1 files changed, 142 insertions, 101 deletions

diff --git a/fs/btrfs/scrub.c b/fs/btrfs/scrub.c
index 4cae41bd6de0..b877203f1dc5 100644
--- a/fs/btrfs/scrub.c
+++ b/fs/btrfs/scrub.c
@@ -43,9 +43,20 @@ struct scrub_ctx;
 /*
  * The following value only influences the performance.
  *
- * This determines the batch size for stripe submitted in one go.
+ * This detemines how many stripes would be submitted in one go,
+ * which is 512KiB (BTRFS_STRIPE_LEN * SCRUB_STRIPES_PER_GROUP).
  */
-#define SCRUB_STRIPES_PER_SCTX	8	/* That would be 8 64K stripe per-device. */
+#define SCRUB_STRIPES_PER_GROUP		8
+
+/*
+ * How many groups we have for each sctx.
+ *
+ * This would be 8M per device, the same value as the old scrub in-flight bios
+ * size limit.
+ */
+#define SCRUB_GROUPS_PER_SCTX		16
+
+#define SCRUB_TOTAL_STRIPES		(SCRUB_GROUPS_PER_SCTX * SCRUB_STRIPES_PER_GROUP)
 
 /*
  * The following value times PAGE_SIZE needs to be large enough to match the
@@ -172,9 +183,11 @@ struct scrub_stripe {
 };
 
 struct scrub_ctx {
-	struct scrub_stripe	stripes[SCRUB_STRIPES_PER_SCTX];
+	struct scrub_stripe	stripes[SCRUB_TOTAL_STRIPES];
 	struct scrub_stripe	*raid56_data_stripes;
 	struct btrfs_fs_info	*fs_info;
+	struct btrfs_path	extent_path;
+	struct btrfs_path	csum_path;
 	int			first_free;
 	int			cur_stripe;
 	atomic_t		cancel_req;
@@ -315,10 +328,10 @@ static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx)
 	if (!sctx)
 		return;
 
-	for (i = 0; i < SCRUB_STRIPES_PER_SCTX; i++)
+	for (i = 0; i < SCRUB_TOTAL_STRIPES; i++)
 		release_scrub_stripe(&sctx->stripes[i]);
 
-	kfree(sctx);
+	kvfree(sctx);
 }
 
 static void scrub_put_ctx(struct scrub_ctx *sctx)
@@ -333,13 +346,20 @@ static noinline_for_stack struct scrub_ctx *scrub_setup_ctx(
 	struct scrub_ctx *sctx;
 	int		i;
 
-	sctx = kzalloc(sizeof(*sctx), GFP_KERNEL);
+	/* Since sctx has inline 128 stripes, it can go beyond 64K easily.  Use
+	 * kvzalloc().
+	 */
+	sctx = kvzalloc(sizeof(*sctx), GFP_KERNEL);
 	if (!sctx)
 		goto nomem;
 	refcount_set(&sctx->refs, 1);
 	sctx->is_dev_replace = is_dev_replace;
 	sctx->fs_info = fs_info;
-	for (i = 0; i < SCRUB_STRIPES_PER_SCTX; i++) {
+	sctx->extent_path.search_commit_root = 1;
+	sctx->extent_path.skip_locking = 1;
+	sctx->csum_path.search_commit_root = 1;
+	sctx->csum_path.skip_locking = 1;
+	for (i = 0; i < SCRUB_TOTAL_STRIPES; i++) {
 		int ret;
 
 		ret = init_scrub_stripe(fs_info, &sctx->stripes[i]);
@@ -605,7 +625,8 @@ static void scrub_verify_one_metadata(struct scrub_stripe *stripe, int sector_nr
 			      btrfs_stack_header_bytenr(header), logical);
 		return;
 	}
-	if (memcmp(header->fsid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE) != 0) {
+	if (memcmp(header->fsid, fs_info->fs_devices->metadata_uuid,
+		   BTRFS_FSID_SIZE) != 0) {
 		bitmap_set(&stripe->meta_error_bitmap, sector_nr, sectors_per_tree);
 		bitmap_set(&stripe->error_bitmap, sector_nr, sectors_per_tree);
 		btrfs_warn_rl(fs_info,
@@ -969,6 +990,9 @@ skip:
 	spin_unlock(&sctx->stat_lock);
 }
 
