mount.cifs: Use systemd's mechanism for getting password, if present

If the mount requests are "from" cifstab, then just asking for "Password:" would be unclear, this asks it as:
"Password for user@.. :"

I'm not subscribed to the mailing list.

---
mount.cifs: Use systemd's mechanism for getting password, if present.

If systemd is running, then use /bin/systemd-ask-password to get the password instead of get_pass(..) .

Reference: bug: https://bugzilla.novell.com/show_bug.cgi?id=767894

mount.cifs.c +19 -0

index 6f3f382..d721de6 100644
@@ -1687,6 +1687,58 @@ drop_child_privs(void)
 	return 0;
 }
 
+/*
+ * If systemd is present, then try to get password via
+ * /bin/systemd-ask-password, else just use getpass(..)
+ */
+static char*
+get_password(const char *prompt, char *input, int capacity)
+{
+	int is_systemd_running;
+	struct stat a, b;
+
+	/* We simply test whether the systemd cgroup hierarchy is
+	 * mounted */
+	is_systemd_running = (lstat("/sys/fs/cgroup", &a) == 0)
+		&& (lstat("/sys/fs/cgroup/systemd", &b) == 0)
+		&& (a.st_dev != b.st_dev);
+
+	if (is_systemd_running) {
+		/* systemd */
+		char *cmd;

mount.cifs.c +34 -5

+		if (fgets(input, capacity, fp)) {
+			int len = strlen(input);
+			if (input[len - 1] == '\n')
+				input[len - 1] = '\0';
+		}
+
+		fclose(fp);
+	} else {
+		/* getpass is obsolete, but there's apparently nothing that replaces it */
+		char *tmp_pass = getpass(prompt);
+		if (!tmp_pass)
+			return NULL;
+
+		strncpy(input, tmp_pass, capacity - 1);
+		input[capacity - 1] = '\0';
+
+		/* zero-out the static buffer */
+		memset(tmp_pass, 0, strlen(tmp_pass));
+	}
+
+	return input;
+}
+
 static int
 assemble_mountinfo(struct parsed_mount_info *parsed_info,
 		   const char *thisprogram, const char *mountpoint,

@@ -1768,14 +1820,20 @@ assemble_mountinfo(struct parsed_mount_info *parsed_info,
 	}
 
 	if (!parsed_info->got_password) {
-		/* getpass is obsolete, but there's apparently nothing that replaces it */
-		char *tmp_pass = getpass("Password: ");
-		if (!tmp_pass) {
+		char tmp_pass[MOUNT_PASSWD_SIZE + 1];
+		char *prompt = NULL;
+
+		if(asprintf(&prompt, "Password for %s@%s: ", parsed_info->username, orig_dev) < 0)
+			prompt = NULL;
+
+		if (get_password(prompt ? prompt : "Password: ", tmp_pass, MOUNT_PASSWD_SIZE + 1)) {
+			rc = set_password(parsed_info, tmp_pass);
+		} else {
 			fprintf(stderr, "Error reading password, exiting\n");
 			rc = EX_SYSERR;
-			goto assemble_exit;
 		}
-		rc = set_password(parsed_info, tmp_pass);
+
+		free(prompt);
 		if (rc)
 			goto assemble_exit;
 	}

Quoted 47 lines · On Fri, 22 Jun 2012 15:04:23 +0530 Ankit Jain…

If the mount requests are "from" cifstab, then just asking for "Password:" would be unclear, this asks it as:
"Password for user@.. :"

I'm not subscribed to the mailing list.

---
mount.cifs: Use systemd's mechanism for getting password, if present.

If systemd is running, then use /bin/systemd-ask-password to get the password instead of get_pass(..) .

