Btrfs: Dynamic chunk and block group allocation

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

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 1453d995fef67ccb66d4701c4dae01444c2c1dfd..a877105f5c4756449019f091a38e7acef584afa4 100644 (file)
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -431,9 +431,19 @@ struct btrfs_block_group_item {
        __le64 flags;
 } __attribute__ ((__packed__));
 
+struct btrfs_space_info {
+       u64 flags;
+       u64 total_bytes;
+       u64 bytes_used;
+       u64 bytes_pinned;
+       int full;
+       struct list_head list;
+};
+
 struct btrfs_block_group_cache {
        struct btrfs_key key;
        struct btrfs_block_group_item item;
+       struct btrfs_space_info *space_info;
        u64 pinned;
        u64 flags;
        int cached;
@@ -490,7 +500,7 @@ struct btrfs_fs_info {
        struct list_head dirty_cowonly_roots;
 
        struct list_head devices;
-       struct list_head *last_device;
+       struct list_head space_info;
        spinlock_t delalloc_lock;
        spinlock_t new_trans_lock;
        u64 delalloc_bytes;

diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 8e37fa120cc82f4d7006d740fd650866d7e0fded..2a239ae49f782ac38c0af1eb1af2fdae77f95ef2 100644 (file)
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -700,8 +700,8 @@ struct btrfs_root *open_ctree(struct super_block *sb)
        fs_info->dev_root = dev_root;
        INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
        INIT_LIST_HEAD(&fs_info->devices);
+       INIT_LIST_HEAD(&fs_info->space_info);
        btrfs_mapping_init(&fs_info->mapping_tree);
-       fs_info->last_device = &fs_info->devices;
        fs_info->sb = sb;
        fs_info->throttles = 0;
        fs_info->mount_opt = 0;

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 2cd957d6e8d8acb8243a289a7469730911bb1f9a..15082b1087beb553c43ed2605a4e2ab529b53e33 100644 (file)
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -36,6 +36,10 @@ static int finish_current_insert(struct btrfs_trans_handle *trans, struct
                                 btrfs_root *extent_root);
 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
                               btrfs_root *extent_root);
+int btrfs_make_block_group(struct btrfs_trans_handle *trans,
+                          struct btrfs_root *root, u64 bytes_used,
+                          u64 type, u64 chunk_tree, u64 chunk_objectid,
+                          u64 size);
 
 
 static int cache_block_group(struct btrfs_root *root,
@@ -168,16 +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)
 {
-       if ((bits & BLOCK_GROUP_DATA) &&
-           (cache->flags & BTRFS_BLOCK_GROUP_DATA))
-               return 1;
-       if ((bits & BLOCK_GROUP_METADATA) &&
-            (cache->flags & BTRFS_BLOCK_GROUP_METADATA))
-               return 1;
-       if ((bits & BLOCK_GROUP_SYSTEM) &&
-            (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM))
-               return 1;
-       return 0;
+       return (cache->flags & bits);
 }
 
 static int noinline find_search_start(struct btrfs_root *root,
@@ -276,6 +271,18 @@ static u64 div_factor(u64 num, int factor)
        return num;
 }
 
+static int block_group_state_bits(u64 flags)
+{
+       int bits = 0;
+       if (flags & BTRFS_BLOCK_GROUP_DATA)
+               bits |= BLOCK_GROUP_DATA;
+       if (flags & BTRFS_BLOCK_GROUP_METADATA)
+               bits |= BLOCK_GROUP_METADATA;
+       if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
+               bits |= BLOCK_GROUP_SYSTEM;
+       return bits;
+}
+
 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
                                                 struct btrfs_block_group_cache
                                                 *hint, u64 search_start,
@@ -304,7 +311,7 @@ struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
        if (!owner)
                factor = 8;
 
-       bit = data;
+       bit = block_group_state_bits(data);
 
        if (search_start && search_start < total_fs_bytes) {
                struct btrfs_block_group_cache *shint;
@@ -358,10 +365,15 @@ again:
                        free_check = cache->key.offset;
                else
                        free_check = div_factor(cache->key.offset, factor);
+
                if (used + cache->pinned < free_check) {
                        found_group = cache;
                        goto found;
                }
+               if (full_search) {
+                       printk("failed on cache %Lu used %Lu total %Lu\n",
+                              cache->key.objectid, used, cache->key.offset);
+               }
                cond_resched();
        }
        if (!full_search) {
@@ -983,6 +995,58 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
        return werr;
 }
 
