[PATCH 0/2][V12] btrfs-progs: allocation_hint disk property

[PATCH 0/2][V12] btrfs-progs: allocation_hint disk property

[Goffredo Baroncelli]

From: Goffredo Baroncelli <kreijack@inwind.it>

This patches set is the userspace portion of the serie
"[PATCH V12] btrfs: allocation_hint".

Look this patches set for further information.

Changelog:
V12:
- removed support for major_minor: now even to query the allocation_hint
property requires root privileges.
V11:
- find the correct /sys/fs/btrfs/<UUID> directory using the
devinfo/major_minor sysfs file.
- update the man page to the .rst file format

G.Baroncelli

Goffredo Baroncelli (2):
  btrfs-progs: new "allocation_hint" property.
  Update man page for allocator_hint property.

 Documentation/btrfs-property.rst |  21 ++++
 cmds/property.c                  | 162 +++++++++++++++++++++++++++++++
 kernel-shared/ctree.h            |  13 +++
 3 files changed, 196 insertions(+)

-- 
2.35.1

[PATCH 1/2] btrfs-progs: new "allocation_hint" property.

[Goffredo Baroncelli]

From: Goffredo Baroncelli <kreijack@inwind.it>

Handle the property allocation_hint of a btrfs device. Below
an example of use:

$ # set a new value
$ sudo btrfs property set /dev/vde allocation_hint DATA_ONLY

$ # get the current value
$ sudo btrfs property get /dev/vde allocation_hint
devid=4, path=/dev/vde: allocation_hint=DATA_ONLY

The following values are availables:
- DATA_ONLY
- DATA_PREFERRED (default)
- METADATA_PREFERRED
- METADATA_ONLY

Root privileges are required.

Signed-off-by: Goffredo Baroncelli <kreijack@inwind.it>
---
 cmds/property.c       | 162 ++++++++++++++++++++++++++++++++++++++++++
 kernel-shared/ctree.h |  13 ++++
 2 files changed, 175 insertions(+)

diff --git a/cmds/property.c b/cmds/property.c
index b3ccc0ff..b35c8ed3 100644
--- a/cmds/property.c
+++ b/cmds/property.c
@@ -22,6 +22,7 @@
 #include <sys/ioctl.h>
 #include <sys/stat.h>
 #include <sys/xattr.h>
+#include <sys/sysmacros.h>
 #include <uuid/uuid.h>
 #include <btrfsutil.h>
 #include "cmds/commands.h"
@@ -30,6 +31,7 @@
 #include "common/open-utils.h"
 #include "common/utils.h"
 #include "common/help.h"
+#include "common/path-utils.h"
 
 #define XATTR_BTRFS_PREFIX     "btrfs."
 #define XATTR_BTRFS_PREFIX_LEN (sizeof(XATTR_BTRFS_PREFIX) - 1)
@@ -232,6 +234,160 @@ out:
 	return ret;
 }
 
+static int btrfs_find_devid_and_mnt(const char *devpath, int *devid,
+				    char *path, int maxpath)
+{
+	int ret, i, fd;
+	DIR *dir;
+	struct stat stdevpath;
+	struct btrfs_ioctl_fs_info_args fi_args;
+	struct btrfs_ioctl_dev_info_args dev_info;
+
+	ret = get_btrfs_mount(devpath, path, maxpath);
+	if (ret)
+		return ret;
+
+	fd = btrfs_open_dir(path, &dir, 1);
+	if (fd < 0)
+		return fd;
+
+	ret = stat(devpath, &stdevpath);
+	if (ret) {
+		error("cannot stat '%s'", devpath);
+		goto out;
+	}
+
+	ret = ioctl(fd, BTRFS_IOC_FS_INFO, &fi_args);
+	if (ret < 0) {
+		if (errno == EPERM)
+			return -errno;
+		error("cannot get filesystem info: %m");
+		ret = -10;
+		goto out;
+	}
+
+	for (i = 0 ; i <= fi_args.max_id ; i++) {
+		struct stat st;
+
+		memset(&dev_info, 0, sizeof(dev_info));
+		ret = get_device_info(fd, i, &dev_info);
+		if (ret == -ENODEV)
+			continue;
+		if (ret) {
+			error("cannot get info about device devid=%d", i);
+			goto out;
+		}
+
+		if (!dev_info.path)
+			/* missing devices */
+			continue;
+
+		ret = stat((char *)dev_info.path, &st);
+		if (ret) {
+			error("cannot stat '%s'", devpath);
+			goto out;
+		}
+
+		if (major(st.st_rdev) == major(stdevpath.st_rdev) &&
+		    minor(st.st_rdev) == minor(stdevpath.st_rdev)) {
+			*devid = dev_info.devid;
+			ret = 0;
+			goto out;
+		}
+	}
+
+	ret = -12;
+
+out:
+	close_file_or_dir(fd, dir);
+	return ret;
+}
+
+static int prop_allocation_hint(enum prop_object_type type,
+				const char *object,
+				const char *name,
+				const char *value,
+				bool force)
+{
+	int ret, devid, fd, fd2;
+	char path[PATH_MAX];
+	DIR *dir;
+	u8 fsid[BTRFS_UUID_SIZE];
+	char fsid_str[BTRFS_UUID_UNPARSED_SIZE];
+	char sysfs_file[PATH_MAX];
+	char filename[PATH_MAX];
+	char buf[1024];
+
+	ret = btrfs_find_devid_and_mnt(object, &devid, path, sizeof(path));
+	if (ret)
+		return -5;
+
+	fd = btrfs_open_dir(path, &dir, 1);
+	if (fd < 0)
+		return fd;
+
+	ret = get_fsid_fd(fd, fsid);
+	if (ret < 0)
+		goto out;
+
+	uuid_unparse(fsid, fsid_str);
+	sprintf(filename, "devinfo/%d/allocation_hint", devid);
+
+	/* build /sys/fs/btrfs/<UUID>/devinfo/<DEVID>/type */
+	ret = path_cat3_out(sysfs_file, "/sys/fs/btrfs", fsid_str, filename);
+	if (ret < 0)
+		goto out;
+
+	fd2 = open(sysfs_file, O_RDONLY);
+	if (fd2 < 0) {
+		error("'allocation_hint' property not available or accessible.");
+		ret = -errno;
+		goto out;
+	}
+
+	ret = read(fd2, buf, sizeof(buf) - 1);
+	close(fd2);
+	if (ret < 0) {
+		error("Unable to read the 'allocation_hint' property.");
+		ret = -errno;
+		goto out;
+	}
+
+	buf[sizeof(buf) - 1] = 0;
+
+	if (!value) {
+		/* READ */
+		printf("devid=%d, path=%s: allocation_hint=%s\n",
+			devid, object, buf);
+		ret = 0;
+	} else {
+		/* WRITE */
+		fd2 = open(sysfs_file, O_RDWR);
+		if (fd2 < 0) {
+			error("'allocation_hint' not accessible for updating.");
+			ret = -errno;
+			goto out;
+		}
+
+		ret = write(fd2, value, strlen(value));
+		close(fd2);
+		if (ret < 0) {
+			if (errno == EINVAL)
+				error("Invalid '%s' allocation_hint property.", value);
+			else
+				error("Unable to update 'allocation_hint' property.");
+			ret = -errno;
+			goto out;
+		}
+
+	}
+
+	ret = 0;
+out:
+	close_file_or_dir(fd, dir);
+	return ret;
+}
+
 const struct prop_handler prop_handlers[] = {
 	{
 		.name ="ro",
@@ -254,6 +410,12 @@ const struct prop_handler prop_handlers[] = {
 		.types = prop_object_inode,
 		.handler = prop_compression
 	},
+	{
+		.name = "allocation_hint",
+		.desc = "hint to store the data/metadata chunks",
+		.types = prop_object_dev,
+		.handler = prop_allocation_hint
+	},
 	{NULL, NULL, 0, 0, NULL}
 };
 
