hostapd: fix IEEE 802.11r (fast roaming) defaults #940
Conversation
|
Thanks @devianceluka for doing the real work here; I'm just tidying up the commit message a bit and making sure it doesn't get lost. |
|
I've been using this (the original) patchset since november 2017, it would be great to get this merged. |
|
Also enable 11r support in wpad-mini since it is fairly small. |
|
Could you please rebase this on top of current master as the first part was already applied to master. |
|
Done, and added LTO to claw back some of the size increase. Not that it was massive anyway. |
Thats a great idea... I hate it to install full wpad for only 802.11r |
| @@ -133,6 +133,10 @@ ifneq ($(LOCAL_TYPE),hostapd) | |||
| ifdef CONFIG_WPA_SUPPLICANT_NO_TIMESTAMP_CHECK | |||
| TARGET_CFLAGS += -DNO_TIMESTAMP_CHECK | |||
| endif | |||
| ifeq ($(LOCAL_VARIANT),mini) | |||
| TARGET_CFLAGS += -flto -Os | |||
There was a problem hiding this comment.
Remove -Os, it's already in CFLAGS by default.
With so many devices having dual 2.4GHz + 5GHz radios, this isn't a particularly esoteric use case any more. Luci offers it by default too. On my test build the size increase is relatively modest: text data bss dec hex filename 425067 2704 356 428127 6885f wpad-mini 452003 2684 372 455059 6f193 wpad-mini+11r 717339 2796 588 720723 aff53 wpad-full Let's just enable it by default. We can offset some of the size increase by enabling LTO: 437615 2496 372 440483 6b8a3 wpad-mini+11r+lto Signed-off-by: David Woodhouse <dwmw2@infradead.org>
|
I hate to see this dying a death. The stumbling block here is the size (and functionality) increase for what is a mini(malist) package with attendant concerns that the size increase breaks the bank on 'legacy' routers that are short on space. The way forward that I can certainly vote for is to create (yet another) variant, I suggest 'wpad-basic' which gives you the mini(malist) defaults but adds a few finesse functionality points, such as 11r & 11w. Make wpad-basic the new default, so that 11r/w works out of the box with luci. I'm happy to work with whoever to get this going, though wpad/hostapd is going to be a voyage of discovery for me :-) |
|
2018-07-09 13:26 GMT+02:00 Kevin D-B <notifications@github.com>:
Make wpad-basic the new default, so that 11r/w works out of the
box with luci.
If you add "except for targets with the small_flash feature flag set": ACK.
|
|
As mips is the concern, what's the difference in size between with 11r (incl -flto) and without? |
|
that is not what is being proposed at all. |
|
Ahh, so there's going to be yet another variant? |
|
yes |
|
Wouldn't it be a good idea to look at the size first? If it's ends up being the ~same as the current version/variant is it really worth bothering making another one? |
|
The commit has some size info included, also with -lto for comparison - I doubt the mips numbers are significantly different in % terms. I've seen pressure (in IRC) that an increase in size caused by an increase in functionality on 'mini' packages is not desirable. The only way out of that is to create a new package. There was talk of renaming wpad-mini to wpad-tiny and have a new wpad-mini. Personally, if mini is short for minimalist, then it should provide the minimum required functionality without finesse features for the smallest size. Basic can take that as a starting point and add some finesse features, obviously with a size penalty. |
|
I just pushed a binutils update + upstream backport, which fixes some LTO issues in my build, along with some changes to enable full LTO for all hostapd/wpa_supplicant variants, including make jobserver support |
|
@ldir-EDB0 |
|
With LTO enabled for both the 11r and non-11r wpad-mini build, the size difference for the full gzip-compressed .ipk file is around 25k. I'd estimate that this feature costs around 15-20k of flash space after LZMA. |
|
@nbd168 |
|
duplicate of what we have pending in patchwork |
Use ft_psk_generate_local=1 by default, as it makes everything else fairly
trivial. All of the r0kh/r1kh and key management stuff goes away and hostapd
fairly much does it all for us.
We do need to provide nas_identifier, which can be derived from the BSSID,
and we need to generate a mobility_domain, for which we default to the first
four chars of the md5sum of the SSID.
The complex manual setup should also still work, but the defaults also
now work easily out of the box. Verified by manually running hostapd
(with the autogenerated config) and watching the debug output:
wlan2: STA ac:37:43:a0:a6:ae WPA: FT authentication already completed - do not start 4-way handshake
This was previous submitted to LEDE in
lede-project/source#1382
[dwmw2: Rewrote commit message]
Signed-off-by: Gospod Nassa devianca@gmail.com
Signed-off-by: David Woodhouse dwmw2@infradead.org