diff mbox series

[v2,03/10] blktrace: fix debugfs use after free

Message ID 20200419194529.4872-4-mcgrof@kernel.org
State New, archived
Headers show
Series block: fix blktrace debugfs use after free | expand

Commit Message

Luis Chamberlain April 19, 2020, 7:45 p.m. UTC
On commit 6ac93117ab00 ("blktrace: use existing disk debugfs directory")
merged on v4.12 Omar fixed the original blktrace code for request-based
drivers (multiqueue). This however left in place a possible crash, if you
happen to abuse blktrace in a way it was not intended, and even more so
with our current asynchronous request_queue removal.

Namely, if you loop adding a device, setup the blktrace with BLKTRACESETUP,
forget to BLKTRACETEARDOWN, and then just remove the device you end up
with a panic:

[  107.193134] debugfs: Directory 'loop0' with parent 'block' already present!
[  107.254615] BUG: kernel NULL pointer dereference, address: 00000000000000a0
[  107.258785] #PF: supervisor write access in kernel mode
[  107.262035] #PF: error_code(0x0002) - not-present page
[  107.264106] PGD 0 P4D 0
[  107.264404] Oops: 0002 [#1] SMP NOPTI
[  107.264803] CPU: 8 PID: 674 Comm: kworker/8:2 Tainted: G            E 5.6.0-rc7-next-20200327 #1
[  107.265712] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.13.0-1 04/01/2014
[  107.266553] Workqueue: events __blk_release_queue
[  107.267051] RIP: 0010:down_write+0x15/0x40
[  107.267488] Code: eb ca e8 ee a5 8d ff cc cc cc cc cc cc cc cc cc cc cc cc cc cc 0f 1f 44 00 00 55 48 89 fd e8 52 db ff ff 31 c0 ba 01 00 00 00 <f0> 48 0f b1 55 00 75 0f  65 48 8b 04 25 c0 8b 01 00 48 89 45 08 5d
[  107.269300] RSP: 0018:ffff9927c06efda8 EFLAGS: 00010246
[  107.269841] RAX: 0000000000000000 RBX: ffff8be7e73b0600 RCX: ffffff8100000000
[  107.270559] RDX: 0000000000000001 RSI: ffffff8100000000 RDI: 00000000000000a0
[  107.271281] RBP: 00000000000000a0 R08: ffff8be7ebc80fa8 R09: ffff8be7ebc80fa8
[  107.272001] R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000000
[  107.272722] R13: ffff8be7efc30400 R14: ffff8be7e0571200 R15: 00000000000000a0
[  107.273475] FS:  0000000000000000(0000) GS:ffff8be7efc00000(0000) knlGS:0000000000000000
[  107.274346] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  107.274968] CR2: 00000000000000a0 CR3: 000000042abee003 CR4: 0000000000360ee0
[  107.275710] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  107.276465] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[  107.277214] Call Trace:
[  107.277532]  simple_recursive_removal+0x4e/0x2e0
[  107.278049]  ? debugfs_remove+0x60/0x60
[  107.278493]  debugfs_remove+0x40/0x60
[  107.278922]  blk_trace_free+0xd/0x50
[  107.279339]  __blk_trace_remove+0x27/0x40
[  107.279797]  blk_trace_shutdown+0x30/0x40
[  107.280256]  __blk_release_queue+0xab/0x110
[  107.280734]  process_one_work+0x1b4/0x380
[  107.281194]  worker_thread+0x50/0x3c0
[  107.281622]  kthread+0xf9/0x130
[  107.281994]  ? process_one_work+0x380/0x380
[  107.282467]  ? kthread_park+0x90/0x90
[  107.282895]  ret_from_fork+0x1f/0x40
[  107.283316] Modules linked in: loop(E) <etc>
[  107.288562] CR2: 00000000000000a0
[  107.288957] ---[ end trace b885d243d441bbce ]---

This splat happens to be very similar to the one reported via
kernel.org korg#205713, only that korg#205713 was for v4.19.83
and the above now includes the simple_recursive_removal() introduced
via commit a3d1e7eb5abe ("simple_recursive_removal(): kernel-side rm
-rf for ramfs-style filesystems") merged on v5.6.

korg#205713 then was used to create CVE-2019-19770 and claims that
the bug is in a use-after-free in the debugfs core code. The
implications of this being a generic UAF on debugfs would be
much more severe, as it would imply parent dentries can sometimes
not be positive, which we hold by design is just not possible.

Below is the splat explained with a bit more details, explaining
what is happening in userspace, kernel, and a print of the CPU on,
which the code runs on:

load loopback module
[   13.603371] == blk_mq_debugfs_register(12) start
[   13.604040] == blk_mq_debugfs_register(12) q->debugfs_dir created
[   13.604934] == blk_mq_debugfs_register(12) end
[   13.627382] == blk_mq_debugfs_register(12) start
[   13.628041] == blk_mq_debugfs_register(12) q->debugfs_dir created
[   13.629240] == blk_mq_debugfs_register(12) end
[   13.651667] == blk_mq_debugfs_register(12) start
[   13.652836] == blk_mq_debugfs_register(12) q->debugfs_dir created
[   13.655107] == blk_mq_debugfs_register(12) end
[   13.684917] == blk_mq_debugfs_register(12) start
[   13.687876] == blk_mq_debugfs_register(12) q->debugfs_dir created
[   13.691588] == blk_mq_debugfs_register(13) end
[   13.707320] == blk_mq_debugfs_register(13) start
[   13.707863] == blk_mq_debugfs_register(13) q->debugfs_dir created
[   13.708856] == blk_mq_debugfs_register(13) end
[   13.735623] == blk_mq_debugfs_register(13) start
[   13.736656] == blk_mq_debugfs_register(13) q->debugfs_dir created
[   13.738411] == blk_mq_debugfs_register(13) end
[   13.763326] == blk_mq_debugfs_register(13) start
[   13.763972] == blk_mq_debugfs_register(13) q->debugfs_dir created
[   13.765167] == blk_mq_debugfs_register(13) end
[   13.779510] == blk_mq_debugfs_register(13) start
[   13.780522] == blk_mq_debugfs_register(13) q->debugfs_dir created
[   13.782338] == blk_mq_debugfs_register(13) end
[   13.783521] loop: module loaded

LOOP_CTL_DEL(loop0) #1
[   13.803550] = __blk_release_queue(4) start
[   13.807772] == blk_trace_shutdown(4) start
[   13.810749] == blk_trace_shutdown(4) end
[   13.813437] = __blk_release_queue(4) calling blk_mq_debugfs_unregister()
[   13.817593] ==== blk_mq_debugfs_unregister(4) begin
[   13.817621] ==== blk_mq_debugfs_unregister(4) debugfs_remove_recursive(q->debugfs_dir)
[   13.821203] ==== blk_mq_debugfs_unregister(4) end q->debugfs_dir is NULL
[   13.826166] = __blk_release_queue(4) blk_mq_debugfs_unregister() end
[   13.832992] = __blk_release_queue(4) end

LOOP_CTL_ADD(loop0) #1
[   13.843742] == blk_mq_debugfs_register(7) start
[   13.845569] == blk_mq_debugfs_register(7) q->debugfs_dir created
[   13.848628] == blk_mq_debugfs_register(7) end

BLKTRACE_SETUP(loop0) #1
[   13.850924] == blk_trace_ioctl(7, BLKTRACESETUP) start
[   13.852852] === do_blk_trace_setup(7) start
[   13.854580] === do_blk_trace_setup(7) creating directory
[   13.856620] === do_blk_trace_setup(7) using what debugfs_lookup() gave
[   13.860635] === do_blk_trace_setup(7) end with ret: 0
[   13.862615] == blk_trace_ioctl(7, BLKTRACESETUP) end

LOOP_CTL_DEL(loop0) #2
[   13.883304] = __blk_release_queue(7) start
[   13.885324] == blk_trace_shutdown(7) start
[   13.887197] == blk_trace_shutdown(7) calling __blk_trace_remove()
[   13.889807] == __blk_trace_remove(7) start
[   13.891669] === blk_trace_cleanup(7) start
[   13.911656] ====== blk_trace_free(7) start

This is a scheduled __blk_release_queue(). Note that at this point
we know that the old device is gone as the gendisk removal is
synchronous.

LOOP_CTL_ADD(loop0) #2
[   13.912709] == blk_mq_debugfs_register(2) start

This is a *new* device, as we know the last one was removed synchronously.

---> From LOOP_CTL_DEL(loop0) #2
[   13.915887] ====== blk_trace_free(7) end

---> From LOOP_CTL_ADD(loop0) #2
[   13.918359] debugfs: Directory 'loop0' with parent 'block' already present!

Above we have a case where the upon LOOP_CTL_ADD(loop) #2 the
debugfs_create_dir() start_creating() call will have found an existing dentry,
and return ERR_PTR(-EEXIST), and so the q->debugfs_dir will be initialized to
NULL. This clash happens because an asynchronous __blk_release_queue() has
been scheduled from the above LOOP_CTL_DEL(loop0) #2 call, and it has not yet
removed the old loop0 debugfs directory.

[   13.926433] == blk_mq_debugfs_register(2) q->debugfs_dir created

This is a huge assumption, since we never check for errors on
blk_mq_debugfs_register(), and so the child files we need to
create *can* simply be created using NULL as as the parent, in
which case debugfs would use debugfs_mount->mnt_root as the parent.

[   13.930373] == blk_mq_debugfs_register(2) end

BLKTRACE_SETUP(loop0) #2
[   13.933961] == blk_trace_ioctl(2, BLKTRACESETUP) start
[   13.936758] === do_blk_trace_setup(2) start
[   13.938944] === do_blk_trace_setup(2) creating directory
[   13.941029] === do_blk_trace_setup(2) using what debugfs_lookup() gave

Here debugfs_lookup() will have found the old debugfs directory, even
though q->debugfs_dir was empty when initializing this device. So the
files created can actually be created where they should have.

---> From LOOP_CTL_DEL(loop0) #2
[   13.971046] === blk_trace_cleanup(7) end
[   13.973175] == __blk_trace_remove(7) end
[   13.975352] == blk_trace_shutdown(7) end
[   13.977415] = __blk_release_queue(7) calling blk_mq_debugfs_unregister()
[   13.980645] ==== blk_mq_debugfs_unregister(7) begin
[   13.980696] ==== blk_mq_debugfs_unregister(7) debugfs_remove_recursive(q->debugfs_dir)

This is a continuation from the old scheduled  __blk_release_queue(). It
is using another request_queue structure than what the currently activated
blktrace setup is using.

Here now the dentry where the blktrace setup files were created on top
of is gone.  In fact, since debugfs_remove_recursive() is used, *all* of
those blktrace debugfs files should have been removed as well, including
the relay. This is despite the fact that the removal and blktrace setup
are working on two different request_queue structures.

[   13.983118] ==== blk_mq_debugfs_unregister(7) end q->debugfs_dir is NULL
[   13.986945] = __blk_release_queue(7) blk_mq_debugfs_unregister() end
[   13.993155] = __blk_release_queue(7) end

The scheduled __blk_release_queue completes finally.

---> From BLKTRACE_SETUP(loop0) #2
[   13.995928] === do_blk_trace_setup(2) end with ret: 0
[   13.997623] == blk_trace_ioctl(2, BLKTRACESETUP) end

At this point the old blktrace *setup* is complete though, and let us
recall that the old dentry is still used. But recall that we just had
our old device removed, and since debugfs_remove_recursive() was used,
so are all of its files. This request_queue however is still active.

LOOP_CTL_DEL(loop0) #3
[   14.035119] = __blk_release_queue(2) start
[   14.036925] == blk_trace_shutdown(2) start
[   14.038518] == blk_trace_shutdown(2) calling __blk_trace_remove()
[   14.040829] == __blk_trace_remove(2) start
[   14.042413] === blk_trace_cleanup(2) start

This removal will cleanup the blktrace setup. Note the trap though,
the blktrace setup will already have been cleaned up since the
scheduled removal from instance #2 removed the files on the same
debugfs path.

The __blk_release_queue() is scheduled and is asynchronous so can
continue later, and that is where the crash happens, on CPU 2.

LOOP_CTL_ADD(loop0) #3
[   14.072522] == blk_mq_debugfs_register(6) start

---> From LOOP_CTL_DEL(loop0) #3
[   14.075151] ====== blk_trace_free(2) start

---> From LOOP_CTL_ADD(loop0) #3
[   14.075882] == blk_mq_debugfs_register(6) q->debugfs_dir created

---> From LOOP_CTL_DEL(loop0) #3
The panic happens on CPU 2 when we continue with the old scheduled
removal of the request_queue, which will run into a blktrace setup
with NULL pointers.

[   14.078624] BUG: kernel NULL pointer dereference, address: 00000000000000a0
[   14.084332] == blk_mq_debugfs_register(6) end
[   14.086971] #PF: supervisor write access in kernel mode
[   14.086974] #PF: error_code(0x0002) - not-present page
[   14.086977] PGD 0 P4D 0
[   14.086984] Oops: 0002 [#1] SMP NOPTI
[   14.086990] CPU: 2 PID: 287 Comm: kworker/2:2 Tainted: G            E 5.6.0-next-20200403+ #54
[   14.086991] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.13.0-1 04/01/2014
[   14.087002] Workqueue: events __blk_release_queue
[   14.087011] RIP: 0010:down_write+0x15/0x40
[   14.090300] == blk_trace_ioctl(6, BLKTRACESETUP) start
[   14.093277] Code: eb ca e8 3e 34 8d ff cc cc cc cc cc cc cc cc cc cc
cc cc cc cc 0f 1f 44 00 00 55 48 89 fd e8 52 db ff ff 31 c0 ba 01 00 00
00 <f0> 48 0f b1 55 00 75 0f 65 48 8b 04 25 c0 8b 01 00 48 89 45 08 5d
[   14.093280] RSP: 0018:ffffc28a00533da8 EFLAGS: 00010246
[   14.093284] RAX: 0000000000000000 RBX: ffff9f7a24d07980 RCX: ffffff8100000000
[   14.093286] RDX: 0000000000000001 RSI: ffffff8100000000 RDI: 00000000000000a0
[   14.093287] RBP: 00000000000000a0 R08: 0000000000000000 R09: 0000000000000019
[   14.093289] R10: 0000000000000774 R11: 0000000000000000 R12: 0000000000000000
[   14.093291] R13: ffff9f7a2fab0400 R14: ffff9f7a21dd1140 R15: 00000000000000a0
[   14.093294] FS:  0000000000000000(0000) GS:ffff9f7a2fa80000(0000) knlGS:0000000000000000
[   14.093296] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   14.093298] CR2: 00000000000000a0 CR3: 00000004293d2003 CR4: 0000000000360ee0
[   14.093307] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[   14.093308] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[   14.093310] Call Trace:
[   14.093324]  simple_recursive_removal+0x4e/0x2e0
[   14.093330]  ? debugfs_remove+0x60/0x60
[   14.093334]  debugfs_remove+0x40/0x60
[   14.093339]  blk_trace_free+0x20/0x70
[   14.093346]  __blk_trace_remove+0x54/0x90
[   14.096704] === do_blk_trace_setup(6) start
[   14.098534]  blk_trace_shutdown+0x74/0x80
[   14.100958] === do_blk_trace_setup(6) creating directory
[   14.104575]  __blk_release_queue+0xbe/0x160
[   14.104580]  process_one_work+0x1b4/0x380
[   14.104585]  worker_thread+0x50/0x3c0
[   14.104589]  kthread+0xf9/0x130
[   14.104593]  ? process_one_work+0x380/0x380
[   14.104596]  ? kthread_park+0x90/0x90
[   14.104599]  ret_from_fork+0x1f/0x40
[   14.104603] Modules linked in: loop(E) xfs(E) libcrc32c(E)
crct10dif_pclmul(E) crc32_pclmul(E) ghash_clmulni_intel(E) joydev(E)
serio_raw(E) aesni_intel(E) glue_helper(E) virtio_balloon(E) evdev(E)
crypto_simd(E) pcspkr(E) cryptd(E) i6300esb(E) button(E) ip_tables(E)
x_tables(E) autofs4(E) ext4(E) crc32c_generic(E) crc16(E) mbcache(E)
jbd2(E) virtio_net(E) net_failover(E) failover(E) virtio_blk(E)
ata_generic(E) uhci_hcd(E) ata_piix(E) ehci_hcd(E) nvme(E) libata(E)
crc32c_intel(E) usbcore(E) psmouse(E) nvme_core(E) virtio_pci(E)
scsi_mod(E) virtio_ring(E) t10_pi(E) virtio(E) i2c_piix4(E) floppy(E)
[   14.107400] === do_blk_trace_setup(6) using what debugfs_lookup() gave
[   14.108939] CR2: 00000000000000a0
[   14.110589] === do_blk_trace_setup(6) end with ret: 0
[   14.111592] ---[ end trace 7a783b33b9614db9 ]---

The root cause to this issue is that debugfs_lookup() can find a
previous incarnation's dir of the same name which is about to get
removed from a not yet schedule work. When that happens, the the files
are taken underneath the nose of the blktrace, and when it comes time to
cleanup, these dentries are long gone because of a scheduled removal.

