Skip to contentSkip to conversation

f2fs: avoid inifinite loop to wait for flushing node pages at cp_error

PATCH In review Jaegeuk Kim 2 participants 12 messages May 22, 2020 – May 28, 2020 F2FS file system

lore
Share & tools
lore

Jaegeuk Kim

author maintainer
#0 lore raw

PATCH 1/1 f2fs: avoid inifinite loop to wait for flushing node pages at cp_error

Shutdown test is somtime hung, since dirty node pages weren't flushed out. Let's drop dirty pages at umount after shutdown.

---
 fs/f2fs/node.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
fs/f2fs/node.c +8 -1
index e632de10aedab..8c63964a82fd0 100644
@@ -1520,8 +1520,15 @@ static int __write_node_page(struct page *page, bool atomic, bool *submitted,
 
 	trace_f2fs_writepage(page, NODE);
 
-	if (unlikely(f2fs_cp_error(sbi)))
+	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			dec_page_count(sbi, F2FS_DIRTY_NODES);
+			up_read(&sbi->node_write);
+			unlock_page(page);
+			return 0;
+		}
 		goto redirty_out;
+	}
 
 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 		goto redirty_out;
Signature · 1 line
2.27.0.rc0.183.gde8f92d652-goog
author maintainer
#1 lore raw

Shutdown test is somtimes hung, since it keeps trying to flush dirty node pages in an inifinite loop. Let's drop dirty pages at umount in that case.

---
v2:
- fix typos

 fs/f2fs/node.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
fs/f2fs/node.c +8 -1
index e632de10aedab..8c63964a82fd0 100644
@@ -1520,8 +1520,15 @@ static int __write_node_page(struct page *page, bool atomic, bool *submitted,
 
 	trace_f2fs_writepage(page, NODE);
 
-	if (unlikely(f2fs_cp_error(sbi)))
+	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			dec_page_count(sbi, F2FS_DIRTY_NODES);
+			up_read(&sbi->node_write);
+			unlock_page(page);
+			return 0;
+		}
 		goto redirty_out;
+	}
 
 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 		goto redirty_out;
Signature · 1 line
2.27.0.rc0.183.gde8f92d652-goog
author maintainer
#2 lore raw

Shutdown test is somtimes hung, since it keeps trying to flush dirty node pages in an inifinite loop. Let's drop dirty pages at umount in that case.

---
v3:
- fix wrong unlock

v2:
- fix typos

 fs/f2fs/node.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
fs/f2fs/node.c +8 -1
index e632de10aedab..e0bb0f7e0506e 100644
@@ -1520,8 +1520,15 @@ static int __write_node_page(struct page *page, bool atomic, bool *submitted,
 
 	trace_f2fs_writepage(page, NODE);
 
-	if (unlikely(f2fs_cp_error(sbi)))
+	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			ClearPageUptodate(page);
+			dec_page_count(sbi, F2FS_DIRTY_NODES);
+			unlock_page(page);
+			return 0;
+		}
 		goto redirty_out;
+	}
 
 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 		goto redirty_out;
Signature · 1 line
2.27.0.rc0.183.gde8f92d652-goog
#3 lore raw
Quoted 1 line · Jaegeuk Kim

Shutdown test is somtimes hung, since it keeps trying to flush dirty node pages

IMO, for umount case, we should drop dirty reference and dirty pages on meta/data pages like we change for node pages to avoid potential dead loop...

Thanks,

Quoted 35 lines

in an inifinite loop. Let's drop dirty pages at umount in that case.

---
v3:
- fix wrong unlock

v2:
- fix typos

 fs/f2fs/node.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
fs/f2fs/node.c +8 -1
index e632de10aedab..e0bb0f7e0506e 100644
@@ -1520,8 +1520,15 @@ static int __write_node_page(struct page *page, bool atomic, bool *submitted,
 
 	trace_f2fs_writepage(page, NODE);
 
-	if (unlikely(f2fs_cp_error(sbi)))
+	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			ClearPageUptodate(page);
+			dec_page_count(sbi, F2FS_DIRTY_NODES);
+			unlock_page(page);
+			return 0;
+		}
 		goto redirty_out;
+	}
 
 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 		goto redirty_out;
author maintainer
#4 lore raw
Quoted 5 lines · Chao Yu
Quoted 1 line · Jaegeuk Kim

Shutdown test is somtimes hung, since it keeps trying to flush dirty node pages

IMO, for umount case, we should drop dirty reference and dirty pages on meta/data pages like we change for node pages to avoid potential dead loop...