+static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
+                                                 u64 flags)
+{
+       struct list_head *head = &info->space_info;
+       struct list_head *cur;
+       struct btrfs_space_info *found;
+       list_for_each(cur, head) {
+               found = list_entry(cur, struct btrfs_space_info, list);
+               if (found->flags == flags)
+                       return found;
+       }
+       return NULL;
+
+}
+
+static int do_chunk_alloc(struct btrfs_trans_handle *trans,
+                         struct btrfs_root *extent_root, u64 alloc_bytes,
+                         u64 flags)
+{
+       struct btrfs_space_info *space_info;
+       u64 thresh;
+       u64 start;
+       u64 num_bytes;
+       int ret;
+
+       space_info = __find_space_info(extent_root->fs_info, flags);
+       BUG_ON(!space_info);
+
+       if (space_info->full)
+               return 0;
+
+       thresh = div_factor(space_info->total_bytes, 7);
+       if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
+           thresh)
+               return 0;
+
+       ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
+       if (ret == -ENOSPC) {
+printk("space info full %Lu\n", flags);
+               space_info->full = 1;
+               return 0;
+       }
+
+       BUG_ON(ret);
+
+       ret = btrfs_make_block_group(trans, extent_root, 0, flags,
+                    extent_root->fs_info->chunk_root->root_key.objectid,
+                    start, num_bytes);
+       BUG_ON(ret);
+       return 0;
+}
+
 static int update_block_group(struct btrfs_trans_handle *trans,
                              struct btrfs_root *root,
                              u64 bytenr, u64 num_bytes, int alloc,
@@ -1012,8 +1076,10 @@ static int update_block_group(struct btrfs_trans_handle *trans,
                num_bytes = min(total, cache->key.offset - byte_in_group);
                if (alloc) {
                        old_val += num_bytes;
+                       cache->space_info->bytes_used += num_bytes;
                } else {
                        old_val -= num_bytes;
+                       cache->space_info->bytes_used -= num_bytes;
                        if (mark_free) {
                                set_extent_dirty(&info->free_space_cache,
                                                 bytenr, bytenr + num_bytes - 1,
@@ -1026,6 +1092,7 @@ static int update_block_group(struct btrfs_trans_handle *trans,
        }
        return 0;
 }
+
 static int update_pinned_extents(struct btrfs_root *root,
                                u64 bytenr, u64 num, int pin)
 {
@@ -1047,9 +1114,11 @@ static int update_pinned_extents(struct btrfs_root *root,
                          (bytenr - cache->key.objectid));
                if (pin) {
                        cache->pinned += len;
+                       cache->space_info->bytes_pinned += len;
                        fs_info->total_pinned += len;
                } else {
                        cache->pinned -= len;
+                       cache->space_info->bytes_pinned -= len;
                        fs_info->total_pinned -= len;
                }
                bytenr += len;
@@ -1472,7 +1541,7 @@ check_failed:
                goto new_group;
        }
 
-       if (!(data & BLOCK_GROUP_DATA)) {
+       if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
                block_group = btrfs_lookup_block_group(info, ins->objectid);
                if (block_group)
                        trans->block_group = block_group;
@@ -1532,12 +1601,25 @@ int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
        struct btrfs_path *path;
        struct btrfs_key keys[2];
 
-       if (data)
-               data = BLOCK_GROUP_DATA;
-       else if (root == root->fs_info->chunk_root)
-               data = BLOCK_GROUP_SYSTEM;
-       else
-               data = BLOCK_GROUP_METADATA;
+       if (data) {
+               data = BTRFS_BLOCK_GROUP_DATA;
+       } else if (root == root->fs_info->chunk_root) {
+               data = BTRFS_BLOCK_GROUP_SYSTEM;
+       } else {
+               data = BTRFS_BLOCK_GROUP_METADATA;
+       }
+
+       if (root->ref_cows) {
+               if (data != BTRFS_BLOCK_GROUP_METADATA) {
+                       ret = do_chunk_alloc(trans, root->fs_info->extent_root,
+                                            num_bytes,
+                                            BTRFS_BLOCK_GROUP_METADATA);
+                       BUG_ON(ret);
+               }
+               ret = do_chunk_alloc(trans, root->fs_info->extent_root,
+                                    num_bytes, data);
+               BUG_ON(ret);
+       }
 
        new_hint = max(hint_byte, root->fs_info->alloc_start);
        if (new_hint < btrfs_super_total_bytes(&info->super_copy))
@@ -2490,6 +2572,34 @@ 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;
@@ -2497,6 +2607,7 @@ int btrfs_read_block_groups(struct btrfs_root *root)
        int bit;
        struct btrfs_block_group_cache *cache;
        struct btrfs_fs_info *info = root->fs_info;
+       struct btrfs_space_info *space_info;
        struct extent_io_tree *block_group_cache;
        struct btrfs_key key;
        struct btrfs_key found_key;
@@ -2547,6 +2658,12 @@ int btrfs_read_block_groups(struct btrfs_root *root)
                        bit = BLOCK_GROUP_METADATA;
                }
 
