Received-SPF: pass (google.com: domain of linux-ext4-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Fengnan Chang <changfengnan@bytedance.com>
To:     wangjianchao@kuaishou.com, tytso@mit.edu
Cc:     linux-ext4@vger.kernel.org,
        Fengnan Chang <changfengnan@bytedance.com>
Subject: [PATCH] ext4: improve discard efficiency
Date:   Tue,  4 Jul 2023 19:56:13 +0800
Message-Id: <20230704115613.88313-1-changfengnan@bytedance.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

In commit a015434480dc("ext4: send parallel discards on commit
completions"), issue all discard commands in parallel make all
bios could merged into one request, so lowlevel drive can issue
multi segments in one time which is more efficiency, but commit
55cdd0af2bc5 ("ext4: get discard out of jbd2 commit kthread contex")
seems broke this way, let's fix it.
In my test, the time of fstrim fs with multi big sparse file
reduce from 6.7s to 1.3s.

Signed-off-by: Fengnan Chang <changfengnan@bytedance.com>
---
 fs/ext4/mballoc.c | 31 +++++++++++++++++++++++++++----
 1 file changed, 27 insertions(+), 4 deletions(-)

diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
index a2475b8c9fb5..e8fbc5f7d541 100644
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -6790,7 +6790,7 @@ int ext4_group_add_blocks(handle_t *handle, struct super_block *sb,
  * be called with under the group lock.
  */
 static int ext4_trim_extent(struct super_block *sb,
-		int start, int count, struct ext4_buddy *e4b)
+		int start, int count, struct ext4_buddy *e4b, struct bio **biop)
 __releases(bitlock)
 __acquires(bitlock)
 {
@@ -6812,9 +6812,8 @@ __acquires(bitlock)
 	 */
 	mb_mark_used(e4b, &ex);
 	ext4_unlock_group(sb, group);
-	ret = ext4_issue_discard(sb, group, start, count, NULL);
+	ret = ext4_issue_discard(sb, group, start, count, biop);
 	ext4_lock_group(sb, group);
-	mb_free_blocks(NULL, e4b, start, ex.fe_len);
 	return ret;
 }
 
@@ -6826,6 +6825,12 @@ __releases(ext4_group_lock_ptr(sb, e4b->bd_group))
 {
 	ext4_grpblk_t next, count, free_count;
 	void *bitmap;
+	struct ext4_free_data *entry = NULL, *fd, *nfd;
+	struct list_head discard_data_list;
+	struct bio *discard_bio = NULL;
+	struct blk_plug plug;
+
+	INIT_LIST_HEAD(&discard_data_list);
 
 	bitmap = e4b->bd_bitmap;
 	start = (e4b->bd_info->bb_first_free > start) ?
@@ -6833,6 +6838,7 @@ __releases(ext4_group_lock_ptr(sb, e4b->bd_group))
 	count = 0;
 	free_count = 0;
 
+	blk_start_plug(&plug);
 	while (start <= max) {
 		start = mb_find_next_zero_bit(bitmap, max + 1, start);
 		if (start > max)
@@ -6840,10 +6846,15 @@ __releases(ext4_group_lock_ptr(sb, e4b->bd_group))
 		next = mb_find_next_bit(bitmap, max + 1, start);
 
 		if ((next - start) >= minblocks) {
-			int ret = ext4_trim_extent(sb, start, next - start, e4b);
+			int ret = ext4_trim_extent(sb, start, next - start, e4b, &discard_bio);
 
 			if (ret && ret != -EOPNOTSUPP)
 				break;
+			entry = kmem_cache_alloc(ext4_free_data_cachep,
+				GFP_NOFS|__GFP_NOFAIL);
+			entry->efd_start_cluster = start;
+			entry->efd_count = next - start;
+			list_add_tail(&entry->efd_list, &discard_data_list);
 			count += next - start;
 		}
 		free_count += next - start;
@@ -6863,6 +6874,18 @@ __releases(ext4_group_lock_ptr(sb, e4b->bd_group))
 		if ((e4b->bd_info->bb_free - free_count) < minblocks)
 			break;
 	}
+	if (discard_bio) {
+		ext4_unlock_group(sb, e4b->bd_group);
+		submit_bio_wait(discard_bio);
+		bio_put(discard_bio);
+		ext4_lock_group(sb, e4b->bd_group);
+	}
+	blk_finish_plug(&plug);
+
+	list_for_each_entry_safe(fd, nfd, &discard_data_list, efd_list) {
+		mb_free_blocks(NULL, e4b, fd->efd_start_cluster, fd->efd_count);
+		kmem_cache_free(ext4_free_data_cachep, fd);
+	}
 
 	return count;
 }
-- 
2.37.1 (Apple Git-137.1)