Bug 71073
| Summary: | [vlan] a media ioctl to vlan(4) causes panic if parent is bge(4) | ||
|---|---|---|---|
| Product: | Base System | Reporter: | Yar Tikhiy <yar> |
| Component: | kern | Assignee: | Andre Oppermann <andre> |
| Status: | Closed FIXED | ||
| Severity: | Affects Only Me | CC: | andre |
| Priority: | Normal | ||
| Version: | 6.0-CURRENT | ||
| Hardware: | Any | ||
| OS: | Any | ||
Responsible Changed From-To: freebsd-bugs->andre Take over. State Changed From-To: open->analyzed Bill Paul and me are working on this. A possible fix is in the works. State Changed From-To: analyzed->closed The problem has been fixed in sys/net/if_vlan.c rev. 1.69. The bad news are that if_vlan.c rev. 1.69 hardly could fix this problem. First, I saw this problem *after* rev. 1.69 had been committed. Second, this problem had to do with vlan-bge interoperation over media control and link state, not with multicast stuff. The good news are that I've just heard of vlan(4) working OK over bge(4) in RELENG_6. Alas, I cannot confirm that myself right now because my test machine with bge(4) has been dismantled. I hope I'll be able to stick my bge(4) in another spare machine RSN. Perhaps we should put this PR in a state like "feedback" for now, eh? -- Yar Yar Tikhiy wrote:
>
> The bad news are that if_vlan.c rev. 1.69 hardly could fix this
> problem. First, I saw this problem *after* rev. 1.69 had been
> committed. Second, this problem had to do with vlan-bge interoperation
> over media control and link state, not with multicast stuff.
>
> The good news are that I've just heard of vlan(4) working OK over
> bge(4) in RELENG_6. Alas, I cannot confirm that myself right now
> because my test machine with bge(4) has been dismantled. I hope
> I'll be able to stick my bge(4) in another spare machine RSN.
>
> Perhaps we should put this PR in a state like "feedback" for now,
> eh?
I've got two bge interfaces in my main FreeBSD development box and
they are just fine including vlan and link state changes. If not
the change I've singled out, then some other must have fixed it.
However that doesn't change the overvation that it reliably works
now. Which is good. :-)
--
Andre
On Tue, Oct 11, 2005 at 05:59:11PM +0200, Andre Oppermann wrote:
> >
> > The bad news are that if_vlan.c rev. 1.69 hardly could fix this
> > problem. First, I saw this problem *after* rev. 1.69 had been
> > committed. Second, this problem had to do with vlan-bge interoperation
> > over media control and link state, not with multicast stuff.
> >
> > The good news are that I've just heard of vlan(4) working OK over
> > bge(4) in RELENG_6. Alas, I cannot confirm that myself right now
> > because my test machine with bge(4) has been dismantled. I hope
> > I'll be able to stick my bge(4) in another spare machine RSN.
> >
> > Perhaps we should put this PR in a state like "feedback" for now,
> > eh?
>
> I've got two bge interfaces in my main FreeBSD development box and
> they are just fine including vlan and link state changes. If not
> the change I've singled out, then some other must have fixed it.
> However that doesn't change the overvation that it reliably works
> now. Which is good. :-)
It's great news since I'm still in the hope of putting my bge back
to work and development (I just ought to, I got it as a donation :-)
Thanks a lot!
--
Yar
|
Upon receiving a SIOCGIFMEDIA ioctl, vlan(4) will grab the vlan mutex and call the bge(4) parent's if_ioctl handler. The sequence of calls will end up in vlan_link_state(), which tries to obtain the vlan mutex again and so panics the system. The stack trace is as follows: #24 0xc049de23 in panic ( fmt=0xc05d4b68 "_mtx_lock_sleep: recursed on non-recursive mutex %s @ %s:%d\n") at /usr/src/sys/kern/kern_shutdown.c:542 #25 0xc0496232 in _mtx_lock_sleep (m=0xc064f500, td=0xc0d889a0, opts=0, file=0xc05dead1 "/usr/src/sys/net/if_vlan.c", line=819) at /usr/src/sys/kern/kern_mutex.c:436 #26 0xc0495f50 in _mtx_lock_flags (m=0xc064f500, opts=0, file=0xc05dead1 "/usr/src/sys/net/if_vlan.c", line=819) at /usr/src/sys/kern/kern_mutex.c:253 #27 0xc0502657 in vlan_link_state (ifp=0xc0e5e000, link=1) at /usr/src/sys/net/if_vlan.c:819 #28 0xc04500d1 in miibus_linkchg (dev=0xc0df8000) at /usr/src/sys/dev/mii/mii.c:270 #29 0xc04508f1 in mii_phy_update (sc=0xc0e7a580, cmd=3) at miibus_if.h:62 #30 0xc044d38b in brgphy_service (sc=0xc0e7a580, mii=0xc0e76940, cmd=3) at /usr/src/sys/dev/mii/brgphy.c:391 #31 0xc04502cf in mii_pollstat (mii=0xc0e76940) at /usr/src/sys/dev/mii/mii.c:384 #32 0xc0445583 in bge_ifmedia_sts (ifp=0x0, ifmr=0xc5836c60) at /usr/src/sys/dev/bge/if_bge.c:3411 #33 0xc0501553 in ifmedia_ioctl (ifp=0xc0e5e000, ifr=0x0, ifm=0xc0e76940, cmd=0) at /usr/src/sys/net/if_media.c:281 #34 0xc04457bb in bge_ioctl (ifp=0xc0e5e000, command=3223873848, data=0x0) at /usr/src/sys/dev/bge/if_bge.c:3489 #35 0xc05027d3 in vlan_ioctl (ifp=0xc0f04000, cmd=0, data=0xc5836c60 "vlan99") at /usr/src/sys/net/if_vlan.c:872 #36 0xc04fd348 in ifhwioctl (cmd=3223873848, ifp=0xc0f04000, data=0xc5836c60 "vlan99", td=0x0) at /usr/src/sys/net/if.c:1288 #37 0xc04fd44f in ifioctl (so=0xc0f0eb64, cmd=3223873848, data=0xc5836c60 "vlan99", td=0xc0d889a0) at /usr/src/sys/net/if.c:1341 #38 0xc04c39b1 in soo_ioctl (fp=0x0, cmd=0, data=0xc5836c60, active_cred=0xc0d7be00, td=0x0) at /usr/src/sys/kern/sys_socket.c:202 #39 0xc04be444 in ioctl (td=0xc0d889a0, uap=0xc5836d14) at file.h:258 I found that using fxp(4) instead of bge(4) didn't result in system panic. Fix: Unfortunately, I don't understand yet which code is actually incorrect, bge's or vlan's. I hope andre@ (CC'd) could shed light on the issue. Andre, would you mind dropping a word? That seems to be your code ;-) How-To-Repeat: Just boot a system with vlan interfaces set up in rc.conf to attach to a bge. The system will panic as soon as the rc script `netif' will try to display post-configuration status of the interfaces.