I believe we're doing for them. :P

Quoted 38 lines

Thanks,

Quoted 35 lines

in an inifinite loop. Let's drop dirty pages at umount in that case.

---
v3:
- fix wrong unlock

v2:
- fix typos

 fs/f2fs/node.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
fs/f2fs/node.c +8 -1
index e632de10aedab..e0bb0f7e0506e 100644
@@ -1520,8 +1520,15 @@ static int __write_node_page(struct page *page, bool atomic, bool *submitted,
 
 	trace_f2fs_writepage(page, NODE);
 
-	if (unlikely(f2fs_cp_error(sbi)))
+	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			ClearPageUptodate(page);
+			dec_page_count(sbi, F2FS_DIRTY_NODES);
+			unlock_page(page);
+			return 0;
+		}
 		goto redirty_out;
+	}
 
 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 		goto redirty_out;
#5 lore raw
Quoted 8 lines · Jaegeuk Kim
Quoted 5 lines · Chao Yu
Quoted 1 line · Jaegeuk Kim

Shutdown test is somtimes hung, since it keeps trying to flush dirty node pages

IMO, for umount case, we should drop dirty reference and dirty pages on meta/data pages like we change for node pages to avoid potential dead loop...

I believe we're doing for them. :P

Actually, I mean do we need to drop dirty meta/data pages explicitly as below:

fs/f2fs/checkpoint.c +8 -1
index 3dc3ac6fe143..4c08fd0a680a 100644
@@ -299,8 +299,15 @@ static int __f2fs_write_meta_page(struct page *page,
 
 	trace_f2fs_writepage(page, META);
 
-	if (unlikely(f2fs_cp_error(sbi)))
+	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			ClearPageUptodate(page);
+			dec_page_count(sbi, F2FS_DIRTY_META);
+			unlock_page(page);
+			return 0;
+		}
 		goto redirty_out;
+	}
 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 		goto redirty_out;
 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
fs/f2fs/data.c +6 -0
index 48a622b95b76..94b342802513 100644
@@ -2682,6 +2682,12 @@ int f2fs_write_single_data_page(struct page *page, int *submitted,
 
 	/* we should bypass data pages to proceed the kworkder jobs */
 	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			ClearPageUptodate(page);
+			inode_dec_dirty_pages(inode);
+			unlock_page(page);
+			return 0;
+		}
 		mapping_set_error(page->mapping, -EIO);
 		/*
 		 * don't drop any dirty dentry pages for keeping lastest
Quoted 41 lines
Quoted 36 lines

Thanks,

Quoted 32 lines

in an inifinite loop. Let's drop dirty pages at umount in that case.

---
v3:
- fix wrong unlock

v2:
- fix typos

 fs/f2fs/node.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
fs/f2fs/node.c +8 -1
index e632de10aedab..e0bb0f7e0506e 100644
@@ -1520,8 +1520,15 @@ static int __write_node_page(struct page *page, bool atomic, bool *submitted,
 
 	trace_f2fs_writepage(page, NODE);
 
-	if (unlikely(f2fs_cp_error(sbi)))
+	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			ClearPageUptodate(page);
+			dec_page_count(sbi, F2FS_DIRTY_NODES);
+			unlock_page(page);
+			return 0;
+		}
 		goto redirty_out;
+	}
 
Quoted 3 lines

fs/f2fs/node.c static int __write_node_page(struct page *page, bool atomic, bool *submitted, · line 1534

Quoted 3 lines
fs/f2fs/node.c +0 -0
 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 		goto redirty_out;

.

#6 lore raw
Quoted 38 lines · Chao Yu
Quoted 8 lines · Jaegeuk Kim
Quoted 5 lines · Chao Yu
Quoted 1 line · Jaegeuk Kim

Shutdown test is somtimes hung, since it keeps trying to flush dirty node pages

IMO, for umount case, we should drop dirty reference and dirty pages on meta/data pages like we change for node pages to avoid potential dead loop...

