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
create two new transaction types for the two flags which need
transactional-level support (RENAME_EXCHANGE and RENAME_WHITEOUT).
RENAME_NOREPLACE does not need any ZIL support because we know that if
the operation succeeded before creating the ZIL entry, there was no file
to be clobbered and thus it can be treated as a regular TX_RENAME.

Cc: Pavel Snajdr <[email protected]>
Signed-off-by: Aleksa Sarai <[email protected]>

Author: cyphar

AUTHORS

@@ -20,6 +20,7 @@ CONTRIBUTORS:
     Alec Salazar <[email protected]>
     Alejandro R. Sedeño <[email protected]>
     Alek Pinchuk <[email protected]>
+    Aleksa Sarai <[email protected]>
     Alex Braunegg <[email protected]>
     Alex McWhirter <[email protected]>
     Alex Reece <[email protected]>
@@ -236,6 +237,7 @@ CONTRIBUTORS:
     Paul Dagnelie <[email protected]>
     Paul Zuchowski <[email protected]>
     Pavel Boldin <[email protected]>
+    Pavel Snajdr <[email protected]>
     Pavel Zakharov <[email protected]>
     Pawel Jakub Dawidek <[email protected]>
     Pedro Giffuni <[email protected]>

cmd/zdb/zdb_il.c

@@ -128,6 +128,14 @@ zil_prt_rec_rename(zilog_t *zilog, int txtype, const void *arg)
 	(void) printf("%ssdoid %llu, tdoid %llu\n", tab_prefix,
 	    (u_longlong_t)lr->lr_sdoid, (u_longlong_t)lr->lr_tdoid);
 	(void) printf("%ssrc %s tgt %s\n", tab_prefix, snm, tnm);
+	switch (txtype) {
+	case TX_RENAME_EXCHANGE:
+		(void) printf("%sflags RENAME_EXCHANGE\n", tab_prefix);
+		break;
+	case TX_RENAME_WHITEOUT:
+		(void) printf("%sflags RENAME_WHITEOUT\n", tab_prefix);
+		break;
+	}
 }
 
 static int
@@ -330,6 +338,8 @@ static zil_rec_info_t zil_rec_info[TX_MAX_TYPE] = {
 	{.zri_print = zil_prt_rec_write,    .zri_name = "TX_WRITE2          "},
 	{.zri_print = zil_prt_rec_setsaxattr,
 	    .zri_name = "TX_SETSAXATTR      "},
+	{.zri_print = zil_prt_rec_rename,   .zri_name = "TX_RENAME_EXCHANGE "},
+	{.zri_print = zil_prt_rec_rename,   .zri_name = "TX_RENAME_WHITEOUT "},
 };
 
 static int

cmd/ztest.c

@@ -2368,6 +2368,8 @@ static zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
 	NULL,			/* TX_MKDIR_ACL_ATTR */
 	NULL,			/* TX_WRITE2 */
 	NULL,			/* TX_SETSAXATTR */
+	NULL,			/* TX_RENAME_EXCHANGE */
+	NULL,			/* TX_RENAME_WHITEOUT */
 };
 
 /*

config/kernel-rename.m4

@@ -1,8 +1,28 @@
 AC_DEFUN([ZFS_AC_KERNEL_SRC_RENAME], [
+	dnl #
+	dnl # 3.9 (to 4.9) API change,
+	dnl #
+	dnl # A new version of iops->rename() was added (rename2) that takes a flag
+	dnl # argument (to support renameat2). However this separate function was
+	dnl # merged back into iops->rename() in Linux 4.9.
+	dnl #
+	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,
+		};
+	],[])
+
 	dnl #
 	dnl # 4.9 API change,
-	dnl # iops->rename2() merged into iops->rename(), and iops->rename() now wants
-	dnl # flags.
+	dnl #
+	dnl # iops->rename2() merged into iops->rename(), and iops->rename() now
+	dnl # wants flags.
 	dnl #
 	ZFS_LINUX_TEST_SRC([inode_operations_rename_flags], [
 		#include <linux/fs.h>
@@ -16,11 +36,29 @@ AC_DEFUN([ZFS_AC_KERNEL_SRC_RENAME], [
 		};
 	],[])
 
