btrfs: introduce RAID1 round-robin read balancing

Add round-robin read policy that balances reads over available devices
(all RAID1 block group profiles). Switch to the next devices is done
after a number of blocks is read, which is 256K by default and is
configurable in sysfs.

The format is "round-robin:<min-contig-read>" and can be set in file

  /sys/fs/btrfs/FSID/read_policy

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>

fs/btrfs/sysfs.c

@@ -1305,7 +1305,12 @@ static ssize_t btrfs_temp_fsid_show(struct kobject *kobj,
 }
 BTRFS_ATTR(, temp_fsid, btrfs_temp_fsid_show);
 
-static const char * const btrfs_read_policy_name[] = { "pid" };
+static const char *btrfs_read_policy_name[] = {
+	"pid",
+#ifdef CONFIG_BTRFS_EXPERIMENTAL
+	"round-robin",
+#endif
+};
 
 static int btrfs_read_policy_to_enum(const char *str, s64 *value_ret)
 {
@@ -1355,6 +1360,12 @@ static ssize_t btrfs_read_policy_show(struct kobject *kobj,
 
 		ret += sysfs_emit_at(buf, ret, "%s", btrfs_read_policy_name[i]);
 
+#ifdef CONFIG_BTRFS_EXPERIMENTAL
+		if (i == BTRFS_READ_POLICY_RR)
+			ret += sysfs_emit_at(buf, ret, ":%u",
+					     READ_ONCE(fs_devices->rr_min_contig_read));
+#endif
+
 		if (i == policy)
 			ret += sysfs_emit_at(buf, ret, "]");
 	}
@@ -1376,6 +1387,41 @@ static ssize_t btrfs_read_policy_store(struct kobject *kobj,
 	if (index < 0)
 		return -EINVAL;
 
+#ifdef CONFIG_BTRFS_EXPERIMENTAL
+	/* If moving from RR then disable collecting fs stats. */
+	if (fs_devices->read_policy == BTRFS_READ_POLICY_RR && index != BTRFS_READ_POLICY_RR)
+		fs_devices->collect_fs_stats = false;
+
+	if (index == BTRFS_READ_POLICY_RR) {
+		if (value != -1) {
+			const u32 sectorsize = fs_devices->fs_info->sectorsize;
+
+			if (!IS_ALIGNED(value, sectorsize)) {
+				u64 temp_value = round_up(value, sectorsize);
+
+				btrfs_debug(fs_devices->fs_info,
+"read_policy: min contig read %lld should be multiple of sectorsize %u, rounded to %llu",
+					  value, sectorsize, temp_value);
+				value = temp_value;
+			}
+		} else {
+			value = BTRFS_DEFAULT_RR_MIN_CONTIG_READ;
+		}
+
+		if (index != READ_ONCE(fs_devices->read_policy) ||
+		    value != READ_ONCE(fs_devices->rr_min_contig_read)) {
+			WRITE_ONCE(fs_devices->read_policy, index);
+			WRITE_ONCE(fs_devices->rr_min_contig_read, value);
+
+			btrfs_info(fs_devices->fs_info, "read policy set to '%s:%lld'",
+				   btrfs_read_policy_name[index], value);
+		}
+
+		fs_devices->collect_fs_stats = true;
+
+		return len;
+	}
+#endif
 	if (index != READ_ONCE(fs_devices->read_policy)) {
 		WRITE_ONCE(fs_devices->read_policy, index);
 		btrfs_info(fs_devices->fs_info, "read policy set to '%s'",

fs/btrfs/volumes.c

@@ -1335,6 +1335,9 @@ static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
 	fs_devices->total_rw_bytes = 0;
 	fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
 	fs_devices->read_policy = BTRFS_READ_POLICY_PID;
+#ifdef CONFIG_BTRFS_EXPERIMENTAL
+	fs_devices->rr_min_contig_read = BTRFS_DEFAULT_RR_MIN_CONTIG_READ;
+#endif
 
 	return 0;
 }
@@ -5967,6 +5970,62 @@ unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
 	return len;
 }
 
