Skip to contentSkip to conversation

writeback: clear auto_free in initializaiton

PATCH In review [email protected] 3 participants 3 messages Aug 18, 2020 – Sep 2, 2020 VFS

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 writeback: clear auto_free in initializaiton

From: Tom Rix <[email protected]>

Review fs/fs-writeback.c bdi_split_work_to_wbs The CONFIG_CGROUP_WRITEBACK version contains this line

	base_work->auto_free = 0;

Which seems like a strange place to set auto_free as it is not where the rest of base_work is initialized.

In the default version of bdi_split_work_to_wbs, if a successful malloc happens, base_work is copied and auto_free is set to 1, else the base_work is copied to another local valarible and its auto_free is set to 0.

So move the clearing of auto_free to the initialization of the local base_work structures.

---
 fs/fs-writeback.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)
fs/fs-writeback.c +3 -4
index a605c3dddabc..fa1106de2ab0 100644
@@ -881,7 +881,6 @@ static void bdi_split_work_to_wbs(struct backing_dev_info *bdi,
 		work = &fallback_work;
 		*work = *base_work;
 		work->nr_pages = nr_pages;
-		work->auto_free = 0;
 		work->done = &fallback_work_done;
 
 		wb_queue_work(wb, work);
@@ -1055,10 +1054,8 @@ static void bdi_split_work_to_wbs(struct backing_dev_info *bdi,
 {
 	might_sleep();
 
-	if (!skip_if_busy || !writeback_in_progress(&bdi->wb)) {
-		base_work->auto_free = 0;
+	if (!skip_if_busy || !writeback_in_progress(&bdi->wb))
 		wb_queue_work(&bdi->wb, base_work);
-	}
 }
 
 #endif	/* CONFIG_CGROUP_WRITEBACK */
@@ -2459,6 +2456,7 @@ static void __writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr,
 		.done			= &done,
 		.nr_pages		= nr,
 		.reason			= reason,
+		.auto_free		= 0,
 	};
 
 	if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info)
@@ -2538,6 +2536,7 @@ void sync_inodes_sb(struct super_block *sb)
 		.done		= &done,
 		.reason		= WB_REASON_SYNC,
 		.for_sync	= 1,
+		.auto_free	= 0,
 	};
 
 	/*
Signature · 1 line
2.18.1
reviewer
#1 lore raw
Quoted 5 lines · On Tue 18-08-20 07:13:30, [email protected] wrote:

From: Tom Rix <[email protected]>

Review fs/fs-writeback.c bdi_split_work_to_wbs The CONFIG_CGROUP_WRITEBACK version contains this line

	base_work->auto_free = 0;

It is actually the !CONFIG_CGROUP_WRITEBACK version...

Quoted 2 lines

Which seems like a strange place to set auto_free as it is not where the rest of base_work is initialized.

Otherwise I agree it's a strange place. I've added Tejun to CC just in case he remembers why he's added that.

Quoted 10 lines

In the default version of bdi_split_work_to_wbs, if a successful malloc happens, base_work is copied and auto_free is set to 1, else the base_work is copied to another local valarible and its auto_free is set to 0.

So move the clearing of auto_free to the initialization of the local base_work structures.

Some more comments below.

Quoted 9 lines
---
 fs/fs-writeback.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)
fs/fs-writeback.c +0 -0
index a605c3dddabc..fa1106de2ab0 100644

fs/fs-writeback.c static void bdi_split_work_to_wbs(struct backing_dev_info *bdi, · line 884

Quoted 5 lines
fs/fs-writeback.c +0 -1
@@ -881,7 +881,6 @@ static void bdi_split_work_to_wbs(struct backing_dev_info *bdi,
 		work = &fallback_work;
 		*work = *base_work;
 		work->nr_pages = nr_pages;
-		work->auto_free = 0;
 		work->done = &fallback_work_done;

Honestly, I'd leave this alone. Although base_work should have auto_free == 0, this assignment IMO helps readability.

fs/fs-writeback.c static void bdi_split_work_to_wbs(struct backing_dev_info *bdi, · line 1061

Quoted 9 lines
fs/fs-writeback.c +1 -3
@@ -1055,10 +1054,8 @@ static void bdi_split_work_to_wbs(struct backing_dev_info *bdi,
 {
 	might_sleep();
 
-	if (!skip_if_busy || !writeback_in_progress(&bdi->wb)) {
-		base_work->auto_free = 0;
+	if (!skip_if_busy || !writeback_in_progress(&bdi->wb))
 		wb_queue_work(&bdi->wb, base_work);
-	}

Agreed with this.

Quoted 9 lines
fs/fs-writeback.c +1 -0
@@ -2459,6 +2456,7 @@ static void __writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr,
 		.done			= &done,
 		.nr_pages		= nr,
 		.reason			= reason,
+		.auto_free		= 0,
 	};
 
 	if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info)

fs/fs-writeback.c void sync_inodes_sb(struct super_block *sb) · line 2540

Quoted 5 lines
fs/fs-writeback.c +1 -0
@@ -2538,6 +2536,7 @@ void sync_inodes_sb(struct super_block *sb)
 		.done		= &done,
 		.reason		= WB_REASON_SYNC,
 		.for_sync	= 1,
+		.auto_free	= 0,
 	};

No need for explicit initialization to 0 - that is implicit with the initializers.

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
#2 lore raw

Hello,

Quoted 5 lines · Jan Kara
Quoted 2 lines

Which seems like a strange place to set auto_free as it is not where the rest of base_work is initialized.

Otherwise I agree it's a strange place. I've added Tejun to CC just in case he remembers why he's added that.

I don't remember exactly but maybe I was trying to mirror the CGROUP_WRITEBACK counterpart without actually thinking about it? The patch looks good to me.

Thanks.

Signature · 1 line
tejun