This issue is happening because of two reasons:

  1) The request_queue is currently removed asynchronously as of commit
     dc9edc44de6c ("block: Fix a blk_exit_rl() regression") merged on
     v4.12, this allows races with userspace which were not possible
     before unless as removal of a block device used to happen
     synchronously with its request_queue. One could however still
     parallelize blksetup calls while one loops on device addition and
     removal.

  2) There are no errors checks when we create the debugfs directory,
     be it on init or for blktrace. The concept of when the directory
     *should* really exist is further complicated by the fact that
     we have asynchronous request_queue removal. And, we have no
     real sanity checks to ensure we don't re-create the queue debugfs
     directory.

We can fix the UAF by using a debugfs directory which moving forward
will always be accessible if debugfs is enabled, this way, its allocated
and avaialble always for both request-based block drivers or
make_request drivers (multiqueue) block drivers.

We also put sanity checks in place to ensure that if the directory is
found with debugfs_lookup() it is the dentry we expect. When doing a
blktrace against a parition, we will always be creating a temporary
debugfs directory, so ensure that only exists once as well to avoid
issues against concurrent blktrace runs.

Lastly, since we are now always creating the needed request_queue
debugfs directory upon init, we can also take the initiative to
proactively check against errors. We currently do not check for
errors on add_disk() and friends, but we shouldn't make the issue
any worse.

This also simplifies the code considerably, with the only penalty now
being that we're always creating the request queue debugfs directory for
the request-based block device drivers.

The UAF then is not a core debugfs issue, but instead a complex misuse
of debugfs, and this issue can only be triggered if you are root.

This issue can be reproduced with break-blktrace [2] using:

  break-blktrace -c 10 -d -s

This patch fixes this issue. Note that there is also another
respective UAF but from the ioctl path [3], this should also fix
that issue.

This patch then also disputes the severity of CVE-2019-19770 as
this issue is only possible by being root and using blktrace.

It is not a core debugfs issue.

[0] https://bugzilla.kernel.org/show_bug.cgi?id=205713
[1] https://nvd.nist.gov/vuln/detail/CVE-2019-19770
[2] https://github.com/mcgrof/break-blktrace
[3] https://lore.kernel.org/lkml/000000000000ec635b059f752700@google.com/

Cc: Bart Van Assche <bvanassche@acm.org>
Cc: Omar Sandoval <osandov@fb.com>
Cc: Hannes Reinecke <hare@suse.com>
Cc: Nicolai Stange <nstange@suse.de>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: yu kuai <yukuai3@huawei.com>
Reported-by: syzbot+603294af2d01acfdd6da@syzkaller.appspotmail.com
Fixes: 6ac93117ab00 ("blktrace: use existing disk debugfs directory")
Signed-off-by: Luis Chamberlain <mcgrof@kernel.org>
---
 block/blk-debugfs.c          | 30 +++++++++++
 block/blk-mq-debugfs.c       |  5 --
 block/blk-sysfs.c            |  9 ++++
 block/blk.h                  | 11 ++++
 include/linux/blkdev.h       |  5 +-
 include/linux/blktrace_api.h |  1 +
 kernel/trace/blktrace.c      | 98 +++++++++++++++++++++++++++++++++---
 7 files changed, 146 insertions(+), 13 deletions(-)

Comments

Bart Van Assche April 19, 2020, 9:55 p.m. UTC | #1
On 4/19/20 12:45 PM, Luis Chamberlain wrote:
> +int __must_check blk_queue_debugfs_register(struct request_queue *q)
> +{
> +	struct dentry *dir = NULL;
> +
> +	/* This can happen if we have a bug in the lower layers */

What does "this" refer to? Which layers does "lower layers" refer to? 
Most software developers consider a module that calls directly into 
another module as a higher layer (callbacks through function pointers do 
not count; see also https://en.wikipedia.org/wiki/Modular_programming). 
According to that definition block drivers are a software layer 
immediately above the block layer core.

How about changing that comment into the following to make it 
unambiguous (if this is what you meant)?

	/*
	 * Check whether the debugfs directory already exists. This can
	 * only happen as the result of a bug in a block driver.
	 */

> +	dir = debugfs_lookup(kobject_name(q->kobj.parent), blk_debugfs_root);
> +	if (dir) {
> +		pr_warn("%s: registering request_queue debugfs directory twice is not allowed\n",
> +			kobject_name(q->kobj.parent));
> +		dput(dir);
> +		return -EALREADY;
> +	}
> +
> +	q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
> +					    blk_debugfs_root);
> +	if (!q->debugfs_dir)
> +		return -ENOMEM;
> +
> +	return 0;
> +}

kobject_name(q->kobj.parent) is used three times in the above function. 
How about introducing a local variable that holds the result of that 
expression?

> +static bool blk_trace_target_disk(const char *target, const char *diskname)
> +{
> +	if (strlen(target) != strlen(diskname))
> +		return false;
> +
> +	if (!strncmp(target, diskname,
> +		     min_t(size_t, strlen(target), strlen(diskname))))
> +		return true;
> +
> +	return false;
> +}

The above code looks weird to me. When the second if-statement is 
reached, it is guaranteed that 'target' and 'diskname' have the same 
length. So why to calculate the minimum length in the second 
if-statement of two strings that have the same length?

Independent of what the purpose of the above code is, can that code be 
rewritten such that it does not depend on the details of how names are 
assigned to disks and partitions? Would disk_get_part() be useful here?

Thanks,

Bart.
Luis Chamberlain April 20, 2020, 12:04 a.m. UTC | #2
On Sun, Apr 19, 2020 at 02:55:42PM -0700, Bart Van Assche wrote:
> On 4/19/20 12:45 PM, Luis Chamberlain wrote:
> > +int __must_check blk_queue_debugfs_register(struct request_queue *q)
> > +{
> > +	struct dentry *dir = NULL;
> > +
> > +	/* This can happen if we have a bug in the lower layers */
> 
> What does "this" refer to? Which layers does "lower layers" refer to? Most
> software developers consider a module that calls directly into another
> module as a higher layer (callbacks through function pointers do not count;
> see also https://en.wikipedia.org/wiki/Modular_programming). According to
> that definition block drivers are a software layer immediately above the
> block layer core.
> 
> How about changing that comment into the following to make it unambiguous
> (if this is what you meant)?
> 
> 	/*
> 	 * Check whether the debugfs directory already exists. This can
> 	 * only happen as the result of a bug in a block driver.
> 	 */

But I didn't mean on a block driver. I meant the block core. In our
case, the async request_queue removal is an example. There is nothing
that block drivers could have done to help much with that.

I could just change "lower layers" to "block layer" ?

> > +	dir = debugfs_lookup(kobject_name(q->kobj.parent), blk_debugfs_root);
> > +	if (dir) {
> > +		pr_warn("%s: registering request_queue debugfs directory twice is not allowed\n",
> > +			kobject_name(q->kobj.parent));
> > +		dput(dir);
> > +		return -EALREADY;
> > +	}
> > +
> > +	q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
> > +					    blk_debugfs_root);
> > +	if (!q->debugfs_dir)
> > +		return -ENOMEM;
> > +
> > +	return 0;
> > +}
> 
> kobject_name(q->kobj.parent) is used three times in the above function. How
> about introducing a local variable that holds the result of that expression?

Sure.

> > +static bool blk_trace_target_disk(const char *target, const char *diskname)
> > +{
> > +	if (strlen(target) != strlen(diskname))
> > +		return false;
> > +
> > +	if (!strncmp(target, diskname,
> > +		     min_t(size_t, strlen(target), strlen(diskname))))
> > +		return true;
> > +
> > +	return false;
> > +}
> 
> The above code looks weird to me. When the second if-statement is reached,
> it is guaranteed that 'target' and 'diskname' have the same length. So why
> to calculate the minimum length in the second if-statement of two strings
> that have the same length?

True, no need that that point. Will fix.

> Independent of what the purpose of the above code is, can that code be
> rewritten such that it does not depend on the details of how names are
> assigned to disks and partitions? Would disk_get_part() be useful here?

I did try, but couldn't figure out a way. I'll keep looking but likewise
let me know if you find a way.

  Luis
Bart Van Assche April 20, 2020, 12:38 a.m. UTC | #3
On 4/19/20 5:04 PM, Luis Chamberlain wrote:
> On Sun, Apr 19, 2020 at 02:55:42PM -0700, Bart Van Assche wrote:
>> On 4/19/20 12:45 PM, Luis Chamberlain wrote:
>>> +int __must_check blk_queue_debugfs_register(struct request_queue *q)
>>> +{
>>> +	struct dentry *dir = NULL;
>>> +
>>> +	/* This can happen if we have a bug in the lower layers */
>>
>> What does "this" refer to? Which layers does "lower layers" refer to? Most
>> software developers consider a module that calls directly into another
>> module as a higher layer (callbacks through function pointers do not count;
>> see also https://en.wikipedia.org/wiki/Modular_programming). According to
>> that definition block drivers are a software layer immediately above the
>> block layer core.
>>
>> How about changing that comment into the following to make it unambiguous
>> (if this is what you meant)?
>>
>> 	/*
>> 	 * Check whether the debugfs directory already exists. This can
>> 	 * only happen as the result of a bug in a block driver.
>> 	 */
> 
> But I didn't mean on a block driver. I meant the block core. In our
> case, the async request_queue removal is an example. There is nothing
> that block drivers could have done to help much with that.
> 
> I could just change "lower layers" to "block layer" ?

That sounds good to me.

>> Independent of what the purpose of the above code is, can that code be
>> rewritten such that it does not depend on the details of how names are
>> assigned to disks and partitions? Would disk_get_part() be useful here?
> 
> I did try, but couldn't figure out a way. I'll keep looking but likewise
> let me know if you find a way.

How about making blk_trace_setup() pass the result of the following 
expression as an additional argument to blk_trace_setup():

	bdev != bdev->bd_contains

I think that is a widely used approach to verify after a block device 
has been opened whether or not 'bdev' refers to a partition (bdev != 
bdev->bd_contains means that 'bdev' represents a partition).

Thanks,

Bart.
Luis Chamberlain April 20, 2020, 6:46 p.m. UTC | #4
On Sun, Apr 19, 2020 at 05:38:59PM -0700, Bart Van Assche wrote:
> On 4/19/20 5:04 PM, Luis Chamberlain wrote:
> > > Independent of what the purpose of the above code is, can that code be
> > > rewritten such that it does not depend on the details of how names are
> > > assigned to disks and partitions? Would disk_get_part() be useful here?
> > 
> > I did try, but couldn't figure out a way. I'll keep looking but likewise
> > let me know if you find a way.
> 
> How about making blk_trace_setup() pass the result of the following
> expression as an additional argument to blk_trace_setup():
> 
> 	bdev != bdev->bd_contains
> 
> I think that is a widely used approach to verify after a block device has
> been opened whether or not 'bdev' refers to a partition (bdev !=
> bdev->bd_contains means that 'bdev' represents a partition).

Sweet, that should simplify this considerbly, thanks!

