zfs_rename: support RENAME_* flags

Implement support for Linux's RENAME_* flags (for renameat2). Aside from
being quite useful for userspace (providing race-free ways to exchange
paths and implement mv --no-clobber), they are used by overlayfs and are
thus required in order to use overlayfs-on-ZFS.

In order for us to represent the new renameat2(2) flags in the ZIL, we
need to create a new transaction type (to be backwards-compatible).
Since RENAME_EXCHANGE and RENAME_WHITEOUT are mutually exclusive they
deserve separate types. We just re-use the logic of
zfs_{log,replay}_rename() with the only change being the transaction
types and the associate vfsflags passed to zfs_rename().

RENAME_NOREPLACE doesn't need an entry because if the renameat2(2) fails
because of RENAME_NOREPLACE there won't be a ZIL entry for the operation
(and if it succeeds then it should also succeed on-replay).

Unfortunately, more work is required in order use overlayfs-on-ZFS
(namely we have to remove our .d_revalidate hook, since overlayfs
refuses to use a filesystem with d_revalidate as an upperdir).

Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
Signed-off-by: Pavel Snajdr <snajpa@snajpa.net>

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>

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

@@ -547,6 +547,19 @@ static inline void zfs_gid_write(struct inode *ip, gid_t gid)
 #define	zpl_follow_up(path)			follow_up(path)
 #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/sys/zil.h

@@ -162,7 +162,9 @@ typedef enum zil_create {
 #define	TX_MKDIR_ATTR		18	/* mkdir with attr */
 #define	TX_MKDIR_ACL_ATTR	19	/* mkdir with ACL + attrs */
 #define	TX_WRITE2		20	/* dmu_sync EALREADY write */
-#define	TX_MAX_TYPE		21	/* Max transaction type */
+#define	TX_EXCHANGE		21	/* Exchange two paths */
+#define	TX_WHITEOUT		22	/* Rename a file, leaving a whiteout */
+#define	TX_MAX_TYPE		23	/* Max transaction type */
 
 /*
  * The transactions for mkdir, symlink, remove, rmdir, link, and rename

module/os/linux/zfs/zfs_vnops.c

@@ -3671,10 +3671,24 @@ zfs_rename(struct inode *sdip, char *snm, struct inode *tdip, char *tnm,
 	int		error = 0;
 	int		zflg = 0;
 	boolean_t	waited = B_FALSE;
+	uint64_t	txtype;
+	/* Needed for whiteout inode creation. */
+	vattr_t		wo_vap;
+	uint64_t	wo_projid;
+	boolean_t	fuid_dirtied;
+	zfs_acl_ids_t	acl_ids;
+	boolean_t	have_acl = B_FALSE;
 
 	if (snm == NULL || tnm == NULL)
 		return (SET_ERROR(EINVAL));
 
+	if (flags & RENAME_EXCHANGE)
+		txtype = TX_EXCHANGE;
+	else if (flags & RENAME_WHITEOUT)
+		txtype = TX_WHITEOUT;
+	else
+		txtype = TX_RENAME;
+
 	ZFS_ENTER(zfsvfs);
 	zilog = zfsvfs->z_log;
 
@@ -3842,14 +3856,14 @@ zfs_rename(struct inode *sdip, char *snm, struct inode *tdip, char *tnm,
 		error = SET_ERROR(EXDEV);
 		goto out;
 	}
+	wo_projid = szp->z_projid;
 
 	/*
 	 * Must have write access at the source to remove the old entry
 	 * and write access at the target to create the new entry.
 	 * Note that if target and source are the same, this can be
 	 * done in a single check.
 	 */
-
 	if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)))
 		goto out;
 
@@ -3866,15 +3880,21 @@ zfs_rename(struct inode *sdip, char *snm, struct inode *tdip, char *tnm,
 	 * Does target exist?
 	 */
 	if (tzp) {
+		if (flags & RENAME_NOREPLACE) {
+			error = SET_ERROR(EEXIST);
+			goto out;
+		}
 		/*
-		 * Source and target must be the same type.
+		 * Source and target must be the same type (unless exchanging).
 		 */
-		boolean_t s_is_dir = S_ISDIR(ZTOI(szp)->i_mode) != 0;
-		boolean_t t_is_dir = S_ISDIR(ZTOI(tzp)->i_mode) != 0;
+		if (txtype != TX_EXCHANGE) {
+			boolean_t s_is_dir = S_ISDIR(ZTOI(szp)->i_mode) != 0;
+			boolean_t t_is_dir = S_ISDIR(ZTOI(tzp)->i_mode) != 0;
 
