X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Fjournal.c;fp=libbcachefs%2Fjournal.c;h=214c8030048292430b07721bd04bac8ea3c44f50;hb=f3f005c76eb5636542a8f5b137bd1904d57e8f86;hp=bc890776eb57933a5931edd2a2f07570f52b7ab3;hpb=1ef396b684a419b5a50ab215103486d189068800;p=bcachefs-tools-debian

diff --git a/libbcachefs/journal.c b/libbcachefs/journal.c
index bc89077..214c803 100644
--- a/libbcachefs/journal.c
+++ b/libbcachefs/journal.c
@@ -27,6 +27,26 @@ static const char * const bch2_journal_errors[] = {
 	NULL
 };
 
+static inline bool journal_seq_unwritten(struct journal *j, u64 seq)
+{
+	return seq > j->seq_ondisk;
+}
+
+static bool __journal_entry_is_open(union journal_res_state state)
+{
+	return state.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL;
+}
+
+static inline unsigned nr_unwritten_journal_entries(struct journal *j)
+{
+	return atomic64_read(&j->seq) - j->seq_ondisk;
+}
+
+static bool journal_entry_is_open(struct journal *j)
+{
+	return __journal_entry_is_open(j->reservations);
+}
+
 static void bch2_journal_buf_to_text(struct printbuf *out, struct journal *j, u64 seq)
 {
 	union journal_res_state s = READ_ONCE(j->reservations);
@@ -54,6 +74,13 @@ static void bch2_journal_buf_to_text(struct printbuf *out, struct journal *j, u6
 	prt_printf(out, "%li jiffies", buf->expires - jiffies);
 	prt_newline(out);
 
+	if (buf->write_done)
+		prt_printf(out, "write done\n");
+	else if (buf->write_allocated)
+		prt_printf(out, "write allocated\n");
+	else if (buf->write_started)
+		prt_printf(out, "write started\n");
+
 	printbuf_indent_sub(out, 2);
 }
 
@@ -66,26 +93,7 @@ static void bch2_journal_bufs_to_text(struct printbuf *out, struct journal *j)
 	     seq <= journal_cur_seq(j);
 	     seq++)
 		bch2_journal_buf_to_text(out, j, seq);
-}
-
-static inline bool journal_seq_unwritten(struct journal *j, u64 seq)
-{
-	return seq > j->seq_ondisk;
-}
-
-static bool __journal_entry_is_open(union journal_res_state state)
-{
-	return state.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL;
-}
-
-static inline unsigned nr_unwritten_journal_entries(struct journal *j)
-{
-	return atomic64_read(&j->seq) - j->seq_ondisk;
-}
-
-static bool journal_entry_is_open(struct journal *j)
-{
-	return __journal_entry_is_open(j->reservations);
+	prt_printf(out, "last buf %s\n", journal_entry_is_open(j) ? "open" : "closed");
 }
 
 static inline struct journal_buf *
@@ -174,21 +182,40 @@ journal_error_check_stuck(struct journal *j, int error, unsigned flags)
 	return stuck;
 }
 
+void bch2_journal_do_writes(struct journal *j)
+{
+	for (u64 seq = journal_last_unwritten_seq(j);
+	     seq <= journal_cur_seq(j);
+	     seq++) {
+		unsigned idx = seq & JOURNAL_BUF_MASK;
+		struct journal_buf *w = j->buf + idx;
+
+		if (w->write_started && !w->write_allocated)
+			break;
+		if (w->write_started)
+			continue;
+
+		if (!journal_state_count(j->reservations, idx)) {
+			w->write_started = true;
+			closure_call(&w->io, bch2_journal_write, j->wq, NULL);
+		}
+
+		break;
+	}
+}
+
 /*
  * Final processing when the last reference of a journal buffer has been
  * dropped. Drop the pin list reference acquired at journal entry open and write
  * the buffer, if requested.
  */
-void bch2_journal_buf_put_final(struct journal *j, u64 seq, bool write)
+void bch2_journal_buf_put_final(struct journal *j, u64 seq)
 {
-	struct bch_fs *c = container_of(j, struct bch_fs, journal);
-
 	lockdep_assert_held(&j->lock);
 
 	if (__bch2_journal_pin_put(j, seq))
 		bch2_journal_reclaim_fast(j);
-	if (write)
-		closure_call(&j->io, bch2_journal_write, c->io_complete_wq, NULL);
+	bch2_journal_do_writes(j);
 }
 
