Change /etc/mtab to /proc/self/mounts

Fix misleading error message:

 "The /dev/zfs device is missing and must be created.", if /etc/mtab is missing.

Reviewed-by: Richard Laager <[email protected]>
Reviewed-by: Brian Behlendorf <[email protected]>
Signed-off-by: Eric Desrochers <[email protected]>
Closes #4680 
Closes #5029

13 files changed, 53 insertions, 43 deletions

diff --git a/cmd/mount_zfs/mount_zfs.c b/cmd/mount_zfs/mount_zfs.c
index b78c91c7c..f6631a527 100644
--- a/cmd/mount_zfs/mount_zfs.c
+++ b/cmd/mount_zfs/mount_zfs.c
@@ -294,11 +294,11 @@ mtab_is_writeable(void)
 	struct stat st;
 	int error, fd;
 
-	error = lstat(MNTTAB, &st);
+	error = lstat("/etc/mtab", &st);
 	if (error || S_ISLNK(st.st_mode))
 		return (0);
 
-	fd = open(MNTTAB, O_RDWR | O_CREAT, 0644);
+	fd = open("/etc/mtab", O_RDWR | O_CREAT, 0644);
 	if (fd < 0)
 		return (0);
 
@@ -320,21 +320,21 @@ mtab_update(char *dataset, char *mntpoint, char *type, char *mntopts)
 	mnt.mnt_freq = 0;
 	mnt.mnt_passno = 0;
 
-	fp = setmntent(MNTTAB, "a+");
+	fp = setmntent("/etc/mtab", "a+");
 	if (!fp) {
 		(void) fprintf(stderr, gettext(
-		    "filesystem '%s' was mounted, but %s "
+		    "filesystem '%s' was mounted, but /etc/mtab "
 		    "could not be opened due to error %d\n"),
-		    dataset, MNTTAB, errno);
+		    dataset, errno);
 		return (MOUNT_FILEIO);
 	}
 
 	error = addmntent(fp, &mnt);
 	if (error) {
 		(void) fprintf(stderr, gettext(
-		    "filesystem '%s' was mounted, but %s "
+		    "filesystem '%s' was mounted, but /etc/mtab "
 		    "could not be updated due to error %d\n"),
-		    dataset, MNTTAB, errno);
+		    dataset, errno);
 		return (MOUNT_FILEIO);
 	}
 
diff --git a/cmd/zfs/zfs_main.c b/cmd/zfs/zfs_main.c
index dd165da0e..cca52dbe7 100644
--- a/cmd/zfs/zfs_main.c
+++ b/cmd/zfs/zfs_main.c
@@ -6146,9 +6146,10 @@ share_mount(int op, int argc, char **argv)
 		}
 
 		/*
-		 * When mount is given no arguments, go through /etc/mtab and
-		 * display any active ZFS mounts.  We hide any snapshots, since
-		 * they are controlled automatically.
+		 * When mount is given no arguments, go through
+		 * /proc/self/mounts and display any active ZFS mounts.
+		 * We hide any snapshots, since they are controlled
+		 * automatically.
 		 */
 
 		/* Reopen MNTTAB to prevent reading stale data from open file */
@@ -6228,8 +6229,8 @@ unshare_unmount_compare(const void *larg, const void *rarg, void *unused)
 
 /*
  * Convenience routine used by zfs_do_umount() and manual_unmount().  Given an
- * absolute path, find the entry /etc/mtab, verify that its a ZFS filesystem,
- * and unmount it appropriately.
+ * absolute path, find the entry /proc/self/mounts, verify that its a
+ * ZFS filesystems, and unmount it appropriately.
  */
 static int
 unshare_unmount_path(int op, char *path, int flags, boolean_t is_manual)
@@ -6242,7 +6243,7 @@ unshare_unmount_path(int op, char *path, int flags, boolean_t is_manual)
 	ino_t path_inode;
 
