Re: [PATCH] RDMA/ucma: Put a lock around every call to the rdma_cm layer

From: Jason Gunthorpe <jgg@mellanox.com>
To: Eric Biggers <ebiggers@kernel.org>
Cc: "linux-rdma@vger.kernel.org" <linux-rdma@vger.kernel.org>
Subject: Re: [PATCH] RDMA/ucma: Put a lock around every call to the rdma_cm layer
Date: Mon, 9 Mar 2020 16:30:12 -0300	[thread overview]
Message-ID: <20200309193012.GA13183@mellanox.com> (raw)
In-Reply-To: <20200307204153.GJ15444@sol.localdomain>

On Sat, Mar 07, 2020 at 12:41:53PM -0800, Eric Biggers wrote:
> On Wed, Feb 19, 2020 at 08:22:25PM +0000, Jason Gunthorpe wrote:
> > On Tue, Feb 18, 2020 at 10:07:01PM -0800, Eric Biggers wrote:
> > > > these 11 lets include them as  well. I wasn't able to find a way to
> > > > search for things, this list is from your past email, thanks.
> > > > 
> > > 
> > > Unfortunately I haven't had time to work on syzkaller bugs lately, so I can't
> > > provide an updated list until I go through the long backlog of bugs.
> > 
> > Ok
> 
> Here's an updated list:
> 
> --------------------------------------------------------------------------------
> Title:              general protection fault in rds_ib_add_one
> Last occurred:      0 days ago
> Reported:           12 days ago
> Branches:           Mainline and others
> Dashboard link:     https://syzkaller.appspot.com/bug?id=15f96d171c64999196ac7db3de107f24b9182a8e
> Original thread:    https://lore.kernel.org/lkml/000000000000b9b7d4059f4e4ac7@google.com/T/#u
> 
> This bug has a C reproducer.

Looks like this is fixed by Hillf

> --------------------------------------------------------------------------------
> Title:              INFO: trying to register non-static key in xa_destroy
> Last occurred:      0 days ago
> Reported:           11 days ago
> Branches:           Mainline and others
> Dashboard link:     https://syzkaller.appspot.com/bug?id=c0a75a31c5fa84e6e5d3131fd98a5b56e2141b9a
> Original thread:    https://lore.kernel.org/lkml/00000000000046895c059f5cae37@google.com/T/#u
> 
> This bug has a C reproducer.

Fixed in v5.6-rc5

> --------------------------------------------------------------------------------
> Title:              general protection fault in nldev_stat_set_doit
> Last occurred:      4 days ago
> Reported:           11 days ago
> Branches:           Mainline and others
> Dashboard link:     https://syzkaller.appspot.com/bug?id=1fbcb607cf49d8b5a3c8e056971f045f9bfa34f3
> Original thread:    https://lore.kernel.org/lkml/0000000000004aa34d059f5caedc@google.com/T/#u
> 
> This bug has a C reproducer.

Fixed in v5.6-rc5

> --------------------------------------------------------------------------------
> Title:              BUG: corrupted list in _cma_attach_to_dev
> Last occurred:      2 days ago
> Reported:           6 days ago
> Branches:           Mainline
> Dashboard link:     https://syzkaller.appspot.com/bug?id=067b1e60bab1b617c1208f078cd76c9087f070e0
> Original thread:    https://lore.kernel.org/lkml/000000000000cfed90059fcfdccb@google.com/T/#u
> 
> This bug has a C reproducer.

Most likely fixed by this patch, syzkaller is re-testing

> --------------------------------------------------------------------------------
> Title:              WARNING: kobject bug in ib_register_device
> Last occurred:      1 day ago
> Reported:           12 days ago
> Branches:           Mainline and others
> Dashboard link:     https://syzkaller.appspot.com/bug?id=805ad726feb6910e35088ae7bbe61f4125e573b7
> Original thread:    https://lore.kernel.org/lkml/000000000000026ac5059f4e27f3@google.com/T/#u
> 
> This bug has a C reproducer.

