Re: [PATCH 134/190] Revert "md: Fix failed allocation of md_register_thread"

[Greg Kroah-Hartman]

On Fri, Apr 30, 2021 at 06:10:24AM +0000, Song Liu wrote:
> 
> 
> > On Apr 27, 2021, at 10:46 PM, Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> > 
> > On Wed, Apr 21, 2021 at 03:00:09PM +0200, Greg Kroah-Hartman wrote:
> >> This reverts commit e406f12dde1a8375d77ea02d91f313fb1a9c6aec.
> >> 
> >> Commits from @umn.edu addresses have been found to be submitted in "bad
> >> faith" to try to test the kernel community's ability to review "known
> >> malicious" changes.  The result of these submissions can be found in a
> >> paper published at the 42nd IEEE Symposium on Security and Privacy
> >> entitled, "Open Source Insecurity: Stealthily Introducing
> >> Vulnerabilities via Hypocrite Commits" written by Qiushi Wu (University
> >> of Minnesota) and Kangjie Lu (University of Minnesota).
> >> 
> >> Because of this, all submissions from this group must be reverted from
> >> the kernel tree and will need to be re-reviewed again to determine if
> >> they actually are a valid fix.  Until that work is complete, remove this
> >> change to ensure that no problems are being introduced into the
> >> codebase.
> >> 
> >> Cc: stable@xxxxxxxxxxxxxxx # v3.16+
> >> Cc: Guoqing Jiang <gqjiang@xxxxxxxx>
> >> Cc: Aditya Pakki <pakki001@xxxxxxx>
> >> Cc: Song Liu <songliubraving@xxxxxx>
> >> Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
> >> ---
> >> drivers/md/raid10.c | 2 --
> >> drivers/md/raid5.c  | 2 --
> >> 2 files changed, 4 deletions(-)
> >> 
> >> diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c
> >> index a9ae7d113492..4fec1cdd4207 100644
> >> --- a/drivers/md/raid10.c
> >> +++ b/drivers/md/raid10.c
> >> @@ -3896,8 +3896,6 @@ static int raid10_run(struct mddev *mddev)
> >> 		set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
> >> 		mddev->sync_thread = md_register_thread(md_do_sync, mddev,
> >> 							"reshape");
> >> -		if (!mddev->sync_thread)
> >> -			goto out_free_conf;
> >> 	}
> >> 
> >> 	return 0;
> >> diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
> >> index 5d57a5bd171f..9b2bd50beee7 100644
> >> --- a/drivers/md/raid5.c
> >> +++ b/drivers/md/raid5.c
> >> @@ -7677,8 +7677,6 @@ static int raid5_run(struct mddev *mddev)
> >> 		set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
> >> 		mddev->sync_thread = md_register_thread(md_do_sync, mddev,
> >> 							"reshape");
> >> -		if (!mddev->sync_thread)
> >> -			goto abort;
> >> 	}
> >> 
> >> 	/* Ok, everything is just fine now */
> >> -- 
> >> 2.31.1
> >> 
> > 
> > These changes look ok, but the error handling logic seems to be freeing
> > the incorrect thread, not the one that these functions create.  That's
> > independant of this change, but seems odd.  If someone cares about it,
> > it should probably be looked at, or if correct, a comment would be nice
> > as it's really confusing.
> 
> I don't think this is confusing. raid[5|10]_run() creates two threads:
> first mddev->thread, then mddev->sync_thread. If we fail to create the
> second thread (sync_thread), we free the first thread (mddev->thread) in 
> the error handling logic. 

Look a bit lower, in "abort:" you do clean up "->thread", but never
"->sync_thread"  You can get to "abort:" after sync_thread is properly
registered.

That is what I was trying to say above, as it's not obvious because
"->thread" was created in a different function, setup_conf(), to see it
be freed here in the error path of raid[5|10]_run() wasn't all that
clear to me.

Anyway, these are code paths that are obviously never hit, so it's
probably not a big deal.

thanks,

greg k-h