X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Fjournal.h;h=011711e99c8d825ec968cf513f82c08a66ecabc5;hb=d320a4e927fd706b34c714b77130965a385ea4fb;hp=9ad82c6081c18524a70fbdc3530e92cd09f46ac4;hpb=e783d814e83b2309930e1f6459212da6da8c8a54;p=bcachefs-tools-debian

diff --git a/libbcachefs/journal.h b/libbcachefs/journal.h
index 9ad82c6..011711e 100644
--- a/libbcachefs/journal.h
+++ b/libbcachefs/journal.h
@@ -1,5 +1,6 @@
-#ifndef _BCACHE_JOURNAL_H
-#define _BCACHE_JOURNAL_H
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BCACHEFS_JOURNAL_H
+#define _BCACHEFS_JOURNAL_H
 
 /*
  * THE JOURNAL:
@@ -28,8 +29,8 @@
  *
  * Synchronous updates are specified by passing a closure (@flush_cl) to
  * bch2_btree_insert() or bch_btree_insert_node(), which then pass that parameter
- * down to the journalling code. That closure will will wait on the journal
- * write to complete (via closure_wait()).
+ * down to the journalling code. That closure will wait on the journal write to
+ * complete (via closure_wait()).
  *
  * If the index update wasn't synchronous, the journal entry will be
  * written out after 10 ms have elapsed, by default (the delay_ms field
@@ -112,62 +113,56 @@
 
 #include "journal_types.h"
 
-/*
- * Only used for holding the journal entries we read in btree_journal_read()
- * during cache_registration
- */
-struct journal_replay {
-	struct list_head	list;
-	struct jset		j;
-};
-
-#define JOURNAL_PIN	(32 * 1024)
+struct bch_fs;
 
-static inline bool journal_pin_active(struct journal_entry_pin *pin)
+static inline void journal_wake(struct journal *j)
 {
-	return pin->pin_list != NULL;
+	wake_up(&j->wait);
+	closure_wake_up(&j->async_wait);
+	closure_wake_up(&j->preres_wait);
 }
 
-static inline struct journal_entry_pin_list *
-journal_seq_pin(struct journal *j, u64 seq)
+static inline struct journal_buf *journal_cur_buf(struct journal *j)
 {
-	return &j->pin.data[(size_t) seq & j->pin.mask];
+	return j->buf + j->reservations.idx;
 }
 
-void bch2_journal_pin_add(struct journal *, struct journal_res *,
-			  struct journal_entry_pin *, journal_pin_flush_fn);
-void bch2_journal_pin_drop(struct journal *, struct journal_entry_pin *);
-void bch2_journal_pin_add_if_older(struct journal *,
-				  struct journal_entry_pin *,
-				  struct journal_entry_pin *,
-				  journal_pin_flush_fn);
-void bch2_journal_flush_pins(struct journal *);
+/* Sequence number of oldest dirty journal entry */
 
-struct closure;
-struct bch_fs;
-struct keylist;
+static inline u64 journal_last_seq(struct journal *j)
+{
+	return j->pin.front;
+}
 
-struct bkey_i *bch2_journal_find_btree_root(struct bch_fs *, struct jset *,
-					   enum btree_id, unsigned *);
+static inline u64 journal_cur_seq(struct journal *j)
+{
+	EBUG_ON(j->pin.back - 1 != atomic64_read(&j->seq));
 
-int bch2_journal_seq_should_ignore(struct bch_fs *, u64, struct btree *);
+	return j->pin.back - 1;
+}
 
-u64 bch2_inode_journal_seq(struct journal *, u64);
+static inline u64 journal_last_unwritten_seq(struct journal *j)
+{
+	return j->seq_ondisk + 1;
+}
 
 static inline int journal_state_count(union journal_res_state s, int idx)
 {
-	return idx == 0 ? s.buf0_count : s.buf1_count;
+	switch (idx) {
+	case 0: return s.buf0_count;
+	case 1: return s.buf1_count;
+	case 2: return s.buf2_count;
+	case 3: return s.buf3_count;
+	}
+	BUG();
 }
 
 static inline void journal_state_inc(union journal_res_state *s)
 {
 	s->buf0_count += s->idx == 0;
 	s->buf1_count += s->idx == 1;
-}
-
-static inline void bch2_journal_set_has_inode(struct journal_buf *buf, u64 inum)
-{
-	set_bit(hash_64(inum, ilog2(sizeof(buf->has_inode) * 8)), buf->has_inode);
+	s->buf2_count += s->idx == 2;
+	s->buf3_count += s->idx == 3;
 }
 
