IW_RETRY_LIFETIME;
 		vwrq->value = (int)local->config.txLifetime * 1024;
-	} else if((vwrq->flags & IW_RETRY_LONG)) {
-		vwrq->flags = IW_RETRY_LIMIT | IW_RETRY_LONG;
+	} else if((vwrq->flags & IW_RETRY_MAX)) {
+		vwrq->flags = IW_RETRY_LIMIT | IW_RETRY_MAX;
 		vwrq->value = (int)local->config.longRetryLimit;
 	} else {
 		vwrq->flags = IW_RETRY_LIMIT;
 		vwrq->value = (int)local->config.shortRetryLimit;
 		if((int)local->config.shortRetryLimit != (int)local->config.longRetryLimit)
-			vwrq->flags |= IW_RETRY_SHORT;
+			vwrq->flags |= IW_RETRY_MIN;
 	}
 
 	return 0;
@@ -7005,7 +7005,6 @@ static int airo_set_power(struct net_dev
 			local->config.rmode |= RXMODE_BC_MC_ADDR;
 			set_bit (FLAG_COMMIT, &local->flags);
 		case IW_POWER_ON:
-			/* This is broken, fixme ;-) */
 			break;
 		default:
 			return -EINVAL;
drivers/net/wireless/atmel.c +9 -9
index 0fc267d..995c7be 100644
@@ -1656,13 +1656,13 @@ static int atmel_set_essid(struct net_de
 		priv->connect_to_any_BSS = 0;
 
 		/* Check the size of the string */
-		if (dwrq->length > MAX_SSID_LENGTH)
+		if (dwrq->length > MAX_SSID_LENGTH + 1)
 			 return -E2BIG;
 		if (index != 0)
 			return -EINVAL;
 
-		memcpy(priv->new_SSID, extra, dwrq->length);
-		priv->new_SSID_size = dwrq->length;
+		memcpy(priv->new_SSID, extra, dwrq->length - 1);
+		priv->new_SSID_size = dwrq->length - 1;
 	}
 
 	return -EINPROGRESS;
@@ -2120,9 +2120,9 @@ static int atmel_set_retry(struct net_de
 	struct atmel_private *priv = netdev_priv(dev);
 
 	if (!vwrq->disabled && (vwrq->flags & IW_RETRY_LIMIT)) {
-		if (vwrq->flags & IW_RETRY_LONG)
+		if (vwrq->flags & IW_RETRY_MAX)
 			priv->long_retry = vwrq->value;
-		else if (vwrq->flags & IW_RETRY_SHORT)
+		else if (vwrq->flags & IW_RETRY_MIN)
 			priv->short_retry = vwrq->value;
 		else {
 			/* No modifier : set both */
@@ -2144,15 +2144,15 @@ static int atmel_get_retry(struct net_de
 
 	vwrq->disabled = 0;      /* Can't be disabled */
 
-	/* Note : by default, display the short retry number */
-	if (vwrq->flags & IW_RETRY_LONG) {
-		vwrq->flags = IW_RETRY_LIMIT | IW_RETRY_LONG;
+	/* Note : by default, display the min retry number */
+	if (vwrq->flags & IW_RETRY_MAX) {
+		vwrq->flags = IW_RETRY_LIMIT | IW_RETRY_MAX;
 		vwrq->value = priv->long_retry;
 	} else {
 		vwrq->flags = IW_RETRY_LIMIT;
 		vwrq->value = priv->short_retry;
 		if (priv->long_retry != priv->short_retry)
-			vwrq->flags |= IW_RETRY_SHORT;
+			vwrq->flags |= IW_RETRY_MIN;
 	}
 
 	return 0;
drivers/net/wireless/bcm43xx/bcm43xx_wx.c +1 -1
index 9b7b15c..888077f 100644
@@ -334,7 +334,7 @@ static int bcm43xx_wx_get_nick(struct ne
 	size_t len;
 
 	mutex_lock(&bcm->mutex);
-	len = strlen(bcm->nick);
+	len = strlen(bcm->nick) + 1;
 	memcpy(extra, bcm->nick, len);
 	data->data.length = (__u16)len;
 	data->data.flags = 1;
drivers/net/wireless/hostap/hostap_ioctl.c +5 -5
index d061fb3..7a49785 100644
@@ -1412,9 +1412,9 @@ #if 0
 	/* what could be done, if firmware would support this.. */
 
 	if (rrq->flags & IW_RETRY_LIMIT) {
-		if (rrq->flags & IW_RETRY_LONG)
+		if (rrq->flags & IW_RETRY_MAX)
 			HFA384X_RID_LONGRETRYLIMIT = rrq->value;
-		else if (rrq->flags & IW_RETRY_SHORT)
+		else if (rrq->flags & IW_RETRY_MIN)
 			HFA384X_RID_SHORTRETRYLIMIT = rrq->value;
 		else {
 			HFA384X_RID_LONGRETRYLIMIT = rrq->value;
@@ -1468,14 +1468,14 @@ static int prism2_ioctl_giwretry(struct 
 				rrq->value = le16_to_cpu(altretry);
 			else
 				rrq->value = local->manual_retry_count;
-		} else if ((rrq->flags & IW_RETRY_LONG)) {
-			rrq->flags = IW_RETRY_LIMIT | IW_RETRY_LONG;
+		} else if ((rrq->flags & IW_RETRY_MAX)) {
+			rrq->flags = IW_RETRY_LIMIT | IW_RETRY_MAX;
 			rrq->value = longretry;
 		} else {
 			rrq->flags = IW_RETRY_LIMIT;
 			rrq->value = shortretry;
 			if (shortretry != longretry)
-				rrq->flags |= IW_RETRY_SHORT;
+				rrq->flags |= IW_RETRY_MIN;
 		}
 	}
 	return 0;
drivers/net/wireless/ipw2100.c +7 -7
index 599e2fe..3e01c3d 100644
@@ -6972,7 +6972,7 @@ static int ipw2100_wx_set_essid(struct n
 	}
 
 	if (wrqu->essid.flags && wrqu->essid.length) {
-		length = wrqu->essid.length;
+		length = wrqu->essid.length - 1;
 		essid = extra;
 	}
 