Reference: bug: https://bugzilla.novell.com/show_bug.cgi?id=767894

mount.cifs.c +25 -0

index 6f3f382..d721de6 100644
@@ -1687,6 +1687,58 @@ drop_child_privs(void)
 	return 0;
 }
 
+/*
+ * If systemd is present, then try to get password via
+ * /bin/systemd-ask-password, else just use getpass(..)
+ */
+static char*
+get_password(const char *prompt, char *input, int capacity)
+{
+	int is_systemd_running;
+	struct stat a, b;
+
+	/* We simply test whether the systemd cgroup hierarchy is
+	 * mounted */
+	is_systemd_running = (lstat("/sys/fs/cgroup", &a) == 0)
+		&& (lstat("/sys/fs/cgroup/systemd", &b) == 0)
+		&& (a.st_dev != b.st_dev);
+
+	if (is_systemd_running) {
+		/* systemd */
+		char *cmd;
+		FILE *fp = NULL;
+
+		if (asprintf(&cmd, "/bin/systemd-ask-password \"%s\"", prompt) >= 0) {
+			fp = popen (cmd, "re");
+			free (cmd);
+		}

mount.cifs.c drop_child_privs(void) · line 1718

Quoted 4 lines

mount.cifs.c +4 -0

+
+		if (!fp)
+			return NULL;
+

What if systemd is running but we can't call /bin/systemd-ask-password for some reason? Like, maybe it doesn't exist? Should this then fall back to trying to get the password the old-fashioned way?

Hmmm...the manpage for this command also says:

The purpose of this tool is to query system-wide passwords -- that is passwords not attached to a specific user account. Examples include: unlocking encrypted hard disks when they are plugged in or at boot, entering an SSL certificate passphrase for web and VPN servers.

...does this really match that use-case? Hypothetically...

Suppose a user mount is set up in /etc/fstab and then the user calls:

    $ mount /mnt/cifs

...or something along those lines. It used to be that he'd get a password prompr on his terminal if one wasn't in the fstab. Will that still be the case here? Or will this "broadcast" some sort of password request all over the machine?

Quoted 54 lines

mount.cifs.c +34 -5

+		if (fgets(input, capacity, fp)) {
+			int len = strlen(input);
+			if (input[len - 1] == '\n')
+				input[len - 1] = '\0';
+		}
+
+		fclose(fp);
+	} else {
+		/* getpass is obsolete, but there's apparently nothing that replaces it */
+		char *tmp_pass = getpass(prompt);
+		if (!tmp_pass)
+			return NULL;
+
+		strncpy(input, tmp_pass, capacity - 1);
+		input[capacity - 1] = '\0';
+
+		/* zero-out the static buffer */
+		memset(tmp_pass, 0, strlen(tmp_pass));
+	}
+
+	return input;
+}
+
 static int
 assemble_mountinfo(struct parsed_mount_info *parsed_info,
 		   const char *thisprogram, const char *mountpoint,

@@ -1768,14 +1820,20 @@ assemble_mountinfo(struct parsed_mount_info *parsed_info,
 	}
 
 	if (!parsed_info->got_password) {
-		/* getpass is obsolete, but there's apparently nothing that replaces it */
-		char *tmp_pass = getpass("Password: ");
-		if (!tmp_pass) {
+		char tmp_pass[MOUNT_PASSWD_SIZE + 1];
+		char *prompt = NULL;
+
+		if(asprintf(&prompt, "Password for %s@%s: ", parsed_info->username, orig_dev) < 0)
+			prompt = NULL;
+
+		if (get_password(prompt ? prompt : "Password: ", tmp_pass, MOUNT_PASSWD_SIZE + 1)) {
+			rc = set_password(parsed_info, tmp_pass);
+		} else {
 			fprintf(stderr, "Error reading password, exiting\n");
 			rc = EX_SYSERR;
-			goto assemble_exit;
 		}
-		rc = set_password(parsed_info, tmp_pass);
+
+		free(prompt);
 		if (rc)
 			goto assemble_exit;
 	}

Quoted 50 lines · Jeff Layton

Quoted 47 lines · On Fri, 22 Jun 2012 15:04:23 +0530 Ankit Jain…

If the mount requests are "from" cifstab, then just asking for "Password:" would be unclear, this asks it as:
"Password for user@.. :"

I'm not subscribed to the mailing list.

