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.

Reviewed-by: Ryan Moeller <[email protected]>
Reviewed-by: Alexander Motin <[email protected]>
Reviewed-by: Brian Behlendorf <[email protected]>
Reviewed-by: Pavel Snajdr <[email protected]>
Signed-off-by: Aleksa Sarai <[email protected]>
Closes #12209
Closes #14070

7 files changed, 53 insertions, 3 deletions

diff --git a/include/os/freebsd/zfs/sys/zfs_vnops_os.h b/include/os/freebsd/zfs/sys/zfs_vnops_os.h
index 460aecd2e..839ee629a 100644
--- a/include/os/freebsd/zfs/sys/zfs_vnops_os.h
+++ b/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,
diff --git a/include/os/linux/kernel/linux/vfs_compat.h b/include/os/linux/kernel/linux/vfs_compat.h
index eeed0a388..fd0b9e8e1 100644
--- a/include/os/linux/kernel/linux/vfs_compat.h
+++ b/include/os/linux/kernel/linux/vfs_compat.h
@@ -325,6 +325,19 @@ static inline void zfs_gid_write(struct inode *ip, gid_t 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
  */
 #if !(defined(HAVE_SETATTR_PREPARE_NO_USERNS) || \
diff --git a/include/os/linux/spl/sys/sysmacros.h b/include/os/linux/spl/sys/sysmacros.h
index be1f77e43..99e3a6fb4 100644
--- a/include/os/linux/spl/sys/sysmacros.h
+++ b/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)
 
diff --git a/include/os/linux/zfs/sys/zfs_vnops_os.h b/include/os/linux/zfs/sys/zfs_vnops_os.h
index 787d258e1..197ea9bec 100644
--- a/include/os/linux/zfs/sys/zfs_vnops_os.h
+++ b/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);
diff --git a/include/os/linux/zfs/sys/zpl.h b/include/os/linux/zfs/sys/zpl.h
index 83416d647..c3ee0ae4a 100644
--- a/include/os/linux/zfs/sys/zpl.h
+++ b/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;
 
diff --git a/include/sys/zfs_znode.h b/include/sys/zfs_znode.h
index c8656b3f6..88d642350 100644
--- a/include/sys/zfs_znode.h
+++ b/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);
diff --git a/include/sys/zil.h b/include/sys/zil.h
index cec04f120..9591fb4f6 100644
--- a/include/sys/zil.h
+++ b/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
@@ -318,6 +320,19 @@ typedef struct {
 } 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 */
 	uint64_t	lr_offset;	/* offset to write to */