@@ -7065,7 +7065,7 @@ static int ipw2100_wx_get_nick(struct ne
 
 	struct ipw2100_priv *priv = ieee80211_priv(dev);
 
-	wrqu->data.length = strlen(priv->nick);
+	wrqu->data.length = strlen(priv->nick) + 1;
 	memcpy(extra, priv->nick, wrqu->data.length);
 	wrqu->data.flags = 1;	/* active */
 
@@ -7357,14 +7357,14 @@ static int ipw2100_wx_set_retry(struct n
 		goto done;
 	}
 
-	if (wrqu->retry.flags & IW_RETRY_SHORT) {
+	if (wrqu->retry.flags & IW_RETRY_MIN) {
 		err = ipw2100_set_short_retry(priv, wrqu->retry.value);
 		IPW_DEBUG_WX("SET Short Retry Limit -> %d \n",
 			     wrqu->retry.value);
 		goto done;
 	}
 
-	if (wrqu->retry.flags & IW_RETRY_LONG) {
+	if (wrqu->retry.flags & IW_RETRY_MAX) {
 		err = ipw2100_set_long_retry(priv, wrqu->retry.value);
 		IPW_DEBUG_WX("SET Long Retry Limit -> %d \n",
 			     wrqu->retry.value);
@@ -7397,14 +7397,14 @@ static int ipw2100_wx_get_retry(struct n
 	if ((wrqu->retry.flags & IW_RETRY_TYPE) == IW_RETRY_LIFETIME)
 		return -EINVAL;
 
-	if (wrqu->retry.flags & IW_RETRY_LONG) {
-		wrqu->retry.flags = IW_RETRY_LIMIT | IW_RETRY_LONG;
+	if (wrqu->retry.flags & IW_RETRY_MAX) {
+		wrqu->retry.flags = IW_RETRY_LIMIT | IW_RETRY_MAX;
 		wrqu->retry.value = priv->long_retry_limit;
 	} else {
 		wrqu->retry.flags =
 		    (priv->short_retry_limit !=
 		     priv->long_retry_limit) ?
-		    IW_RETRY_LIMIT | IW_RETRY_SHORT : IW_RETRY_LIMIT;
+		    IW_RETRY_LIMIT | IW_RETRY_MIN : IW_RETRY_LIMIT;
 
 		wrqu->retry.value = priv->short_retry_limit;
 	}
drivers/net/wireless/ipw2200.c +9 -7
index 5685d7b..7358664 100644
@@ -8875,6 +8875,8 @@ static int ipw_wx_set_essid(struct net_d
         }
 
 	length = min((int)wrqu->essid.length, IW_ESSID_MAX_SIZE);
+	if (!extra[length - 1])
+		length--;
 
 	priv->config |= CFG_STATIC_ESSID;
 
@@ -8951,7 +8953,7 @@ static int ipw_wx_get_nick(struct net_de
 	struct ipw_priv *priv = ieee80211_priv(dev);
 	IPW_DEBUG_WX("Getting nick\n");
 	mutex_lock(&priv->mutex);
-	wrqu->data.length = strlen(priv->nick);
+	wrqu->data.length = strlen(priv->nick) + 1;
 	memcpy(extra, priv->nick, wrqu->data.length);
 	wrqu->data.flags = 1;	/* active */
 	mutex_unlock(&priv->mutex);
@@ -9274,9 +9276,9 @@ static int ipw_wx_set_retry(struct net_d
 		return -EINVAL;
 
 	mutex_lock(&priv->mutex);
-	if (wrqu->retry.flags & IW_RETRY_SHORT)
+	if (wrqu->retry.flags & IW_RETRY_MIN)
 		priv->short_retry_limit = (u8) wrqu->retry.value;
-	else if (wrqu->retry.flags & IW_RETRY_LONG)
+	else if (wrqu->retry.flags & IW_RETRY_MAX)
 		priv->long_retry_limit = (u8) wrqu->retry.value;
 	else {
 		priv->short_retry_limit = (u8) wrqu->retry.value;
@@ -9305,11 +9307,11 @@ static int ipw_wx_get_retry(struct net_d
 		return -EINVAL;
 	}
 
-	if (wrqu->retry.flags & IW_RETRY_LONG) {
-		wrqu->retry.flags = IW_RETRY_LIMIT | IW_RETRY_LONG;
+	if (wrqu->retry.flags & IW_RETRY_MAX) {
+		wrqu->retry.flags = IW_RETRY_LIMIT | IW_RETRY_MAX;
 		wrqu->retry.value = priv->long_retry_limit;
-	} else if (wrqu->retry.flags & IW_RETRY_SHORT) {
-		wrqu->retry.flags = IW_RETRY_LIMIT | IW_RETRY_SHORT;
+	} else if (wrqu->retry.flags & IW_RETRY_MIN) {
+		wrqu->retry.flags = IW_RETRY_LIMIT | IW_RETRY_MIN;
 		wrqu->retry.value = priv->short_retry_limit;
 	} else {
 		wrqu->retry.flags = IW_RETRY_LIMIT;
drivers/net/wireless/orinoco.c +5 -5
index 9e19a96..1840b69 100644
@@ -3037,7 +3037,7 @@ static int orinoco_ioctl_getessid(struct
 	}
 
 	erq->flags = 1;
-	erq->length = strlen(essidbuf);
+	erq->length = strlen(essidbuf) + 1;
 
 	return 0;
 }
@@ -3078,7 +3078,7 @@ static int orinoco_ioctl_getnick(struct 
 	memcpy(nickbuf, priv->nick, IW_ESSID_MAX_SIZE+1);
 	orinoco_unlock(priv, &flags);
 