 	/*
-	 * Search for the path in /etc/mtab.  Rather than looking for the
+	 * Search for the path in /proc/self/mounts. Rather than looking for the
 	 * specific path, which can be fooled by non-standard paths (i.e. ".."
 	 * or "//"), we stat() the path and search for the corresponding
 	 * (major,minor) device pair.
@@ -6273,8 +6274,8 @@ unshare_unmount_path(int op, char *path, int flags, boolean_t is_manual)
 			    "currently mounted\n"), cmdname, path);
 			return (1);
 		}
-		(void) fprintf(stderr, gettext("warning: %s not in mtab\n"),
-		    path);
+		(void) fprintf(stderr, gettext("warning: %s not in"
+		    "/proc/self/mounts\n"), path);
 		if ((ret = umount2(path, flags)) != 0)
 			(void) fprintf(stderr, gettext("%s: %s\n"), path,
 			    strerror(errno));
@@ -6385,9 +6386,9 @@ unshare_unmount(int op, int argc, char **argv)
 		/*
 		 * We could make use of zfs_for_each() to walk all datasets in
 		 * the system, but this would be very inefficient, especially
-		 * since we would have to linearly search /etc/mtab for each
-		 * one.  Instead, do one pass through /etc/mtab looking for
-		 * zfs entries and call zfs_unmount() for each one.
+		 * since we would have to linearly search /proc/self/mounts for
+		 * each one. Instead, do one pass through /proc/self/mounts
+		 * looking for zfs entries and call zfs_unmount() for each one.
 		 *
 		 * Things get a little tricky if the administrator has created
 		 * mountpoints beneath other ZFS filesystems.  In this case, we
diff --git a/cmd/zinject/translate.c b/cmd/zinject/translate.c
index 5cc9d9fdc..adace72ff 100644
--- a/cmd/zinject/translate.c
+++ b/cmd/zinject/translate.c
@@ -120,7 +120,7 @@ parse_pathname(const char *inpath, char *dataset, char *relpath,
 #else
 	if ((fp = fopen(MNTTAB, "r")) == NULL) {
 #endif
-		(void) fprintf(stderr, "cannot open /etc/mtab\n");
+		(void) fprintf(stderr, "cannot open %s\n", MNTTAB);
 		return (-1);
 	}
 
diff --git a/config/user-commands.m4 b/config/user-commands.m4
index 58633b4a0..6e9c3d103 100644
--- a/config/user-commands.m4
+++ b/config/user-commands.m4
@@ -127,7 +127,7 @@ AC_DEFUN([ZFS_AC_CONFIG_USER_COMMANDS_LINUX], [
 	PAGESIZE=$($GETCONF PAGESIZE)
 	AC_SUBST(PAGESIZE)
 
-	MNTTAB=/etc/mtab
+	MNTTAB=/proc/self/mounts
 	AC_SUBST(MNTTAB)
 ])
 
diff --git a/contrib/initramfs/scripts/zfs b/contrib/initramfs/scripts/zfs
index 6a78a46d2..250dc5717 100644
--- a/contrib/initramfs/scripts/zfs
+++ b/contrib/initramfs/scripts/zfs
@@ -288,9 +288,8 @@ load_module_initrd()
 		wait_for_dev
 	fi
 
-	# zpool import refuse to import without a valid mtab
-	[ ! -f /proc/mounts ] && mount proc /proc
-	[ ! -f /etc/mtab ] && cat /proc/mounts > /etc/mtab
+	# zpool import refuse to import without a valid /proc/self/mounts
+	[ ! -f /proc/self/mounts ] && mount proc /proc
 