---
mount.cifs: Use systemd's mechanism for getting password, if present.

If systemd is running, then use /bin/systemd-ask-password to get the password instead of get_pass(..) .

Reference: bug: https://bugzilla.novell.com/show_bug.cgi?id=767894

mount.cifs.c +25 -0

index 6f3f382..d721de6 100644
@@ -1687,6 +1687,58 @@ drop_child_privs(void)
 	return 0;
 }
 
+/*
+ * If systemd is present, then try to get password via
+ * /bin/systemd-ask-password, else just use getpass(..)
+ */
+static char*
+get_password(const char *prompt, char *input, int capacity)
+{
+	int is_systemd_running;
+	struct stat a, b;
+
+	/* We simply test whether the systemd cgroup hierarchy is
+	 * mounted */
+	is_systemd_running = (lstat("/sys/fs/cgroup", &a) == 0)
+		&& (lstat("/sys/fs/cgroup/systemd", &b) == 0)
+		&& (a.st_dev != b.st_dev);
+
+	if (is_systemd_running) {
+		/* systemd */
+		char *cmd;
+		FILE *fp = NULL;
+
+		if (asprintf(&cmd, "/bin/systemd-ask-password \"%s\"", prompt) >= 0) {
+			fp = popen (cmd, "re");
+			free (cmd);
+		}

Quoted 8 lines

mount.cifs.c drop_child_privs(void) · line 1718

Quoted 4 lines

mount.cifs.c +4 -0

+
+		if (!fp)
+			return NULL;
+

What if systemd is running but we can't call /bin/systemd-ask-password for some reason? Like, maybe it doesn't exist? Should this then fall back to trying to get the password the old-fashioned way?

My first draft of this patch, in fact, did a fallback to get_pass(..) in case that binary wasn't present. But AFAIU, if systemd is present and we just use get_pass(..), user won't get that prompt and can't really interactively give the password. This is the current problem infact. Also, /bin/systemd-ask-password seems to be part of the core systemd package (on opensuse 12.1 atleast).

Quoted 11 lines

Hmmm...the manpage for this command also says:

The purpose of this tool is to query system-wide passwords -- that is passwords not attached to a specific user account. Examples include: unlocking encrypted hard disks when they are plugged in or at boot, entering an SSL certificate passphrase for web and VPN servers.

...does this really match that use-case? Hypothetically...

Quoted 8 lines

Suppose a user mount is set up in /etc/fstab and then the user calls:

    $ mount /mnt/cifs

...or something along those lines. It used to be that he'd get a password prompr on his terminal if one wasn't in the fstab. Will that still be the case here? Or will this "broadcast" some sort of password request all over the machine?

Yeah, mount.cifs will use /bin/systemd-ask-password, which would basically cause all the systemd password agents to try and get the password. One of those is asking for password on that terminal. And one of the agents does a wall(1) for the request. It would depend on what that distro installs, i guess.

So, one of the agents *will* indeed ask for password on the same terminal, so that should mimic (almost) the older behavior. Incase we don't have this fix, or systemd-ask-password is not present, then the user will have to specify the password in cifstab or in mount options.

I have tested only on openSUSE 12.1 btw.

Thanks,
-Ankit

Quoted 56 lines

Quoted 54 lines

mount.cifs.c +34 -5

+		if (fgets(input, capacity, fp)) {
+			int len = strlen(input);
+			if (input[len - 1] == '\n')
+				input[len - 1] = '\0';
+		}
+
+		fclose(fp);
+	} else {
+		/* getpass is obsolete, but there's apparently nothing that replaces it */
+		char *tmp_pass = getpass(prompt);
+		if (!tmp_pass)
+			return NULL;
+
+		strncpy(input, tmp_pass, capacity - 1);
+		input[capacity - 1] = '\0';
+
+		/* zero-out the static buffer */
+		memset(tmp_pass, 0, strlen(tmp_pass));
+	}
+
+	return input;
+}
+
 static int
 assemble_mountinfo(struct parsed_mount_info *parsed_info,
 		   const char *thisprogram, const char *mountpoint,

