[Linux-cachefs] [PATCH v9 2/2] xfs: avoid transaction reservation recursion

[Dave Chinner]

On Sun, Dec 06, 2020 at 02:40:46PM +0800, Yafang Shao wrote:
> PF_FSTRANS which is used to avoid transaction reservation recursion, is
> dropped since commit 9070733b4efa ("xfs: abstract PF_FSTRANS to
> PF_MEMALLOC_NOFS") and commit 7dea19f9ee63 ("mm: introduce
> memalloc_nofs_{save,restore} API") and replaced by PF_MEMALLOC_NOFS which
> means to avoid filesystem reclaim recursion. That change is subtle.
> Let's take the exmple of the check of WARN_ON_ONCE(current->flags &
> PF_MEMALLOC_NOFS)) to explain why this abstraction from PF_FSTRANS to
> PF_MEMALLOC_NOFS is not proper.
> Below comment is quoted from Dave,
> > It wasn't for memory allocation recursion protection in XFS - it was for
> > transaction reservation recursion protection by something trying to flush
> > data pages while holding a transaction reservation. Doing
> > this could deadlock the journal because the existing reservation
> > could prevent the nested reservation for being able to reserve space
> > in the journal and that is a self-deadlock vector.
> > IOWs, this check is not protecting against memory reclaim recursion
> > bugs at all (that's the previous check [1]). This check is
> > protecting against the filesystem calling writepages directly from a
> > context where it can self-deadlock.
> > So what we are seeing here is that the PF_FSTRANS ->
> > PF_MEMALLOC_NOFS abstraction lost all the actual useful information
> > about what type of error this check was protecting against.
> 
> As a result, we should reintroduce PF_FSTRANS. As current->journal_info
> isn't used in XFS, we can reuse it to indicate whehter the task is in
> fstrans or not, Per Willy. To achieve that, some new helpers are introduce
> in this patch, per Dave:
> - xfs_trans_context_set()
>   Used in xfs_trans_alloc()
> - xfs_trans_context_clear()
>   Used in xfs_trans_commit() and xfs_trans_cancel()
> - xfs_trans_context_active()
>   To check whehter current is in fs transcation or not
> 
> Darrick helped fix the error occurred in xfs/141.[2]
> 
> No obvious error occurred when I run xfstests in my test machine.
> 
> [1]. Below check is to avoid memory reclaim recursion.
> if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
>         PF_MEMALLOC))
>         goto redirty;
> 
> [2]. https://lore.kernel.org/linux-xfs/20201104001649.GN7123@magnolia/
> 
> Cc: Darrick J. Wong <darrick.wong at oracle.com>
> Cc: Matthew Wilcox (Oracle) <willy at infradead.org>
> Cc: Christoph Hellwig <hch at lst.de>
> Cc: Dave Chinner <david at fromorbit.com>
> Cc: Michal Hocko <mhocko at kernel.org>
> Cc: David Howells <dhowells at redhat.com>
> Cc: Jeff Layton <jlayton at redhat.com>
> Signed-off-by: Yafang Shao <laoar.shao at gmail.com>
> ---
>  fs/iomap/buffered-io.c |  7 -------
>  fs/xfs/xfs_aops.c      | 23 +++++++++++++++++++++--
>  fs/xfs/xfs_linux.h     |  4 ----
>  fs/xfs/xfs_trans.c     | 25 +++++++++++++------------
>  fs/xfs/xfs_trans.h     | 23 +++++++++++++++++++++++
>  5 files changed, 57 insertions(+), 25 deletions(-)
> 
> diff --git a/fs/iomap/buffered-io.c b/fs/iomap/buffered-io.c
> index 10cc7979ce38..3c53fa6ce64d 100644
> --- a/fs/iomap/buffered-io.c
> +++ b/fs/iomap/buffered-io.c
> @@ -1458,13 +1458,6 @@ iomap_do_writepage(struct page *page, struct writeback_control *wbc, void *data)
>  			PF_MEMALLOC))
>  		goto redirty;
>  
> -	/*
> -	 * Given that we do not allow direct reclaim to call us, we should
> -	 * never be called in a recursive filesystem reclaim context.
> -	 */
> -	if (WARN_ON_ONCE(current->flags & PF_MEMALLOC_NOFS))
> -		goto redirty;
> -

So this would have given us a warning if something went wrong...

