On Wed, Feb 10, 2016 at 02:23:00PM -0700, Chris Murphy wrote: > On Wed, Feb 10, 2016 at 1:39 PM, Михаил Гаврилов > <mikhail.v.gavri...@gmail.com> wrote: > > > > > > Here full log: > > http://btrfs.sy24.ru/kernel-sysrqw-btrfscleaner770blocked-2.txt > > > > I am so sorry if this log is useless. > > Looks good to me. The blocked task happens out of no where with > nothing reported for almost an hour before the blocking. And I see the > sysrq: SysRq : Show Blocked State was issued and lots of information > is in the file. > > > If "sysrq" is needed enabled before hang then I need set this > > permanently because as I said I not having exactly reproducing this. > > echo 1 > /proc/sys/kernel/sysrq can happen anytime, it just enables > sysrq triggering functions which on Fedora kernels is not enabled by > default. The main thing is that the echo w to the sysrq trigger needs > to happen at the time of the problem to show the state. You did that. > Let's see what Liu Bo has to say about it.
Really appreciate for collecting these, it should be helpful. Unfortunately I still could not figure out who's holding fs tree's root WRITE_LOCK so that others are blocked. A possible bug in log code (the follwing patch addressed it), - log_new_dir_dentries() is holding log tree's leaf READ_LOCK and may try to get fs tree's READ_LOCK via btrfs_iget() -> btrfs_lookup(). (This is shown in the backtrac) - btrfs_log_inode() can call btrfs_search_forward() to get fs tree's leaf READ_LOCK and then call copy_items() -> btrfs_insert_empty_items() to acquire WRITE_LOCK of log tree's leaf and leaf's parent. (In the backtrace, this is blocked by item 1 because log_new_dir_dentries is holding a log tree leaf's READ_LOCK() which happens to be sibling to the leaf that btrfs_insert_empty_items() is accessing, when doing split_leaf() it needs to get the sibling's WRITE_LOCK(). ) Thanks, -liubo diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c index 323e12c..4a64fdd 100644 --- a/fs/btrfs/tree-log.c +++ b/fs/btrfs/tree-log.c @@ -4956,6 +4956,7 @@ process_leaf: if (di_key.type == BTRFS_ROOT_ITEM_KEY) continue; + btrfs_release_path(path); di_inode = btrfs_iget(root->fs_info->sb, &di_key, root, NULL); if (IS_ERR(di_inode)) { @@ -4971,7 +4972,6 @@ process_leaf: ctx->log_new_dentries = false; if (type == BTRFS_FT_DIR) log_mode = LOG_INODE_ALL; - btrfs_release_path(path); ret = btrfs_log_inode(trans, root, di_inode, log_mode, 0, LLONG_MAX, ctx); iput(di_inode); -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html