 /*
@@ -179,63 +174,121 @@ static inline unsigned jset_u64s(unsigned u64s)
 	return u64s + sizeof(struct jset_entry) / sizeof(u64);
 }
 
-static inline void bch2_journal_add_entry_at(struct journal_buf *buf,
-					    const void *data, size_t u64s,
-					    unsigned type, enum btree_id id,
-					    unsigned level, unsigned offset)
+static inline int journal_entry_overhead(struct journal *j)
 {
-	struct jset_entry *entry = vstruct_idx(buf->data, offset);
+	return sizeof(struct jset) / sizeof(u64) + j->entry_u64s_reserved;
+}
 
+static inline struct jset_entry *
+bch2_journal_add_entry_noreservation(struct journal_buf *buf, size_t u64s)
+{
+	struct jset *jset = buf->data;
+	struct jset_entry *entry = vstruct_idx(jset, le32_to_cpu(jset->u64s));
+
+	memset(entry, 0, sizeof(*entry));
 	entry->u64s = cpu_to_le16(u64s);
-	entry->btree_id = id;
-	entry->level = level;
-	entry->flags = 0;
-	SET_JOURNAL_ENTRY_TYPE(entry, type);
 
-	memcpy_u64s(entry->_data, data, u64s);
+	le32_add_cpu(&jset->u64s, jset_u64s(u64s));
+
+	return entry;
 }
 
-static inline void bch2_journal_add_keys(struct journal *j, struct journal_res *res,
-					enum btree_id id, const struct bkey_i *k)
+static inline struct jset_entry *
+journal_res_entry(struct journal *j, struct journal_res *res)
 {
-	struct journal_buf *buf = &j->buf[res->idx];
-	unsigned actual = jset_u64s(k->k.u64s);
+	return vstruct_idx(j->buf[res->idx].data, res->offset);
+}
 
-	EBUG_ON(!res->ref);
-	BUG_ON(actual > res->u64s);
+static inline unsigned journal_entry_init(struct jset_entry *entry, unsigned type,
+					  enum btree_id id, unsigned level,
+					  unsigned u64s)
+{
+	entry->u64s	= cpu_to_le16(u64s);
+	entry->btree_id = id;
+	entry->level	= level;
+	entry->type	= type;
+	entry->pad[0]	= 0;
+	entry->pad[1]	= 0;
+	entry->pad[2]	= 0;
+	return jset_u64s(u64s);
+}
+
+static inline unsigned journal_entry_set(struct jset_entry *entry, unsigned type,
+					  enum btree_id id, unsigned level,
+					  const void *data, unsigned u64s)
+{
+	unsigned ret = journal_entry_init(entry, type, id, level, u64s);
 
-	bch2_journal_set_has_inode(buf, k->k.p.inode);
+	memcpy_u64s_small(entry->_data, data, u64s);
+	return ret;
+}
+
+static inline struct jset_entry *
+bch2_journal_add_entry(struct journal *j, struct journal_res *res,
+			 unsigned type, enum btree_id id,
+			 unsigned level, unsigned u64s)
+{
+	struct jset_entry *entry = journal_res_entry(j, res);
+	unsigned actual = journal_entry_init(entry, type, id, level, u64s);
 
-	bch2_journal_add_entry_at(buf, k, k->k.u64s,
-				 JOURNAL_ENTRY_BTREE_KEYS, id,
-				 0, res->offset);
+	EBUG_ON(!res->ref);
+	EBUG_ON(actual > res->u64s);
 
 	res->offset	+= actual;
 	res->u64s	-= actual;
+	return entry;
 }
 
-void bch2_journal_buf_put_slowpath(struct journal *, bool);
+static inline bool journal_entry_empty(struct jset *j)
+{
+	struct jset_entry *i;
+
+	if (j->seq != j->last_seq)
+		return false;
+
+	vstruct_for_each(j, i)
+		if (i->type == BCH_JSET_ENTRY_btree_keys && i->u64s)
+			return false;
+	return true;
+}
 