+static void scrub_write_sectors(struct scrub_ctx *sctx, struct scrub_stripe *stripe,
+				unsigned long write_bitmap, bool dev_replace);
+
 /*
  * The main entrance for all read related scrub work, including:
  *
@@ -977,13 +1001,16 @@ skip:
  * - Go through the remaining mirrors and try to read as large blocksize as
  *   possible
  * - Go through all mirrors (including the failed mirror) sector-by-sector
+ * - Submit writeback for repaired sectors
  *
- * Writeback does not happen here, it needs extra synchronization.
+ * Writeback for dev-replace does not happen here, it needs extra
+ * synchronization for zoned devices.
  */
 static void scrub_stripe_read_repair_worker(struct work_struct *work)
 {
 	struct scrub_stripe *stripe = container_of(work, struct scrub_stripe, work);
-	struct btrfs_fs_info *fs_info = stripe->bg->fs_info;
+	struct scrub_ctx *sctx = stripe->sctx;
+	struct btrfs_fs_info *fs_info = sctx->fs_info;
 	int num_copies = btrfs_num_copies(fs_info, stripe->bg->start,
 					  stripe->bg->length);
 	int mirror;
@@ -1048,7 +1075,23 @@ static void scrub_stripe_read_repair_worker(struct work_struct *work)
 			goto out;
 	}
 out:
-	scrub_stripe_report_errors(stripe->sctx, stripe);
+	/*
+	 * Submit the repaired sectors.  For zoned case, we cannot do repair
+	 * in-place, but queue the bg to be relocated.
+	 */
+	if (btrfs_is_zoned(fs_info)) {
+		if (!bitmap_empty(&stripe->error_bitmap, stripe->nr_sectors))
+			btrfs_repair_one_zone(fs_info, sctx->stripes[0].bg->start);
+	} else if (!sctx->readonly) {
+		unsigned long repaired;
+
+		bitmap_andnot(&repaired, &stripe->init_error_bitmap,
+			      &stripe->error_bitmap, stripe->nr_sectors);
+		scrub_write_sectors(sctx, stripe, repaired, false);
+		wait_scrub_stripe_io(stripe);
+	}
+
+	scrub_stripe_report_errors(sctx, stripe);
 	set_bit(SCRUB_STRIPE_FLAG_REPAIR_DONE, &stripe->state);
 	wake_up(&stripe->repair_wait);
 }
@@ -1261,7 +1304,6 @@ static int get_raid56_logic_offset(u64 physical, int num,
 
 		/* Work out the disk rotation on this stripe-set */
 		rot = stripe_nr % map->num_stripes;
-		stripe_nr /= map->num_stripes;
 		/* calculate which stripe this data locates */
 		rot += i;
 		stripe_index = rot % map->num_stripes;
@@ -1467,6 +1509,8 @@ static void scrub_stripe_reset_bitmaps(struct scrub_stripe *stripe)
  * Return <0 for error.
  */
 static int scrub_find_fill_first_stripe(struct btrfs_block_group *bg,
+					struct btrfs_path *extent_path,
+					struct btrfs_path *csum_path,
 					struct btrfs_device *dev, u64 physical,
 					int mirror_num, u64 logical_start,
 					u32 logical_len,
@@ -1476,7 +1520,6 @@ static int scrub_find_fill_first_stripe(struct btrfs_block_group *bg,
 	struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bg->start);
 	struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bg->start);
 	const u64 logical_end = logical_start + logical_len;
-	struct btrfs_path path = { 0 };
 	u64 cur_logical = logical_start;
 	u64 stripe_end;
 	u64 extent_start;
@@ -1492,14 +1535,13 @@ static int scrub_find_fill_first_stripe(struct btrfs_block_group *bg,
 	/* The range must be inside the bg. */
 	ASSERT(logical_start >= bg->start && logical_end <= bg->start + bg->length);
 
