btrfs: clean snapshots one by one
authorDavid Sterba <dsterba@suse.cz>
Tue, 12 Mar 2013 15:13:28 +0000 (15:13 +0000)
committerJosef Bacik <jbacik@fusionio.com>
Mon, 6 May 2013 19:54:21 +0000 (15:54 -0400)
Each time pick one dead root from the list and let the caller know if
it's needed to continue. This should improve responsiveness during
umount and balance which at some point waits for cleaning all currently
queued dead roots.

A new dead root is added to the end of the list, so the snapshots
disappear in the order of deletion.

The snapshot cleaning work is now done only from the cleaner thread and the
others wake it if needed.

Signed-off-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
fs/btrfs/disk-io.c
fs/btrfs/extent-tree.c
fs/btrfs/relocation.c
fs/btrfs/transaction.c
fs/btrfs/transaction.h

index 898263f56d96705960cee25501bf20d16a19ba7c..b8ed1d4fe50966142e01945799046835b5f256ec 100644 (file)
@@ -1658,15 +1658,20 @@ static int cleaner_kthread(void *arg)
        struct btrfs_root *root = arg;
 
        do {
+               int again = 0;
+
                if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
-                   mutex_trylock(&root->fs_info->cleaner_mutex)) {
-                       btrfs_run_delayed_iputs(root);
-                       btrfs_clean_old_snapshots(root);
-                       mutex_unlock(&root->fs_info->cleaner_mutex);
+                   down_read_trylock(&root->fs_info->sb->s_umount)) {
+                       if (mutex_trylock(&root->fs_info->cleaner_mutex)) {
+                               btrfs_run_delayed_iputs(root);
+                               again = btrfs_clean_one_deleted_snapshot(root);
+                               mutex_unlock(&root->fs_info->cleaner_mutex);
+                       }
                        btrfs_run_defrag_inodes(root->fs_info);
+                       up_read(&root->fs_info->sb->s_umount);
                }
 
-               if (!try_to_freeze()) {
+               if (!try_to_freeze() && !again) {
                        set_current_state(TASK_INTERRUPTIBLE);
                        if (!kthread_should_stop())
                                schedule();
@@ -3358,8 +3363,8 @@ int btrfs_commit_super(struct btrfs_root *root)
 
        mutex_lock(&root->fs_info->cleaner_mutex);
        btrfs_run_delayed_iputs(root);
-       btrfs_clean_old_snapshots(root);
        mutex_unlock(&root->fs_info->cleaner_mutex);
+       wake_up_process(root->fs_info->cleaner_kthread);
 
        /* wait until ongoing cleanup work done */
        down_write(&root->fs_info->cleanup_work_sem);
index 7505856df9f38cc53d9399b7b05da07417342b46..be09157ff91ba97a5728638fee5cd1a1ea3e4f95 100644 (file)
@@ -7286,6 +7286,8 @@ static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
  * reference count by one. if update_ref is true, this function
  * also make sure backrefs for the shared block and all lower level
  * blocks are properly updated.
+ *
+ * If called with for_reloc == 0, may exit early with -EAGAIN
  */
 int btrfs_drop_snapshot(struct btrfs_root *root,
                         struct btrfs_block_rsv *block_rsv, int update_ref,
@@ -7386,6 +7388,12 @@ int btrfs_drop_snapshot(struct btrfs_root *root,
        wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
 
        while (1) {
+               if (!for_reloc && btrfs_fs_closing(root->fs_info)) {
+                       pr_debug("btrfs: drop snapshot early exit\n");
+                       err = -EAGAIN;
+                       goto out_end_trans;
+               }
+
                ret = walk_down_tree(trans, root, path, wc);
                if (ret < 0) {
                        err = ret;
index 86f192ffc2128e87d0045aa895079f7fa399d6cb..4ef5f7455fb3dae14521af7b375a61bcd7d0cd6f 100644 (file)
@@ -4148,10 +4148,7 @@ int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start)
 
