On 10/09/2015 07:14 PM, Ivan Shapovalov wrote:
On 2015-10-09 at 16:55 +0200, Edward Shishkin wrote:
On 10/09/2015 03:50 PM, Ivan Shapovalov wrote:
On 2015-10-09 at 15:27 +0200, Edward Shishkin wrote:
Hi Ivan,
On 10/09/2015 01:16 PM, Ivan Shapovalov wrote:
Ref.: https://www.mail-archive.com/linux-f2fs-devel%40lists.sou
rcef
orge.net/msg02745.html
Do you have a stack trace for reiser4?
How to reproduce it?
I'll rebuild the kernel without the fix and provide you with the
oops'
stacktrace asap.
I guess that it's tied to the config. In my case, it is
reproducible on
each boot, just as the DE starts up and something issues the first
fsync().
Yes, let's try to find the culprit who doesn't set i_wb...
So, here are the traces I've got after adding an
assert(PageDirty(node->pg)) to queue_jnode():
/* captured by hand as these are panics, not oopses */
1.
queue_jnode()
unformatted_make_reloc()
assign_real_blocknrs()
forward_relocate_unformatted()
forward_alloc_unformatted_journal()
? coord_num_units()
handle_pos_on_twig()
flush_current_atom()
flush_some_atom()
reiser4_writeout()
reiser4_writeback_inodes()
<...>
2.
znode_make_reloc()
forward_alloc_formatted_wa()
? zload_ra()
allocate_znode()
alloc_pos_and_ancestors()
flush_current_atom()
reiser4_txn_end()
? reiser4_txn_end()
reiser4_txn_restart_current()
force_commit_atom()
? reiser4_txn_restart_current()
txnmgr_force_commit_all()
writepages_cryptcompress()
reiser4_writepages_dispatch()
<...>
sys_fsync()
Thanks Ivan.
Not a good news, TBH...
For formatted nodes we can continue to narrow down the problem
(see the attached patch). For unformatted nodes only code review
can help. Normally, all modifications of unformatted nodes should
look like the following:
struct page *page = jnode_page(node);
lock_page(page);
char *data = kmap(page);
/* modifications are going here */
kunmap(page);
set_page_dirty_nobuffers(page); /* somebody forgets to do this */
unlock_page(page);
Modifications of formatted nodes should look like the following:
longterm_lock_znode(node);
zload(node);
/* modifications are going here */
zrelse(node);
znode_make_dirty(node); /* somebody forgets to do this */
longterm_unlock_znode();
Anyway, we can use your patch 3 as a temporal fixup.
Thanks,
Edward.
---
fs/reiser4/lock.c | 5 +++++
1 file changed, 5 insertions(+)
--- a/fs/reiser4/lock.c
+++ b/fs/reiser4/lock.c
@@ -641,6 +641,11 @@ void longterm_unlock_znode(lock_handle *
zput(node);
return;
}
+ if (znode_page(node) != NULL)
+ if (!PageDirty(znode_page(node))) {
+ warning("", "releasing last write-lock, but page is fishily clean");
+ dump_stack();
+ }
}
if (handle->signaled)
