btrfs: zoned: cache reported zone during mount

When mounting a device, we are reporting the zones twice: once for
checking the zone attributes in btrfs_get_dev_zone_info and once for
loading block groups' zone info in
btrfs_load_block_group_zone_info(). With a lot of block groups, that
leads to a lot of REPORT ZONE commands and slows down the mount
process.

This patch introduces a zone info cache in struct
btrfs_zoned_device_info. The cache is populated while in
btrfs_get_dev_zone_info() and used for
btrfs_load_block_group_zone_info() to reduce the number of REPORT ZONE
commands. The zone cache is then released after loading the block
groups, as it will not be much effective during the run time.

Benchmark: Mount an HDD with 57,007 block groups
Before patch: 171.368 seconds
After patch: 64.064 seconds

While it still takes a minute due to the slowness of loading all the
block groups, the patch reduces the mount time by 1/3.

Link: https://lore.kernel.org/linux-btrfs/CAHQ7scUiLtcTqZOMMY5kbWUBOhGRwKo6J6wYPT5WY+C=cD49nQ@mail.gmail.com/
Fixes: 5b316468983d ("btrfs: get zone information of zoned block devices")
CC: stable@vger.kernel.org
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>

1 files changed, 77 insertions, 9 deletions

diff --git a/fs/btrfs/zoned.c b/fs/btrfs/zoned.c
index 678a29469511..b06059a5db2a 100644
--- a/fs/btrfs/zoned.c
+++ b/fs/btrfs/zoned.c
@@ -5,6 +5,7 @@
 #include <linux/blkdev.h>
 #include <linux/sched/mm.h>
 #include <linux/atomic.h>
+#include <linux/vmalloc.h>
 #include "ctree.h"
 #include "volumes.h"
 #include "zoned.h"
@@ -213,6 +214,8 @@ static int emulate_report_zones(struct btrfs_device *device, u64 pos,
 static int btrfs_get_dev_zones(struct btrfs_device *device, u64 pos,
 			       struct blk_zone *zones, unsigned int *nr_zones)
 {
+	struct btrfs_zoned_device_info *zinfo = device->zone_info;
+	u32 zno;
 	int ret;
 
 	if (!*nr_zones)
@@ -224,6 +227,34 @@ static int btrfs_get_dev_zones(struct btrfs_device *device, u64 pos,
 		return 0;
 	}
 
+	/* Check cache */
+	if (zinfo->zone_cache) {
+		unsigned int i;
+
+		ASSERT(IS_ALIGNED(pos, zinfo->zone_size));
+		zno = pos >> zinfo->zone_size_shift;
+		/*
+		 * We cannot report zones beyond the zone end. So, it is OK to
+		 * cap *nr_zones to at the end.
+		 */
+		*nr_zones = min_t(u32, *nr_zones, zinfo->nr_zones - zno);
+
+		for (i = 0; i < *nr_zones; i++) {
+			struct blk_zone *zone_info;
+
+			zone_info = &zinfo->zone_cache[zno + i];
+			if (!zone_info->len)
+				break;
+		}
+
+		if (i == *nr_zones) {
+			/* Cache hit on all the zones */
+			memcpy(zones, zinfo->zone_cache + zno,
+			       sizeof(*zinfo->zone_cache) * *nr_zones);
+			return 0;
+		}
+	}
+
 	ret = blkdev_report_zones(device->bdev, pos >> SECTOR_SHIFT, *nr_zones,
 				  copy_zone_info_cb, zones);
 	if (ret < 0) {
@@ -237,6 +268,11 @@ static int btrfs_get_dev_zones(struct btrfs_device *device, u64 pos,
 	if (!ret)
 		return -EIO;
 
+	/* Populate cache */
+	if (zinfo->zone_cache)
+		memcpy(zinfo->zone_cache + zno, zones,
+		       sizeof(*zinfo->zone_cache) * *nr_zones);
+
 	return 0;
 }
 
@@ -300,7 +336,7 @@ int btrfs_get_dev_zone_info_all_devices(struct btrfs_fs_info *fs_info)
 		if (!device->bdev)
 			continue;
 
