Btrfs: Implement raid0 when multiple devices are present

Signed-off-by: Chris Mason <chris.mason@oracle.com>

fs/btrfs/ctree.h

@@ -413,6 +413,7 @@ struct btrfs_csum_item {
 #define BTRFS_BLOCK_GROUP_DATA     (1 << 0)
 #define BTRFS_BLOCK_GROUP_SYSTEM   (1 << 1)
 #define BTRFS_BLOCK_GROUP_METADATA (1 << 2)
+#define BTRFS_BLOCK_GROUP_RAID0    (1 << 3)
 
 
 struct btrfs_block_group_item {
@@ -498,6 +499,8 @@ struct btrfs_fs_info {
 	u64 delalloc_bytes;
 	u64 last_alloc;
 	u64 last_data_alloc;
+	int extra_data_alloc_bits;
+	int extra_alloc_bits;
 };
 
 /*

fs/btrfs/disk-io.c

@@ -736,6 +736,8 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	fs_info->total_pinned = 0;
 	fs_info->last_alloc = 0;
 	fs_info->last_data_alloc = 0;
+	fs_info->extra_alloc_bits = 0;
+	fs_info->extra_data_alloc_bits = 0;
 
 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 	INIT_WORK(&fs_info->trans_work, btrfs_transaction_cleaner, fs_info);

fs/btrfs/extent-tree.c

@@ -172,7 +172,7 @@ struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
 
 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
 {
-	return (cache->flags & bits);
+	return (cache->flags & bits) == bits;
 }
 
 static int noinline find_search_start(struct btrfs_root *root,
@@ -1010,6 +1010,35 @@ static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 
 }
 
+static int update_space_info(struct btrfs_fs_info *info, u64 flags,
+			     u64 total_bytes, u64 bytes_used,
+			     struct btrfs_space_info **space_info)
+{
+	struct btrfs_space_info *found;
+
+	found = __find_space_info(info, flags);
+	if (found) {
+		found->total_bytes += total_bytes;
+		found->bytes_used += bytes_used;
+		WARN_ON(found->total_bytes < found->bytes_used);
+		*space_info = found;
+		return 0;
+	}
+	found = kmalloc(sizeof(*found), GFP_NOFS);
+	if (!found)
+		return -ENOMEM;
+
+	list_add(&found->list, &info->space_info);
+	found->flags = flags;
+	found->total_bytes = total_bytes;
+	found->bytes_used = bytes_used;
+	found->bytes_pinned = 0;
+	found->full = 0;
+	*space_info = found;
+	return 0;
+}
+
+
 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
 			  struct btrfs_root *extent_root, u64 alloc_bytes,
 			  u64 flags)
@@ -1021,6 +1050,11 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans,
 	int ret;
 
 	space_info = __find_space_info(extent_root->fs_info, flags);
+	if (!space_info) {
+		ret = update_space_info(extent_root->fs_info, flags,
+					0, 0, &space_info);
+		BUG_ON(ret);
+	}
 	BUG_ON(!space_info);
 
 	if (space_info->full)
@@ -1044,6 +1078,17 @@ printk("space info full %Lu\n", flags);
 		     extent_root->fs_info->chunk_root->root_key.objectid,
 		     start, num_bytes);
 	BUG_ON(ret);
+
+	if (flags & BTRFS_BLOCK_GROUP_RAID0) {
+		if (flags & BTRFS_BLOCK_GROUP_DATA) {
+			extent_root->fs_info->extra_data_alloc_bits =
+				BTRFS_BLOCK_GROUP_RAID0;
+		}
+		if (flags & BTRFS_BLOCK_GROUP_METADATA) {
+			extent_root->fs_info->extra_alloc_bits =
+				BTRFS_BLOCK_GROUP_RAID0;
+		}
+	}
 	return 0;
 }
 
