bcachefs: Deduplicate keys in the journal before replay

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/alloc_background.c

@@ -12,7 +12,7 @@
 #include "debug.h"
 #include "ec.h"
 #include "error.h"
-#include "journal_io.h"
+#include "recovery.h"
 #include "trace.h"
 
 #include <linux/kthread.h>
@@ -261,13 +261,13 @@ static void bch2_alloc_read_key(struct bch_fs *c, struct bkey_s_c k)
 	percpu_up_read(&c->mark_lock);
 }
 
-int bch2_alloc_read(struct bch_fs *c, struct list_head *journal_replay_list)
+int bch2_alloc_read(struct bch_fs *c, struct journal_keys *journal_keys)
 {
-	struct journal_replay *r;
 	struct btree_trans trans;
 	struct btree_iter *iter;
 	struct bkey_s_c k;
 	struct bch_dev *ca;
+	struct journal_key *j;
 	unsigned i;
 	int ret;
 
@@ -282,14 +282,9 @@ int bch2_alloc_read(struct bch_fs *c, struct list_head *journal_replay_list)
 	if (ret)
 		return ret;
 
-	list_for_each_entry(r, journal_replay_list, list) {
-		struct bkey_i *k, *n;
-		struct jset_entry *entry;
-
-		for_each_jset_key(k, n, entry, &r->j)
-			if (entry->btree_id == BTREE_ID_ALLOC)
-				bch2_alloc_read_key(c, bkey_i_to_s_c(k));
-	}
+	for_each_journal_key(*journal_keys, j)
+		if (j->btree_id == BTREE_ID_ALLOC)
+			bch2_alloc_read_key(c, bkey_i_to_s_c(j->k));
 
 	percpu_down_write(&c->mark_lock);
 	bch2_dev_usage_from_buckets(c);

fs/bcachefs/alloc_background.h

@@ -25,7 +25,8 @@ void bch2_alloc_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 	.val_to_text	= bch2_alloc_to_text,		\
 }
 
-int bch2_alloc_read(struct bch_fs *, struct list_head *);
+struct journal_keys;
+int bch2_alloc_read(struct bch_fs *, struct journal_keys *);
 int bch2_alloc_replay_key(struct bch_fs *, struct bkey_i *);
 
 static inline void bch2_wake_allocator(struct bch_dev *ca)

fs/bcachefs/btree_gc.c

@@ -19,9 +19,9 @@
 #include "error.h"
 #include "extents.h"
 #include "journal.h"
-#include "journal_io.h"
 #include "keylist.h"
 #include "move.h"
+#include "recovery.h"
 #include "replicas.h"
 #include "super-io.h"
 #include "trace.h"
@@ -273,7 +273,7 @@ static inline int btree_id_gc_phase_cmp(enum btree_id l, enum btree_id r)
 		(int) btree_id_to_gc_phase(r);
 }
 
-static int bch2_gc_btrees(struct bch_fs *c, struct list_head *journal,
+static int bch2_gc_btrees(struct bch_fs *c, struct journal_keys *journal_keys,
 			  bool initial, bool metadata_only)
 {
 	enum btree_id ids[BTREE_ID_NR];
@@ -292,25 +292,21 @@ static int bch2_gc_btrees(struct bch_fs *c, struct list_head *journal,
 		if (ret)
 			return ret;
 
-		if (journal && !metadata_only &&
+		if (journal_keys && !metadata_only &&
 		    btree_node_type_needs_gc(type)) {
-			struct bkey_i *k, *n;
-			struct jset_entry *j;
-			struct journal_replay *r;
+			struct journal_key *j;
 			int ret;
 
-			list_for_each_entry(r, journal, list)
-				for_each_jset_key(k, n, j, &r->j) {
-					if (type == __btree_node_type(j->level, j->btree_id)) {
+			for_each_journal_key(*journal_keys, j)
+				if (j->btree_id == id) {
 					ret = bch2_gc_mark_key(c,
-							bkey_i_to_s_c(k),
+						bkey_i_to_s_c(j->k),
 						&max_stale, initial);
 					if (ret)
 						return ret;
 				}
 		}
 	}
-	}
 
 	return 0;
 }
@@ -695,7 +691,7 @@ static int bch2_gc_start(struct bch_fs *c,
  *    move around - if references move backwards in the ordering GC
  *    uses, GC could skip past them
  */
-int bch2_gc(struct bch_fs *c, struct list_head *journal,
+int bch2_gc(struct bch_fs *c, struct journal_keys *journal_keys,
 	    bool initial, bool metadata_only)
 {
 	struct bch_dev *ca;
@@ -716,7 +712,7 @@ int bch2_gc(struct bch_fs *c, struct list_head *journal,
 
 	bch2_mark_superblocks(c);
 
