Btrfs: fix BUG_ON() casued by the reserved space migration
authorMiao Xie <miaox@cn.fujitsu.com>
Wed, 25 Sep 2013 13:47:45 +0000 (21:47 +0800)
committerChris Mason <chris.mason@fusionio.com>
Tue, 12 Nov 2013 02:54:28 +0000 (21:54 -0500)
When we did space balance and snapshot creation at the same time, we might
meet the following oops:
 kernel BUG at fs/btrfs/inode.c:3038!
 [SNIP]
 Call Trace:
 [<ffffffffa0411ec7>] btrfs_orphan_cleanup+0x293/0x407 [btrfs]
 [<ffffffffa042dc45>] btrfs_mksubvol.isra.28+0x259/0x373 [btrfs]
 [<ffffffffa042de85>] btrfs_ioctl_snap_create_transid+0x126/0x156 [btrfs]
 [<ffffffffa042dff1>] btrfs_ioctl_snap_create_v2+0xd0/0x121 [btrfs]
 [<ffffffffa0430b2c>] btrfs_ioctl+0x414/0x1854 [btrfs]
 [<ffffffff813b60b7>] ? __do_page_fault+0x305/0x379
 [<ffffffff811215a9>] vfs_ioctl+0x1d/0x39
 [<ffffffff81121d7c>] do_vfs_ioctl+0x32d/0x3e2
 [<ffffffff81057fe7>] ? finish_task_switch+0x80/0xb8
 [<ffffffff81121e88>] SyS_ioctl+0x57/0x83
 [<ffffffff813b39ff>] ? do_device_not_available+0x12/0x14
 [<ffffffff813b99c2>] system_call_fastpath+0x16/0x1b
 [SNIP]
 RIP  [<ffffffffa040da40>] btrfs_orphan_add+0xc3/0x126 [btrfs]

The reason of the problem is that the relocation root creation stole
the reserved space, which was reserved for orphan item deletion.

There are several ways to fix this problem, one is to increasing
the reserved space size of the space balace, and then we can use
that space to create the relocation tree for each fs/file trees.
But it is hard to calculate the suitable size because we doesn't
know how many fs/file trees we need relocate.

We fixed this problem by reserving the space for relocation root creation
actively since the space it need is very small (one tree block, used for
root node copy), then we use that reserved space to create the
relocation tree. If we don't reserve space for relocation tree creation,
we will use the reserved space of the balance.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>
fs/btrfs/relocation.c
fs/btrfs/transaction.c
fs/btrfs/transaction.h

index 0359eecac95d3c849c6499d13743b56950fc2b76..a945374993f7d54a4768a92e25fb2ff2054b22a2 100644 (file)
@@ -1383,6 +1383,7 @@ int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
 {
        struct btrfs_root *reloc_root;
        struct reloc_control *rc = root->fs_info->reloc_ctl;
+       struct btrfs_block_rsv *rsv;
        int clear_rsv = 0;
        int ret;
 
@@ -1396,13 +1397,14 @@ int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
            root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
                return 0;
 
-       if (!trans->block_rsv) {
+       if (!trans->reloc_reserved) {
+               rsv = trans->block_rsv;
                trans->block_rsv = rc->block_rsv;
                clear_rsv = 1;
        }
        reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
        if (clear_rsv)
-               trans->block_rsv = NULL;
+               trans->block_rsv = rsv;
 
        ret = __add_reloc_root(reloc_root);
        BUG_ON(ret < 0);
index 134039fd59bbf9a094f5ca30638be893fd7a27f5..a213bafe68ec56caf61048462f597774a867b919 100644 (file)
@@ -353,6 +353,17 @@ static int may_wait_transaction(struct btrfs_root *root, int type)
        return 0;
 }
 
+static inline bool need_reserve_reloc_root(struct btrfs_root *root)
+{
+       if (!root->fs_info->reloc_ctl ||
+           !root->ref_cows ||
+           root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
+           root->reloc_root)
+               return false;
+
+       return true;
+}
+
 static struct btrfs_trans_handle *
 start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type,
                  enum btrfs_reserve_flush_enum flush)
@@ -360,8 +371,9 @@ start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type,
        struct btrfs_trans_handle *h;
        struct btrfs_transaction *cur_trans;
        u64 num_bytes = 0;
-       int ret;
        u64 qgroup_reserved = 0;
+       bool reloc_reserved = false;
+       int ret;
 
        if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
                return ERR_PTR(-EROFS);
@@ -390,6 +402,14 @@ start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type,
                }
 
                num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
+               /*
+                * Do the reservation for the relocation root creation
+                */
+               if (unlikely(need_reserve_reloc_root(root))) {
+                       num_bytes += root->nodesize;
+                       reloc_reserved = true;
+               }
+
                ret = btrfs_block_rsv_add(root,
                                          &root->fs_info->trans_block_rsv,
                                          num_bytes, flush);
@@ -451,6 +471,7 @@ again:
        h->delayed_ref_elem.seq = 0;
        h->type = type;
        h->allocating_chunk = false;
+       h->reloc_reserved = false;
        INIT_LIST_HEAD(&h->qgroup_ref_list);
        INIT_LIST_HEAD(&h->new_bgs);
 
@@ -466,6 +487,7 @@ again:
                                              h->transid, num_bytes, 1);
                h->block_rsv = &root->fs_info->trans_block_rsv;
                h->bytes_reserved = num_bytes;
+               h->reloc_reserved = reloc_reserved;
        }
        h->qgroup_reserved = qgroup_reserved;
 
index 306f88ae1de3c1e50d2435d82145e41c689577d2..7657d115067d3f8ae1d5cc39a53410d8a1011673 100644 (file)
@@ -92,6 +92,7 @@ struct btrfs_trans_handle {
        short aborted;
        short adding_csums;
        bool allocating_chunk;
+       bool reloc_reserved;
        unsigned int type;
        /*
         * this root is only needed to validate that the root passed to