I believe we're doing for them. :P

Actually, I mean do we need to drop dirty meta/data pages explicitly as below:

fs/f2fs/checkpoint.c +8 -1
index 3dc3ac6fe143..4c08fd0a680a 100644
@@ -299,8 +299,15 @@ static int __f2fs_write_meta_page(struct page *page,
 
 	trace_f2fs_writepage(page, META);
 
-	if (unlikely(f2fs_cp_error(sbi)))
+	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			ClearPageUptodate(page);
+			dec_page_count(sbi, F2FS_DIRTY_META);
+			unlock_page(page);
+			return 0;
+		}
 		goto redirty_out;
+	}
 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 		goto redirty_out;
 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
fs/f2fs/data.c +0 -0
index 48a622b95b76..94b342802513 100644

fs/f2fs/data.c int f2fs_write_single_data_page(struct page *page, int *submitted, · line 2690

Quoted 9 lines
fs/f2fs/data.c +6 -0
@@ -2682,6 +2682,12 @@ int f2fs_write_single_data_page(struct page *page, int *submitted,
 
 	/* we should bypass data pages to proceed the kworkder jobs */
 	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			ClearPageUptodate(page);
+			inode_dec_dirty_pages(inode);
+			unlock_page(page);
+			return 0;
+		}

Oh, I notice previously, we will drop non-directory inode's dirty pages directly, however, during umount, we'd better drop directory inode's dirty pages as well, right?

Quoted 53 lines
fs/f2fs/data.c +0 -0
 		mapping_set_error(page->mapping, -EIO);
 		/*
 		 * don't drop any dirty dentry pages for keeping lastest
Quoted 37 lines
Quoted 37 lines

Thanks,

Quoted 32 lines

in an inifinite loop. Let's drop dirty pages at umount in that case.

---
v3:
- fix wrong unlock

v2:
- fix typos

 fs/f2fs/node.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
fs/f2fs/node.c +8 -1
index e632de10aedab..e0bb0f7e0506e 100644
@@ -1520,8 +1520,15 @@ static int __write_node_page(struct page *page, bool atomic, bool *submitted,
 
 	trace_f2fs_writepage(page, NODE);
 
-	if (unlikely(f2fs_cp_error(sbi)))
+	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			ClearPageUptodate(page);
+			dec_page_count(sbi, F2FS_DIRTY_NODES);
+			unlock_page(page);
+			return 0;
+		}
 		goto redirty_out;
+	}
 
Quoted 5 lines
Quoted 3 lines

fs/f2fs/node.c static int __write_node_page(struct page *page, bool atomic, bool *submitted, · line 1534

Quoted 3 lines
fs/f2fs/node.c +0 -0
 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 		goto redirty_out;

.

_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
.

author maintainer
#7 lore raw
Quoted 40 lines · Chao Yu
Quoted 39 lines · Chao Yu
Quoted 8 lines · Jaegeuk Kim
Quoted 5 lines · Chao Yu
Quoted 1 line · Jaegeuk Kim

Shutdown test is somtimes hung, since it keeps trying to flush dirty node pages

IMO, for umount case, we should drop dirty reference and dirty pages on meta/data pages like we change for node pages to avoid potential dead loop...

I believe we're doing for them. :P

Actually, I mean do we need to drop dirty meta/data pages explicitly as below:

fs/f2fs/checkpoint.c +8 -1
index 3dc3ac6fe143..4c08fd0a680a 100644
@@ -299,8 +299,15 @@ static int __f2fs_write_meta_page(struct page *page,
 
 	trace_f2fs_writepage(page, META);
 
-	if (unlikely(f2fs_cp_error(sbi)))
+	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			ClearPageUptodate(page);
+			dec_page_count(sbi, F2FS_DIRTY_META);
+			unlock_page(page);
+			return 0;
+		}
 		goto redirty_out;
+	}
 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 		goto redirty_out;
 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
fs/f2fs/data.c +0 -0
index 48a622b95b76..94b342802513 100644
@@ -2682,6 +2682,12 @@ int f2fs_write_single_data_page(struct page *page, int *submitted,
 
Quoted 11 lines

fs/f2fs/data.c int f2fs_write_single_data_page(struct page *page, int *submitted, · line 2690

Quoted 8 lines
fs/f2fs/data.c +6 -0
 	/* we should bypass data pages to proceed the kworkder jobs */
 	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			ClearPageUptodate(page);
+			inode_dec_dirty_pages(inode);
+			unlock_page(page);
+			return 0;
+		}

Oh, I notice previously, we will drop non-directory inode's dirty pages directly, however, during umount, we'd better drop directory inode's dirty pages as well, right?