-	ret = bch2_gc_btrees(c, journal, initial, metadata_only);
+	ret = bch2_gc_btrees(c, journal_keys, initial, metadata_only);
 	if (ret)
 		goto out;
 

fs/bcachefs/btree_gc.h

@@ -5,7 +5,9 @@
 #include "btree_types.h"
 
 void bch2_coalesce(struct bch_fs *);
-int bch2_gc(struct bch_fs *, struct list_head *, bool, bool);
+
+struct journal_keys;
+int bch2_gc(struct bch_fs *, struct journal_keys *, bool, bool);
 void bch2_gc_thread_stop(struct bch_fs *);
 int bch2_gc_thread_start(struct bch_fs *);
 void bch2_mark_dev_superblock(struct bch_fs *, struct bch_dev *, unsigned);

fs/bcachefs/ec.c

@@ -12,8 +12,8 @@
 #include "ec.h"
 #include "error.h"
 #include "io.h"
-#include "journal_io.h"
 #include "keylist.h"
+#include "recovery.h"
 #include "super-io.h"
 #include "util.h"
 
@@ -1235,9 +1235,9 @@ static void bch2_stripe_read_key(struct bch_fs *c, struct bkey_s_c k)
 	bch2_mark_key(c, k, true, 0, NULL, 0, 0);
 }
 
-int bch2_stripes_read(struct bch_fs *c, struct list_head *journal_replay_list)
+int bch2_stripes_read(struct bch_fs *c, struct journal_keys *journal_keys)
 {
-	struct journal_replay *r;
+	struct journal_key *i;
 	struct btree_trans trans;
 	struct btree_iter *iter;
 	struct bkey_s_c k;
@@ -1258,14 +1258,9 @@ int bch2_stripes_read(struct bch_fs *c, struct list_head *journal_replay_list)
 	if (ret)
 		return ret;
 
-	list_for_each_entry(r, journal_replay_list, list) {
-		struct bkey_i *k, *n;
-		struct jset_entry *entry;
-
-		for_each_jset_key(k, n, entry, &r->j)
-			if (entry->btree_id == BTREE_ID_EC)
-				bch2_stripe_read_key(c, bkey_i_to_s_c(k));
-	}
+	for_each_journal_key(*journal_keys, i)
+		if (i->btree_id == BTREE_ID_EC)
+			bch2_stripe_read_key(c, bkey_i_to_s_c(i->k));
 
 	return 0;
 }

fs/bcachefs/ec.h

@@ -150,7 +150,8 @@ void bch2_ec_stop_dev(struct bch_fs *, struct bch_dev *);
 
 void bch2_ec_flush_new_stripes(struct bch_fs *);
 
-int bch2_stripes_read(struct bch_fs *, struct list_head *);
+struct journal_keys;
+int bch2_stripes_read(struct bch_fs *, struct journal_keys *);
 int bch2_stripes_write(struct bch_fs *, unsigned, bool *);
 
 int bch2_ec_mem_alloc(struct bch_fs *, bool);

fs/bcachefs/recovery.h

@@ -2,6 +2,22 @@
 #ifndef _BCACHEFS_RECOVERY_H
 #define _BCACHEFS_RECOVERY_H
 
+struct journal_keys {
+	struct journal_key {
+		enum btree_id	btree_id:8;
+		unsigned	allocated:1;
+		struct bpos	pos;
+		struct bkey_i	*k;
+		u32		journal_seq;
+		u32		journal_offset;
+	}			*d;
+	size_t			nr;
+	u64			journal_seq_base;
+};
+
+#define for_each_journal_key(keys, i)				\
+	for (i = (keys).d; i < (keys).d + (keys).nr; (i)++)
+
 int bch2_fs_recovery(struct bch_fs *);
 int bch2_fs_initialize(struct bch_fs *);