btrfs: remove dead create_space_info calls

From: Jeff Mahoney <[email protected]>

Any time the first block group of a new type is created, we add a new kobject to sysfs to hold the attributes for that type. Kobject-internal allocations always use GFP_KERNEL, making them prone to fs-reclaim races. While it appears as if this can occur any time a block group is created, the only times the first block group of a new type can be created in memory is at mount and when we create the first new block group during raid conversion.

This patch adds a new list to track pending kobject additions and then handles them after we do chunk relocation. Between relocating the target chunk (or forcing allocation of a new chunk in the case of data) and removing the old chunk, we're in a safe place for fs-reclaim to occur. We're holding the volume mutex, which is already held across page faults, and the delete_unused_bgs_mutex, which will only stall the cleaner thread.

Signed-off-by Jeff Mahoney <[email protected]>

---
 fs/btrfs/ctree.h       |  6 ++++-
 fs/btrfs/disk-io.c     |  2 ++
 fs/btrfs/extent-tree.c | 59 +++++++++++++++++++++++++++++++++++---------------
 fs/btrfs/sysfs.c       |  2 +-
 fs/btrfs/volumes.c     | 12 ++++++++++
 5 files changed, 61 insertions(+), 20 deletions(-)

fs/btrfs/ctree.h +5 -1

index ffbb05aa66fa..75dbdf1bbead 100644
@@ -381,8 +381,9 @@ struct btrfs_dev_replace {
 
 /* For raid type sysfs entries */
 struct raid_kobject {
-	int raid_type;
+	u64 flags;
 	struct kobject kobj;
+	struct list_head list;
 };
 
 struct btrfs_space_info {

@@ -938,6 +939,8 @@ struct btrfs_fs_info {
 	int thread_pool_size;
 
 	struct kobject *space_info_kobj;
+	struct list_head pending_raid_kobjs;
+	spinlock_t pending_raid_kobjs_lock; /* uncontended */
 
 	u64 total_pinned;
 

@@ -2688,6 +2691,7 @@ int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr);
 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
 			   struct btrfs_fs_info *fs_info, u64 bytes_used,
 			   u64 type, u64 chunk_offset, u64 size);
+void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info);
 struct btrfs_trans_handle *btrfs_start_trans_remove_block_group(
 				struct btrfs_fs_info *fs_info,
 				const u64 chunk_offset);

fs/btrfs/disk-io.c +2 -0

index eb6bb3169a9e..d5e1c2ff71ff 100644
@@ -2447,6 +2447,8 @@ int open_ctree(struct super_block *sb,
 	INIT_LIST_HEAD(&fs_info->delayed_iputs);
 	INIT_LIST_HEAD(&fs_info->delalloc_roots);
 	INIT_LIST_HEAD(&fs_info->caching_block_groups);
+	INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
+	spin_lock_init(&fs_info->pending_raid_kobjs_lock);
 	spin_lock_init(&fs_info->delalloc_root_lock);
 	spin_lock_init(&fs_info->trans_lock);
 	spin_lock_init(&fs_info->fs_roots_radix_lock);

fs/btrfs/extent-tree.c +6 -0

index 0e9a21230217..bb5368faa937 100644
@@ -9908,9 +9908,39 @@ int btrfs_free_block_groups(struct btrfs_fs_info *info)
 	return 0;
 }
 
+/* link_block_group will queue up kobjects to add when we're reclaim-safe */
+void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info)
+{
+	struct btrfs_space_info *space_info;
+	struct raid_kobject *rkobj;
+	LIST_HEAD(list);

fs/btrfs/extent-tree.c +25 -18

+
+		ret = kobject_add(&rkobj->kobj, &space_info->kobj,
+				  "%s", get_raid_name(index));
+		if (ret) {
+			kobject_put(&rkobj->kobj);
+			break;
+		}
+	}
+	if (ret)
+		btrfs_warn(fs_info,
+			   "failed to add kobject for block cache, ignoring");
+}
+
 static void link_block_group(struct btrfs_block_group_cache *cache)
 {
 	struct btrfs_space_info *space_info = cache->space_info;
+	struct btrfs_fs_info *fs_info = cache->fs_info;
 	int index = get_block_group_index(cache);
 	bool first = false;
 