@@ -1768,14 +1820,20 @@ assemble_mountinfo(struct parsed_mount_info *parsed_info,
 	}
 
 	if (!parsed_info->got_password) {
-		/* getpass is obsolete, but there's apparently nothing that replaces it */
-		char *tmp_pass = getpass("Password: ");
-		if (!tmp_pass) {
+		char tmp_pass[MOUNT_PASSWD_SIZE + 1];
+		char *prompt = NULL;
+
+		if(asprintf(&prompt, "Password for %s@%s: ", parsed_info->username, orig_dev) < 0)
+			prompt = NULL;
+
+		if (get_password(prompt ? prompt : "Password: ", tmp_pass, MOUNT_PASSWD_SIZE + 1)) {
+			rc = set_password(parsed_info, tmp_pass);
+		} else {
 			fprintf(stderr, "Error reading password, exiting\n");
 			rc = EX_SYSERR;
-			goto assemble_exit;
 		}
-		rc = set_password(parsed_info, tmp_pass);
+
+		free(prompt);
 		if (rc)
 			goto assemble_exit;
 	}

It does, but we do have to concern ourselves with older versions of systemd that might not, and with distros that might use systemd but not add the tool (consider embedded distros). I think we'll need a mechanism to fall back to the legacy password mechanism.

Also, a way to disable this at compile-time would nice. Maybe a --enable-systemd autoconf option would be good that defaults to "on" with a simple test to see if the build machine is running systemd?

Hi Jeff,

On 06/26/2012 04:58 PM, Jeff Layton wrote:
[snip]

Quoted 4 lines

It does, but we do have to concern ourselves with older versions of systemd that might not, and with distros that might use systemd but not add the tool (consider embedded distros). I think we'll need a mechanism to fall back to the legacy password mechanism.

I'm guessing that by "legacy" you mean get_pass(..)? But on a regular systemd install, it takes over stdin/stdout, and the prompt won't come up on a terminal. But I don't know how it is set up on the other scenarios you mentioned. So, I'll do that but you will have to test it ;) I tried using /dev/tty, but in/out are redirected AFAIU, so, this didn't help.

Quoted 3 lines

Also, a way to disable this at compile-time would nice. Maybe a --enable-systemd autoconf option would be good that defaults to "on" with a simple test to see if the build machine is running systemd?

Sure, autoconf option sounds good. But do we want to check for systemd running on the *build* machine (and choose default value for --enable-systemd accordingly)? Systemd running or not sounds like a runtime behavior, and we fall back to getpass anyway. I haven't added this in the attached patch, but if you feel that this makes sense, then I can add it.

Hi Jeff,

On 06/26/2012 04:58 PM, Jeff Layton wrote:
[snip]

Quoted 4 lines

It does, but we do have to concern ourselves with older versions of systemd that might not, and with distros that might use systemd but not add the tool (consider embedded distros). I think we'll need a mechanism to fall back to the legacy password mechanism.

I'm guessing that by "legacy" you mean get_pass(..)? But on a regular systemd install, it takes over stdin/stdout, and the prompt won't come up on a terminal. But I don't know how it is set up on the other scenarios you mentioned. So, I'll do that but you will have to test it ;) I tried using /dev/tty, but in/out are redirected AFAIU, so, this didn't help.

Quoted 3 lines

Also, a way to disable this at compile-time would nice. Maybe a --enable-systemd autoconf option would be good that defaults to "on" with a simple test to see if the build machine is running systemd?

Sure, autoconf option sounds good. But do we want to check for systemd running on the *build* machine (and choose default value for --enable-systemd accordingly)? Systemd running or not sounds like a runtime behavior, and we fall back to getpass anyway. I haven't added this in the attached patch, but if you feel that this makes sense, then I can add it.

Regards,