+               ret = update_space_info(info, cache->flags, found_key.offset,
+                                       btrfs_block_group_used(&cache->item),
+                                       &space_info);
+               BUG_ON(ret);
+               cache->space_info = space_info;
+
                /* use EXTENT_LOCKED to prevent merging */
                set_extent_bits(block_group_cache, found_key.objectid,
                                found_key.objectid + found_key.offset - 1,
@@ -2563,3 +2680,58 @@ error:
        btrfs_free_path(path);
        return ret;
 }
+
+int btrfs_make_block_group(struct btrfs_trans_handle *trans,
+                          struct btrfs_root *root, u64 bytes_used,
+                          u64 type, u64 chunk_tree, u64 chunk_objectid,
+                          u64 size)
+{
+       int ret;
+       int bit = 0;
+       struct btrfs_root *extent_root;
+       struct btrfs_block_group_cache *cache;
+       struct extent_io_tree *block_group_cache;
+
+       extent_root = root->fs_info->extent_root;
+       block_group_cache = &root->fs_info->block_group_cache;
+
+       cache = kmalloc(sizeof(*cache), GFP_NOFS);
+       BUG_ON(!cache);
+       cache->key.objectid = chunk_objectid;
+       cache->key.offset = size;
+       cache->cached = 0;
+       cache->pinned = 0;
+       btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
+       memset(&cache->item, 0, sizeof(cache->item));
+       btrfs_set_block_group_used(&cache->item, bytes_used);
+       btrfs_set_block_group_chunk_tree(&cache->item, chunk_tree);
+       btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
+       cache->flags = type;
+       btrfs_set_block_group_flags(&cache->item, type);
+
+       ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
+                               &cache->space_info);
+       BUG_ON(ret);
+
+       if (type & BTRFS_BLOCK_GROUP_DATA) {
+               bit = BLOCK_GROUP_DATA;
+       } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
+               bit = BLOCK_GROUP_SYSTEM;
+       } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
+               bit = BLOCK_GROUP_METADATA;
+       }
+       set_extent_bits(block_group_cache, chunk_objectid,
+                       chunk_objectid + size - 1,
+                       bit | EXTENT_LOCKED, GFP_NOFS);
+       set_state_private(block_group_cache, chunk_objectid,
+                         (unsigned long)cache);
+
+       ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
+                               sizeof(cache->item));
+       BUG_ON(ret);
+
+       finish_current_insert(trans, extent_root);
+       ret = del_pending_extents(trans, extent_root);
+       BUG_ON(ret);
+       return 0;
+}

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 17063cd2cb73db467466ccbf27b0561b5366a14a..109576b57f69d0ee56fe0ba23d852005d3868aaa 100644 (file)
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -1527,6 +1527,7 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
 {
        struct inode *inode;
        struct btrfs_inode_item *inode_item;
+       struct btrfs_block_group_cache *new_inode_group;
        struct btrfs_key *location;
        struct btrfs_path *path;
        struct btrfs_inode_ref *ref;
@@ -1553,9 +1554,13 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
                owner = 0;
        else
                owner = 1;
-       group = btrfs_find_block_group(root, group, 0,
+       new_inode_group = btrfs_find_block_group(root, group, 0,
                                       BTRFS_BLOCK_GROUP_METADATA, owner);
-       BTRFS_I(inode)->block_group = group;
+       if (!new_inode_group) {
+               printk("find_block group failed\n");
+               new_inode_group = group;
+       }
+       BTRFS_I(inode)->block_group = new_inode_group;
        BTRFS_I(inode)->flags = 0;
 
        key[0].objectid = objectid;

diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index 90a8d45dc6d772c0334b79e37e885d687f3af2c6..a52a13f365d6e5396c9da98f704f9eb126297de8 100644 (file)
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -131,7 +131,7 @@ check_pending:
        btrfs_release_path(root, path);
        BUG_ON(*start < search_start);
 
-       if (*start + num_bytes >= search_end) {
+       if (*start + num_bytes > search_end) {
                ret = -ENOSPC;
                goto error;
        }
@@ -159,8 +159,9 @@ int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
                return -ENOMEM;
 
        ret = find_free_dev_extent(trans, device, path, num_bytes, start);
-       if (ret)
+       if (ret) {
                goto err;
+       }
 
        key.objectid = device->devid;
        key.offset = *start;
@@ -214,22 +215,6 @@ error:
        return ret;
 }
 
-static struct btrfs_device *next_device(struct list_head *head,
-                                       struct list_head *last)
-{
-       struct list_head *next = last->next;
-       struct btrfs_device *dev;
-
-       if (list_empty(head))
-               return NULL;
-
-       if (next == head)
-               next = next->next;
-
-       dev = list_entry(next, struct btrfs_device, dev_list);
-       return dev;
-}
-
 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
                           u64 *objectid)
 {
@@ -397,31 +382,63 @@ int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
 
