Skip to contentSkip to conversation

btrfs-progs: cmds/tune: add set/clear features

PATCH In review Qu Wenruo 5 participants 20 messages Sep 5, 2023 – Oct 30, 2023

Subsystems (3)
lore
Share & tools
lore
author
#13 lore raw

PATCH 7/7 btrfs-progs: tests/cli: add a test case for "btrfs tune" subcommand

In reply to Qu Wenruo

The new test case would test all supported features of both set and clear subcommands.

Also test the error handling of unknown features.

---
 tests/cli-tests/018-btrfs-tune/test.sh | 40 ++++++++++++++++++++++++++
 1 file changed, 40 insertions(+)
 create mode 100755 tests/cli-tests/018-btrfs-tune/test.sh
tests/cli-tests/018-btrfs-tune/test.sh new file +40 -0
new file mode 100755
index 000000000000..3b2d1ebb3e59
@@ -0,0 +1,40 @@
+#!/bin/bash
+# test all command line options of btrfstune
+
+source "$TEST_TOP/common" || exit
+
+check_prereq btrfstune
+
+setup_root_helper
+prepare_test_dev
+
+run_mayfail "$TOP/btrfstune" || true
+run_check "$TOP/btrfstune" --help
+
+run_mustfail "must not work on non-existent device" \
+	"$TOP/btrfstune" -r file-does-not-exist
+
+run_check_mkfs_test_dev -O ^extref
+run_check "$TOP/btrfs" tune set extref "$TEST_DEV"
+run_check "$TOP/btrfs" check "$TEST_DEV"
+
+run_check_mkfs_test_dev -O ^skinny-metadata
+run_check "$TOP/btrfs" tune set skinny-metadata "$TEST_DEV"
+run_check "$TOP/btrfs" check "$TEST_DEV"
+
+run_check_mkfs_test_dev -O ^no-holes
+run_check "$TOP/btrfs" tune set no-holes "$TEST_DEV"
+run_check "$TOP/btrfs" check "$TEST_DEV"
+
+run_check_mkfs_test_dev -O ^block-group-tree
+run_check "$TOP/btrfs" tune set block-group-tree "$TEST_DEV"
+run_check "$TOP/btrfs" check "$TEST_DEV"
+run_check "$TOP/btrfs" tune clear block-group-tree "$TEST_DEV"
+run_check "$TOP/btrfs" check "$TEST_DEV"
+
+run_check_mkfs_test_dev
+run_check "$TOP/btrfs" tune set -f seed "$TEST_DEV"
+run_check "$TOP/btrfs" tune clear seed "$TEST_DEV"
+
+run_mustfail "Unknown features" "$TOP/btrfs" tune set unknown-feature "$TEST_DEV"
+run_mustfail "Unknown features" "$TOP/btrfs" tune clear unknown-feature "$TEST_DEV"
Signature · 1 line
2.42.0
author
#9 lore raw

PATCH 3/7 btrfs-progs: cmds/tune: add set support for free-space-tree feature

In reply to Qu Wenruo

This patch allows "btrfs tune set" to enable free-space-tree feature, using pretty much the same code from btrfstune.

---
 cmds/tune.c | 52 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 52 insertions(+)
cmds/tune.c +52 -0
index 92c7b9f1502c..9906cde393f4 100644
@@ -17,11 +17,13 @@
 #include <unistd.h>
 #include "kerncompat.h"
 #include "cmds/commands.h"
+#include "check/clear-cache.h"
 #include "common/help.h"
 #include "common/fsfeatures.h"
 #include "kernel-shared/messages.h"
 #include "kernel-shared/disk-io.h"
 #include "kernel-shared/transaction.h"
+#include "kernel-shared/free-space-tree.h"
 
 static const char * const cmd_tune_set_usage[] = {
 	"btrfs tune set <feature> [<device>]",
@@ -57,6 +59,15 @@ static const struct btrfs_feature set_features[] = {
 		VERSION_TO_STRING2(safe, 4,0),
 		VERSION_TO_STRING2(default, 5,15),
 		.desc		= "no explicit hole extents for files"
+	}, {
+		.name		= "free-space-tree",
+		.compat_ro_flag	= BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE |
+				  BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE_VALID,
+		.sysfs_name = "free_space_tree",
+		VERSION_TO_STRING2(compat, 4,5),
+		VERSION_TO_STRING2(safe, 4,9),
+		VERSION_TO_STRING2(default, 5,15),
+		.desc		= "free space tree (space_cache=v2)"
 	},
 	/* Keep this one last */
 	{
@@ -134,6 +145,36 @@ static int set_super_incompat_flags(struct btrfs_fs_info *fs_info, u64 flags)
 	return ret;
 }
 
+static int convert_to_fst(struct btrfs_fs_info *fs_info)
+{
+	int ret;
+
+	/* We may have invalid old v2 cache, clear them first. */
+	if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
+		ret = btrfs_clear_free_space_tree(fs_info);
+		if (ret < 0) {
+			errno = -ret;
+			error("failed to clear stale v2 free space cache: %m");
+			return ret;
+		}
+	}
+	ret = btrfs_clear_v1_cache(fs_info);
+	if (ret < 0) {
+		errno = -ret;
+		error("failed to clear v1 free space cache: %m");
+		return ret;
+	}
+
+	ret = btrfs_create_free_space_tree(fs_info);
+	if (ret < 0) {
+		errno = -ret;
+		error("failed to create free space tree: %m");
+		return ret;
+	}
+	pr_verbose(LOG_DEFAULT, "Converted to free space tree feature\n");
+	return ret;
+}
+
 static int cmd_tune_set(const struct cmd_struct *cmd, int argc, char **argv)
 {
 	struct btrfs_fs_info *fs_info;
@@ -202,6 +243,17 @@ static int cmd_tune_set(const struct cmd_struct *cmd, int argc, char **argv)
 		}
 		goto out;
 	}
+	if (!strcmp(feature, "free-space-tree")) {
+		if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
+			error("filesystem already has free-space-tree feature");
+			ret = -EINVAL;
+			goto out;
+		}
+		ret = convert_to_fst(fs_info);
+		if (ret < 0)
+			error("failed to convert the filesystem to free-space-tree feature");
+		goto out;
+	}
 
 out:
 	if (fs_info)
Signature · 1 line
2.42.0
author
#10 lore raw

PATCH 4/7 btrfs-progs: cmds/tune: add set support for block-group-tree feature

In reply to Qu Wenruo

Most of the code is reusing the existing code from btrfs-tune, only needs to add extra linkage dependency on tune/convert-bgt.o.

---
 Makefile    |  2 +-
 cmds/tune.c | 25 +++++++++++++++++++++++++
 2 files changed, 26 insertions(+), 1 deletion(-)
Makefile +1 -1
index 9857daaa42ac..45d64129352c 100644
@@ -241,7 +241,7 @@ cmds_objects = cmds/subvolume.o cmds/subvolume-list.o \
 	       cmds/inspect-dump-super.o cmds/inspect-tree-stats.o cmds/filesystem-du.o \
 	       cmds/reflink.o cmds/tune.o \
 	       mkfs/common.o check/mode-common.o check/mode-lowmem.o \
-	       check/clear-cache.o
+	       check/clear-cache.o tune/convert-bgt.o
 
 libbtrfs_objects = \
 		kernel-lib/rbtree.o	\
cmds/tune.c +25 -0
index 9906cde393f4..76ad9ebfc39e 100644
@@ -24,6 +24,7 @@
 #include "kernel-shared/disk-io.h"
 #include "kernel-shared/transaction.h"
 #include "kernel-shared/free-space-tree.h"
+#include "tune/tune.h"
 
 static const char * const cmd_tune_set_usage[] = {
 	"btrfs tune set <feature> [<device>]",
@@ -68,6 +69,14 @@ static const struct btrfs_feature set_features[] = {
 		VERSION_TO_STRING2(safe, 4,9),
 		VERSION_TO_STRING2(default, 5,15),
 		.desc		= "free space tree (space_cache=v2)"
+	}, {
+		.name		= "block-group-tree",
+		.compat_ro_flag	= BTRFS_FEATURE_COMPAT_RO_BLOCK_GROUP_TREE,
+		.sysfs_name	= "block_group_tree",
+		VERSION_TO_STRING2(compat, 6,1),
+		VERSION_NULL(safe),
+		VERSION_NULL(default),
+		.desc		= "block group tree to reduce mount time"
 	},
 	/* Keep this one last */
 	{
@@ -254,6 +263,22 @@ static int cmd_tune_set(const struct cmd_struct *cmd, int argc, char **argv)
 			error("failed to convert the filesystem to free-space-tree feature");
 		goto out;
 	}
+	if (!strcmp(feature, "block-group-tree")) {
+		if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
+			error("filesystem already has block-group-tree feature");
+			ret = -EINVAL;
+			goto out;
+		}
+		if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
+			error("the filesystem doesn't have free-space-tree feature, please enable it first");
+			ret = -EINVAL;
+			goto out;
+		}
+		ret = convert_to_bg_tree(fs_info);
+		if (ret < 0)
+			error("failed to convert the filesystem to free-space-tree feature");
+		goto out;
+	}
 
 out:
 	if (fs_info)
Signature · 1 line
2.42.0
author
#1 lore raw

PATCH 1/7 btrfs-progs: export btrfs_feature structure

In reply to Qu Wenruo

For the incoming "btrfs tune" subcommand, we will have different features supported by that subcommand.

Instead of bloating the runtime and mkfs features, here we just export btrfs_feature, so each subcommand can have their own definition of supported features.

And since we're here, also add needed headers for future users of "fsfeatures.h".