  Luis
Greg Kroah-Hartman April 20, 2020, 8:16 p.m. UTC | #5
On Sun, Apr 19, 2020 at 07:45:22PM +0000, Luis Chamberlain wrote:
> On commit 6ac93117ab00 ("blktrace: use existing disk debugfs directory")
> merged on v4.12 Omar fixed the original blktrace code for request-based
> drivers (multiqueue). This however left in place a possible crash, if you
> happen to abuse blktrace in a way it was not intended, and even more so
> with our current asynchronous request_queue removal.
> 
> Namely, if you loop adding a device, setup the blktrace with BLKTRACESETUP,
> forget to BLKTRACETEARDOWN, and then just remove the device you end up
> with a panic:
> 
> [  107.193134] debugfs: Directory 'loop0' with parent 'block' already present!
> [  107.254615] BUG: kernel NULL pointer dereference, address: 00000000000000a0
> [  107.258785] #PF: supervisor write access in kernel mode
> [  107.262035] #PF: error_code(0x0002) - not-present page
> [  107.264106] PGD 0 P4D 0
> [  107.264404] Oops: 0002 [#1] SMP NOPTI
> [  107.264803] CPU: 8 PID: 674 Comm: kworker/8:2 Tainted: G            E 5.6.0-rc7-next-20200327 #1
> [  107.265712] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.13.0-1 04/01/2014
> [  107.266553] Workqueue: events __blk_release_queue
> [  107.267051] RIP: 0010:down_write+0x15/0x40
> [  107.267488] Code: eb ca e8 ee a5 8d ff cc cc cc cc cc cc cc cc cc cc cc cc cc cc 0f 1f 44 00 00 55 48 89 fd e8 52 db ff ff 31 c0 ba 01 00 00 00 <f0> 48 0f b1 55 00 75 0f  65 48 8b 04 25 c0 8b 01 00 48 89 45 08 5d
> [  107.269300] RSP: 0018:ffff9927c06efda8 EFLAGS: 00010246
> [  107.269841] RAX: 0000000000000000 RBX: ffff8be7e73b0600 RCX: ffffff8100000000
> [  107.270559] RDX: 0000000000000001 RSI: ffffff8100000000 RDI: 00000000000000a0
> [  107.271281] RBP: 00000000000000a0 R08: ffff8be7ebc80fa8 R09: ffff8be7ebc80fa8
> [  107.272001] R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000000
> [  107.272722] R13: ffff8be7efc30400 R14: ffff8be7e0571200 R15: 00000000000000a0
> [  107.273475] FS:  0000000000000000(0000) GS:ffff8be7efc00000(0000) knlGS:0000000000000000
> [  107.274346] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [  107.274968] CR2: 00000000000000a0 CR3: 000000042abee003 CR4: 0000000000360ee0
> [  107.275710] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> [  107.276465] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
> [  107.277214] Call Trace:
> [  107.277532]  simple_recursive_removal+0x4e/0x2e0
> [  107.278049]  ? debugfs_remove+0x60/0x60
> [  107.278493]  debugfs_remove+0x40/0x60
> [  107.278922]  blk_trace_free+0xd/0x50
> [  107.279339]  __blk_trace_remove+0x27/0x40
> [  107.279797]  blk_trace_shutdown+0x30/0x40
> [  107.280256]  __blk_release_queue+0xab/0x110
> [  107.280734]  process_one_work+0x1b4/0x380
> [  107.281194]  worker_thread+0x50/0x3c0
> [  107.281622]  kthread+0xf9/0x130
> [  107.281994]  ? process_one_work+0x380/0x380
> [  107.282467]  ? kthread_park+0x90/0x90
> [  107.282895]  ret_from_fork+0x1f/0x40
> [  107.283316] Modules linked in: loop(E) <etc>
> [  107.288562] CR2: 00000000000000a0
> [  107.288957] ---[ end trace b885d243d441bbce ]---
> 
> This splat happens to be very similar to the one reported via
> kernel.org korg#205713, only that korg#205713 was for v4.19.83
> and the above now includes the simple_recursive_removal() introduced
> via commit a3d1e7eb5abe ("simple_recursive_removal(): kernel-side rm
> -rf for ramfs-style filesystems") merged on v5.6.
> 
> korg#205713 then was used to create CVE-2019-19770 and claims that
> the bug is in a use-after-free in the debugfs core code. The
> implications of this being a generic UAF on debugfs would be
> much more severe, as it would imply parent dentries can sometimes
> not be positive, which we hold by design is just not possible.
> 
> Below is the splat explained with a bit more details, explaining
> what is happening in userspace, kernel, and a print of the CPU on,
> which the code runs on:
> 
> load loopback module
> [   13.603371] == blk_mq_debugfs_register(12) start
> [   13.604040] == blk_mq_debugfs_register(12) q->debugfs_dir created
> [   13.604934] == blk_mq_debugfs_register(12) end
> [   13.627382] == blk_mq_debugfs_register(12) start
> [   13.628041] == blk_mq_debugfs_register(12) q->debugfs_dir created
> [   13.629240] == blk_mq_debugfs_register(12) end
> [   13.651667] == blk_mq_debugfs_register(12) start
> [   13.652836] == blk_mq_debugfs_register(12) q->debugfs_dir created
> [   13.655107] == blk_mq_debugfs_register(12) end
> [   13.684917] == blk_mq_debugfs_register(12) start
> [   13.687876] == blk_mq_debugfs_register(12) q->debugfs_dir created
> [   13.691588] == blk_mq_debugfs_register(13) end
> [   13.707320] == blk_mq_debugfs_register(13) start
> [   13.707863] == blk_mq_debugfs_register(13) q->debugfs_dir created
> [   13.708856] == blk_mq_debugfs_register(13) end
> [   13.735623] == blk_mq_debugfs_register(13) start
> [   13.736656] == blk_mq_debugfs_register(13) q->debugfs_dir created
> [   13.738411] == blk_mq_debugfs_register(13) end
> [   13.763326] == blk_mq_debugfs_register(13) start
> [   13.763972] == blk_mq_debugfs_register(13) q->debugfs_dir created
> [   13.765167] == blk_mq_debugfs_register(13) end
> [   13.779510] == blk_mq_debugfs_register(13) start
> [   13.780522] == blk_mq_debugfs_register(13) q->debugfs_dir created
> [   13.782338] == blk_mq_debugfs_register(13) end
> [   13.783521] loop: module loaded
> 
> LOOP_CTL_DEL(loop0) #1
> [   13.803550] = __blk_release_queue(4) start
> [   13.807772] == blk_trace_shutdown(4) start
> [   13.810749] == blk_trace_shutdown(4) end
> [   13.813437] = __blk_release_queue(4) calling blk_mq_debugfs_unregister()
> [   13.817593] ==== blk_mq_debugfs_unregister(4) begin
> [   13.817621] ==== blk_mq_debugfs_unregister(4) debugfs_remove_recursive(q->debugfs_dir)
> [   13.821203] ==== blk_mq_debugfs_unregister(4) end q->debugfs_dir is NULL
> [   13.826166] = __blk_release_queue(4) blk_mq_debugfs_unregister() end
> [   13.832992] = __blk_release_queue(4) end
> 
> LOOP_CTL_ADD(loop0) #1
> [   13.843742] == blk_mq_debugfs_register(7) start
> [   13.845569] == blk_mq_debugfs_register(7) q->debugfs_dir created
> [   13.848628] == blk_mq_debugfs_register(7) end
> 
> BLKTRACE_SETUP(loop0) #1
> [   13.850924] == blk_trace_ioctl(7, BLKTRACESETUP) start
> [   13.852852] === do_blk_trace_setup(7) start
> [   13.854580] === do_blk_trace_setup(7) creating directory
> [   13.856620] === do_blk_trace_setup(7) using what debugfs_lookup() gave
> [   13.860635] === do_blk_trace_setup(7) end with ret: 0
> [   13.862615] == blk_trace_ioctl(7, BLKTRACESETUP) end
> 
> LOOP_CTL_DEL(loop0) #2
> [   13.883304] = __blk_release_queue(7) start
> [   13.885324] == blk_trace_shutdown(7) start
> [   13.887197] == blk_trace_shutdown(7) calling __blk_trace_remove()
> [   13.889807] == __blk_trace_remove(7) start
> [   13.891669] === blk_trace_cleanup(7) start
> [   13.911656] ====== blk_trace_free(7) start
> 
> This is a scheduled __blk_release_queue(). Note that at this point
> we know that the old device is gone as the gendisk removal is
> synchronous.
> 
> LOOP_CTL_ADD(loop0) #2
> [   13.912709] == blk_mq_debugfs_register(2) start
> 
> This is a *new* device, as we know the last one was removed synchronously.
> 
> ---> From LOOP_CTL_DEL(loop0) #2
> [   13.915887] ====== blk_trace_free(7) end
> 
> ---> From LOOP_CTL_ADD(loop0) #2
> [   13.918359] debugfs: Directory 'loop0' with parent 'block' already present!
> 
> Above we have a case where the upon LOOP_CTL_ADD(loop) #2 the
> debugfs_create_dir() start_creating() call will have found an existing dentry,
> and return ERR_PTR(-EEXIST), and so the q->debugfs_dir will be initialized to
> NULL. This clash happens because an asynchronous __blk_release_queue() has
> been scheduled from the above LOOP_CTL_DEL(loop0) #2 call, and it has not yet
> removed the old loop0 debugfs directory.
> 
> [   13.926433] == blk_mq_debugfs_register(2) q->debugfs_dir created
> 
> This is a huge assumption, since we never check for errors on
> blk_mq_debugfs_register(), and so the child files we need to
> create *can* simply be created using NULL as as the parent, in
> which case debugfs would use debugfs_mount->mnt_root as the parent.
> 
> [   13.930373] == blk_mq_debugfs_register(2) end
> 
> BLKTRACE_SETUP(loop0) #2
> [   13.933961] == blk_trace_ioctl(2, BLKTRACESETUP) start
> [   13.936758] === do_blk_trace_setup(2) start
> [   13.938944] === do_blk_trace_setup(2) creating directory
> [   13.941029] === do_blk_trace_setup(2) using what debugfs_lookup() gave
> 
> Here debugfs_lookup() will have found the old debugfs directory, even
> though q->debugfs_dir was empty when initializing this device. So the
> files created can actually be created where they should have.
> 
> ---> From LOOP_CTL_DEL(loop0) #2
> [   13.971046] === blk_trace_cleanup(7) end
> [   13.973175] == __blk_trace_remove(7) end
> [   13.975352] == blk_trace_shutdown(7) end
> [   13.977415] = __blk_release_queue(7) calling blk_mq_debugfs_unregister()
> [   13.980645] ==== blk_mq_debugfs_unregister(7) begin
> [   13.980696] ==== blk_mq_debugfs_unregister(7) debugfs_remove_recursive(q->debugfs_dir)
> 
> This is a continuation from the old scheduled  __blk_release_queue(). It
> is using another request_queue structure than what the currently activated
> blktrace setup is using.
> 
> Here now the dentry where the blktrace setup files were created on top
> of is gone.  In fact, since debugfs_remove_recursive() is used, *all* of
> those blktrace debugfs files should have been removed as well, including
> the relay. This is despite the fact that the removal and blktrace setup
> are working on two different request_queue structures.
> 
> [   13.983118] ==== blk_mq_debugfs_unregister(7) end q->debugfs_dir is NULL
> [   13.986945] = __blk_release_queue(7) blk_mq_debugfs_unregister() end
> [   13.993155] = __blk_release_queue(7) end
> 
> The scheduled __blk_release_queue completes finally.
> 
> ---> From BLKTRACE_SETUP(loop0) #2
> [   13.995928] === do_blk_trace_setup(2) end with ret: 0
> [   13.997623] == blk_trace_ioctl(2, BLKTRACESETUP) end
> 
> At this point the old blktrace *setup* is complete though, and let us
> recall that the old dentry is still used. But recall that we just had
> our old device removed, and since debugfs_remove_recursive() was used,
> so are all of its files. This request_queue however is still active.
> 
> LOOP_CTL_DEL(loop0) #3
> [   14.035119] = __blk_release_queue(2) start
> [   14.036925] == blk_trace_shutdown(2) start
> [   14.038518] == blk_trace_shutdown(2) calling __blk_trace_remove()
> [   14.040829] == __blk_trace_remove(2) start
> [   14.042413] === blk_trace_cleanup(2) start
> 
> This removal will cleanup the blktrace setup. Note the trap though,
> the blktrace setup will already have been cleaned up since the
> scheduled removal from instance #2 removed the files on the same
> debugfs path.
> 
> The __blk_release_queue() is scheduled and is asynchronous so can
> continue later, and that is where the crash happens, on CPU 2.
> 
> LOOP_CTL_ADD(loop0) #3
> [   14.072522] == blk_mq_debugfs_register(6) start
> 
> ---> From LOOP_CTL_DEL(loop0) #3
> [   14.075151] ====== blk_trace_free(2) start
> 
> ---> From LOOP_CTL_ADD(loop0) #3
> [   14.075882] == blk_mq_debugfs_register(6) q->debugfs_dir created
> 
> ---> From LOOP_CTL_DEL(loop0) #3
> The panic happens on CPU 2 when we continue with the old scheduled
> removal of the request_queue, which will run into a blktrace setup
> with NULL pointers.
> 
> [   14.078624] BUG: kernel NULL pointer dereference, address: 00000000000000a0
> [   14.084332] == blk_mq_debugfs_register(6) end
> [   14.086971] #PF: supervisor write access in kernel mode
> [   14.086974] #PF: error_code(0x0002) - not-present page
> [   14.086977] PGD 0 P4D 0
> [   14.086984] Oops: 0002 [#1] SMP NOPTI
> [   14.086990] CPU: 2 PID: 287 Comm: kworker/2:2 Tainted: G            E 5.6.0-next-20200403+ #54
> [   14.086991] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.13.0-1 04/01/2014
> [   14.087002] Workqueue: events __blk_release_queue
> [   14.087011] RIP: 0010:down_write+0x15/0x40
> [   14.090300] == blk_trace_ioctl(6, BLKTRACESETUP) start
> [   14.093277] Code: eb ca e8 3e 34 8d ff cc cc cc cc cc cc cc cc cc cc
> cc cc cc cc 0f 1f 44 00 00 55 48 89 fd e8 52 db ff ff 31 c0 ba 01 00 00
> 00 <f0> 48 0f b1 55 00 75 0f 65 48 8b 04 25 c0 8b 01 00 48 89 45 08 5d
> [   14.093280] RSP: 0018:ffffc28a00533da8 EFLAGS: 00010246
> [   14.093284] RAX: 0000000000000000 RBX: ffff9f7a24d07980 RCX: ffffff8100000000
> [   14.093286] RDX: 0000000000000001 RSI: ffffff8100000000 RDI: 00000000000000a0
> [   14.093287] RBP: 00000000000000a0 R08: 0000000000000000 R09: 0000000000000019
> [   14.093289] R10: 0000000000000774 R11: 0000000000000000 R12: 0000000000000000
> [   14.093291] R13: ffff9f7a2fab0400 R14: ffff9f7a21dd1140 R15: 00000000000000a0
> [   14.093294] FS:  0000000000000000(0000) GS:ffff9f7a2fa80000(0000) knlGS:0000000000000000
> [   14.093296] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [   14.093298] CR2: 00000000000000a0 CR3: 00000004293d2003 CR4: 0000000000360ee0
> [   14.093307] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> [   14.093308] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
> [   14.093310] Call Trace:
> [   14.093324]  simple_recursive_removal+0x4e/0x2e0
> [   14.093330]  ? debugfs_remove+0x60/0x60
> [   14.093334]  debugfs_remove+0x40/0x60
> [   14.093339]  blk_trace_free+0x20/0x70
> [   14.093346]  __blk_trace_remove+0x54/0x90
> [   14.096704] === do_blk_trace_setup(6) start
> [   14.098534]  blk_trace_shutdown+0x74/0x80
> [   14.100958] === do_blk_trace_setup(6) creating directory
> [   14.104575]  __blk_release_queue+0xbe/0x160
> [   14.104580]  process_one_work+0x1b4/0x380
> [   14.104585]  worker_thread+0x50/0x3c0
> [   14.104589]  kthread+0xf9/0x130
> [   14.104593]  ? process_one_work+0x380/0x380
> [   14.104596]  ? kthread_park+0x90/0x90
> [   14.104599]  ret_from_fork+0x1f/0x40
> [   14.104603] Modules linked in: loop(E) xfs(E) libcrc32c(E)
> crct10dif_pclmul(E) crc32_pclmul(E) ghash_clmulni_intel(E) joydev(E)
> serio_raw(E) aesni_intel(E) glue_helper(E) virtio_balloon(E) evdev(E)
> crypto_simd(E) pcspkr(E) cryptd(E) i6300esb(E) button(E) ip_tables(E)
> x_tables(E) autofs4(E) ext4(E) crc32c_generic(E) crc16(E) mbcache(E)
> jbd2(E) virtio_net(E) net_failover(E) failover(E) virtio_blk(E)
> ata_generic(E) uhci_hcd(E) ata_piix(E) ehci_hcd(E) nvme(E) libata(E)
> crc32c_intel(E) usbcore(E) psmouse(E) nvme_core(E) virtio_pci(E)
> scsi_mod(E) virtio_ring(E) t10_pi(E) virtio(E) i2c_piix4(E) floppy(E)
> [   14.107400] === do_blk_trace_setup(6) using what debugfs_lookup() gave
> [   14.108939] CR2: 00000000000000a0
> [   14.110589] === do_blk_trace_setup(6) end with ret: 0
> [   14.111592] ---[ end trace 7a783b33b9614db9 ]---
> 
> The root cause to this issue is that debugfs_lookup() can find a
> previous incarnation's dir of the same name which is about to get
> removed from a not yet schedule work. When that happens, the the files
> are taken underneath the nose of the blktrace, and when it comes time to
> cleanup, these dentries are long gone because of a scheduled removal.
> 
> This issue is happening because of two reasons:
> 
>   1) The request_queue is currently removed asynchronously as of commit
>      dc9edc44de6c ("block: Fix a blk_exit_rl() regression") merged on
>      v4.12, this allows races with userspace which were not possible
>      before unless as removal of a block device used to happen
>      synchronously with its request_queue. One could however still
>      parallelize blksetup calls while one loops on device addition and
>      removal.
> 
>   2) There are no errors checks when we create the debugfs directory,
>      be it on init or for blktrace. The concept of when the directory
>      *should* really exist is further complicated by the fact that
>      we have asynchronous request_queue removal. And, we have no
>      real sanity checks to ensure we don't re-create the queue debugfs
>      directory.
> 
> We can fix the UAF by using a debugfs directory which moving forward
> will always be accessible if debugfs is enabled, this way, its allocated
> and avaialble always for both request-based block drivers or
> make_request drivers (multiqueue) block drivers.
> 
> We also put sanity checks in place to ensure that if the directory is
> found with debugfs_lookup() it is the dentry we expect. When doing a
> blktrace against a parition, we will always be creating a temporary
> debugfs directory, so ensure that only exists once as well to avoid
> issues against concurrent blktrace runs.
> 
> Lastly, since we are now always creating the needed request_queue
> debugfs directory upon init, we can also take the initiative to
> proactively check against errors. We currently do not check for
> errors on add_disk() and friends, but we shouldn't make the issue
> any worse.
> 
> This also simplifies the code considerably, with the only penalty now
> being that we're always creating the request queue debugfs directory for
> the request-based block device drivers.
> 
> The UAF then is not a core debugfs issue, but instead a complex misuse
> of debugfs, and this issue can only be triggered if you are root.
> 
> This issue can be reproduced with break-blktrace [2] using:
> 
>   break-blktrace -c 10 -d -s
> 
> This patch fixes this issue. Note that there is also another
> respective UAF but from the ioctl path [3], this should also fix
> that issue.
> 
> This patch then also disputes the severity of CVE-2019-19770 as
> this issue is only possible by being root and using blktrace.
> 
> It is not a core debugfs issue.
> 
> [0] https://bugzilla.kernel.org/show_bug.cgi?id=205713
> [1] https://nvd.nist.gov/vuln/detail/CVE-2019-19770
> [2] https://github.com/mcgrof/break-blktrace
> [3] https://lore.kernel.org/lkml/000000000000ec635b059f752700@google.com/
> 
> Cc: Bart Van Assche <bvanassche@acm.org>
> Cc: Omar Sandoval <osandov@fb.com>
> Cc: Hannes Reinecke <hare@suse.com>
> Cc: Nicolai Stange <nstange@suse.de>
> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> Cc: Michal Hocko <mhocko@kernel.org>
> Cc: yu kuai <yukuai3@huawei.com>
> Reported-by: syzbot+603294af2d01acfdd6da@syzkaller.appspotmail.com
> Fixes: 6ac93117ab00 ("blktrace: use existing disk debugfs directory")
> Signed-off-by: Luis Chamberlain <mcgrof@kernel.org>
> ---
>  block/blk-debugfs.c          | 30 +++++++++++
>  block/blk-mq-debugfs.c       |  5 --
>  block/blk-sysfs.c            |  9 ++++
>  block/blk.h                  | 11 ++++
>  include/linux/blkdev.h       |  5 +-
>  include/linux/blktrace_api.h |  1 +
>  kernel/trace/blktrace.c      | 98 +++++++++++++++++++++++++++++++++---
>  7 files changed, 146 insertions(+), 13 deletions(-)

This patch triggered gmail's spam detection, your changelog text is
whack...