@@ -1655,24 +1700,31 @@ int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
 	struct btrfs_extent_ref *ref;
 	struct btrfs_path *path;
 	struct btrfs_key keys[2];
+	int extra_chunk_alloc_bits = 0;
 
 	if (data) {
-		data = BTRFS_BLOCK_GROUP_DATA;
+		data = BTRFS_BLOCK_GROUP_DATA | info->extra_data_alloc_bits;
 	} else if (root == root->fs_info->chunk_root) {
 		data = BTRFS_BLOCK_GROUP_SYSTEM;
 	} else {
-		data = BTRFS_BLOCK_GROUP_METADATA;
+		data = BTRFS_BLOCK_GROUP_METADATA | info->extra_alloc_bits;
 	}
+	if (btrfs_super_num_devices(&info->super_copy) > 1 &&
+	    !(data & BTRFS_BLOCK_GROUP_SYSTEM))
+		extra_chunk_alloc_bits = BTRFS_BLOCK_GROUP_RAID0;
 
 	if (root->ref_cows) {
-		if (data != BTRFS_BLOCK_GROUP_METADATA) {
+		if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
 			ret = do_chunk_alloc(trans, root->fs_info->extent_root,
 					     2 * 1024 * 1024,
-					     BTRFS_BLOCK_GROUP_METADATA);
+					     BTRFS_BLOCK_GROUP_METADATA |
+					     info->extra_alloc_bits |
+					     extra_chunk_alloc_bits);
 			BUG_ON(ret);
 		}
 		ret = do_chunk_alloc(trans, root->fs_info->extent_root,
-				     num_bytes + 2 * 1024 * 1024, data);
+				     num_bytes + 2 * 1024 * 1024, data |
+				     extra_chunk_alloc_bits);
 		BUG_ON(ret);
 	}
 
@@ -2627,34 +2679,6 @@ error:
 	return ret;
 }
 
-static int update_space_info(struct btrfs_fs_info *info, u64 flags,
-			     u64 total_bytes, u64 bytes_used,
-			     struct btrfs_space_info **space_info)
-{
-	struct btrfs_space_info *found;
-
-	found = __find_space_info(info, flags);
-	if (found) {
-		found->total_bytes += total_bytes;
-		found->bytes_used += bytes_used;
-		WARN_ON(found->total_bytes < found->bytes_used);
-		*space_info = found;
-		return 0;
-	}
-	found = kmalloc(sizeof(*found), GFP_NOFS);
-	if (!found)
-		return -ENOMEM;
-
-	list_add(&found->list, &info->space_info);
-	found->flags = flags;
-	found->total_bytes = total_bytes;
-	found->bytes_used = bytes_used;
-	found->bytes_pinned = 0;
-	found->full = 0;
-	*space_info = found;
-	return 0;
-}
-
 int btrfs_read_block_groups(struct btrfs_root *root)
 {
 	struct btrfs_path *path;
@@ -2712,6 +2736,16 @@ int btrfs_read_block_groups(struct btrfs_root *root)
 		} else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
 			bit = BLOCK_GROUP_METADATA;
 		}
+		if (cache->flags & BTRFS_BLOCK_GROUP_RAID0) {
+			if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
+				info->extra_data_alloc_bits =
+					BTRFS_BLOCK_GROUP_RAID0;
+			}
+			if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
+				info->extra_alloc_bits =
+					BTRFS_BLOCK_GROUP_RAID0;
+			}
+		}
 
 		ret = update_space_info(info, cache->flags, found_key.offset,
 					btrfs_block_group_used(&cache->item),

fs/btrfs/inode.c

@@ -317,8 +317,6 @@ int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
 	map_length = length;
 	ret = btrfs_map_block(map_tree, logical, &physical, &map_length, &dev);
 	if (map_length < length + size) {
-		printk("merge bio hook logical %Lu bio len %Lu physical %Lu "
-		       "len %Lu\n", logical, length, physical, map_length);
 		return 1;
 	}
 	return 0;

fs/btrfs/volumes.c

@@ -18,6 +18,7 @@
 #include <linux/sched.h>
 #include <linux/bio.h>
 #include <linux/buffer_head.h>
+#include <asm/div64.h>
 #include "ctree.h"
 #include "extent_map.h"
 #include "disk-io.h"
@@ -25,10 +26,24 @@
 #include "print-tree.h"
 #include "volumes.h"
 