---
 common/fsfeatures.c | 53 ---------------------------------------------
 common/fsfeatures.h | 50 ++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 50 insertions(+), 53 deletions(-)
common/fsfeatures.c +0 -53
index 9ee392d3a8a6..f8eeea7695c1 100644
@@ -32,64 +32,11 @@
 #include "common/sysfs-utils.h"
 #include "common/messages.h"
 
-/*
- * Insert a root item for temporary tree root
- *
- * Only used in make_btrfs_v2().
- */
-#define VERSION_TO_STRING3(name, a,b,c)				\
-	.name ## _str = #a "." #b "." #c,			\
-	.name ## _ver = KERNEL_VERSION(a,b,c)
-#define VERSION_TO_STRING2(name, a,b)				\
-	.name ## _str = #a "." #b,				\
-	.name ## _ver = KERNEL_VERSION(a,b,0)
-#define VERSION_NULL(name)					\
-	.name ## _str = NULL,					\
-	.name ## _ver = 0
-
 enum feature_source {
 	FS_FEATURES,
 	RUNTIME_FEATURES,
 };
 
-/*
- * Feature stability status and versions: compat <= safe <= default
- */
-struct btrfs_feature {
-	const char *name;
-
-	/*
-	 * At least one of the bit must be set in the following *_flag member.
-	 *
-	 * For features like list-all and quota which don't have any
-	 * incompat/compat_ro bit set, it go to runtime_flag.
-	 */
-	u64 incompat_flag;
-	u64 compat_ro_flag;
-	u64 runtime_flag;
-
-	const char *sysfs_name;
-	/*
-	 * Compatibility with kernel of given version. Filesystem can be
-	 * mounted.
-	 */
-	const char *compat_str;
-	u32 compat_ver;
-	/*
-	 * Considered safe for use, but is not on by default, even if the
-	 * kernel supports the feature.
-	 */
-	const char *safe_str;
-	u32 safe_ver;
-	/*
-	 * Considered safe for use and will be turned on by default if
-	 * supported by the running kernel.
-	 */
-	const char *default_str;
-	u32 default_ver;
-	const char *desc;
-};
-
 static const struct btrfs_feature mkfs_features[] = {
 	{
 		.name		= "mixed-bg",
common/fsfeatures.h +50 -0

63 lines not on this page · raw message

Signature · 1 line
2.42.0
author
#11 lore raw

PATCH 5/7 btrfs-progs: cmds/tune: add set support for seeding device

In reply to Qu Wenruo

With the new "btrfs tune set" command, seeding device support can be added just like all other features.

---
 Makefile    |  2 +-
 cmds/tune.c | 35 ++++++++++++++++++++++++++++++++++-
 2 files changed, 35 insertions(+), 2 deletions(-)
Makefile +1 -1
index 45d64129352c..b502336d7441 100644
@@ -241,7 +241,7 @@ cmds_objects = cmds/subvolume.o cmds/subvolume-list.o \
 	       cmds/inspect-dump-super.o cmds/inspect-tree-stats.o cmds/filesystem-du.o \
 	       cmds/reflink.o cmds/tune.o \
 	       mkfs/common.o check/mode-common.o check/mode-lowmem.o \
-	       check/clear-cache.o tune/convert-bgt.o
+	       check/clear-cache.o tune/convert-bgt.o tune/seeding.o
 
 libbtrfs_objects = \
 		kernel-lib/rbtree.o	\
cmds/tune.c +34 -1
index 76ad9ebfc39e..195d0a235fc8 100644
@@ -19,6 +19,7 @@
 #include "cmds/commands.h"
 #include "check/clear-cache.h"
 #include "common/help.h"
+#include "common/utils.h"
 #include "common/fsfeatures.h"
 #include "kernel-shared/messages.h"
 #include "kernel-shared/disk-io.h"
@@ -30,6 +31,7 @@ static const char * const cmd_tune_set_usage[] = {
 	"btrfs tune set <feature> [<device>]",
 	"Set/enable specified feature for the unmounted filesystem",
 	"",
+	OPTLINE("-f", "force dangerous operations."),
 	HELPINFO_INSERT_GLOBALS,
 	HELPINFO_INSERT_VERBOSE,
 	NULL,
@@ -77,6 +79,13 @@ static const struct btrfs_feature set_features[] = {
 		VERSION_NULL(safe),
 		VERSION_NULL(default),
 		.desc		= "block group tree to reduce mount time"
+	}, {
+		.name		= "seed",
+		.sysfs_name	= NULL,
+		VERSION_TO_STRING3(compat, 2,6,29),
+		VERSION_NULL(safe),
+		VERSION_NULL(default),
+		.desc		= "seeding device support"
 	},
 	/* Keep this one last */
 	{
@@ -190,15 +199,19 @@ static int cmd_tune_set(const struct cmd_struct *cmd, int argc, char **argv)
 	struct open_ctree_args oca = { 0 };
 	char *feature;
 	char *path;
+	bool force = false;
 	int ret = 0;
 
 	optind = 0;
 	while (1) {
-		int c = getopt(argc, argv, "");
+		int c = getopt(argc, argv, "f");
 		if (c < 0)
 			break;
 
 		switch (c) {
+		case 'f':
+			force = true;
+			break;
 		default:
 			usage_unknown_option(cmd, argv);
 		}
@@ -279,6 +292,26 @@ static int cmd_tune_set(const struct cmd_struct *cmd, int argc, char **argv)
 			error("failed to convert the filesystem to free-space-tree feature");
 		goto out;
 	}
+	if (!strcmp(feature, "seed")) {
+		if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
+			error("SEED flag cannot be changed on a metadata-uuid changed fs");
+			ret = -EINVAL;
+			goto out;
+		}
+		if (!force) {
+			warning(
+"this is dangerous, clearing the seeding flag may cause the derived device not to be mountable!");
+			ret = ask_user("We are going to clear the seeding flag, are you sure?");
+			if (!ret) {
+				error("clear seeding flag canceled");
+				ret = -EINVAL;
+				goto out;
+			}
+		}
+
+		ret = update_seeding_flag(fs_info->tree_root, path, 1, force);
+		goto out;
+	}
 
 out:
 	if (fs_info)
Signature · 1 line
2.42.0
author
#12 lore raw

PATCH 6/7 btrfs-progs: cmds/tune: add "btrfs tune clear" subcommand

In reply to Qu Wenruo

This new "btrfs tune clear" subcommand is mostly the same as "btrfs tune set", but with much less features supported:

- seed
- block-group-tree

For the features in "btrfs tune set" but not in "btrfs tune clear", they are mostly default features now, and we won't support disabling those features anymore.

---
 Documentation/btrfs-tune.rst |   7 +++
 cmds/tune.c                  | 111 +++++++++++++++++++++++++++++++++++
 2 files changed, 118 insertions(+)
Documentation/btrfs-tune.rst +7 -0
index 827c92eadb72..4a556a197540 100644
@@ -21,6 +21,13 @@ set <feature> [<device>]
         If *feature* is `list-all`, all supported features would be listed, and
 	no *device* parameter is needed.
 
+clear <feature> [<device>]
+        Set/enable a feature.
+
+        If *feature* is `list-all`, all supported features would be listed, and
+	no *device* parameter is needed. And the listed features may be different
+	compared to the ones from `btrfs tune set list-all`.
+
 EXIT STATUS
 -----------
 
cmds/tune.c +111 -0

123 lines not on this page · raw message

Signature · 1 line
2.42.0
author
#4 lore raw

PATCH 2/7 btrfs-progs: cmds: add "btrfs tune set" subcommand group

In reply to Qu Wenruo

As the first step to convert btrfstune into "btrfs tune" subcommand group, this patch would add the following subcommand group:

btrfs tune set <feature> [<device>]

For now the following features are supported:

- extref
- skinny-metadata
- no-holes
All those are simple super block flags toggle.

- list-all
This acts the same way as "mkfs.btrfs -O list-all", the difference is it would only list the supported features.
(In the future, there will be "btrfs tune clear" subcommand, which
would support a different set of features).

---
 Documentation/btrfs-tune.rst |  40 ++++++
 Documentation/btrfs.rst      |   5 +
 Documentation/conf.py        |   1 +
 Documentation/man-index.rst  |   1 +
 Makefile                     |   2 +-
 btrfs.c                      |   1 +
 cmds/commands.h              |   1 +
 cmds/tune.c                  | 227 +++++++++++++++++++++++++++++++++++
 8 files changed, 277 insertions(+), 1 deletion(-)
 create mode 100644 Documentation/btrfs-tune.rst
 create mode 100644 cmds/tune.c
Documentation/btrfs-tune.rst new file +40 -0
new file mode 100644
index 000000000000..827c92eadb72
@@ -0,0 +1,40 @@
+btrfs-tune(8)
+==================
+
+SYNOPSIS
+--------
+
+**btrfs tune** <subcommand> [<args>]
+
+DESCRIPTION
+-----------
+
+:command:`btrfs tune` is used to tweak various btrfs features on a
+unmounted filesystem.
+
+SUBCOMMAND
+-----------
+
+set <feature> [<device>]
+        Set/enable a feature.
+
+        If *feature* is `list-all`, all supported features would be listed, and
+	no *device* parameter is needed.
+
+EXIT STATUS
+-----------
+
+**btrfs tune** returns a zero exit status if it succeeds. A non-zero value is
+returned in case of failure.
+
+AVAILABILITY
+------------
+
+**btrfs** is part of btrfs-progs.  Please refer to the documentation at
+`https://btrfs.readthedocs.io <https://btrfs.readthedocs.io>`_.
+
+SEE ALSO
+--------
+
+:doc:`mkfs.btrfs`,
+``mount(8)``
Documentation/btrfs.rst +5 -0
index e878f158aaa1..5aea0d1a208c 100644
@@ -134,6 +134,10 @@ subvolume
 	Create/delete/list/manage btrfs subvolume.
 	See :doc:`btrfs-subvolume` for details.
 