> diff --git a/block/blk-debugfs.c b/block/blk-debugfs.c
> index 19091e1effc0..d84038bce0a5 100644
> --- a/block/blk-debugfs.c
> +++ b/block/blk-debugfs.c
> @@ -3,6 +3,9 @@
>  /*
>   * Shared request-based / make_request-based functionality
>   */
> +
> +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
> +
>  #include <linux/kernel.h>
>  #include <linux/blkdev.h>
>  #include <linux/debugfs.h>
> @@ -13,3 +16,30 @@ void blk_debugfs_register(void)
>  {
>  	blk_debugfs_root = debugfs_create_dir("block", NULL);
>  }
> +
> +int __must_check blk_queue_debugfs_register(struct request_queue *q)
> +{
> +	struct dentry *dir = NULL;
> +
> +	/* This can happen if we have a bug in the lower layers */
> +	dir = debugfs_lookup(kobject_name(q->kobj.parent), blk_debugfs_root);
> +	if (dir) {
> +		pr_warn("%s: registering request_queue debugfs directory twice is not allowed\n",
> +			kobject_name(q->kobj.parent));
> +		dput(dir);
> +		return -EALREADY;
> +	}
> +
> +	q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
> +					    blk_debugfs_root);
> +	if (!q->debugfs_dir)
> +		return -ENOMEM;

Why doesn't the directory just live in the request queue, or somewhere
else, so that you save it when it is created and then that's it.  No
need to "look it up" anywhere else.

Or do you do that in later patches?  I only see this one at the moment,
sorry...

>  static struct dentry *blk_trace_debugfs_dir(struct blk_user_trace_setup *buts,
> +					    struct request_queue *q,
>  					    struct blk_trace *bt)
>  {
>  	struct dentry *dir = NULL;
>  
> +	/* This can only happen if we have a bug on our lower layers */
> +	if (!q->kobj.parent) {
> +		pr_warn("%s: request_queue parent is gone\n", buts->name);

A kobject always has a parent, unless it has not been registered yet, so
I don't know what you are testing could ever happen.



> +		return NULL;
> +	}
> +
> +	/*
> +	 * From a sysfs kobject perspective, the request_queue sits on top of
> +	 * the gendisk, which has the name of the disk. We always create a
> +	 * debugfs directory upon init for this gendisk kobject, so we re-use
> +	 * that if blktrace is going to be done for it.
> +	 */
> +	if (blk_trace_target_disk(buts->name, kobject_name(q->kobj.parent))) {
> +		if (!q->debugfs_dir) {
> +			pr_warn("%s: expected request_queue debugfs_dir is not set\n",
> +				buts->name);

What is userspace supposed to be able to do if they see this warning?

> +			return NULL;
> +		}
> +		/*
> +		 * debugfs_lookup() is used to ensure the directory is not
> +		 * taken from underneath us. We must dput() it later once
> +		 * done with it within blktrace.
> +		 */
> +		dir = debugfs_lookup(buts->name, blk_debugfs_root);
> +		if (!dir) {
> +			pr_warn("%s: expected request_queue debugfs_dir dentry is gone\n",
> +				buts->name);

Again, can't we just save the pointer when we create it and not have to
look it up again?

> +			return NULL;
> +		}
> +		 /*
> +		 * This is a reaffirmation that debugfs_lookup() shall always
> +		 * return the same dentry if it was already set.
> +		 */

I'm all for reaffirmation and the like, but really, is this needed???

> +		if (dir != q->debugfs_dir) {
> +			dput(dir);
> +			pr_warn("%s: expected dentry dir != q->debugfs_dir\n",
> +				buts->name);
> +			return NULL;

Why are you testing to see if debugfs really is working properly?
SHould all users do crazy things like this (hint, rhetorical
question...)

> +		}
> +		bt->backing_dir = q->debugfs_dir;
> +		return bt->backing_dir;
> +	}
> +
> +	/*
> +	 * If not using blktrace on the gendisk, we are going to create a
> +	 * temporary debugfs directory for it, however this cannot be shared
> +	 * between two concurrent blktraces since the path is not unique, so
> +	 * ensure this is only done once.
> +	 */
>  	dir = debugfs_lookup(buts->name, blk_debugfs_root);
> -	if (!dir)
> -		bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
> +	if (dir) {
> +		pr_warn("%s: temporary blktrace debugfs directory already present\n",
> +			buts->name);
> +		dput(dir);
> +		return NULL;
> +	}
> +
> +	bt->dir = debugfs_create_dir(buts->name, blk_debugfs_root);
> +	if (!bt->dir) {
> +		pr_warn("%s: temporary blktrace debugfs directory could not be created\n",
> +			buts->name);

Again, do not test the return value, you do not care.  I've been
removing these checks from everywhere.

thanks,

greg k-h
Luis Chamberlain April 20, 2020, 8:41 p.m. UTC | #6
On Mon, Apr 20, 2020 at 10:16:15PM +0200, Greg KH wrote:
> On Sun, Apr 19, 2020 at 07:45:22PM +0000, Luis Chamberlain wrote:
> 
> This patch triggered gmail's spam detection, your changelog text is
> whack...

Oh? What do you think triggered it?

> > diff --git a/block/blk-debugfs.c b/block/blk-debugfs.c
> > index 19091e1effc0..d84038bce0a5 100644
> > --- a/block/blk-debugfs.c
> > +++ b/block/blk-debugfs.c
> > @@ -3,6 +3,9 @@
> >  /*
> >   * Shared request-based / make_request-based functionality
> >   */
> > +
> > +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
> > +
> >  #include <linux/kernel.h>
> >  #include <linux/blkdev.h>
> >  #include <linux/debugfs.h>
> > @@ -13,3 +16,30 @@ void blk_debugfs_register(void)
> >  {
> >  	blk_debugfs_root = debugfs_create_dir("block", NULL);
> >  }
> > +
> > +int __must_check blk_queue_debugfs_register(struct request_queue *q)
> > +{
> > +	struct dentry *dir = NULL;
> > +
> > +	/* This can happen if we have a bug in the lower layers */
> > +	dir = debugfs_lookup(kobject_name(q->kobj.parent), blk_debugfs_root);
> > +	if (dir) {
> > +		pr_warn("%s: registering request_queue debugfs directory twice is not allowed\n",
> > +			kobject_name(q->kobj.parent));
> > +		dput(dir);
> > +		return -EALREADY;
> > +	}
> > +
> > +	q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
> > +					    blk_debugfs_root);
> > +	if (!q->debugfs_dir)
> > +		return -ENOMEM;
> 
> Why doesn't the directory just live in the request queue, or somewhere
> else, so that you save it when it is created and then that's it.  No
> need to "look it up" anywhere else.

Its already there. And yes, after my changes it is technically possible
to just re-use it directly. But this is complicated by a few things. One
is that at this point in time, asynchronous request_queue removal is
still possible, and so a race was exposed where a requeust_queue may be
lingering but its old device is gone. That race is fixed by reverting us
back to synchronous request_queue removal, therefore ensuring that the
debugfs dir exists so long as the device does.

I can remove the debugfs_lookup() *after* we revert to synchronous
request_queue removal, or we just re-order the patches so that the
revert happens first. That should simplify this patch.

The code in this patch was designed to help dispute the logic behind
the CVE, in particular it shows exactly where debugfs_dir *is* the
one found by debugfs_lookup(), and shows the real issue behind the
removal.

But yeah, now that that is done, I hope its clear to all, and I think
this patch can be simplified if we just revert the async requeust_queue
removal first.

> Or do you do that in later patches?  I only see this one at the moment,
> sorry...
> 
> >  static struct dentry *blk_trace_debugfs_dir(struct blk_user_trace_setup *buts,
> > +					    struct request_queue *q,
> >  					    struct blk_trace *bt)
> >  {
> >  	struct dentry *dir = NULL;
> >  
> > +	/* This can only happen if we have a bug on our lower layers */
> > +	if (!q->kobj.parent) {
> > +		pr_warn("%s: request_queue parent is gone\n", buts->name);
> 
> A kobject always has a parent, unless it has not been registered yet, so
> I don't know what you are testing could ever happen.

Or it has been kobject_del()'d?

A deferred requeust_queue removal shows this is possible, the parent is
taken underneath from us because the refcounting of this kobject is
already kobject_del()'d, and its actual removal scheduled for later.
The parent is taken underneath from us prior to the scheduled removal
completing.

> 
> > +		return NULL;
> > +	}
> > +
> > +	/*
> > +	 * From a sysfs kobject perspective, the request_queue sits on top of
> > +	 * the gendisk, which has the name of the disk. We always create a
> > +	 * debugfs directory upon init for this gendisk kobject, so we re-use
> > +	 * that if blktrace is going to be done for it.
> > +	 */
> > +	if (blk_trace_target_disk(buts->name, kobject_name(q->kobj.parent))) {
> > +		if (!q->debugfs_dir) {
> > +			pr_warn("%s: expected request_queue debugfs_dir is not set\n",
> > +				buts->name);
> 
> What is userspace supposed to be able to do if they see this warning?

Userspace doesn't parse warnings, but the NULL ensures it won't crash
the kernel. The warn informs the kernel of a possible block layer bug.

> > +			return NULL;
> > +		}
> > +		/*
> > +		 * debugfs_lookup() is used to ensure the directory is not
> > +		 * taken from underneath us. We must dput() it later once
> > +		 * done with it within blktrace.
> > +		 */
> > +		dir = debugfs_lookup(buts->name, blk_debugfs_root);
> > +		if (!dir) {
> > +			pr_warn("%s: expected request_queue debugfs_dir dentry is gone\n",
> > +				buts->name);
> 
> Again, can't we just save the pointer when we create it and not have to
> look it up again?

Only if we do the revert of the requeust_queue removal first.

> > +			return NULL;
> > +		}
> > +		 /*
> > +		 * This is a reaffirmation that debugfs_lookup() shall always
> > +		 * return the same dentry if it was already set.
> > +		 */
> 
> I'm all for reaffirmation and the like, but really, is this needed???

To those who were still not sure that the issue was not a debugfs issue
I hoped this to make it clear. But indeed, if we revert back to
synchronous request_queue removal, that should simplify this.

> > +		if (dir != q->debugfs_dir) {
> > +			dput(dir);
> > +			pr_warn("%s: expected dentry dir != q->debugfs_dir\n",
> > +				buts->name);
> > +			return NULL;
> 
> Why are you testing to see if debugfs really is working properly?
> SHould all users do crazy things like this (hint, rhetorical
> question...)

No, this can happen with the race I mentioned above.

> > +		}
> > +		bt->backing_dir = q->debugfs_dir;
> > +		return bt->backing_dir;
> > +	}
> > +
> > +	/*
> > +	 * If not using blktrace on the gendisk, we are going to create a
> > +	 * temporary debugfs directory for it, however this cannot be shared
> > +	 * between two concurrent blktraces since the path is not unique, so
> > +	 * ensure this is only done once.
> > +	 */
> >  	dir = debugfs_lookup(buts->name, blk_debugfs_root);
> > -	if (!dir)
> > -		bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
> > +	if (dir) {
> > +		pr_warn("%s: temporary blktrace debugfs directory already present\n",
> > +			buts->name);
> > +		dput(dir);
> > +		return NULL;
> > +	}
> > +
> > +	bt->dir = debugfs_create_dir(buts->name, blk_debugfs_root);
> > +	if (!bt->dir) {
> > +		pr_warn("%s: temporary blktrace debugfs directory could not be created\n",
> > +			buts->name);
> 
> Again, do not test the return value, you do not care.  I've been
> removing these checks from everywhere.

Sure, the question still stands on what *should* the kernel do if the
blktrace setup failed to create the debugfs files.

  Luis
Greg Kroah-Hartman April 21, 2020, 7 a.m. UTC | #7
On Mon, Apr 20, 2020 at 08:41:56PM +0000, Luis Chamberlain wrote:
> On Mon, Apr 20, 2020 at 10:16:15PM +0200, Greg KH wrote:
> > On Sun, Apr 19, 2020 at 07:45:22PM +0000, Luis Chamberlain wrote:
> > 
> > This patch triggered gmail's spam detection, your changelog text is
> > whack...
> 
> Oh? What do you think triggered it?

No idea.

> 
> > > diff --git a/block/blk-debugfs.c b/block/blk-debugfs.c
> > > index 19091e1effc0..d84038bce0a5 100644
> > > --- a/block/blk-debugfs.c
> > > +++ b/block/blk-debugfs.c
> > > @@ -3,6 +3,9 @@
> > >  /*
> > >   * Shared request-based / make_request-based functionality
> > >   */
> > > +
> > > +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
> > > +
> > >  #include <linux/kernel.h>
> > >  #include <linux/blkdev.h>
> > >  #include <linux/debugfs.h>
> > > @@ -13,3 +16,30 @@ void blk_debugfs_register(void)
> > >  {
> > >  	blk_debugfs_root = debugfs_create_dir("block", NULL);
> > >  }
> > > +
> > > +int __must_check blk_queue_debugfs_register(struct request_queue *q)
> > > +{
> > > +	struct dentry *dir = NULL;
> > > +
> > > +	/* This can happen if we have a bug in the lower layers */
> > > +	dir = debugfs_lookup(kobject_name(q->kobj.parent), blk_debugfs_root);
> > > +	if (dir) {
> > > +		pr_warn("%s: registering request_queue debugfs directory twice is not allowed\n",
> > > +			kobject_name(q->kobj.parent));
> > > +		dput(dir);
> > > +		return -EALREADY;
> > > +	}
> > > +
> > > +	q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
> > > +					    blk_debugfs_root);
> > > +	if (!q->debugfs_dir)
> > > +		return -ENOMEM;
> > 
> > Why doesn't the directory just live in the request queue, or somewhere
> > else, so that you save it when it is created and then that's it.  No
> > need to "look it up" anywhere else.
> 
> Its already there. And yes, after my changes it is technically possible
> to just re-use it directly. But this is complicated by a few things. One
> is that at this point in time, asynchronous request_queue removal is
> still possible, and so a race was exposed where a requeust_queue may be
> lingering but its old device is gone. That race is fixed by reverting us
> back to synchronous request_queue removal, therefore ensuring that the
> debugfs dir exists so long as the device does.
> 
> I can remove the debugfs_lookup() *after* we revert to synchronous
> request_queue removal, or we just re-order the patches so that the
> revert happens first. That should simplify this patch.
> 
> The code in this patch was designed to help dispute the logic behind
> the CVE, in particular it shows exactly where debugfs_dir *is* the
> one found by debugfs_lookup(), and shows the real issue behind the
> removal.
> 
> But yeah, now that that is done, I hope its clear to all, and I think
> this patch can be simplified if we just revert the async requeust_queue
> removal first.

