linux-kernel - [PATCH] f2fs: fix to avoid REQ_TIME and CP

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [day] [month] [year] [list]

Message-ID: <20201125025736.58540-1-yuchao0@huawei.com>
Date:   Wed, 25 Nov 2020 10:57:36 +0800
From:   Chao Yu <yuchao0@...wei.com>
To:     <jaegeuk@...nel.org>
CC:     <linux-f2fs-devel@...ts.sourceforge.net>,
        <linux-kernel@...r.kernel.org>, <chao@...nel.org>,
        Chao Yu <yuchao0@...wei.com>, Lei Li <noctis.akm@...il.com>
Subject: [PATCH] f2fs: fix to avoid REQ_TIME and CP_TIME collision

Lei Li reported a issue: if foreground operations are frequent, background
checkpoint may be always skipped due to below check, result in losing more
data after sudden power-cut.

f2fs_balance_fs_bg()
...
	if (!is_idle(sbi, REQ_TIME) &&
		(!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
		return;

E.g:
cp_interval = 5 second
idle_interval = 2 second
foreground operation interval = 1 second (append 1 byte per second into file)

In such case, no matter when it calls f2fs_balance_fs_bg(), is_idle(, REQ_TIME)
returns false, result in skipping background checkpoint.

This patch changes as below to make trigger condition being more reasonable:
- trigger sync_fs() if dirty_{nats,nodes} and prefree segs exceeds threshold;
- skip triggering sync_fs() if there is any background inflight IO or there is
foreground operation recently and meanwhile cp_rwsem is being held by someone;

Reported-by: Lei Li <noctis.akm@...il.com>
Signed-off-by: Chao Yu <yuchao0@...wei.com>
---
 fs/f2fs/f2fs.h    | 19 +++++++++++++------
 fs/f2fs/segment.c | 47 +++++++++++++++++++++++++++--------------------
 2 files changed, 40 insertions(+), 26 deletions(-)

diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index 4f2766f3d2c1..b5054ffe3e1c 100644
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -2416,24 +2416,31 @@ static inline void *f2fs_kmem_cache_alloc(struct kmem_cache *cachep,
 	return entry;
 }
 
-static inline bool is_idle(struct f2fs_sb_info *sbi, int type)
+static inline bool is_inflight_io(struct f2fs_sb_info *sbi, int type)
 {
-	if (sbi->gc_mode == GC_URGENT_HIGH)
-		return true;
-
 	if (get_pages(sbi, F2FS_RD_DATA) || get_pages(sbi, F2FS_RD_NODE) ||
 		get_pages(sbi, F2FS_RD_META) || get_pages(sbi, F2FS_WB_DATA) ||
 		get_pages(sbi, F2FS_WB_CP_DATA) ||
 		get_pages(sbi, F2FS_DIO_READ) ||
 		get_pages(sbi, F2FS_DIO_WRITE))
-		return false;
+		return true;
 
 	if (type != DISCARD_TIME && SM_I(sbi) && SM_I(sbi)->dcc_info &&
 			atomic_read(&SM_I(sbi)->dcc_info->queued_discard))
-		return false;
+		return true;
 
 	if (SM_I(sbi) && SM_I(sbi)->fcc_info &&
 			atomic_read(&SM_I(sbi)->fcc_info->queued_flush))
+		return true;
+	return false;
+}
+
+static inline bool is_idle(struct f2fs_sb_info *sbi, int type)
+{
+	if (sbi->gc_mode == GC_URGENT_HIGH)
+		return true;
+
+	if (is_inflight_io(sbi, type))
 		return false;
 
 	if (sbi->gc_mode == GC_URGENT_LOW &&
diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
index 39d6dce09fb7..bd066cff0bf9 100644
--- a/fs/f2fs/segment.c
+++ b/fs/f2fs/segment.c
@@ -529,31 +529,38 @@ void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
 	else
 		f2fs_build_free_nids(sbi, false, false);
 
-	if (!is_idle(sbi, REQ_TIME) &&
-		(!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
+	if (excess_dirty_nats(sbi) || excess_dirty_nodes(sbi) ||
+		excess_prefree_segs(sbi))
+		goto do_sync;
+
+	/* there is background inflight IO or foreground operation recently */
+	if (is_inflight_io(sbi, REQ_TIME) ||
+		(!f2fs_time_over(sbi, REQ_TIME) && rwsem_is_locked(&sbi->cp_rwsem)))
 		return;
 
+	/* exceed periodical checkpoint timeout threshold */
+	if (f2fs_time_over(sbi, CP_TIME))
+		goto do_sync;
+
 	/* checkpoint is the only way to shrink partial cached entries */
-	if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
-			!f2fs_available_free_memory(sbi, INO_ENTRIES) ||
-			excess_prefree_segs(sbi) ||
-			excess_dirty_nats(sbi) ||
-			excess_dirty_nodes(sbi) ||
-			f2fs_time_over(sbi, CP_TIME)) {
-		if (test_opt(sbi, DATA_FLUSH) && from_bg) {
-			struct blk_plug plug;
-
-			mutex_lock(&sbi->flush_lock);
-
-			blk_start_plug(&plug);
-			f2fs_sync_dirty_inodes(sbi, FILE_INODE);
-			blk_finish_plug(&plug);
+	if (f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
+		f2fs_available_free_memory(sbi, INO_ENTRIES))
+		return;
 
-			mutex_unlock(&sbi->flush_lock);
-		}
-		f2fs_sync_fs(sbi->sb, true);
-		stat_inc_bg_cp_count(sbi->stat_info);
+do_sync:
+	if (test_opt(sbi, DATA_FLUSH) && from_bg) {
+		struct blk_plug plug;
+
+		mutex_lock(&sbi->flush_lock);
+
+		blk_start_plug(&plug);
+		f2fs_sync_dirty_inodes(sbi, FILE_INODE);
+		blk_finish_plug(&plug);
+
+		mutex_unlock(&sbi->flush_lock);
 	}
+	f2fs_sync_fs(sbi->sb, true);
+	stat_inc_bg_cp_count(sbi->stat_info);
 }
 
 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
-- 
2.26.2