+tune
+	Change various btrfs features.
+	See :doc:`btrfs-tune` for details.
+
 .. _man-btrfs8-standalone-tools:
 
 STANDALONE TOOLS
@@ -150,6 +154,7 @@ btrfs-convert
         in-place conversion from ext2/3/4 filesystems to btrfs
 btrfstune
         tweak some filesystem properties on a unmounted filesystem
+	(will be replaced by `btrfs-tune`)
 btrfs-select-super
         rescue tool to overwrite primary superblock from a spare copy
 btrfs-find-root
Documentation/conf.py +1 -0
index 1025e10d7206..e0801bca4686 100644
@@ -66,6 +66,7 @@ man_pages = [
     ('btrfs-check', 'btrfs-check', 'check or repair a btrfs filesystem', '', 8),
     ('btrfs-balance', 'btrfs-balance', 'balance block groups on a btrfs filesystem', '', 8),
     ('btrfs-subvolume', 'btrfs-subvolume', 'manage btrfs subvolumes', '', 8),
+    ('btrfs-tune', 'btrfs-tune', 'tweak btrfs features', '', 8),
     ('btrfs-map-logical', 'btrfs-map-logical', 'map btrfs logical extent to physical extent', '', 8),
     ('btrfs', 'btrfs', 'a toolbox to manage btrfs filesystems', '', 8),
     ('mkfs.btrfs', 'mkfs.btrfs', 'create a btrfs filesystem', '', 8),
Documentation/man-index.rst +1 -0
index 36d45d2903ea..5fcd4cbc4bee 100644
@@ -28,6 +28,7 @@ Manual pages
    btrfs-select-super
    btrfs-send
    btrfs-subvolume
+   btrfs-tune
    btrfstune
    fsck.btrfs
    mkfs.btrfs
Makefile +1 -1
index f4feb1fff8e1..9857daaa42ac 100644
@@ -239,7 +239,7 @@ cmds_objects = cmds/subvolume.o cmds/subvolume-list.o \
 	       cmds/rescue-super-recover.o \
 	       cmds/property.o cmds/filesystem-usage.o cmds/inspect-dump-tree.o \
 	       cmds/inspect-dump-super.o cmds/inspect-tree-stats.o cmds/filesystem-du.o \
-	       cmds/reflink.o \
+	       cmds/reflink.o cmds/tune.o \
 	       mkfs/common.o check/mode-common.o check/mode-lowmem.o \
 	       check/clear-cache.o
 
btrfs.c +1 -0
index 751f193ee2e0..c2dae0303ffe 100644
@@ -389,6 +389,7 @@ static const struct cmd_group btrfs_cmd_group = {
 		&cmd_struct_scrub,
 		&cmd_struct_send,
 		&cmd_struct_subvolume,
+		&cmd_struct_tune,
 
 		/* Help and version stay last */
 		&cmd_struct_help,
cmds/commands.h +1 -0
index 5ab7c881f634..aebacd718a7b 100644
@@ -151,5 +151,6 @@ DECLARE_COMMAND(qgroup);
 DECLARE_COMMAND(replace);
 DECLARE_COMMAND(restore);
 DECLARE_COMMAND(rescue);
+DECLARE_COMMAND(tune);
 
 #endif
cmds/tune.c new file +227 -0

227 lines not on this page · raw message

Signature · 1 line
2.42.0

Qu Wenruo

author
#0 lore raw

cover btrfs-progs: cmds/tune: add set/clear features

This is the first step to convert btrfstune functionality to "btrfs tune" subcommand group.

For now only binary features, aka set and clear, is supported, thus uuid and csum change is not yet implemented.
(Both need their own subcommand groups other than set/clear groups)

And even for set/clear, there is some changes to btrfstune:

- Merge seed feature into set/clear
To enable seeding, just go "btrfs tune set seed <device>".

- All supported features can be checked by "list-all" feature Please note that, "btrfs tune set list-all" and "btrfs tune clear list-all" will have different output.

The reason is some fundamental features like no-holes can not be disabled.

Qu Wenruo (7):
btrfs-progs: export btrfs_feature structure
btrfs-progs: cmds: add "btrfs tune set" subcommand group
btrfs-progs: cmds/tune: add set support for free-space-tree feature
btrfs-progs: cmds/tune: add set support for block-group-tree feature
btrfs-progs: cmds/tune: add set support for seeding device
btrfs-progs: cmds/tune: add "btrfs tune clear" subcommand
btrfs-progs: tests/cli: add a test case for "btrfs tune" subcommand

 Documentation/btrfs-tune.rst           |  47 +++
 Documentation/btrfs.rst                |   5 +
 Documentation/conf.py                  |   1 +
 Documentation/man-index.rst            |   1 +
 Makefile                               |   4 +-
 btrfs.c                                |   1 +
 cmds/commands.h                        |   1 +
 cmds/tune.c                            | 448 +++++++++++++++++++++++++
 common/fsfeatures.c                    |  53 ---
 common/fsfeatures.h                    |  50 +++
 tests/cli-tests/018-btrfs-tune/test.sh |  40 +++
 11 files changed, 596 insertions(+), 55 deletions(-)
 create mode 100644 Documentation/btrfs-tune.rst
 create mode 100644 cmds/tune.c
 create mode 100755 tests/cli-tests/018-btrfs-tune/test.sh

--
2.42.0

author
#14 lore raw

Hi David,

Mind to give some feedback on this?
As this is the basis for all later "btrfs tune" subcommand.

Thanks,
Qu

Quoted 48 lines · Qu Wenruo

This is the first step to convert btrfstune functionality to "btrfs tune" subcommand group.

For now only binary features, aka set and clear, is supported, thus uuid and csum change is not yet implemented.
(Both need their own subcommand groups other than set/clear groups)

And even for set/clear, there is some changes to btrfstune:

- Merge seed feature into set/clear
To enable seeding, just go "btrfs tune set seed <device>".

- All supported features can be checked by "list-all" feature Please note that, "btrfs tune set list-all" and "btrfs tune clear list-all" will have different output.

The reason is some fundamental features like no-holes can not be disabled.

Qu Wenruo (7):
btrfs-progs: export btrfs_feature structure
btrfs-progs: cmds: add "btrfs tune set" subcommand group
btrfs-progs: cmds/tune: add set support for free-space-tree feature
btrfs-progs: cmds/tune: add set support for block-group-tree feature
btrfs-progs: cmds/tune: add set support for seeding device
btrfs-progs: cmds/tune: add "btrfs tune clear" subcommand
btrfs-progs: tests/cli: add a test case for "btrfs tune" subcommand

  Documentation/btrfs-tune.rst           |  47 +++
  Documentation/btrfs.rst                |   5 +
  Documentation/conf.py                  |   1 +
  Documentation/man-index.rst            |   1 +
  Makefile                               |   4 +-
  btrfs.c                                |   1 +
  cmds/commands.h                        |   1 +
  cmds/tune.c                            | 448 +++++++++++++++++++++++++
  common/fsfeatures.c                    |  53 ---
  common/fsfeatures.h                    |  50 +++
  tests/cli-tests/018-btrfs-tune/test.sh |  40 +++
  11 files changed, 596 insertions(+), 55 deletions(-)

create mode 100644 Documentation/btrfs-tune.rst create mode 100644 cmds/tune.c
create mode 100755 tests/cli-tests/018-btrfs-tune/test.sh

--
2.42.0

#2 lore raw

Re: [PATCH 1/7] btrfs-progs: export btrfs_feature structure

In reply to Qu Wenruo
Quoted 7 lines · Qu Wenruo

For the incoming "btrfs tune" subcommand, we will have different features supported by that subcommand.

Instead of bloating the runtime and mkfs features, here we just export btrfs_feature, so each subcommand can have their own definition of supported features.

Looks good.

Quoted 3 lines

And since we're here, also add needed headers for future users of "fsfeatures.h".

I don't see anything added, missed?

Thanks, Anand

Quoted 144 lines
---
  common/fsfeatures.c | 53 ---------------------------------------------
  common/fsfeatures.h | 50 ++++++++++++++++++++++++++++++++++++++++++
  2 files changed, 50 insertions(+), 53 deletions(-)
common/fsfeatures.c +0 -53
index 9ee392d3a8a6..f8eeea7695c1 100644
@@ -32,64 +32,11 @@
  #include "common/sysfs-utils.h"
  #include "common/messages.h"
 
-/*
- * Insert a root item for temporary tree root
- *
- * Only used in make_btrfs_v2().
- */
-#define VERSION_TO_STRING3(name, a,b,c)				\
-	.name ## _str = #a "." #b "." #c,			\
-	.name ## _ver = KERNEL_VERSION(a,b,c)
-#define VERSION_TO_STRING2(name, a,b)				\
-	.name ## _str = #a "." #b,				\
-	.name ## _ver = KERNEL_VERSION(a,b,0)
-#define VERSION_NULL(name)					\
-	.name ## _str = NULL,					\
-	.name ## _ver = 0
-
  enum feature_source {
  	FS_FEATURES,
  	RUNTIME_FEATURES,
  };
 
-/*
- * Feature stability status and versions: compat <= safe <= default
- */
-struct btrfs_feature {
-	const char *name;
-
-	/*
-	 * At least one of the bit must be set in the following *_flag member.
-	 *
-	 * For features like list-all and quota which don't have any
-	 * incompat/compat_ro bit set, it go to runtime_flag.
-	 */
-	u64 incompat_flag;
-	u64 compat_ro_flag;
-	u64 runtime_flag;
-
-	const char *sysfs_name;
-	/*
-	 * Compatibility with kernel of given version. Filesystem can be
-	 * mounted.
-	 */
-	const char *compat_str;
-	u32 compat_ver;
-	/*
-	 * Considered safe for use, but is not on by default, even if the
-	 * kernel supports the feature.
-	 */
-	const char *safe_str;
-	u32 safe_ver;
-	/*
-	 * Considered safe for use and will be turned on by default if
-	 * supported by the running kernel.
-	 */
-	const char *default_str;
-	u32 default_ver;
-	const char *desc;
-};
-
  static const struct btrfs_feature mkfs_features[] = {
  	{
  		.name		= "mixed-bg",
common/fsfeatures.h +50 -0
index c4ab704862cd..c9fb489d2d79 100644
@@ -19,7 +19,9 @@
 
