linux: renameat2 flags support

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>
@@ -236,6 +237,7 @@ 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>

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
@@ -329,6 +337,8 @@ 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 @@ 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

@@ -40,7 +40,7 @@ extern int zfs_rmdir(znode_t *dzp, const char *name, znode_t *cwd,
     cred_t *cr, int flags);
 extern int zfs_setattr(znode_t *zp, vattr_t *vap, int flag, cred_t *cr);
 extern int zfs_rename(znode_t *sdzp, const char *snm, znode_t *tdzp,
-    const char *tnm, cred_t *cr, int flags);
+    const char *tnm, cred_t *cr, int flags, uint64_t rflags, vattr_t *wo_vap);
 extern int zfs_symlink(znode_t *dzp, const char *name, vattr_t *vap,
     const char *link, znode_t **zpp, cred_t *cr, int flags);
 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/debug.h

@@ -127,6 +127,16 @@ spl_assert(const char *buf, const char *file, const char *func, int line)
 		    (long long) (_verify3_right));			\
 	} while (0)
 
+#define	VERIFY_IMPLY(A, B) \
+	((void)(likely((!(A)) || (B)) ||				\
+	    spl_assert("(" #A ") implies (" #B ")",			\
+	    __FILE__, __FUNCTION__, __LINE__)))
+
+#define	VERIFY_EQUIV(A, B) \
+	((void)(likely(!!(A) == !!(B)) || 				\
+	    spl_assert("(" #A ") is equivalent to (" #B ")",		\
+	    __FILE__, __FUNCTION__, __LINE__)))
+
 /*
  * Debugging disabled (--disable-debug)
  */
@@ -158,14 +168,8 @@ spl_assert(const char *buf, const char *file, const char *func, int line)
 #define	ASSERT3P	VERIFY3P
 #define	ASSERT0		VERIFY0
 #define	ASSERT		VERIFY
-#define	IMPLY(A, B) \
-	((void)(likely((!(A)) || (B)) ||				\
-	    spl_assert("(" #A ") implies (" #B ")",			\
-	    __FILE__, __FUNCTION__, __LINE__)))
-#define	EQUIV(A, B) \
-	((void)(likely(!!(A) == !!(B)) || 				\
-	    spl_assert("(" #A ") is equivalent to (" #B ")",		\
-	    __FILE__, __FUNCTION__, __LINE__)))
+#define	IMPLY		VERIFY_IMPLY
+#define	EQUIV		VERIFY_EQUIV
 
 #endif /* NDEBUG */
 

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_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/zfs_vnops_os.h

@@ -58,7 +58,7 @@ extern int zfs_getattr_fast(struct user_namespace *, struct inode *ip,
 	struct kstat *sp);
 extern int zfs_setattr(znode_t *zp, vattr_t *vap, int flag, cred_t *cr);
 extern int zfs_rename(znode_t *sdzp, char *snm, znode_t *tdzp,
-    char *tnm, cred_t *cr, int flags);
+    char *tnm, cred_t *cr, int flags, uint64_t rflags, vattr_t *wo_vap);
 extern int zfs_symlink(znode_t *dzp, char *name, vattr_t *vap,
     char *link, znode_t **zpp, cred_t *cr, int flags);
 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);
 
 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;
 extern dentry_operations_t zpl_dentry_operations;

include/sys/zfs_znode.h

@@ -300,6 +300,12 @@ 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 */

include/zfeature_common.h

@@ -78,6 +78,8 @@ typedef enum spa_feature {
 	SPA_FEATURE_ZILSAXATTR,
 	SPA_FEATURE_HEAD_ERRLOG,
 	SPA_FEATURE_BLAKE3,
+	SPA_FEATURE_RENAME_EXCHANGE,
+	SPA_FEATURE_RENAME_WHITEOUT,
 	SPA_FEATURES
 } spa_feature_t;