REQ_FUA did not flush metadata in SSD mode, this commit fixes it.
Signed-off-by: Mikulas Patocka <[email protected]>
Cc: [email protected]
Fixes: c1005322ff02 ("dm writecache: handle REQ_FUA")
Assisted-by: Claude:claude-opus-5
Assisted-by: Codex:gpt-6-sol
---
drivers/md/dm-writecache.c | 45 +++++++++++++++++++++++++++++++++++----------
1 file changed, 35 insertions(+), 10 deletions(-)
Index: linux-2.6/drivers/md/dm-writecache.c
===================================================================
--- linux-2.6.orig/drivers/md/dm-writecache.c
+++ linux-2.6/drivers/md/dm-writecache.c
@@ -194,6 +194,7 @@ struct dm_writecache {
struct task_struct *endio_thread;
struct task_struct *flush_thread;
+ spinlock_t flush_list_lock;
struct bio_list flush_list;
struct dm_kcopyd_client *dm_kcopyd;
@@ -1288,9 +1289,11 @@ static int writecache_flush_thread(void
struct bio *bio;
wc_lock(wc);
+ spin_lock_irq(&wc->flush_list_lock);
bio = bio_list_pop(&wc->flush_list);
if (!bio) {
set_current_state(TASK_INTERRUPTIBLE);
+ spin_unlock_irq(&wc->flush_list_lock);
wc_unlock(wc);
if (unlikely(kthread_should_stop())) {
@@ -1301,6 +1304,7 @@ static int writecache_flush_thread(void
schedule();
continue;
}
+ spin_unlock_irq(&wc->flush_list_lock);
if (bio_op(bio) == REQ_OP_DISCARD) {
writecache_discard(wc, bio->bi_iter.bi_sector,
@@ -1322,9 +1326,12 @@ static int writecache_flush_thread(void
static void writecache_offload_bio(struct dm_writecache *wc, struct bio *bio)
{
+ unsigned long flags;
+ spin_lock_irqsave(&wc->flush_list_lock, flags);
if (bio_list_empty(&wc->flush_list))
wake_up_process(wc->flush_thread);
bio_list_add(&wc->flush_list, bio);
+ spin_unlock_irqrestore(&wc->flush_list_lock, flags);
}
enum wc_map_op {
@@ -1436,6 +1443,8 @@ static void writecache_bio_copy_ssd(stru
static enum wc_map_op writecache_map_write(struct dm_writecache *wc, struct
bio *bio)
{
struct wc_entry *e;
+ enum wc_map_op ret = WC_MAP_SUBMIT;
+ bool need_flush = false;
do {
bool found_entry = false;
@@ -1472,7 +1481,8 @@ direct_write:
writecache_map_remap_origin(wc, bio, e);
wc->stats.writes_around += bio->bi_iter.bi_size
>> wc->block_size_bits;
wc->stats.writes += bio->bi_iter.bi_size >>
wc->block_size_bits;
- return WC_MAP_REMAP_ORIGIN;
+ ret = WC_MAP_REMAP_ORIGIN;
+ goto flush_pmem_and_ret;
}
wc->stats.writes_blocked_on_freelist++;
writecache_wait_on_freelist(wc);
@@ -1486,21 +1496,25 @@ bio_copy:
if (WC_MODE_PMEM(wc)) {
bio_copy_block(wc, bio, memory_data(wc, e));
wc->stats.writes++;
+ need_flush = true;
} else {
writecache_bio_copy_ssd(wc, bio, e, search_used);
return WC_MAP_REMAP;
}
} while (bio->bi_iter.bi_size);
- if (unlikely(bio->bi_opf & REQ_FUA || wc->uncommitted_blocks >=
wc->autocommit_blocks)) {
- writecache_flush(wc);
- if (writecache_has_error(wc))
- bio->bi_status = BLK_STS_IOERR;
- } else {
- writecache_schedule_autocommit(wc);
+flush_pmem_and_ret:
+ if (need_flush) {
+ if (unlikely(bio->bi_opf & REQ_FUA || wc->uncommitted_blocks >=
wc->autocommit_blocks)) {
+ writecache_flush(wc);
+ if (writecache_has_error(wc))
+ ret = WC_MAP_ERROR;
+ } else {
+ writecache_schedule_autocommit(wc);
+ }
}
- return WC_MAP_SUBMIT;
+ return ret;
}
static enum wc_map_op writecache_map_flush(struct dm_writecache *wc, struct
bio *bio)
@@ -1590,7 +1604,10 @@ done:
case WC_MAP_REMAP:
/* make sure that writecache_end_io decrements bio_in_progress:
*/
- bio->bi_private = (void *)1;
+ if (unlikely(bio->bi_opf & REQ_FUA) && bio_op(bio) ==
REQ_OP_WRITE)
+ bio->bi_private = (void *)3;
+ else
+ bio->bi_private = (void *)1;
atomic_inc(&wc->bio_in_progress[bio_data_dir(bio)]);
wc_unlock(wc);
return DM_MAPIO_REMAPPED;
@@ -1620,12 +1637,19 @@ static int writecache_end_io(struct dm_t
{
struct dm_writecache *wc = ti->private;
- if (bio->bi_private == (void *)1) {
+ if (bio->bi_private == (void *)1 || bio->bi_private == (void *)3) {
int dir = bio_data_dir(bio);
+ bool fua = bio->bi_private == (void *)3;
if (atomic_dec_and_test(&wc->bio_in_progress[dir]))
if
(unlikely(waitqueue_active(&wc->bio_in_progress_wait[dir])))
wake_up(&wc->bio_in_progress_wait[dir]);
+
+ if (fua && likely(*status == BLK_STS_OK)) {
+ bio->bi_private = NULL;
+ writecache_offload_bio(wc, bio);
+ return DM_ENDIO_INCOMPLETE;
+ }
} else if (bio->bi_private == (void *)2) {
dm_iot_io_end(&wc->iot, 1);
}
@@ -2522,6 +2546,7 @@ invalid_optional:
wc->memory_map_size -= (uint64_t)wc->start_sector <<
SECTOR_SHIFT;
+ spin_lock_init(&wc->flush_list_lock);
bio_list_init(&wc->flush_list);
wc->flush_thread = kthread_run(writecache_flush_thread, wc,
"dm_writecache_flush");
if (IS_ERR(wc->flush_thread)) {
