From bced5623e3fc8a47aee58ea2f34cdb3d6439be92 Mon Sep 17 00:00:00 2001
From: Lukas Czerner <lczerner@redhat.com>
Date: Tue, 23 Jan 2018 16:07:11 -0800
Subject: [PATCH 3/3] fs: Avoid invalidation in interrupt context in
 dio_complete()
Git-commit: ffe51f0142a291a957eebb9687cafb15f2b3fc14
Patch-mainline: v4.14-rc6
References: bsc#1073407 bsc#1069135

Currently we try to defer completion of async DIO to the process context
in case there are any mapped pages associated with the inode so that we
can invalidate the pages when the IO completes. However the check is racy
and the pages can be mapped afterwards. If this happens we might end up
calling invalidate_inode_pages2_range() in dio_complete() in interrupt
context which could sleep. This can be reproduced by generic/451.

Fix this by passing the information whether we can or can't invalidate
to the dio_complete(). Thanks Eryu Guan for reporting this and Jan Kara
for suggesting a fix.

Fixes: 332391a9935d ("fs: Fix page cache inconsistency when mixing buffered and AIO DIO")
Reported-by: Eryu Guan <eguan@redhat.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Tested-by: Eryu Guan <eguan@redhat.com>
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Luis R. Rodriguez <mcgrof@suse.com>
---
 fs/direct-io.c | 19 +++++++++++++------
 1 file changed, 13 insertions(+), 6 deletions(-)

diff --git a/fs/direct-io.c b/fs/direct-io.c
index 1d1dc78e9263..80a0e3bf581b 100644
--- a/fs/direct-io.c
+++ b/fs/direct-io.c
@@ -44,6 +44,12 @@
  */
 #define DIO_PAGES	64
 
+/*
+ * Flags for dio_complete()
+ */
+#define DIO_COMPLETE_ASYNC		0x01	/* This is async IO */
+#define DIO_COMPLETE_INVALIDATE		0x02	/* Can invalidate pages */
+
 /*
  * This code generally works in units of "dio_blocks".  A dio_block is
  * somewhere between the hard sector size and the filesystem block size.  it
@@ -225,7 +231,7 @@ static inline struct page *dio_get_page(struct dio *dio,
  * filesystems can use it to hold additional state between get_block calls and
  * dio_complete.
  */
-static ssize_t dio_complete(struct dio *dio, ssize_t ret, bool is_async)
+static ssize_t dio_complete(struct dio *dio, ssize_t ret, unsigned int flags)
 {
 	loff_t offset = dio->iocb->ki_pos;
 	ssize_t transferred = 0;
@@ -278,7 +284,8 @@ static ssize_t dio_complete(struct dio *dio, ssize_t ret, bool is_async)
 	 * one is a pretty crazy thing to do, so we don't support it 100%.  If
 	 * this invalidation fails, tough, the write still worked...
 	 */
-	if (ret > 0 && dio->op == REQ_OP_WRITE &&
+	if (flags & DIO_COMPLETE_INVALIDATE &&
+	    ret > 0 && dio->op == REQ_OP_WRITE &&
 	    dio->inode->i_mapping->nrpages) {
 		err = invalidate_inode_pages2_range(dio->inode->i_mapping,
 					offset >> PAGE_SHIFT,
@@ -289,7 +296,7 @@ static ssize_t dio_complete(struct dio *dio, ssize_t ret, bool is_async)
 	if (!(dio->flags & DIO_SKIP_DIO_COUNT))
 		inode_dio_end(dio->inode);
 
-	if (is_async) {
+	if (flags & DIO_COMPLETE_ASYNC) {
 		/*
 		 * generic_write_sync expects ki_pos to have been updated
 		 * already, but the submission path only does this for
@@ -310,7 +317,7 @@ static void dio_aio_complete_work(struct work_struct *work)
 {
 	struct dio *dio = container_of(work, struct dio, complete_work);
 
-	dio_complete(dio, 0, true);
+	dio_complete(dio, 0, DIO_COMPLETE_ASYNC | DIO_COMPLETE_INVALIDATE);
 }
 
 static blk_status_t dio_bio_complete(struct dio *dio, struct bio *bio);
@@ -352,7 +359,7 @@ static void dio_bio_end_aio(struct bio *bio)
 			queue_work(dio->inode->i_sb->s_dio_done_wq,
 				   &dio->complete_work);
 		} else {
-			dio_complete(dio, 0, true);
+			dio_complete(dio, 0, DIO_COMPLETE_ASYNC);
 		}
 	}
 }
@@ -1358,7 +1365,7 @@ do_blockdev_direct_IO(struct kiocb *iocb, struct inode *inode,
 		dio_await_completion(dio);
 
 	if (drop_refcount(dio) == 0) {
-		retval = dio_complete(dio, retval, false);
+		retval = dio_complete(dio, retval, DIO_COMPLETE_INVALIDATE);
 	} else
 		BUG_ON(retval != -EIOCBQUEUED);
 
-- 
2.15.0