Skip to contentSkip to conversation

f2fs: fix race condition in between free nid allocator/initializer

PATCH Superseded Chao Yu 2 participants 3 messages Mar 21, 2017 – Mar 22, 2017 F2FS file system

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Chao Yu

author
#0 lore raw

PATCH v2 1/1 f2fs: fix race condition in between free nid allocator/initializer

In below concurrent case, allocated nid can be loaded into free nid cache and be allocated again.

Thread A Thread B
- f2fs_create
- f2fs_new_inode
- alloc_nid
- __insert_nid_to_list(ALLOC_NID_LIST)
- f2fs_balance_fs_bg
- build_free_nids
- __build_free_nids
- scan_nat_page
- add_free_nid
- __lookup_nat_cache
- f2fs_add_link
- init_inode_metadata
- new_inode_page
- new_node_page
- set_node_addr
- alloc_nid_done
- __remove_nid_from_list(ALLOC_NID_LIST)
- __insert_nid_to_list(FREE_NID_LIST)

This patch uses build_lock covering free nid allocation and initialization to avoid this race condition.

---
v2: fix deadlock in remove_free_nid.

 fs/f2fs/node.c | 16 +++++++++++++---
 1 file changed, 13 insertions(+), 3 deletions(-)
fs/f2fs/node.c +13 -3
index 4f89478be683..d12b00dc9d26 100644
@@ -22,8 +22,6 @@
 #include "trace.h"
 #include <trace/events/f2fs.h>
 
-#define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
-
 static struct kmem_cache *nat_entry_slab;
 static struct kmem_cache *free_nid_slab;
 static struct kmem_cache *nat_entry_set_slab;
@@ -2000,15 +1998,18 @@ bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
 		return false;
 	}
 #endif
+
+	mutex_lock(&nm_i->build_lock);
 	spin_lock(&nm_i->nid_list_lock);
 
 	if (unlikely(nm_i->available_nids == 0)) {
 		spin_unlock(&nm_i->nid_list_lock);
+		mutex_unlock(&nm_i->build_lock);
 		return false;
 	}
 
 	/* We should not use stale free nids created by build_free_nids */
-	if (nm_i->nid_cnt[FREE_NID_LIST] && !on_build_free_nids(nm_i)) {
+	if (nm_i->nid_cnt[FREE_NID_LIST]) {
 		f2fs_bug_on(sbi, list_empty(&nm_i->nid_list[FREE_NID_LIST]));
 		i = list_first_entry(&nm_i->nid_list[FREE_NID_LIST],
 					struct free_nid, list);
@@ -2022,9 +2023,11 @@ bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
 		update_free_nid_bitmap(sbi, *nid, false, false);
 
 		spin_unlock(&nm_i->nid_list_lock);
+		mutex_unlock(&nm_i->build_lock);
 		return true;
 	}
 	spin_unlock(&nm_i->nid_list_lock);
+	mutex_unlock(&nm_i->build_lock);
 
 	/* Let's scan nat pages and its caches to get free nids */
 	build_free_nids(sbi, true, false);
@@ -2039,11 +2042,13 @@ void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
 	struct f2fs_nm_info *nm_i = NM_I(sbi);
 	struct free_nid *i;
 
+	mutex_lock(&nm_i->build_lock);
 	spin_lock(&nm_i->nid_list_lock);
 	i = __lookup_free_nid_list(nm_i, nid);
 	f2fs_bug_on(sbi, !i);
 	__remove_nid_from_list(sbi, i, ALLOC_NID_LIST, false);
 	spin_unlock(&nm_i->nid_list_lock);
+	mutex_unlock(&nm_i->build_lock);
 
 	kmem_cache_free(free_nid_slab, i);
 }
@@ -2060,6 +2065,7 @@ void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
 	if (!nid)
 		return;
 
+	mutex_lock(&nm_i->build_lock);
 	spin_lock(&nm_i->nid_list_lock);
 	i = __lookup_free_nid_list(nm_i, nid);
 	f2fs_bug_on(sbi, !i);
@@ -2078,6 +2084,7 @@ void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
 	update_free_nid_bitmap(sbi, nid, true, false);
 
 	spin_unlock(&nm_i->nid_list_lock);
+	mutex_unlock(&nm_i->build_lock);
 
 	if (need_free)
 		kmem_cache_free(free_nid_slab, i);
@@ -2445,6 +2452,7 @@ void flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 	if (!nm_i->dirty_nat_cnt)
 		return;
 
+	mutex_lock(&nm_i->build_lock);
 	down_write(&nm_i->nat_tree_lock);
 
 	/*
@@ -2470,6 +2478,8 @@ void flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 		__flush_nat_entry_set(sbi, set, cpc);
 
 	up_write(&nm_i->nat_tree_lock);
+	mutex_unlock(&nm_i->build_lock);
+
 	/* Allow dirty nats by node block allocation in write_begin */
 }
 
Signature · 6 lines
2.8.2.295.g3f1c1d0


------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
maintainer
#1 lore raw

Hi Chao,

Quoted 23 lines · Chao Yu

In below concurrent case, allocated nid can be loaded into free nid cache and be allocated again.

