All of lore.kernel.org
 help / color / mirror / Atom feed
From: Zheng Yan <zheng.yan@oracle.com>
To: linux-btrfs@vger.kernel.org, Chris Mason <chris.mason@oracle.com>,
	yanzheng@21cn.com
Subject: [PATCH 4/4] update allocator the handle 'hole' in address space
Date: Tue, 09 Sep 2008 02:56:42 +0800	[thread overview]
Message-ID: <48C5756A.6090308@oracle.com> (raw)

Hello,

Space balancing creates some 'holes' in the address space, this confuses
the allocator. This patch removes all checks against the super_total_bytes.

Regards
Yan Zheng

---
diff -r 16615727163e ctree.h
--- a/ctree.h	Tue Sep 09 02:16:25 2008 +0800
+++ b/ctree.h	Tue Sep 09 02:16:35 2008 +0800
@@ -500,7 +500,6 @@
 	u64 bytes_used;
 	u64 bytes_pinned;
 	u64 bytes_reserved;
-	int full;
 	int force_alloc;
 	struct list_head list;
 };
diff -r 16615727163e extent-tree.c
--- a/extent-tree.c	Tue Sep 09 02:16:25 2008 +0800
+++ b/extent-tree.c	Tue Sep 09 02:16:35 2008 +0800
@@ -51,6 +51,8 @@
 				 btrfs_root *extent_root);
 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
 			       btrfs_root *extent_root);
+static int __alloc_chunk_for_extent_op(struct btrfs_trans_handle *trans,
+				       struct btrfs_root *root);
 static struct btrfs_block_group_cache *
 __btrfs_find_block_group(struct btrfs_root *root,
 			 struct btrfs_block_group_cache *hint,
@@ -254,13 +256,10 @@
 	struct extent_state *state;
 	u64 last;
 	u64 start = 0;
-	u64 cache_miss = 0;
-	u64 total_fs_bytes;
 	u64 search_start = *start_ret;
 	int wrapped = 0;
 
 	WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
-	total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 	free_space_cache = &root->fs_info->free_space_cache;
 
 	if (!cache)
@@ -280,8 +279,6 @@
 	state = find_first_extent_bit_state(free_space_cache, last, EXTENT_DIRTY);
 	while(1) {
 		if (!state) {
-			if (!cache_miss)
-				cache_miss = last;
 			spin_unlock_irq(&free_space_cache->lock);
 			goto new_group;
 		}
@@ -318,7 +315,7 @@
 	last = cache->key.objectid + cache->key.offset;
 wrapped:
 	cache = btrfs_lookup_first_block_group(root->fs_info, last);
-	if (!cache || cache->key.objectid >= total_fs_bytes) {
+	if (!cache) {
 no_cache:
 		if (!wrapped) {
 			wrapped = 1;
@@ -327,12 +324,6 @@
 		}
 		goto out;
 	}
-	if (cache_miss && !cache->cached) {
-		cache_block_group(root, cache);
-		last = cache_miss;
-		cache = btrfs_lookup_first_block_group(root->fs_info, last);
-	}
-	cache_miss = 0;
 	cache = btrfs_find_block_group(root, cache, last, data, 0);
 	if (!cache)
 		goto no_cache;
@@ -719,6 +710,7 @@
 	if (!path)
 		return -ENOMEM;
 
+	__alloc_chunk_for_extent_op(trans, root);
 	ret = lookup_extent_backref(trans, extent_root, path,
 				    bytenr, orig_parent,
 				    root_objectid, ref_generation, 1);
@@ -779,6 +771,7 @@
 	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 	key.offset = num_bytes;
 
+	__alloc_chunk_for_extent_op(trans, root);
 	ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 				0, 1);
 	if (ret < 0)
@@ -1277,6 +1270,7 @@
 	unsigned long bi;
 	struct extent_buffer *leaf;
 
+	__alloc_chunk_for_extent_op(trans, extent_root);
 	ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
 	if (ret < 0)
 		goto fail;
@@ -1374,7 +1368,6 @@
 	if (found) {
 		found->total_bytes += total_bytes;
 		found->bytes_used += bytes_used;
-		found->full = 0;
 		*space_info = found;
 		return 0;
 	}
@@ -1388,7 +1381,6 @@
 	found->bytes_used = bytes_used;
 	found->bytes_pinned = 0;
 	found->bytes_reserved = 0;
-	found->full = 0;
 	found->force_alloc = 0;
 	*space_info = found;
 	return 0;
@@ -1462,8 +1454,6 @@
 		force = 1;
 		space_info->force_alloc = 0;
 	}
-	if (space_info->full)
-		goto out;
 
 	thresh = div_factor(space_info->total_bytes, 6);
 	if (!force &&
@@ -1473,12 +1463,8 @@
 
 	mutex_lock(&extent_root->fs_info->chunk_mutex);
 	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;
+	if (ret == -ENOSPC)
 		goto out_unlock;
-	}
-	BUG_ON(ret);
 
 	ret = btrfs_make_block_group(trans, extent_root, 0, flags,
 		     BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
@@ -1487,6 +1473,24 @@
 	mutex_unlock(&extent_root->fs_info->chunk_mutex);
 out:
 	return 0;
+}
+
+static int __alloc_chunk_for_extent_op(struct btrfs_trans_handle *trans,
+				       struct btrfs_root *root)
+{
+	struct btrfs_fs_info *info = root->fs_info;
+	int ret;
+
+	if (root == root->fs_info->chunk_root ||
+	    root == root->fs_info->dev_root)
+		return 0;
+
+	ret = do_chunk_alloc(trans, root->fs_info->extent_root,
+			     2 * 1024 * 1024,
+			     BTRFS_BLOCK_GROUP_METADATA |
+			     (info->metadata_alloc_profile &
+			      info->avail_metadata_alloc_bits), 0);
+	return ret;
 }
 
 static int update_block_group(struct btrfs_trans_handle *trans,
@@ -2232,6 +2236,7 @@
 	if (ref_generation != trans->transid)
 		pin = 1;
 