-static inline void bch2_journal_buf_put(struct journal *j, unsigned idx,
-				       bool need_write_just_set)
+/*
+ * Drop reference on a buffer index and return true if the count has hit zero.
+ */
+static inline union journal_res_state journal_state_buf_put(struct journal *j, unsigned idx)
 {
 	union journal_res_state s;
 
 	s.v = atomic64_sub_return(((union journal_res_state) {
 				    .buf0_count = idx == 0,
 				    .buf1_count = idx == 1,
+				    .buf2_count = idx == 2,
+				    .buf3_count = idx == 3,
 				    }).v, &j->reservations.counter);
+	return s;
+}
+
+void bch2_journal_buf_put_final(struct journal *, u64, bool);
+
+static inline void __bch2_journal_buf_put(struct journal *j, unsigned idx, u64 seq)
+{
+	union journal_res_state s;
 
-	EBUG_ON(s.idx != idx && !s.prev_buf_unwritten);
+	s = journal_state_buf_put(j, idx);
+	if (!journal_state_count(s, idx))
+		bch2_journal_buf_put_final(j, seq, idx == s.unwritten_idx);
+}
 
-	/*
-	 * Do not initiate a journal write if the journal is in an error state
-	 * (previous journal entry write may have failed)
-	 */
-	if (s.idx != idx &&
-	    !journal_state_count(s, idx) &&
-	    s.cur_entry_offset != JOURNAL_ENTRY_ERROR_VAL)
-		bch2_journal_buf_put_slowpath(j, need_write_just_set);
+static inline void bch2_journal_buf_put(struct journal *j, unsigned idx, u64 seq)
+{
+	union journal_res_state s;
+
+	s = journal_state_buf_put(j, idx);
+	if (!journal_state_count(s, idx)) {
+		spin_lock(&j->lock);
+		bch2_journal_buf_put_final(j, seq, idx == s.unwritten_idx);
+		spin_unlock(&j->lock);
+	}
 }
 
 /*
@@ -248,28 +301,33 @@ static inline void bch2_journal_res_put(struct journal *j,
 	if (!res->ref)
 		return;
 
-	lock_release(&j->res_map, 0, _RET_IP_);
+	lock_release(&j->res_map, _THIS_IP_);
 
-	while (res->u64s) {
-		bch2_journal_add_entry_at(&j->buf[res->idx], NULL, 0,
-					 JOURNAL_ENTRY_BTREE_KEYS,
-					 0, 0, res->offset);
-		res->offset	+= jset_u64s(0);
-		res->u64s	-= jset_u64s(0);
-	}
+	while (res->u64s)
+		bch2_journal_add_entry(j, res,
+				       BCH_JSET_ENTRY_btree_keys,
+				       0, 0, 0);
 
-	bch2_journal_buf_put(j, res->idx, false);
+	bch2_journal_buf_put(j, res->idx, res->seq);
 
 	res->ref = 0;
 }
 
 int bch2_journal_res_get_slowpath(struct journal *, struct journal_res *,
-				 unsigned, unsigned);
+				  unsigned);
+
+/* First bits for BCH_WATERMARK: */
+enum journal_res_flags {
+	__JOURNAL_RES_GET_NONBLOCK	= BCH_WATERMARK_BITS,
+	__JOURNAL_RES_GET_CHECK,
+};
+
+#define JOURNAL_RES_GET_NONBLOCK	(1 << __JOURNAL_RES_GET_NONBLOCK)
+#define JOURNAL_RES_GET_CHECK		(1 << __JOURNAL_RES_GET_CHECK)
 
 static inline int journal_res_get_fast(struct journal *j,
 				       struct journal_res *res,
-				       unsigned u64s_min,
-				       unsigned u64s_max)
+				       unsigned flags)
 {
 	union journal_res_state old, new;
 	u64 v = atomic64_read(&j->reservations.counter);
@@ -281,51 +339,172 @@ static inline int journal_res_get_fast(struct journal *j,
 		 * Check if there is still room in the current journal
 		 * entry:
 		 */
-		if (old.cur_entry_offset + u64s_min > j->cur_entry_u64s)
+		if (new.cur_entry_offset + res->u64s > j->cur_entry_u64s)
 			return 0;
 
-		res->offset	= old.cur_entry_offset;
-		res->u64s	= min(u64s_max, j->cur_entry_u64s -
-				      old.cur_entry_offset);
+		EBUG_ON(!journal_state_count(new, new.idx));
+
+		if ((flags & BCH_WATERMARK_MASK) < j->watermark)
+			return 0;
 