-		ret = btrfs_get_dev_zone_info(device);
+		ret = btrfs_get_dev_zone_info(device, true);
 		if (ret)
 			break;
 	}
@@ -309,7 +345,7 @@ int btrfs_get_dev_zone_info_all_devices(struct btrfs_fs_info *fs_info)
 	return ret;
 }
 
-int btrfs_get_dev_zone_info(struct btrfs_device *device)
+int btrfs_get_dev_zone_info(struct btrfs_device *device, bool populate_cache)
 {
 	struct btrfs_fs_info *fs_info = device->fs_info;
 	struct btrfs_zoned_device_info *zone_info = NULL;
@@ -339,6 +375,8 @@ int btrfs_get_dev_zone_info(struct btrfs_device *device)
 	if (!zone_info)
 		return -ENOMEM;
 
+	device->zone_info = zone_info;
+
 	if (!bdev_is_zoned(bdev)) {
 		if (!fs_info->zone_size) {
 			ret = calculate_emulated_zone_size(fs_info);
@@ -407,6 +445,23 @@ int btrfs_get_dev_zone_info(struct btrfs_device *device)
 		goto out;
 	}
 
+	/*
+	 * Enable zone cache only for a zoned device. On a non-zoned device, we
+	 * fill the zone info with emulated CONVENTIONAL zones, so no need to
+	 * use the cache.
+	 */
+	if (populate_cache && bdev_is_zoned(device->bdev)) {
+		zone_info->zone_cache = vzalloc(sizeof(struct blk_zone) *
+						zone_info->nr_zones);
+		if (!zone_info->zone_cache) {
+			btrfs_err_in_rcu(device->fs_info,
+				"zoned: failed to allocate zone cache for %s",
+				rcu_str_deref(device->name));
+			ret = -ENOMEM;
+			goto out;
+		}
+	}
+
 	/* Get zones type */
 	nactive = 0;
 	while (sector < nr_sectors) {
@@ -505,8 +560,6 @@ int btrfs_get_dev_zone_info(struct btrfs_device *device)
 
 	kfree(zones);
 
-	device->zone_info = zone_info;
-
 	switch (bdev_zoned_model(bdev)) {
 	case BLK_ZONED_HM:
 		model = "host-managed zoned";
@@ -539,11 +592,7 @@ int btrfs_get_dev_zone_info(struct btrfs_device *device)
 out:
 	kfree(zones);
 out_free_zone_info:
-	bitmap_free(zone_info->active_zones);
-	bitmap_free(zone_info->empty_zones);
-	bitmap_free(zone_info->seq_zones);
-	kfree(zone_info);
-	device->zone_info = NULL;
+	btrfs_destroy_dev_zone_info(device);
 
 	return ret;
 }
@@ -558,6 +607,7 @@ void btrfs_destroy_dev_zone_info(struct btrfs_device *device)
 	bitmap_free(zone_info->active_zones);
 	bitmap_free(zone_info->seq_zones);
 	bitmap_free(zone_info->empty_zones);
+	vfree(zone_info->zone_cache);
 	kfree(zone_info);
 	device->zone_info = NULL;
 }
@@ -1975,3 +2025,21 @@ void btrfs_clear_data_reloc_bg(struct btrfs_block_group *bg)
 		fs_info->data_reloc_bg = 0;
 	spin_unlock(&fs_info->relocation_bg_lock);
 }
+
+void btrfs_free_zone_cache(struct btrfs_fs_info *fs_info)
+{
+	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
+	struct btrfs_device *device;
+
+	if (!btrfs_is_zoned(fs_info))
+		return;
+
+	mutex_lock(&fs_devices->device_list_mutex);
+	list_for_each_entry(device, &fs_devices->devices, dev_list) {
+		if (device->zone_info) {
+			vfree(device->zone_info->zone_cache);
+			device->zone_info->zone_cache = NULL;
+		}
+	}
+	mutex_unlock(&fs_devices->device_list_mutex);
+}