 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
                      struct btrfs_root *extent_root, u64 *start,
-                     u64 *num_bytes, u32 type)
+                     u64 *num_bytes, u64 type)
 {
        u64 dev_offset;
        struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
        struct btrfs_stripe *stripes;
        struct btrfs_device *device = NULL;
        struct btrfs_chunk *chunk;
+       struct list_head private_devs;
        struct list_head *dev_list = &extent_root->fs_info->devices;
-       struct list_head *last_dev = extent_root->fs_info->last_device;
+       struct list_head *cur;
        struct extent_map_tree *em_tree;
        struct map_lookup *map;
        struct extent_map *em;
        u64 physical;
        u64 calc_size = 1024 * 1024 * 1024;
-       int num_stripes;
+       u64 avail;
+       u64 max_avail = 0;
+       int num_stripes = 1;
+       int looped = 0;
        int ret;
-       int index = 0;
+       int index;
        struct btrfs_key key;
 
+       if (list_empty(dev_list))
+               return -ENOSPC;
+again:
+       INIT_LIST_HEAD(&private_devs);
+       cur = dev_list->next;
+       index = 0;
+       /* build a private list of devices we will allocate from */
+       while(index < num_stripes) {
+               device = list_entry(cur, struct btrfs_device, dev_list);
+               avail = device->total_bytes - device->bytes_used;
+               cur = cur->next;
+               if (avail > max_avail)
+                       max_avail = avail;
+               if (avail >= calc_size) {
+                       list_move_tail(&device->dev_list, &private_devs);
+                       index++;
+               }
+               if (cur == dev_list)
+                       break;
+       }
+       if (index < num_stripes) {
+               list_splice(&private_devs, dev_list);
+               if (!looped && max_avail > 0) {
+                       looped = 1;
+                       calc_size = max_avail;
+                       goto again;
+               }
+               return -ENOSPC;
+       }
 
        ret = find_next_chunk(chunk_root, &key.objectid);
        if (ret)
                return ret;
 
-       num_stripes = 1;
        chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
        if (!chunk)
                return -ENOMEM;
@@ -429,11 +446,12 @@ int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
        stripes = &chunk->stripe;
 
        *num_bytes = calc_size;
+       index = 0;
        while(index < num_stripes) {
-               device = next_device(dev_list, last_dev);
-               BUG_ON(!device);
-               last_dev = &device->dev_list;
-               extent_root->fs_info->last_device = last_dev;
+               BUG_ON(list_empty(&private_devs));
+               cur = private_devs.next;
+               device = list_entry(cur, struct btrfs_device, dev_list);
+               list_move_tail(&device->dev_list, dev_list);
 
                ret = btrfs_alloc_dev_extent(trans, device,
                                             key.objectid,
@@ -449,6 +467,7 @@ int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
                physical = dev_offset;
                index++;
        }
+       BUG_ON(!list_empty(&private_devs));
 
        /* key.objectid was set above */
        key.offset = *num_bytes;
@@ -692,17 +711,17 @@ static int read_one_dev(struct btrfs_root *root, struct btrfs_key *key,
        int ret;
 
        devid = btrfs_device_id(leaf, dev_item);
-       if (btrfs_find_device(root, devid))
-               return 0;
-
-       device = kmalloc(sizeof(*device), GFP_NOFS);
-       if (!device)
-               return -ENOMEM;
+       device = btrfs_find_device(root, devid);
+       if (!device) {
+               device = kmalloc(sizeof(*device), GFP_NOFS);
+               if (!device)
+                       return -ENOMEM;
+               list_add(&device->dev_list, &root->fs_info->devices);
+       }
 
        fill_device_from_item(leaf, dev_item, device);
        device->dev_root = root->fs_info->dev_root;
        device->bdev = root->fs_info->sb->s_bdev;
-       list_add(&device->dev_list, &root->fs_info->devices);
        memcpy(&device->dev_key, key, sizeof(*key));
        ret = 0;
 #if 0

diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
index 4a47dcb00a73ab2a43bd60ad98379904ea52834b..4ab9891878a86d8ce87fc004550401f2f0e5e260 100644 (file)
--- a/fs/btrfs/volumes.h
+++ b/fs/btrfs/volumes.h
@@ -71,7 +71,7 @@ int btrfs_read_sys_array(struct btrfs_root *root);
 int btrfs_read_chunk_tree(struct btrfs_root *root);
 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
                      struct btrfs_root *extent_root, u64 *start,
-                     u64 *num_bytes, u32 type);
+                     u64 *num_bytes, u64 type);
 void btrfs_mapping_init(struct btrfs_mapping_tree *tree);
 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree);
 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio);