OverlayFS support (d_revalidate out and support renameat2 flags)

AUTHORS

@@ -19,6 +19,7 @@ CONTRIBUTORS:
     Albert Lee <trisk@nexenta.com>
     Alec Salazar <alec.j.salazar@gmail.com>
     Alejandro R. Sedeño <asedeno@mit.edu>
+    Aleksa Sarai <cyphar@cyphar.com>
     Alek Pinchuk <alek@nexenta.com>
     Alex Braunegg <alex.braunegg@gmail.com>
     Alex McWhirter <alexmcwhirter@triadic.us>
@@ -233,6 +234,7 @@ CONTRIBUTORS:
     Paul Dagnelie <pcd@delphix.com>
     Paul Zuchowski <pzuchowski@datto.com>
     Pavel Boldin <boldin.pavel@gmail.com>
+    Pavel Snajdr <snajpa@snajpa.net>
     Pavel Zakharov <pavel.zakharov@delphix.com>
     Pawel Jakub Dawidek <pjd@FreeBSD.org>
     Pedro Giffuni <pfg@freebsd.org>

config/kernel-dentry-alias.m4

@@ -0,0 +1,30 @@
+dnl #
+dnl # 3.18 API change
+dnl # Dentry aliases are in d_u struct dentry member
+dnl #
+AC_DEFUN([ZFS_AC_KERNEL_SRC_DENTRY_ALIAS_D_U], [
+	ZFS_LINUX_TEST_SRC([dentry_alias_d_u], [
+		#include <linux/fs.h>
+		#include <linux/dcache.h>
+		#include <linux/list.h>
+	], [
+		struct inode *inode __attribute__ ((unused)) = NULL;
+		struct dentry *dentry __attribute__ ((unused)) = NULL;
+		hlist_for_each_entry(dentry, &inode->i_dentry,
+		    d_u.d_alias) {
+			d_drop(dentry);
+		}
+	])
+])
+
+AC_DEFUN([ZFS_AC_KERNEL_DENTRY_ALIAS_D_U], [
+	AC_MSG_CHECKING([whether dentry aliases are in d_u member])
+	ZFS_LINUX_TEST_RESULT([dentry_alias_d_u], [
+		AC_MSG_RESULT(yes)
+		AC_DEFINE(HAVE_DENTRY_D_U_ALIASES, 1,
+		    [dentry aliases are in d_u member])
+	],[
+		AC_MSG_RESULT(no)
+	])
+])
+

config/kernel-rename.m4

@@ -3,8 +3,20 @@ dnl # 4.9 API change,
 dnl # iops->rename2() merged into iops->rename(), and iops->rename() now wants
 dnl # flags.
 dnl #
-AC_DEFUN([ZFS_AC_KERNEL_SRC_RENAME_WANTS_FLAGS], [
-	ZFS_LINUX_TEST_SRC([inode_operations_rename], [
+AC_DEFUN([ZFS_AC_KERNEL_SRC_RENAME], [
+	ZFS_LINUX_TEST_SRC([inode_operations_rename2], [
+		#include <linux/fs.h>
+		int rename2_fn(struct inode *sip, struct dentry *sdp,
+			struct inode *tip, struct dentry *tdp,
+			unsigned int flags) { return 0; }
+
+		static const struct inode_operations
+		    iops __attribute__ ((unused)) = {
+			.rename2 = rename2_fn,
+		};
+	],[])
+
+	ZFS_LINUX_TEST_SRC([inode_operations_rename_flags], [
 		#include <linux/fs.h>
 		int rename_fn(struct inode *sip, struct dentry *sdp,
 			struct inode *tip, struct dentry *tdp,
@@ -15,15 +27,45 @@ AC_DEFUN([ZFS_AC_KERNEL_SRC_RENAME_WANTS_FLAGS], [
 			.rename = rename_fn,
 		};
 	],[])
+
+	ZFS_LINUX_TEST_SRC([dir_inode_operations_wrapper_rename2], [
+		#include <linux/fs.h>
+		int rename2_fn(struct inode *sip, struct dentry *sdp,
+			struct inode *tip, struct dentry *tdp,
+			unsigned int flags) { return 0; }
+
+		static const struct inode_operations_wrapper
+		    iops __attribute__ ((unused)) = {
+			.rename2 = rename2_fn,
+		};
+	],[])
 ])
 
-AC_DEFUN([ZFS_AC_KERNEL_RENAME_WANTS_FLAGS], [
-	AC_MSG_CHECKING([whether iops->rename() wants flags])
-	ZFS_LINUX_TEST_RESULT([inode_operations_rename], [
+AC_DEFUN([ZFS_AC_KERNEL_RENAME], [
+	AC_MSG_CHECKING([whether iops->rename2() exists])
+	ZFS_LINUX_TEST_RESULT([inode_operations_rename2], [
 		AC_MSG_RESULT(yes)
-		AC_DEFINE(HAVE_RENAME_WANTS_FLAGS, 1,
-		    [iops->rename() wants flags])
+		AC_DEFINE(HAVE_RENAME2, 1, [iops->rename2() exists])
 	],[
 		AC_MSG_RESULT(no)
+
+		AC_MSG_CHECKING([whether iops->rename() wants flags])
+		ZFS_LINUX_TEST_RESULT([inode_operations_rename_flags], [
+			AC_MSG_RESULT(yes)
+			AC_DEFINE(HAVE_RENAME_WANTS_FLAGS, 1,
+			    [iops->rename() wants flags])
+		],[
+			AC_MSG_RESULT(no)
+
+			AC_MSG_CHECKING([whether struct inode_operations_wrapper takes .rename2()])
+			ZFS_LINUX_TEST_RESULT([dir_inode_operations_wrapper_rename2], [
+				AC_MSG_RESULT(yes)
+				AC_DEFINE(HAVE_RENAME2_OPERATIONS_WRAPPER, 1,
+				    [struct inode_operations_wrapper takes .rename2()])
+			],[
+				AC_MSG_RESULT(no)
+			])
+		])
 	])
 ])
+

config/kernel.m4

@@ -87,6 +87,7 @@ AC_DEFUN([ZFS_AC_KERNEL_TEST_SRC], [
 	ZFS_AC_KERNEL_SRC_SETATTR_PREPARE
 	ZFS_AC_KERNEL_SRC_INSERT_INODE_LOCKED
 	ZFS_AC_KERNEL_SRC_DENTRY
+	ZFS_AC_KERNEL_SRC_DENTRY_ALIAS_D_U
 	ZFS_AC_KERNEL_SRC_TRUNCATE_SETSIZE
 	ZFS_AC_KERNEL_SRC_SECURITY_INODE
 	ZFS_AC_KERNEL_SRC_FST_MOUNT
@@ -110,7 +111,7 @@ AC_DEFUN([ZFS_AC_KERNEL_TEST_SRC], [
 	ZFS_AC_KERNEL_SRC_KUIDGID_T
 	ZFS_AC_KERNEL_SRC_KUID_HELPERS
 	ZFS_AC_KERNEL_SRC_MODULE_PARAM_CALL_CONST
-	ZFS_AC_KERNEL_SRC_RENAME_WANTS_FLAGS
+	ZFS_AC_KERNEL_SRC_RENAME
 	ZFS_AC_KERNEL_SRC_CURRENT_TIME
 	ZFS_AC_KERNEL_SRC_USERNS_CAPABILITIES
 	ZFS_AC_KERNEL_SRC_IN_COMPAT_SYSCALL
@@ -181,6 +182,7 @@ AC_DEFUN([ZFS_AC_KERNEL_TEST_RESULT], [
 	ZFS_AC_KERNEL_SETATTR_PREPARE
 	ZFS_AC_KERNEL_INSERT_INODE_LOCKED
 	ZFS_AC_KERNEL_DENTRY
+	ZFS_AC_KERNEL_DENTRY_ALIAS_D_U
 	ZFS_AC_KERNEL_TRUNCATE_SETSIZE
 	ZFS_AC_KERNEL_SECURITY_INODE
 	ZFS_AC_KERNEL_FST_MOUNT
@@ -204,7 +206,7 @@ AC_DEFUN([ZFS_AC_KERNEL_TEST_RESULT], [
 	ZFS_AC_KERNEL_KUIDGID_T
 	ZFS_AC_KERNEL_KUID_HELPERS
 	ZFS_AC_KERNEL_MODULE_PARAM_CALL_CONST
-	ZFS_AC_KERNEL_RENAME_WANTS_FLAGS
+	ZFS_AC_KERNEL_RENAME
 	ZFS_AC_KERNEL_CURRENT_TIME
 	ZFS_AC_KERNEL_USERNS_CAPABILITIES
 	ZFS_AC_KERNEL_IN_COMPAT_SYSCALL

configure.ac

@@ -207,6 +207,7 @@ AC_CONFIG_FILES([
 	tests/zfs-tests/cmd/randfree_file/Makefile
 	tests/zfs-tests/cmd/randwritecomp/Makefile
 	tests/zfs-tests/cmd/readmmap/Makefile
+	tests/zfs-tests/cmd/renameat2/Makefile
 	tests/zfs-tests/cmd/rename_dir/Makefile
 	tests/zfs-tests/cmd/rm_lnkcnt_zero_file/Makefile
 	tests/zfs-tests/cmd/stride_dd/Makefile
@@ -332,6 +333,7 @@ AC_CONFIG_FILES([
 	tests/zfs-tests/tests/functional/no_space/Makefile
 	tests/zfs-tests/tests/functional/nopwrite/Makefile
 	tests/zfs-tests/tests/functional/online_offline/Makefile
+	tests/zfs-tests/tests/functional/overlayfs/Makefile
 	tests/zfs-tests/tests/functional/pool_checkpoint/Makefile
 	tests/zfs-tests/tests/functional/pool_names/Makefile
 	tests/zfs-tests/tests/functional/poolversion/Makefile
@@ -346,6 +348,7 @@ AC_CONFIG_FILES([
 	tests/zfs-tests/tests/functional/refquota/Makefile
 	tests/zfs-tests/tests/functional/refreserv/Makefile
 	tests/zfs-tests/tests/functional/removal/Makefile
+	tests/zfs-tests/tests/functional/renameat2/Makefile
 	tests/zfs-tests/tests/functional/rename_dirs/Makefile
 	tests/zfs-tests/tests/functional/replacement/Makefile
 	tests/zfs-tests/tests/functional/reservation/Makefile

include/os/linux/kernel/linux/dcache_compat.h

@@ -61,4 +61,25 @@ d_clear_d_op(struct dentry *dentry)
 	    DCACHE_OP_REVALIDATE | DCACHE_OP_DELETE);
 }
 
+/*
+ * Walk and invalidate all dentry aliases of an inode
+ * unless it's a mountpoint
+ */
+static inline void
+zpl_d_drop_aliases(struct inode *inode)
+{
+	struct dentry *dentry;
+	spin_lock(&inode->i_lock);
+#ifdef HAVE_DENTRY_D_U_ALIASES
+	hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
+#else
+	hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
+#endif
+		if (!IS_ROOT(dentry) && !d_mountpoint(dentry) &&
+		    (dentry->d_inode == inode)) {
+			d_drop(dentry);
+		}
+	}
+	spin_unlock(&inode->i_lock);
+}
 #endif /* _ZFS_DCACHE_H */

include/os/linux/kernel/linux/vfs_compat.h

@@ -340,6 +340,19 @@ static inline void zfs_gid_write(struct inode *ip, gid_t gid)
 #endif
 }
 
+/*
+ * 3.15 API change
+ */
+#ifndef RENAME_NOREPLACE
+#define	RENAME_NOREPLACE	(1 << 0) /* Don't overwrite target */
+#endif
+#ifndef RENAME_EXCHANGE
+#define	RENAME_EXCHANGE		(1 << 1) /* Exchange source and dest */
+#endif
+#ifndef RENAME_WHITEOUT
+#define	RENAME_WHITEOUT		(1 << 2) /* Whiteout source */
+#endif
+
 /*
  * 4.9 API change
  */

include/os/linux/zfs/sys/zfs_dir.h

@@ -52,6 +52,7 @@ extern "C" {
 extern int zfs_dirent_lock(zfs_dirlock_t **, znode_t *, char *, znode_t **,
     int, int *, pathname_t *);
 extern void zfs_dirent_unlock(zfs_dirlock_t *);
+extern int zfs_drop_nlink(znode_t *, dmu_tx_t *, boolean_t *);
 extern int zfs_link_create(zfs_dirlock_t *, znode_t *, dmu_tx_t *, int);
 extern int zfs_link_destroy(zfs_dirlock_t *, znode_t *, dmu_tx_t *, int,
     boolean_t *);

include/os/linux/zfs/sys/zpl.h

@@ -42,10 +42,16 @@ extern void zpl_vap_init(vattr_t *vap, struct inode *dir,
     umode_t mode, cred_t *cr);
 
 extern const struct inode_operations zpl_inode_operations;
+#ifdef HAVE_RENAME2_OPERATIONS_WRAPPER
+extern const struct inode_operations_wrapper zpl_dir_inode_operations;
+#else
 extern const struct inode_operations zpl_dir_inode_operations;
+#endif
 extern const struct inode_operations zpl_symlink_inode_operations;
 extern const struct inode_operations zpl_special_inode_operations;
+#if 0
 extern dentry_operations_t zpl_dentry_operations;
+#endif
 
 /* zpl_file.c */
 extern ssize_t zpl_read_common(struct inode *ip, const char *buf,

include/sys/zfs_znode.h

@@ -273,6 +273,13 @@ extern void zfs_log_link(zilog_t *zilog, dmu_tx_t *tx, uint64_t txtype,
     znode_t *dzp, znode_t *zp, char *name);
 extern void zfs_log_symlink(zilog_t *zilog, dmu_tx_t *tx, uint64_t txtype,
     znode_t *dzp, znode_t *zp, char *name, char *link);
+extern void zfs_log_rename_exchange(zilog_t *zilog, dmu_tx_t *tx,
+    uint64_t txtype, znode_t *sdzp, char *sname, znode_t *tdzp,
+    char *dname, znode_t *szp);
+extern void zfs_log_rename_whiteout(zilog_t *zilog, dmu_tx_t *tx,
+    uint64_t txtype, znode_t *sdzp, char *sname, znode_t *tdzp,
+    char *dname, znode_t *szp, znode_t *wzp, vsecattr_t *vsecp,
+    zfs_fuid_info_t *fuidp, vattr_t *vap);
 extern void zfs_log_rename(zilog_t *zilog, dmu_tx_t *tx, uint64_t txtype,
     znode_t *sdzp, char *sname, znode_t *tdzp, char *dname, znode_t *szp);
 extern void zfs_log_write(zilog_t *zilog, dmu_tx_t *tx, int txtype,

module/os/linux/zfs/zfs_dir.c

@@ -905,6 +905,74 @@ zfs_dropname(zfs_dirlock_t *dl, znode_t *zp, znode_t *dzp, dmu_tx_t *tx,
 	return (error);
 }
 
