Update bcachefs sources to bcca1c557b bcachefs: fixes for getting stuck flushing journal pins

4 files changed, 29 insertions, 33 deletions

diff --git a/.bcachefs_revision b/.bcachefs_revision
index 3bcc585a..088f645c 100644
--- a/.bcachefs_revision
+++ b/.bcachefs_revision
@@ -1 +1 @@
-ef60854e9912d24c0ba83e0760552c98257d2b07
+bcca1c557b1897ecc3aeb1f89ab91865487d91ab
diff --git a/libbcachefs/fs-io.c b/libbcachefs/fs-io.c
index fdc24be1..c1739f53 100644
--- a/libbcachefs/fs-io.c
+++ b/libbcachefs/fs-io.c
@@ -56,7 +56,7 @@ struct bch_writepage_io {
 struct dio_write {
 	struct closure			cl;
 	struct kiocb			*req;
-	struct task_struct		*task;
+	struct mm_struct		*mm;
 	unsigned			loop:1,
 					sync:1,
 					free_iov:1;
@@ -1753,6 +1753,7 @@ static void bch2_dio_write_loop_async(struct closure *);
 
 static long bch2_dio_write_loop(struct dio_write *dio)
 {
+	bool kthread = (current->flags & PF_KTHREAD) != 0;
 	struct kiocb *req = dio->req;
 	struct address_space *mapping = req->ki_filp->f_mapping;
 	struct bch_inode_info *inode = dio->iop.inode;
@@ -1777,13 +1778,13 @@ static long bch2_dio_write_loop(struct dio_write *dio)
 	while (1) {
 		BUG_ON(current->pagecache_lock);
 		current->pagecache_lock = &mapping->add_lock;
-		if (current != dio->task)
-			use_mm(dio->task->mm);
+		if (kthread)
+			use_mm(dio->mm);
 
 		ret = bio_iov_iter_get_pages(bio, &dio->iter);
 
-		if (current != dio->task)
-			unuse_mm(dio->task->mm);
+		if (kthread)
+			unuse_mm(dio->mm);
 		current->pagecache_lock = NULL;
 
 		if (unlikely(ret < 0))
@@ -1894,7 +1895,7 @@ static int bch2_direct_IO_write(struct kiocb *req,
 	dio = container_of(bio, struct dio_write, iop.op.wbio.bio);
 	closure_init(&dio->cl, NULL);
 	dio->req		= req;
-	dio->task		= current;
+	dio->mm			= current->mm;
 	dio->loop		= false;
 	dio->sync		= is_sync_kiocb(req) ||
 		offset + iter->count > inode->v.i_size;
@@ -1902,7 +1903,7 @@ static int bch2_direct_IO_write(struct kiocb *req,
 	dio->quota_res.sectors	= 0;
 	dio->iter		= *iter;
 	bch2_fswrite_op_init(&dio->iop, c, inode, io_opts(c, inode), true);
-	dio->iop.op.write_point	= writepoint_hashed((unsigned long) dio->task);
+	dio->iop.op.write_point	= writepoint_hashed((unsigned long) current);
 	dio->iop.op.flags |= BCH_WRITE_NOPUT_RESERVATION;
 
 	if ((req->ki_flags & IOCB_DSYNC) &&
diff --git a/libbcachefs/journal_reclaim.c b/libbcachefs/journal_reclaim.c
index d8d1b6b0..4a997366 100644
--- a/libbcachefs/journal_reclaim.c
+++ b/libbcachefs/journal_reclaim.c
@@ -64,6 +64,9 @@ static inline void __journal_pin_drop(struct journal *j,
 	if (atomic_dec_and_test(&pin_list->count) &&
 	    pin_list == &fifo_peek_front(&j->pin))
 		bch2_journal_reclaim_fast(j);
+	else if (fifo_used(&j->pin) == 1 &&
+		 atomic_read(&pin_list->count) == 1)
+		journal_wake(j);
 }
 
 void bch2_journal_pin_drop(struct journal *j,
@@ -336,56 +339,48 @@ void bch2_journal_reclaim_work(struct work_struct *work)
 				   msecs_to_jiffies(j->reclaim_delay_ms));
 }
 
-static int journal_flush_done(struct journal *j, u64 seq_to_flush,
-			      struct journal_entry_pin **pin,
-			      u64 *pin_seq)
+static int journal_flush_done(struct journal *j, u64 seq_to_flush)
 {
+	struct journal_entry_pin *pin;
+	u64 pin_seq;
 	int ret;
 
-	*pin = NULL;
-
 	ret = bch2_journal_error(j);
 	if (ret)
 		return ret;
 
+	mutex_lock(&j->reclaim_lock);
 	spin_lock(&j->lock);
+
+	while ((pin = journal_get_next_pin(j, seq_to_flush, &pin_seq))) {
+		journal_pin_mark_flushing(j, pin, pin_seq);
+		spin_unlock(&j->lock);
+
+		journal_pin_flush(j, pin, pin_seq);
+
+		spin_lock(&j->lock);
+	}
 	/*
 	 * If journal replay hasn't completed, the unreplayed journal entries
 	 * hold refs on their corresponding sequence numbers
 	 */
-	ret = (*pin = journal_get_next_pin(j, seq_to_flush, pin_seq)) != NULL ||
-		!test_bit(JOURNAL_REPLAY_DONE, &j->flags) ||
+	ret = !test_bit(JOURNAL_REPLAY_DONE, &j->flags) ||
 		journal_last_seq(j) > seq_to_flush ||
 		(fifo_used(&j->pin) == 1 &&
 		 atomic_read(&fifo_peek_front(&j->pin).count) == 1);
-	if (*pin)
-		journal_pin_mark_flushing(j, *pin, *pin_seq);
 
 	spin_unlock(&j->lock);
+	mutex_unlock(&j->reclaim_lock);
 
 	return ret;
 }
 
 void bch2_journal_flush_pins(struct journal *j, u64 seq_to_flush)
 {
-	struct journal_entry_pin *pin;
-	u64 pin_seq;
-
 	if (!test_bit(JOURNAL_STARTED, &j->flags))
 		return;
 
-	mutex_lock(&j->reclaim_lock);
-
-	while (1) {
-		wait_event(j->wait, journal_flush_done(j, seq_to_flush,
-						       &pin, &pin_seq));
-		if (!pin)
-			break;
-
-		journal_pin_flush(j, pin, pin_seq);
-	}
-
-	mutex_unlock(&j->reclaim_lock);
+	closure_wait_event(&j->async_wait, journal_flush_done(j, seq_to_flush));
 }
 
 int bch2_journal_flush_device_pins(struct journal *j, int dev_idx)
diff --git a/linux/sched.c b/linux/sched.c
index b3fc7739..353c11db 100644
--- a/linux/sched.c
+++ b/linux/sched.c
@@ -37,7 +37,7 @@ void schedule(void)
 
 	rcu_quiescent_state();
 
-	while ((v = current->state) != TASK_RUNNING)
+	while ((v = READ_ONCE(current->state)) != TASK_RUNNING)
 		futex(&current->state, FUTEX_WAIT|FUTEX_PRIVATE_FLAG,
 		      v, NULL, NULL, 0);
 }