 	# Load the module
 	load_module "zfs" || return 1
@@ -919,7 +918,7 @@ mountroot()
 	#
 	#   but the MOUNTPOINT prefix is preserved on descendent filesystem
 	#   after the pivot into the regular root, which later breaks things
-	#   like `zfs mount -a` and the /etc/mtab refresh.
+	#   like `zfs mount -a` and the /proc/self/mounts refresh.
 	#
 	# * Mount additional filesystems required
 	#   Such as /usr, /var, /usr/local etc.
diff --git a/etc/init.d/zfs-functions.in b/etc/init.d/zfs-functions.in
index acfdf9926..97f2ea039 100644
--- a/etc/init.d/zfs-functions.in
+++ b/etc/init.d/zfs-functions.in
@@ -368,7 +368,7 @@ read_mtab()
 			# Set the variable.
 			eval export MTAB_$mntpnt=\"$fs\"
 		fi
-	done < /proc/mounts
+	done < /proc/self/mounts
 }
 
 in_mtab()
diff --git a/etc/init.d/zfs-mount.in b/etc/init.d/zfs-mount.in
index 34db057c8..2722a31e4 100755
--- a/etc/init.d/zfs-mount.in
+++ b/etc/init.d/zfs-mount.in
@@ -39,7 +39,7 @@ chkroot() {
 		if [ "$2" = "/" ]; then
 			return 0
 		fi
-	done < /etc/mtab
+	done < /proc/self/mounts
 
 	return 1
 }
@@ -178,7 +178,7 @@ do_start()
 
 	check_module_loaded "zfs" || exit 0
 
-	# Ensure / exists in /etc/mtab, if not update mtab accordingly.
+	# Ensure / exists in /proc/self/mounts.
 	# This should be handled by rc.sysinit but lets be paranoid.
 	if ! chkroot
 	then
diff --git a/lib/libspl/include/sys/mnttab.h b/lib/libspl/include/sys/mnttab.h
index 21d89658c..026a8fa7b 100644
--- a/lib/libspl/include/sys/mnttab.h
+++ b/lib/libspl/include/sys/mnttab.h
@@ -38,7 +38,7 @@
 #undef MNTTAB
 #endif /* MNTTAB */
 
-#define	MNTTAB		"/etc/mtab"
+#define	MNTTAB		"/proc/self/mounts"
 #define	MNT_LINE_MAX	4096
 
 #define	MNT_TOOLONG	1	/* entry exceeds MNT_LINE_MAX */
diff --git a/lib/libzfs/libzfs_dataset.c b/lib/libzfs/libzfs_dataset.c
index 3ac31ad1d..84fe71734 100755
--- a/lib/libzfs/libzfs_dataset.c
+++ b/lib/libzfs/libzfs_dataset.c
@@ -1908,9 +1908,9 @@ zfs_unset_recvd_props_mode(zfs_handle_t *zhp, uint64_t *cookie)
  * zfs_prop_get_int() are built using this interface.
  *
  * Certain properties can be overridden using 'mount -o'.  In this case, scan
- * the contents of the /etc/mtab entry, searching for the appropriate options.
- * If they differ from the on-disk values, report the current values and mark
- * the source "temporary".
+ * the contents of the /proc/self/mounts entry, searching for the
+ * appropriate options. If they differ from the on-disk values, report the
+ * current values and mark the source "temporary".
  */
 static int
 get_numeric_property(zfs_handle_t *zhp, zfs_prop_t prop, zprop_source_t *src,
@@ -1981,8 +1981,9 @@ get_numeric_property(zfs_handle_t *zhp, zfs_prop_t prop, zprop_source_t *src,
 
 	/*
 	 * Because looking up the mount options is potentially expensive
-	 * (iterating over all of /etc/mtab), we defer its calculation until
-	 * we're looking up a property which requires its presence.
+	 * (iterating over all of /proc/self/mounts), we defer its
+	 * calculation until we're looking up a property which requires
+	 * its presence.
 	 */
 	if (!zhp->zfs_mntcheck &&
 	    (mntopt_on != NULL || prop == ZFS_PROP_MOUNTED)) {
diff --git a/lib/libzfs/libzfs_mount.c b/lib/libzfs/libzfs_mount.c
index 8c7f32638..44781679d 100644
--- a/lib/libzfs/libzfs_mount.c
+++ b/lib/libzfs/libzfs_mount.c
@@ -347,8 +347,8 @@ zfs_add_option(zfs_handle_t *zhp, char *options, int len,
 		return (0);
 
