[Cluster-devel] [PATCH v2 6/6] gfs2: Fix mmap + page fault deadlocks for direct I/O

[Andreas Gruenbacher]

Direct I/O differs from buffered I/O in that it uses bio_iov_iter_get_pages for
grabbing page references and for manually faulting in pages instead of
triggering actual page faults.  For disabling these manual page faults, it's
not enough to call pagefault_disable(); instead, we use the new
ITER_FLAG_FAST_ONLY flag for telling iomap_dio_rw to stop faulting pages in for
us.

Signed-off-by: Andreas Gruenbacher <agruenba at redhat.com>
---
 fs/gfs2/file.c | 35 ++++++++++++++++++++++++++++++++++-
 1 file changed, 34 insertions(+), 1 deletion(-)

diff --git a/fs/gfs2/file.c b/fs/gfs2/file.c
index 99df7934b4d8..6feb857a8a1c 100644
--- a/fs/gfs2/file.c
+++ b/fs/gfs2/file.c
@@ -763,21 +763,42 @@ static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to,
 	struct file *file = iocb->ki_filp;
 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
 	size_t count = iov_iter_count(to);
+	size_t written = 0;
 	ssize_t ret;
 
+	/*
+	 * In this function, we disable page faults when whe're holding the
+	 * inode glock while doing I/O.  If a page fault occurs, we drop the
+	 * inode glock, fault in the pages manually, and then we retry.  Other
+	 * than in gfs2_file_read_iter, iomap_dio_rw can trigger implicit as
+	 * well as manual page faults, and we need to disable both kinds
+	 * separately.
+	 */
+
 	if (!count)
 		return 0; /* skip atime */
 
 	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
+retry:
 	ret = gfs2_glock_nq(gh);
 	if (ret)
 		goto out_uninit;
 
+	pagefault_disable();
+	to->type |= ITER_FLAG_FAST_ONLY;
 	ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL, 0);
+	to->type &= ~ITER_FLAG_FAST_ONLY;
+	pagefault_enable();
+
 	gfs2_glock_dq(gh);
+	if (ret > 0)
+		written += ret;
+	if (unlikely(iov_iter_count(to) && (ret > 0 || ret == -EFAULT)) &&
+	    fault_in_iov_iter(to))
+		goto retry;
 out_uninit:
 	gfs2_holder_uninit(gh);
-	return ret;
+	return written ? written : ret;
 }
 
 static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from,
@@ -790,6 +811,12 @@ static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from,
 	loff_t offset = iocb->ki_pos;
 	ssize_t ret;
 
+	/*
+	 * In this function, we disable page faults when whe're holding the
+	 * inode glock while doing I/O.  If a page fault occurs, we drop the
+	 * inode glock, fault in the pages manually, and then we retry.
+	 */
+
 	/*
 	 * Deferred lock, even if its a write, since we do no allocation on
 	 * this path. All we need to change is the atime, and this lock mode
@@ -799,6 +826,7 @@ static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from,
 	 * VFS does.
 	 */
 	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
+retry:
 	ret = gfs2_glock_nq(gh);
 	if (ret)
 		goto out_uninit;
@@ -807,11 +835,16 @@ static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from,
 	if (offset + len > i_size_read(&ip->i_inode))
 		goto out;
 
+	from->type |= ITER_FLAG_FAST_ONLY;
 	ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL, 0);
+	from->type &= ~ITER_FLAG_FAST_ONLY;
+
 	if (ret == -ENOTBLK)
 		ret = 0;
 out:
 	gfs2_glock_dq(gh);
+	if (unlikely(ret == -EFAULT) && fault_in_iov_iter(from))
+		goto retry;
 out_uninit:
 	gfs2_holder_uninit(gh);
 	return ret;
-- 
2.26.3