Thread A Thread B
- f2fs_create
- f2fs_new_inode
- alloc_nid
- __insert_nid_to_list(ALLOC_NID_LIST)
- f2fs_balance_fs_bg
- build_free_nids
- __build_free_nids
- scan_nat_page
- add_free_nid
- __lookup_nat_cache
- f2fs_add_link
- init_inode_metadata
- new_inode_page
- new_node_page
- set_node_addr
- alloc_nid_done
- __remove_nid_from_list(ALLOC_NID_LIST)
- __insert_nid_to_list(FREE_NID_LIST)

Quoted 2 lines

This patch uses build_lock covering free nid allocation and initialization to avoid this race condition.

I have a concern about mutex contention.
How about this?

---
 fs/f2fs/node.c | 60 ++++++++++++++++++++++++++++++++++++++++------------------
 1 file changed, 42 insertions(+), 18 deletions(-)
fs/f2fs/node.c +42 -18
index 347aa30ef0cf..25bc47087f91 100644
@@ -1759,40 +1759,64 @@ static void __remove_nid_from_list(struct f2fs_sb_info *sbi,
 static bool add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
 {
 	struct f2fs_nm_info *nm_i = NM_I(sbi);
-	struct free_nid *i;
+	struct free_nid *i, *e;
 	struct nat_entry *ne;
-	int err;
+	int err = -EINVAL;
+	bool ret = false;
 
 	/* 0 nid should not be used */
 	if (unlikely(nid == 0))
 		return false;
 
-	if (build) {
-		/* do not add allocated nids */
-		ne = __lookup_nat_cache(nm_i, nid);
-		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
-				nat_get_blkaddr(ne) != NULL_ADDR))
-			return false;
-	}
-
 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
 	i->nid = nid;
 	i->state = NID_NEW;
 
-	if (radix_tree_preload(GFP_NOFS)) {
-		kmem_cache_free(free_nid_slab, i);
-		return true;
-	}
+	if (radix_tree_preload(GFP_NOFS))
+		goto err;
 
 	spin_lock(&nm_i->nid_list_lock);
+
+	if (build) {
+		/*
+		 *   Thread A             Thread B
+		 *  - f2fs_create
+		 *   - f2fs_new_inode
+		 *    - alloc_nid
+		 *     - __insert_nid_to_list(ALLOC_NID_LIST)
+		 *                     - f2fs_balance_fs_bg
+		 *                      - build_free_nids
+		 *                       - __build_free_nids
+		 *                        - scan_nat_page
+		 *                         - add_free_nid
+		 *                          - __lookup_nat_cache
+		 *  - f2fs_add_link
+		 *   - init_inode_metadata
+		 *    - new_inode_page
+		 *     - new_node_page
+		 *      - set_node_addr
+		 *  - alloc_nid_done
+		 *   - __remove_nid_from_list(ALLOC_NID_LIST)
+		 *                         - __insert_nid_to_list(FREE_NID_LIST)
+		 */
+		ne = __lookup_nat_cache(nm_i, nid);
+		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
+				nat_get_blkaddr(ne) != NULL_ADDR))
+			goto err_out;
+
+		e = __lookup_free_nid_list(nm_i, nid);
+		if (e && e->state == NID_ALLOC)
+			goto err_out;
+	}
+	ret = true;
 	err = __insert_nid_to_list(sbi, i, FREE_NID_LIST, true);
+err_out:
 	spin_unlock(&nm_i->nid_list_lock);
 	radix_tree_preload_end();
-	if (err) {
+err:
+	if (err)
 		kmem_cache_free(free_nid_slab, i);
-		return true;
-	}
-	return true;
+	return ret;
 }
 
 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
Signature · 6 lines
2.11.0


------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
author
#2 lore raw

Hi Jaegeuk,

Quoted 26 lines · Jaegeuk Kim

Hi Chao,

Quoted 23 lines · Chao Yu

In below concurrent case, allocated nid can be loaded into free nid cache and be allocated again.

