]> rtime.felk.cvut.cz Git - linux-imx.git/commitdiff
Btrfs: fix wrong handle at error path of create_snapshot() when the commit fails
authorMiao Xie <miaox@cn.fujitsu.com>
Mon, 4 Mar 2013 09:44:29 +0000 (09:44 +0000)
committerJosef Bacik <jbacik@fusionio.com>
Mon, 4 Mar 2013 21:33:22 +0000 (16:33 -0500)
There are several bugs at error path of create_snapshot() when the
transaction commitment failed.
- access the freed transaction handler. At the end of the
  transaction commitment, the transaction handler was freed, so we
  should not access it after the transaction commitment.
- we were not aware of the error which happened during the snapshot
  creation if we submitted a async transaction commitment.
- pending snapshot access vs pending snapshot free. when something
  wrong happened after we submitted a async transaction commitment,
  the transaction committer would cleanup the pending snapshots and
  free them. But the snapshot creators were not aware of it, they
  would access the freed pending snapshots.

This patch fixes the above problems by:
- remove the dangerous code that accessed the freed handler
- assign ->error if the error happens during the snapshot creation
- the transaction committer doesn't free the pending snapshots,
  just assigns the error number and evicts them before we unblock
  the transaction.

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
fs/btrfs/disk-io.c
fs/btrfs/ioctl.c
fs/btrfs/transaction.c

index 02369a3c162e0e891a09c3f893f277c5ec907635..7d84651e850b22cb937b29c68d34c1119af1e4a0 100644 (file)
@@ -62,7 +62,7 @@ static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
                                      struct btrfs_root *root);
-static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
+static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t);
 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
                                        struct extent_io_tree *dirty_pages,
@@ -3687,7 +3687,7 @@ int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
        return ret;
 }
 
-static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
+static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
 {
        struct btrfs_pending_snapshot *snapshot;
        struct list_head splice;
@@ -3700,10 +3700,8 @@ static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
                snapshot = list_entry(splice.next,
                                      struct btrfs_pending_snapshot,
                                      list);
-
+               snapshot->error = -ECANCELED;
                list_del_init(&snapshot->list);
-
-               kfree(snapshot);
        }
 }
 
@@ -3840,6 +3838,8 @@ void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
        cur_trans->blocked = 1;
        wake_up(&root->fs_info->transaction_blocked_wait);
 
+       btrfs_evict_pending_snapshots(cur_trans);
+
        cur_trans->blocked = 0;
        wake_up(&root->fs_info->transaction_wait);
 
@@ -3849,8 +3849,6 @@ void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
        btrfs_destroy_delayed_inodes(root);
        btrfs_assert_delayed_root_empty(root);
 
-       btrfs_destroy_pending_snapshots(cur_trans);
-
        btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
                                     EXTENT_DIRTY);
        btrfs_destroy_pinned_extent(root,
@@ -3894,6 +3892,8 @@ int btrfs_cleanup_transaction(struct btrfs_root *root)
                if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
                        wake_up(&root->fs_info->transaction_blocked_wait);
 
+               btrfs_evict_pending_snapshots(t);
+
                t->blocked = 0;
                smp_mb();
                if (waitqueue_active(&root->fs_info->transaction_wait))
@@ -3907,8 +3907,6 @@ int btrfs_cleanup_transaction(struct btrfs_root *root)
                btrfs_destroy_delayed_inodes(root);
                btrfs_assert_delayed_root_empty(root);
 
-               btrfs_destroy_pending_snapshots(t);
-
                btrfs_destroy_delalloc_inodes(root);
 
                spin_lock(&root->fs_info->trans_lock);
index b908960c97466612a3ec6e631b6134536c8d86fb..94c0e42dfa1e1f0f760908f296df01b67077471e 100644 (file)
@@ -596,12 +596,8 @@ static int create_snapshot(struct btrfs_root *root, struct inode *dir,
                ret = btrfs_commit_transaction(trans,
                                               root->fs_info->extent_root);
        }
-       if (ret) {
-               /* cleanup_transaction has freed this for us */
-               if (trans->aborted)
-                       pending_snapshot = NULL;
+       if (ret)
                goto fail;
-       }
 
        ret = pending_snapshot->error;
        if (ret)
index f11c2e0a37465e989eaacca2111fbb99753ac3c1..d8fce6fe9cf8361ceda3062615cf5170d37ec57e 100644 (file)
@@ -1053,7 +1053,12 @@ int btrfs_defrag_root(struct btrfs_root *root)
 