-struct map_lookup {
+struct stripe {
 	struct btrfs_device *dev;
 	u64 physical;
 };
+
+struct map_lookup {
+	u64 type;
+	int io_align;
+	int io_width;
+	int stripe_len;
+	int sector_size;
+	int num_stripes;
+	struct stripe stripes[];
+};
+
+#define map_lookup_size(n) (sizeof(struct map_lookup) + \
+			    (sizeof(struct stripe) * (n)))
+
 static DEFINE_MUTEX(uuid_mutex);
 static LIST_HEAD(fs_uuids);
 
@@ -592,6 +607,7 @@ int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
 		      u64 *num_bytes, u64 type)
 {
 	u64 dev_offset;
+	struct btrfs_fs_info *info = extent_root->fs_info;
 	struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
 	struct btrfs_stripe *stripes;
 	struct btrfs_device *device = NULL;
@@ -610,10 +626,18 @@ int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
 	int looped = 0;
 	int ret;
 	int index;
+	int stripe_len = 64 * 1024;
 	struct btrfs_key key;
 
 	if (list_empty(dev_list))
 		return -ENOSPC;
+
+	if (type & BTRFS_BLOCK_GROUP_RAID0)
+		num_stripes = btrfs_super_num_devices(&info->super_copy);
+	if (type & BTRFS_BLOCK_GROUP_DATA)
+		stripe_len = 64 * 1024;
+	if (type & (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_SYSTEM))
+		stripe_len = 32 * 1024;
 again:
 	INIT_LIST_HEAD(&private_devs);
 	cur = dev_list->next;
@@ -650,9 +674,15 @@ again:
 	if (!chunk)
 		return -ENOMEM;
 
+	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
+	if (!map) {
+		kfree(chunk);
+		return -ENOMEM;
+	}
+
 	stripes = &chunk->stripe;
 
-	*num_bytes = calc_size;
+	*num_bytes = calc_size * num_stripes;
 	index = 0;
 	while(index < num_stripes) {
 		BUG_ON(list_empty(&private_devs));
@@ -669,6 +699,8 @@ printk("alloc chunk size %Lu from dev %Lu\n", calc_size, device->devid);
 		ret = btrfs_update_device(trans, device);
 		BUG_ON(ret);
 
+		map->stripes[index].dev = device;
+		map->stripes[index].physical = dev_offset;
 		btrfs_set_stack_stripe_devid(stripes + index, device->devid);
 		btrfs_set_stack_stripe_offset(stripes + index, dev_offset);
 		physical = dev_offset;
@@ -680,12 +712,18 @@ printk("alloc chunk size %Lu from dev %Lu\n", calc_size, device->devid);
 	key.offset = *num_bytes;
 	key.type = BTRFS_CHUNK_ITEM_KEY;
 	btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
-	btrfs_set_stack_chunk_stripe_len(chunk, 64 * 1024);
+	btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
 	btrfs_set_stack_chunk_type(chunk, type);
 	btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
-	btrfs_set_stack_chunk_io_align(chunk, extent_root->sectorsize);
+	btrfs_set_stack_chunk_io_align(chunk, stripe_len);
-	btrfs_set_stack_chunk_io_width(chunk, extent_root->sectorsize);
+	btrfs_set_stack_chunk_io_width(chunk, stripe_len);
 	btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
+	map->sector_size = extent_root->sectorsize;
+	map->stripe_len = stripe_len;
+	map->io_align = stripe_len;
+	map->io_width = stripe_len;
+	map->type = type;
+	map->num_stripes = num_stripes;
 
 	ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
 				btrfs_chunk_item_size(num_stripes));
@@ -695,25 +733,11 @@ printk("alloc chunk size %Lu from dev %Lu\n", calc_size, device->devid);
 	em = alloc_extent_map(GFP_NOFS);
 	if (!em)
 		return -ENOMEM;
-	map = kmalloc(sizeof(*map), GFP_NOFS);
-	if (!map) {
-		free_extent_map(em);
-		return -ENOMEM;
-	}
-
 	em->bdev = (struct block_device *)map;
 	em->start = key.objectid;
 	em->len = key.offset;
 	em->block_start = 0;
 