  #include "kerncompat.h"
  #include <stdio.h>
+#include <linux/version.h>
  #include "kernel-lib/sizes.h"
+#include "kernel-shared/uapi/btrfs.h"
 
  #define BTRFS_MKFS_DEFAULT_NODE_SIZE SZ_16K
 
@@ -43,6 +45,54 @@ struct btrfs_mkfs_features {
   */
  #define BTRFS_FEATURE_STRING_BUF_SIZE		(160)
 
+#define VERSION_TO_STRING3(name, a,b,c)				\
+	.name ## _str = #a "." #b "." #c,			\
+	.name ## _ver = KERNEL_VERSION(a,b,c)
+#define VERSION_TO_STRING2(name, a,b)				\
+	.name ## _str = #a "." #b,				\
+	.name ## _ver = KERNEL_VERSION(a,b,0)
+#define VERSION_NULL(name)					\
+	.name ## _str = NULL,					\
+	.name ## _ver = 0
+
+/*
+ * Feature stability status and versions: compat <= safe <= default
+ */
+struct btrfs_feature {
+	const char *name;
+
+	/*
+	 * At least one of the bit must be set in the following *_flag member.
+	 *
+	 * For features like list-all and quota which don't have any
+	 * incompat/compat_ro bit set, it go to runtime_flag.
+	 */
+	u64 incompat_flag;
+	u64 compat_ro_flag;
+	u64 runtime_flag;
+
+	const char *sysfs_name;
+	/*
+	 * Compatibility with kernel of given version. Filesystem can be
+	 * mounted.
+	 */
+	const char *compat_str;
+	u32 compat_ver;
+	/*
+	 * Considered safe for use, but is not on by default, even if the
+	 * kernel supports the feature.
+	 */
+	const char *safe_str;
+	u32 safe_ver;
+	/*
+	 * Considered safe for use and will be turned on by default if
+	 * supported by the running kernel.
+	 */
+	const char *default_str;
+	u32 default_ver;
+	const char *desc;
+};
+
  static const struct btrfs_mkfs_features btrfs_mkfs_default_features = {
  	.compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE |
  			   BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE_VALID,
#5 lore raw

Re: [PATCH 2/7] btrfs-progs: cmds: add "btrfs tune set" subcommand group

In reply to Qu Wenruo
Quoted 12 lines · Qu Wenruo

As the first step to convert btrfstune into "btrfs tune" subcommand group, this patch would add the following subcommand group:

btrfs tune set <feature> [<device>]

For now the following features are supported:

- extref
- skinny-metadata
- no-holes
All those are simple super block flags toggle.

Quoted 6 lines

- list-all
This acts the same way as "mkfs.btrfs -O list-all", the difference is it would only list the supported features.
(In the future, there will be "btrfs tune clear" subcommand, which
would support a different set of features).

With this patchset, the syntax is structured as follows:

   $ btrfs tune --help
   usage: btrfs tune <command> <args>

btrfs tune set <feature> [<device>]
Set/enable specified feature for the unmounted filesystem
btrfs tune clear <feature> [<device>]
Clear/disable specified feature for the unmounted filesystem

change various btrfs features

However, for consistency, I suggest the following syntax:

set:

   $ btrfs tune <feature> /dev/sda
  clear:
   $ btrfs tune <feature> --clear /dev/sda

  list:
   $ btrfs tune --list-all

This syntax aligns with the:

   $ btrfs device scan --forget

Thanks, Anand

Quoted 376 lines
---
  Documentation/btrfs-tune.rst |  40 ++++++
  Documentation/btrfs.rst      |   5 +
  Documentation/conf.py        |   1 +
  Documentation/man-index.rst  |   1 +
  Makefile                     |   2 +-
  btrfs.c                      |   1 +
  cmds/commands.h              |   1 +
  cmds/tune.c                  | 227 +++++++++++++++++++++++++++++++++++
  8 files changed, 277 insertions(+), 1 deletion(-)

create mode 100644 Documentation/btrfs-tune.rst
create mode 100644 cmds/tune.c

Documentation/btrfs-tune.rst new file +40 -0
new file mode 100644
index 000000000000..827c92eadb72
@@ -0,0 +1,40 @@
+btrfs-tune(8)
+==================
+
+SYNOPSIS
+--------
+
+**btrfs tune** <subcommand> [<args>]
+
+DESCRIPTION
+-----------
+
+:command:`btrfs tune` is used to tweak various btrfs features on a
+unmounted filesystem.
+
+SUBCOMMAND
+-----------
+
+set <feature> [<device>]
+        Set/enable a feature.
+
+        If *feature* is `list-all`, all supported features would be listed, and
+	no *device* parameter is needed.
+
+EXIT STATUS
+-----------
+
+**btrfs tune** returns a zero exit status if it succeeds. A non-zero value is
+returned in case of failure.
+
+AVAILABILITY
+------------
+
+**btrfs** is part of btrfs-progs.  Please refer to the documentation at
+`https://btrfs.readthedocs.io <https://btrfs.readthedocs.io>`_.
+
+SEE ALSO
+--------
+
+:doc:`mkfs.btrfs`,
+``mount(8)``
Documentation/btrfs.rst +5 -0
index e878f158aaa1..5aea0d1a208c 100644
@@ -134,6 +134,10 @@ subvolume
  	Create/delete/list/manage btrfs subvolume.
  	See :doc:`btrfs-subvolume` for details.
 
+tune
+	Change various btrfs features.
+	See :doc:`btrfs-tune` for details.
+
  .. _man-btrfs8-standalone-tools:
 
  STANDALONE TOOLS
@@ -150,6 +154,7 @@ btrfs-convert
          in-place conversion from ext2/3/4 filesystems to btrfs
  btrfstune
          tweak some filesystem properties on a unmounted filesystem
+	(will be replaced by `btrfs-tune`)
  btrfs-select-super
          rescue tool to overwrite primary superblock from a spare copy
  btrfs-find-root
Documentation/conf.py +1 -0
index 1025e10d7206..e0801bca4686 100644
@@ -66,6 +66,7 @@ man_pages = [
      ('btrfs-check', 'btrfs-check', 'check or repair a btrfs filesystem', '', 8),
      ('btrfs-balance', 'btrfs-balance', 'balance block groups on a btrfs filesystem', '', 8),
      ('btrfs-subvolume', 'btrfs-subvolume', 'manage btrfs subvolumes', '', 8),
+    ('btrfs-tune', 'btrfs-tune', 'tweak btrfs features', '', 8),
      ('btrfs-map-logical', 'btrfs-map-logical', 'map btrfs logical extent to physical extent', '', 8),
      ('btrfs', 'btrfs', 'a toolbox to manage btrfs filesystems', '', 8),
      ('mkfs.btrfs', 'mkfs.btrfs', 'create a btrfs filesystem', '', 8),
Documentation/man-index.rst +1 -0
index 36d45d2903ea..5fcd4cbc4bee 100644
@@ -28,6 +28,7 @@ Manual pages
     btrfs-select-super
     btrfs-send
     btrfs-subvolume
+   btrfs-tune
     btrfstune
     fsck.btrfs
     mkfs.btrfs
Makefile +1 -1
index f4feb1fff8e1..9857daaa42ac 100644
@@ -239,7 +239,7 @@ cmds_objects = cmds/subvolume.o cmds/subvolume-list.o \
  	       cmds/rescue-super-recover.o \
  	       cmds/property.o cmds/filesystem-usage.o cmds/inspect-dump-tree.o \
  	       cmds/inspect-dump-super.o cmds/inspect-tree-stats.o cmds/filesystem-du.o \
-	       cmds/reflink.o \
+	       cmds/reflink.o cmds/tune.o \
  	       mkfs/common.o check/mode-common.o check/mode-lowmem.o \
  	       check/clear-cache.o
 
btrfs.c +1 -0
index 751f193ee2e0..c2dae0303ffe 100644
@@ -389,6 +389,7 @@ static const struct cmd_group btrfs_cmd_group = {
  		&cmd_struct_scrub,
  		&cmd_struct_send,
  		&cmd_struct_subvolume,
+		&cmd_struct_tune,
 
  		/* Help and version stay last */
  		&cmd_struct_help,
cmds/commands.h +1 -0
index 5ab7c881f634..aebacd718a7b 100644
@@ -151,5 +151,6 @@ DECLARE_COMMAND(qgroup);
  DECLARE_COMMAND(replace);
  DECLARE_COMMAND(restore);
  DECLARE_COMMAND(rescue);
+DECLARE_COMMAND(tune);
 