-	path.search_commit_root = 1;
-	path.skip_locking = 1;
-
-	ret = find_first_extent_item(extent_root, &path, logical_start, logical_len);
+	ret = find_first_extent_item(extent_root, extent_path, logical_start,
+				     logical_len);
 	/* Either error or not found. */
 	if (ret)
 		goto out;
-	get_extent_info(&path, &extent_start, &extent_len, &extent_flags, &extent_gen);
+	get_extent_info(extent_path, &extent_start, &extent_len, &extent_flags,
+			&extent_gen);
 	if (extent_flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
 		stripe->nr_meta_extents++;
 	if (extent_flags & BTRFS_EXTENT_FLAG_DATA)
@@ -1527,7 +1569,7 @@ static int scrub_find_fill_first_stripe(struct btrfs_block_group *bg,
 
 	/* Fill the extent info for the remaining sectors. */
 	while (cur_logical <= stripe_end) {
-		ret = find_first_extent_item(extent_root, &path, cur_logical,
+		ret = find_first_extent_item(extent_root, extent_path, cur_logical,
 					     stripe_end - cur_logical + 1);
 		if (ret < 0)
 			goto out;
@@ -1535,7 +1577,7 @@ static int scrub_find_fill_first_stripe(struct btrfs_block_group *bg,
 			ret = 0;
 			break;
 		}
-		get_extent_info(&path, &extent_start, &extent_len,
+		get_extent_info(extent_path, &extent_start, &extent_len,
 				&extent_flags, &extent_gen);
 		if (extent_flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
 			stripe->nr_meta_extents++;
@@ -1560,9 +1602,9 @@ static int scrub_find_fill_first_stripe(struct btrfs_block_group *bg,
 		 */
 		ASSERT(BITS_PER_LONG >= BTRFS_STRIPE_LEN >> fs_info->sectorsize_bits);
 
-		ret = btrfs_lookup_csums_bitmap(csum_root, stripe->logical,
-						stripe_end, stripe->csums,
-						&csum_bitmap, true);
+		ret = btrfs_lookup_csums_bitmap(csum_root, csum_path,
+						stripe->logical, stripe_end,
+						stripe->csums, &csum_bitmap);
 		if (ret < 0)
 			goto out;
 		if (ret > 0)
@@ -1575,7 +1617,6 @@ static int scrub_find_fill_first_stripe(struct btrfs_block_group *bg,
 	}
 	set_bit(SCRUB_STRIPE_FLAG_INITIALIZED, &stripe->state);
 out:
-	btrfs_release_path(&path);
 	return ret;
 }
 
@@ -1653,6 +1694,28 @@ static bool stripe_has_metadata_error(struct scrub_stripe *stripe)
 	return false;
 }
 
+static void submit_initial_group_read(struct scrub_ctx *sctx,
+				      unsigned int first_slot,
+				      unsigned int nr_stripes)
+{
+	struct blk_plug plug;
+
+	ASSERT(first_slot < SCRUB_TOTAL_STRIPES);
+	ASSERT(first_slot + nr_stripes <= SCRUB_TOTAL_STRIPES);
+
+	scrub_throttle_dev_io(sctx, sctx->stripes[0].dev,
+			      btrfs_stripe_nr_to_offset(nr_stripes));
+	blk_start_plug(&plug);
+	for (int i = 0; i < nr_stripes; i++) {
+		struct scrub_stripe *stripe = &sctx->stripes[first_slot + i];
+
+		/* Those stripes should be initialized. */
+		ASSERT(test_bit(SCRUB_STRIPE_FLAG_INITIALIZED, &stripe->state));
+		scrub_submit_initial_read(sctx, stripe);
+	}
+	blk_finish_plug(&plug);
+}
+
 static int flush_scrub_stripes(struct scrub_ctx *sctx)
 {
 	struct btrfs_fs_info *fs_info = sctx->fs_info;
@@ -1665,11 +1728,11 @@ static int flush_scrub_stripes(struct scrub_ctx *sctx)
 