> @@ -568,6 +569,16 @@ xfs_vm_writepage(
>  {
>  	struct xfs_writepage_ctx wpc = { };
>  
> +	/*
> +	 * Given that we do not allow direct reclaim to call us, we should
> +	 * never be called while in a filesystem transaction.
> +	 */
> +	if (xfs_trans_context_active()) {
> +		redirty_page_for_writepage(wbc, page);
> +		unlock_page(page);
> +		return 0;
> +	}
> +
>  	return iomap_writepage(page, wbc, &wpc.ctx, &xfs_writeback_ops);
>  }
>  
> @@ -579,6 +590,14 @@ xfs_vm_writepages(
>  	struct xfs_writepage_ctx wpc = { };
>  
>  	xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
> +
> +	/*
> +	 * Given that we do not allow direct reclaim to call us, we should
> +	 * never be called while in a filesystem transaction.
> +	 */
> +	if (xfs_trans_context_active())
> +		return 0;
> +
>  	return iomap_writepages(mapping, wbc, &wpc.ctx, &xfs_writeback_ops);
>  }

But neither of these will trigger a warning at all, so we won't know
that there's a bug in the code at all. Given this is primarily a "we
have a bug in the code" deadlock avoidance check, we really need the
noisy warnings if these fire...

> diff --git a/fs/xfs/xfs_trans.c b/fs/xfs/xfs_trans.c
> index c94e71f741b6..09ae5c181299 100644
> --- a/fs/xfs/xfs_trans.c
> +++ b/fs/xfs/xfs_trans.c
> @@ -67,6 +67,11 @@ xfs_trans_free(
>  	xfs_extent_busy_sort(&tp->t_busy);
>  	xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false);
>  
> +	/* Detach the transaction from this thread. */
> +	ASSERT(current->journal_info != NULL);
> +	if (current->journal_info == tp)
> +		xfs_trans_context_clear(tp);

>From the context of this patch and the code it is replacing, I have
no idea why this condition could occur, so this needs a comment
explaining when current->journal_info is not equal to the
transaction we are freeing.

> +
>  	trace_xfs_trans_free(tp, _RET_IP_);
>  	if (!(tp->t_flags & XFS_TRANS_NO_WRITECOUNT))
>  		sb_end_intwrite(tp->t_mountp->m_super);
> @@ -119,7 +124,11 @@ xfs_trans_dup(
>  
>  	ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
>  	tp->t_rtx_res = tp->t_rtx_res_used;
> +
> +	/* Associate the new transaction with this thread. */
> +	ASSERT(current->journal_info == tp);
>  	ntp->t_pflags = tp->t_pflags;
> +	current->journal_info = ntp;

Why is this open coded and not in a helper like all the
current->journal_info manipulations? Something like
xfs_trans_context_swap(tp, ntp) with a comment explaining that it is
used to transfer the transaction context when rolling a permanent
transaction?

>  
>  	/* move deferred ops over to the new tp */
>  	xfs_defer_move(ntp, tp);
> @@ -153,8 +162,6 @@ xfs_trans_reserve(
>  	int			error = 0;
>  	bool			rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
>  
> -	/* Mark this thread as being in a transaction */
> -	current_set_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
>  
>  	/*
>  	 * Attempt to reserve the needed disk blocks by decrementing
> @@ -163,10 +170,8 @@ xfs_trans_reserve(
>  	 */
>  	if (blocks > 0) {
>  		error = xfs_mod_fdblocks(mp, -((int64_t)blocks), rsvd);
> -		if (error != 0) {
> -			current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
> +		if (error != 0)
>  			return -ENOSPC;
> -		}
>  		tp->t_blk_res += blocks;
>  	}
>  
> @@ -241,8 +246,6 @@ xfs_trans_reserve(
>  		tp->t_blk_res = 0;
>  	}
>  
> -	current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
> -
>  	return error;
>  }
>  
> @@ -284,6 +287,8 @@ xfs_trans_alloc(
>  	INIT_LIST_HEAD(&tp->t_dfops);
>  	tp->t_firstblock = NULLFSBLOCK;
>  
> +	/* Mark this thread as being in a transaction */
> +	xfs_trans_context_set(tp);
>  	error = xfs_trans_reserve(tp, resp, blocks, rtextents);
>  	if (error) {
>  		xfs_trans_cancel(tp);

This refactoring should probably be a separate patch, done first.

> @@ -878,7 +883,6 @@ __xfs_trans_commit(
>  
>  	xfs_log_commit_cil(mp, tp, &commit_lsn, regrant);
>  
> -	current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
>  	xfs_trans_free(tp);
>  
>  	/*
> @@ -910,7 +914,7 @@ __xfs_trans_commit(
>  			xfs_log_ticket_ungrant(mp->m_log, tp->t_ticket);
>  		tp->t_ticket = NULL;
>  	}
> -	current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
> +
>  	xfs_trans_free_items(tp, !!error);
>  	xfs_trans_free(tp);
>  
> @@ -970,9 +974,6 @@ xfs_trans_cancel(
>  		tp->t_ticket = NULL;
>  	}
>  
> -	/* mark this thread as no longer being in a transaction */
> -	current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS);
> -
>  	xfs_trans_free_items(tp, dirty);
>  	xfs_trans_free(tp);
>  }

And moving current_restore_flags_nested() into xfs_trans_free()
should also probably be a separate patch.

That way this patch is simply changing all the flags manipulations
to use the new wrappers, and not a mix of refactoring and API
rework...

Cheers,

Dave.
-- 
Dave Chinner
david at fromorbit.com