-	nrq->length = strlen(nickbuf);
+	nrq->length = strlen(nickbuf)+1;
 
 	return 0;
 }
@@ -3575,14 +3575,14 @@ static int orinoco_ioctl_getretry(struct
 		rrq->value = lifetime * 1000;	/* ??? */
 	} else {
 		/* By default, display the min number */
-		if ((rrq->flags & IW_RETRY_LONG)) {
-			rrq->flags = IW_RETRY_LIMIT | IW_RETRY_LONG;
+		if ((rrq->flags & IW_RETRY_MAX)) {
+			rrq->flags = IW_RETRY_LIMIT | IW_RETRY_MAX;
 			rrq->value = long_limit;
 		} else {
 			rrq->flags = IW_RETRY_LIMIT;
 			rrq->value = short_limit;
 			if(short_limit != long_limit)
-				rrq->flags |= IW_RETRY_SHORT;
+				rrq->flags |= IW_RETRY_MIN;
 		}
 	}
 
drivers/net/wireless/prism54/isl_ioctl.c +8 -8
index 286325c..c09fbf7 100644
@@ -742,9 +742,9 @@ prism54_set_essid(struct net_device *nde
 
 	/* Check if we were asked for `any' */
 	if (dwrq->flags && dwrq->length) {
-		if (dwrq->length > 32)
+		if (dwrq->length > min(33, IW_ESSID_MAX_SIZE + 1))
 			return -E2BIG;
-		essid.length = dwrq->length;
+		essid.length = dwrq->length - 1;
 		memcpy(essid.octets, extra, dwrq->length);
 	} else
 		essid.length = 0;
@@ -814,7 +814,7 @@ prism54_get_nick(struct net_device *ndev
 	dwrq->length = 0;
 
 	down_read(&priv->mib_sem);
-	dwrq->length = strlen(priv->nickname);
+	dwrq->length = strlen(priv->nickname) + 1;
 	memcpy(extra, priv->nickname, dwrq->length);
 	up_read(&priv->mib_sem);
 
@@ -992,9 +992,9 @@ prism54_set_retry(struct net_device *nde
 		return -EINVAL;
 
 	if (vwrq->flags & IW_RETRY_LIMIT) {
-		if (vwrq->flags & IW_RETRY_SHORT)
+		if (vwrq->flags & IW_RETRY_MIN)
 			slimit = vwrq->value;
-		else if (vwrq->flags & IW_RETRY_LONG)
+		else if (vwrq->flags & IW_RETRY_MAX)
 			llimit = vwrq->value;
 		else {
 			/* we are asked to set both */
@@ -1035,18 +1035,18 @@ prism54_get_retry(struct net_device *nde
 		    mgt_get_request(priv, DOT11_OID_MAXTXLIFETIME, 0, NULL, &r);
 		vwrq->value = r.u * 1024;
 		vwrq->flags = IW_RETRY_LIFETIME;
-	} else if ((vwrq->flags & IW_RETRY_LONG)) {
+	} else if ((vwrq->flags & IW_RETRY_MAX)) {
 		/* we are asked for the long retry limit */
 		rvalue |=
 		    mgt_get_request(priv, DOT11_OID_LONGRETRIES, 0, NULL, &r);
 		vwrq->value = r.u;
-		vwrq->flags = IW_RETRY_LIMIT | IW_RETRY_LONG;
+		vwrq->flags = IW_RETRY_LIMIT | IW_RETRY_MAX;
 	} else {
 		/* default. get the  short retry limit */
 		rvalue |=
 		    mgt_get_request(priv, DOT11_OID_SHORTRETRIES, 0, NULL, &r);
 		vwrq->value = r.u;
-		vwrq->flags = IW_RETRY_LIMIT | IW_RETRY_SHORT;
+		vwrq->flags = IW_RETRY_LIMIT | IW_RETRY_MIN;
 	}
 
 	return rvalue;
drivers/net/wireless/ray_cs.c +1 -1
index e82548e..4574290 100644
@@ -1173,7 +1173,7 @@ static int ray_set_essid(struct net_devi
 		return -EOPNOTSUPP;
 	} else {
 		/* Check the size of the string */
-		if(dwrq->length > IW_ESSID_MAX_SIZE) {
+		if(dwrq->length > IW_ESSID_MAX_SIZE + 1) {
 			return -E2BIG;
 		}
 
