bcachefs: Switch fsync to use bi_journal_seq

Now that we're recording in each inode the journal sequence number of
the most recent update, fsync becomes a lot simpler and we can delete
all the plumbing for ei_journal_seq.

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>

9 files changed, 65 insertions, 96 deletions

diff --git a/fs/bcachefs/acl.c b/fs/bcachefs/acl.c
index 2afa15b26700..51a0b48a5313 100644
--- a/fs/bcachefs/acl.c
+++ b/fs/bcachefs/acl.c
@@ -330,8 +330,7 @@ retry:
 	inode_u.bi_mode		= mode;
 
 	ret =   bch2_inode_write(&trans, &inode_iter, &inode_u) ?:
-		bch2_trans_commit(&trans, NULL,
-				  &inode->ei_journal_seq, 0);
+		bch2_trans_commit(&trans, NULL, NULL, 0);
 btree_err:
 	bch2_trans_iter_exit(&trans, &inode_iter);
 
diff --git a/fs/bcachefs/fs-io.c b/fs/bcachefs/fs-io.c
index f4c97fc0e3d1..7de6b7a7aa60 100644
--- a/fs/bcachefs/fs-io.c
+++ b/fs/bcachefs/fs-io.c
@@ -1096,7 +1096,6 @@ static void bch2_writepage_io_alloc(struct bch_fs *c,
 	op			= &w->io->op;
 	bch2_write_op_init(op, c, w->opts);
 	op->target		= w->opts.foreground_target;
-	op_journal_seq_set(op, &inode->ei_journal_seq);
 	op->nr_replicas		= nr_replicas;
 	op->res.nr_replicas	= nr_replicas;
 	op->write_point		= writepoint_hashed(inode->ei_last_dirtied);
@@ -1947,7 +1946,6 @@ static long bch2_dio_write_loop(struct dio_write *dio)
 		bch2_write_op_init(&dio->op, c, io_opts(c, &inode->ei_inode));
 		dio->op.end_io		= bch2_dio_write_loop_async;
 		dio->op.target		= dio->op.opts.foreground_target;
-		op_journal_seq_set(&dio->op, &inode->ei_journal_seq);
 		dio->op.write_point	= writepoint_hashed((unsigned long) current);
 		dio->op.nr_replicas	= dio->op.opts.data_replicas;
 		dio->op.subvol		= inode->ei_subvol;
@@ -2164,29 +2162,36 @@ unlock:
 
 /* fsync: */
 
-int bch2_fsync(struct file *file, loff_t start, loff_t end, int datasync)
+/*
+ * inode->ei_inode.bi_journal_seq won't be up to date since it's set in an
+ * insert trigger: look up the btree inode instead
+ */
+static int bch2_flush_inode(struct bch_fs *c, subvol_inum inum)
 {
-	struct bch_inode_info *inode = file_bch_inode(file);
-	struct bch_fs *c = inode->v.i_sb->s_fs_info;
-	int ret, ret2;
+	struct bch_inode_unpacked inode;
+	int ret;
 
-	ret = file_write_and_wait_range(file, start, end);
+	if (c->opts.journal_flush_disabled)
+		return 0;
+
+	ret = bch2_inode_find_by_inum(c, inum, &inode);
 	if (ret)
 		return ret;
 
-	if (datasync && !(inode->v.i_state & I_DIRTY_DATASYNC))
-		goto out;
+	return bch2_journal_flush_seq(&c->journal, inode.bi_journal_seq);
+}
 
-	ret = sync_inode_metadata(&inode->v, 1);
-	if (ret)
-		return ret;
-out:
-	if (!c->opts.journal_flush_disabled)
-		ret = bch2_journal_flush_seq(&c->journal,
-					     inode->ei_journal_seq);
-	ret2 = file_check_and_advance_wb_err(file);
+int bch2_fsync(struct file *file, loff_t start, loff_t end, int datasync)
+{
+	struct bch_inode_info *inode = file_bch_inode(file);
+	struct bch_fs *c = inode->v.i_sb->s_fs_info;
+	int ret, ret2, ret3;
+
+	ret = file_write_and_wait_range(file, start, end);
+	ret2 = sync_inode_metadata(&inode->v, 1);
+	ret3 = bch2_flush_inode(c, inode_inum(inode));
 