Thread A Thread B
- f2fs_create
- f2fs_new_inode
- alloc_nid
- __insert_nid_to_list(ALLOC_NID_LIST)
- f2fs_balance_fs_bg
- build_free_nids
- __build_free_nids
- scan_nat_page
- add_free_nid
- __lookup_nat_cache
- f2fs_add_link
- init_inode_metadata
- new_inode_page
- new_node_page
- set_node_addr
- alloc_nid_done
- __remove_nid_from_list(ALLOC_NID_LIST)
- __insert_nid_to_list(FREE_NID_LIST)

Quoted 4 lines
Quoted 2 lines

This patch uses build_lock covering free nid allocation and initialization to avoid this race condition.

I have a concern about mutex contention.

Alright, till now I haven't caught any other race problem, so I think it would be OK as well.

Quoted 1 line

How about this?

I change it a bit to avoid unneeded radix-tree operation, see below.

Quoted 64 lines
---
 fs/f2fs/node.c | 60 ++++++++++++++++++++++++++++++++++++++++------------------
 1 file changed, 42 insertions(+), 18 deletions(-)
fs/f2fs/node.c +25 -14
index 347aa30ef0cf..25bc47087f91 100644
@@ -1759,40 +1759,64 @@ static void __remove_nid_from_list(struct f2fs_sb_info *sbi,
 static bool add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
 {
 	struct f2fs_nm_info *nm_i = NM_I(sbi);
-	struct free_nid *i;
+	struct free_nid *i, *e;
 	struct nat_entry *ne;
-	int err;
+	int err = -EINVAL;
+	bool ret = false;
 
 	/* 0 nid should not be used */
 	if (unlikely(nid == 0))
 		return false;
 
-	if (build) {
-		/* do not add allocated nids */
-		ne = __lookup_nat_cache(nm_i, nid);
-		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
-				nat_get_blkaddr(ne) != NULL_ADDR))
-			return false;
-	}
-
 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
 	i->nid = nid;
 	i->state = NID_NEW;
 
-	if (radix_tree_preload(GFP_NOFS)) {
-		kmem_cache_free(free_nid_slab, i);
-		return true;
-	}
+	if (radix_tree_preload(GFP_NOFS))
+		goto err;
 
 	spin_lock(&nm_i->nid_list_lock);
+
+	if (build) {
+		/*
+		 *   Thread A             Thread B
+		 *  - f2fs_create
+		 *   - f2fs_new_inode
+		 *    - alloc_nid
+		 *     - __insert_nid_to_list(ALLOC_NID_LIST)
+		 *                     - f2fs_balance_fs_bg
+		 *                      - build_free_nids
+		 *                       - __build_free_nids
+		 *                        - scan_nat_page
+		 *                         - add_free_nid
+		 *                          - __lookup_nat_cache
+		 *  - f2fs_add_link
+		 *   - init_inode_metadata
+		 *    - new_inode_page
+		 *     - new_node_page
+		 *      - set_node_addr
+		 *  - alloc_nid_done

fs/f2fs/node.c static void __remove_nid_from_list(struct f2fs_sb_info *sbi, · line 1808

Quoted 10 lines
fs/f2fs/node.c +10 -0
+		 *   - __remove_nid_from_list(ALLOC_NID_LIST)
+		 *                         - __insert_nid_to_list(FREE_NID_LIST)
+		 */
+		ne = __lookup_nat_cache(nm_i, nid);
+		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
+				nat_get_blkaddr(ne) != NULL_ADDR))
+			goto err_out;
+
+		e = __lookup_free_nid_list(nm_i, nid);
+		if (e && e->state == NID_ALLOC)
		e = __lookup_free_nid_list(nm_i, nid);
		if (e) {
			if (e->state == NID_NEW)
				ret = true;
			goto err_out;
		}

Thanks,

Quoted 9 lines
fs/f2fs/node.c +4 -1
+			goto err_out;
+	}
+	ret = true;
 	err = __insert_nid_to_list(sbi, i, FREE_NID_LIST, true);
+err_out:
 	spin_unlock(&nm_i->nid_list_lock);
 	radix_tree_preload_end();
-	if (err) {

fs/f2fs/node.c static void __remove_nid_from_list(struct f2fs_sb_info *sbi, · line 1822

Quoted 10 lines
fs/f2fs/node.c +3 -3
+err:
+	if (err)
 		kmem_cache_free(free_nid_slab, i);
-		return true;
-	}
-	return true;
+	return ret;
 }
 
 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)

------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most engaging tech sites, Slashdot.org! http://sdm.link/slashdot