diff --git a/kernel-shared/ctree.h b/kernel-shared/ctree.h
index de5452f2..7bff3952 100644
--- a/kernel-shared/ctree.h
+++ b/kernel-shared/ctree.h
@@ -216,6 +216,19 @@ struct btrfs_mapping_tree {
 	struct cache_tree cache_tree;
 };
 
+/* btrfs chunk allocation hints */
+#define BTRFS_DEV_ALLOCATION_HINT_BIT_COUNT    2
+#define BTRFS_DEV_ALLOCATION_HINT_MASK ((1ULL << \
+	       BTRFS_DEV_ALLOCATION_HINT_BIT_COUNT) - 1)
+/* preferred metadata chunk, but data chunk allowed */
+#define BTRFS_DEV_ALLOCATION_HINT_METADATA_PREFERRED        (1ULL)
+/* only metadata chunk are allowed */
+#define BTRFS_DEV_ALLOCATION_HINT_METADATA_ONLY     (2ULL)
+/* only data chunk allowed */
+#define BTRFS_DEV_ALLOCATION_HINT_DATA_ONLY         (3ULL)
+/* preferred data chunk, but metadata chunk allowed */
+#define BTRFS_DEV_ALLOCATION_HINT_DATA_PREFERRED    (0ULL)
+
 #define BTRFS_UUID_SIZE 16
 struct btrfs_dev_item {
 	/* the internal btrfs device id */
-- 
2.35.1

[PATCH 2/2] Update man page for allocator_hint property.

[Goffredo Baroncelli]

From: Goffredo Baroncelli <kreijack@inwind.it>

Update the man page of the btrfs property subcommand to show the use
of the device property "allocation_hint".

Signed-off-by: Goffredo Baroncelli <kreijack@inwind.it>
---
 Documentation/btrfs-property.rst | 21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)

diff --git a/Documentation/btrfs-property.rst b/Documentation/btrfs-property.rst
index 5896faa2..155839fe 100644
--- a/Documentation/btrfs-property.rst
+++ b/Documentation/btrfs-property.rst
@@ -48,6 +48,27 @@ get [-t <type>] <object> [<name>]
         compression
                 compression algorithm set for an inode, possible values: *lzo*, *zlib*, *zstd*.
                 To disable compression use "" (empty string), *no* or *none*.
+        allocation_hint
+                a device property that instructs how and when the allocator should use a
+                block device.
+                Possible values are:
+
+                METADATA_PREFERRED
+                        the device has an higher priority when a new metadata
+                        chunk is allocated. Data chunk is allowed only if there is no other
+                        possibility.
+                METADATA_ONLY
+                        the device is used only for metadata chunk.
+                        Data chunk is never allowed.
+                DATA_PREFERRED (default)
+                        the device has an higher priority when a new data
+                        chunk is allocated. Metadata chunk is allowed only if there is no other
+                        possibility.
+                DATA_ONLY
+                        the device is used only for data chunk. Metadata chunk is never allowed.
+
+                The default is 'DATA_PREFERRED'; if all the disks have this setting the
+                allocator uses all of them with the same priority.
 
 list [-t <type>] <object>
         Lists available properties with their descriptions for the given object.
-- 
2.35.1