On Wed, Feb 05, 2020 at 08:49:16AM +0900, Naohiro Aota wrote:
> On Tue, Feb 04, 2020 at 07:42:29AM -0800, Darrick J. Wong wrote:
> > On Tue, Feb 04, 2020 at 06:59:43PM +0900, Naohiro Aota wrote:
> > > claim_swapfile() currently keeps the inode locked when it is successful, or
> > > the file is already swapfile (with -EBUSY). And, on the other error cases,
> > > it does not lock the inode.
> > >
> > > This inconsistency of the lock state and return value is quite confusing
> > > and actually causing a bad unlock balance as below in the "bad_swap"
> > > section of __do_sys_swapon().
> > >
> > > This commit fixes this issue by unlocking the inode on the error path. It
> > > also reverts blocksize and releases bdev, so that the caller can safely
> > > forget about the inode.
> > >
> > > =====================================
> > > WARNING: bad unlock balance detected!
> > > 5.5.0-rc7+ #176 Not tainted
> > > -------------------------------------
> > > swapon/4294 is trying to release lock (&sb->s_type->i_mutex_key) at:
> > > [<ffffffff8173a6eb>] __do_sys_swapon+0x94b/0x3550
> > > but there are no more locks to release!
> > >
> > > other info that might help us debug this:
> > > no locks held by swapon/4294.
> > >
> > > stack backtrace:
> > > CPU: 5 PID: 4294 Comm: swapon Not tainted 5.5.0-rc7-BTRFS-ZNS+ #176
> > > Hardware name: ASUS All Series/H87-PRO, BIOS 2102 07/29/2014
> > > Call Trace:
> > > dump_stack+0xa1/0xea
> > > ? __do_sys_swapon+0x94b/0x3550
> > > print_unlock_imbalance_bug.cold+0x114/0x123
> > > ? __do_sys_swapon+0x94b/0x3550
> > > lock_release+0x562/0xed0
> > > ? kvfree+0x31/0x40
> > > ? lock_downgrade+0x770/0x770
> > > ? kvfree+0x31/0x40
> > > ? rcu_read_lock_sched_held+0xa1/0xd0
> > > ? rcu_read_lock_bh_held+0xb0/0xb0
> > > up_write+0x2d/0x490
> > > ? kfree+0x293/0x2f0
> > > __do_sys_swapon+0x94b/0x3550
> > > ? putname+0xb0/0xf0
> > > ? kmem_cache_free+0x2e7/0x370
> > > ? do_sys_open+0x184/0x3e0
> > > ? generic_max_swapfile_size+0x40/0x40
> > > ? do_syscall_64+0x27/0x4b0
> > > ? entry_SYSCALL_64_after_hwframe+0x49/0xbe
> > > ? lockdep_hardirqs_on+0x38c/0x590
> > > __x64_sys_swapon+0x54/0x80
> > > do_syscall_64+0xa4/0x4b0
> > > entry_SYSCALL_64_after_hwframe+0x49/0xbe
> > > RIP: 0033:0x7f15da0a0dc7
> > >
> > > Fixes: 1638045c3677 ("mm: set S_SWAPFILE on blockdev swap devices")
> > > Signed-off-by: Naohiro Aota <naohiro.aota@xxxxxxx>
> > > ---
> > > mm/swapfile.c | 29 ++++++++++++++++++++++-------
> > > 1 file changed, 22 insertions(+), 7 deletions(-)
> > >
> > > diff --git a/mm/swapfile.c b/mm/swapfile.c
> > > index bb3261d45b6a..dd5d7fa42282 100644
> > > --- a/mm/swapfile.c
> > > +++ b/mm/swapfile.c
> > > @@ -2886,24 +2886,37 @@ static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
> > > p->old_block_size = block_size(p->bdev);
> > > error = set_blocksize(p->bdev, PAGE_SIZE);
> > > if (error < 0)
> > > - return error;
> > > + goto err;
> > > /*
> > > * Zoned block devices contain zones that have a sequential
> > > * write only restriction. Hence zoned block devices are not
> > > * suitable for swapping. Disallow them here.
> > > */
> > > - if (blk_queue_is_zoned(p->bdev->bd_queue))
> > > - return -EINVAL;
> > > + if (blk_queue_is_zoned(p->bdev->bd_queue)) {
> > > + error = -EINVAL;
> > > + goto err;
> > > + }
> > > p->flags |= SWP_BLKDEV;
> > > } else if (S_ISREG(inode->i_mode)) {
> > > p->bdev = inode->i_sb->s_bdev;
> > > }
> > >
> > > inode_lock(inode);
> > > - if (IS_SWAPFILE(inode))
> > > - return -EBUSY;
> > > + if (IS_SWAPFILE(inode)) {
> > > + inode_unlock(inode);
> > > + error = -EBUSY;
> > > + goto err;
> > > + }
> > >
> > > return 0;
> > > +
> > > +err:
> > > + if (S_ISBLK(inode->i_mode)) {
> > > + set_blocksize(p->bdev, p->old_block_size);
> > > + blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
> > > + }
> > > +
> > > + return error;
> > > }
> > >
> > >
> > > @@ -3157,10 +3170,12 @@ SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
> > > mapping = swap_file->f_mapping;
> > > inode = mapping->host;
> > >
> > > - /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */
> > > + /* do inode_lock(inode); */
> >
> > What if we made this function responsible for calling inode_lock (and
> > unlock) instead of splitting it between sys_swapon and claim_swapfile?
>
> I think we cannot take inode_lock before claim_swapfile() because we can
> have circular locking dependency as:
>
> claim_swapfile()
> -> blkdev_get() -> __blkdev_get()
> -> mutex_lock(&bdev->bd_mutex)
> -> bd_set_size()
> -> inode_lock(&bdev->bd_inode);
Ah, good point. Thank you for doing the research on that. :)
> So, one thing we can do is to move inode_lock() and "if (IS_SWAPFILE(..))
> ..." out of claim_swapfile(). In this case, the "bad_swap" section must
> check if "inode_is_locked" to call "inode_unlock".
I think I wouldn't rely on inode_is_locked and structure the error
escape as follows:
err = claim_swapfile()
if (err)
goto bad_swap;
inode_lock()
if (IS_SWAPFILE)
goto unlock_swap;
other_stuff()
unlock_swap:
inode_unlock()
bad_swap:
fput()
since that's how we (well, XFS anyway :)) tend to do it.
--D
> >
> > --D
> >
> > > error = claim_swapfile(p, inode);
> > > - if (unlikely(error))
> > > + if (unlikely(error)) {
> > > + inode = NULL;
> > > goto bad_swap;
> > > + }
> > >
> > > /*
> > > * Read the swap header.
> > > --
> > > 2.25.0
> > >