 /*
@@ -380,11 +407,14 @@ static int journal_entry_open(struct journal *j)
 	BUG_ON(j->buf + (journal_cur_seq(j) & JOURNAL_BUF_MASK) != buf);
 
 	bkey_extent_init(&buf->key);
-	buf->noflush	= false;
-	buf->must_flush	= false;
-	buf->separate_flush = false;
-	buf->flush_time	= 0;
+	buf->noflush		= false;
+	buf->must_flush		= false;
+	buf->separate_flush	= false;
+	buf->flush_time		= 0;
 	buf->need_flush_to_write_buffer = true;
+	buf->write_started	= false;
+	buf->write_allocated	= false;
+	buf->write_done		= false;
 
 	memset(buf->data, 0, sizeof(*buf->data));
 	buf->data->seq	= cpu_to_le64(journal_cur_seq(j));
@@ -418,9 +448,10 @@ static int journal_entry_open(struct journal *j)
 	} while ((v = atomic64_cmpxchg(&j->reservations.counter,
 				       old.v, new.v)) != old.v);
 
-	mod_delayed_work(c->io_complete_wq,
-			 &j->write_work,
-			 msecs_to_jiffies(c->opts.journal_flush_delay));
+	if (nr_unwritten_journal_entries(j) == 1)
+		mod_delayed_work(j->wq,
+				 &j->write_work,
+				 msecs_to_jiffies(c->opts.journal_flush_delay));
 	journal_wake(j);
 
 	if (j->early_journal_entries.nr)
@@ -445,20 +476,16 @@ static void journal_quiesce(struct journal *j)
 static void journal_write_work(struct work_struct *work)
 {
 	struct journal *j = container_of(work, struct journal, write_work.work);
-	struct bch_fs *c = container_of(j, struct bch_fs, journal);
-	long delta;
 
 	spin_lock(&j->lock);
-	if (!__journal_entry_is_open(j->reservations))
-		goto unlock;
-
-	delta = journal_cur_buf(j)->expires - jiffies;
+	if (__journal_entry_is_open(j->reservations)) {
+		long delta = journal_cur_buf(j)->expires - jiffies;
 
-	if (delta > 0)
-		mod_delayed_work(c->io_complete_wq, &j->write_work, delta);
-	else
-		__journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL, true);
-unlock:
+		if (delta > 0)
+			mod_delayed_work(j->wq, &j->write_work, delta);
+		else
+			__journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL, true);
+	}
 	spin_unlock(&j->lock);
 }
 
@@ -473,33 +500,32 @@ retry:
 	if (journal_res_get_fast(j, res, flags))
 		return 0;
 
-	if (bch2_journal_error(j))
-		return -BCH_ERR_erofs_journal_err;
+	if ((flags & BCH_WATERMARK_MASK) < j->watermark) {
+		ret = JOURNAL_ERR_journal_full;
+		can_discard = j->can_discard;
+		goto out;
+	}
 
-	spin_lock(&j->lock);
+	if (j->blocked)
+		return -BCH_ERR_journal_res_get_blocked;
 
-	/* check once more in case somebody else shut things down... */
-	if (bch2_journal_error(j)) {
-		spin_unlock(&j->lock);
+	if (bch2_journal_error(j))
 		return -BCH_ERR_erofs_journal_err;
+
+	if (nr_unwritten_journal_entries(j) == ARRAY_SIZE(j->buf) && !journal_entry_is_open(j)) {
+		ret = JOURNAL_ERR_max_in_flight;
+		goto out;
 	}
 
+	spin_lock(&j->lock);
+
 	/*
 	 * Recheck after taking the lock, so we don't race with another thread
 	 * that just did journal_entry_open() and call bch2_journal_entry_close()
 	 * unnecessarily
 	 */
 	if (journal_res_get_fast(j, res, flags)) {
-		spin_unlock(&j->lock);
-		return 0;
-	}
-
-	if ((flags & BCH_WATERMARK_MASK) < j->watermark) {
-		/*
-		 * Don't want to close current journal entry, just need to
-		 * invoke reclaim:
-		 */
-		ret = JOURNAL_ERR_journal_full;
+		ret = 0;
 		goto unlock;
 	}
 
@@ -515,30 +541,30 @@ retry:
 		j->buf_size_want = max(j->buf_size_want, buf->buf_size << 1);
 
 	__journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL, false);
-	ret = journal_entry_open(j);
-
-	if (ret == JOURNAL_ERR_max_in_flight) {
-		track_event_change(&c->times[BCH_TIME_blocked_journal_max_in_flight],
-				   &j->max_in_flight_start, true);
-		if (trace_journal_entry_full_enabled()) {
-			struct printbuf buf = PRINTBUF;
-			buf.atomic++;
-
-			bch2_journal_bufs_to_text(&buf, j);
-			trace_journal_entry_full(c, buf.buf);
-			printbuf_exit(&buf);
-		}
-		count_event(c, journal_entry_full);
-	}
+	ret = journal_entry_open(j) ?: JOURNAL_ERR_retry;
 unlock:
 	can_discard = j->can_discard;
 	spin_unlock(&j->lock);
-
-	if (!ret)
+out:
+	if (ret == JOURNAL_ERR_retry)
 		goto retry;
+	if (!ret)
+		return 0;
+
 	if (journal_error_check_stuck(j, ret, flags))
 		ret = -BCH_ERR_journal_res_get_blocked;
 