@@ -9921,27 +9951,19 @@ static void link_block_group(struct btrfs_block_group_cache *cache)
 	up_write(&space_info->groups_sem);
 
 	if (first) {
-		struct raid_kobject *rkobj;
-		int ret;
-
-		rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
-		if (!rkobj)
-			goto out_err;
-		rkobj->raid_type = index;
-		kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
-		ret = kobject_add(&rkobj->kobj, &space_info->kobj,
-				  "%s", get_raid_name(index));
-		if (ret) {
-			kobject_put(&rkobj->kobj);
-			goto out_err;
+		struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
+		if (!rkobj) {
+			btrfs_warn(cache->fs_info, "couldn't alloc memory for raid level kobject");
+			return;
 		}
+		rkobj->flags = cache->flags;
+		kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
+
+		spin_lock(&fs_info->pending_raid_kobjs_lock);
+		list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs);
+		spin_unlock(&fs_info->pending_raid_kobjs_lock);
 		space_info->block_group_kobjs[index] = &rkobj->kobj;
 	}
-
-	return;
-out_err:
-	btrfs_warn(cache->fs_info,
-		   "failed to add kobject for block cache, ignoring");
 }
 
 static struct btrfs_block_group_cache *

@@ -10157,6 +10179,7 @@ int btrfs_read_block_groups(struct btrfs_fs_info *info)
 			inc_block_group_ro(cache, 1);
 	}
 
+	btrfs_add_raid_kobjects(info);
 	init_global_block_rsv(info);
 	ret = 0;
 error:

fs/btrfs/sysfs.c +1 -1

index a8bafed931f4..b975fda719e7 100644
@@ -272,7 +272,7 @@ static ssize_t raid_bytes_show(struct kobject *kobj,
 {
 	struct btrfs_space_info *sinfo = to_space_info(kobj->parent);
 	struct btrfs_block_group_cache *block_group;
-	int index = to_raid_kobj(kobj)->raid_type;
+	int index = __get_raid_index(to_raid_kobj(kobj)->flags);
 	u64 val = 0;
 
 	down_read(&sinfo->groups_sem);

fs/btrfs/volumes.c +12 -0

index b2d05c6b1c56..b0cffce168cf 100644
@@ -2997,6 +2997,16 @@ static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
 	if (ret)
 		return ret;
 
+	/*
+	 * We add the kobjects here (and after forcing data chunk creation)
+	 * since relocation is the only place we'll create chunks of a new
+	 * type at runtime.  The only place where we'll remove the last
+	 * chunk of a type is the call immediately below this one.  Even
+	 * so, we're protected against races with the cleaner thread since
+	 * we're covered by the delete_unused_bgs_mutex.
+	 */
+	btrfs_add_raid_kobjects(fs_info);
+
 	trans = btrfs_start_trans_remove_block_group(root->fs_info,
 						     chunk_offset);
 	if (IS_ERR(trans)) {

@@ -3124,6 +3134,8 @@ static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
 			if (ret < 0)
 				return ret;
 
+			btrfs_add_raid_kobjects(fs_info);
+
 			return 1;
 		}
 	}

Quoted 86 lines · On 20.03.2018 21:25, [email protected] wrote:

From: Jeff Mahoney <[email protected]>

Any time the first block group of a new type is created, we add a new kobject to sysfs to hold the attributes for that type. Kobject-internal allocations always use GFP_KERNEL, making them prone to fs-reclaim races. While it appears as if this can occur any time a block group is created, the only times the first block group of a new type can be created in memory is at mount and when we create the first new block group during raid conversion.

This patch adds a new list to track pending kobject additions and then handles them after we do chunk relocation. Between relocating the target chunk (or forcing allocation of a new chunk in the case of data) and removing the old chunk, we're in a safe place for fs-reclaim to occur. We're holding the volume mutex, which is already held across page faults, and the delete_unused_bgs_mutex, which will only stall the cleaner thread.

Signed-off-by Jeff Mahoney <[email protected]>

---
 fs/btrfs/ctree.h       |  6 ++++-
 fs/btrfs/disk-io.c     |  2 ++
 fs/btrfs/extent-tree.c | 59 +++++++++++++++++++++++++++++++++++---------------
 fs/btrfs/sysfs.c       |  2 +-
 fs/btrfs/volumes.c     | 12 ++++++++++
 5 files changed, 61 insertions(+), 20 deletions(-)

fs/btrfs/ctree.h +5 -1

index ffbb05aa66fa..75dbdf1bbead 100644
@@ -381,8 +381,9 @@ struct btrfs_dev_replace {
 
 /* For raid type sysfs entries */
 struct raid_kobject {
-	int raid_type;
+	u64 flags;
 	struct kobject kobj;
+	struct list_head list;
 };
 
 struct btrfs_space_info {

@@ -938,6 +939,8 @@ struct btrfs_fs_info {
 	int thread_pool_size;
 
 	struct kobject *space_info_kobj;
+	struct list_head pending_raid_kobjs;
+	spinlock_t pending_raid_kobjs_lock; /* uncontended */
 
 	u64 total_pinned;
 