-	return ret ?: ret2;
+	return ret ?: ret2 ?: ret3;
 }
 
 /* truncate: */
@@ -2448,7 +2453,7 @@ int bch2_truncate(struct mnt_idmap *idmap,
 
 	ret = bch2_fpunch(c, inode_inum(inode),
 			round_up(iattr->ia_size, block_bytes(c)) >> 9,
-			U64_MAX, &inode->ei_journal_seq, &i_sectors_delta);
+			U64_MAX, &i_sectors_delta);
 	i_sectors_acct(c, inode, NULL, i_sectors_delta);
 
 	if (unlikely(ret))
@@ -2508,7 +2513,6 @@ static long bchfs_fpunch(struct bch_inode_info *inode, loff_t offset, loff_t len
 
 		ret = bch2_fpunch(c, inode_inum(inode),
 				  discard_start, discard_end,
-				  &inode->ei_journal_seq,
 				  &i_sectors_delta);
 		i_sectors_acct(c, inode, NULL, i_sectors_delta);
 	}
@@ -2587,7 +2591,6 @@ static long bchfs_fcollapse_finsert(struct bch_inode_info *inode,
 
 		ret = bch2_fpunch(c, inode_inum(inode),
 				  offset >> 9, (offset + len) >> 9,
-				  &inode->ei_journal_seq,
 				  &i_sectors_delta);
 		i_sectors_acct(c, inode, NULL, i_sectors_delta);
 
@@ -2691,8 +2694,7 @@ reassemble:
 		ret =   bch2_btree_iter_traverse(&del) ?:
 			bch2_trans_update(&trans, &del, &delete, trigger_flags) ?:
 			bch2_trans_update(&trans, &dst, copy.k, trigger_flags) ?:
-			bch2_trans_commit(&trans, &disk_res,
-					  &inode->ei_journal_seq,
+			bch2_trans_commit(&trans, &disk_res, NULL,
 					  BTREE_INSERT_NOFAIL);
 		bch2_disk_reservation_put(c, &disk_res);
 
@@ -2803,7 +2805,7 @@ static int __bchfs_fallocate(struct bch_inode_info *inode, int mode,
 
 		ret = bch2_extent_update(&trans, inode_inum(inode), &iter,
 					 &reservation.k_i,
-				&disk_res, &inode->ei_journal_seq,
+				&disk_res, NULL,
 				0, &i_sectors_delta, true);
 		i_sectors_acct(c, inode, &quota_res, i_sectors_delta);
 bkey_err:
@@ -3003,7 +3005,6 @@ loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src,
 			       inode_inum(dst), pos_dst >> 9,
 			       inode_inum(src), pos_src >> 9,
 			       aligned_len >> 9,
-			       &dst->ei_journal_seq,
 			       pos_dst + len, &i_sectors_delta);
 	if (ret < 0)
 		goto err;
@@ -3021,10 +3022,9 @@ loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src,
 		i_size_write(&dst->v, pos_dst + ret);
 	spin_unlock(&dst->v.i_lock);
 
-	if (((file_dst->f_flags & (__O_SYNC | O_DSYNC)) ||
-	     IS_SYNC(file_inode(file_dst))) &&
-	    !c->opts.journal_flush_disabled)
-		ret = bch2_journal_flush_seq(&c->journal, dst->ei_journal_seq);
+	if ((file_dst->f_flags & (__O_SYNC | O_DSYNC)) ||
+	    IS_SYNC(file_inode(file_dst)))
+		ret = bch2_flush_inode(c, inode_inum(dst));
 err:
 	bch2_unlock_inodes(INODE_LOCK|INODE_PAGECACHE_BLOCK, src, dst);
 
diff --git a/fs/bcachefs/fs.c b/fs/bcachefs/fs.c
index 12178bd15c34..92919b16f2f5 100644
--- a/fs/bcachefs/fs.c
+++ b/fs/bcachefs/fs.c
@@ -41,25 +41,6 @@ static void bch2_vfs_inode_init(struct bch_fs *, subvol_inum,
 				struct bch_inode_info *,
 				struct bch_inode_unpacked *);
 