drivers/net/wireless/wl3501_cs.c +3 -3
index e3ae5f6..e0d294c 100644
@@ -1802,15 +1802,15 @@ static int wl3501_get_retry(struct net_d
 				      &retry, sizeof(retry));
 	if (rc)
 		goto out;
-	if (wrqu->retry.flags & IW_RETRY_LONG) {
-		wrqu->retry.flags = IW_RETRY_LIMIT | IW_RETRY_LONG;
+	if (wrqu->retry.flags & IW_RETRY_MAX) {
+		wrqu->retry.flags = IW_RETRY_LIMIT | IW_RETRY_MAX;
 		goto set_value;
 	}
 	rc = wl3501_get_mib_value(this, WL3501_MIB_ATTR_SHORT_RETRY_LIMIT,
 				  &retry, sizeof(retry));
 	if (rc)
 		goto out;
-	wrqu->retry.flags = IW_RETRY_LIMIT | IW_RETRY_SHORT;
+	wrqu->retry.flags = IW_RETRY_LIMIT | IW_RETRY_MIN;
 set_value:
 	wrqu->retry.value = retry;
 	wrqu->retry.disabled = 0;
drivers/net/wireless/zd1201.c +1 -1
index 80af9a9..f50ec10 100644
@@ -1218,7 +1218,7 @@ static int zd1201_set_essid(struct net_d
 		return -EINVAL;
 	if (data->length < 1)
 		data->length = 1;
-	zd->essidlen = data->length;
+	zd->essidlen = data->length-1;
 	memset(zd->essid, 0, IW_ESSID_MAX_SIZE+1);
 	memcpy(zd->essid, essid, data->length);
 	return zd1201_join(zd, zd->essid, zd->essidlen);
drivers/net/wireless/zd1211rw/zd_netdev.c +1 -1
index af3a7b3..440ef24 100644
@@ -82,7 +82,7 @@ static int iw_get_nick(struct net_device
 		       union iwreq_data *req, char *extra)
 {
 	strcpy(extra, "zd1211");
-	req->data.length = strlen(extra);
+	req->data.length = strlen(extra) + 1;
 	req->data.flags = 1;
 	return 0;
 }
include/linux/netdevice.h +1 -0
index 9264139..f159c72 100644
@@ -334,6 +334,7 @@ #define NETIF_F_ALL_CSUM	(NETIF_F_IP_CSU
 
 
 	struct net_device_stats* (*get_stats)(struct net_device *dev);
+	struct iw_statistics*	(*get_wireless_stats)(struct net_device *dev);
 
 	/* List of functions to handle Wireless Extensions (instead of ioctl).
 	 * See <net/iw_handler.h> for details. Jean II */
include/linux/wireless.h +4 -20
index a50a013..1358856 100644
@@ -1,7 +1,7 @@
 /*
  * This file define a set of standard wireless extensions
  *
- * Version :	21	14.3.06
+ * Version :	20	17.2.06
  *
  * Authors :	Jean Tourrilhes - HPL - <[email protected]>
  * Copyright (c) 1997-2006 Jean Tourrilhes, All Rights Reserved.
@@ -69,14 +69,9 @@ #define _LINUX_WIRELESS_H
 
 /***************************** INCLUDES *****************************/
 
-/* This header is used in user-space, therefore need to be sanitised
- * for that purpose. Those includes are usually not compatible with glibc.
- * To know which includes to use in user-space, check iwlib.h. */
-#ifdef __KERNEL__
 #include <linux/types.h>		/* for "caddr_t" et al		*/
 #include <linux/socket.h>		/* for "struct sockaddr" et al	*/
 #include <linux/if.h>			/* for IFNAMSIZ and co... */
-#endif	/* __KERNEL__ */
 
 /***************************** VERSION *****************************/
 /*
@@ -85,7 +80,7 @@ #endif	/* __KERNEL__ */
  * (there is some stuff that will be added in the future...)
  * I just plan to increment with each new version.
  */
-#define WIRELESS_EXT	21
+#define WIRELESS_EXT	20
 
 /*
  * Changes :
@@ -213,14 +208,6 @@ #define WIRELESS_EXT	21
  * V19 to V20
  * ----------
  *	- RtNetlink requests support (SET/GET)
- *
- * V20 to V21
- * ----------
- *	- Remove (struct net_device *)->get_wireless_stats()
- *	- Change length in ESSID and NICK to strlen() instead of strlen()+1
- *	- Add IW_RETRY_SHORT/IW_RETRY_LONG retry modifiers
- *	- Power/Retry relative values no longer * 100000
- *	- Add explicit flag to tell stats are in 802.11k RCPI : IW_QUAL_RCPI
  */
 
 /**************************** CONSTANTS ****************************/
@@ -461,7 +448,6 @@ #define IW_QUAL_DBM		0x08	/* Level + Noi
 #define IW_QUAL_QUAL_INVALID	0x10	/* Driver doesn't provide value */
 #define IW_QUAL_LEVEL_INVALID	0x20
 #define IW_QUAL_NOISE_INVALID	0x40
-#define IW_QUAL_RCPI		0x80	/* Level + Noise are 802.11k RCPI */
 #define IW_QUAL_ALL_INVALID	0x70
 
 /* Frequency flags */
@@ -514,12 +500,10 @@ #define IW_RETRY_ON		0x0000	/* No detail
 #define IW_RETRY_TYPE		0xF000	/* Type of parameter */
 #define IW_RETRY_LIMIT		0x1000	/* Maximum number of retries*/
 #define IW_RETRY_LIFETIME	0x2000	/* Maximum duration of retries in us */
-#define IW_RETRY_MODIFIER	0x00FF	/* Modify a parameter */
+#define IW_RETRY_MODIFIER	0x000F	/* Modify a parameter */
 #define IW_RETRY_MIN		0x0001	/* Value is a minimum  */
 #define IW_RETRY_MAX		0x0002	/* Value is a maximum */
 #define IW_RETRY_RELATIVE	0x0004	/* Value is not in seconds/ms/us */
-#define IW_RETRY_SHORT		0x0010	/* Value is for short packets  */
-#define IW_RETRY_LONG		0x0020	/* Value is for long packets */
 
 /* Scanning request flags */
 #define IW_SCAN_DEFAULT		0x0000	/* Default scan of the driver */
@@ -1033,7 +1017,7 @@ struct	iw_range
 	/* Note : this frequency list doesn't need to fit channel numbers,
 	 * because each entry contain its channel index */
 
-	__u32		enc_capa;	/* IW_ENC_CAPA_* bit field */
+	__u32		enc_capa; /* IW_ENC_CAPA_* bit field */
 };
 