        while (1) {
                mutex_lock(&fs_info->cleaner_mutex);
-
-               btrfs_clean_old_snapshots(fs_info->tree_root);
                ret = relocate_block_group(rc);
-
                mutex_unlock(&fs_info->cleaner_mutex);
                if (ret < 0) {
                        err = ret;
index 6c0a72ab6de04ea415a6f5ba5332050fc0ee9266..5a5ea99f29ed6cd741b27241e754322c4f55b26a 100644 (file)
@@ -948,7 +948,7 @@ static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
 int btrfs_add_dead_root(struct btrfs_root *root)
 {
        spin_lock(&root->fs_info->trans_lock);
-       list_add(&root->root_list, &root->fs_info->dead_roots);
+       list_add_tail(&root->root_list, &root->fs_info->dead_roots);
        spin_unlock(&root->fs_info->trans_lock);
        return 0;
 }
@@ -1873,31 +1873,49 @@ cleanup_transaction:
 }
 
 /*
- * interface function to delete all the snapshots we have scheduled for deletion
+ * return < 0 if error
+ * 0 if there are no more dead_roots at the time of call
+ * 1 there are more to be processed, call me again
+ *
+ * The return value indicates there are certainly more snapshots to delete, but
+ * if there comes a new one during processing, it may return 0. We don't mind,
+ * because btrfs_commit_super will poke cleaner thread and it will process it a
+ * few seconds later.
  */
-int btrfs_clean_old_snapshots(struct btrfs_root *root)
+int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
 {
-       LIST_HEAD(list);
+       int ret;
        struct btrfs_fs_info *fs_info = root->fs_info;
 
+       if (fs_info->sb->s_flags & MS_RDONLY) {
+               pr_debug("btrfs: cleaner called for RO fs!\n");
+               return 0;
+       }
+
        spin_lock(&fs_info->trans_lock);
-       list_splice_init(&fs_info->dead_roots, &list);
+       if (list_empty(&fs_info->dead_roots)) {
+               spin_unlock(&fs_info->trans_lock);
+               return 0;
+       }
+       root = list_first_entry(&fs_info->dead_roots,
+                       struct btrfs_root, root_list);
+       list_del(&root->root_list);
        spin_unlock(&fs_info->trans_lock);
 
-       while (!list_empty(&list)) {
-               int ret;
-
-               root = list_entry(list.next, struct btrfs_root, root_list);
-               list_del(&root->root_list);
+       pr_debug("btrfs: cleaner removing %llu\n",
+                       (unsigned long long)root->objectid);
 
-               btrfs_kill_all_delayed_nodes(root);
+       btrfs_kill_all_delayed_nodes(root);
 
-               if (btrfs_header_backref_rev(root->node) <
-                   BTRFS_MIXED_BACKREF_REV)
-                       ret = btrfs_drop_snapshot(root, NULL, 0, 0);
-               else
-                       ret =btrfs_drop_snapshot(root, NULL, 1, 0);
-               BUG_ON(ret < 0);
-       }
-       return 0;
+       if (btrfs_header_backref_rev(root->node) <
+                       BTRFS_MIXED_BACKREF_REV)
+               ret = btrfs_drop_snapshot(root, NULL, 0, 0);
+       else
+               ret = btrfs_drop_snapshot(root, NULL, 1, 0);
+       /*
+        * If we encounter a transaction abort during snapshot cleaning, we
+        * don't want to crash here
+        */
+       BUG_ON(ret < 0 && ret != -EAGAIN && ret != -EROFS);
+       return 1;
 }
index 3c8e0d25c8e4d1bbe88d7404150bcb1d3db1a8f6..f6edd5e6baa3d759e89738280f5e2cbc44385d25 100644 (file)
@@ -123,7 +123,7 @@ int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
 
 int btrfs_add_dead_root(struct btrfs_root *root);
 int btrfs_defrag_root(struct btrfs_root *root);
-int btrfs_clean_old_snapshots(struct btrfs_root *root);
+int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root);
 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
                             struct btrfs_root *root);
 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,