+	dnl #
+	dnl # EL7 compatibility
+	dnl #
+	dnl # EL7 has backported renameat2 support, but it's done by defining a
+	dnl # separate iops wrapper structure that takes the .renameat2 function.
+	dnl #
+	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,
+		};
+	],[])
+
 	dnl #
 	dnl # 5.12 API change,
 	dnl #
-	dnl # Linux 5.12 introduced passing struct user_namespace* as the first argument
-	dnl # of the rename() and other inode_operations members.
+	dnl # Linux 5.12 introduced passing struct user_namespace* as the first
+	dnl # argument of the rename() and other inode_operations members.
 	dnl #
 	ZFS_LINUX_TEST_SRC([inode_operations_rename_userns], [
 		#include <linux/fs.h>
@@ -44,13 +82,30 @@ AC_DEFUN([ZFS_AC_KERNEL_RENAME], [
 	],[
 		AC_MSG_RESULT(no)
 
-		AC_MSG_CHECKING([whether iop->rename() wants flags])
-		ZFS_LINUX_TEST_RESULT([inode_operations_rename_flags], [
+		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)
+				])
+			])
 		])
 	])
 ])

include/os/freebsd/zfs/sys/zfs_vnops_os.h

@@ -41,7 +41,8 @@ extern int zfs_rmdir(znode_t *dzp, const char *name, znode_t *cwd,
 extern int zfs_setattr(znode_t *zp, vattr_t *vap, int flag, cred_t *cr,
     zuserns_t *mnt_ns);
 extern int zfs_rename(znode_t *sdzp, const char *snm, znode_t *tdzp,
-    const char *tnm, cred_t *cr, int flags, zuserns_t *mnt_ns);
+    const char *tnm, cred_t *cr, int flags, uint64_t rflags, vattr_t *wo_vap,
+    zuserns_t *mnt_ns);
 extern int zfs_symlink(znode_t *dzp, const char *name, vattr_t *vap,
     const char *link, znode_t **zpp, cred_t *cr, int flags, zuserns_t *mnt_ns);
 extern int zfs_link(znode_t *tdzp, znode_t *sp,

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

@@ -324,6 +324,19 @@ static inline void zfs_gid_write(struct inode *ip, gid_t gid)
 	ip->i_gid = make_kgid(kcred->user_ns, gid);
 }
 
+/*
+ * 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/spl/sys/sysmacros.h

@@ -120,6 +120,16 @@ extern uint32_t zone_get_hostid(void *zone);
 extern void spl_setup(void);
 extern void spl_cleanup(void);
 
+/*
+ * Only handles the first 4096 majors and first 256 minors. We don't have a
+ * libc for the kernel module so we define this inline.
+ */
+static inline dev_t
+makedev(unsigned int major, unsigned int minor)
+{
+	return ((major & 0xFFF) << 8) | (minor & 0xFF);
+}
+
 #define	highbit(x)		__fls(x)
 #define	lowbit(x)		__ffs(x)
 

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

@@ -61,7 +61,8 @@ extern int zfs_getattr_fast(struct user_namespace *, struct inode *ip,
 extern int zfs_setattr(znode_t *zp, vattr_t *vap, int flag, cred_t *cr,
     zuserns_t *mnt_ns);
 extern int zfs_rename(znode_t *sdzp, char *snm, znode_t *tdzp,
-    char *tnm, cred_t *cr, int flags, zuserns_t *mnt_ns);
+    char *tnm, cred_t *cr, int flags, uint64_t rflags, vattr_t *wo_vap,
+    zuserns_t *mnt_ns);
 extern int zfs_symlink(znode_t *dzp, char *name, vattr_t *vap,
     char *link, znode_t **zpp, cred_t *cr, int flags, zuserns_t *mnt_ns);
 extern int zfs_readlink(struct inode *ip, zfs_uio_t *uio, cred_t *cr);

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

@@ -42,7 +42,11 @@ extern void zpl_vap_init(vattr_t *vap, struct inode *dir,
     umode_t mode, cred_t *cr, zuserns_t *mnt_ns);
 