 /*
net/core/net-sysfs.c +4 -1
index f47f319..1347276 100644
@@ -344,6 +344,8 @@ static ssize_t wireless_show(struct clas
 		if(dev->wireless_handlers &&
 		   dev->wireless_handlers->get_wireless_stats)
 			iw = dev->wireless_handlers->get_wireless_stats(dev);
+		else if (dev->get_wireless_stats)
+			iw = dev->get_wireless_stats(dev);
 		if (iw != NULL)
 			ret = (*format)(iw, buf);
 	}
@@ -463,7 +465,8 @@ int netdev_register_sysfs(struct net_dev
 		*groups++ = &netstat_group;
 
 #ifdef WIRELESS_EXT
-	if (net->wireless_handlers && net->wireless_handlers->get_wireless_stats)
+	if (net->get_wireless_stats
+	    || (net->wireless_handlers && net->wireless_handlers->get_wireless_stats))
 		*groups++ = &wireless_group;
 #endif
 
net/core/wireless.c +26 -41
index ffff0da..3168fca 100644
@@ -68,14 +68,6 @@
  *
  * v8 - 17.02.06 - Jean II
  *	o RtNetlink requests support (SET/GET)
- *
- * v8b - 03.08.06 - Herbert Xu
- *	o Fix Wireless Event locking issues.
- *
- * v9 - 14.3.06 - Jean II
- *	o Change length in ESSID and NICK to strlen() instead of strlen()+1
- *	o Make standard_ioctl_num and standard_event_num unsigned
- *	o Remove (struct net_device *)->get_wireless_stats()
  */
 
 /***************************** INCLUDES *****************************/
@@ -242,24 +234,24 @@ static const struct iw_ioctl_description
 	[SIOCSIWESSID	- SIOCIWFIRST] = {
 		.header_type	= IW_HEADER_TYPE_POINT,
 		.token_size	= 1,
-		.max_tokens	= IW_ESSID_MAX_SIZE,
+		.max_tokens	= IW_ESSID_MAX_SIZE + 1,
 		.flags		= IW_DESCR_FLAG_EVENT,
 	},
 	[SIOCGIWESSID	- SIOCIWFIRST] = {
 		.header_type	= IW_HEADER_TYPE_POINT,
 		.token_size	= 1,
-		.max_tokens	= IW_ESSID_MAX_SIZE,
+		.max_tokens	= IW_ESSID_MAX_SIZE + 1,
 		.flags		= IW_DESCR_FLAG_DUMP,
 	},
 	[SIOCSIWNICKN	- SIOCIWFIRST] = {
 		.header_type	= IW_HEADER_TYPE_POINT,
 		.token_size	= 1,
-		.max_tokens	= IW_ESSID_MAX_SIZE,
+		.max_tokens	= IW_ESSID_MAX_SIZE + 1,
 	},
 	[SIOCGIWNICKN	- SIOCIWFIRST] = {
 		.header_type	= IW_HEADER_TYPE_POINT,
 		.token_size	= 1,
-		.max_tokens	= IW_ESSID_MAX_SIZE,
+		.max_tokens	= IW_ESSID_MAX_SIZE + 1,
 	},
 	[SIOCSIWRATE	- SIOCIWFIRST] = {
 		.header_type	= IW_HEADER_TYPE_PARAM,
@@ -346,8 +338,8 @@ static const struct iw_ioctl_description
 		.max_tokens	= sizeof(struct iw_pmksa),
 	},
 };
-static const unsigned standard_ioctl_num = (sizeof(standard_ioctl) /
-					    sizeof(struct iw_ioctl_description));
+static const int standard_ioctl_num = (sizeof(standard_ioctl) /
+				       sizeof(struct iw_ioctl_description));
 
 /*
  * Meta-data about all the additional standard Wireless Extension events
@@ -397,8 +389,8 @@ static const struct iw_ioctl_description
 		.max_tokens	= sizeof(struct iw_pmkid_cand),
 	},
 };
-static const unsigned standard_event_num = (sizeof(standard_event) /
-					    sizeof(struct iw_ioctl_description));
+static const int standard_event_num = (sizeof(standard_event) /
+				       sizeof(struct iw_ioctl_description));
 
 /* Size (in bytes) of the various private data types */
 static const char iw_priv_type_size[] = {
@@ -473,6 +465,17 @@ static inline struct iw_statistics *get_
 	   (dev->wireless_handlers->get_wireless_stats != NULL))
 		return dev->wireless_handlers->get_wireless_stats(dev);
 
+	/* Old location, field to be removed in next WE */
+	if(dev->get_wireless_stats) {
+		static int printed_message;
+
+		if (!printed_message++)
+			printk(KERN_DEBUG "%s (WE) : Driver using old /proc/net/wireless support, please fix driver !\n",
+				dev->name);
+
+		return dev->get_wireless_stats(dev);
+	}
+
 	/* Not found */
 	return (struct iw_statistics *) NULL;
 }
@@ -1840,33 +1843,8 @@ #endif	/* CONFIG_NET_WIRELESS_RTNETLINK 
  */
 
 #ifdef WE_EVENT_RTNETLINK
-/* ---------------------------------------------------------------- */
-/*
- * Locking...
- * ----------
- *
- * Thanks to Herbert Xu <[email protected]> for fixing
- * the locking issue in here and implementing this code !
- *
- * The issue : wireless_send_event() is often called in interrupt context,
- * while the Netlink layer can never be called in interrupt context.
- * The fully formed RtNetlink events are queued, and then a tasklet is run
- * to feed those to Netlink.
- * The skb_queue is interrupt safe, and its lock is not held while calling
- * Netlink, so there is no possibility of dealock.
- * Jean II
- */
-
 static struct sk_buff_head wireless_nlevent_queue;
 
-static int __init wireless_nlevent_init(void)
-{
-	skb_queue_head_init(&wireless_nlevent_queue);
-	return 0;
-}
-
-subsys_initcall(wireless_nlevent_init);
-
 static void wireless_nlevent_process(unsigned long data)
 {
 	struct sk_buff *skb;
@@ -1943,6 +1921,13 @@ static inline void rtmsg_iwinfo(struct n
 	tasklet_schedule(&wireless_nlevent_tasklet);
 }
 