-static void journal_seq_copy(struct bch_fs *c,
-			     struct bch_inode_info *dst,
-			     u64 journal_seq)
-{
-	/*
-	 * atomic64_cmpxchg has a fallback for archs that don't support it,
-	 * cmpxchg does not:
-	 */
-	atomic64_t *dst_seq = (void *) &dst->ei_journal_seq;
-	u64 old, v = READ_ONCE(dst->ei_journal_seq);
-
-	do {
-		old = v;
-
-		if (old >= journal_seq)
-			break;
-	} while ((v = atomic64_cmpxchg(dst_seq, old, journal_seq)) != old);
-}
-
 static void __pagecache_lock_put(struct pagecache_lock *lock, long i)
 {
 	BUG_ON(atomic_long_read(&lock->v) == 0);
@@ -152,9 +133,7 @@ retry:
 				BTREE_ITER_INTENT) ?:
 		(set ? set(inode, &inode_u, p) : 0) ?:
 		bch2_inode_write(&trans, &iter, &inode_u) ?:
-		bch2_trans_commit(&trans, NULL,
-				  &inode->ei_journal_seq,
-				  BTREE_INSERT_NOFAIL);
+		bch2_trans_commit(&trans, NULL, NULL, BTREE_INSERT_NOFAIL);
 
 	/*
 	 * the btree node lock protects inode->ei_inode, not ei_update_lock;
@@ -329,7 +308,6 @@ err_before_quota:
 	if (!(flags & BCH_CREATE_TMPFILE)) {
 		bch2_inode_update_after_write(c, dir, &dir_u,
 					      ATTR_MTIME|ATTR_CTIME);
-		journal_seq_copy(c, dir, journal_seq);
 		mutex_unlock(&dir->ei_update_lock);
 	}
 
@@ -337,7 +315,6 @@ err_before_quota:
 	inum.inum = inode_u.bi_inum;
 
 	bch2_vfs_inode_init(c, inum, inode, &inode_u);
-	journal_seq_copy(c, inode, journal_seq);
 
 	set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 	set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
@@ -362,7 +339,6 @@ err_before_quota:
 		 * We raced, another process pulled the new inode into cache
 		 * before us:
 		 */
-		journal_seq_copy(c, old, journal_seq);
 		make_bad_inode(&inode->v);
 		iput(&inode->v);
 
@@ -446,7 +422,7 @@ static int __bch2_link(struct bch_fs *c,
 	mutex_lock(&inode->ei_update_lock);
 	bch2_trans_init(&trans, c, 4, 1024);
 
-	ret = __bch2_trans_do(&trans, NULL, &inode->ei_journal_seq, 0,
+	ret = __bch2_trans_do(&trans, NULL, NULL, 0,
 			bch2_link_trans(&trans,
 					inode_inum(dir),   &dir_u,
 					inode_inum(inode), &inode_u,
@@ -455,7 +431,6 @@ static int __bch2_link(struct bch_fs *c,
 	if (likely(!ret)) {
 		BUG_ON(inode_u.bi_inum != inode->v.i_ino);
 
-		journal_seq_copy(c, inode, dir->ei_journal_seq);
 		bch2_inode_update_after_write(c, dir, &dir_u,
 					      ATTR_MTIME|ATTR_CTIME);
 		bch2_inode_update_after_write(c, inode, &inode_u, ATTR_CTIME);
@@ -498,7 +473,7 @@ int __bch2_unlink(struct inode *vdir, struct dentry *dentry,
 	bch2_lock_inodes(INODE_UPDATE_LOCK, dir, inode);
 	bch2_trans_init(&trans, c, 4, 1024);
 
-	ret = __bch2_trans_do(&trans, NULL, &dir->ei_journal_seq,
+	ret = __bch2_trans_do(&trans, NULL, NULL,
 			      BTREE_INSERT_NOFAIL,
 			bch2_unlink_trans(&trans,
 					  inode_inum(dir), &dir_u,
@@ -508,7 +483,6 @@ int __bch2_unlink(struct inode *vdir, struct dentry *dentry,
 	if (likely(!ret)) {
 		BUG_ON(inode_u.bi_inum != inode->v.i_ino);
 