Hmm, I remember I dropped them before. Need to double check.

Quoted 54 lines
Quoted 54 lines
fs/f2fs/data.c +0 -0
 		mapping_set_error(page->mapping, -EIO);
 		/*
 		 * don't drop any dirty dentry pages for keeping lastest
Quoted 37 lines
Quoted 37 lines

Thanks,

Quoted 32 lines

in an inifinite loop. Let's drop dirty pages at umount in that case.

---
v3:
- fix wrong unlock

v2:
- fix typos

 fs/f2fs/node.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
fs/f2fs/node.c +8 -1
index e632de10aedab..e0bb0f7e0506e 100644
@@ -1520,8 +1520,15 @@ static int __write_node_page(struct page *page, bool atomic, bool *submitted,
 
 	trace_f2fs_writepage(page, NODE);
 
-	if (unlikely(f2fs_cp_error(sbi)))
+	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			ClearPageUptodate(page);
+			dec_page_count(sbi, F2FS_DIRTY_NODES);
+			unlock_page(page);
+			return 0;
+		}
 		goto redirty_out;
+	}
 
Quoted 5 lines
Quoted 3 lines

fs/f2fs/node.c static int __write_node_page(struct page *page, bool atomic, bool *submitted, · line 1534

Quoted 3 lines
fs/f2fs/node.c +0 -0
 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 		goto redirty_out;

.

_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
.

#8 lore raw
Quoted 6 lines · Jaegeuk Kim
Quoted 5 lines · Chao Yu
Quoted 4 lines · Chao Yu
Quoted 3 lines · Jaegeuk Kim
Quoted 2 lines · Chao Yu
Quoted 1 line · Jaegeuk Kim

Shutdown test is somtimes hung, since it keeps trying to flush dirty node pages

71.07% 0.01% kworker/u256:1+ [kernel.kallsyms] [k] wb_writeback
|
--71.06%--wb_writeback
| |--68.96%--__writeback_inodes_wb
| |
| --68.95%--writeback_sb_inodes
| |
| |--65.08%--__writeback_single_inode
| | |
| | --64.35%--do_writepages
| | |
| | |--59.83%--f2fs_write_node_pages
| | | |
| | | --59.74%--f2fs_sync_node_pages
| | | |
| | | |--27.91%--pagevec_lookup_range_tag
| | | | |
| | | | --27.90%--find_get_pages_range_tag

Before umount, kworker will always hold one core, that looks not reasonable, to avoid that, could we just allow node write, since it's out-place-update, and cp is not allowed, we don't need to worry about its effect on data on previous checkpoint, and it can decrease memory footprint cost by node pages.

Thanks,

Quoted 105 lines
Quoted 35 lines
Quoted 35 lines
Quoted 5 lines
Quoted 3 lines

IMO, for umount case, we should drop dirty reference and dirty pages on meta/data pages like we change for node pages to avoid potential dead loop...

I believe we're doing for them. :P

Actually, I mean do we need to drop dirty meta/data pages explicitly as below:

fs/f2fs/checkpoint.c +8 -1
index 3dc3ac6fe143..4c08fd0a680a 100644
@@ -299,8 +299,15 @@ static int __f2fs_write_meta_page(struct page *page,
 
 	trace_f2fs_writepage(page, META);
 
-	if (unlikely(f2fs_cp_error(sbi)))
+	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			ClearPageUptodate(page);
+			dec_page_count(sbi, F2FS_DIRTY_META);
+			unlock_page(page);
+			return 0;
+		}
 		goto redirty_out;
+	}
 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 		goto redirty_out;
 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
fs/f2fs/data.c +0 -0
index 48a622b95b76..94b342802513 100644
@@ -2682,6 +2682,12 @@ int f2fs_write_single_data_page(struct page *page, int *submitted,
 
Quoted 11 lines

fs/f2fs/data.c int f2fs_write_single_data_page(struct page *page, int *submitted, · line 2690

Quoted 8 lines
fs/f2fs/data.c +6 -0
 	/* we should bypass data pages to proceed the kworkder jobs */
 	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			ClearPageUptodate(page);
+			inode_dec_dirty_pages(inode);
+			unlock_page(page);
+			return 0;
+		}

Oh, I notice previously, we will drop non-directory inode's dirty pages directly, however, during umount, we'd better drop directory inode's dirty pages as well, right?