 	ASSERT(test_bit(SCRUB_STRIPE_FLAG_INITIALIZED, &sctx->stripes[0].state));
 
-	scrub_throttle_dev_io(sctx, sctx->stripes[0].dev,
-			      btrfs_stripe_nr_to_offset(nr_stripes));
-	for (int i = 0; i < nr_stripes; i++) {
-		stripe = &sctx->stripes[i];
-		scrub_submit_initial_read(sctx, stripe);
+	/* Submit the stripes which are populated but not submitted. */
+	if (nr_stripes % SCRUB_STRIPES_PER_GROUP) {
+		const int first_slot = round_down(nr_stripes, SCRUB_STRIPES_PER_GROUP);
+
+		submit_initial_group_read(sctx, first_slot, nr_stripes - first_slot);
 	}
 
 	for (int i = 0; i < nr_stripes; i++) {
@@ -1679,32 +1742,6 @@ static int flush_scrub_stripes(struct scrub_ctx *sctx)
 			   test_bit(SCRUB_STRIPE_FLAG_REPAIR_DONE, &stripe->state));
 	}
 
-	/*
-	 * Submit the repaired sectors.  For zoned case, we cannot do repair
-	 * in-place, but queue the bg to be relocated.
-	 */
-	if (btrfs_is_zoned(fs_info)) {
-		for (int i = 0; i < nr_stripes; i++) {
-			stripe = &sctx->stripes[i];
-
-			if (!bitmap_empty(&stripe->error_bitmap, stripe->nr_sectors)) {
-				btrfs_repair_one_zone(fs_info,
-						      sctx->stripes[0].bg->start);
-				break;
-			}
-		}
-	} else if (!sctx->readonly) {
-		for (int i = 0; i < nr_stripes; i++) {
-			unsigned long repaired;
-
-			stripe = &sctx->stripes[i];
-
-			bitmap_andnot(&repaired, &stripe->init_error_bitmap,
-				      &stripe->error_bitmap, stripe->nr_sectors);
-			scrub_write_sectors(sctx, stripe, repaired, false);
-		}
-	}
-
 	/* Submit for dev-replace. */
 	if (sctx->is_dev_replace) {
 		/*
@@ -1749,28 +1786,40 @@ static void raid56_scrub_wait_endio(struct bio *bio)
 
 static int queue_scrub_stripe(struct scrub_ctx *sctx, struct btrfs_block_group *bg,
 			      struct btrfs_device *dev, int mirror_num,
-			      u64 logical, u32 length, u64 physical)
+			      u64 logical, u32 length, u64 physical,
+			      u64 *found_logical_ret)
 {
 	struct scrub_stripe *stripe;
 	int ret;
 
-	/* No available slot, submit all stripes and wait for them. */
-	if (sctx->cur_stripe >= SCRUB_STRIPES_PER_SCTX) {
-		ret = flush_scrub_stripes(sctx);
-		if (ret < 0)
-			return ret;
-	}
+	/*
+	 * There should always be one slot left, as caller filling the last
+	 * slot should flush them all.
+	 */
+	ASSERT(sctx->cur_stripe < SCRUB_TOTAL_STRIPES);
 
