Re: md/bitmap: move some fields of 'struct bitmap' into a 'storage' substruct.

[NeilBrown <neilb@suse.de>]

[-- Attachment #1: Type: text/plain, Size: 2461 bytes --]

On Fri, 20 Apr 2012 16:13:36 +0300 Dan Carpenter <dan.carpenter@ORACLE.COM>
wrote:

> Hello NeilBrown,
> 
> This is a semi-automatic email about new static checker warnings.
> 
> The patch 9159d8a35d0a: "md/bitmap: move some fields of 'struct
> bitmap' into a 'storage' substruct." from Apr 19, 2012, leads to the
> following Smatch complaint:
> 
> drivers/md/bitmap.c:1212 bitmap_daemon_work()
> 	error: we previously assumed 'bitmap->storage.filemap' could be
> 	       null (see line 1145)
> 
> drivers/md/bitmap.c
>   1144			bitmap->need_sync = 0;
>   1145			if (bitmap->storage.filemap) {
>                             ^^^^^^^^^^^^^^^^^^^^^^^
> Renamed check.
> 
>   1146				sb = kmap_atomic(bitmap->storage.sb_page);
>   1147				sb->events_cleared =
>   1148					cpu_to_le64(bitmap->events_cleared);
>   1149				kunmap_atomic(sb);
>   1150				set_page_attr(bitmap, 0,
>   1151					      BITMAP_PAGE_NEEDWRITE);
>   1152			}
>   1153		}
> 
> [snip]
> 
>   1210			if (test_and_clear_page_attr(bitmap, j,
>   1211						     BITMAP_PAGE_NEEDWRITE)) {
>   1212				write_page(bitmap, bitmap->storage.filemap[j], 0);
>                                                    ^^^^^^^^^^^^^^^^^^^^^^^^^^
> Renamed dereference.

There is a relationship between storage.filemap and storage.file_pages.
If the later is 0, the former must be non-NULL.

This dereference only happens if file_pages > 0...  Maybe I should check
file_pages up above so as not to confuse smatch??

> 
>   1213				if (!bitmap->storage.filemap)
>                                      ^^^^^^^^^^^^^^^^^^^^^^^
> Another check.

Yes, that does look odd.  The check isn't needed any more and I have just
removed it.
Previously if write_page() got an error it would free the filemap.  At that
time there was locking here.
   spin_unlock
   write_page()
   spin_lock
   if (!bitmap->storage.filemap) ....

so it was more obvious that filemap could well change between the write_page
and the test.
When I change write_page to not free the filemap any more, I could remove the
locking.  But I didn't notice that I could remove the test as well.

New code will be in my for-next shortly.

Thanks,
NeilBrown


> 
>   1214					break;
> 
> Really, this isn't a new warning, it's just shows up as a new warning
> because of the rename.  Still, I was curious about the check after the
> dereference.
> 
> regards,
> dan carpenter


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 828 bytes --]