@@ -2688,6 +2691,7 @@ int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr);
 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
 			   struct btrfs_fs_info *fs_info, u64 bytes_used,
 			   u64 type, u64 chunk_offset, u64 size);
+void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info);
 struct btrfs_trans_handle *btrfs_start_trans_remove_block_group(
 				struct btrfs_fs_info *fs_info,
 				const u64 chunk_offset);

fs/btrfs/disk-io.c +2 -0

index eb6bb3169a9e..d5e1c2ff71ff 100644
@@ -2447,6 +2447,8 @@ int open_ctree(struct super_block *sb,
 	INIT_LIST_HEAD(&fs_info->delayed_iputs);
 	INIT_LIST_HEAD(&fs_info->delalloc_roots);
 	INIT_LIST_HEAD(&fs_info->caching_block_groups);
+	INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
+	spin_lock_init(&fs_info->pending_raid_kobjs_lock);
 	spin_lock_init(&fs_info->delalloc_root_lock);
 	spin_lock_init(&fs_info->trans_lock);
 	spin_lock_init(&fs_info->fs_roots_radix_lock);

fs/btrfs/extent-tree.c +6 -0

index 0e9a21230217..bb5368faa937 100644
@@ -9908,9 +9908,39 @@ int btrfs_free_block_groups(struct btrfs_fs_info *info)
 	return 0;
 }
 
+/* link_block_group will queue up kobjects to add when we're reclaim-safe */
+void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info)
+{
+	struct btrfs_space_info *space_info;
+	struct raid_kobject *rkobj;
+	LIST_HEAD(list);

fs/btrfs/extent-tree.c int btrfs_free_block_groups(struct btrfs_fs_info *info) · line 9926

Quoted 10 lines

fs/btrfs/extent-tree.c +10 -0

+	int index;
+	int ret = 0;
+
+	spin_lock(&fs_info->pending_raid_kobjs_lock);
+	list_splice_init(&fs_info->pending_raid_kobjs, &list);
+	spin_unlock(&fs_info->pending_raid_kobjs_lock);
+
+	list_for_each_entry(rkobj, &list, list) {
+		space_info = __find_space_info(fs_info, rkobj->flags);
+		index = __get_raid_index(rkobj->flags);

This function is no more. It was refactored in 24bc2843edd5 ("btrfs:
Refactor __get_raid_index() to btrfs_bg_flags_to_raid_index()")

So I guess you should use btrfs_bg_flags_to_raid_index instead.

From: Jeff Mahoney <[email protected]>

Since commit 2be12ef79 (btrfs: Separate space_info create/update), we've separated out the creation and updating of the space info structures. That commit was a straightforward refactoring of the two parts of update_space_info, but we can go a step further. Since commits c59021f84 (Btrfs: fix OOPS of empty filesystem after balance) and b742bb82f (Btrfs: Link block groups of different raid types), we know that the space_info structures will be created at mount and there will only ever be, at most, three of them.

This patch cleans out the create_space_info calls after __find_space_info returns NULL since __find_space_info *can't* return NULL.

The initial cause for reviewing this was the kobject_add calls from create_space_info occuring in sites where fs-reclaim wasn't allowed. Now we are certain they occur only early in the mount process and are safe.

Signed-off-by Jeff Mahoney <[email protected]>

---
 fs/btrfs/ctree.h       |  6 +++---
 fs/btrfs/extent-tree.c | 16 ++--------------
 2 files changed, 5 insertions(+), 17 deletions(-)

fs/btrfs/ctree.h +3 -3

index da308774b8a4..ffbb05aa66fa 100644
@@ -952,9 +952,9 @@ struct btrfs_fs_info {
 	struct btrfs_fs_devices *fs_devices;
 
 	/*
-	 * the space_info list is almost entirely read only.  It only changes
-	 * when we add a new raid type to the FS, and that happens
-	 * very rarely.  RCU is used to protect it.
+	 * The space_info list is effectively read only after initial
+	 * setup.  It is populated at mount time and cleaned up after
+	 * all block groups are removed.  RCU is used to protect it.
 	 */
 	struct list_head space_info;
 

fs/btrfs/extent-tree.c +2 -14

index 2d5e963fae88..0e9a21230217 100644
@@ -4603,11 +4603,7 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans,
 		return -ENOSPC;
 
 	space_info = __find_space_info(fs_info, flags);
-	if (!space_info) {
-		ret = create_space_info(fs_info, flags, &space_info);
-		if (ret)
-			return ret;
-	}
+	ASSERT(space_info);
 
 again:
 	spin_lock(&space_info->lock);

@@ -10255,15 +10251,7 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
 	 * with its ->space_info set.
 	 */
 	cache->space_info = __find_space_info(fs_info, cache->flags);
-	if (!cache->space_info) {
-		ret = create_space_info(fs_info, cache->flags,
-				       &cache->space_info);
-		if (ret) {
-			btrfs_remove_free_space_cache(cache);
-			btrfs_put_block_group(cache);
-			return ret;
-		}
-	}
+	ASSERT(cache->space_info);
 
 	ret = btrfs_add_block_group_cache(fs_info, cache);
 	if (ret) {