-		journal_seq_copy(c, inode, dir->ei_journal_seq);
 		bch2_inode_update_after_write(c, dir, &dir_u,
 					      ATTR_MTIME|ATTR_CTIME);
 		bch2_inode_update_after_write(c, inode, &inode_u,
@@ -550,8 +524,6 @@ static int bch2_symlink(struct mnt_idmap *idmap,
 	if (unlikely(ret))
 		goto err;
 
-	journal_seq_copy(c, dir, inode->ei_journal_seq);
-
 	ret = __bch2_link(c, inode, dir, dentry);
 	if (unlikely(ret))
 		goto err;
@@ -586,7 +558,6 @@ static int bch2_rename2(struct mnt_idmap *idmap,
 		? BCH_RENAME_EXCHANGE
 		: dst_dentry->d_inode
 		? BCH_RENAME_OVERWRITE : BCH_RENAME;
-	u64 journal_seq = 0;
 	int ret;
 
 	if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
@@ -626,7 +597,7 @@ static int bch2_rename2(struct mnt_idmap *idmap,
 			goto err;
 	}
 
-	ret = __bch2_trans_do(&trans, NULL, &journal_seq, 0,
+	ret = __bch2_trans_do(&trans, NULL, NULL, 0,
 			bch2_rename_trans(&trans,
 					  inode_inum(src_dir), &src_dir_u,
 					  inode_inum(dst_dir), &dst_dir_u,
@@ -644,23 +615,17 @@ static int bch2_rename2(struct mnt_idmap *idmap,
 
 	bch2_inode_update_after_write(c, src_dir, &src_dir_u,
 				      ATTR_MTIME|ATTR_CTIME);
-	journal_seq_copy(c, src_dir, journal_seq);
 
-	if (src_dir != dst_dir) {
+	if (src_dir != dst_dir)
 		bch2_inode_update_after_write(c, dst_dir, &dst_dir_u,
 					      ATTR_MTIME|ATTR_CTIME);
-		journal_seq_copy(c, dst_dir, journal_seq);
-	}
 
 	bch2_inode_update_after_write(c, src_inode, &src_inode_u,
 				      ATTR_CTIME);
-	journal_seq_copy(c, src_inode, journal_seq);
 
-	if (dst_inode) {
+	if (dst_inode)
 		bch2_inode_update_after_write(c, dst_inode, &dst_inode_u,
 					      ATTR_CTIME);
-		journal_seq_copy(c, dst_inode, journal_seq);
-	}
 err:
 	bch2_trans_exit(&trans);
 
@@ -767,8 +732,7 @@ retry:
 	}
 
 	ret =   bch2_inode_write(&trans, &inode_iter, &inode_u) ?:
-		bch2_trans_commit(&trans, NULL,
-				  &inode->ei_journal_seq,
+		bch2_trans_commit(&trans, NULL, NULL,
 				  BTREE_INSERT_NOFAIL);
 btree_err:
 	bch2_trans_iter_exit(&trans, &inode_iter);
@@ -1203,7 +1167,6 @@ static void bch2_vfs_inode_init(struct bch_fs *c, subvol_inum inum,
 	inode->v.i_size		= bi->bi_size;
 
 	inode->ei_flags		= 0;
-	inode->ei_journal_seq	= bi->bi_journal_seq;
 	inode->ei_quota_reserved = 0;
 	inode->ei_qid		= bch_qid(bi);
 	inode->ei_subvol	= inum.subvol;
@@ -1242,7 +1205,6 @@ static struct inode *bch2_alloc_inode(struct super_block *sb)
 	mutex_init(&inode->ei_update_lock);
 	pagecache_lock_init(&inode->ei_pagecache_lock);
 	mutex_init(&inode->ei_quota_lock);
-	inode->ei_journal_seq = 0;
 
 	return &inode->v;
 }
diff --git a/fs/bcachefs/fs.h b/fs/bcachefs/fs.h
index 38c04282da64..1c8936df9fbb 100644
--- a/fs/bcachefs/fs.h
+++ b/fs/bcachefs/fs.h
@@ -36,7 +36,6 @@ struct bch_inode_info {
 	unsigned long		ei_flags;
 