Don't try to "dispute" crazyness, that's not what kernel code is for.
Just do the right thing, and simply saving off the pointer to the
debugfs file when created is the "correct" thing to do, no matter what.
No race conditions or anything else can happen when you do that.

> > Or do you do that in later patches?  I only see this one at the moment,
> > sorry...
> > 
> > >  static struct dentry *blk_trace_debugfs_dir(struct blk_user_trace_setup *buts,
> > > +					    struct request_queue *q,
> > >  					    struct blk_trace *bt)
> > >  {
> > >  	struct dentry *dir = NULL;
> > >  
> > > +	/* This can only happen if we have a bug on our lower layers */
> > > +	if (!q->kobj.parent) {
> > > +		pr_warn("%s: request_queue parent is gone\n", buts->name);
> > 
> > A kobject always has a parent, unless it has not been registered yet, so
> > I don't know what you are testing could ever happen.
> 
> Or it has been kobject_del()'d?

If that happened, how in the world are you in this function anyway, as
the request_queue is an invalid pointer at that point in time???

> A deferred requeust_queue removal shows this is possible, the parent is
> taken underneath from us because the refcounting of this kobject is
> already kobject_del()'d, and its actual removal scheduled for later.
> The parent is taken underneath from us prior to the scheduled removal
> completing.

No, a parent's reference is always valid while the child pointer is
alive.

There is an odd race condition right now that we are working on fixing
if you notice another lkml thread, but that race will soon be fixed.  So
no need for every single user in the kernel to try to test for something
like this (hint, this check is still wrong as with this logic, what
could prevent parent from going away right _after_ you check it???)

Just remove this invalid check please.

> > > +		return NULL;
> > > +	}
> > > +
> > > +	/*
> > > +	 * From a sysfs kobject perspective, the request_queue sits on top of
> > > +	 * the gendisk, which has the name of the disk. We always create a
> > > +	 * debugfs directory upon init for this gendisk kobject, so we re-use
> > > +	 * that if blktrace is going to be done for it.
> > > +	 */
> > > +	if (blk_trace_target_disk(buts->name, kobject_name(q->kobj.parent))) {
> > > +		if (!q->debugfs_dir) {
> > > +			pr_warn("%s: expected request_queue debugfs_dir is not set\n",
> > > +				buts->name);
> > 
> > What is userspace supposed to be able to do if they see this warning?
> 
> Userspace doesn't parse warnings, but the NULL ensures it won't crash
> the kernel. The warn informs the kernel of a possible block layer bug.

Again, the code should not care if the pointer is NULL, as only debugfs
deals with those pointers (and it can handle a NULL pointer just fine.)

> > > +			return NULL;
> > > +		}
> > > +		/*
> > > +		 * debugfs_lookup() is used to ensure the directory is not
> > > +		 * taken from underneath us. We must dput() it later once
> > > +		 * done with it within blktrace.
> > > +		 */
> > > +		dir = debugfs_lookup(buts->name, blk_debugfs_root);
> > > +		if (!dir) {
> > > +			pr_warn("%s: expected request_queue debugfs_dir dentry is gone\n",
> > > +				buts->name);
> > 
> > Again, can't we just save the pointer when we create it and not have to
> > look it up again?
> 
> Only if we do the revert of the requeust_queue removal first.

Your end goal should be no more debugfs_lookup() calls.  Hopefully by
the end of this patch series, that is the result.

> > > +			return NULL;
> > > +		}
> > > +		 /*
> > > +		 * This is a reaffirmation that debugfs_lookup() shall always
> > > +		 * return the same dentry if it was already set.
> > > +		 */
> > 
> > I'm all for reaffirmation and the like, but really, is this needed???
> 
> To those who were still not sure that the issue was not a debugfs issue
> I hoped this to make it clear. But indeed, if we revert back to
> synchronous request_queue removal, that should simplify this.

Again, don't pander to crazies :)

thanks,

greg k-h
Christoph Hellwig April 22, 2020, 7:27 a.m. UTC | #8
On Sun, Apr 19, 2020 at 07:45:22PM +0000, Luis Chamberlain wrote:
> On commit 6ac93117ab00 ("blktrace: use existing disk debugfs directory")
> merged on v4.12 Omar fixed the original blktrace code for request-based
> drivers (multiqueue). This however left in place a possible crash, if you
> happen to abuse blktrace in a way it was not intended, and even more so
> with our current asynchronous request_queue removal.
> 
> Namely, if you loop adding a device, setup the blktrace with BLKTRACESETUP,
> forget to BLKTRACETEARDOWN, and then just remove the device you end up
> with a panic:

FYI, I find all this backtrace garbage not hepful at all.  It requires
me to scroll for so long that I've forgot what was written above by
the time I'm past it.

> This splat happens to be very similar to the one reported via
> kernel.org korg#205713, only that korg#205713 was for v4.19.83
> and the above now includes the simple_recursive_removal() introduced
> via commit a3d1e7eb5abe ("simple_recursive_removal(): kernel-side rm
> -rf for ramfs-style filesystems") merged on v5.6.
> 
> korg#205713 then was used to create CVE-2019-19770 and claims that
> the bug is in a use-after-free in the debugfs core code. The
> implications of this being a generic UAF on debugfs would be
> much more severe, as it would imply parent dentries can sometimes
> not be positive, which we hold by design is just not possible.
> 
> Below is the splat explained with a bit more details, explaining
> what is happening in userspace, kernel, and a print of the CPU on,
> which the code runs on:
> 
> load loopback module
> [   13.603371] == blk_mq_debugfs_register(12) start
> [   13.604040] == blk_mq_debugfs_register(12) q->debugfs_dir created

Same for this..  I think the real valuable changelog only stars below
this 'trace'.

> The root cause to this issue is that debugfs_lookup() can find a
> previous incarnation's dir of the same name which is about to get
> removed from a not yet schedule work. When that happens, the the files
> are taken underneath the nose of the blktrace, and when it comes time to
> cleanup, these dentries are long gone because of a scheduled removal.
> 
> This issue is happening because of two reasons:
> 
>   1) The request_queue is currently removed asynchronously as of commit
>      dc9edc44de6c ("block: Fix a blk_exit_rl() regression") merged on
>      v4.12, this allows races with userspace which were not possible
>      before unless as removal of a block device used to happen
>      synchronously with its request_queue. One could however still
>      parallelize blksetup calls while one loops on device addition and
>      removal.
> 
>   2) There are no errors checks when we create the debugfs directory,
>      be it on init or for blktrace. The concept of when the directory
>      *should* really exist is further complicated by the fact that
>      we have asynchronous request_queue removal. And, we have no
>      real sanity checks to ensure we don't re-create the queue debugfs
>      directory.
> 
> We can fix the UAF by using a debugfs directory which moving forward
> will always be accessible if debugfs is enabled, this way, its allocated
> and avaialble always for both request-based block drivers or
> make_request drivers (multiqueue) block drivers.
> 
> We also put sanity checks in place to ensure that if the directory is
> found with debugfs_lookup() it is the dentry we expect. When doing a
> blktrace against a parition, we will always be creating a temporary
> debugfs directory, so ensure that only exists once as well to avoid
> issues against concurrent blktrace runs.
> 
> Lastly, since we are now always creating the needed request_queue
> debugfs directory upon init, we can also take the initiative to
> proactively check against errors. We currently do not check for
> errors on add_disk() and friends, but we shouldn't make the issue
> any worse.
> 
> This also simplifies the code considerably, with the only penalty now
> being that we're always creating the request queue debugfs directory for
> the request-based block device drivers.
> 
> The UAF then is not a core debugfs issue, but instead a complex misuse
> of debugfs, and this issue can only be triggered if you are root.
> 
> This issue can be reproduced with break-blktrace [2] using:
> 
>   break-blktrace -c 10 -d -s
> 
> This patch fixes this issue. Note that there is also another
> respective UAF but from the ioctl path [3], this should also fix
> that issue.
> 
> This patch then also disputes the severity of CVE-2019-19770 as
> this issue is only possible by being root and using blktrace.
> 
> It is not a core debugfs issue.
> 
> [0] https://bugzilla.kernel.org/show_bug.cgi?id=205713
> [1] https://nvd.nist.gov/vuln/detail/CVE-2019-19770
> [2] https://github.com/mcgrof/break-blktrace
> [3] https://lore.kernel.org/lkml/000000000000ec635b059f752700@google.com/

> +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt

This looks like an unrelated change.

> +
>  #include <linux/kernel.h>
>  #include <linux/blkdev.h>
>  #include <linux/debugfs.h>
> @@ -13,3 +16,30 @@ void blk_debugfs_register(void)
>  {
>  	blk_debugfs_root = debugfs_create_dir("block", NULL);
>  }
> +
> +int __must_check blk_queue_debugfs_register(struct request_queue *q)

__must_check for a function with a single caller looks silly.

> +{
> +	struct dentry *dir = NULL;
> +
> +	/* This can happen if we have a bug in the lower layers */
> +	dir = debugfs_lookup(kobject_name(q->kobj.parent), blk_debugfs_root);
> +	if (dir) {
> +		pr_warn("%s: registering request_queue debugfs directory twice is not allowed\n",
> +			kobject_name(q->kobj.parent));
> +		dput(dir);
> +		return -EALREADY;
> +	}

I don't see why we need this check.  If it is valueable enough we
should have a debugfs_create_dir_exclusive or so that retunrns an error
for an exsting directory, instead of reimplementing it in the caller in
a racy way.  But I'm not really sure we need it to start with.

> +
> +	q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
> +					    blk_debugfs_root);
> +	if (!q->debugfs_dir)
> +		return -ENOMEM;
> +
> +	return 0;
> +}
> +
> +void blk_queue_debugfs_unregister(struct request_queue *q)
> +{
> +	debugfs_remove_recursive(q->debugfs_dir);
> +	q->debugfs_dir = NULL;
> +}

Which to me suggests we can just fold these two into the callers,
with an IS_ENABLED for the creation case given that we check for errors
and the stub will always return an error.

>  	debugfs_create_files(q->debugfs_dir, q, blk_mq_debugfs_queue_attrs);
>  
>  	/*
> @@ -856,9 +853,7 @@ void blk_mq_debugfs_register(struct request_queue *q)
>  
>  void blk_mq_debugfs_unregister(struct request_queue *q)
>  {
> -	debugfs_remove_recursive(q->debugfs_dir);
>  	q->sched_debugfs_dir = NULL;
> -	q->debugfs_dir = NULL;
>  }

This function is weird - the sched dir gets removed by the
debugfs_remove_recursive, so just leaving a function that clears
a pointer is rather odd.  In fact I don't think we need to clear
either sched_debugfs_dir or debugfs_dir anywhere.

>  
> @@ -975,6 +976,14 @@ int blk_register_queue(struct gendisk *disk)
>  		goto unlock;
>  	}
>  
> +	ret = blk_queue_debugfs_register(q);
> +	if (ret) {
> +		blk_trace_remove_sysfs(dev);
> +		kobject_del(&q->kobj);
> +		kobject_put(&dev->kobj);
> +		goto unlock;
> +	}
> +

Please use a goto label to consolidate the common cleanup code.

Also I think these generic debugfs changes probably should be separate
to the blktrace changes.

> +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
> +
>  #include <linux/kernel.h>
>  #include <linux/blkdev.h>
>  #include <linux/blktrace_api.h>
> @@ -311,7 +314,15 @@ static void blk_trace_free(struct blk_trace *bt)
>  	debugfs_remove(bt->msg_file);
>  	debugfs_remove(bt->dropped_file);
>  	relay_close(bt->rchan);
> -	debugfs_remove(bt->dir);
> +	/*
> +	 * backing_dir is set when we use the request_queue debugfs directory.
> +	 * Otherwise we are using a temporary directory created only for the
> +	 * purpose of blktrace.
> +	 */
> +	if (bt->backing_dir)
> +		dput(bt->backing_dir);
> +	else
> +		debugfs_remove(bt->dir);
>  	free_percpu(bt->sequence);
>  	free_percpu(bt->msg_data);
>  	kfree(bt);
> @@ -468,16 +479,89 @@ static void blk_trace_setup_lba(struct blk_trace *bt,
>  	}
>  }
>  
> +static bool blk_trace_target_disk(const char *target, const char *diskname)
> +{
> +	if (strlen(target) != strlen(diskname))
> +		return false;
> +
> +	if (!strncmp(target, diskname,
> +		     min_t(size_t, strlen(target), strlen(diskname))))
> +		return true;
> +
> +	return false;
> +}
> +
>  static struct dentry *blk_trace_debugfs_dir(struct blk_user_trace_setup *buts,
> +					    struct request_queue *q,
>  					    struct blk_trace *bt)
>  {
>  	struct dentry *dir = NULL;
>  
> +	/* This can only happen if we have a bug on our lower layers */
> +	if (!q->kobj.parent) {
> +		pr_warn("%s: request_queue parent is gone\n", buts->name);
> +		return NULL;
> +	}

Why is this not simply a WARN_ON_ONCE()?

> +	/*
> +	 * From a sysfs kobject perspective, the request_queue sits on top of
> +	 * the gendisk, which has the name of the disk. We always create a
> +	 * debugfs directory upon init for this gendisk kobject, so we re-use
> +	 * that if blktrace is going to be done for it.
> +	 */

-EPARSE.

> +	if (blk_trace_target_disk(buts->name, kobject_name(q->kobj.parent))) {
> +		if (!q->debugfs_dir) {
> +			pr_warn("%s: expected request_queue debugfs_dir is not set\n",
> +				buts->name);
> +			return NULL;
> +		}
> +		/*
> +		 * debugfs_lookup() is used to ensure the directory is not
> +		 * taken from underneath us. We must dput() it later once
> +		 * done with it within blktrace.
> +		 */
> +		dir = debugfs_lookup(buts->name, blk_debugfs_root);
> +		if (!dir) {
> +			pr_warn("%s: expected request_queue debugfs_dir dentry is gone\n",
> +				buts->name);
> +			return NULL;
> +		}
> +		 /*
> +		 * This is a reaffirmation that debugfs_lookup() shall always
> +		 * return the same dentry if it was already set.
> +		 */
> +		if (dir != q->debugfs_dir) {
> +			dput(dir);
> +			pr_warn("%s: expected dentry dir != q->debugfs_dir\n",
> +				buts->name);
> +			return NULL;
> +		}
> +		bt->backing_dir = q->debugfs_dir;
> +		return bt->backing_dir;
> +	}

Even with the gigantic commit log I don't get the point of this
code.  It looks rather sketchy and I can't find a rationale for it.
Luis Chamberlain April 22, 2020, 7:28 a.m. UTC | #9
On Tue, Apr 21, 2020 at 09:00:48AM +0200, Greg KH wrote:
> On Mon, Apr 20, 2020 at 08:41:56PM +0000, Luis Chamberlain wrote:
> > On Mon, Apr 20, 2020 at 10:16:15PM +0200, Greg KH wrote:
> > > On Sun, Apr 19, 2020 at 07:45:22PM +0000, Luis Chamberlain wrote:
> > > 
> > > This patch triggered gmail's spam detection, your changelog text is
> > > whack...
> > 
> > Oh? What do you think triggered it?
> 
> No idea.

Alright, well I'm going to move most of the analysis to the bug report
and be as concise as possible on the commit log.