+	if (ret == JOURNAL_ERR_max_in_flight &&
+	    track_event_change(&c->times[BCH_TIME_blocked_journal_max_in_flight], true)) {
+
+		struct printbuf buf = PRINTBUF;
+		prt_printf(&buf, "seq %llu\n", journal_cur_seq(j));
+		bch2_journal_bufs_to_text(&buf, j);
+		trace_journal_entry_full(c, buf.buf);
+		printbuf_exit(&buf);
+		count_event(c, journal_entry_full);
+	}
+
 	/*
 	 * Journal is full - can't rely on reclaim from work item due to
 	 * freezing:
@@ -727,7 +753,7 @@ int bch2_journal_flush_seq(struct journal *j, u64 seq)
 	ret = wait_event_interruptible(j->wait, (ret2 = bch2_journal_flush_seq_async(j, seq, NULL)));
 
 	if (!ret)
-		bch2_time_stats_update(j->flush_seq_time, start_time);
+		time_stats_update(j->flush_seq_time, start_time);
 
 	return ret ?: ret2 < 0 ? ret2 : 0;
 }
@@ -1157,7 +1183,6 @@ int bch2_fs_journal_start(struct journal *j, u64 cur_seq)
 	struct journal_replay *i, **_i;
 	struct genradix_iter iter;
 	bool had_entries = false;
-	unsigned ptr;
 	u64 last_seq = cur_seq, nr, seq;
 
 	genradix_for_each_reverse(&c->journal_entries, iter, _i) {
@@ -1211,8 +1236,8 @@ int bch2_fs_journal_start(struct journal *j, u64 cur_seq)
 		p = journal_seq_pin(j, seq);
 
 		p->devs.nr = 0;
-		for (ptr = 0; ptr < i->nr_ptrs; ptr++)
-			bch2_dev_list_add_dev(&p->devs, i->ptrs[ptr].dev);
+		darray_for_each(i->ptrs, ptr)
+			bch2_dev_list_add_dev(&p->devs, ptr->dev);
 
 		had_entries = true;
 	}
@@ -1240,13 +1265,17 @@ int bch2_fs_journal_start(struct journal *j, u64 cur_seq)
 
 void bch2_dev_journal_exit(struct bch_dev *ca)
 {
-	kfree(ca->journal.bio);
-	kfree(ca->journal.buckets);
-	kfree(ca->journal.bucket_seq);
+	struct journal_device *ja = &ca->journal;
+
+	for (unsigned i = 0; i < ARRAY_SIZE(ja->bio); i++) {
+		kfree(ja->bio[i]);
+		ja->bio[i] = NULL;
+	}
 
-	ca->journal.bio		= NULL;
-	ca->journal.buckets	= NULL;
-	ca->journal.bucket_seq	= NULL;
+	kfree(ja->buckets);
+	kfree(ja->bucket_seq);
+	ja->buckets	= NULL;
+	ja->bucket_seq	= NULL;
 }
 
 int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb)
@@ -1256,14 +1285,13 @@ int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb)
 		bch2_sb_field_get(sb, journal);
 	struct bch_sb_field_journal_v2 *journal_buckets_v2 =
 		bch2_sb_field_get(sb, journal_v2);
-	unsigned i, nr_bvecs;
 
 	ja->nr = 0;
 
 	if (journal_buckets_v2) {
 		unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
 
-		for (i = 0; i < nr; i++)
+		for (unsigned i = 0; i < nr; i++)
 			ja->nr += le64_to_cpu(journal_buckets_v2->d[i].nr);
 	} else if (journal_buckets) {
 		ja->nr = bch2_nr_journal_buckets(journal_buckets);
@@ -1273,13 +1301,18 @@ int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb)
 	if (!ja->bucket_seq)
 		return -BCH_ERR_ENOMEM_dev_journal_init;
 
-	nr_bvecs = DIV_ROUND_UP(JOURNAL_ENTRY_SIZE_MAX, PAGE_SIZE);
+	unsigned nr_bvecs = DIV_ROUND_UP(JOURNAL_ENTRY_SIZE_MAX, PAGE_SIZE);
 