  #endif
cmds/tune.c new file +227 -0
new file mode 100644
index 000000000000..92c7b9f1502c
@@ -0,0 +1,227 @@
+/*
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <unistd.h>
+#include "kerncompat.h"
+#include "cmds/commands.h"
+#include "common/help.h"
+#include "common/fsfeatures.h"
+#include "kernel-shared/messages.h"
+#include "kernel-shared/disk-io.h"
+#include "kernel-shared/transaction.h"
+
+static const char * const cmd_tune_set_usage[] = {
+	"btrfs tune set <feature> [<device>]",
+	"Set/enable specified feature for the unmounted filesystem",
+	"",
+	HELPINFO_INSERT_GLOBALS,
+	HELPINFO_INSERT_VERBOSE,
+	NULL,
+};
+
+static const struct btrfs_feature set_features[] = {
+	{
+		.name		= "extref",
+		.incompat_flag	= BTRFS_FEATURE_INCOMPAT_EXTENDED_IREF,
+		.sysfs_name	= "extended_iref",
+		VERSION_TO_STRING2(compat, 3,7),
+		VERSION_TO_STRING2(safe, 3,12),
+		VERSION_TO_STRING2(default, 3,12),
+		.desc		= "increased hardlink limit per file to 65536"
+	}, {
+		.name		= "skinny-metadata",
+		.incompat_flag	= BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA,
+		.sysfs_name	= "skinny_metadata",
+		VERSION_TO_STRING2(compat, 3,10),
+		VERSION_TO_STRING2(safe, 3,18),
+		VERSION_TO_STRING2(default, 3,18),
+		.desc		= "reduced-size metadata extent refs"
+	}, {
+		.name		= "no-holes",
+		.incompat_flag	= BTRFS_FEATURE_INCOMPAT_NO_HOLES,
+		.sysfs_name	= "no_holes",
+		VERSION_TO_STRING2(compat, 3,14),
+		VERSION_TO_STRING2(safe, 4,0),
+		VERSION_TO_STRING2(default, 5,15),
+		.desc		= "no explicit hole extents for files"
+	},
+	/* Keep this one last */
+	{
+		.name		= "list-all",
+		.runtime_flag	= BTRFS_FEATURE_RUNTIME_LIST_ALL,
+		.sysfs_name	= NULL,
+		VERSION_NULL(compat),
+		VERSION_NULL(safe),
+		VERSION_NULL(default),
+		.desc		= NULL
+	}
+};
+
+static void list_all_features(const char *prefix,
+			      const struct btrfs_feature *features,
+			      int nr_features)
+{
+	/* We should have at least one empty feature. */
+	ASSERT(nr_features > 1);
+
+	printf("features available to %s:\n", prefix);
+	for (int i = 0; i < nr_features - 1; i++) {
+		const struct btrfs_feature *feat = features + i;
+		const char *sep = "";
+
+		fprintf(stderr, "%-20s- %s (", feat->name, feat->desc);
+		if (feat->compat_ver) {
+			fprintf(stderr, "compat=%s", feat->compat_str);
+			sep = ", ";
+		}
+		if (feat->safe_ver) {
+			fprintf(stderr, "%ssafe=%s", sep, feat->safe_str);
+			sep = ", ";
+		}
+		if (feat->default_ver)
+			fprintf(stderr, "%sdefault=%s", sep, feat->default_str);
+		fprintf(stderr, ")\n");
+	}
+}
+
+static int check_features(const char *name, const struct btrfs_feature *features,
+			  int nr_features)
+{
+	bool found = false;
+
+	for (int i = 0; i < nr_features; i++) {
+		const struct btrfs_feature *feat = features + i;
+
+		if (!strcmp(feat->name, name)) {
+			found = true;
+			break;
+		}
+	}
+	if (found)
+		return 0;
+	return -EINVAL;
+}
+
+static int set_super_incompat_flags(struct btrfs_fs_info *fs_info, u64 flags)
+{
+	struct btrfs_root *root = fs_info->tree_root;
+	struct btrfs_trans_handle *trans;
+	struct btrfs_super_block *disk_super;
+	u64 super_flags;
+	int ret;
+
+	disk_super = fs_info->super_copy;
+	super_flags = btrfs_super_incompat_flags(disk_super);
+	super_flags |= flags;
+	trans = btrfs_start_transaction(root, 1);
+	BUG_ON(IS_ERR(trans));
+	btrfs_set_super_incompat_flags(disk_super, super_flags);
+	ret = btrfs_commit_transaction(trans, root);
+
+	return ret;
+}
+
+static int cmd_tune_set(const struct cmd_struct *cmd, int argc, char **argv)
+{
+	struct btrfs_fs_info *fs_info;
+	struct open_ctree_args oca = { 0 };
+	char *feature;
+	char *path;
+	int ret = 0;
+
+	optind = 0;
+	while (1) {
+		int c = getopt(argc, argv, "");
+		if (c < 0)
+			break;
+
+		switch (c) {
+		default:
+			usage_unknown_option(cmd, argv);
+		}
+	}
+
+	if (check_argc_min(argc - optind, 1))
+		return 1;
+
+	feature = argv[optind];
+
+	if (check_features(feature, set_features, ARRAY_SIZE(set_features))) {
+		error("Unknown feature to set: %s", feature);
+		return 1;
+	}
+	if (!strcmp(feature, "list-all")) {
+		list_all_features("set", set_features, ARRAY_SIZE(set_features));
+		return 0;
+	}
+
+	if (check_argc_exact(argc - optind, 2))
+		return 1;
+
+	path = argv[optind + 1];
+	oca.flags = OPEN_CTREE_WRITES;
+	oca.filename = path;
+	fs_info = open_ctree_fs_info(&oca);
+	if (!fs_info) {
+		error("failed to open btrfs");
+		ret = -EIO;
+		goto out;
+	}
+	/*
+	 * For those 3 features, we only need to update the superblock to add
+	 * the new feature flags.
+	 */
+	if (!strcmp(feature, "extref") ||
+	    !strcmp(feature, "skinny-metadata") ||
+	    !strcmp(feature, "no-holes")) {
+		u64 incompat_flags = 0;
+
+		if (!strcmp(feature, "extref"))
+			incompat_flags |= BTRFS_FEATURE_INCOMPAT_EXTENDED_IREF;
+		if (!strcmp(feature, "skinny-metadata"))
+			incompat_flags |= BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA;
+		if (!strcmp(feature, "no-holes"))
+			incompat_flags |= BTRFS_FEATURE_INCOMPAT_NO_HOLES;
+		ret = set_super_incompat_flags(fs_info, incompat_flags);
+		if (ret < 0) {
+			errno = -ret;
+			error("failed to set feature '%s': %m", feature);
+		}
+		goto out;
+	}
+
+out:
+	if (fs_info)
+		close_ctree_fs_info(fs_info);
+	return !!ret;
+}
+
+static DEFINE_SIMPLE_COMMAND(tune_set, "set");
+
+static const char * const tune_cmd_group_usage[] = {
+	"btrfs tune <command> <args>",
+	NULL,
+};
+
+static const char tune_cmd_group_info[] = "change various btrfs features";
+
+static const struct cmd_group tune_cmd_group = {
+	tune_cmd_group_usage, tune_cmd_group_info, {
+		&cmd_struct_tune_set,
+		NULL
+	}
+};
+DEFINE_GROUP_COMMAND_TOKEN(tune);
author
#6 lore raw

Re: [PATCH 2/7] btrfs-progs: cmds: add "btrfs tune set" subcommand group

Quoted 37 lines · Anand Jain
Quoted 12 lines · Qu Wenruo

As the first step to convert btrfstune into "btrfs tune" subcommand group, this patch would add the following subcommand group:

  btrfs tune set <feature> [<device>]

For now the following features are supported:

- extref
- skinny-metadata
- no-holes
   All those are simple super block flags toggle.

Quoted 6 lines

- list-all
   This acts the same way as "mkfs.btrfs -O list-all", the difference is it would only list the supported features.
   (In the future, there will be "btrfs tune clear" subcommand, which would support a different set of features).

With this patchset, the syntax is structured as follows:

    $ btrfs tune --help
    usage: btrfs tune <command> <args>

      btrfs tune set <feature> [<device>]
          Set/enable specified feature for the unmounted filesystem btrfs tune clear <feature> [<device>]
          Clear/disable specified feature for the unmounted filesystem

   change various btrfs features

However, for consistency, I suggest the following syntax:

Quoted 8 lines

  set:

    $ btrfs tune <feature> /dev/sda

  clear:

    $ btrfs tune <feature> --clear /dev/sda

  list:

    $ btrfs tune --list-all

This can be confusing instead.

Remember set and clear have different supported feature set. E.g. we can not disable no-holes/extref features.

Thus here we want set/clear as subcommands, not the features name itself.

Furthermore, using <feature> as the subcommand wil later conflicts with other subcommand like "change-csum" and "change-fsid".

Quoted 4 lines

This syntax aligns with the:

    $ btrfs device scan --forget

In fact, the current "set/clear" follows the existing behavior, it's more clear if you scan a single device:

btrfs device scan <device_name>

As for btrfs tune it goes:

btrfs tune set <feature_name> <mnt>

We should put the variable parameter last, not between some fixed subcommand.

Thanks,
Qu

Quoted 387 lines

Thanks, Anand

Quoted 381 lines

---
  Documentation/btrfs-tune.rst |  40 ++++++ Documentation/btrfs.rst      |   5 +
  Documentation/conf.py        |   1 + Documentation/man-index.rst  |   1 +
  Makefile                     |   2 +-
  btrfs.c                      |   1 +
  cmds/commands.h              |   1 +
  cmds/tune.c                  | 227 +++++++++++++++++++++++++++++++++++

  8 files changed, 277 insertions(+), 1 deletion(-)

  create mode 100644 Documentation/btrfs-tune.rst
  create mode 100644 cmds/tune.c

Documentation/btrfs-tune.rst new file +21 -0
new file mode 100644
index 000000000000..827c92eadb72
@@ -0,0 +1,40 @@
+btrfs-tune(8)
+==================
+
+SYNOPSIS
+--------
+
+**btrfs tune** <subcommand> [<args>]
+
+DESCRIPTION
+-----------
+
+:command:`btrfs tune` is used to tweak various btrfs features on a
+unmounted filesystem.
+
+SUBCOMMAND
+-----------
+
+set <feature> [<device>]
+        Set/enable a feature.
+
+        If *feature* is `list-all`, all supported features would be 

listed, and
+    no *device* parameter is needed.

Documentation/btrfs-tune.rst new file +5 -0
+
+EXIT STATUS
+-----------
+
+**btrfs tune** returns a zero exit status if it succeeds. A non-zero 

value is

Documentation/btrfs-tune.rst new file +13 -0
+returned in case of failure.
+
+AVAILABILITY
+------------
+
+**btrfs** is part of btrfs-progs.  Please refer to the documentation at
+`https://btrfs.readthedocs.io <https://btrfs.readthedocs.io>`_.
+
+SEE ALSO
+--------
+
+:doc:`mkfs.btrfs`,
+``mount(8)``
Documentation/btrfs.rst +0 -0
index e878f158aaa1..5aea0d1a208c 100644
@@ -134,6 +134,10 @@ subvolume

      Create/delete/list/manage btrfs subvolume.
      See :doc:btrfs-subvolume for details.

