On Thu, 2018-05-03 at 16:42 +0200, Jan Kara wrote:

On Wed 25-04-18 17:07:48, Fabiano Rosas wrote:

quoted

I'm looking into an issue where removing a virtio disk via sysfs while another
process is issuing write() calls results in the writing task going into a
livelock:


root@guest # cat test.sh
#!/bin/bash

dd if=/dev/zero of=/dev/vda bs=1M count=10000 &
sleep 1
echo 1 > /sys/bus/pci/devices/0000:00:04.0/remove

root@guest # ls /dev/vd*
/dev/vda

root@guest # grep Dirty /proc/meminfo
Dirty:                 0 kB

root@guest # sh test.sh

root@guest # ps aux | grep "[d]d if"
root      1699 38.6  0.0 111424  1216 hvc0     D+   10:48   0:01 dd if=/dev/zero of=/dev/vda bs=1M count=10000

root@guest # ls /dev/vd*
ls: cannot access /dev/vd*: No such file or directory

root@guest # cat /proc/1699/stack
[<0>] 0xc0000000ffe28218
[<0>] __switch_to+0x31c/0x480
[<0>] balance_dirty_pages+0x990/0xb90
[<0>] balance_dirty_pages_ratelimited+0x50c/0x6c0
[<0>] generic_perform_write+0x1b0/0x260
[<0>] __generic_file_write_iter+0x200/0x240
[<0>] blkdev_write_iter+0xa4/0x150
[<0>] __vfs_write+0x14c/0x240
[<0>] vfs_write+0xd0/0x240
[<0>] ksys_write+0x6c/0x110
[<0>] system_call+0x58/0x6c

root@guest # grep Dirty /proc/meminfo
Dirty:           1506816 kB

---

I have done some tracing and I believe this is caused by the clearing of
'WB_registered' in 'wb_shutdown':

<snip>
sh-1697  [000] ....  3994.541664: sysfs_remove_link <-del_gendisk
sh-1697  [000] ....  3994.541671: wb_shutdown <-bdi_unregister
<snip>

Later, when 'balance_dirty_pages' tries to start writeback, it doesn't happen
because 'WB_registered' is not set:

fs/fs-writeback.c

static void wb_wakeup(struct bdi_writeback *wb)
{
        spin_lock_bh(&wb->work_lock);
        if (test_bit(WB_registered, &wb->state))
                mod_delayed_work(bdi_wq, &wb->dwork, 0);
        spin_unlock_bh(&wb->work_lock);
}

So we get stuck in a loop in 'balance_dirty_pages':

root@guest # cat /sys/kernel/debug/tracing/set_ftrace_filter
balance_dirty_pages_ratelimited
balance_dirty_pages
wb_wakeup
wb_workfn
io_schedule_timeout

<snip>
dd-1699  [000] .... 11192.535946: wb_wakeup <-balance_dirty_pages
dd-1699  [000] .... 11192.535950: io_schedule_timeout <-balance_dirty_pages
dd-1699  [000] .... 11192.745968: wb_wakeup <-balance_dirty_pages
dd-1699  [000] .... 11192.745972: io_schedule_timeout <-balance_dirty_pages
<snip>


The test on 'WB_registered' before starting the writeback task is introduced
by: "5acda9 bdi: avoid oops on device removal".

I have made a *naive* attempt at fixing it by allowing writeback to happen even
without 'WB_registered':

diff --git a/fs/fs-writeback.c b/fs/fs-writeback.c
index d4d04fe..050b067 100644
--- a/fs/fs-writeback.c
+++ b/fs/fs-writeback.c

@@ -982,7 +982,7 @@ void wb_start_background_writeback(struct bdi_writeback *wb)
 	 * writeback as soon as there is no other work to do.
 	 */
 	trace_writeback_wake_background(wb);
-	wb_wakeup(wb);
+	mod_delayed_work(bdi_wq, &wb->dwork, 0);
 }
 
 /*

@@ -1933,7 +1933,7 @@ void wb_workfn(struct work_struct *work)
 						struct bdi_writeback, dwork);
 	long pages_written;
 
-	set_worker_desc("flush-%s", dev_name(wb->bdi->dev));
+	set_worker_desc("flush-%s", wb->bdi->dev ? dev_name(wb->bdi->dev) : "?" );
 	current->flags |= PF_SWAPWRITE;
 
 	if (likely(!current_is_workqueue_rescuer() ||
--

The effect of that is that the 'dd' process now finishes successfully and we
get "Buffer I/O error"s for the dirty pages that remain. I believe this to be
in conformance with existing interfaces since dd does not issue any fsync()
calls.


Does my analysis make any sense and would something along these lines be
acceptable as a solution?

Thanks for the debugging of the problem. I agree with your analysis however
I don't like your fix. The issue is that when bdi is unregistered we don't
really expect any writeback to happen after that moment. This is what
prevents various use-after-free issues and I'd like that to stay the way it
is.

What I think we should do is that we'll prevent dirtying of new pages when
we know the underlying device is gone. Because that will fix your problem
and also make sure user sees the IO errors directly instead of just in the
kernel log. The question is how to make this happen in the least painful
way. I think we could intercept writes in grab_cache_page_write_begin()
(which however requires that function to return a proper error code and not
just NULL / non-NULL). And we should also intercept write faults to not
allow page dirtying via mmap - probably somewhere in do_shared_fault() and
do_wp_page(). I've added Jeff to CC since he's dealing with IO error
handling a lot these days. Jeff, what do you think?

								Honza

(cc'ing Willy too since he's given this more thought than me)

For the record, I've mostly been looking at error _reporting_. Handling
errors at this level is not something I've really considered in great
detail as of yet.

Still, I think the basic idea sounds reasonable. Not allowing pages to
be dirtied when we can't clean them seems like a reasonable thing to
do.

The big question is how we'll report this to userland:

Would your approach have it return an error on write() and such? What
sort of error if so? ENODEV? Would we have to SIGBUS when someone tries
to dirty the page through mmap?
--
Jeff Layton [off-list ref]