 /*
  * new snapshots need to be created at a very specific time in the
- * transaction commit.  This does the actual creation
+ * transaction commit.  This does the actual creation.
+ *
+ * Note:
+ * If the error which may affect the commitment of the current transaction
+ * happens, we should return the error number. If the error which just affect
+ * the creation of the pending snapshots, just return 0.
  */
 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
                                   struct btrfs_fs_info *fs_info,
@@ -1072,7 +1077,7 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
        struct extent_buffer *tmp;
        struct extent_buffer *old;
        struct timespec cur_time = CURRENT_TIME;
-       int ret;
+       int ret = 0;
        u64 to_reserve = 0;
        u64 index = 0;
        u64 objectid;
@@ -1081,40 +1086,36 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 
        path = btrfs_alloc_path();
        if (!path) {
-               ret = pending->error = -ENOMEM;
-               return ret;
+               pending->error = -ENOMEM;
+               return 0;
        }
 
        new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
        if (!new_root_item) {
-               ret = pending->error = -ENOMEM;
+               pending->error = -ENOMEM;
                goto root_item_alloc_fail;
        }
 
-       ret = btrfs_find_free_objectid(tree_root, &objectid);
-       if (ret) {
-               pending->error = ret;
+       pending->error = btrfs_find_free_objectid(tree_root, &objectid);
+       if (pending->error)
                goto no_free_objectid;
-       }
 
        btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
 
        if (to_reserve > 0) {
-               ret = btrfs_block_rsv_add(root, &pending->block_rsv,
-                                         to_reserve,
-                                         BTRFS_RESERVE_NO_FLUSH);
-               if (ret) {
-                       pending->error = ret;
+               pending->error = btrfs_block_rsv_add(root,
+                                                    &pending->block_rsv,
+                                                    to_reserve,
+                                                    BTRFS_RESERVE_NO_FLUSH);
+               if (pending->error)
                        goto no_free_objectid;
-               }
        }
 
-       ret = btrfs_qgroup_inherit(trans, fs_info, root->root_key.objectid,
-                                  objectid, pending->inherit);
-       if (ret) {
-               pending->error = ret;
+       pending->error = btrfs_qgroup_inherit(trans, fs_info,
+                                             root->root_key.objectid,
+                                             objectid, pending->inherit);
+       if (pending->error)
                goto no_free_objectid;
-       }
 
        key.objectid = objectid;
        key.offset = (u64)-1;
@@ -1142,7 +1143,7 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
                                         dentry->d_name.len, 0);
        if (dir_item != NULL && !IS_ERR(dir_item)) {
                pending->error = -EEXIST;
-               goto fail;
+               goto dir_item_existed;
        } else if (IS_ERR(dir_item)) {
                ret = PTR_ERR(dir_item);
                btrfs_abort_transaction(trans, root, ret);
@@ -1273,6 +1274,8 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
        if (ret)
                btrfs_abort_transaction(trans, root, ret);
 fail:
+       pending->error = ret;
+dir_item_existed:
        trans->block_rsv = rsv;
        trans->bytes_reserved = 0;
 no_free_objectid:
@@ -1288,12 +1291,17 @@ root_item_alloc_fail:
 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
                                             struct btrfs_fs_info *fs_info)
 {
-       struct btrfs_pending_snapshot *pending;
+       struct btrfs_pending_snapshot *pending, *next;
        struct list_head *head = &trans->transaction->pending_snapshots;
+       int ret = 0;
 
-       list_for_each_entry(pending, head, list)
-               create_pending_snapshot(trans, fs_info, pending);
-       return 0;
+       list_for_each_entry_safe(pending, next, head, list) {
+               list_del(&pending->list);
+               ret = create_pending_snapshot(trans, fs_info, pending);
+               if (ret)
+                       break;
+       }
+       return ret;
 }
 
 static void update_super_roots(struct btrfs_root *root)