Documentation/btrfs.rst +4 -0
+tune
+    Change various btrfs features.
+    See :doc:`btrfs-tune` for details.
+

  .. _man-btrfs8-standalone-tools:
  STANDALONE TOOLS

Documentation/btrfs.rst +0 -0
@@ -150,6 +154,7 @@ btrfs-convert

          in-place conversion from ext2/3/4 filesystems to btrfs btrfstune
          tweak some filesystem properties on a unmounted filesystem +    (will be replaced by btrfs-tune)
  btrfs-select-super
          rescue tool to overwrite primary superblock from a spare copy btrfs-find-root

Documentation/conf.py +1 -0
index 1025e10d7206..e0801bca4686 100644
@@ -66,6 +66,7 @@ man_pages = [
       ('btrfs-check', 'btrfs-check', 'check or repair a btrfs filesystem', '', 8),
       ('btrfs-balance', 'btrfs-balance', 'balance block groups on a btrfs filesystem', '', 8),
       ('btrfs-subvolume', 'btrfs-subvolume', 'manage btrfs subvolumes', '', 8),
+    ('btrfs-tune', 'btrfs-tune', 'tweak btrfs features', '', 8),
       ('btrfs-map-logical', 'btrfs-map-logical', 'map btrfs logical extent to physical extent', '', 8),
       ('btrfs', 'btrfs', 'a toolbox to manage btrfs filesystems', '', 8),
       ('mkfs.btrfs', 'mkfs.btrfs', 'create a btrfs filesystem', '', 8),
Documentation/man-index.rst +0 -0
index 36d45d2903ea..5fcd4cbc4bee 100644
@@ -28,6 +28,7 @@ Manual pages

     btrfs-select-super
     btrfs-send
     btrfs-subvolume
+   btrfs-tune
     btrfstune
     fsck.btrfs
     mkfs.btrfs

Makefile +1 -1
index f4feb1fff8e1..9857daaa42ac 100644
@@ -239,7 +239,7 @@ cmds_objects = cmds/subvolume.o 
 cmds/subvolume-list.o \
              cmds/rescue-super-recover.o \
              cmds/property.o cmds/filesystem-usage.o cmds/inspect-dump-tree.o \
              cmds/inspect-dump-super.o cmds/inspect-tree-stats.o cmds/filesystem-du.o \
-           cmds/reflink.o \
+           cmds/reflink.o cmds/tune.o \
              mkfs/common.o check/mode-common.o check/mode-lowmem.o \

             check/clear-cache.o

btrfs.c +1 -0
index 751f193ee2e0..c2dae0303ffe 100644
@@ -389,6 +389,7 @@ static const struct cmd_group btrfs_cmd_group = {
           &cmd_struct_scrub,
           &cmd_struct_send,
           &cmd_struct_subvolume,
+        &cmd_struct_tune,
           /* Help and version stay last */
           &cmd_struct_help,
cmds/commands.h +1 -0
index 5ab7c881f634..aebacd718a7b 100644
@@ -151,5 +151,6 @@ DECLARE_COMMAND(qgroup);
   DECLARE_COMMAND(replace);
   DECLARE_COMMAND(restore);
   DECLARE_COMMAND(rescue);
+DECLARE_COMMAND(tune);
   #endif
cmds/tune.c new file +100 -0
new file mode 100644
index 000000000000..92c7b9f1502c
@@ -0,0 +1,227 @@
+/*
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <unistd.h>
+#include "kerncompat.h"
+#include "cmds/commands.h"
+#include "common/help.h"
+#include "common/fsfeatures.h"
+#include "kernel-shared/messages.h"
+#include "kernel-shared/disk-io.h"
+#include "kernel-shared/transaction.h"
+
+static const char * const cmd_tune_set_usage[] = {
+    "btrfs tune set <feature> [<device>]",
+    "Set/enable specified feature for the unmounted filesystem",
+    "",
+    HELPINFO_INSERT_GLOBALS,
+    HELPINFO_INSERT_VERBOSE,
+    NULL,
+};
+
+static const struct btrfs_feature set_features[] = {
+    {
+        .name        = "extref",
+        .incompat_flag    = BTRFS_FEATURE_INCOMPAT_EXTENDED_IREF,
+        .sysfs_name    = "extended_iref",
+        VERSION_TO_STRING2(compat, 3,7),
+        VERSION_TO_STRING2(safe, 3,12),
+        VERSION_TO_STRING2(default, 3,12),
+        .desc        = "increased hardlink limit per file to 65536"
+    }, {
+        .name        = "skinny-metadata",
+        .incompat_flag    = BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA,
+        .sysfs_name    = "skinny_metadata",
+        VERSION_TO_STRING2(compat, 3,10),
+        VERSION_TO_STRING2(safe, 3,18),
+        VERSION_TO_STRING2(default, 3,18),
+        .desc        = "reduced-size metadata extent refs"
+    }, {
+        .name        = "no-holes",
+        .incompat_flag    = BTRFS_FEATURE_INCOMPAT_NO_HOLES,
+        .sysfs_name    = "no_holes",
+        VERSION_TO_STRING2(compat, 3,14),
+        VERSION_TO_STRING2(safe, 4,0),
+        VERSION_TO_STRING2(default, 5,15),
+        .desc        = "no explicit hole extents for files"
+    },
+    /* Keep this one last */
+    {
+        .name        = "list-all",
+        .runtime_flag    = BTRFS_FEATURE_RUNTIME_LIST_ALL,
+        .sysfs_name    = NULL,
+        VERSION_NULL(compat),
+        VERSION_NULL(safe),
+        VERSION_NULL(default),
+        .desc        = NULL
+    }
+};
+
+static void list_all_features(const char *prefix,
+                  const struct btrfs_feature *features,
+                  int nr_features)
+{
+    /* We should have at least one empty feature. */
+    ASSERT(nr_features > 1);
+
+    printf("features available to %s:\n", prefix);
+    for (int i = 0; i < nr_features - 1; i++) {
+        const struct btrfs_feature *feat = features + i;
+        const char *sep = "";
+
+        fprintf(stderr, "%-20s- %s (", feat->name, feat->desc);
+        if (feat->compat_ver) {
+            fprintf(stderr, "compat=%s", feat->compat_str);
+            sep = ", ";
+        }
+        if (feat->safe_ver) {
+            fprintf(stderr, "%ssafe=%s", sep, feat->safe_str);
+            sep = ", ";
+        }
+        if (feat->default_ver)
+            fprintf(stderr, "%sdefault=%s", sep, feat->default_str);
+        fprintf(stderr, ")\n");
+    }
+}
+
+static int check_features(const char *name, const struct 

btrfs_feature *features,
+              int nr_features)

cmds/tune.c new file +17 -0
+{
+    bool found = false;
+
+    for (int i = 0; i < nr_features; i++) {
+        const struct btrfs_feature *feat = features + i;
+
+        if (!strcmp(feat->name, name)) {
+            found = true;
+            break;
+        }
+    }
+    if (found)
+        return 0;
+    return -EINVAL;
+}
+
+static int set_super_incompat_flags(struct btrfs_fs_info *fs_info, 

u64 flags)

cmds/tune.c new file +19 -0
+{
+    struct btrfs_root *root = fs_info->tree_root;
+    struct btrfs_trans_handle *trans;
+    struct btrfs_super_block *disk_super;
+    u64 super_flags;
+    int ret;
+
+    disk_super = fs_info->super_copy;
+    super_flags = btrfs_super_incompat_flags(disk_super);
+    super_flags |= flags;
+    trans = btrfs_start_transaction(root, 1);
+    BUG_ON(IS_ERR(trans));
+    btrfs_set_super_incompat_flags(disk_super, super_flags);
+    ret = btrfs_commit_transaction(trans, root);
+
+    return ret;
+}
+
+static int cmd_tune_set(const struct cmd_struct *cmd, int argc, char 

**argv)

cmds/tune.c new file +25 -0
+{
+    struct btrfs_fs_info *fs_info;
+    struct open_ctree_args oca = { 0 };
+    char *feature;
+    char *path;
+    int ret = 0;
+
+    optind = 0;
+    while (1) {
+        int c = getopt(argc, argv, "");
+        if (c < 0)
+            break;
+
+        switch (c) {
+        default:
+            usage_unknown_option(cmd, argv);
+        }
+    }
+
+    if (check_argc_min(argc - optind, 1))
+        return 1;
+
+    feature = argv[optind];
+
+    if (check_features(feature, set_features, 