Hmm, I remember I dropped them before. Need to double check.

Quoted 48 lines
Quoted 48 lines
fs/f2fs/data.c +0 -0
 		mapping_set_error(page->mapping, -EIO);
 		/*
 		 * don't drop any dirty dentry pages for keeping lastest
Quoted 42 lines
Quoted 37 lines

Thanks,

Quoted 32 lines

in an inifinite loop. Let's drop dirty pages at umount in that case.

---
v3:
- fix wrong unlock

v2:
- fix typos

 fs/f2fs/node.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
fs/f2fs/node.c +8 -1
index e632de10aedab..e0bb0f7e0506e 100644
@@ -1520,8 +1520,15 @@ static int __write_node_page(struct page *page, bool atomic, bool *submitted,
 
 	trace_f2fs_writepage(page, NODE);
 
-	if (unlikely(f2fs_cp_error(sbi)))
+	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			ClearPageUptodate(page);
+			dec_page_count(sbi, F2FS_DIRTY_NODES);
+			unlock_page(page);
+			return 0;
+		}
 		goto redirty_out;
+	}
 
Quoted 3 lines

fs/f2fs/node.c static int __write_node_page(struct page *page, bool atomic, bool *submitted, · line 1534

Quoted 3 lines
fs/f2fs/node.c +0 -0
 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 		goto redirty_out;

.

Quoted 6 lines
Quoted 6 lines

_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
.

.

author maintainer
#9 lore raw
Quoted 28 lines · Chao Yu
Quoted 6 lines · Jaegeuk Kim
Quoted 5 lines · Chao Yu
Quoted 4 lines · Chao Yu
Quoted 3 lines · Jaegeuk Kim
Quoted 2 lines · Chao Yu
Quoted 1 line · Jaegeuk Kim

Shutdown test is somtimes hung, since it keeps trying to flush dirty node pages

71.07% 0.01% kworker/u256:1+ [kernel.kallsyms] [k] wb_writeback
|
--71.06%--wb_writeback
| |--68.96%--__writeback_inodes_wb
| |
| --68.95%--writeback_sb_inodes
| |
| |--65.08%--__writeback_single_inode
| | |
| | --64.35%--do_writepages
| | |
| | |--59.83%--f2fs_write_node_pages
| | | |
| | | --59.74%--f2fs_sync_node_pages
| | | |
| | | |--27.91%--pagevec_lookup_range_tag
| | | | |
| | | | --27.90%--find_get_pages_range_tag

Quoted 4 lines

Before umount, kworker will always hold one core, that looks not reasonable, to avoid that, could we just allow node write, since it's out-place-update, and cp is not allowed, we don't need to worry about its effect on data on previous checkpoint, and it can decrease memory footprint cost by node pages.

It can cause some roll-forward recovery?

Quoted 108 lines

Thanks,

Quoted 105 lines
Quoted 35 lines
Quoted 35 lines
Quoted 5 lines
Quoted 3 lines

IMO, for umount case, we should drop dirty reference and dirty pages on meta/data pages like we change for node pages to avoid potential dead loop...

I believe we're doing for them. :P

Actually, I mean do we need to drop dirty meta/data pages explicitly as below:

fs/f2fs/checkpoint.c +8 -1
index 3dc3ac6fe143..4c08fd0a680a 100644
@@ -299,8 +299,15 @@ static int __f2fs_write_meta_page(struct page *page,
 
 	trace_f2fs_writepage(page, META);
 
-	if (unlikely(f2fs_cp_error(sbi)))
+	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			ClearPageUptodate(page);
+			dec_page_count(sbi, F2FS_DIRTY_META);
+			unlock_page(page);
+			return 0;
+		}
 		goto redirty_out;
+	}
 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 		goto redirty_out;
 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
fs/f2fs/data.c +0 -0
index 48a622b95b76..94b342802513 100644
@@ -2682,6 +2682,12 @@ int f2fs_write_single_data_page(struct page *page, int *submitted,
 
