bcachefs: switch to rhashtable for vfs inodes hash

the standard vfs inode hash table suffers from painful lock contention -
this is long overdue
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/acl.c

@@ -361,7 +361,7 @@ int bch2_set_acl(struct mnt_idmap *idmap,
 	bch2_trans_begin(trans);
 	acl = _acl;
 
-	ret   = bch2_subvol_is_ro_trans(trans, inode->ei_subvol) ?:
+	ret   = bch2_subvol_is_ro_trans(trans, inode->ei_inum.subvol) ?:
 		bch2_inode_peek(trans, &inode_iter, &inode_u, inode_inum(inode),
 			      BTREE_ITER_intent);
 	if (ret)

fs/bcachefs/bcachefs.h

@@ -1023,6 +1023,7 @@ struct bch_fs {
 	/* fs.c */
 	struct list_head	vfs_inodes_list;
 	struct mutex		vfs_inodes_lock;
+	struct rhashtable	vfs_inodes_table;
 
 	/* VFS IO PATH - fs-io.c */
 	struct bio_set		writepage_bioset;

fs/bcachefs/fs-io-buffered.c

@@ -486,7 +486,7 @@ static void bch2_writepage_io_alloc(struct bch_fs *c,
 	op->nr_replicas		= nr_replicas;
 	op->res.nr_replicas	= nr_replicas;
 	op->write_point		= writepoint_hashed(inode->ei_last_dirtied);
-	op->subvol		= inode->ei_subvol;
+	op->subvol		= inode->ei_inum.subvol;
 	op->pos			= POS(inode->v.i_ino, sector);
 	op->end_io		= bch2_writepage_io_done;
 	op->devs_need_flush	= &inode->ei_devs_need_flush;

fs/bcachefs/fs-io-direct.c

@@ -500,7 +500,7 @@ static __always_inline long bch2_dio_write_loop(struct dio_write *dio)
 		dio->op.target		= dio->op.opts.foreground_target;
 		dio->op.write_point	= writepoint_hashed((unsigned long) current);
 		dio->op.nr_replicas	= dio->op.opts.data_replicas;
-		dio->op.subvol		= inode->ei_subvol;
+		dio->op.subvol		= inode->ei_inum.subvol;
 		dio->op.pos		= POS(inode->v.i_ino, (u64) req->ki_pos >> 9);
 		dio->op.devs_need_flush	= &inode->ei_devs_need_flush;
 

fs/bcachefs/fs-io.c

@@ -267,7 +267,7 @@ static int __bch2_truncate_folio(struct bch_inode_info *inode,
 		 * XXX: we're doing two index lookups when we end up reading the
 		 * folio
 		 */
-		ret = range_has_data(c, inode->ei_subvol,
+		ret = range_has_data(c, inode->ei_inum.subvol,
 				POS(inode->v.i_ino, (index << PAGE_SECTORS_SHIFT)),
 				POS(inode->v.i_ino, (index << PAGE_SECTORS_SHIFT) + PAGE_SECTORS));
 		if (ret <= 0)
@@ -618,7 +618,7 @@ static noinline int __bchfs_fallocate(struct bch_inode_info *inode, int mode,
 		bch2_trans_begin(trans);
 
 		ret = bch2_subvolume_get_snapshot(trans,
-					inode->ei_subvol, &snapshot);
+					inode->ei_inum.subvol, &snapshot);
 		if (ret)
 			goto bkey_err;
 
@@ -823,7 +823,7 @@ static int quota_reserve_range(struct bch_inode_info *inode,
 retry:
 	bch2_trans_begin(trans);
 
-	ret = bch2_subvolume_get_snapshot(trans, inode->ei_subvol, &snapshot);
+	ret = bch2_subvolume_get_snapshot(trans, inode->ei_inum.subvol, &snapshot);
 	if (ret)
 		goto err;
 

fs/bcachefs/fs-ioctl.c

@@ -100,7 +100,7 @@ static int bch2_ioc_setflags(struct bch_fs *c,
 	}
 
 	mutex_lock(&inode->ei_update_lock);
-	ret   = bch2_subvol_is_ro(c, inode->ei_subvol) ?:
+	ret   = bch2_subvol_is_ro(c, inode->ei_inum.subvol) ?:
 		bch2_write_inode(c, inode, bch2_inode_flags_set, &s,
 			       ATTR_CTIME);
 	mutex_unlock(&inode->ei_update_lock);
@@ -184,7 +184,7 @@ static int bch2_ioc_fssetxattr(struct bch_fs *c,
 	}
 
 	mutex_lock(&inode->ei_update_lock);
-	ret   = bch2_subvol_is_ro(c, inode->ei_subvol) ?:
+	ret   = bch2_subvol_is_ro(c, inode->ei_inum.subvol) ?:
 		bch2_set_projid(c, inode, fa.fsx_projid) ?:
 		bch2_write_inode(c, inode, fssetxattr_inode_update_fn, &s,
 			       ATTR_CTIME);

fs/bcachefs/fs.h

@@ -13,6 +13,9 @@
 
 struct bch_inode_info {
 	struct inode		v;
+	struct rhash_head	hash;
+	subvol_inum		ei_inum;
+
 	struct list_head	ei_vfs_inode_list;
 	unsigned long		ei_flags;
 
