max open zone may be larger than log header number of f2fs, for
such case, it doesn't need to wait last IO in previous zone, let's
introduce available_open_zone semaphore, and reduce it once we
submit first write IO in a zone, and increase it after completion
of last IO in the zone.
Cc: Daeho Jeong <daeh...@gmail.com>
Signed-off-by: Chao Yu <c...@kernel.org>
---
v3:
- avoid race condition in between __submit_merged_bio()
and __allocate_new_segment().
fs/f2fs/data.c | 105 ++++++++++++++++++++++++++++++----------------
fs/f2fs/f2fs.h | 34 ++++++++++++---
fs/f2fs/iostat.c | 7 ++++
fs/f2fs/iostat.h | 2 +
fs/f2fs/segment.c | 43 ++++++++++++++++---
fs/f2fs/segment.h | 12 +++++-
fs/f2fs/super.c | 2 +
7 files changed, 156 insertions(+), 49 deletions(-)
diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index 0d88649c60a5..18a4ac0a06bc 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -373,11 +373,10 @@ static void f2fs_write_end_io(struct bio *bio)
#ifdef CONFIG_BLK_DEV_ZONED
static void f2fs_zone_write_end_io(struct bio *bio)
{
- struct f2fs_bio_info *io = (struct f2fs_bio_info *)bio->bi_private;
+ struct f2fs_sb_info *sbi = iostat_get_bio_private(bio);
- bio->bi_private = io->bi_private;
- complete(&io->zone_wait);
f2fs_write_end_io(bio);
+ up(&sbi->available_open_zones);
}
#endif
@@ -531,6 +530,24 @@ static void __submit_merged_bio(struct f2fs_bio_info *io)
if (!io->bio)
return;
+#ifdef CONFIG_BLK_DEV_ZONED
+ if (io->open_zone) {
+ /*
+ * if there is no open zone, it will wait for last IO in
+ * previous zone before submitting new IO.
+ */
+ down(&fio->sbi->available_open_zones);
+ io->open_zone = false;
+ io->zone_openned = true;
+ }
+
+ if (io->close_zone) {
+ io->bio->bi_end_io = f2fs_zone_write_end_io;
+ io->zone_openned = false;
+ io->close_zone = false;
+ }
+#endif
+
if (is_read_io(fio->op)) {
trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
f2fs_submit_read_bio(io->sbi, io->bio, fio->type);
@@ -601,9 +618,9 @@ int f2fs_init_write_merge_io(struct f2fs_sb_info *sbi)
INIT_LIST_HEAD(&sbi->write_io[i][j].bio_list);
init_f2fs_rwsem(&sbi->write_io[i][j].bio_list_lock);
#ifdef CONFIG_BLK_DEV_ZONED
- init_completion(&sbi->write_io[i][j].zone_wait);
- sbi->write_io[i][j].zone_pending_bio = NULL;
- sbi->write_io[i][j].bi_private = NULL;
+ sbi->write_io[i][j].open_zone = false;
+ sbi->write_io[i][j].zone_openned = false;
+ sbi->write_io[i][j].close_zone = false;
#endif
}
}
@@ -634,6 +651,31 @@ static void __f2fs_submit_merged_write(struct f2fs_sb_info
*sbi,
f2fs_up_write(&io->io_rwsem);
}
+void f2fs_blkzoned_submit_merged_write(struct f2fs_sb_info *sbi, int type)
+{
+#ifdef CONFIG_BLK_DEV_ZONED
+ struct f2fs_bio_info *io;
+
+ if (!f2fs_sb_has_blkzoned(sbi))
+ return;
+
+ io = sbi->write_io[PAGE_TYPE(type)] + type_to_temp(type);
+
+ f2fs_down_write(&io->io_rwsem);
+ if (io->zone_openned) {
+ if (io->bio) {
+ io->close_zone = true;
+ __submit_merged_bio(io);
+ } else if (io->zone_openned) {
+ up(&sbi->available_open_zones);
+ io->zone_openned = false;
+ }
+ }
+ f2fs_up_write(&io->io_rwsem);
+#endif
+
+}
+
static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
struct inode *inode, struct page *page,
nid_t ino, enum page_type type, bool force)
@@ -918,22 +960,16 @@ int f2fs_merge_page_bio(struct f2fs_io_info *fio)
}
#ifdef CONFIG_BLK_DEV_ZONED
-static bool is_end_zone_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr)
+static bool is_blkaddr_zone_boundary(struct f2fs_sb_info *sbi,
+ block_t blkaddr, bool start)
{
- int devi = 0;
+ if (!f2fs_blkaddr_in_seqzone(sbi, blkaddr))
+ return false;
+
+ if (start)
+ return (blkaddr % sbi->blocks_per_blkz) == 0;
+ return (blkaddr % sbi->blocks_per_blkz == sbi->blocks_per_blkz - 1);
- if (f2fs_is_multi_device(sbi)) {
- devi = f2fs_target_device_index(sbi, blkaddr);
- if (blkaddr < FDEV(devi).start_blk ||
- blkaddr > FDEV(devi).end_blk) {
- f2fs_err(sbi, "Invalid block %x", blkaddr);
- return false;
- }
- blkaddr -= FDEV(devi).start_blk;
- }
- return bdev_is_zoned(FDEV(devi).bdev) &&
- f2fs_blkz_is_seq(sbi, devi, blkaddr) &&
- (blkaddr % sbi->blocks_per_blkz == sbi->blocks_per_blkz - 1);
}
#endif