 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;
 

include/sys/zfs_znode.h

@@ -299,6 +299,12 @@ extern void zfs_log_symlink(zilog_t *zilog, dmu_tx_t *tx, uint64_t txtype,
 extern void zfs_log_rename(zilog_t *zilog, dmu_tx_t *tx, uint64_t txtype,
     znode_t *sdzp, const char *sname, znode_t *tdzp, const char *dname,
     znode_t *szp);
+extern void zfs_log_rename_exchange(zilog_t *zilog, dmu_tx_t *tx,
+    uint64_t txtype, znode_t *sdzp, const char *sname, znode_t *tdzp,
+    const char *dname, znode_t *szp);
+extern void zfs_log_rename_whiteout(zilog_t *zilog, dmu_tx_t *tx,
+    uint64_t txtype, znode_t *sdzp, const char *sname, znode_t *tdzp,
+    const char *dname, znode_t *szp, znode_t *wzp);
 extern void zfs_log_write(zilog_t *zilog, dmu_tx_t *tx, int txtype,
     znode_t *zp, offset_t off, ssize_t len, int ioflag,
     zil_callback_t callback, void *callback_data);

include/sys/zil.h

@@ -164,7 +164,9 @@ typedef enum zil_create {
 #define	TX_MKDIR_ACL_ATTR	19	/* mkdir with ACL + attrs */
 #define	TX_WRITE2		20	/* dmu_sync EALREADY write */
 #define	TX_SETSAXATTR		21	/* Set sa xattrs on file */
-#define	TX_MAX_TYPE		22	/* Max transaction type */
+#define	TX_RENAME_EXCHANGE	22	/* Atomic swap via renameat2 */
+#define	TX_RENAME_WHITEOUT	23	/* Atomic whiteout via renameat2 */
+#define	TX_MAX_TYPE		24	/* Max transaction type */
 
 /*
  * The transactions for mkdir, symlink, remove, rmdir, link, and rename
@@ -317,6 +319,19 @@ typedef struct {
 	/* 2 strings: names of source and destination follow this */
 } lr_rename_t;
 
+typedef struct {
+	lr_rename_t	lr_rename;	/* common rename portion */
+	/* members related to the whiteout file (based on lr_create_t) */
+	uint64_t	lr_wfoid;	/* obj id of the new whiteout file */
+	uint64_t	lr_wmode;	/* mode of object */
+	uint64_t	lr_wuid;	/* uid of whiteout */
+	uint64_t	lr_wgid;	/* gid of whiteout */
+	uint64_t	lr_wgen;	/* generation (txg of creation) */
+	uint64_t	lr_wcrtime[2];	/* creation time */
+	uint64_t	lr_wrdev;	/* always makedev(0, 0) */
+	/* 2 strings: names of source and destination follow this */
+} lr_rename_whiteout_t;
+
 typedef struct {
 	lr_t		lr_common;	/* common portion of log record */
 	uint64_t	lr_foid;	/* file object to write */

module/os/freebsd/zfs/zfs_vnops_os.c

@@ -3420,14 +3420,17 @@ zfs_do_rename_impl(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp,
 
 int
 zfs_rename(znode_t *sdzp, const char *sname, znode_t *tdzp, const char *tname,
-    cred_t *cr, int flags, zuserns_t *mnt_ns)
+    cred_t *cr, int flags, uint64_t rflags, vattr_t *wo_vap, zuserns_t *mnt_ns)
 {
 	struct componentname scn, tcn;
 	vnode_t *sdvp, *tdvp;
 	vnode_t *svp, *tvp;
 	int error;
 	svp = tvp = NULL;
 
+	if (rflags != 0 || wo_vap != NULL)
+		return (SET_ERROR(EINVAL));
+
 	sdvp = ZTOV(sdzp);
 	tdvp = ZTOV(tdzp);
 	error = zfs_lookup_internal(sdzp, sname, &svp, &scn, DELETE);

module/os/linux/zfs/zfs_dir.c

@@ -1035,7 +1035,8 @@ zfs_link_destroy(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag,
 		}
 
 		/* The only error is !zfs_dirempty() and we checked earlier. */
-		ASSERT3U(zfs_drop_nlink_locked(zp, tx, &unlinked), ==, 0);
+		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);