-		journal_state_inc(&new);
 		new.cur_entry_offset += res->u64s;
+		journal_state_inc(&new);
+
+		/*
+		 * If the refcount would overflow, we have to wait:
+		 * XXX - tracepoint this:
+		 */
+		if (!journal_state_count(new, new.idx))
+			return 0;
+
+		if (flags & JOURNAL_RES_GET_CHECK)
+			return 1;
 	} while ((v = atomic64_cmpxchg(&j->reservations.counter,
 				       old.v, new.v)) != old.v);
 
-	res->ref = true;
-	res->idx = new.idx;
-	res->seq = le64_to_cpu(j->buf[res->idx].data->seq);
+	res->ref	= true;
+	res->idx	= old.idx;
+	res->offset	= old.cur_entry_offset;
+	res->seq	= le64_to_cpu(j->buf[old.idx].data->seq);
 	return 1;
 }
 
 static inline int bch2_journal_res_get(struct journal *j, struct journal_res *res,
-				      unsigned u64s_min, unsigned u64s_max)
+				       unsigned u64s, unsigned flags)
 {
 	int ret;
 
 	EBUG_ON(res->ref);
-	EBUG_ON(u64s_max < u64s_min);
+	EBUG_ON(!test_bit(JOURNAL_STARTED, &j->flags));
+
+	res->u64s = u64s;
 
-	if (journal_res_get_fast(j, res, u64s_min, u64s_max))
+	if (journal_res_get_fast(j, res, flags))
 		goto out;
 
-	ret = bch2_journal_res_get_slowpath(j, res, u64s_min, u64s_max);
+	ret = bch2_journal_res_get_slowpath(j, res, flags);
 	if (ret)
 		return ret;
 out:
-	lock_acquire_shared(&j->res_map, 0, 0, NULL, _THIS_IP_);
-	EBUG_ON(!res->ref);
+	if (!(flags & JOURNAL_RES_GET_CHECK)) {
+		lock_acquire_shared(&j->res_map, 0,
+				    (flags & JOURNAL_RES_GET_NONBLOCK) != 0,
+				    NULL, _THIS_IP_);
+		EBUG_ON(!res->ref);
+	}
 	return 0;
 }
 
-void bch2_journal_wait_on_seq(struct journal *, u64, struct closure *);
-void bch2_journal_flush_seq_async(struct journal *, u64, struct closure *);
+/* journal_preres: */
+
+static inline void journal_set_watermark(struct journal *j)
+{
+	union journal_preres_state s = READ_ONCE(j->prereserved);
+	unsigned watermark = BCH_WATERMARK_stripe;
+
+	if (fifo_free(&j->pin) < j->pin.size / 4)
+		watermark = max_t(unsigned, watermark, BCH_WATERMARK_copygc);
+	if (fifo_free(&j->pin) < j->pin.size / 8)
+		watermark = max_t(unsigned, watermark, BCH_WATERMARK_reclaim);
+
+	if (s.reserved > s.remaining)
+		watermark = max_t(unsigned, watermark, BCH_WATERMARK_copygc);
+	if (!s.remaining)
+		watermark = max_t(unsigned, watermark, BCH_WATERMARK_reclaim);
+
+	if (watermark == j->watermark)
+		return;
+
+	swap(watermark, j->watermark);
+	if (watermark > j->watermark)
+		journal_wake(j);
+}
+
+static inline void bch2_journal_preres_put(struct journal *j,
+					   struct journal_preres *res)
+{
+	union journal_preres_state s = { .reserved = res->u64s };
+
+	if (!res->u64s)
+		return;
+
+	s.v = atomic64_sub_return(s.v, &j->prereserved.counter);
+	res->u64s = 0;
+
+	if (unlikely(s.waiting)) {
+		clear_bit(ilog2((((union journal_preres_state) { .waiting = 1 }).v)),
+			  (unsigned long *) &j->prereserved.v);
+		closure_wake_up(&j->preres_wait);
+	}
+
+	if (s.reserved <= s.remaining && j->watermark)
+		journal_set_watermark(j);
+}
+
+int __bch2_journal_preres_get(struct journal *,
+			struct journal_preres *, unsigned, unsigned);
+
+static inline int bch2_journal_preres_get_fast(struct journal *j,
+					       struct journal_preres *res,
+					       unsigned new_u64s,
+					       unsigned flags,
+					       bool set_waiting)
+{
+	int d = new_u64s - res->u64s;
+	union journal_preres_state old, new;
+	u64 v = atomic64_read(&j->prereserved.counter);
+	enum bch_watermark watermark = flags & BCH_WATERMARK_MASK;
+	int ret;
+
+	do {
+		old.v = new.v = v;
+		ret = 0;
+
+		if (watermark == BCH_WATERMARK_reclaim ||
+		    new.reserved + d < new.remaining) {
+			new.reserved += d;
+			ret = 1;
+		} else if (set_waiting && !new.waiting)
+			new.waiting = true;
+		else
+			return 0;
+	} while ((v = atomic64_cmpxchg(&j->prereserved.counter,
+				       old.v, new.v)) != old.v);
+
+	if (ret)
+		res->u64s += d;
+	return ret;
+}
+
+static inline int bch2_journal_preres_get(struct journal *j,
+					  struct journal_preres *res,
+					  unsigned new_u64s,
+					  unsigned flags)
+{
+	if (new_u64s <= res->u64s)
+		return 0;
+
+	if (bch2_journal_preres_get_fast(j, res, new_u64s, flags, false))
+		return 0;
+
+	if (flags & JOURNAL_RES_GET_NONBLOCK)
+		return -BCH_ERR_journal_preres_get_blocked;
+
+	return __bch2_journal_preres_get(j, res, new_u64s, flags);
+}
+
+/* journal_entry_res: */
+
+void bch2_journal_entry_res_resize(struct journal *,
+				   struct journal_entry_res *,
+				   unsigned);
+
+int bch2_journal_flush_seq_async(struct journal *, u64, struct closure *);
 void bch2_journal_flush_async(struct journal *, struct closure *);
-void bch2_journal_meta_async(struct journal *, struct closure *);
 