 	stripe = &sctx->stripes[sctx->cur_stripe];
-
-	/* We can queue one stripe using the remaining slot. */
 	scrub_reset_stripe(stripe);
-	ret = scrub_find_fill_first_stripe(bg, dev, physical, mirror_num,
-					   logical, length, stripe);
+	ret = scrub_find_fill_first_stripe(bg, &sctx->extent_path,
+					   &sctx->csum_path, dev, physical,
+					   mirror_num, logical, length, stripe);
 	/* Either >0 as no more extents or <0 for error. */
 	if (ret)
 		return ret;
+	if (found_logical_ret)
+		*found_logical_ret = stripe->logical;
 	sctx->cur_stripe++;
+
+	/* We filled one group, submit it. */
+	if (sctx->cur_stripe % SCRUB_STRIPES_PER_GROUP == 0) {
+		const int first_slot = sctx->cur_stripe - SCRUB_STRIPES_PER_GROUP;
+
+		submit_initial_group_read(sctx, first_slot, SCRUB_STRIPES_PER_GROUP);
+	}
+
+	/* Last slot used, flush them all. */
+	if (sctx->cur_stripe == SCRUB_TOTAL_STRIPES)
+		return flush_scrub_stripes(sctx);
 	return 0;
 }
 
@@ -1784,6 +1833,8 @@ static int scrub_raid56_parity_stripe(struct scrub_ctx *sctx,
 	struct btrfs_fs_info *fs_info = sctx->fs_info;
 	struct btrfs_raid_bio *rbio;
 	struct btrfs_io_context *bioc = NULL;
+	struct btrfs_path extent_path = { 0 };
+	struct btrfs_path csum_path = { 0 };
 	struct bio *bio;
 	struct scrub_stripe *stripe;
 	bool all_empty = true;
@@ -1794,6 +1845,16 @@ static int scrub_raid56_parity_stripe(struct scrub_ctx *sctx,
 
 	ASSERT(sctx->raid56_data_stripes);
 
+	/*
+	 * For data stripe search, we cannot re-use the same extent/csum paths,
+	 * as the data stripe bytenr may be smaller than previous extent.  Thus
+	 * we have to use our own extent/csum paths.
+	 */
+	extent_path.search_commit_root = 1;
+	extent_path.skip_locking = 1;
+	csum_path.search_commit_root = 1;
+	csum_path.skip_locking = 1;
+
 	for (int i = 0; i < data_stripes; i++) {
 		int stripe_index;
 		int rot;
@@ -1808,7 +1869,7 @@ static int scrub_raid56_parity_stripe(struct scrub_ctx *sctx,
 
 		scrub_reset_stripe(stripe);
 		set_bit(SCRUB_STRIPE_FLAG_NO_REPORT, &stripe->state);
-		ret = scrub_find_fill_first_stripe(bg,
+		ret = scrub_find_fill_first_stripe(bg, &extent_path, &csum_path,
 				map->stripes[stripe_index].dev, physical, 1,
 				full_stripe_start + btrfs_stripe_nr_to_offset(i),
 				BTRFS_STRIPE_LEN, stripe);
@@ -1853,24 +1914,6 @@ static int scrub_raid56_parity_stripe(struct scrub_ctx *sctx,
 	/* For now, no zoned support for RAID56. */
 	ASSERT(!btrfs_is_zoned(sctx->fs_info));
 
-	/* Writeback for the repaired sectors. */
-	for (int i = 0; i < data_stripes; i++) {
-		unsigned long repaired;
-
-		stripe = &sctx->raid56_data_stripes[i];
-
-		bitmap_andnot(&repaired, &stripe->init_error_bitmap,
-			      &stripe->error_bitmap, stripe->nr_sectors);
-		scrub_write_sectors(sctx, stripe, repaired, false);
-	}
-
-	/* Wait for the above writebacks to finish. */
-	for (int i = 0; i < data_stripes; i++) {
-		stripe = &sctx->raid56_data_stripes[i];
-
-		wait_scrub_stripe_io(stripe);
-	}
-
 	/*
 	 * Now all data stripes are properly verified. Check if we have any
 	 * unrepaired, if so abort immediately or we could further corrupt the
@@ -1936,6 +1979,8 @@ static int scrub_raid56_parity_stripe(struct scrub_ctx *sctx,
 	bio_put(bio);
 	btrfs_bio_counter_dec(fs_info);
 