@@ -24,8 +27,6 @@ struct bch_inode_info {
 	struct mutex		ei_quota_lock;
 	struct bch_qid		ei_qid;
 
-	u32			ei_subvol;
-
 	/*
 	 * When we've been doing nocow writes we'll need to issue flushes to the
 	 * underlying block devices
@@ -50,10 +51,7 @@ struct bch_inode_info {
 
 static inline subvol_inum inode_inum(struct bch_inode_info *inode)
 {
-	return (subvol_inum) {
-		.subvol	= inode->ei_subvol,
-		.inum	= inode->ei_inode.bi_inum,
-	};
+	return inode->ei_inum;
 }
 
 struct bch_inode_info *__bch2_inode_hash_find(struct bch_fs *, subvol_inum);
@@ -69,6 +67,7 @@ struct bch_inode_info *__bch2_inode_hash_find(struct bch_fs *, subvol_inum);
  * those:
  */
 #define EI_INODE_SNAPSHOT		1
+#define EI_INODE_HASHED			2
 
 #define to_bch_ei(_inode)					\
 	container_of_or_null(_inode, struct bch_inode_info, v)
@@ -189,6 +188,9 @@ int __bch2_unlink(struct inode *, struct dentry *, bool);
 
 void bch2_evict_subvolume_inodes(struct bch_fs *, snapshot_id_list *);
 
+void bch2_fs_vfs_exit(struct bch_fs *);
+int bch2_fs_vfs_init(struct bch_fs *);
+
 void bch2_vfs_exit(void);
 int bch2_vfs_init(void);
 
@@ -203,6 +205,10 @@ static inline struct bch_inode_info *__bch2_inode_hash_find(struct bch_fs *c, su
 
 static inline void bch2_evict_subvolume_inodes(struct bch_fs *c,
 					       snapshot_id_list *s) {}
+
+static inline void bch2_fs_vfs_exit(struct bch_fs *c) {}
+static inline int bch2_fs_vfs_init(struct bch_fs *c) { return 0; }
+
 static inline void bch2_vfs_exit(void) {}
 static inline int bch2_vfs_init(void) { return 0; }
 

fs/bcachefs/inode.c

@@ -365,7 +365,7 @@ int bch2_inode_peek(struct btree_trans *trans,
 		    subvol_inum inum, unsigned flags)
 {
 	int ret = bch2_inode_peek_nowarn(trans, iter, inode, inum, flags);
-	bch_err_msg(trans->c, ret, "looking up inum %u:%llu:", inum.subvol, inum.inum);
+	bch_err_msg(trans->c, ret, "looking up inum %llu:%llu:", inum.subvol, inum.inum);
 	return ret;
 }
 

fs/bcachefs/subvolume_types.h

@@ -30,7 +30,8 @@ struct snapshot_table {
 };
 
 typedef struct {
-	u32		subvol;
+	/* we can't have padding in this struct: */
+	u64		subvol;
 	u64		inum;
 } subvol_inum;
 

fs/bcachefs/super.c

@@ -543,6 +543,7 @@ static void __bch2_fs_free(struct bch_fs *c)
 	bch2_fs_fs_io_direct_exit(c);
 	bch2_fs_fs_io_buffered_exit(c);
 	bch2_fs_fsio_exit(c);
+	bch2_fs_vfs_exit(c);
 	bch2_fs_ec_exit(c);
 	bch2_fs_encryption_exit(c);
 	bch2_fs_nocow_locking_exit(c);
@@ -926,6 +927,7 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 	    bch2_fs_encryption_init(c) ?:
 	    bch2_fs_compress_init(c) ?:
 	    bch2_fs_ec_init(c) ?:
+	    bch2_fs_vfs_init(c) ?:
 	    bch2_fs_fsio_init(c) ?:
 	    bch2_fs_fs_io_buffered_init(c) ?:
 	    bch2_fs_fs_io_direct_init(c);

fs/bcachefs/xattr.c

@@ -306,7 +306,7 @@ ssize_t bch2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 	bch2_trans_begin(trans);
 	iter = (struct btree_iter) { NULL };
 
-	ret = bch2_subvolume_get_snapshot(trans, inode->ei_subvol, &snapshot);
+	ret = bch2_subvolume_get_snapshot(trans, inode->ei_inum.subvol, &snapshot);
 	if (ret)
 		goto err;