mkfs: add mkfs config version specifier for the config file

We may opt later to bump this and add new features which we don't want a v1 parser to use. So peg a version number. We add a new [global] section dedicated for special configuration file option specifiers, with the only subtopt supported being 'version'.

The only we support for now is version 1.

Suggested-by Dave Chinner <[email protected]>

Signed-off-by Luis R. Rodriguez <[email protected]>

---

As suggested by Sandeed, added a [global] section which makes the code easier to read and maintain.

 man/man8/mkfs.xfs.8.in |  9 ++++-
 mkfs/config.c          | 91 ++++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 99 insertions(+), 1 deletion(-)

man/man8/mkfs.xfs.8.in +8 -1

index cf4bdf827d9c..c86d559ba3e1 100644
@@ -965,12 +965,19 @@ option follow a simple ini-style format as shown below.
 Available options consist of a small subset of the parameters available
 via the
 .BR mkfs.xfs (8)
-command line.
+command line. All configuration files must start with an annotation of
+the configuration file format, specified in the global section with the keyword
+.B version
+and the only supported version format is 1.
 Currently all default parameters can only be either enabled or disabled,
 with a value of 1 to enable or 0 to disable.
 See below for a list of all supported configuration parameters and their
 current built-in default settings.
 .PP
+.BI [global]
+.br
+.BI version=1
+.PP
 .BI [data]
 .br
 .BI noalign=0

mkfs/config.c +91 -0

index 835adc45f02d..4bb371d6b1e8 100644
@@ -32,6 +32,8 @@
  * We only provide definitions for what we currently support parsing.
  */
 
+static uint64_t mkfs_config_version;
+
 enum data_subopts {
 	D_NOALIGN = 0,
 };

@@ -42,6 +44,14 @@ enum inode_subopts {
 	I_SPINODES,
 };
 
+/*
+ * These are not part of the defaults but rather global generic configuration
+ * file options.
+ */
+enum global_subopts {
+	G_VERSION = 0,
+};
+
 enum log_subopts {
 	L_LAZYSBCNTR = 0,
 };

@@ -122,6 +132,24 @@ inode_config_parser(
 	return -1;
 }
 
+static int
+global_config_parser(
+	struct mkfs_default_params	*dft,
+	int				psubopt,
+	uint64_t			value)
+{
+	enum global_subopts		subopt = psubopt;
+
+	switch (subopt) {
+	case G_VERSION:
+		if (value != 1)
+			return -1;
+		mkfs_config_version = value;
+		return 0;
+	}
+	return -1;
+}
+
 static int
 log_config_parser(
 	struct mkfs_default_params	*dft,

@@ -234,6 +262,14 @@ struct confopts {
 		},
 		.parser = inode_config_parser,
 	},
+	{
+		.name = "global",
+		.subopts = {
+			[G_VERSION] = "version",
+			NULL
+		},
+		.parser = global_config_parser,
+	},
 	{
 		.name = "log",
 		.subopts = {

@@ -287,6 +323,24 @@ get_confopts(
 	return NULL;
 }
 
+static bool config_version_set(void)
+{
+	/* cannot fail */
+	struct confopts *opts = get_confopts("global");
+
+	return opts->seen;
+}
+
+static uint64_t config_version(void)
+{
+	/* cannot fail */
+	struct confopts *opts = get_confopts("global");
+
+	if (!opts->seen)
+		return 0;
+	return mkfs_config_version;
+}
+
 enum parse_line_type {
 	PARSE_COMMENT = 0,
 	PARSE_EMPTY,

@@ -379,6 +433,32 @@ parse_get_line_type(
 	return PARSE_INVALID;
 }
 
+static bool
+check_version_supported(
+		void)
+{
+	uint64_t version = config_version();
+
+	if (!version) {
+		fprintf(stderr,
+_("Version for configuration file not specified\n"));
+		goto out;
+	}
+
+	if (version != 1) {
+		errno = EOPNOTSUPP;
+		fprintf(stderr,
+_("Only version 1 supported, you specified version %lu\n"), version);
+		goto out;
+	}
+
+	return true;
+
+out:
+	errno = EOPNOTSUPP;
+	return false;
+}
+
 static int
 parse_config_stream(
 	struct mkfs_default_params	*dft,

@@ -420,6 +500,7 @@ parse_config_stream(
 					  config_file, lineno, line);
 			goto out;
 		case PARSE_SECTION:
+
 			confopt = get_confopts(tag);
 			if (!confopt) {
 				fprintf(stderr,

@@ -451,6 +532,13 @@ _("No section specified yet on line %s:%zu : %s\n"),
 				goto out_free_tag;
 			}
 
+			if (!config_version_set() &&
+			    strcmp(confopt->name, "global") != 0) {
+				fprintf(stderr,
+_("A global section with a version set was not detected and must be set first\n"));
+				goto out_free_tag;
+			}
+
 			/*
 			 * We re-use the line buffer allocated by getline(),
 			 * however line must be kept pointing to its original

@@ -486,6 +574,9 @@ _("Error parsing line %s:%zu : %s\n"),
 				goto out;
 			}
 
+			if (!check_version_supported())
+				goto out;
+
 			break;
 		}
 		free(line);

Soo....

