Btrfs: fix deadlock on metadata reservation when evicting a inode

When I ran the xfstests, I found the test tasks was blocked on meta-data
reservation.

By debugging, I found the reason of this bug:
   start transaction
        |
	v
   reserve meta-data space
	|
	v
   flush delay allocation -> iput inode -> evict inode
	^					|
	|					v
   wait for delay allocation flush <- reserve meta-data space

And besides that, the flush on evicting inode will block the thread, which
is reclaiming the memory, and make oom happen easily.

Fix this bug by skipping the flush step when evicting inode.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
(cherry picked from commit aa38a711a893accf5b5192f3d705a120deaa81e0)

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 3d546d73a1049bff39a829afa8b2a748a4e72eeb..a2a1f94492e4f1a9e4442cdcc9b08352eb4bec7a 100644 (file)
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -2369,6 +2369,9 @@ int btrfs_block_rsv_check(struct btrfs_root *root,
 int btrfs_block_rsv_refill(struct btrfs_root *root,
                          struct btrfs_block_rsv *block_rsv,
                          u64 min_reserved);
+int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
+                                  struct btrfs_block_rsv *block_rsv,
+                                  u64 min_reserved);
 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
                            struct btrfs_block_rsv *dst_rsv,
                            u64 num_bytes);

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 5d86877f10e1ca90074a71ef50c9d051730986c3..b7e5f6898d07c00cadef581ccefb967acc13dcae 100644 (file)
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -3887,9 +3887,9 @@ int btrfs_block_rsv_check(struct btrfs_root *root,
        return ret;
 }
 
-int btrfs_block_rsv_refill(struct btrfs_root *root,
-                         struct btrfs_block_rsv *block_rsv,
-                         u64 min_reserved)
+static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
+                                          struct btrfs_block_rsv *block_rsv,
+                                          u64 min_reserved, int flush)
 {
        u64 num_bytes = 0;
        int ret = -ENOSPC;
@@ -3908,7 +3908,7 @@ int btrfs_block_rsv_refill(struct btrfs_root *root,
        if (!ret)
                return 0;
 
-       ret = reserve_metadata_bytes(root, block_rsv, num_bytes, 1);
+       ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
        if (!ret) {
                block_rsv_add_bytes(block_rsv, num_bytes, 0);
                return 0;
@@ -3917,6 +3917,20 @@ int btrfs_block_rsv_refill(struct btrfs_root *root,
        return ret;
 }
 
+int btrfs_block_rsv_refill(struct btrfs_root *root,
+                          struct btrfs_block_rsv *block_rsv,
+                          u64 min_reserved)
+{
+       return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
+}
+
+int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
+                                  struct btrfs_block_rsv *block_rsv,
+                                  u64 min_reserved)
+{
+       return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
+}
+
 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
                            struct btrfs_block_rsv *dst_rsv,
                            u64 num_bytes)

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index c78b3a9f087e5d7757d07ea79503fafc8db9d167..dfa540ea8f44e61ac164ebe9fb083b6a7831afe5 100644 (file)
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -3490,7 +3490,7 @@ void btrfs_evict_inode(struct inode *inode)
         * doing the truncate.
         */
        while (1) {
-               ret = btrfs_block_rsv_refill(root, rsv, min_size);
+               ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
 
                /*
                 * Try and steal from the global reserve since we will