+static int __init wireless_nlevent_init(void)
+{
+	skb_queue_head_init(&wireless_nlevent_queue);
+	return 0;
+}
+
+subsys_initcall(wireless_nlevent_init);
 #endif	/* WE_EVENT_RTNETLINK */
 
 /* ---------------------------------------------------------------- */
net/ieee80211/softmac/ieee80211softmac_wx.c +4 -4
index 2aa779d..75320b6 100644
@@ -80,10 +80,10 @@ ieee80211softmac_wx_set_essid(struct net
 	 * If it's our network, ignore the change, we're already doing it!
 	 */
 	if((sm->associnfo.associating || sm->associated) &&
-	   (data->essid.flags && data->essid.length)) {
+	   (data->essid.flags && data->essid.length && extra)) {
 		/* Get the associating network */
 		n = ieee80211softmac_get_network_by_bssid(sm, sm->associnfo.bssid);
-		if(n && n->essid.len == data->essid.length &&
+		if(n && n->essid.len == (data->essid.length - 1) &&
 		   !memcmp(n->essid.data, extra, n->essid.len)) {
 			dprintk(KERN_INFO PFX "Already associating or associated to "MAC_FMT"\n",
 				MAC_ARG(sm->associnfo.bssid));
@@ -109,8 +109,8 @@ ieee80211softmac_wx_set_essid(struct net
 	sm->associnfo.static_essid = 0;
 	sm->associnfo.assoc_wait = 0;
 
-	if (data->essid.flags && data->essid.length) {
-		length = min((int)data->essid.length, IW_ESSID_MAX_SIZE);
+	if (data->essid.flags && data->essid.length && extra /*required?*/) {
+		length = min(data->essid.length - 1, IW_ESSID_MAX_SIZE);
 		if (length) {
 			memcpy(sm->associnfo.req_essid.data, extra, length);
 			sm->associnfo.static_essid = 1;
#80 lore raw
In reply to John W. Linville
Quoted 5 lines · John W. Linville

Unfortunately, I don't see any way to "fix" WE-21 without similarly breaking wireless-tools 29 and other "WE-21 aware" apps. And since I'll bet that the various WE-aware apps have checks like "if WE > 20" for managing ESSID length settings, we may have painted ourselves into a korner (sic).

OK, I'm going to ask a stupid question. Why is the kernel<->wireless driver interface have to be tied to the userspace<->wireless interface? The first is internal to the kernel and is free to change, and if it breaks out-of-tree drivers, I'll complain to Intel about why the !@#@ the ipw3945 driver hasn't been merged, but we've never made any guarantees about break out-of-tree drivers, so I wouldn't have the right to complain to anyone else.

The second is an external userspace ABI, and that should remain constant. It sounds like right now one gets pushed straight out to the other, but what if the wireless infrastructure layer in the kernel provided "interface glue" so you can translate between multiple interface versions --- and then force the userspace (or at least new versions of the userspace) to declare to the kernel what version of the interface they are expecting?

That's what we do with the stat system call. Userspace tells the kernel whether they want the v1, v2, v3, v4, or v5 version of the stat structure, and we have interface glue to support all of the old versions.

Is there some reason why this would be too hard to do with the current interface? Or is the arguement that if you're going to invest that much energy in fixing the userspace interface code, you would rather go to d80211/nl80211?

Regards,

- Ted

#81 lore raw
Quoted 10 lines · Theodore Tso
Quoted 5 lines · John W. Linville

Unfortunately, I don't see any way to "fix" WE-21 without similarly breaking wireless-tools 29 and other "WE-21 aware" apps. And since I'll bet that the various WE-aware apps have checks like "if WE > 20" for managing ESSID length settings, we may have painted ourselves into a korner (sic).

OK, I'm going to ask a stupid question. Why is the kernel<->wireless driver interface have to be tied to the userspace<->wireless interface?

They are not tied since WE-13. But you need a certain amount
of consistency, otherwise it's pure madness. If the driver does X (no-NUL), but the userspace sees Y (mandatory NUL), then both driver writer and application writer will go insane. You really want the API to be as transparent as possible.

But that's not the issue. The issue is that the userspace API
change was decided at the wireless summit last spring, and this was something that most wireless people were very strongly advocating for, including userspace people (Dan, Jouni). And most of it has already been implemented.
I think we can trust both Dan and Jouni to have a pretty good
idea of the impact of such changes. They are the one having to implement it and dealing with the angry users.
I've done many incompatible changes to the WE API over the
years, and it all went fine and dandy (anybody did notice those changes ?). They key is to get userspace in place when you do the kernel change.
As I said, I was against that change, and I'm the one being
flamed.

Quoted 2 lines

Is there some reason why this would be too hard to do with the current interface?

It's already done with the current interface, you can access
the API with either ioctls or RtNetlink.

Quoted 1 line

- Ted

	Have fun...

	Jean
#82 lore raw
Quoted 2 lines · On Tue, 3 Oct 2006 16:31:38 -0700 Jean Tourrilhes <[email protected]> wrote:

It's already done with the current interface, you can access
the API with either ioctls or RtNetlink.

Does this answer Ted's question though? Why can't userspace request the new v48 interface to get the changed API while older tools get the v47 interface and continue to work without needing to be upgraded?

Sean

A message that is not in the archive (sent off-list or not yet received): [email protected]

#84 lore raw
Quoted 10 lines · Sean
Quoted 2 lines · On Tue, 3 Oct 2006 16:31:38 -0700 Jean Tourrilhes <[email protected]> wrote:

It's already done with the current interface, you can access
the API with either ioctls or RtNetlink.