I'm really on the fence about versioning at all, because we have yet to identify any concrete scenario in which we'd need to bump the version; we haven't even defined what exactly the "version" implies.

So unless anyone has a concrete example of how a version would be needed, functional, reliably-parsed, and well-defined I'm inclined to leave this out.

Quoted 5 lines · Eric Sandeen

Soo....

I'm really on the fence about versioning at all, because we have yet to identify any concrete scenario in which we'd need to bump the version; we haven't even defined what exactly the "version" implies.

Right.

Quoted 2 lines

So unless anyone has a concrete example of how a version would be needed, functional, reliably-parsed, and well-defined I'm inclined to leave this out.

Dave asked for it, so I would imagine he could make a better case for it.

But from a generic file format point of view, a version typically does make sense for a slew of reasons, but it is typically the not known, the unexpected, for which it can be a life saver.

Things that I can mention other than what you mentioned (whether or not they are good, this is just a list of crazy things):

1) We move from using our own parser to e2fsprogs profile parser (recall that my studies revealed it was the smallest) or lib_iniconfig (more robust, and widely used, however a kitchen sink compared to e2fsprogs profile parser)

Both however support the INI format we are using.

The version however may help to make the old parser not barf if say we added some magic in the future which would make the old parser barf.

2) Data type change for an existing token. Say crc=2 was invented (again just an example), it would barf on the old parser, but with the version check it would immediately tell you a more meaningful message.

Perhaps the biggest advantage I can think of then then is a better error message for a set of supported features, starting with what we have today reflected as parsed as version 1. This does differ from what we do with the CLI, however as with the respecification simplifcation done on config.c in contrast to the CLI respecification checks, I would see this as a win for users, and yet keeps things simple.

If this seems acceptable and reasonable, then version would be a reflection of supporting the format and layout of the file, as well was indicative of supporting only a specific feature set.

Thoughts?

Luis

Quoted 12 lines · Luis R. Rodriguez

But from a generic file format point of view, a version typically does make sense for a slew of reasons, but it is typically the not known, the unexpected, for which it can be a life saver.

Things that I can mention other than what you mentioned (whether or not they are good, this is just a list of crazy things):

1) We move from using our own parser to e2fsprogs profile parser (recall that my studies revealed it was the smallest) or lib_iniconfig (more robust, and widely used, however a kitchen sink compared to e2fsprogs profile parser)

Quoted 5 lines

Both however support the INI format we are using.

The version however may help to make the old parser not barf if say we added some magic in the future which would make the old parser barf.

... if the file format does not change, a parser change doesn't matter, and a version change would be pointless.

Quoted 4 lines

2) Data type change for an existing token. Say crc=2 was invented (again just an example), it would barf on the old parser, but with the version check it would immediately tell you a more meaningful message.

But how is that any different from what it would do on the command line for a similar change?

"Invalid value 2 for token crc=2 on line XYZ" is completely reasonable, and does not require any version information.

In fact, what you are suggesting here means we'd bump the version if and only if the config file contained "crc=2" which makes little sense.

Quoted 5 lines

message for a set of supported features, starting with what we have today reflected as parsed as version 1. This does differ from what we do with the CLI, however as with the respecification simplifcation done on config.c in contrast to the CLI respecification checks, I would see this as a win for users, and yet keeps things simple.

See my earlier email for why versioning cannot depend on the presence or absence of any parseable tokens ... stating "token XYZ not supported" should suffice IMHO.

(now, it's true that today we don't distinguish between unknown vs. unsupported tokens, I don't know if I should try to fix that before 4.17 or not; again, though,

that is on the parser side, not the config file side.)

Quoted 3 lines

If this seems acceptable and reasonable, then version would be a reflection of supporting the format and layout of the file, as well was indicative of supporting only a specific feature set.

What is or is not a supported token is wholly a function of the mkfs/parser version, not the config file version, right?

So, not to be argumentative ... but I remain unconvinced of the need for config file versioning.

Quoted 15 lines · Eric Sandeen

Quoted 14 lines · Luis R. Rodriguez

But from a generic file format point of view, a version typically does make sense for a slew of reasons, but it is typically the not known, the unexpected, for which it can be a life saver.

Things that I can mention other than what you mentioned (whether or not they are good, this is just a list of crazy things):

1) We move from using our own parser to e2fsprogs profile parser (recall that my studies revealed it was the smallest) or lib_iniconfig (more robust, and widely used, however a kitchen sink compared to e2fsprogs profile parser)

Both however support the INI format we are using.

Quoted 6 lines

Quoted 3 lines

The version however may help to make the old parser not barf if say we added some magic in the future which would make the old parser barf.

... if the file format does not change, a parser change doesn't matter, and a version change would be pointless.

It can. Consider support for different comment styles.

Quoted 5 lines · Eric Sandeen

Soo....

I'm really on the fence about versioning at all, because we have yet to identify any concrete scenario in which we'd need to bump the version; we haven't even defined what exactly the "version" implies.

Right.

Quoted 2 lines

So unless anyone has a concrete example of how a version would be needed, functional, reliably-parsed, and well-defined I'm inclined to leave this out.

Dave asked for it, so I would imagine he could make a better case for it.