 int bch2_journal_flush_seq(struct journal *, u64);
 int bch2_journal_flush(struct journal *);
+bool bch2_journal_noflush_seq(struct journal *, u64);
 int bch2_journal_meta(struct journal *);
 
 void bch2_journal_halt(struct journal *);
@@ -336,16 +515,7 @@ static inline int bch2_journal_error(struct journal *j)
 		? -EIO : 0;
 }
 
-static inline bool journal_flushes_device(struct bch_dev *ca)
-{
-	return true;
-}
-
-void bch2_journal_start(struct bch_fs *);
-int bch2_journal_mark(struct bch_fs *, struct list_head *);
-void bch2_journal_entries_free(struct list_head *);
-int bch2_journal_read(struct bch_fs *, struct list_head *);
-int bch2_journal_replay(struct bch_fs *, struct list_head *);
+struct bch_dev;
 
 static inline void bch2_journal_set_replay_done(struct journal *j)
 {
@@ -353,23 +523,27 @@ static inline void bch2_journal_set_replay_done(struct journal *j)
 	set_bit(JOURNAL_REPLAY_DONE, &j->flags);
 }
 
-ssize_t bch2_journal_print_debug(struct journal *, char *);
+void bch2_journal_unblock(struct journal *);
+void bch2_journal_block(struct journal *);
 
-int bch2_dev_journal_alloc(struct bch_dev *);
+void __bch2_journal_debug_to_text(struct printbuf *, struct journal *);
+void bch2_journal_debug_to_text(struct printbuf *, struct journal *);
+void bch2_journal_pins_to_text(struct printbuf *, struct journal *);
+bool bch2_journal_seq_pins_to_text(struct printbuf *, struct journal *, u64 *);
 
-static inline unsigned bch2_nr_journal_buckets(struct bch_sb_field_journal *j)
-{
-	return j
-		? (__le64 *) vstruct_end(&j->field) - j->buckets
-		: 0;
-}
+int bch2_set_nr_journal_buckets(struct bch_fs *, struct bch_dev *,
+				unsigned nr);
+int bch2_dev_journal_alloc(struct bch_dev *);
+int bch2_fs_journal_alloc(struct bch_fs *);
 
-int bch2_journal_move(struct bch_dev *);
+void bch2_dev_journal_stop(struct journal *, struct bch_dev *);
 
 void bch2_fs_journal_stop(struct journal *);
+int bch2_fs_journal_start(struct journal *, u64);
+
 void bch2_dev_journal_exit(struct bch_dev *);
 int bch2_dev_journal_init(struct bch_dev *, struct bch_sb *);
 void bch2_fs_journal_exit(struct journal *);
 int bch2_fs_journal_init(struct journal *);
 
-#endif /* _BCACHE_JOURNAL_H */
+#endif /* _BCACHEFS_JOURNAL_H */