-		if (s_is_dir != t_is_dir) {
-			error = SET_ERROR(s_is_dir ? ENOTDIR : EISDIR);
-			goto out;
+			if (s_is_dir != t_is_dir) {
+				error = SET_ERROR(s_is_dir ? ENOTDIR : EISDIR);
+				goto out;
+			}
 		}
 		/*
 		 * POSIX dictates that when the source and target
@@ -3886,11 +3906,38 @@ zfs_rename(struct inode *sdip, char *snm, struct inode *tdip, char *tnm,
 			goto out;
 		}
 	}
+	/* Target must exist for RENAME_EXCHANGE. */
+	if (!tzp && txtype == TX_EXCHANGE) {
+		error = SET_ERROR(ENOENT);
+		goto out;
+	}
+
+	/* Set up inode creation for RENAME_WHITEOUT. */
+	if (txtype == TX_WHITEOUT) {
+		error = zfs_zaccess(sdzp, ACE_ADD_FILE, 0, B_FALSE, cr);
+		if (error)
+			goto out;
+
+		zpl_vap_init(&wo_vap, ZTOI(sdzp), S_IFCHR, cr);
+		/* Can't use of makedevice() here, so hard-code it. */
+		wo_vap.va_rdev = 0;
+
+		error = zfs_acl_ids_create(sdzp, 0, &wo_vap, cr, NULL,
+		    &acl_ids);
+		if (error)
+			goto out;
+		have_acl = B_TRUE;
+
+		if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, wo_projid)) {
+			error = SET_ERROR(EDQUOT);
+			goto out;
+		}
+	}
 
 	tx = dmu_tx_create(zfsvfs->z_os);
 	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
 	dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
-	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
+	dmu_tx_hold_zap(tx, sdzp->z_id, txtype == TX_EXCHANGE, snm);
 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
 	if (sdzp != tdzp) {
 		dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
@@ -3900,7 +3947,21 @@ zfs_rename(struct inode *sdip, char *snm, struct inode *tdip, char *tnm,
 		dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
 		zfs_sa_upgrade_txholds(tx, tzp);
 	}
+	if (txtype == TX_WHITEOUT) {
+		dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
+		    ZFS_SA_BASE_ATTR_SIZE);
 
+		dmu_tx_hold_zap(tx, sdzp->z_id, TRUE, snm);
+		dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
+		if (!zfsvfs->z_use_sa &&
+		    acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
+			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
+			    0, acl_ids.z_aclp->z_acl_bytes);
+		}
+	}
+	fuid_dirtied = zfsvfs->z_fuid_dirty;
+	if (fuid_dirtied)
+		zfs_fuid_txhold(zfsvfs, tx);
 	zfs_sa_upgrade_txholds(tx, szp);
 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
 	error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
@@ -3948,28 +4009,77 @@ zfs_rename(struct inode *sdip, char *snm, struct inode *tdip, char *tnm,
 	 * Unlink the target.
 	 */
 	if (tzp) {
-		error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
+		int tzflg = zflg;
+
+		if (txtype == TX_EXCHANGE) {
+			/* This inode will be re-linked soon. */
+			tzflg |= ZRENAMING;
+
+			tzp->z_pflags |= ZFS_AV_MODIFIED;
+			if (sdzp->z_pflags & ZFS_PROJINHERIT)
+				tzp->z_pflags |= ZFS_PROJINHERIT;
+
+			error = sa_update(tzp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
+			    (void *)&tzp->z_pflags, sizeof (uint64_t), tx);
+			ASSERT0(error);
+		}
+		error = zfs_link_destroy(tdl, tzp, tx, tzflg, NULL);
 		ASSERT0(error);
 	}
 
 	/*
-	 * Create a new link at the target.
+	 * Create the new target links:
+	 *   * We always link the target.
+	 *   * RENAME_WHITEOUT: Create a whiteout inode in-place of the source.
+	 *   * RENAME_EXCHANGE: Link the old target to the source.
 	 */
 	error = zfs_link_create(tdl, szp, tx, ZRENAMING);
-	if (error) {
+
 	/*
 	 * If we have removed the existing target, a subsequent call to
 	 * zfs_link_create() to add back the same entry, but with a new
 	 * dnode (szp), should not fail.
 	 */
 	ASSERT0(error);
 