+static int
+zfs_drop_nlink_locked(znode_t *zp, dmu_tx_t *tx, boolean_t *unlinkedp)
+{
+	zfsvfs_t	*zfsvfs = ZTOZSB(zp);
+	int		zp_is_dir = S_ISDIR(ZTOI(zp)->i_mode);
+	boolean_t	unlinked = B_FALSE;
+	sa_bulk_attr_t	bulk[3];
+	uint64_t	mtime[2], ctime[2];
+	uint64_t	links;
+	int		count = 0;
+	int		error;
+
+	if (zp_is_dir && !zfs_dirempty(zp))
+		return (SET_ERROR(ENOTEMPTY));
+
+	if (ZTOI(zp)->i_nlink <= zp_is_dir) {
+		zfs_panic_recover("zfs: link count on %lu is %u, "
+		    "should be at least %u", zp->z_id,
+		    (int)ZTOI(zp)->i_nlink, zp_is_dir + 1);
+		set_nlink(ZTOI(zp), zp_is_dir + 1);
+	}
+	drop_nlink(ZTOI(zp));
+	if (ZTOI(zp)->i_nlink == zp_is_dir) {
+		zp->z_unlinked = B_TRUE;
+		clear_nlink(ZTOI(zp));
+		unlinked = B_TRUE;
+	} else {
+		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs),
+		    NULL, &ctime, sizeof (ctime));
+		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
+		    NULL, &zp->z_pflags, sizeof (zp->z_pflags));
+		zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime,
+		    ctime);
+	}
+	links = ZTOI(zp)->i_nlink;
+	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
+	    NULL, &links, sizeof (links));
+	error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
+	ASSERT3U(error, ==, 0);
+
+	if (unlinkedp != NULL)
+		*unlinkedp = unlinked;
+	else if (unlinked)
+		zfs_unlinked_add(zp, tx);
+
+	return (0);
+}
+
+/*
+ * Forcefully drop an nlink reference from (zp) and mark it for deletion if it
+ * was the last link. This *must* only be done to znodes which have already
+ * been zfs_link_destroy()'d with ZRENAMING. This is explicitly only used in
+ * the error path of zfs_rename(), where we have to correct the nlink count if
+ * we failed to link the target as well as failing to re-link the original
+ * znodes.
+ */
+int
+zfs_drop_nlink(znode_t *zp, dmu_tx_t *tx, boolean_t *unlinkedp)
+{
+	int error;
+
+	mutex_enter(&zp->z_lock);
+	error = zfs_drop_nlink_locked(zp, tx, unlinkedp);
+	mutex_exit(&zp->z_lock);
+
+	return (error);
+}
+
 /*
  * Unlink zp from dl, and mark zp for deletion if this was the last link. Can
  * fail if zp is a mount point (EBUSY) or a non-empty directory (ENOTEMPTY).
@@ -945,31 +1013,9 @@ zfs_link_destroy(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag,
 			return (error);
 		}
 
-		if (ZTOI(zp)->i_nlink <= zp_is_dir) {
-			zfs_panic_recover("zfs: link count on %lu is %u, "
-			    "should be at least %u", zp->z_id,
-			    (int)ZTOI(zp)->i_nlink, zp_is_dir + 1);
-			set_nlink(ZTOI(zp), zp_is_dir + 1);
-		}
-		drop_nlink(ZTOI(zp));
-		if (ZTOI(zp)->i_nlink == zp_is_dir) {
-			zp->z_unlinked = B_TRUE;
-			clear_nlink(ZTOI(zp));
-			unlinked = B_TRUE;
-		} else {
-			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs),
-			    NULL, &ctime, sizeof (ctime));
-			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
-			    NULL, &zp->z_pflags, sizeof (zp->z_pflags));
-			zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime,
-			    ctime);
-		}
-		links = ZTOI(zp)->i_nlink;
-		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
-		    NULL, &links, sizeof (links));
-		error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
-		count = 0;
-		ASSERT(error == 0);
+		/* The only error is !zfs_dirempty() and we checked earlier. */
+		error = zfs_drop_nlink_locked(zp, tx, &unlinked);
+		ASSERT3U(error, ==, 0);
 		mutex_exit(&zp->z_lock);
 	} else {
 		error = zfs_dropname(dl, zp, dzp, tx, flag);

module/os/linux/zfs/zfs_vfsops.c

@@ -1926,7 +1926,9 @@ zfs_domount(struct super_block *sb, zfs_mnt_t *zm, int silent)
 	sb->s_op = &zpl_super_operations;
 	sb->s_xattr = zpl_xattr_handlers;
 	sb->s_export_op = &zpl_export_operations;
+#if 0
 	sb->s_d_op = &zpl_dentry_operations;
+#endif
 
 	/* Set features for file system. */
 	zfs_set_fuid_feature(zfsvfs);
@@ -2275,6 +2277,7 @@ zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
 	    zp = list_next(&zfsvfs->z_all_znodes, zp)) {
 		err2 = zfs_rezget(zp);
 		if (err2) {
+			zpl_d_drop_aliases(ZTOI(zp));
 			remove_inode_hash(ZTOI(zp));
 			zp->z_is_stale = B_TRUE;
 		}