ARRAY_SIZE(set_features))) {
+        error("Unknown feature to set: %s", feature); +        return 1;
+    }
+    if (!strcmp(feature, "list-all")) {
+        list_all_features("set", set_features, ARRAY_SIZE(set_features));
+        return 0;
+    }

cmds/tune.c new file +15 -0
+
+    if (check_argc_exact(argc - optind, 2))
+        return 1;
+
+    path = argv[optind + 1];
+    oca.flags = OPEN_CTREE_WRITES;
+    oca.filename = path;
+    fs_info = open_ctree_fs_info(&oca);
+    if (!fs_info) {
+        error("failed to open btrfs");
+        ret = -EIO;
+        goto out;
+    }
+    /*
+     * For those 3 features, we only need to update the superblock to 

add
+     * the new feature flags.
+     */
+    if (!strcmp(feature, "extref") ||
+        !strcmp(feature, "skinny-metadata") ||
+        !strcmp(feature, "no-holes")) {
+        u64 incompat_flags = 0;

cmds/tune.c new file +29 -0
+
+        if (!strcmp(feature, "extref"))
+            incompat_flags |= BTRFS_FEATURE_INCOMPAT_EXTENDED_IREF;
+        if (!strcmp(feature, "skinny-metadata"))
+            incompat_flags |= BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA;
+        if (!strcmp(feature, "no-holes"))
+            incompat_flags |= BTRFS_FEATURE_INCOMPAT_NO_HOLES;
+        ret = set_super_incompat_flags(fs_info, incompat_flags);
+        if (ret < 0) {
+            errno = -ret;
+            error("failed to set feature '%s': %m", feature);
+        }
+        goto out;
+    }
+
+out:
+    if (fs_info)
+        close_ctree_fs_info(fs_info);
+    return !!ret;
+}
+
+static DEFINE_SIMPLE_COMMAND(tune_set, "set");
+
+static const char * const tune_cmd_group_usage[] = {
+    "btrfs tune <command> <args>",
+    NULL,
+};
+
+static const char tune_cmd_group_info[] = "change various btrfs 

features";

cmds/tune.c new file +8 -0
+
+static const struct cmd_group tune_cmd_group = {
+    tune_cmd_group_usage, tune_cmd_group_info, {
+        &cmd_struct_tune_set,
+        NULL
+    }
+};
+DEFINE_GROUP_COMMAND_TOKEN(tune);
#3 lore raw

Re: [PATCH 1/7] btrfs-progs: export btrfs_feature structure

In reply to Anand Jain
Quoted 9 lines · Anand Jain
Quoted 7 lines · Qu Wenruo

For the incoming "btrfs tune" subcommand, we will have different features supported by that subcommand.

Instead of bloating the runtime and mkfs features, here we just export btrfs_feature, so each subcommand can have their own definition of supported features.

Quoted 6 lines

Looks good.

Quoted 3 lines

And since we're here, also add needed headers for future users of "fsfeatures.h".

I don't see anything added, missed?

My bad, the line is not clear enough.

There are two added #include lines inside "fsfeatures.h", as during development I found if we just include "fsfeatures.h" by itself, there would be some missing definitions.

Thus we have the following lines:

common/fsfeatures.h +2 -0
@@ -19,7 +19,9 @@
 
 #include "kerncompat.h"
 #include <stdio.h>
+#include <linux/version.h>
 #include "kernel-lib/sizes.h"
+#include "kernel-shared/uapi/btrfs.h"
 

Thanks,
Qu

Quoted 144 lines

Thanks, Anand

Quoted 140 lines

---
  common/fsfeatures.c | 53 --------------------------------------------- common/fsfeatures.h | 50 ++++++++++++++++++++++++++++++++++++++++++

  2 files changed, 50 insertions(+), 53 deletions(-)
common/fsfeatures.c +0 -23
index 9ee392d3a8a6..f8eeea7695c1 100644
@@ -32,64 +32,11 @@
   #include "common/sysfs-utils.h"
   #include "common/messages.h"
-/*
- * Insert a root item for temporary tree root
- *
- * Only used in make_btrfs_v2().
- */
-#define VERSION_TO_STRING3(name, a,b,c)                \
-    .name ## _str = #a "." #b "." #c,            \
-    .name ## _ver = KERNEL_VERSION(a,b,c)
-#define VERSION_TO_STRING2(name, a,b)                \
-    .name ## _str = #a "." #b,                \
-    .name ## _ver = KERNEL_VERSION(a,b,0)
-#define VERSION_NULL(name)                    \
-    .name ## _str = NULL,                    \
-    .name ## _ver = 0
-
   enum feature_source {
       FS_FEATURES,
       RUNTIME_FEATURES,
   };
-/*
- * Feature stability status and versions: compat <= safe <= default
- */
-struct btrfs_feature {
-    const char *name;
-
-    /*
-     * At least one of the bit must be set in the following *_flag

member.
-     *
-     * For features like list-all and quota which don't have any -     * incompat/compat_ro bit set, it go to runtime_flag. -     */
-    u64 incompat_flag;
-    u64 compat_ro_flag;
-    u64 runtime_flag;

common/fsfeatures.c +0 -23
-
-    const char *sysfs_name;
-    /*
-     * Compatibility with kernel of given version. Filesystem can be
-     * mounted.
-     */
-    const char *compat_str;
-    u32 compat_ver;
-    /*
-     * Considered safe for use, but is not on by default, even if the
-     * kernel supports the feature.
-     */
-    const char *safe_str;
-    u32 safe_ver;
-    /*
-     * Considered safe for use and will be turned on by default if
-     * supported by the running kernel.
-     */
-    const char *default_str;
-    u32 default_ver;
-    const char *desc;
-};
-

  static const struct btrfs_feature mkfs_features[] = { {
          .name        = "mixed-bg",

common/fsfeatures.h +20 -0
index c4ab704862cd..c9fb489d2d79 100644
@@ -19,7 +19,9 @@
   #include "kerncompat.h"
   #include <stdio.h>
+#include <linux/version.h>
   #include "kernel-lib/sizes.h"
+#include "kernel-shared/uapi/btrfs.h"
   #define BTRFS_MKFS_DEFAULT_NODE_SIZE SZ_16K
@@ -43,6 +45,54 @@ struct btrfs_mkfs_features {
    */
   #define BTRFS_FEATURE_STRING_BUF_SIZE        (160)
+#define VERSION_TO_STRING3(name, a,b,c)                \
+    .name ## _str = #a "." #b "." #c,            \
+    .name ## _ver = KERNEL_VERSION(a,b,c)
+#define VERSION_TO_STRING2(name, a,b)                \
+    .name ## _str = #a "." #b,                \
+    .name ## _ver = KERNEL_VERSION(a,b,0)
+#define VERSION_NULL(name)                    \
+    .name ## _str = NULL,                    \
+    .name ## _ver = 0
+
+/*
+ * Feature stability status and versions: compat <= safe <= default
+ */
+struct btrfs_feature {
+    const char *name;
+
+    /*
+     * At least one of the bit must be set in the following *_flag

member.
+     *
+     * For features like list-all and quota which don't have any +     * incompat/compat_ro bit set, it go to runtime_flag. +     */
+    u64 incompat_flag;
+    u64 compat_ro_flag;
+    u64 runtime_flag;

common/fsfeatures.h +23 -0
+
+    const char *sysfs_name;
+    /*
+     * Compatibility with kernel of given version. Filesystem can be
+     * mounted.
+     */
+    const char *compat_str;
+    u32 compat_ver;
+    /*
+     * Considered safe for use, but is not on by default, even if the
+     * kernel supports the feature.
+     */
+    const char *safe_str;
+    u32 safe_ver;
+    /*
+     * Considered safe for use and will be turned on by default if
+     * supported by the running kernel.
+     */
+    const char *default_str;
+    u32 default_ver;
+    const char *desc;
+};
+

  static const struct btrfs_mkfs_features btrfs_mkfs_default_features = { .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE | BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE_VALID,

#15 lore raw
In reply to Qu Wenruo
Quoted 7 lines · Qu Wenruo

This is the first step to convert btrfstune functionality to "btrfs tune" subcommand group.

For now only binary features, aka set and clear, is supported, thus uuid and csum change is not yet implemented.
(Both need their own subcommand groups other than set/clear groups)

Quoted 9 lines

And even for set/clear, there is some changes to btrfstune:

- Merge seed feature into set/clear
To enable seeding, just go "btrfs tune set seed <device>".

- All supported features can be checked by "list-all" feature Please note that, "btrfs tune set list-all" and "btrfs tune clear list-all" will have different output.

As just a regular BTRFS user I would like to please ask you to consider having functions that may have consequences e.g. "dangerous" functions disabled until enabled.

In other words much like you need to arm a missile or flip a safety cover on a toggle switch

So if btrfs on every mount set a bit that disabled access to dangerous functions (such as the tune functionality you would have to explicitly enable it to be able to do scary stuff).

btrfs tune enable
btrfs tune whatever_needs_tuning=123
btrfs tune disable (manually disable scary functions)

And yes, I am aware that root users have powers to do scary stuff, but sometimes even root users need to be protected against free will.

Please consider something like the above...

author
#16 lore raw
Quoted 19 lines · waxhead
Quoted 7 lines · Qu Wenruo

This is the first step to convert btrfstune functionality to "btrfs tune" subcommand group.

For now only binary features, aka set and clear, is supported, thus uuid and csum change is not yet implemented.
(Both need their own subcommand groups other than set/clear groups)

Quoted 9 lines

And even for set/clear, there is some changes to btrfstune:

- Merge seed feature into set/clear
   To enable seeding, just go "btrfs tune set seed <device>".

- All supported features can be checked by "list-all" feature Please note that, "btrfs tune set list-all" and "btrfs tune clear list-all" will have different output.

As just a regular BTRFS user I would like to please ask you to consider having functions that may have consequences e.g. "dangerous" functions disabled until enabled.

Quoted 7 lines

In other words much like you need to arm a missile or flip a safety cover on a toggle switch

So if btrfs on every mount set a bit that disabled access to dangerous functions (such as the tune functionality you would have to explicitly enable it to be able to do scary stuff).

btrfs tune enable
btrfs tune whatever_needs_tuning=123
btrfs tune disable (manually disable scary functions)

We hide unsafe features behind the experimental flag of btrfs-progs, which is determined at compiling time.

Furthermore, all those features can be set/clear in "btrfs tune" must be done offline (unmounted).

Thus I think there is not much I can do to further improve the situation.

Thanks,
Qu

Quoted 4 lines

And yes, I am aware that root users have powers to do scary stuff, but sometimes even root users need to be protected against free will.

Please consider something like the above...

#7 lore raw

Re: [PATCH 2/7] btrfs-progs: cmds: add "btrfs tune set" subcommand group

In reply to Anand Jain
Quoted 31 lines · Anand Jain
Quoted 12 lines · Qu Wenruo

As the first step to convert btrfstune into "btrfs tune" subcommand group, this patch would add the following subcommand group:

btrfs tune set <feature> [<device>]

For now the following features are supported:

- extref
- skinny-metadata
- no-holes
All those are simple super block flags toggle.