-	zfs_log_rename(zilog, tx, TX_RENAME |
+	switch (txtype) {
+		case TX_EXCHANGE:
+			error = zfs_link_create(sdl, tzp, tx, ZRENAMING);
+			/*
+			 * The same argument as zfs_link_create() failing for
+			 * szp applies here, since the source directory must
+			 * have had an entry we are replacing.
+			 */
+			ASSERT3U(error, ==, 0);
+			if (error)
+				goto commit_unlink_td_szp;
+			break;
+		case TX_WHITEOUT: {
+			znode_t		*wzp;
+
+			zfs_mknode(sdzp, &wo_vap, tx, cr, 0, &wzp, &acl_ids);
+			error = zfs_link_create(sdl, wzp, tx, ZNEW);
+			if (error) {
+				zfs_znode_delete(wzp, tx);
+				remove_inode_hash(ZTOI(wzp));
+				goto commit_unlink_td_szp;
+			}
+			/* No need to zfs_log_create_txtype here. */
+		}
+	}
+
+	if (fuid_dirtied)
+		zfs_fuid_sync(zfsvfs, tx);
+
+	zfs_log_rename(zilog, tx, txtype |
 	    (flags & FIGNORECASE ? TX_CI : 0), sdzp,
 	    sdl->dl_name, tdzp, tdl->dl_name, szp);
 
 	dmu_tx_commit(tx);
 out:
+	if (have_acl)
+		zfs_acl_ids_free(&acl_ids);
+
 	if (zl != NULL)
 		zfs_rename_unlock(&zl);
 

module/os/linux/zfs/zpl_inode.c

@@ -407,13 +407,13 @@ zpl_rename2(struct inode *sdip, struct dentry *sdentry,
 	int error;
 	fstrans_cookie_t cookie;
 
-	/* We don't have renameat2(2) support */
-	if (flags)
+	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
 		return (-EINVAL);
 
 	crhold(cr);
 	cookie = spl_fstrans_mark();
-	error = -zfs_rename(sdip, dname(sdentry), tdip, dname(tdentry), cr, 0);
+	error = -zfs_rename(sdip, dname(sdentry), tdip, dname(tdentry), cr,
+	    flags);
 	spl_fstrans_unmark(cookie);
 	crfree(cr);
 	ASSERT3S(error, <=, 0);

module/zfs/zfs_log.c

@@ -474,7 +474,9 @@ zfs_log_symlink(zilog_t *zilog, dmu_tx_t *tx, uint64_t txtype,
 }
 
 /*
- * Handles TX_RENAME transactions.
+ * Handles TX_{RENAME,EXCHANGE,WHITEOUT} transactions. They all have the same
+ * underyling structure (lr_rename_t) but have different txtypes to indicate
+ * different renameat2(2) flags.
  */
 void
 zfs_log_rename(zilog_t *zilog, dmu_tx_t *tx, uint64_t txtype,

module/zfs/zfs_replay.c

@@ -625,15 +625,14 @@ zfs_replay_link(void *arg1, void *arg2, boolean_t byteswap)
 }
 
 static int
-zfs_replay_rename(void *arg1, void *arg2, boolean_t byteswap)
+_zfs_replay_renameat2(void *arg1, void *arg2, boolean_t byteswap, int vflg)
 {
 	zfsvfs_t *zfsvfs = arg1;
 	lr_rename_t *lr = arg2;
 	char *sname = (char *)(lr + 1);	/* sname and tname follow lr_rename_t */
 	char *tname = sname + strlen(sname) + 1;
 	znode_t *sdzp, *tdzp;
 	int error;
-	int vflg = 0;
 
 	if (byteswap)
 		byteswap_uint64_array(lr, sizeof (*lr));
@@ -657,6 +656,24 @@ zfs_replay_rename(void *arg1, void *arg2, boolean_t byteswap)
 	return (error);
 }
 
+static int
+zfs_replay_rename(void *arg1, void *arg2, boolean_t byteswap)
+{
+	return (_zfs_replay_renameat2(arg1, arg2, byteswap, 0));
+}
+
+static int
+zfs_replay_exchange(void *arg1, void *arg2, boolean_t byteswap)
+{
+	return (_zfs_replay_renameat2(arg1, arg2, byteswap, RENAME_EXCHANGE));
+}
+
+static int
+zfs_replay_whiteout(void *arg1, void *arg2, boolean_t byteswap)
+{
+	return (_zfs_replay_renameat2(arg1, arg2, byteswap, RENAME_WHITEOUT));
+}
+
 static int
 zfs_replay_write(void *arg1, void *arg2, boolean_t byteswap)
 {
@@ -981,4 +998,6 @@ zil_replay_func_t *zfs_replay_vector[TX_MAX_TYPE] = {
 	zfs_replay_create,	/* TX_MKDIR_ATTR */
 	zfs_replay_create_acl,	/* TX_MKDIR_ACL_ATTR */
 	zfs_replay_write2,	/* TX_WRITE2 */
+	zfs_replay_exchange,	/* TX_EXCHANGE */
+	zfs_replay_whiteout,	/* TX_WHITEOUT */
 };