f2fs: check skipped write in f2fs_enable_checkpoint()

This patch introduces sbi->nr_pages[F2FS_SKIPPED_WRITE] to record any
skipped write during data flush in f2fs_enable_checkpoint().

So in the loop of data flush, if there is any skipped write in previous
flush, let's retry sync_inode_sb(), otherwise, all dirty data written
before f2fs_enable_checkpoint() should have been persisted, then break
the retry loop.

Signed-off-by: Chao Yu <chao@kernel.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

fs/f2fs/data.c

@@ -3495,6 +3495,16 @@ static inline void account_writeback(struct inode *inode, bool inc)
 	f2fs_up_read(&F2FS_I(inode)->i_sem);
 }
 
+static inline void update_skipped_write(struct f2fs_sb_info *sbi,
+						struct writeback_control *wbc)
+{
+	long skipped = wbc->pages_skipped;
+
+	if (is_sbi_flag_set(sbi, SBI_ENABLE_CHECKPOINT) && skipped &&
+		wbc->sync_mode == WB_SYNC_ALL)
+		atomic_add(skipped, &sbi->nr_pages[F2FS_SKIPPED_WRITE]);
+}
+
 static int __f2fs_write_data_pages(struct address_space *mapping,
 						struct writeback_control *wbc,
 						enum iostat_type io_type)
@@ -3559,10 +3569,19 @@ static int __f2fs_write_data_pages(struct address_space *mapping,
 	 */
 
 	f2fs_remove_dirty_inode(inode);
+
+	/*
+	 * f2fs_write_cache_pages() has retry logic for EAGAIN case which is
+	 * common when racing w/ checkpoint, so only update skipped write
+	 * when ret is non-zero.
+	 */
+	if (ret)
+		update_skipped_write(sbi, wbc);
 	return ret;
 
 skip_write:
 	wbc->pages_skipped += get_dirty_pages(inode);
+	update_skipped_write(sbi, wbc);
 	trace_f2fs_writepages(mapping->host, wbc, DATA);
 	return 0;
 }

fs/f2fs/debug.c

@@ -423,6 +423,7 @@ static const char *s_flag[MAX_SBI_FLAG] = {
 	[SBI_IS_RESIZEFS]	= "resizefs",
 	[SBI_IS_FREEZING]	= "freezefs",
 	[SBI_IS_WRITABLE]	= "writable",
+	[SBI_ENABLE_CHECKPOINT]	= "enable_checkpoint",
 };
 
 static const char *ipu_mode_names[F2FS_IPU_MAX] = {

fs/f2fs/f2fs.h

@@ -1238,6 +1238,7 @@ enum count_type {
 	F2FS_RD_META,
 	F2FS_DIO_WRITE,
 	F2FS_DIO_READ,
+	F2FS_SKIPPED_WRITE,	/* skip or fail during f2fs_enable_checkpoint() */
 	NR_COUNT_TYPE,
 };
 
@@ -1476,6 +1477,7 @@ enum {
 	SBI_IS_RESIZEFS,			/* resizefs is in process */
 	SBI_IS_FREEZING,			/* freezefs is in process */
 	SBI_IS_WRITABLE,			/* remove ro mountoption transiently */
+	SBI_ENABLE_CHECKPOINT,			/* indicate it's during f2fs_enable_checkpoint() */
 	MAX_SBI_FLAG,
 };
 

fs/f2fs/super.c

@@ -2690,6 +2690,7 @@ static int f2fs_enable_checkpoint(struct f2fs_sb_info *sbi)
 	long long start, writeback, end;
 	int ret;
 	struct f2fs_lock_context lc;
+	long long skipped_write, dirty_data;
 
 	f2fs_info(sbi, "f2fs_enable_checkpoint() starts, meta: %lld, node: %lld, data: %lld",
 					get_pages(sbi, F2FS_DIRTY_META),
@@ -2698,17 +2699,45 @@ static int f2fs_enable_checkpoint(struct f2fs_sb_info *sbi)
 
 	start = ktime_get();
 
+	set_sbi_flag(sbi, SBI_ENABLE_CHECKPOINT);
+
 	/* we should flush all the data to keep data consistency */
 	do {
+		skipped_write = get_pages(sbi, F2FS_SKIPPED_WRITE);
+		dirty_data = get_pages(sbi, F2FS_DIRTY_DATA);
+
 		sync_inodes_sb(sbi->sb);
 		f2fs_io_schedule_timeout(DEFAULT_SCHEDULE_TIMEOUT);
-	} while (get_pages(sbi, F2FS_DIRTY_DATA) && retry--);
+
+		f2fs_info(sbi, "sync_inode_sb done, dirty_data: %lld, %lld, "
+				"skipped write: %lld, %lld, retry: %d",
+				get_pages(sbi, F2FS_DIRTY_DATA),
+				dirty_data,
+				get_pages(sbi, F2FS_SKIPPED_WRITE),
+				skipped_write, retry);
+
+		/*
+		 * sync_inodes_sb() has retry logic, so let's check dirty_data
+		 * in prior to skipped_write in case there is no dirty data.
+		 */
+		if (!get_pages(sbi, F2FS_DIRTY_DATA))
+			break;
+		if (get_pages(sbi, F2FS_SKIPPED_WRITE) == skipped_write)
+			break;
+	} while (retry--);
+
+	clear_sbi_flag(sbi, SBI_ENABLE_CHECKPOINT);
 
 	writeback = ktime_get();
 
-	if (unlikely(get_pages(sbi, F2FS_DIRTY_DATA)))
+	if (unlikely(get_pages(sbi, F2FS_DIRTY_DATA) ||
-		f2fs_warn(sbi, "checkpoint=enable has some unwritten data: %lld",
+			get_pages(sbi, F2FS_SKIPPED_WRITE)))
-					get_pages(sbi, F2FS_DIRTY_DATA));
+		f2fs_warn(sbi, "checkpoint=enable unwritten data: %lld, skipped data: %lld, retry: %d",
+				get_pages(sbi, F2FS_DIRTY_DATA),
+				get_pages(sbi, F2FS_SKIPPED_WRITE), retry);
+
+	if (get_pages(sbi, F2FS_SKIPPED_WRITE))
+		atomic_set(&sbi->nr_pages[F2FS_SKIPPED_WRITE], 0);
 
 	f2fs_down_write_trace(&sbi->gc_lock, &lc);
 	f2fs_dirty_to_prefree(sbi);