+#ifdef CONFIG_BTRFS_EXPERIMENTAL
+struct stripe_mirror {
+	u64 devid;
+	int num;
+};
+
+static int btrfs_cmp_devid(const void *a, const void *b)
+{
+	const struct stripe_mirror *s1 = (const struct stripe_mirror *)a;
+	const struct stripe_mirror *s2 = (const struct stripe_mirror *)b;
+
+	if (s1->devid < s2->devid)
+		return -1;
+	if (s1->devid > s2->devid)
+		return 1;
+	return 0;
+}
+
+/*
+ * Select a stripe for reading using the round-robin algorithm.
+ *
+ *  1. Compute the read cycle as the total sectors read divided by the minimum
+ *     sectors per device.
+ *  2. Determine the stripe number for the current read by taking the modulus
+ *     of the read cycle with the total number of stripes:
+ *
+ *      stripe index = (total sectors / min sectors per dev) % num stripes
+ *
+ * The calculated stripe index is then used to select the corresponding device
+ * from the list of devices, which is ordered by devid.
+ */
+static int btrfs_read_rr(const struct btrfs_chunk_map *map, int first, int num_stripes)
+{
+	struct stripe_mirror stripes[BTRFS_RAID1_MAX_MIRRORS] = { 0 };
+	struct btrfs_device *device  = map->stripes[first].dev;
+	struct btrfs_fs_devices *fs_devices = device->fs_devices;
+	unsigned int read_cycle;
+	unsigned int total_reads;
+	unsigned int min_reads_per_dev;
+
+	total_reads = percpu_counter_sum(&fs_devices->stats_read_blocks);
+	min_reads_per_dev = READ_ONCE(fs_devices->rr_min_contig_read) >>
+			    fs_devices->fs_info->sectorsize_bits;
+
+	for (int index = 0, i = first; i < first + num_stripes; i++) {
+		stripes[index].devid = map->stripes[i].dev->devid;
+		stripes[index].num = i;
+		index++;
+	}
+	sort(stripes, num_stripes, sizeof(struct stripe_mirror), btrfs_cmp_devid, NULL);
+
+	read_cycle = total_reads / min_reads_per_dev;
+	return stripes[read_cycle % num_stripes].num;
+}
+#endif
+
 static int find_live_mirror(struct btrfs_fs_info *fs_info,
 			    struct btrfs_chunk_map *map, int first,
 			    int dev_replace_is_ongoing)
@@ -5996,6 +6055,11 @@ static int find_live_mirror(struct btrfs_fs_info *fs_info,
 	case BTRFS_READ_POLICY_PID:
 		preferred_mirror = first + (current->pid % num_stripes);
 		break;
+#ifdef CONFIG_BTRFS_EXPERIMENTAL
+	case BTRFS_READ_POLICY_RR:
+		preferred_mirror = btrfs_read_rr(map, first, num_stripes);
+		break;
+#endif
 	}
 
 	if (dev_replace_is_ongoing &&

fs/btrfs/volumes.h

@@ -296,13 +296,20 @@ enum btrfs_chunk_allocation_policy {
 	BTRFS_CHUNK_ALLOC_ZONED,
 };
 
+#define BTRFS_DEFAULT_RR_MIN_CONTIG_READ	(SZ_256K)
+/* Keep in sync with raid_attr table, current maximum is RAID1C4. */
+#define BTRFS_RAID1_MAX_MIRRORS			(4)
 /*
  * Read policies for mirrored block group profiles, read picks the stripe based
  * on these policies.
  */
 enum btrfs_read_policy {
 	/* Use process PID to choose the stripe */
 	BTRFS_READ_POLICY_PID,
+#ifdef CONFIG_BTRFS_EXPERIMENTAL
+	/* Balancing RAID1 reads across all striped devices (round-robin). */
+	BTRFS_READ_POLICY_RR,
+#endif
 	BTRFS_NR_READ_POLICY,
 };
 
@@ -436,6 +443,12 @@ struct btrfs_fs_devices {
 	enum btrfs_read_policy read_policy;
 
 #ifdef CONFIG_BTRFS_EXPERIMENTAL
+	/*
+	 * Minimum contiguous reads before switching to next device, the unit
+	 * is one block/sectorsize.
+	 */
+	u32 rr_min_contig_read;
+
 	/* Checksum mode - offload it or do it synchronously. */
 	enum btrfs_offload_csum_mode offload_csum_mode;
 #endif