+	__alloc_chunk_for_extent_op(trans, root);
 	ret = __free_extent(trans, root, bytenr, num_bytes, parent,
 			    root_objectid, ref_generation,
 			    owner_objectid, owner_offset, pin, pin == 0);
@@ -2324,9 +2329,6 @@
 
 	search_start = max(search_start, first_logical_byte(root, 0));
 	orig_search_start = search_start;
-
-	if (search_end == (u64)-1)
-		search_end = btrfs_super_total_bytes(&info->super_copy);
 
 	if (hint_byte) {
 		block_group = btrfs_lookup_first_block_group(info, hint_byte);
@@ -2399,9 +2401,6 @@
 	ins->objectid = search_start;
 	ins->offset = num_bytes;
 
-	if (ins->objectid + num_bytes >= search_end)
-		goto enospc;
-
 	if (ins->objectid + num_bytes >
 	    block_group->key.objectid + block_group->key.offset) {
 		search_start = block_group->key.objectid +
@@ -2443,7 +2442,8 @@
 	return 0;
 
 new_group:
-	if (search_start + num_bytes >= search_end) {
+	block_group = btrfs_lookup_first_block_group(info, search_start);
+	if (!block_group) {
 enospc:
 		search_start = orig_search_start;
 		if (full_scan) {
@@ -2457,7 +2457,6 @@
 		} else
 			wrapped = 1;
 	}
-	block_group = btrfs_lookup_first_block_group(info, search_start);
 	cond_resched();
 	block_group = btrfs_find_block_group(root, block_group,
 					     search_start, data, 0);
@@ -2613,6 +2612,7 @@
 	path = btrfs_alloc_path();
 	BUG_ON(!path);
 
+	__alloc_chunk_for_extent_op(trans, root);
 	ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
 				       sizes, 2);
 
@@ -3795,6 +3795,8 @@
 		needs_lock = 1;
 		mutex_lock(&root->fs_info->alloc_mutex);
 	}
+	if (root == root->fs_info->extent_root)
+		__alloc_chunk_for_extent_op(trans, root);
 
 	path->lowest_level = level;
 	ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
@@ -4406,9 +4408,6 @@
 		set_extent_bits(block_group_cache, found_key.objectid,
 				found_key.objectid + found_key.offset - 1,
 				bit | EXTENT_LOCKED, GFP_NOFS);
-		if (key.objectid >=
-		    btrfs_super_total_bytes(&info->super_copy))
-			break;
 	}
 	ret = 0;
 error:
diff -r 16615727163e volumes.c
--- a/volumes.c	Tue Sep 09 02:16:25 2008 +0800
+++ b/volumes.c	Tue Sep 09 02:16:35 2008 +0800
@@ -1094,6 +1094,7 @@
 		 &root->fs_info->fs_devices->alloc_list);
 	root->fs_info->fs_devices->num_devices++;
 	root->fs_info->fs_devices->open_devices++;
+	root->fs_info->fs_devices->full = 0;
 out:
 	unlock_chunks(root);
 	btrfs_end_transaction(trans, root);
@@ -1376,6 +1377,8 @@
 		BUG_ON(ret);
 
 		btrfs_end_transaction(trans, dev_root);
+
+		dev_root->fs_info->fs_devices->full = 0;
 	}
 
 	/* step two, relocate all the chunks */
@@ -1422,6 +1425,8 @@
 					   found_key.objectid,
 					   found_key.offset);
 		BUG_ON(ret);
+
+		dev_root->fs_info->fs_devices->full = 0;
 	}
 	ret = 0;
 error:
@@ -1601,6 +1606,10 @@
 		WARN_ON(1);
 		type &= ~BTRFS_BLOCK_GROUP_DUP;
 	}
+
+	if (extent_root->fs_info->fs_devices->full)
+		return -ENOSPC;
+
 	dev_list = &extent_root->fs_info->fs_devices->alloc_list;
 	if (list_empty(dev_list))
 		return -ENOSPC;
@@ -1717,7 +1726,12 @@
 			calc_size = max_avail;
 			goto again;
 		}
+		if (calc_size > min_stripe_size) {
+			calc_size >>= 1;
+			goto again;
+		}
 		btrfs_free_path(path);
+		extent_root->fs_info->fs_devices->full = 1;
 		return -ENOSPC;
 	}
 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
diff -r 16615727163e volumes.h
--- a/volumes.h	Tue Sep 09 02:16:25 2008 +0800
+++ b/volumes.h	Tue Sep 09 02:16:35 2008 +0800
@@ -85,6 +85,7 @@
 	struct list_head alloc_list;
 	struct list_head list;
 	int mounted;
+	int full;
 };
 
 struct btrfs_bio_stripe {

                 reply	other threads:[~2008-09-08 18:56 UTC|newest]

Thread overview: [no followups] expand[flat|nested]  mbox.gz  Atom feed

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=48C5756A.6090308@oracle.com \
    --to=zheng.yan@oracle.com \
    --cc=chris.mason@oracle.com \
    --cc=linux-btrfs@vger.kernel.org \
    --cc=yanzheng@21cn.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.