Quoted 11 lines

fs/f2fs/data.c int f2fs_write_single_data_page(struct page *page, int *submitted, · line 2690

Quoted 8 lines
fs/f2fs/data.c +6 -0
 	/* we should bypass data pages to proceed the kworkder jobs */
 	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			ClearPageUptodate(page);
+			inode_dec_dirty_pages(inode);
+			unlock_page(page);
+			return 0;
+		}

Oh, I notice previously, we will drop non-directory inode's dirty pages directly, however, during umount, we'd better drop directory inode's dirty pages as well, right?

Hmm, I remember I dropped them before. Need to double check.

Quoted 48 lines
Quoted 48 lines
fs/f2fs/data.c +0 -0
 		mapping_set_error(page->mapping, -EIO);
 		/*
 		 * don't drop any dirty dentry pages for keeping lastest
Quoted 42 lines
Quoted 37 lines

Thanks,

Quoted 32 lines

in an inifinite loop. Let's drop dirty pages at umount in that case.

---
v3:
- fix wrong unlock

v2:
- fix typos

 fs/f2fs/node.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
fs/f2fs/node.c +8 -1
index e632de10aedab..e0bb0f7e0506e 100644
@@ -1520,8 +1520,15 @@ static int __write_node_page(struct page *page, bool atomic, bool *submitted,
 
 	trace_f2fs_writepage(page, NODE);
 
-	if (unlikely(f2fs_cp_error(sbi)))
+	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			ClearPageUptodate(page);
+			dec_page_count(sbi, F2FS_DIRTY_NODES);
+			unlock_page(page);
+			return 0;
+		}
 		goto redirty_out;
+	}
 
Quoted 3 lines

fs/f2fs/node.c static int __write_node_page(struct page *page, bool atomic, bool *submitted, · line 1534

Quoted 3 lines
fs/f2fs/node.c +0 -0
 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 		goto redirty_out;

.

Quoted 6 lines
Quoted 6 lines

_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
.

.

#10 lore raw
Quoted 29 lines · Jaegeuk Kim
Quoted 28 lines · Chao Yu
Quoted 6 lines · Jaegeuk Kim
Quoted 5 lines · Chao Yu
Quoted 4 lines · Chao Yu
Quoted 3 lines · Jaegeuk Kim
Quoted 2 lines · Chao Yu
Quoted 1 line · Jaegeuk Kim

Shutdown test is somtimes hung, since it keeps trying to flush dirty node pages

71.07% 0.01% kworker/u256:1+ [kernel.kallsyms] [k] wb_writeback
|
--71.06%--wb_writeback
| |--68.96%--__writeback_inodes_wb
| |
| --68.95%--writeback_sb_inodes
| |
| |--65.08%--__writeback_single_inode
| | |
| | --64.35%--do_writepages
| | |
| | |--59.83%--f2fs_write_node_pages
| | | |
| | | --59.74%--f2fs_sync_node_pages
| | | |
| | | |--27.91%--pagevec_lookup_range_tag
| | | | |
| | | | --27.90%--find_get_pages_range_tag

Quoted 6 lines
Quoted 4 lines

Before umount, kworker will always hold one core, that looks not reasonable, to avoid that, could we just allow node write, since it's out-place-update, and cp is not allowed, we don't need to worry about its effect on data on previous checkpoint, and it can decrease memory footprint cost by node pages.

It can cause some roll-forward recovery?

Yup, recovery should be considered,

Later fsync() will fail due to:

int f2fs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)

{
	if (unlikely(f2fs_cp_error(F2FS_I_SB(file_inode(file)))))
		return -EIO;

And we need to adjust f2fs_fsync_node_pages() to handle in-process fsyncing node pages as well:

if (f2fs_cp_error()) {
	set_fsync_mark(page, 0);
	set_dentry_mark(page, 0);
	if (atomic) {
		unlock & put page;
		ret = -EIO;
		break;
	}
}

ret = __write_node_page();

Thanks,

Quoted 111 lines
Quoted 109 lines

Thanks,

Quoted 105 lines
Quoted 35 lines
Quoted 35 lines
Quoted 5 lines
Quoted 3 lines

IMO, for umount case, we should drop dirty reference and dirty pages on meta/data pages like we change for node pages to avoid potential dead loop...

