linux: overlayfs support

Signed-off-by: Ryan Moeller <ryan@iXsystems.com>

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;
+	}
 }
 
 /* ARGSUSED */
@@ -305,6 +313,9 @@ static zil_rec_info_t zil_rec_info[TX_MAX_TYPE] = {
 	{.zri_print = zil_prt_rec_create,   .zri_name = "TX_MKDIR_ATTR      "},
 	{.zri_print = zil_prt_rec_create,   .zri_name = "TX_MKDIR_ACL_ATTR  "},
 	{.zri_print = zil_prt_rec_write,    .zri_name = "TX_WRITE2          "},
+	{.zri_print = NULL,		    .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 "},
 };
 
 /* ARGSUSED */
@@ -329,7 +340,14 @@ print_log_record(zilog_t *zilog, const lr_t *lr, void *arg, uint64_t claim_txg)
 
 	if (txtype && verbose >= 3) {
 		if (!zilog->zl_os->os_encrypted) {
-			zil_rec_info[txtype].zri_print(zilog, txtype, lr);
+			zil_prt_rec_func_t print_func =
+			    zil_rec_info[txtype].zri_print;
+			if (print_func != NULL) {
+				print_func(zilog, txtype, lr);
+			} else {
+				(void) printf("%s(%s)\n", tab_prefix,
+				    zil_rec_info[txtype].zri_name);
+			}
 		} else {
 			(void) printf("%s(encrypted)\n", tab_prefix);
 		}

cmd/ztest/ztest.c

@@ -2383,6 +2383,9 @@ zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
 	NULL,			/* TX_MKDIR_ATTR */
 	NULL,			/* TX_MKDIR_ACL_ATTR */
 	NULL,			/* TX_WRITE2 */
+	NULL,			/* TX_SETSAXATTR */
+	NULL,			/* TX_RENAME_EXCHANGE */
+	NULL,			/* TX_RENAME_WHITEOUT */
 };
 
 /*

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

@@ -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)
+				])
+			])
 		])
 	])
 ])

config/kernel.m4

@@ -94,6 +94,7 @@ 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
@@ -212,6 +213,7 @@ 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

configure.ac

@@ -234,6 +234,7 @@ AC_CONFIG_FILES([
 	tests/zfs-tests/cmd/readmmap/Makefile
 	tests/zfs-tests/cmd/read_dos_attributes/Makefile
 	tests/zfs-tests/cmd/rename_dir/Makefile
+	tests/zfs-tests/cmd/renameat2/Makefile
 	tests/zfs-tests/cmd/rm_lnkcnt_zero_file/Makefile
 	tests/zfs-tests/cmd/send_doall/Makefile
 	tests/zfs-tests/cmd/stride_dd/Makefile

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/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/spl/sys/debug.h

@@ -128,6 +128,16 @@ void spl_dumpstack(void);
 	typedef char __attribute__ ((unused))				\
 	__compile_time_assertion__ ## y[(x) ? 1 : -1]
 
+#define	VERIFY_IMPLY(A, B) \
+	((void)(likely((!(A)) || (B)) || \
+	    spl_panic(__FILE__, __FUNCTION__, __LINE__, \
+	    "(" #A ") implies (" #B ")")))
+
+#define	VERIFY_EQUIV(A, B) \
+	((void)(likely(!!(A) == !!(B)) || \
+	    spl_panic(__FILE__, __FUNCTION__, __LINE__, \
+	    "(" #A ") is equivalent to (" #B ")")))
+
 /*
  * Debugging disabled (--disable-debug)
  */
@@ -153,15 +163,8 @@ void spl_dumpstack(void);
 #define	ASSERT3P	VERIFY3P
 #define	ASSERT0		VERIFY0
 #define	ASSERT		VERIFY
-#define	IMPLY(A, B) \
-	((void)(likely((!(A)) || (B)) || \
-	    spl_panic(__FILE__, __FUNCTION__, __LINE__, \
-	    "(" #A ") implies (" #B ")")))
-#define	EQUIV(A, B) \
-	((void)(likely(!!(A) == !!(B)) || \
-	    spl_panic(__FILE__, __FUNCTION__, __LINE__, \
-	    "(" #A ") is equivalent to (" #B ")")))
-/* END CSTYLED */
+#define	IMPLY		VERIFY_IMPLY
+#define	EQUIV		VERIFY_EQUIV
 
 #endif /* NDEBUG */
 

include/os/linux/spl/sys/sysmacros.h

@@ -121,6 +121,16 @@ 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

@@ -42,10 +42,13 @@ 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;
 extern const struct address_space_operations zpl_address_space_operations;
 extern const struct file_operations zpl_file_operations;
 extern const struct file_operations zpl_dir_file_operations;

include/sys/zfs_znode.h

@@ -275,6 +275,12 @@ 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);