Quoted 6 lines

- list-all
This acts the same way as "mkfs.btrfs -O list-all", the difference is it would only list the supported features.
(In the future, there will be "btrfs tune clear" subcommand, which
would support a different set of features).

With this patchset, the syntax is structured as follows:

    $ btrfs tune --help
    usage: btrfs tune <command> <args>

btrfs tune set <feature> [<device>]
Set/enable specified feature for the unmounted filesystem
btrfs tune clear <feature> [<device>]
Clear/disable specified feature for the unmounted filesystem

Quoted 5 lines

change various btrfs features

However, for consistency, I suggest the following syntax:

Consistency with what? If it's with btrfstune then no, the reason why the command is split into subcommands is because the command line parameters became unwieldy and what you suggest below copies that.

Quoted 10 lines
   set:
    $ btrfs tune <feature> /dev/sda

   clear:
    $ btrfs tune <feature> --clear /dev/sda

   list:
    $ btrfs tune --list-all
Quoted 3 lines

This syntax aligns with the:

    $ btrfs device scan --forget

The difference is that scan won't be extended much if ever. The 'tune' interface grows over time so we need properly structure that so we don't end in command line option mess.

As an anti-pattern I can mention the 'mdadm' utility, that comes from times when the commands were not without the initial "--" like we know from git etc, so the main commands like assemble are specified by -A or --assemble. What I find confusing is that it's just another option but means a main action. The command groups logically and visually encapsulate one action and can be possibly fine tuned by parameters without affecting other commands.

#17 lore raw
In reply to waxhead
Quoted 28 lines · waxhead
Quoted 7 lines · Qu Wenruo

This is the first step to convert btrfstune functionality to "btrfs tune" subcommand group.

For now only binary features, aka set and clear, is supported, thus uuid and csum change is not yet implemented.
(Both need their own subcommand groups other than set/clear groups)

Quoted 9 lines

And even for set/clear, there is some changes to btrfstune:

- Merge seed feature into set/clear
To enable seeding, just go "btrfs tune set seed <device>".

- All supported features can be checked by "list-all" feature Please note that, "btrfs tune set list-all" and "btrfs tune clear list-all" will have different output.

As just a regular BTRFS user I would like to please ask you to consider having functions that may have consequences e.g. "dangerous" functions disabled until enabled.

In other words much like you need to arm a missile or flip a safety cover on a toggle switch

So if btrfs on every mount set a bit that disabled access to dangerous functions (such as the tune functionality you would have to explicitly enable it to be able to do scary stuff).

Quoted 8 lines

btrfs tune enable
btrfs tune whatever_needs_tuning=123
btrfs tune disable (manually disable scary functions)

And yes, I am aware that root users have powers to do scary stuff, but sometimes even root users need to be protected against free will.

Please consider something like the above...

We can print a warning, add a delay, require a magic parameter. What you suggest above with the enable/disable commands would require to store the state somewhere so the next call 'tune whatever_needs_tuning' would work (but not another such command run from another terminal).

Not all tune actions are dangerous or irreversible but I agree that there's a lot of potential for damage in case someobody does not read the documentation or understand the consequences. The actions are meant to be one-time and with long term effects, like converting to new structures. Comparable to mkfs, there are checks against accidental overwrite of an existing filesystem, so at minimum we might require --force for each command.

#18 lore raw
In reply to Qu Wenruo
Quoted 6 lines · Qu Wenruo

This is the first step to convert btrfstune functionality to "btrfs tune" subcommand group.

For now only binary features, aka set and clear, is supported, thus uuid and csum change is not yet implemented.
(Both need their own subcommand groups other than set/clear groups)

There are the easy on/off features that can be added first but you also added the block-group-tree _conversion_. That's a different category and IMHO requires a separate command so we can add options to check or reset the unfinished state. And we also have the the conversion 'to' and 'from'.

My current list of 1st level command groups is:

- enable - for the simple on/off features (on)
- disable - dtto, (off)
- convert-to - for the BGT, checksum
- convert-from - subset where the back conversion like BGT may make sense
- analyze - go through the structures and identify features (all, subset)

We can add aliases like set for enable and clear for disable, the two conversion commands may not be necessary or even confusing when it's not clear which direction is the right one.

Actions:

- extref (on/off)
- skinny-metadata (on/off)
- no-holes (on)
- seeding (on/off)
- uuid rewrite (one time)
- metadata_uuid (set value)
- change checksum (convert-to)
- currently there's the simple quota, this may change

This should give examples of all categories of actions we might have in the future.

Quoted 8 lines

And even for set/clear, there is some changes to btrfstune:

- Merge seed feature into set/clear
To enable seeding, just go "btrfs tune set seed <device>".

- All supported features can be checked by "list-all" feature Please note that, "btrfs tune set list-all" and "btrfs tune clear list-all" will have different output.

As the help test works, we would get that for free if all the implemented actions are also command names, i.e. 'btrfs tune convert-to' would list 'block-group-tree'.

author
#19 lore raw
Quoted 14 lines · David Sterba
Quoted 6 lines · Qu Wenruo

This is the first step to convert btrfstune functionality to "btrfs tune" subcommand group.

For now only binary features, aka set and clear, is supported, thus uuid and csum change is not yet implemented.
(Both need their own subcommand groups other than set/clear groups)

There are the easy on/off features that can be added first but you also added the block-group-tree _conversion_. That's a different category and IMHO requires a separate command so we can add options to check or reset the unfinished state. And we also have the the conversion 'to' and 'from'.

Quoted 6 lines

My current list of 1st level command groups is:

- enable - for the simple on/off features (on)
- disable - dtto, (off)
- convert-to - for the BGT, checksum
- convert-from - subset where the back conversion like BGT may make sense

So your idea is, if an conversion is a long running one, and can be interrupted, then we need to go convert instead of simple on/off?

That makes sense, although my question is, does that makes any difference?

If we have an interrupted conversion, the only valid operation we can do is to finish the conversion. No rollback for most conversions.

Thus in the end, it's sill on/off, just with an extra interrupted status, but that status can only lead to the "on" state, thus it makes no difference to have an dedicated state for it.

Thanks,
Qu

Quoted 32 lines

- analyze - go through the structures and identify features (all, subset)

We can add aliases like set for enable and clear for disable, the two conversion commands may not be necessary or even confusing when it's not clear which direction is the right one.

Actions:

- extref (on/off)
- skinny-metadata (on/off)
- no-holes (on)
- seeding (on/off)
- uuid rewrite (one time)
- metadata_uuid (set value)
- change checksum (convert-to)
- currently there's the simple quota, this may change

This should give examples of all categories of actions we might have in the future.

Quoted 8 lines

And even for set/clear, there is some changes to btrfstune:

- Merge seed feature into set/clear
To enable seeding, just go "btrfs tune set seed <device>".

- All supported features can be checked by "list-all" feature Please note that, "btrfs tune set list-all" and "btrfs tune clear list-all" will have different output.

As the help test works, we would get that for free if all the implemented actions are also command names, i.e. 'btrfs tune convert-to' would list 'block-group-tree'.

#8 lore raw

Re: [PATCH 2/7] btrfs-progs: cmds: add "btrfs tune set" subcommand group

In reply to David Sterba
Quoted 52 lines · David Sterba
Quoted 31 lines · Anand Jain
Quoted 12 lines · Qu Wenruo

As the first step to convert btrfstune into "btrfs tune" subcommand group, this patch would add the following subcommand group:

btrfs tune set <feature> [<device>]

For now the following features are supported:

- extref
- skinny-metadata
- no-holes
All those are simple super block flags toggle.

Quoted 6 lines

- list-all
This acts the same way as "mkfs.btrfs -O list-all", the difference is it would only list the supported features.
(In the future, there will be "btrfs tune clear" subcommand, which
would support a different set of features).

With this patchset, the syntax is structured as follows:

     $ btrfs tune --help
     usage: btrfs tune <command> <args>

btrfs tune set <feature> [<device>]
Set/enable specified feature for the unmounted filesystem
btrfs tune clear <feature> [<device>]
Clear/disable specified feature for the unmounted filesystem

Quoted 5 lines

change various btrfs features

However, for consistency, I suggest the following syntax:

Consistency with what? If it's with btrfstune then no, the reason why the command is split into subcommands is because the command line parameters became unwieldy and what you suggest below copies that.

Quoted 10 lines
    set:
     $ btrfs tune <feature> /dev/sda

    clear:
     $ btrfs tune <feature> --clear /dev/sda

    list:
     $ btrfs tune --list-all
Quoted 7 lines
Quoted 3 lines

This syntax aligns with the:

     $ btrfs device scan --forget

The difference is that scan won't be extended much if ever. The 'tune' interface grows over time so we need properly structure that so we don't end in command line option mess.

My apologies for not responding earlier. When designing the 'btrfs tune' command, pls consider we may need '--noscan' and '--device' options at some point. For example:

$ btrfs tune metadata --device=/dev/sda --noscan /dev/sdb

This idea is inspired by the following command:

$ btrfs inspect-internal dump-tree --noscan /dev/sda1

OR

Since some of the commands perform a user-space system-wide scan for btrfs-block-devices, so global options like is also not a bad idea.

'--devices' (specify btrfs image files and continue scanning for
block devices across the system)

and

'--noscan' (to avoid a system-wide block device scan when needed).

Thx, Anand

Quoted 8 lines

As an anti-pattern I can mention the 'mdadm' utility, that comes from times when the commands were not without the initial "--" like we know from git etc, so the main commands like assemble are specified by -A or --assemble. What I find confusing is that it's just another option but means a main action. The command groups logically and visually encapsulate one action and can be possibly fine tuned by parameters without affecting other commands.