 	/*
-	 * zfs_prop_get_int() to not used to ensure our mount options
-	 * are not influenced by the current /etc/mtab contents.
+	 * zfs_prop_get_int() is not used to ensure our mount options
+	 * are not influenced by the current /proc/self/mounts contents.
 	 */
 	value = getprop_uint64(zhp, prop, &source);
 
@@ -1184,8 +1184,8 @@ mountpoint_compare(const void *a, const void *b)
  * Unshare and unmount all datasets within the given pool.  We don't want to
  * rely on traversing the DSL to discover the filesystems within the pool,
  * because this may be expensive (if not all of them are mounted), and can fail
- * arbitrarily (on I/O error, for example).  Instead, we walk /etc/mtab and
- * gather all the filesystems that are currently mounted.
+ * arbitrarily (on I/O error, for example).  Instead, we walk /proc/self/mounts
+ * and gather all the filesystems that are currently mounted.
  */
 int
 zpool_disable_datasets(zpool_handle_t *zhp, boolean_t force)
diff --git a/lib/libzfs/libzfs_util.c b/lib/libzfs/libzfs_util.c
index 3c388a3a0..4988e8115 100755
--- a/lib/libzfs/libzfs_util.c
+++ b/lib/libzfs/libzfs_util.c
@@ -67,9 +67,9 @@ libzfs_error_init(int error)
 		    "loaded.\nTry running '/sbin/modprobe zfs' as root "
 		    "to load them.\n"));
 	case ENOENT:
-		return (dgettext(TEXT_DOMAIN, "The /dev/zfs device is "
-		    "missing and must be created.\nTry running 'udevadm "
-		    "trigger' as root to create it.\n"));
+		return (dgettext(TEXT_DOMAIN, "/dev/zfs and /proc/self/mounts "
+		    "are required.\nTry running 'udevadm trigger' and 'mount "
+		    "-t proc proc /proc' as root.\n"));
 	case ENOEXEC:
 		return (dgettext(TEXT_DOMAIN, "The ZFS modules cannot be "
 		    "auto-loaded.\nTry running '/sbin/modprobe zfs' as "
diff --git a/scripts/ziltest.sh b/scripts/ziltest.sh
index 62cf09e88..bb816293f 100755
--- a/scripts/ziltest.sh
+++ b/scripts/ziltest.sh
@@ -185,7 +185,11 @@ CHECKSUM_BEFORE=$(sha256sum -b "$PAYLOAD")
 #
 # TX_WRITE (small file with ordering)
 #
-cp /etc/mtab $ROOT/small_file
+if is_linux; then
+	cp /proc/self/mounts $ROOT/small_file
+else
+	cp /etc/mtab $ROOT/small_file
+fi
 cp /etc/profile $ROOT/small_file
 
 #
diff --git a/tests/zfs-tests/tests/functional/cli_root/zfs/zfs_003_neg.ksh b/tests/zfs-tests/tests/functional/cli_root/zfs/zfs_003_neg.ksh
index d094d1a7d..63c735f31 100755
--- a/tests/zfs-tests/tests/functional/cli_root/zfs/zfs_003_neg.ksh
+++ b/tests/zfs-tests/tests/functional/cli_root/zfs/zfs_003_neg.ksh
@@ -40,11 +40,16 @@
 
 verify_runnable "global"
 
-log_assert "zfs fails with unexpected scenarios."
+log_assert "zfs fails with unexpected scenario."
 
 #verify zfs failed if ZFS_DEV cannot be opened
 ZFS_DEV=/dev/zfs
 
+if is_linux; then
+	# On Linux, we use /proc/self/mounts, which cannot be moved.
+	MNTTAB=
+fi
+
 for file in $ZFS_DEV $MNTTAB; do
 	if [[ -e $file ]]; then
 		$MV $file ${file}.bak
@@ -55,4 +60,4 @@ for file in $ZFS_DEV $MNTTAB; do
 	$MV ${file}.bak $file
 done
 
-log_pass "zfs fails with unexpected scenarios as expected."
+log_pass "zfs fails with unexpected scenario as expected."