From: Chris Mason <clm@fb.com>
Date: Fri, 9 Dec 2016 01:55:03 +0000 (-0800)
Subject: Revert "Btrfs: adjust len of writes if following a preallocated extent"
X-Git-Tag: v4.10-rc1~68^2~2
X-Git-Url: https://asedeno.scripts.mit.edu/gitweb/?a=commitdiff_plain;h=7c4c71ac8a72aea595f4cc7e09c2bcc61929c4ac;p=linux.git

Revert "Btrfs: adjust len of writes if following a preallocated extent"

This is exposing an existing deadlock between fsync and AIO.  Until we
have the deadlock fixed, I'm pulling this one out.

This reverts commit a23eaa875f0f1d89eb866b8c9860e78273ff5daf.

Signed-off-by: Chris Mason <clm@fb.com>
---

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index dbcec1ced85a..a713d9d324b0 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -7765,12 +7765,10 @@ static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
 	}
 
 	/*
-	 * this will cow the extent, if em is within [start, len], then
-	 * probably we've found a preallocated/existing extent, let's
-	 * give it a chance to use preallocated space.
+	 * this will cow the extent, reset the len in case we changed
+	 * it above
 	 */
-	len = min_t(u64, bh_result->b_size, em->len - (start - em->start));
-	len = ALIGN(len, fs_info->sectorsize);
+	len = bh_result->b_size;
 	free_extent_map(em);
 	em = btrfs_new_extent_direct(inode, start, len);
 	if (IS_ERR(em)) {