-	ca->journal.bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
-	if (!ca->journal.bio)
-		return -BCH_ERR_ENOMEM_dev_journal_init;
+	for (unsigned i = 0; i < ARRAY_SIZE(ja->bio); i++) {
+		ja->bio[i] = kmalloc(struct_size(ja->bio[i], bio.bi_inline_vecs,
+				     nr_bvecs), GFP_KERNEL);
+		if (!ja->bio[i])
+			return -BCH_ERR_ENOMEM_dev_journal_init;
 
-	bio_init(ca->journal.bio, NULL, ca->journal.bio->bi_inline_vecs, nr_bvecs, 0);
+		ja->bio[i]->ca = ca;
+		ja->bio[i]->buf_idx = i;
+		bio_init(&ja->bio[i]->bio, NULL, ja->bio[i]->bio.bi_inline_vecs, nr_bvecs, 0);
+	}
 
 	ja->buckets = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
 	if (!ja->buckets)
@@ -1287,14 +1320,14 @@ int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb)
 
 	if (journal_buckets_v2) {
 		unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
-		unsigned j, dst = 0;
+		unsigned dst = 0;
 
-		for (i = 0; i < nr; i++)
-			for (j = 0; j < le64_to_cpu(journal_buckets_v2->d[i].nr); j++)
+		for (unsigned i = 0; i < nr; i++)
+			for (unsigned j = 0; j < le64_to_cpu(journal_buckets_v2->d[i].nr); j++)
 				ja->buckets[dst++] =
 					le64_to_cpu(journal_buckets_v2->d[i].start) + j;
 	} else if (journal_buckets) {
-		for (i = 0; i < ja->nr; i++)
+		for (unsigned i = 0; i < ja->nr; i++)
 			ja->buckets[i] = le64_to_cpu(journal_buckets->buckets[i]);
 	}
 
@@ -1303,19 +1336,19 @@ int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb)
 
 void bch2_fs_journal_exit(struct journal *j)
 {
-	unsigned i;
+	if (j->wq)
+		destroy_workqueue(j->wq);
 
 	darray_exit(&j->early_journal_entries);
 
-	for (i = 0; i < ARRAY_SIZE(j->buf); i++)
-		kvpfree(j->buf[i].data, j->buf[i].buf_size);
+	for (unsigned i = 0; i < ARRAY_SIZE(j->buf); i++)
+		kvfree(j->buf[i].data);
 	free_fifo(&j->pin);
 }
 
 int bch2_fs_journal_init(struct journal *j)
 {
 	static struct lock_class_key res_key;
-	unsigned i;
 
 	mutex_init(&j->buf_lock);
 	spin_lock_init(&j->lock);
@@ -1336,14 +1369,20 @@ int bch2_fs_journal_init(struct journal *j)
 	if (!(init_fifo(&j->pin, JOURNAL_PIN, GFP_KERNEL)))
 		return -BCH_ERR_ENOMEM_journal_pin_fifo;
 
-	for (i = 0; i < ARRAY_SIZE(j->buf); i++) {
+	for (unsigned i = 0; i < ARRAY_SIZE(j->buf); i++) {
 		j->buf[i].buf_size = JOURNAL_ENTRY_SIZE_MIN;
-		j->buf[i].data = kvpmalloc(j->buf[i].buf_size, GFP_KERNEL);
+		j->buf[i].data = kvmalloc(j->buf[i].buf_size, GFP_KERNEL);
 		if (!j->buf[i].data)
 			return -BCH_ERR_ENOMEM_journal_buf;
+		j->buf[i].idx = i;
 	}
 
 	j->pin.front = j->pin.back = 1;
+
+	j->wq = alloc_workqueue("bcachefs_journal",
+				WQ_HIGHPRI|WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512);
+	if (!j->wq)
+		return -BCH_ERR_ENOMEM_fs_other_alloc;
 	return 0;
 }
 
@@ -1455,7 +1494,6 @@ bool bch2_journal_seq_pins_to_text(struct printbuf *out, struct journal *j, u64
 {
 	struct journal_entry_pin_list *pin_list;
 	struct journal_entry_pin *pin;
-	unsigned i;
 
 	spin_lock(&j->lock);
 	*seq = max(*seq, j->pin.front);
@@ -1473,7 +1511,7 @@ bool bch2_journal_seq_pins_to_text(struct printbuf *out, struct journal *j, u64
 	prt_newline(out);
 	printbuf_indent_add(out, 2);
 
-	for (i = 0; i < ARRAY_SIZE(pin_list->list); i++)
+	for (unsigned i = 0; i < ARRAY_SIZE(pin_list->list); i++)
 		list_for_each_entry(pin, &pin_list->list[i], list) {
 			prt_printf(out, "\t%px %ps", pin, pin->flush);
 			prt_newline(out);