More GFP_NOFS removals

A few more conversions to less strict allocation flags, calls originating from an ioctl and with no chance of getting back with a lock already held. For 4.11.

The following changes since commit 2e94257a41fece589dc4bd737c70dc770e9040ef:

btrfs: Better csum error message for data csum mismatch (2017-02-10 18:04:51 +0100)

are available in the git repository at:

git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux.git dev/less-nofs

for you to fetch changes up to 8d310939a7d232fc71524fdf83f5a8fdbd5eb31c:

btrfs: use GFP_KERNEL in btrfs_add/del_qgroup_relation (2017-02-13 13:11:29 +0100)

----------------------------------------------------------------
David Sterba (4):
btrfs: use GFP_KERNEL in create_snapshot
btrfs: use GFP_KERNEL in btrfs_read_qgroup_config
btrfs: use GFP_KERNEL in btrfs_quota_enable
btrfs: use GFP_KERNEL in btrfs_add/del_qgroup_relation

 fs/btrfs/ioctl.c  | 4 ++--
 fs/btrfs/qgroup.c | 8 ++++----
 2 files changed, 6 insertions(+), 6 deletions(-)

We don't need to use GFP_NOFS here as this is called from ioctls an the only lock held is the subvol_sem, which is of a high level and protects creation/renames/deletion and is never held in the writeout paths.

---
 fs/btrfs/ioctl.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

index 77f93a1e65c7..f65ace58dd2c 100644
@@ -670,12 +670,12 @@ static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 	if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 		return -EINVAL;
 
-	pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS);
+	pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 	if (!pending_snapshot)
 		return -ENOMEM;
 
 	pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
-			GFP_NOFS);
+			GFP_KERNEL);
 	pending_snapshot->path = btrfs_alloc_path();
 	if (!pending_snapshot->root_item || !pending_snapshot->path) {
 		ret = -ENOMEM;

2.10.1

We don't need to use GFP_NOFS here as this is called from ioctls an the only lock held is the subvol_sem, which is of a high level and protects creation/renames/deletion and is never held in the writeout paths.

---
 fs/btrfs/qgroup.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

index cdd0a16bf469..4759fd46cfb5 100644
@@ -876,7 +876,7 @@ int btrfs_quota_enable(struct btrfs_trans_handle *trans,
 		goto out;
 	}
 
-	fs_info->qgroup_ulist = ulist_alloc(GFP_NOFS);
+	fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
 	if (!fs_info->qgroup_ulist) {
 		ret = -ENOMEM;
 		goto out;

2.10.1

The qgroup config is read during mount, we do not have to use NOFS.

---
 fs/btrfs/qgroup.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

index 8496dbf3f38b..cdd0a16bf469 100644
@@ -319,7 +319,7 @@ int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
 	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
 		return 0;
 
-	fs_info->qgroup_ulist = ulist_alloc(GFP_NOFS);
+	fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
 	if (!fs_info->qgroup_ulist) {
 		ret = -ENOMEM;
 		goto out;

2.10.1

Qgroup relations are added/deleted from ioctl, we hold the high level qgroup lock, no deadlocks or recursion from the allocation possible here.

---
 fs/btrfs/qgroup.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

index 4759fd46cfb5..d97353440a70 100644
@@ -1174,7 +1174,7 @@ int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
 	if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
 		return -EINVAL;
 
-	tmp = ulist_alloc(GFP_NOFS);
+	tmp = ulist_alloc(GFP_KERNEL);
 	if (!tmp)
 		return -ENOMEM;
 

@@ -1234,7 +1234,7 @@ int __del_qgroup_relation(struct btrfs_trans_handle *trans,
 	int ret = 0;
 	int err;
 
-	tmp = ulist_alloc(GFP_NOFS);
+	tmp = ulist_alloc(GFP_KERNEL);
 	if (!tmp)
 		return -ENOMEM;
 

2.10.1