Does this answer Ted's question though? Why can't userspace request the new v48 interface to get the changed API while older tools get the v47 interface and continue to work without needing to be upgraded?

How does that happen in practice ? Kernel has no clue on what
userpace version is running.

Quoted 1 line

Sean

Jean

#85 lore raw
Quoted 3 lines · On Tue, 3 Oct 2006 17:30:31 -0700 Jean Tourrilhes <[email protected]> wrote:

How does that happen in practice ? Kernel has no clue on what
userpace version is running.

Ted mentioned that the way it works for stat is that userspace requests an API version and the kernel delivers it. So old versions request old API and new versions request new API. You only ever _add_ new API, and never remove older versions.

Sean

A message that is not in the archive (sent off-list or not yet received): [email protected]

#87 lore raw
Quoted 11 lines · Sean
Quoted 3 lines · On Tue, 3 Oct 2006 17:30:31 -0700 Jean Tourrilhes <[email protected]> wrote:

How does that happen in practice ? Kernel has no clue on what
userpace version is running.

Ted mentioned that the way it works for stat is that userspace requests an API version and the kernel delivers it. So old versions request old API and new versions request new API. You only ever _add_ new API, and never remove older versions.

I think the point is that the API currently has no such facility. Adding a new set of WE ioctls is unpalatable, both for general ioctl-haters and because the WE ioctl collection is quite broad. Maybe that could be solved by forcing the new WE stuff to use the netlink-based WE facilities, but then you are expanding the compatibility nightmare for whatever replaces WE.

John

Signature · 2 lines
John W. Linville
[email protected]
maintainer
#88 lore raw
In reply to Theodore Tso
Quoted 3 lines · Theodore Tso

OK, I'm going to ask a stupid question. Why is the kernel<->wireless driver interface have to be tied to the userspace<->wireless interface?

Haha. Because Jean thinks it isn't and thus everything is fine. But in reality it is.

Quoted 2 lines

Is there some reason why this would be too hard to do with the current interface?

Yes: drivers are expected to mostly handle the ioctls directly without a layer between them and userspace.

Quoted 3 lines

Or is the arguement that if you're going to invest that much energy in fixing the userspace interface code, you would rather go to d80211/nl80211?

cfg80211 and nl80211 actually do this abstraction, nl80211 gets requests and rewrites them to cfg80211 structures that are passed to the driver. I have plans for wext/cfg80211 compat code, essentially replacing the interface between the drivers and wext by cfg80211 and letting userspace not even be aware of it.

johannes

#89 lore raw
Quoted 6 lines · Johannes Berg
Quoted 3 lines · Theodore Tso

OK, I'm going to ask a stupid question. Why is the kernel<->wireless driver interface have to be tied to the userspace<->wireless interface?

Quoted 8 lines

Haha. Because Jean thinks it isn't and thus everything is fine. But in reality it is.

Quoted 2 lines

Is there some reason why this would be too hard to do with the current interface?

Yes: drivers are expected to mostly handle the ioctls directly without a layer between them and userspace.

Once again, your facts are totally wrong about Wireless
Extensions.
Have you ever looked at the code in the kernel ? I guarantee
you that adding whatever specific WE translation is quite easy. In this precise case, this would only increase confusion, so this should be considered bad API practice.
Regards,

Jean

#90 lore raw
Quoted 6 lines · Jean Tourrilhes

Once again, your facts are totally wrong about Wireless
Extensions.
Have you ever looked at the code in the kernel ? I guarantee
you that adding whatever specific WE translation is quite easy. In this precise case, this would only increase confusion, so this should be considered bad API practice.

Wireless Extensions has reached end-of-life, and so we only need to support what's out there in wide distribution.

Jeff

#91 lore raw
Quoted 2 lines · Jeff Garzik

Wireless Extensions has reached end-of-life, and so we only need to support what's out there in wide distribution.

Hmm, so what is going to replace it? I was messing about with my old powerbook G4 titanium, trying to make wpa_supplicant work when I realized the airport/orinoco driver used for my powerbook can't handle WPA since that apparently requires at least WE-18. I started looking into what it would take to teach the orinoco driver about WE>=18. But I suppose there is no point in my looking further if WE is heading to the great bit-bucket in the sky.

Is 'Wireless Extensions The Next Generation' described and documented somewhere? Or am I better off if I just give up and move on to some other more realistic project? :-)

-Erik

--
Erik B. Andersen http://codepoet-consulting.com/
--This message was written using 73% post-consumer electrons--

#92 lore raw
Quoted 13 lines · Erik Andersen
Quoted 2 lines · Jeff Garzik

Wireless Extensions has reached end-of-life, and so we only need to support what's out there in wide distribution.

Hmm, so what is going to replace it? I was messing about with my old powerbook G4 titanium, trying to make wpa_supplicant work when I realized the airport/orinoco driver used for my powerbook can't handle WPA since that apparently requires at least WE-18. I started looking into what it would take to teach the orinoco driver about WE>=18. But I suppose there is no point in my looking further if WE is heading to the great bit-bucket in the sky.

Quoted 3 lines

Is 'Wireless Extensions The Next Generation' described and documented somewhere? Or am I better off if I just give up and move on to some other more realistic project? :-)

Look around for references to nl80211 / cfg80211, particularly on the [email protected] list.

Jeff

P.S. Your mailer is generating buggy Mail-Followup-To lines.

#93 lore raw
In reply to Erik Andersen
Quoted 4 lines · Erik Andersen
Quoted 2 lines · Jeff Garzik

Wireless Extensions has reached end-of-life, and so we only need to support what's out there in wide distribution.