> > > > diff --git a/block/blk-debugfs.c b/block/blk-debugfs.c
> > > > index 19091e1effc0..d84038bce0a5 100644
> > > > --- a/block/blk-debugfs.c
> > > > +++ b/block/blk-debugfs.c
> > > > @@ -3,6 +3,9 @@
> > > >  /*
> > > >   * Shared request-based / make_request-based functionality
> > > >   */
> > > > +
> > > > +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
> > > > +
> > > >  #include <linux/kernel.h>
> > > >  #include <linux/blkdev.h>
> > > >  #include <linux/debugfs.h>
> > > > @@ -13,3 +16,30 @@ void blk_debugfs_register(void)
> > > >  {
> > > >  	blk_debugfs_root = debugfs_create_dir("block", NULL);
> > > >  }
> > > > +
> > > > +int __must_check blk_queue_debugfs_register(struct request_queue *q)
> > > > +{
> > > > +	struct dentry *dir = NULL;
> > > > +
> > > > +	/* This can happen if we have a bug in the lower layers */
> > > > +	dir = debugfs_lookup(kobject_name(q->kobj.parent), blk_debugfs_root);
> > > > +	if (dir) {
> > > > +		pr_warn("%s: registering request_queue debugfs directory twice is not allowed\n",
> > > > +			kobject_name(q->kobj.parent));
> > > > +		dput(dir);
> > > > +		return -EALREADY;
> > > > +	}
> > > > +
> > > > +	q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
> > > > +					    blk_debugfs_root);
> > > > +	if (!q->debugfs_dir)
> > > > +		return -ENOMEM;
> > > 
> > > Why doesn't the directory just live in the request queue, or somewhere
> > > else, so that you save it when it is created and then that's it.  No
> > > need to "look it up" anywhere else.
> > 
> > Its already there. And yes, after my changes it is technically possible
> > to just re-use it directly. But this is complicated by a few things. One
> > is that at this point in time, asynchronous request_queue removal is
> > still possible, and so a race was exposed where a requeust_queue may be
> > lingering but its old device is gone. That race is fixed by reverting us
> > back to synchronous request_queue removal, therefore ensuring that the
> > debugfs dir exists so long as the device does.
> > 
> > I can remove the debugfs_lookup() *after* we revert to synchronous
> > request_queue removal, or we just re-order the patches so that the
> > revert happens first. That should simplify this patch.
> > 
> > The code in this patch was designed to help dispute the logic behind
> > the CVE, in particular it shows exactly where debugfs_dir *is* the
> > one found by debugfs_lookup(), and shows the real issue behind the
> > removal.
> > 
> > But yeah, now that that is done, I hope its clear to all, and I think
> > this patch can be simplified if we just revert the async requeust_queue
> > removal first.
> 
> Don't try to "dispute" crazyness, that's not what kernel code is for.
> Just do the right thing, and simply saving off the pointer to the
> debugfs file when created is the "correct" thing to do, no matter what.
> No race conditions or anything else can happen when you do that.

Nope, races are still possible even if we move revert back to sync
request_queue removal, but I do believe that just reveals a bug
elsewhere, which I'll just fix as I think I know where this is.

> > > Or do you do that in later patches?  I only see this one at the moment,
> > > sorry...
> > > 
> > > >  static struct dentry *blk_trace_debugfs_dir(struct blk_user_trace_setup *buts,
> > > > +					    struct request_queue *q,
> > > >  					    struct blk_trace *bt)
> > > >  {
> > > >  	struct dentry *dir = NULL;
> > > >  
> > > > +	/* This can only happen if we have a bug on our lower layers */
> > > > +	if (!q->kobj.parent) {
> > > > +		pr_warn("%s: request_queue parent is gone\n", buts->name);
> > > 
> > > A kobject always has a parent, unless it has not been registered yet, so
> > > I don't know what you are testing could ever happen.
> > 
> > Or it has been kobject_del()'d?
> 
> If that happened, how in the world are you in this function anyway, as
> the request_queue is an invalid pointer at that point in time???

Nope, the block layer still finishes some work on it.

Drivers are allowed to cleanup a block device in this order, this
example,  is from the loop block driver:

static void loop_remove(struct loop_device *lo)                                 
{
	del_gendisk(lo->lo_disk);
	blk_cleanup_queue(lo->lo_queue);
	blk_mq_free_tag_set(&lo->tag_set);
	put_disk(lo->lo_disk);
	kfree(lo);
}   

At this point in time patch-wise we still haven't reverted back to
synchronous request_queue removal. Considering this, a race with the
parent disappearing can happen because the request_queue removal is
deferred, that is, the request_queue's kobject's release() call used
schedule_work() to finish off its removal. We expect the last
blk_put_queue() to be called at the end of blk_cleanup_queue(). Since
that is deferred and device_del() is called also at the end of
del_gendisk(), it means the release of the queue can happen in a
context where the disk is gone.

Although this issue is triggerable easily with the current async
request_queue removal, I can think of other ways to trigger an issue
here and one of them was suggested as possible by Christoph on the last
v1 patch series.

blk_queue_get() is not atomic and so what it returns can be incorrect:

bool blk_get_queue(struct request_queue *q)
{
	if (likely(!blk_queue_dying(q))) {
		__blk_get_queue(q);
		return true;
	}
	----> we can schedule here easily and move the queue to dying
	----> and race with blk_cleanup_queue() which will then allow
	----> code paths to incorrectly trigger the release of the queue
	----> in places we don't want
	return false;
}
EXPORT_SYMBOL(blk_get_queue);

Another area of concern I am seeing through code inspection is that
since the request_queue life depends on the disk, it seemse odd we call
device_del() before we remove the request_queue. If this is indeed
wrong, we could move the device_del() from del_gendisk() to
disk_release() triggered by the last put_disk().

I have a test now which shows that even if we revert back to synchronous
request_queue removal I can trigger a panic on the same use-after-free
case on debugfs on blktrace, and this is because we are overwriting the
same debugfs directory, and I think the above are its root causes.

I can fix this bug, but that just moves the bug to conflicts within
two sysfs objects already existing, and this is because add_disk()
(which calls __device_add_disk() doesn't return errors). This is both
a blk layer bug in the sense we never check for error and a driver bug
for allowing conflicts. All this just needs to be fixed, and although I
thought this could be done later, I think I'm just going to fix all this
now.

I have reproducer now which enables the same race with synchronous
request_queue removal, I'll work my way towards fixing all this...

> > A deferred requeust_queue removal shows this is possible, the parent is
> > taken underneath from us because the refcounting of this kobject is
> > already kobject_del()'d, and its actual removal scheduled for later.
> > The parent is taken underneath from us prior to the scheduled removal
> > completing.
> 
> No, a parent's reference is always valid while the child pointer is
> alive.

OK I'll keep thsi in mind then.

> There is an odd race condition right now that we are working on fixing
> if you notice another lkml thread, but that race will soon be fixed.

Got a pointer?

> So
> no need for every single user in the kernel to try to test for something
> like this (hint, this check is still wrong as with this logic, what
> could prevent parent from going away right _after_ you check it???)
> 
> Just remove this invalid check please.

Yeah sure, make sense.

> > > > +		return NULL;
> > > > +	}
> > > > +
> > > > +	/*
> > > > +	 * From a sysfs kobject perspective, the request_queue sits on top of
> > > > +	 * the gendisk, which has the name of the disk. We always create a
> > > > +	 * debugfs directory upon init for this gendisk kobject, so we re-use
> > > > +	 * that if blktrace is going to be done for it.
> > > > +	 */
> > > > +	if (blk_trace_target_disk(buts->name, kobject_name(q->kobj.parent))) {
> > > > +		if (!q->debugfs_dir) {
> > > > +			pr_warn("%s: expected request_queue debugfs_dir is not set\n",
> > > > +				buts->name);
> > > 
> > > What is userspace supposed to be able to do if they see this warning?
> > 
> > Userspace doesn't parse warnings, but the NULL ensures it won't crash
> > the kernel. The warn informs the kernel of a possible block layer bug.
> 
> Again, the code should not care if the pointer is NULL, as only debugfs
> deals with those pointers (and it can handle a NULL pointer just fine.)

Alright.

> > > > +			return NULL;
> > > > +		}
> > > > +		/*
> > > > +		 * debugfs_lookup() is used to ensure the directory is not
> > > > +		 * taken from underneath us. We must dput() it later once
> > > > +		 * done with it within blktrace.
> > > > +		 */
> > > > +		dir = debugfs_lookup(buts->name, blk_debugfs_root);
> > > > +		if (!dir) {
> > > > +			pr_warn("%s: expected request_queue debugfs_dir dentry is gone\n",
> > > > +				buts->name);
> > > 
> > > Again, can't we just save the pointer when we create it and not have to
> > > look it up again?
> > 
> > Only if we do the revert of the requeust_queue removal first.
> 
> Your end goal should be no more debugfs_lookup() calls.  Hopefully by
> the end of this patch series, that is the result.

Alrighty.

> > > > +			return NULL;
> > > > +		}
> > > > +		 /*
> > > > +		 * This is a reaffirmation that debugfs_lookup() shall always
> > > > +		 * return the same dentry if it was already set.
> > > > +		 */
> > > 
> > > I'm all for reaffirmation and the like, but really, is this needed???
> > 
> > To those who were still not sure that the issue was not a debugfs issue
> > I hoped this to make it clear. But indeed, if we revert back to
> > synchronous request_queue removal, that should simplify this.
> 
> Again, don't pander to crazies :)

At least I didn't say it :)

  Luis
Christoph Hellwig April 22, 2020, 7:29 a.m. UTC | #10
On Mon, Apr 20, 2020 at 08:41:56PM +0000, Luis Chamberlain wrote:
> Its already there. And yes, after my changes it is technically possible
> to just re-use it directly. But this is complicated by a few things. One
> is that at this point in time, asynchronous request_queue removal is
> still possible, and so a race was exposed where a requeust_queue may be
> lingering but its old device is gone. That race is fixed by reverting us
> back to synchronous request_queue removal, therefore ensuring that the
> debugfs dir exists so long as the device does.
> 
> I can remove the debugfs_lookup() *after* we revert to synchronous
> request_queue removal, or we just re-order the patches so that the
> revert happens first. That should simplify this patch.

Yes, please move the synchronous removal first instead of working around
the current problems.
Luis Chamberlain April 22, 2020, 7:34 a.m. UTC | #11
On Wed, Apr 22, 2020 at 12:29:42AM -0700, Christoph Hellwig wrote:
> On Mon, Apr 20, 2020 at 08:41:56PM +0000, Luis Chamberlain wrote:
> > Its already there. And yes, after my changes it is technically possible
> > to just re-use it directly. But this is complicated by a few things. One
> > is that at this point in time, asynchronous request_queue removal is
> > still possible, and so a race was exposed where a requeust_queue may be
> > lingering but its old device is gone. That race is fixed by reverting us
> > back to synchronous request_queue removal, therefore ensuring that the
> > debugfs dir exists so long as the device does.
> > 
> > I can remove the debugfs_lookup() *after* we revert to synchronous
> > request_queue removal, or we just re-order the patches so that the
> > revert happens first. That should simplify this patch.
> 
> Yes, please move the synchronous removal first instead of working around
> the current problems.

Sounds good. At first it was questionable, now its understood we need it.

  Luis
Luis Chamberlain April 22, 2020, 7:48 a.m. UTC | #12
On Wed, Apr 22, 2020 at 12:27:15AM -0700, Christoph Hellwig wrote:
> On Sun, Apr 19, 2020 at 07:45:22PM +0000, Luis Chamberlain wrote:
> > +{
> > +	struct dentry *dir = NULL;
> > +
> > +	/* This can happen if we have a bug in the lower layers */
> > +	dir = debugfs_lookup(kobject_name(q->kobj.parent), blk_debugfs_root);
> > +	if (dir) {
> > +		pr_warn("%s: registering request_queue debugfs directory twice is not allowed\n",
> > +			kobject_name(q->kobj.parent));
> > +		dput(dir);
> > +		return -EALREADY;
> > +	}
> 
> I don't see why we need this check.  If it is valueable enough we
> should have a debugfs_create_dir_exclusive or so that retunrns an error
> for an exsting directory, instead of reimplementing it in the caller in
> a racy way.  But I'm not really sure we need it to start with.

In short races, and even with synchronous request_queue removal I'm
seeing the race is still possible, but that's due to some other races
I'm going to chase down now.

The easier solution really is to just have a debugfs dir created for
each partition if debugfs is enabled, this way the directory will
always be there, and the lookups are gone.

> > +
> > +	q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
> > +					    blk_debugfs_root);
> > +	if (!q->debugfs_dir)
> > +		return -ENOMEM;
> > +
> > +	return 0;
> > +}
> > +
> > +void blk_queue_debugfs_unregister(struct request_queue *q)
> > +{
> > +	debugfs_remove_recursive(q->debugfs_dir);
> > +	q->debugfs_dir = NULL;
> > +}
> 
> Which to me suggests we can just fold these two into the callers,
> with an IS_ENABLED for the creation case given that we check for errors
> and the stub will always return an error.

Sorry not sure I follow this.

> >  	debugfs_create_files(q->debugfs_dir, q, blk_mq_debugfs_queue_attrs);
> >  
> >  	/*
> > @@ -856,9 +853,7 @@ void blk_mq_debugfs_register(struct request_queue *q)
> >  
> >  void blk_mq_debugfs_unregister(struct request_queue *q)
> >  {
> > -	debugfs_remove_recursive(q->debugfs_dir);
> >  	q->sched_debugfs_dir = NULL;
> > -	q->debugfs_dir = NULL;
> >  }
> 
> This function is weird - the sched dir gets removed by the
> debugfs_remove_recursive, so just leaving a function that clears
> a pointer is rather odd.  In fact I don't think we need to clear
> either sched_debugfs_dir or debugfs_dir anywhere.

Indeed. Will clean it up.

> > @@ -975,6 +976,14 @@ int blk_register_queue(struct gendisk *disk)
> >  		goto unlock;
> >  	}
> >  
> > +	ret = blk_queue_debugfs_register(q);
> > +	if (ret) {
> > +		blk_trace_remove_sysfs(dev);
> > +		kobject_del(&q->kobj);
> > +		kobject_put(&dev->kobj);
> > +		goto unlock;
> > +	}
> > +
> 
> Please use a goto label to consolidate the common cleanup code.

Sure.

> Also I think these generic debugfs changes probably should be separate
> to the blktrace changes.

I'll try to do that.

> >  static struct dentry *blk_trace_debugfs_dir(struct blk_user_trace_setup *buts,
> > +					    struct request_queue *q,
> >  					    struct blk_trace *bt)
> >  {
> >  	struct dentry *dir = NULL;
> >  
> > +	/* This can only happen if we have a bug on our lower layers */
> > +	if (!q->kobj.parent) {
> > +		pr_warn("%s: request_queue parent is gone\n", buts->name);
> > +		return NULL;
> > +	}
> 
> Why is this not simply a WARN_ON_ONCE()?

I'll actually remove it and instead fix the race where it happens.

> > +	if (blk_trace_target_disk(buts->name, kobject_name(q->kobj.parent))) {
> > +		if (!q->debugfs_dir) {
> > +			pr_warn("%s: expected request_queue debugfs_dir is not set\n",
> > +				buts->name);
> > +			return NULL;
> > +		}
> > +		/*
> > +		 * debugfs_lookup() is used to ensure the directory is not
> > +		 * taken from underneath us. We must dput() it later once
> > +		 * done with it within blktrace.
> > +		 */
> > +		dir = debugfs_lookup(buts->name, blk_debugfs_root);
> > +		if (!dir) {
> > +			pr_warn("%s: expected request_queue debugfs_dir dentry is gone\n",
> > +				buts->name);
> > +			return NULL;
> > +		}
> > +		 /*
> > +		 * This is a reaffirmation that debugfs_lookup() shall always
> > +		 * return the same dentry if it was already set.
> > +		 */
> > +		if (dir != q->debugfs_dir) {
> > +			dput(dir);
> > +			pr_warn("%s: expected dentry dir != q->debugfs_dir\n",
> > +				buts->name);
> > +			return NULL;
> > +		}
> > +		bt->backing_dir = q->debugfs_dir;
> > +		return bt->backing_dir;
> > +	}
> 
> Even with the gigantic commit log I don't get the point of this
> code.  It looks rather sketchy and I can't find a rationale for it.

Yeah I think this is going to be much easier on the eyes with the
revert to synchronous request_queue removal first.