@@ -944,20 +980,14 @@ void f2fs_submit_page_write(struct f2fs_io_info *fio)
struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
struct page *bio_page;
enum count_type type;
+#ifdef CONFIG_BLK_DEV_ZONED
+ bool blkzoned = f2fs_sb_has_blkzoned(sbi) && btype < META;
+#endif
f2fs_bug_on(sbi, is_read_io(fio->op));
f2fs_down_write(&io->io_rwsem);
next:
-#ifdef CONFIG_BLK_DEV_ZONED
- if (f2fs_sb_has_blkzoned(sbi) && btype < META && io->zone_pending_bio) {
- wait_for_completion_io(&io->zone_wait);
- bio_put(io->zone_pending_bio);
- io->zone_pending_bio = NULL;
- io->bi_private = NULL;
- }
-#endif
-
if (fio->in_list) {
spin_lock(&io->io_lock);
if (list_empty(&io->io_list)) {
@@ -985,6 +1015,11 @@ void f2fs_submit_page_write(struct f2fs_io_info *fio)
type = WB_DATA_TYPE(bio_page, fio->compressed_page);
inc_page_count(sbi, type);
+#ifdef CONFIG_BLK_DEV_ZONED
+ if (blkzoned && is_blkaddr_zone_boundary(sbi, fio->new_blkaddr, true))
+ io->open_zone = true;
+#endif
+
if (io->bio &&
(!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
fio->new_blkaddr) ||
@@ -1010,15 +1045,11 @@ void f2fs_submit_page_write(struct f2fs_io_info *fio)
io->last_block_in_bio = fio->new_blkaddr;
trace_f2fs_submit_page_write(fio->page, fio);
+
#ifdef CONFIG_BLK_DEV_ZONED
- if (f2fs_sb_has_blkzoned(sbi) && btype < META &&
- is_end_zone_blkaddr(sbi, fio->new_blkaddr)) {
- bio_get(io->bio);
- reinit_completion(&io->zone_wait);
- io->bi_private = io->bio->bi_private;
- io->bio->bi_private = io;
- io->bio->bi_end_io = f2fs_zone_write_end_io;
- io->zone_pending_bio = io->bio;
+ if (blkzoned &&
+ is_blkaddr_zone_boundary(sbi, fio->new_blkaddr, false)) {
+ io->close_zone = true;
__submit_merged_bio(io);
}
#endif
diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index 694f8a52cb84..8a26530cf4fb 100644
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -1234,16 +1234,16 @@ struct f2fs_bio_info {
struct bio *bio; /* bios to merge */
sector_t last_block_in_bio; /* last block number */
struct f2fs_io_info fio; /* store buffered io info. */
-#ifdef CONFIG_BLK_DEV_ZONED
- struct completion zone_wait; /* condition value for the previous
open zone to close */
- struct bio *zone_pending_bio; /* pending bio for the previous zone */
- void *bi_private; /* previous bi_private for pending bio
*/
-#endif
struct f2fs_rwsem io_rwsem; /* blocking op for bio */
spinlock_t io_lock; /* serialize DATA/NODE IOs */
struct list_head io_list; /* track fios */
struct list_head bio_list; /* bio entry list head */
struct f2fs_rwsem bio_list_lock; /* lock to protect bio entry
list */
+#ifdef CONFIG_BLK_DEV_ZONED
+ bool open_zone; /* open a zone */
+ bool zone_openned; /* zone has been openned */
+ bool close_zone; /* close a zone */
+#endif
};
#define FDEV(i) (sbi->devs[i])
@@ -1560,6 +1560,7 @@ struct f2fs_sb_info {
#ifdef CONFIG_BLK_DEV_ZONED
unsigned int blocks_per_blkz; /* F2FS blocks per zone */
unsigned int max_open_zones; /* max open zone resources of
the zoned device */
+ struct semaphore available_open_zones; /* available open zones */
#endif
/* for node-related operations */
@@ -3822,6 +3823,7 @@ void f2fs_destroy_bio_entry_cache(void);
void f2fs_submit_read_bio(struct f2fs_sb_info *sbi, struct bio *bio,
enum page_type type);
int f2fs_init_write_merge_io(struct f2fs_sb_info *sbi);
+void f2fs_blkzoned_submit_merged_write(struct f2fs_sb_info *sbi, int type);
void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type);
void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
struct inode *inode, struct page *page,
@@ -4469,6 +4471,28 @@ static inline bool f2fs_blkz_is_seq(struct f2fs_sb_info
*sbi, int devi,
return test_bit(zno, FDEV(devi).blkz_seq);
}
+
+static inline bool f2fs_blkaddr_in_seqzone(struct f2fs_sb_info *sbi,
+ block_t blkaddr)
+{
+ int devi = 0;
+
+ if (f2fs_is_multi_device(sbi)) {
+ devi = f2fs_target_device_index(sbi, blkaddr);
+ if (blkaddr < FDEV(devi).start_blk ||
+ blkaddr > FDEV(devi).end_blk) {
+ f2fs_err(sbi, "Invalid block %x", blkaddr);
+ return false;
+ }
+ blkaddr -= FDEV(devi).start_blk;
+ }
+
+ if (!bdev_is_zoned(FDEV(devi).bdev) ||
+ !f2fs_blkz_is_seq(sbi, devi, blkaddr))
+ return false;
+
+ return true;