Quoted 8 lines

Hmm, so what is going to replace it? I was messing about with my old powerbook G4 titanium, trying to make wpa_supplicant work when I realized the airport/orinoco driver used for my powerbook can't handle WPA since that apparently requires at least WE-18. I started looking into what it would take to teach the orinoco driver about WE>=18. But I suppose there is no point in my looking further if WE is heading to the great bit-bucket in the sky.

Driver fixes to support later WE versions are still welcome. Information on supporting WPA for that driver will still be needed for cfg80211.

Thanks,

John

Signature · 2 lines
John W. Linville
[email protected]
maintainer
#94 lore raw
In reply to John W. Linville
Quoted 2 lines · John W. Linville

Wireless developers, it's time to get d80211 ready to merge...or figure-out how to get nl80211 on the current stack...hmmm...

Actually, cfg80211/nl80211 are not really tied to the stack, the intent always was to make them replace wext even for legacy drivers.

Hence, all legacy drivers need to do is assign their net_device's struct ieee80211_ptr and register a wiphy operations structure (and profit).

johannes

#95 lore raw
In reply to Jean Tourrilhes
Quoted 8 lines · Jean Tourrilhes

Now it's too late, those changes have propagated to userspace
tools, and are now shipping in some actual release of some distro. So, what are we going to say to Mandriva 2007 and FC6 users, to revert back to an *older* version of the tools ?
Because userspace has already been updated, we have only two
options, merge it now, or in 2.6.20.

Are you saying that compatibility is broken both ways?? Not only one needs new tools for the new kernels but also has to downgrade tools to work with older kernels??

Signature · 1 line
Dmitry
#96 lore raw
Quoted 10 lines · Dmitry Torokhov
Quoted 8 lines · Jean Tourrilhes

Now it's too late, those changes have propagated to userspace
tools, and are now shipping in some actual release of some distro. So, what are we going to say to Mandriva 2007 and FC6 users, to revert back to an *older* version of the tools ?
Because userspace has already been updated, we have only two
options, merge it now, or in 2.6.20.

Quoted 3 lines

Are you saying that compatibility is broken both ways?? Not only one needs new tools for the new kernels but also has to downgrade tools to work with older kernels??

No, it's not. But as soon as *some part* of WE-21 appears in
the kernel, the userspace expect the ESSID change. If we want to have WE-21 without the ESSID change, we need to fix userspace.

Quoted 1 line

Dmitry

Jean

#97 lore raw
Quoted 3 lines · Jean Tourrilhes

No, it's not. But as soon as *some part* of WE-21 appears in
the kernel, the userspace expect the ESSID change. If we want to have WE-21 without the ESSID change, we need to fix userspace.

Or leave WIRELESS_EXT at 20 and come up with a new way of versioning any future changes in WE.. Yes, having two different mechanisms for version number is ugly, but it could prevent userspace breakage.

(And based on the other messages in this thread, it might be useful to include the userspace program's idea of the version in those new commands to allow multiple interface versions to be supported by the kernel).

Signature · 1 line
Jouni Malinen                                            PGP id EFC895FA
#98 lore raw
Quoted 6 lines · Jouni Malinen
Quoted 3 lines · Jean Tourrilhes

No, it's not. But as soon as *some part* of WE-21 appears in
the kernel, the userspace expect the ESSID change. If we want to have WE-21 without the ESSID change, we need to fix userspace.

Quoted 8 lines

Or leave WIRELESS_EXT at 20 and come up with a new way of versioning any future changes in WE.. Yes, having two different mechanisms for version number is ugly, but it could prevent userspace breakage.

(And based on the other messages in this thread, it might be useful to include the userspace program's idea of the version in those new commands to allow multiple interface versions to be supported by the kernel).

or... don't use a NUMBER for this.

If you have a bitmap for supported features, it's much more powerful! That way you can even do this per driver/hardware, and you can add/retract individual capabilities rather than lumping everything into one big number.

#99 lore raw
Quoted 8 lines · Arjan van de Ven
Quoted 4 lines · Jouni Malinen

(And based on the other messages in this thread, it might be useful to include the userspace program's idea of the version in those new commands to allow multiple interface versions to be supported by the kernel).

Quoted 6 lines

or... don't use a NUMBER for this.

If you have a bitmap for supported features, it's much more powerful! That way you can even do this per driver/hardware, and you can add/retract individual capabilities rather than lumping everything into one big number.

Arjan (as usual) makes a very good suggestion here...

Signature · 2 lines
John W. Linville
[email protected]
#100 lore raw
In reply to John W. Linville
Quoted 9 lines · John W. Linville

Today's news seems to indicate that at least the major distros are already shipping the updated tools, or on the verge of shipping them. The window of breakage for most users looks like it will be fairly small, no matter what action taken.

The more we can clean-up the WE API, the easier it will be to implement the cfg80211 WE compatibility layer intended for wireless-dev. I think WE-21 makes things better in that respect.

Please correct me if I'm wrong: isn't it the case that a few of the distros are now coming out with wireless_tools.28 supporting WE-20, but current 2.6.18-git wireless drivers are WE-21, supported only by wireless_tools.29.pre10?

Hugh

#101 lore raw
Quoted 11 lines · Hugh Dickins
Quoted 9 lines · John W. Linville

Today's news seems to indicate that at least the major distros are already shipping the updated tools, or on the verge of shipping them. The window of breakage for most users looks like it will be fairly small, no matter what action taken.

The more we can clean-up the WE API, the easier it will be to implement the cfg80211 WE compatibility layer intended for wireless-dev. I think WE-21 makes things better in that respect.

Quoted 6 lines

Please correct me if I'm wrong: isn't it the case that a few of the distros are now coming out with wireless_tools.28 supporting WE-20, but current 2.6.18-git wireless drivers are WE-21, supported only by wireless_tools.29.pre10?

Hugh

wireless_tools.28 do support WE-21. I'm not a fool.

Jean