[PATCH V2] f2fs: fix the panic in do_checkpoint()

[Sahitya Tummala]

There could be a scenario where f2fs_sync_meta_pages() will not
ensure that all F2FS_DIRTY_META pages are submitted for IO. Thus,
resulting in the below panic in do_checkpoint() -

f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_META) &&
				!f2fs_cp_error(sbi));

This can happen in a low-memory condition, where shrinker could
also be doing the writepage operation (stack shown below)
at the same time when checkpoint is running on another core.

schedule
down_write
f2fs_submit_page_write -> by this time, this page in page cache is tagged
			as PAGECACHE_TAG_WRITEBACK and PAGECACHE_TAG_DIRTY
			is cleared, due to which f2fs_sync_meta_pages()
			cannot sync this page in do_checkpoint() path.
f2fs_do_write_meta_page
__f2fs_write_meta_page
f2fs_write_meta_page
shrink_page_list
shrink_inactive_list
shrink_node_memcg
shrink_node
kswapd

Signed-off-by: Sahitya Tummala <stummala@xxxxxxxxxxxxxx>
---
v2:
- changed the io_schedule_timeout to HZ/50.

 fs/f2fs/checkpoint.c | 18 +++++++++---------
 fs/f2fs/f2fs.h       |  2 +-
 fs/f2fs/super.c      |  2 +-
 3 files changed, 11 insertions(+), 11 deletions(-)

diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c
index ffdaba0..d5601cc 100644
--- a/fs/f2fs/checkpoint.c
+++ b/fs/f2fs/checkpoint.c
@@ -1250,20 +1250,20 @@ static void unblock_operations(struct f2fs_sb_info *sbi)
 	f2fs_unlock_all(sbi);
 }
 
-void f2fs_wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
+void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
 {
 	DEFINE_WAIT(wait);
 
 	for (;;) {
 		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
 
-		if (!get_pages(sbi, F2FS_WB_CP_DATA))
+		if (!get_pages(sbi, type))
 			break;
 
 		if (unlikely(f2fs_cp_error(sbi)))
 			break;
 
-		io_schedule_timeout(5*HZ);
+		io_schedule_timeout(HZ/50);
 	}
 	finish_wait(&sbi->cp_wait, &wait);
 }
@@ -1384,8 +1384,8 @@ static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 
 	/* Flush all the NAT/SIT pages */
 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
-	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_META) &&
-					!f2fs_cp_error(sbi));
+	/* Wait for all dirty meta pages to be submitted for IO */
+	f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
 
 	/*
 	 * modify checkpoint
@@ -1493,11 +1493,11 @@ static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 
 	/* Here, we have one bio having CP pack except cp pack 2 page */
 	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
-	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_META) &&
-					!f2fs_cp_error(sbi));
+	/* Wait for all dirty meta pages to be submitted for IO */
+	f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
 
 	/* wait for previous submitted meta pages writeback */
-	f2fs_wait_on_all_pages_writeback(sbi);
+	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
 
 	/* flush all device cache */
 	err = f2fs_flush_device_cache(sbi);
@@ -1506,7 +1506,7 @@ static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 
 	/* barrier and flush checkpoint cp pack 2 page if it can */
 	commit_checkpoint(sbi, ckpt, start_blk);
-	f2fs_wait_on_all_pages_writeback(sbi);
+	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
 
 	/*
 	 * invalidate intermediate page cache borrowed from meta inode
diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index 5a888a0..b0e0535 100644
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -3196,7 +3196,7 @@ bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 void f2fs_update_dirty_page(struct inode *inode, struct page *page);
 void f2fs_remove_dirty_inode(struct inode *inode);
 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type);
-void f2fs_wait_on_all_pages_writeback(struct f2fs_sb_info *sbi);
+void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type);
 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc);
 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi);
 int __init f2fs_create_checkpoint_caches(void);
diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
index 5111e1f..084633b 100644
--- a/fs/f2fs/super.c
+++ b/fs/f2fs/super.c
@@ -1105,7 +1105,7 @@ static void f2fs_put_super(struct super_block *sb)
 	/* our cp_error case, we can wait for any writeback page */
 	f2fs_flush_merged_writes(sbi);
 
-	f2fs_wait_on_all_pages_writeback(sbi);
+	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
 
 	f2fs_bug_on(sbi, sbi->fsync_node_num);
 
-- 
Qualcomm India Private Limited, on behalf of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project.