 	struct mutex		ei_update_lock;
-	u64			ei_journal_seq;
 	u64			ei_quota_reserved;
 	unsigned long		ei_last_dirtied;
 	struct pagecache_lock	ei_pagecache_lock;
diff --git a/fs/bcachefs/io.c b/fs/bcachefs/io.c
index 0a9cb4d489f4..dc41286c229e 100644
--- a/fs/bcachefs/io.c
+++ b/fs/bcachefs/io.c
@@ -393,7 +393,7 @@ err:
  */
 int bch2_fpunch_at(struct btree_trans *trans, struct btree_iter *iter,
 		   subvol_inum inum, u64 end,
-		   u64 *journal_seq, s64 *i_sectors_delta)
+		   s64 *i_sectors_delta)
 {
 	struct bch_fs *c	= trans->c;
 	unsigned max_sectors	= KEY_SIZE_MAX & (~0 << c->block_bits);
@@ -431,7 +431,7 @@ int bch2_fpunch_at(struct btree_trans *trans, struct btree_iter *iter,
 		bch2_cut_back(end_pos, &delete);
 
 		ret = bch2_extent_update(trans, inum, iter, &delete,
-				&disk_res, journal_seq,
+				&disk_res, NULL,
 				0, i_sectors_delta, false);
 		bch2_disk_reservation_put(c, &disk_res);
 btree_err:
@@ -450,7 +450,7 @@ btree_err:
 }
 
 int bch2_fpunch(struct bch_fs *c, subvol_inum inum, u64 start, u64 end,
-		u64 *journal_seq, s64 *i_sectors_delta)
+		s64 *i_sectors_delta)
 {
 	struct btree_trans trans;
 	struct btree_iter iter;
@@ -461,8 +461,7 @@ int bch2_fpunch(struct bch_fs *c, subvol_inum inum, u64 start, u64 end,
 			     POS(inum.inum, start),
 			     BTREE_ITER_INTENT);
 
-	ret = bch2_fpunch_at(&trans, &iter, inum, end,
-			     journal_seq, i_sectors_delta);
+	ret = bch2_fpunch_at(&trans, &iter, inum, end, i_sectors_delta);
 
 	bch2_trans_iter_exit(&trans, &iter);
 	bch2_trans_exit(&trans);
diff --git a/fs/bcachefs/io.h b/fs/bcachefs/io.h
index ebb0944b4ca3..8be77561badb 100644
--- a/fs/bcachefs/io.h
+++ b/fs/bcachefs/io.h
@@ -68,12 +68,6 @@ static inline u64 *op_journal_seq(struct bch_write_op *op)
 		? op->journal_seq_p : &op->journal_seq;
 }
 
-static inline void op_journal_seq_set(struct bch_write_op *op, u64 *journal_seq)
-{
-	op->journal_seq_p = journal_seq;
-	op->flags |= BCH_WRITE_JOURNAL_SEQ_PTR;
-}
-
 static inline struct workqueue_struct *index_update_wq(struct bch_write_op *op)
 {
 	return op->alloc_reserve == RESERVE_MOVINGGC
@@ -88,8 +82,8 @@ int bch2_extent_update(struct btree_trans *, subvol_inum,
 		       struct disk_reservation *, u64 *, u64, s64 *, bool);
 
 int bch2_fpunch_at(struct btree_trans *, struct btree_iter *,
-		   subvol_inum, u64, u64 *, s64 *);
-int bch2_fpunch(struct bch_fs *c, subvol_inum, u64, u64, u64 *, s64 *);
+		   subvol_inum, u64, s64 *);
+int bch2_fpunch(struct bch_fs *c, subvol_inum, u64, u64, s64 *);
 