  Luis
Christoph Hellwig April 22, 2020, 8:10 a.m. UTC | #13
On Wed, Apr 22, 2020 at 07:48:02AM +0000, Luis Chamberlain wrote:
> > I don't see why we need this check.  If it is valueable enough we
> > should have a debugfs_create_dir_exclusive or so that retunrns an error
> > for an exsting directory, instead of reimplementing it in the caller in
> > a racy way.  But I'm not really sure we need it to start with.
> 
> In short races, and even with synchronous request_queue removal I'm
> seeing the race is still possible, but that's due to some other races
> I'm going to chase down now.
> 
> The easier solution really is to just have a debugfs dir created for
> each partition if debugfs is enabled, this way the directory will
> always be there, and the lookups are gone.

That sounds like the best plan to me.

> 
> > > +
> > > +	q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
> > > +					    blk_debugfs_root);
> > > +	if (!q->debugfs_dir)
> > > +		return -ENOMEM;
> > > +
> > > +	return 0;
> > > +}
> > > +
> > > +void blk_queue_debugfs_unregister(struct request_queue *q)
> > > +{
> > > +	debugfs_remove_recursive(q->debugfs_dir);
> > > +	q->debugfs_dir = NULL;
> > > +}
> > 
> > Which to me suggests we can just fold these two into the callers,
> > with an IS_ENABLED for the creation case given that we check for errors
> > and the stub will always return an error.
> 
> Sorry not sure I follow this.

Don't both with the two above functions and just open code them in
the callers.  IFF you still want to check for errors after the
discussion with Greg, wrap the call in a

	if (IS_ENABLED(CONFIG_DEBUG_FS))

to ensure that you don't fail queue creation in the !DEBUG_FS
case.
Luis Chamberlain April 22, 2020, 8:26 a.m. UTC | #14
On Wed, Apr 22, 2020 at 01:10:11AM -0700, Christoph Hellwig wrote:
> On Wed, Apr 22, 2020 at 07:48:02AM +0000, Luis Chamberlain wrote:
> > > I don't see why we need this check.  If it is valueable enough we
> > > should have a debugfs_create_dir_exclusive or so that retunrns an error
> > > for an exsting directory, instead of reimplementing it in the caller in
> > > a racy way.  But I'm not really sure we need it to start with.
> > 
> > In short races, and even with synchronous request_queue removal I'm
> > seeing the race is still possible, but that's due to some other races
> > I'm going to chase down now.
> > 
> > The easier solution really is to just have a debugfs dir created for
> > each partition if debugfs is enabled, this way the directory will
> > always be there, and the lookups are gone.
> 
> That sounds like the best plan to me.

Groovy.

> > > > +
> > > > +	q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
> > > > +					    blk_debugfs_root);
> > > > +	if (!q->debugfs_dir)
> > > > +		return -ENOMEM;
> > > > +
> > > > +	return 0;
> > > > +}
> > > > +
> > > > +void blk_queue_debugfs_unregister(struct request_queue *q)
> > > > +{
> > > > +	debugfs_remove_recursive(q->debugfs_dir);
> > > > +	q->debugfs_dir = NULL;
> > > > +}
> > > 
> > > Which to me suggests we can just fold these two into the callers,
> > > with an IS_ENABLED for the creation case given that we check for errors
> > > and the stub will always return an error.
> > 
> > Sorry not sure I follow this.
> 
> Don't both with the two above functions and just open code them in
> the callers.  IFF you still want to check for errors after the
> discussion with Greg, wrap the call in a
> 
> 	if (IS_ENABLED(CONFIG_DEBUG_FS))
> 
> to ensure that you don't fail queue creation in the !DEBUG_FS
> case.

Got it, thanks.

  Luis
Ming Lei April 22, 2020, 9:43 a.m. UTC | #15
On Wed, Apr 22, 2020 at 07:28:59AM +0000, Luis Chamberlain wrote:
> On Tue, Apr 21, 2020 at 09:00:48AM +0200, Greg KH wrote:
> > On Mon, Apr 20, 2020 at 08:41:56PM +0000, Luis Chamberlain wrote:
> > > On Mon, Apr 20, 2020 at 10:16:15PM +0200, Greg KH wrote:
> > > > On Sun, Apr 19, 2020 at 07:45:22PM +0000, Luis Chamberlain wrote:
> > > > 
> > > > This patch triggered gmail's spam detection, your changelog text is
> > > > whack...
> > > 
> > > Oh? What do you think triggered it?
> > 
> > No idea.
> 
> Alright, well I'm going to move most of the analysis to the bug report
> and be as concise as possible on the commit log.
> 
> > > > > diff --git a/block/blk-debugfs.c b/block/blk-debugfs.c
> > > > > index 19091e1effc0..d84038bce0a5 100644
> > > > > --- a/block/blk-debugfs.c
> > > > > +++ b/block/blk-debugfs.c
> > > > > @@ -3,6 +3,9 @@
> > > > >  /*
> > > > >   * Shared request-based / make_request-based functionality
> > > > >   */
> > > > > +
> > > > > +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
> > > > > +
> > > > >  #include <linux/kernel.h>
> > > > >  #include <linux/blkdev.h>
> > > > >  #include <linux/debugfs.h>
> > > > > @@ -13,3 +16,30 @@ void blk_debugfs_register(void)
> > > > >  {
> > > > >  	blk_debugfs_root = debugfs_create_dir("block", NULL);
> > > > >  }
> > > > > +
> > > > > +int __must_check blk_queue_debugfs_register(struct request_queue *q)
> > > > > +{
> > > > > +	struct dentry *dir = NULL;
> > > > > +
> > > > > +	/* This can happen if we have a bug in the lower layers */
> > > > > +	dir = debugfs_lookup(kobject_name(q->kobj.parent), blk_debugfs_root);
> > > > > +	if (dir) {
> > > > > +		pr_warn("%s: registering request_queue debugfs directory twice is not allowed\n",
> > > > > +			kobject_name(q->kobj.parent));
> > > > > +		dput(dir);
> > > > > +		return -EALREADY;
> > > > > +	}
> > > > > +
> > > > > +	q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
> > > > > +					    blk_debugfs_root);
> > > > > +	if (!q->debugfs_dir)
> > > > > +		return -ENOMEM;
> > > > 
> > > > Why doesn't the directory just live in the request queue, or somewhere
> > > > else, so that you save it when it is created and then that's it.  No
> > > > need to "look it up" anywhere else.
> > > 
> > > Its already there. And yes, after my changes it is technically possible
> > > to just re-use it directly. But this is complicated by a few things. One
> > > is that at this point in time, asynchronous request_queue removal is
> > > still possible, and so a race was exposed where a requeust_queue may be
> > > lingering but its old device is gone. That race is fixed by reverting us
> > > back to synchronous request_queue removal, therefore ensuring that the
> > > debugfs dir exists so long as the device does.
> > > 
> > > I can remove the debugfs_lookup() *after* we revert to synchronous
> > > request_queue removal, or we just re-order the patches so that the
> > > revert happens first. That should simplify this patch.
> > > 
> > > The code in this patch was designed to help dispute the logic behind
> > > the CVE, in particular it shows exactly where debugfs_dir *is* the
> > > one found by debugfs_lookup(), and shows the real issue behind the
> > > removal.
> > > 
> > > But yeah, now that that is done, I hope its clear to all, and I think
> > > this patch can be simplified if we just revert the async requeust_queue
> > > removal first.
> > 
> > Don't try to "dispute" crazyness, that's not what kernel code is for.
> > Just do the right thing, and simply saving off the pointer to the
> > debugfs file when created is the "correct" thing to do, no matter what.
> > No race conditions or anything else can happen when you do that.
> 
> Nope, races are still possible even if we move revert back to sync
> request_queue removal, but I do believe that just reveals a bug
> elsewhere, which I'll just fix as I think I know where this is.
> 
> > > > Or do you do that in later patches?  I only see this one at the moment,
> > > > sorry...
> > > > 
> > > > >  static struct dentry *blk_trace_debugfs_dir(struct blk_user_trace_setup *buts,
> > > > > +					    struct request_queue *q,
> > > > >  					    struct blk_trace *bt)
> > > > >  {
> > > > >  	struct dentry *dir = NULL;
> > > > >  
> > > > > +	/* This can only happen if we have a bug on our lower layers */
> > > > > +	if (!q->kobj.parent) {
> > > > > +		pr_warn("%s: request_queue parent is gone\n", buts->name);
> > > > 
> > > > A kobject always has a parent, unless it has not been registered yet, so
> > > > I don't know what you are testing could ever happen.
> > > 
> > > Or it has been kobject_del()'d?
> > 
> > If that happened, how in the world are you in this function anyway, as
> > the request_queue is an invalid pointer at that point in time???
> 
> Nope, the block layer still finishes some work on it.
> 
> Drivers are allowed to cleanup a block device in this order, this
> example,  is from the loop block driver:
> 
> static void loop_remove(struct loop_device *lo)                                 
> {
> 	del_gendisk(lo->lo_disk);
> 	blk_cleanup_queue(lo->lo_queue);
> 	blk_mq_free_tag_set(&lo->tag_set);
> 	put_disk(lo->lo_disk);
> 	kfree(lo);
> }   
> 
> At this point in time patch-wise we still haven't reverted back to
> synchronous request_queue removal. Considering this, a race with the
> parent disappearing can happen because the request_queue removal is
> deferred, that is, the request_queue's kobject's release() call used
> schedule_work() to finish off its removal. We expect the last
> blk_put_queue() to be called at the end of blk_cleanup_queue(). Since

Actually no, we expect that request queue is released after disk is
released. Don't forget that gendisk does hold one extra refcount of
request queue.

> that is deferred and device_del() is called also at the end of
> del_gendisk(), it means the release of the queue can happen in a
> context where the disk is gone.
> 
> Although this issue is triggerable easily with the current async
> request_queue removal, I can think of other ways to trigger an issue
> here and one of them was suggested as possible by Christoph on the last
> v1 patch series.
> 
> blk_queue_get() is not atomic and so what it returns can be incorrect:
> 
> bool blk_get_queue(struct request_queue *q)
> {
> 	if (likely(!blk_queue_dying(q))) {
> 		__blk_get_queue(q);
> 		return true;
> 	}
> 	----> we can schedule here easily and move the queue to dying
> 	----> and race with blk_cleanup_queue() which will then allow
> 	----> code paths to incorrectly trigger the release of the queue
> 	----> in places we don't want

Right, actually caller of blk_get_queue() has to guarantee that
the request queue is alive.

Some users of blk_get_queue() aren't necessary, such as rbd and mmc.

> 	return false;
> }
> EXPORT_SYMBOL(blk_get_queue);
> 
> Another area of concern I am seeing through code inspection is that
> since the request_queue life depends on the disk, it seemse odd we call
> device_del() before we remove the request_queue. If this is indeed
> wrong, we could move the device_del() from del_gendisk() to
> disk_release() triggered by the last put_disk().

Why do you think it is odd and want to change this way? Disk has to be
removed first for draining dirty pages to queue, then we can cleanup
queue. Once we start to clean up request queue, all data may not land
disk any more.

> 
> I have a test now which shows that even if we revert back to synchronous
> request_queue removal I can trigger a panic on the same use-after-free
> case on debugfs on blktrace, and this is because we are overwriting the
> same debugfs directory, and I think the above are its root causes.

The reason should be shared debugfs dir between blktrace and blk-mq
debugfs.

> 
> I can fix this bug, but that just moves the bug to conflicts within
> two sysfs objects already existing, and this is because add_disk()
> (which calls __device_add_disk() doesn't return errors). This is both
> a blk layer bug in the sense we never check for error and a driver bug
> for allowing conflicts. All this just needs to be fixed, and although I
> thought this could be done later, I think I'm just going to fix all this
> now.

Yeah, we talked that several times, looks no one tries to post patch to
fix that.


Thanks, 
Ming
Luis Chamberlain April 22, 2020, 10:31 a.m. UTC | #16
On Wed, Apr 22, 2020 at 05:43:20PM +0800, Ming Lei wrote:
> On Wed, Apr 22, 2020 at 07:28:59AM +0000, Luis Chamberlain wrote:
> > On Tue, Apr 21, 2020 at 09:00:48AM +0200, Greg KH wrote:
> > > On Mon, Apr 20, 2020 at 08:41:56PM +0000, Luis Chamberlain wrote:
> > > > On Mon, Apr 20, 2020 at 10:16:15PM +0200, Greg KH wrote:
> > > > > On Sun, Apr 19, 2020 at 07:45:22PM +0000, Luis Chamberlain wrote:
> > > > > 
> > > > > This patch triggered gmail's spam detection, your changelog text is
> > > > > whack...
> > > > 
> > > > Oh? What do you think triggered it?
> > > 
> > > No idea.
> > 
> > Alright, well I'm going to move most of the analysis to the bug report
> > and be as concise as possible on the commit log.
> > 
> > > > > > diff --git a/block/blk-debugfs.c b/block/blk-debugfs.c
> > > > > > index 19091e1effc0..d84038bce0a5 100644
> > > > > > --- a/block/blk-debugfs.c
> > > > > > +++ b/block/blk-debugfs.c
> > > > > > @@ -3,6 +3,9 @@
> > > > > >  /*
> > > > > >   * Shared request-based / make_request-based functionality
> > > > > >   */
> > > > > > +
> > > > > > +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
> > > > > > +
> > > > > >  #include <linux/kernel.h>
> > > > > >  #include <linux/blkdev.h>
> > > > > >  #include <linux/debugfs.h>
> > > > > > @@ -13,3 +16,30 @@ void blk_debugfs_register(void)
> > > > > >  {
> > > > > >  	blk_debugfs_root = debugfs_create_dir("block", NULL);
> > > > > >  }
> > > > > > +
> > > > > > +int __must_check blk_queue_debugfs_register(struct request_queue *q)
> > > > > > +{
> > > > > > +	struct dentry *dir = NULL;
> > > > > > +
> > > > > > +	/* This can happen if we have a bug in the lower layers */
> > > > > > +	dir = debugfs_lookup(kobject_name(q->kobj.parent), blk_debugfs_root);
> > > > > > +	if (dir) {
> > > > > > +		pr_warn("%s: registering request_queue debugfs directory twice is not allowed\n",
> > > > > > +			kobject_name(q->kobj.parent));
> > > > > > +		dput(dir);
> > > > > > +		return -EALREADY;
> > > > > > +	}
> > > > > > +
> > > > > > +	q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
> > > > > > +					    blk_debugfs_root);
> > > > > > +	if (!q->debugfs_dir)
> > > > > > +		return -ENOMEM;
> > > > > 
> > > > > Why doesn't the directory just live in the request queue, or somewhere
> > > > > else, so that you save it when it is created and then that's it.  No
> > > > > need to "look it up" anywhere else.
> > > > 
> > > > Its already there. And yes, after my changes it is technically possible
> > > > to just re-use it directly. But this is complicated by a few things. One
> > > > is that at this point in time, asynchronous request_queue removal is
> > > > still possible, and so a race was exposed where a requeust_queue may be
> > > > lingering but its old device is gone. That race is fixed by reverting us
> > > > back to synchronous request_queue removal, therefore ensuring that the
> > > > debugfs dir exists so long as the device does.
> > > > 
> > > > I can remove the debugfs_lookup() *after* we revert to synchronous
> > > > request_queue removal, or we just re-order the patches so that the
> > > > revert happens first. That should simplify this patch.
> > > > 
> > > > The code in this patch was designed to help dispute the logic behind
> > > > the CVE, in particular it shows exactly where debugfs_dir *is* the
> > > > one found by debugfs_lookup(), and shows the real issue behind the
> > > > removal.
> > > > 
> > > > But yeah, now that that is done, I hope its clear to all, and I think
> > > > this patch can be simplified if we just revert the async requeust_queue
> > > > removal first.
> > > 
> > > Don't try to "dispute" crazyness, that's not what kernel code is for.
> > > Just do the right thing, and simply saving off the pointer to the
> > > debugfs file when created is the "correct" thing to do, no matter what.
> > > No race conditions or anything else can happen when you do that.
> > 
> > Nope, races are still possible even if we move revert back to sync
> > request_queue removal, but I do believe that just reveals a bug
> > elsewhere, which I'll just fix as I think I know where this is.
> > 
> > > > > Or do you do that in later patches?  I only see this one at the moment,
> > > > > sorry...
> > > > > 
> > > > > >  static struct dentry *blk_trace_debugfs_dir(struct blk_user_trace_setup *buts,
> > > > > > +					    struct request_queue *q,
> > > > > >  					    struct blk_trace *bt)
> > > > > >  {
> > > > > >  	struct dentry *dir = NULL;
> > > > > >  
> > > > > > +	/* This can only happen if we have a bug on our lower layers */
> > > > > > +	if (!q->kobj.parent) {
> > > > > > +		pr_warn("%s: request_queue parent is gone\n", buts->name);
> > > > > 
> > > > > A kobject always has a parent, unless it has not been registered yet, so
> > > > > I don't know what you are testing could ever happen.
> > > > 
> > > > Or it has been kobject_del()'d?
> > > 
> > > If that happened, how in the world are you in this function anyway, as
> > > the request_queue is an invalid pointer at that point in time???
> > 
> > Nope, the block layer still finishes some work on it.
> > 
> > Drivers are allowed to cleanup a block device in this order, this
> > example,  is from the loop block driver:
> > 
> > static void loop_remove(struct loop_device *lo)                                 
> > {
> > 	del_gendisk(lo->lo_disk);
> > 	blk_cleanup_queue(lo->lo_queue);
> > 	blk_mq_free_tag_set(&lo->tag_set);
> > 	put_disk(lo->lo_disk);
> > 	kfree(lo);
> > }   
> > 
> > At this point in time patch-wise we still haven't reverted back to
> > synchronous request_queue removal. Considering this, a race with the
> > parent disappearing can happen because the request_queue removal is
> > deferred, that is, the request_queue's kobject's release() call used
> > schedule_work() to finish off its removal. We expect the last
> > blk_put_queue() to be called at the end of blk_cleanup_queue(). Since
> 
> Actually no, we expect that request queue is released after disk is
> released. Don't forget that gendisk does hold one extra refcount of
> request queue.