+	btrfs_release_path(&extent_path);
+	btrfs_release_path(&csum_path);
 out:
 	return ret;
 }
@@ -1957,18 +2002,15 @@ static int scrub_simple_mirror(struct scrub_ctx *sctx,
 {
 	struct btrfs_fs_info *fs_info = sctx->fs_info;
 	const u64 logical_end = logical_start + logical_length;
-	/* An artificial limit, inherit from old scrub behavior */
-	struct btrfs_path path = { 0 };
 	u64 cur_logical = logical_start;
 	int ret;
 
 	/* The range must be inside the bg */
 	ASSERT(logical_start >= bg->start && logical_end <= bg->start + bg->length);
 
-	path.search_commit_root = 1;
-	path.skip_locking = 1;
 	/* Go through each extent items inside the logical range */
 	while (cur_logical < logical_end) {
+		u64 found_logical;
 		u64 cur_physical = physical + cur_logical - logical_start;
 
 		/* Canceled? */
@@ -1993,7 +2035,7 @@ static int scrub_simple_mirror(struct scrub_ctx *sctx,
 
 		ret = queue_scrub_stripe(sctx, bg, device, mirror_num,
 					 cur_logical, logical_end - cur_logical,
-					 cur_physical);
+					 cur_physical, &found_logical);
 		if (ret > 0) {
 			/* No more extent, just update the accounting */
 			sctx->stat.last_physical = physical + logical_length;
@@ -2003,14 +2045,11 @@ static int scrub_simple_mirror(struct scrub_ctx *sctx,
 		if (ret < 0)
 			break;
 
-		ASSERT(sctx->cur_stripe > 0);
-		cur_logical = sctx->stripes[sctx->cur_stripe - 1].logical
-			      + BTRFS_STRIPE_LEN;
+		cur_logical = found_logical + BTRFS_STRIPE_LEN;
 
 		/* Don't hold CPU for too long time */
 		cond_resched();
 	}
-	btrfs_release_path(&path);
 	return ret;
 }
 
@@ -2108,6 +2147,9 @@ static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
 	u64 stripe_logical;
 	int stop_loop = 0;
 
+	/* Extent_path should be released by now. */
+	ASSERT(sctx->extent_path.nodes[0] == NULL);
+
 	scrub_blocked_if_needed(fs_info);
 
 	if (sctx->is_dev_replace &&
@@ -2226,6 +2268,9 @@ out:
 	ret2 = flush_scrub_stripes(sctx);
 	if (!ret)
 		ret = ret2;
+	btrfs_release_path(&sctx->extent_path);
+	btrfs_release_path(&sctx->csum_path);
+
 	if (sctx->raid56_data_stripes) {
 		for (int i = 0; i < nr_data_stripes(map); i++)
 			release_scrub_stripe(&sctx->raid56_data_stripes[i]);
@@ -2710,8 +2755,7 @@ static void scrub_workers_put(struct btrfs_fs_info *fs_info)
 /*
  * get a reference count on fs_info->scrub_workers. start worker if necessary
  */
-static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info,
-						int is_dev_replace)
+static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info)
 {
 	struct workqueue_struct *scrub_workers = NULL;
 	unsigned int flags = WQ_FREEZABLE | WQ_UNBOUND;
@@ -2721,10 +2765,7 @@ static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info,
 	if (refcount_inc_not_zero(&fs_info->scrub_workers_refcnt))
 		return 0;
 
-	if (is_dev_replace)
-		scrub_workers = alloc_ordered_workqueue("btrfs-scrub", flags);
-	else
-		scrub_workers = alloc_workqueue("btrfs-scrub", flags, max_active);
+	scrub_workers = alloc_workqueue("btrfs-scrub", flags, max_active);
 	if (!scrub_workers)
 		return -ENOMEM;
 
@@ -2776,7 +2817,7 @@ int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start,
 	if (IS_ERR(sctx))
 		return PTR_ERR(sctx);
 
-	ret = scrub_workers_get(fs_info, is_dev_replace);
+	ret = scrub_workers_get(fs_info);
 	if (ret)
 		goto out_free_ctx;