 static inline void bch2_write_op_init(struct bch_write_op *op, struct bch_fs *c,
 				      struct bch_io_opts opts)
diff --git a/fs/bcachefs/reflink.c b/fs/bcachefs/reflink.c
index 8e66e6390e62..d003f4088dfc 100644
--- a/fs/bcachefs/reflink.c
+++ b/fs/bcachefs/reflink.c
@@ -210,7 +210,7 @@ static struct bkey_s_c get_next_src(struct btree_iter *iter, struct bpos end)
 s64 bch2_remap_range(struct bch_fs *c,
 		     subvol_inum dst_inum, u64 dst_offset,
 		     subvol_inum src_inum, u64 src_offset,
-		     u64 remap_sectors, u64 *journal_seq,
+		     u64 remap_sectors,
 		     u64 new_i_size, s64 *i_sectors_delta)
 {
 	struct btree_trans trans;
@@ -281,7 +281,7 @@ s64 bch2_remap_range(struct bch_fs *c,
 					min(dst_end.offset,
 					    dst_iter.pos.offset +
 					    src_iter.pos.offset - src_want.offset),
-					journal_seq, i_sectors_delta);
+					i_sectors_delta);
 			continue;
 		}
 
@@ -320,7 +320,7 @@ s64 bch2_remap_range(struct bch_fs *c,
 				    dst_end.offset - dst_iter.pos.offset));
 
 		ret = bch2_extent_update(&trans, dst_inum, &dst_iter,
-					 new_dst.k, &disk_res, journal_seq,
+					 new_dst.k, &disk_res, NULL,
 					 new_i_size, i_sectors_delta,
 					 true);
 		bch2_disk_reservation_put(c, &disk_res);
@@ -347,7 +347,7 @@ s64 bch2_remap_range(struct bch_fs *c,
 		    inode_u.bi_size < new_i_size) {
 			inode_u.bi_size = new_i_size;
 			ret2  = bch2_inode_write(&trans, &inode_iter, &inode_u) ?:
-				bch2_trans_commit(&trans, NULL, journal_seq, 0);
+				bch2_trans_commit(&trans, NULL, NULL, 0);
 		}
 
 		bch2_trans_iter_exit(&trans, &inode_iter);
diff --git a/fs/bcachefs/reflink.h b/fs/bcachefs/reflink.h
index 4c1b82860b0b..3745873fd88d 100644
--- a/fs/bcachefs/reflink.h
+++ b/fs/bcachefs/reflink.h
@@ -58,6 +58,6 @@ static inline __le64 *bkey_refcount(struct bkey_i *k)
 }
 
 s64 bch2_remap_range(struct bch_fs *, subvol_inum, u64,
-		     subvol_inum, u64, u64, u64 *, u64, s64 *);
+		     subvol_inum, u64, u64, u64, s64 *);
 
 #endif /* _BCACHEFS_REFLINK_H */
diff --git a/fs/bcachefs/xattr.c b/fs/bcachefs/xattr.c
index 181af89b0553..21823ce69237 100644
--- a/fs/bcachefs/xattr.c
+++ b/fs/bcachefs/xattr.c
@@ -165,8 +165,24 @@ int bch2_xattr_set(struct btree_trans *trans, subvol_inum inum,
 		   const char *name, const void *value, size_t size,
 		   int type, int flags)
 {
+	struct btree_iter inode_iter = { NULL };
+	struct bch_inode_unpacked inode_u;
 	int ret;
 
+	/*
+	 * We need to do an inode update so that bi_journal_sync gets updated
+	 * and fsync works:
+	 *
+	 * Perhaps we should be updating bi_mtime too?
+	 */
+
+	ret   = bch2_inode_peek(trans, &inode_iter, &inode_u, inum, BTREE_ITER_INTENT) ?:
+		bch2_inode_write(trans, &inode_iter, &inode_u);
+	bch2_trans_iter_exit(trans, &inode_iter);
+
+	if (ret)
+		return ret;
+
 	if (value) {
 		struct bkey_i_xattr *xattr;
 		unsigned namelen = strlen(name);
@@ -352,7 +368,7 @@ static int bch2_xattr_set_handler(const struct xattr_handler *handler,
 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
 	struct bch_hash_info hash = bch2_hash_info_init(c, &inode->ei_inode);
 
-	return bch2_trans_do(c, NULL, &inode->ei_journal_seq, 0,
+	return bch2_trans_do(c, NULL, NULL, 0,
 			bch2_xattr_set(&trans, inode_inum(inode), &hash,
 				       name, value, size,
 				       handler->flags, flags));