Ah yes.

Still the device_del() occurs before, this means the sysfs path
is cleared for a new device to come in as well, and this can happen
even with synchronous request_queue removal.

I have some changes to try to help address this now.

> > that is deferred and device_del() is called also at the end of
> > del_gendisk(), it means the release of the queue can happen in a
> > context where the disk is gone.
> > 
> > Although this issue is triggerable easily with the current async
> > request_queue removal, I can think of other ways to trigger an issue
> > here and one of them was suggested as possible by Christoph on the last
> > v1 patch series.
> > 
> > blk_queue_get() is not atomic and so what it returns can be incorrect:
> > 
> > bool blk_get_queue(struct request_queue *q)
> > {
> > 	if (likely(!blk_queue_dying(q))) {
> > 		__blk_get_queue(q);
> > 		return true;
> > 	}
> > 	----> we can schedule here easily and move the queue to dying
> > 	----> and race with blk_cleanup_queue() which will then allow
> > 	----> code paths to incorrectly trigger the release of the queue
> > 	----> in places we don't want
> 
> Right, actually caller of blk_get_queue() has to guarantee that
> the request queue is alive.

Sure.

> Some users of blk_get_queue() aren't necessary, such as rbd and mmc.

Are you saying we can remove those calls on rbd / mmc or something else?

> 
> > 	return false;
> > }
> > EXPORT_SYMBOL(blk_get_queue);
> > 
> > Another area of concern I am seeing through code inspection is that
> > since the request_queue life depends on the disk, it seemse odd we call
> > device_del() before we remove the request_queue. If this is indeed
> > wrong, we could move the device_del() from del_gendisk() to
> > disk_release() triggered by the last put_disk().
> 
> Why do you think it is odd and want to change this way? Disk has to be
> removed first for draining dirty pages to queue, then we can cleanup
> queue. Once we start to clean up request queue, all data may not land
> disk any more.

I think that all that can be done as-is today, an issue I suspect exists
is that we remove the disk from syfs hierarchy prior to the request
queue, whereas we expect this to be in the inverse order.

> > I have a test now which shows that even if we revert back to synchronous
> > request_queue removal I can trigger a panic on the same use-after-free
> > case on debugfs on blktrace, and this is because we are overwriting the
> > same debugfs directory, and I think the above are its root causes.
> 
> The reason should be shared debugfs dir between blktrace and blk-mq
> debugfs.

I think its something else.

> > I can fix this bug, but that just moves the bug to conflicts within
> > two sysfs objects already existing, and this is because add_disk()
> > (which calls __device_add_disk() doesn't return errors). This is both
> > a blk layer bug in the sense we never check for error and a driver bug
> > for allowing conflicts. All this just needs to be fixed, and although I
> > thought this could be done later, I think I'm just going to fix all this
> > now.
> 
> Yeah, we talked that several times, looks no one tries to post patch to
> fix that.

Consider this done, just need to brush it up now and send for review.

  Luis
Luis Chamberlain April 24, 2020, 11:47 p.m. UTC | #17
On Wed, Apr 22, 2020 at 05:43:20PM +0800, Ming Lei wrote:
> On Wed, Apr 22, 2020 at 07:28:59AM +0000, Luis Chamberlain wrote:
> > At this point in time patch-wise we still haven't reverted back to
> > synchronous request_queue removal. Considering this, a race with the
> > parent disappearing can happen because the request_queue removal is
> > deferred, that is, the request_queue's kobject's release() call used
> > schedule_work() to finish off its removal. We expect the last
> > blk_put_queue() to be called at the end of blk_cleanup_queue(). Since
> 
> Actually no, we expect that request queue is released after disk is
> released. Don't forget that gendisk does hold one extra refcount of
> request queue.

Then by all means using blk_put_queue() from everywhere should be safe
in atomic context, as we have control over that blk_put_queue() on the
block layer.

(Modulo, we accept the races possible today on blk_get_queue(), which
I'll try to address).

  Luis
diff mbox series

Patch

diff --git a/block/blk-debugfs.c b/block/blk-debugfs.c
index 19091e1effc0..d84038bce0a5 100644
--- a/block/blk-debugfs.c
+++ b/block/blk-debugfs.c
@@ -3,6 +3,9 @@ 
 /*
  * Shared request-based / make_request-based functionality
  */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
 #include <linux/kernel.h>
 #include <linux/blkdev.h>
 #include <linux/debugfs.h>
@@ -13,3 +16,30 @@  void blk_debugfs_register(void)
 {
 	blk_debugfs_root = debugfs_create_dir("block", NULL);
 }
+
+int __must_check blk_queue_debugfs_register(struct request_queue *q)
+{
+	struct dentry *dir = NULL;
+
+	/* This can happen if we have a bug in the lower layers */
+	dir = debugfs_lookup(kobject_name(q->kobj.parent), blk_debugfs_root);
+	if (dir) {
+		pr_warn("%s: registering request_queue debugfs directory twice is not allowed\n",
+			kobject_name(q->kobj.parent));
+		dput(dir);
+		return -EALREADY;
+	}
+
+	q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
+					    blk_debugfs_root);
+	if (!q->debugfs_dir)
+		return -ENOMEM;
+
+	return 0;
+}
+
+void blk_queue_debugfs_unregister(struct request_queue *q)
+{
+	debugfs_remove_recursive(q->debugfs_dir);
+	q->debugfs_dir = NULL;
+}
diff --git a/block/blk-mq-debugfs.c b/block/blk-mq-debugfs.c
index b3f2ba483992..bda9378eab90 100644
--- a/block/blk-mq-debugfs.c
+++ b/block/blk-mq-debugfs.c
@@ -823,9 +823,6 @@  void blk_mq_debugfs_register(struct request_queue *q)
 	struct blk_mq_hw_ctx *hctx;
 	int i;
 
-	q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
-					    blk_debugfs_root);
-
 	debugfs_create_files(q->debugfs_dir, q, blk_mq_debugfs_queue_attrs);
 
 	/*
@@ -856,9 +853,7 @@  void blk_mq_debugfs_register(struct request_queue *q)
 
 void blk_mq_debugfs_unregister(struct request_queue *q)
 {
-	debugfs_remove_recursive(q->debugfs_dir);
 	q->sched_debugfs_dir = NULL;
-	q->debugfs_dir = NULL;
 }
 
 static void blk_mq_debugfs_register_ctx(struct blk_mq_hw_ctx *hctx,
diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
index fca9b158f4a0..7072f408e69a 100644
--- a/block/blk-sysfs.c
+++ b/block/blk-sysfs.c
@@ -895,6 +895,7 @@  static void __blk_release_queue(struct work_struct *work)
 
 	blk_trace_shutdown(q);
 
+	blk_queue_debugfs_unregister(q);
 	if (queue_is_mq(q))
 		blk_mq_debugfs_unregister(q);
 
@@ -975,6 +976,14 @@  int blk_register_queue(struct gendisk *disk)
 		goto unlock;
 	}
 
+	ret = blk_queue_debugfs_register(q);
+	if (ret) {
+		blk_trace_remove_sysfs(dev);
+		kobject_del(&q->kobj);
+		kobject_put(&dev->kobj);
+		goto unlock;
+	}
+
 	if (queue_is_mq(q)) {
 		__blk_mq_register_dev(dev, q);
 		blk_mq_debugfs_register(q);
diff --git a/block/blk.h b/block/blk.h
index 86a66b614f08..0cae5a92fe0b 100644
--- a/block/blk.h
+++ b/block/blk.h
@@ -489,10 +489,21 @@  int __bio_add_pc_page(struct request_queue *q, struct bio *bio,
 		bool *same_page);
 #ifdef CONFIG_DEBUG_FS
 void blk_debugfs_register(void);
+int __must_check blk_queue_debugfs_register(struct request_queue *q);
+void blk_queue_debugfs_unregister(struct request_queue *q);
 #else
 static inline void blk_debugfs_register(void)
 {
 }
+
+static inline int __must_check blk_queue_debugfs_register(struct request_queue *q)
+{
+	return 0;
+}
+
+static inline void blk_queue_debugfs_unregister(struct request_queue *q)
+{
+}
 #endif /* CONFIG_DEBUG_FS */
 
 #endif /* BLK_INTERNAL_H */
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 32868fbedc9e..cc43c8e6516c 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -569,8 +569,11 @@  struct request_queue {
 	struct list_head	tag_set_list;
 	struct bio_set		bio_split;
 
-#ifdef CONFIG_BLK_DEBUG_FS
+#ifdef CONFIG_DEBUG_FS
+	/* Used by block/blk-*debugfs.c and kernel/trace/blktrace.c */
 	struct dentry		*debugfs_dir;
+#endif
+#ifdef CONFIG_BLK_DEBUG_FS
 	struct dentry		*sched_debugfs_dir;
 	struct dentry		*rqos_debugfs_dir;
 #endif
diff --git a/include/linux/blktrace_api.h b/include/linux/blktrace_api.h
index 3b6ff5902edc..43a6f98840dc 100644
--- a/include/linux/blktrace_api.h
+++ b/include/linux/blktrace_api.h
@@ -23,6 +23,7 @@  struct blk_trace {
 	u32 pid;
 	u32 dev;
 	struct dentry *dir;
+	struct dentry *backing_dir;
 	struct dentry *dropped_file;
 	struct dentry *msg_file;
 	struct list_head running_list;
diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c
index 2c6e6c386ace..8f87979d0971 100644
--- a/kernel/trace/blktrace.c
+++ b/kernel/trace/blktrace.c
@@ -3,6 +3,9 @@ 
  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
  *
  */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
 #include <linux/kernel.h>
 #include <linux/blkdev.h>
 #include <linux/blktrace_api.h>
@@ -311,7 +314,15 @@  static void blk_trace_free(struct blk_trace *bt)
 	debugfs_remove(bt->msg_file);
 	debugfs_remove(bt->dropped_file);
 	relay_close(bt->rchan);
-	debugfs_remove(bt->dir);
+	/*
+	 * backing_dir is set when we use the request_queue debugfs directory.
+	 * Otherwise we are using a temporary directory created only for the
+	 * purpose of blktrace.
+	 */
+	if (bt->backing_dir)
+		dput(bt->backing_dir);
+	else
+		debugfs_remove(bt->dir);
 	free_percpu(bt->sequence);
 	free_percpu(bt->msg_data);
 	kfree(bt);
@@ -468,16 +479,89 @@  static void blk_trace_setup_lba(struct blk_trace *bt,
 	}
 }
 
+static bool blk_trace_target_disk(const char *target, const char *diskname)
+{
+	if (strlen(target) != strlen(diskname))
+		return false;
+
+	if (!strncmp(target, diskname,
+		     min_t(size_t, strlen(target), strlen(diskname))))
+		return true;
+
+	return false;
+}
+
 static struct dentry *blk_trace_debugfs_dir(struct blk_user_trace_setup *buts,
+					    struct request_queue *q,
 					    struct blk_trace *bt)
 {
 	struct dentry *dir = NULL;
 
+	/* This can only happen if we have a bug on our lower layers */
+	if (!q->kobj.parent) {
+		pr_warn("%s: request_queue parent is gone\n", buts->name);
+		return NULL;
+	}
+
+	/*
+	 * From a sysfs kobject perspective, the request_queue sits on top of
+	 * the gendisk, which has the name of the disk. We always create a
+	 * debugfs directory upon init for this gendisk kobject, so we re-use
+	 * that if blktrace is going to be done for it.
+	 */
+	if (blk_trace_target_disk(buts->name, kobject_name(q->kobj.parent))) {
+		if (!q->debugfs_dir) {
+			pr_warn("%s: expected request_queue debugfs_dir is not set\n",
+				buts->name);
+			return NULL;
+		}
+		/*
+		 * debugfs_lookup() is used to ensure the directory is not
+		 * taken from underneath us. We must dput() it later once
+		 * done with it within blktrace.
+		 */
+		dir = debugfs_lookup(buts->name, blk_debugfs_root);
+		if (!dir) {
+			pr_warn("%s: expected request_queue debugfs_dir dentry is gone\n",
+				buts->name);
+			return NULL;
+		}
+		 /*
+		 * This is a reaffirmation that debugfs_lookup() shall always
+		 * return the same dentry if it was already set.
+		 */
+		if (dir != q->debugfs_dir) {
+			dput(dir);
+			pr_warn("%s: expected dentry dir != q->debugfs_dir\n",
+				buts->name);
+			return NULL;
+		}
+		bt->backing_dir = q->debugfs_dir;
+		return bt->backing_dir;
+	}
+
+	/*
+	 * If not using blktrace on the gendisk, we are going to create a
+	 * temporary debugfs directory for it, however this cannot be shared
+	 * between two concurrent blktraces since the path is not unique, so
+	 * ensure this is only done once.
+	 */
 	dir = debugfs_lookup(buts->name, blk_debugfs_root);
-	if (!dir)
-		bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
+	if (dir) {
+		pr_warn("%s: temporary blktrace debugfs directory already present\n",
+			buts->name);
+		dput(dir);
+		return NULL;
+	}
+
+	bt->dir = debugfs_create_dir(buts->name, blk_debugfs_root);
+	if (!bt->dir) {
+		pr_warn("%s: temporary blktrace debugfs directory could not be created\n",
+			buts->name);
+		return NULL;
+	}
 
-	return dir;
+	return bt->dir;
 }
 
 /*
@@ -521,7 +605,9 @@  static int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 
 	ret = -ENOENT;
 
-	dir = blk_trace_debugfs_dir(buts, bt);
+	dir = blk_trace_debugfs_dir(buts, q, bt);
+	if (!dir)
+		goto err;
 
 	bt->dev = dev;
 	atomic_set(&bt->dropped, 0);
@@ -561,8 +647,6 @@  static int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 
 	ret = 0;
 err:
-	if (dir && !bt->dir)
-		dput(dir);
 	if (ret)
 		blk_trace_free(bt);
 	return ret;