-	map->physical = physical;
-	map->dev = device;
-
-	if (!map->dev) {
-		kfree(map);
-		free_extent_map(em);
-		return -EIO;
-	}
 	kfree(chunk);
 
 	em_tree = &extent_root->fs_info->mapping_tree.map_tree;
@@ -758,6 +782,9 @@ int btrfs_map_block(struct btrfs_mapping_tree *map_tree,
 	struct map_lookup *map;
 	struct extent_map_tree *em_tree = &map_tree->map_tree;
 	u64 offset;
+	u64 stripe_offset;
+	u64 stripe_nr;
+	int stripe_index;
 
 
 	spin_lock(&em_tree->lock);
@@ -767,9 +794,40 @@ int btrfs_map_block(struct btrfs_mapping_tree *map_tree,
 	BUG_ON(em->start > logical || em->start + em->len < logical);
 	map = (struct map_lookup *)em->bdev;
 	offset = logical - em->start;
-	*phys = map->physical + offset;
+
-	*length = em->len - offset;
+	stripe_nr = offset;
-	*dev = map->dev;
+	/*
+	 * stripe_nr counts the total number of stripes we have to stride
+	 * to get to this block
+	 */
+	do_div(stripe_nr, map->stripe_len);
+
+	stripe_offset = stripe_nr * map->stripe_len;
+	BUG_ON(offset < stripe_offset);
+
+	/* stripe_offset is the offset of this block in its stripe*/
+	stripe_offset = offset - stripe_offset;
+
+	/*
+	 * after this do_div call, stripe_nr is the number of stripes
+	 * on this device we have to walk to find the data, and
+	 * stripe_index is the number of our device in the stripe array
+	 */
+	stripe_index = do_div(stripe_nr, map->num_stripes);
+
+	BUG_ON(stripe_index >= map->num_stripes);
+
+	*phys = map->stripes[stripe_index].physical + stripe_offset +
+		stripe_nr * map->stripe_len;
+
+	if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
+		/* we limit the length of each bio to what fits in a stripe */
+		*length = min_t(u64, em->len - offset,
+			      map->stripe_len - stripe_offset);
+	} else {
+		*length = em->len - offset;
+	}
+	*dev = map->stripes[stripe_index].dev;
 	free_extent_map(em);
 	spin_unlock(&em_tree->lock);
 	return 0;
@@ -822,7 +880,9 @@ static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
 	u64 logical;
 	u64 length;
 	u64 devid;
+	int num_stripes;
 	int ret;
+	int i;
 
 	logical = key->objectid;
 	length = key->offset;
@@ -846,7 +906,8 @@ static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
 	em = alloc_extent_map(GFP_NOFS);
 	if (!em)
 		return -ENOMEM;
-	map = kmalloc(sizeof(*map), GFP_NOFS);
+	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
+	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
 	if (!map) {
 		free_extent_map(em);
 		return -ENOMEM;
@@ -857,13 +918,22 @@ static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
 	em->len = length;
 	em->block_start = 0;
 
-	map->physical = btrfs_stripe_offset_nr(leaf, chunk, 0);
+	map->num_stripes = num_stripes;
-	devid = btrfs_stripe_devid_nr(leaf, chunk, 0);
+	map->io_width = btrfs_chunk_io_width(leaf, chunk);
-	map->dev = btrfs_find_device(root, devid);
+	map->io_align = btrfs_chunk_io_align(leaf, chunk);
-	if (!map->dev) {
+	map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
-		kfree(map);
+	map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
-		free_extent_map(em);
+	map->type = btrfs_chunk_type(leaf, chunk);
-		return -EIO;
+	for (i = 0; i < num_stripes; i++) {
+		map->stripes[i].physical =
+			btrfs_stripe_offset_nr(leaf, chunk, i);
+		devid = btrfs_stripe_devid_nr(leaf, chunk, i);
+		map->stripes[i].dev = btrfs_find_device(root, devid);
+		if (!map->stripes[i].dev) {
+			kfree(map);
+			free_extent_map(em);
+			return -EIO;
+		}
 	}
 
 	spin_lock(&map_tree->map_tree.lock);