I believe we're doing for them. :P

Actually, I mean do we need to drop dirty meta/data pages explicitly as below:

fs/f2fs/checkpoint.c +8 -1
index 3dc3ac6fe143..4c08fd0a680a 100644
@@ -299,8 +299,15 @@ static int __f2fs_write_meta_page(struct page *page,
 
 	trace_f2fs_writepage(page, META);
 
-	if (unlikely(f2fs_cp_error(sbi)))
+	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			ClearPageUptodate(page);
+			dec_page_count(sbi, F2FS_DIRTY_META);
+			unlock_page(page);
+			return 0;
+		}
 		goto redirty_out;
+	}
 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 		goto redirty_out;
 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
fs/f2fs/data.c +0 -0
index 48a622b95b76..94b342802513 100644
@@ -2682,6 +2682,12 @@ int f2fs_write_single_data_page(struct page *page, int *submitted,
 
Quoted 11 lines

fs/f2fs/data.c int f2fs_write_single_data_page(struct page *page, int *submitted, · line 2690

Quoted 8 lines
fs/f2fs/data.c +6 -0
 	/* we should bypass data pages to proceed the kworkder jobs */
 	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			ClearPageUptodate(page);
+			inode_dec_dirty_pages(inode);
+			unlock_page(page);
+			return 0;
+		}

Oh, I notice previously, we will drop non-directory inode's dirty pages directly, however, during umount, we'd better drop directory inode's dirty pages as well, right?

Hmm, I remember I dropped them before. Need to double check.

Quoted 48 lines
Quoted 48 lines
fs/f2fs/data.c +0 -0
 		mapping_set_error(page->mapping, -EIO);
 		/*
 		 * don't drop any dirty dentry pages for keeping lastest
Quoted 42 lines
Quoted 37 lines

Thanks,

Quoted 32 lines

in an inifinite loop. Let's drop dirty pages at umount in that case.

---
v3:
- fix wrong unlock

v2:
- fix typos

 fs/f2fs/node.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
fs/f2fs/node.c +8 -1
index e632de10aedab..e0bb0f7e0506e 100644
@@ -1520,8 +1520,15 @@ static int __write_node_page(struct page *page, bool atomic, bool *submitted,
 
 	trace_f2fs_writepage(page, NODE);
 
-	if (unlikely(f2fs_cp_error(sbi)))
+	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			ClearPageUptodate(page);
+			dec_page_count(sbi, F2FS_DIRTY_NODES);
+			unlock_page(page);
+			return 0;
+		}
 		goto redirty_out;
+	}
 
Quoted 3 lines

fs/f2fs/node.c static int __write_node_page(struct page *page, bool atomic, bool *submitted, · line 1534

Quoted 3 lines
fs/f2fs/node.c +0 -0
 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 		goto redirty_out;

.

Quoted 6 lines
Quoted 6 lines

_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
.

.

.

#11 lore raw
In reply to Jaegeuk Kim
Quoted 13 lines · Jaegeuk Kim

Shutdown test is somtime hung, since dirty node pages weren't flushed out. Let's drop dirty pages at umount after shutdown.

---
 fs/f2fs/node.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
fs/f2fs/node.c +0 -0
index e632de10aedab..8c63964a82fd0 100644

fs/f2fs/node.c static int __write_node_page(struct page *page, bool atomic, bool *submitted, · line 1526

Quoted 8 lines
fs/f2fs/node.c +4 -1
@@ -1520,8 +1520,15 @@ static int __write_node_page(struct page *page, bool atomic, bool *submitted,
 
 	trace_f2fs_writepage(page, NODE);
 
-	if (unlikely(f2fs_cp_error(sbi)))
+	if (unlikely(f2fs_cp_error(sbi))) {
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
+			dec_page_count(sbi, F2FS_DIRTY_NODES);
+			up_read(&sbi->node_write);

We don't need to release node_write lock.

Quoted 9 lines
fs/f2fs/node.c +4 -0
+			unlock_page(page);
+			return 0;
+		}
 		goto redirty_out;
+	}
 
 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 		goto redirty_out;