Oh, this wasn't sent to rdma, yes, obvious rdma bug, made a patch

> --------------------------------------------------------------------------------
> Title:              BUG: corrupted list in cma_listen_on_dev
> Last occurred:      4 days ago
> Reported:           4 days ago
> Branches:           Mainline
> Dashboard link:     https://syzkaller.appspot.com/bug?id=e8fcdea4e5a443c597c94fb6eda7d6646eafe6a2
> Original thread:    https://lore.kernel.org/lkml/00000000000020c5d205a001c308@google.com/T/#u
> 
> This bug has a C reproducer.

Fixed by this patch, syzkaller confirmed, now duped to another bug

> --------------------------------------------------------------------------------
> Title:              KASAN: use-after-free Read in rxe_query_port
> Last occurred:      0 days ago
> Reported:           6 days ago
> Branches:           Mainline and others
> Dashboard link:     https://syzkaller.appspot.com/bug?id=f00443e97b44c466dc75edc31601110bf62a6f69
> Original thread:    https://lore.kernel.org/lkml/0000000000000c9e12059fc941ff@google.com/T/#u
> 
> Unfortunately, this bug does not have a reproducer.

Perhaps Yanjun Zhu will look at this

> --------------------------------------------------------------------------------
> Title:              WARNING in ib_free_port_attrs
> Last occurred:      1 day ago
> Reported:           6 days ago
> Branches:           net and net-next
> Dashboard link:     https://syzkaller.appspot.com/bug?id=4ec089798f282f2d2c3219151e420ed1ba10120d
> Original thread:    https://lore.kernel.org/lkml/000000000000460717059fd83734@google.com/T/#u
> 
> Unfortunately, this bug does not have a reproducer.

Parav and I looked at this for a while and couldn't figure how how it
is possible. Hoping for a reproducer

> --------------------------------------------------------------------------------
> Title:              INFO: task hung in rdma_destroy_id
> Last occurred:      3 days ago
> Reported:           5 days ago
> Branches:           Mainline and others
> Dashboard link:     https://syzkaller.appspot.com/bug?id=e89b86960c3636f57dbb16bb25a829377ebdf43d
> Original thread:    https://lore.kernel.org/lkml/00000000000059e701059fe3ec2f@google.com/T/#u
> 
> Unfortunately, this bug does not have a reproducer.

Most likely fixed by this patch

> --------------------------------------------------------------------------------
> Title:              general protection fault in kobject_get
> Last occurred:      6 days ago
> Reported:           6 days ago
> Branches:           net-next
> Dashboard link:     https://syzkaller.appspot.com/bug?id=f8e0f99b310558dd489cc7427711a640c10b93e5
> Original thread:    https://lore.kernel.org/lkml/000000000000c4b371059fd83a92@google.com/T/#u
> 
> Unfortunately, this bug does not have a reproducer.

Really surprised no reproducer, this is not a race bug. I wrote a fix,
it is being tested now.

> --------------------------------------------------------------------------------
> Title:              WARNING: kobject bug in add_one_compat_dev
> Last occurred:      8 days ago
> Reported:           10 days ago
> Branches:           linux-next and net-next
> Dashboard link:     https://syzkaller.appspot.com/bug?id=f8880fdc3cd0ba268421672360cf79bfa7fa4272
> Original thread:    https://lore.kernel.org/lkml/0000000000005f77d6059f888f2e@google.com/T/#u
> 
> Unfortunately, this bug does not have a reproducer.

Hmm. I wonder if this is because 'dev_set_name' failed and we ignored
it? Is that possible with this log? Lets fix that at least - I have no
other idea how we could get an empty name.

> --------------------------------------------------------------------------------
> Title:              WARNING in srp_remove_one
> Last occurred:      9 days ago
> Reported:           6 days ago
> Branches:           Mainline
> Dashboard link:     https://syzkaller.appspot.com/bug?id=16a5827f8f6f6ef0967e6492ffb2e2ca54c8c0fb
> Original thread:    https://lore.kernel.org/lkml/000000000000144d79059fc9415d@google.com/T/#u
> 
> Unfortunately, this bug does not have a reproducer.

This looks a lot like 'WARNING in ib_free_port_attrs' - I don't have a
clear idea how these sysfs errors are possible. I wonder if there is
something strange going on in sysfs land during net ns actions?

Thanks,
Jason