Re: [bug report] fsnotify: pass dir and inode arguments to fsnotify()

[Amir Goldstein <amir73il@gmail.com>]

On Thu, Jul 30, 2020 at 2:13 PM <dan.carpenter@oracle.com> wrote:
>
> Hello Amir Goldstein,
>
> This is a semi-automatic email about new static checker warnings.
>
> The patch 40a100d3adc1: "fsnotify: pass dir and inode arguments to
> fsnotify()" from Jul 22, 2020, leads to the following Smatch
> complaint:

That's an odd report, because...

>
>     fs/notify/fsnotify.c:460 fsnotify()
>     warn: variable dereferenced before check 'inode' (see line 449)
>
> fs/notify/fsnotify.c
>    448          }
>    449          sb = inode->i_sb;
>                      ^^^^^^^^^^^
> New dreference.

First of all, two lines above we have
if (!inode) inode = dir;

This function does not assert (inode || dir), but must it??
This is even documented:

 * @inode:      optional inode associated with event -
 *              either @dir or @inode must be non-NULL.

Second,
The line above was indeed added by:
40a100d3adc1: "fsnotify: pass dir and inode arguments to fsnotify()"

However...

>
>    450
>    451          /*
>    452           * Optimization: srcu_read_lock() has a memory barrier which can
>    453           * be expensive.  It protects walking the *_fsnotify_marks lists.
>    454           * However, if we do not walk the lists, we do not have to do
>    455           * SRCU because we have no references to any objects and do not
>    456           * need SRCU to keep them "alive".
>    457           */
>    458          if (!sb->s_fsnotify_marks &&
>    459              (!mnt || !mnt->mnt_fsnotify_marks) &&
>    460              (!inode || !inode->i_fsnotify_marks) &&
>                      ^^^^^^
> Check too late.  Presumably this check can be removed?

But this line was only added later by:
9b93f33105f5 fsnotify: send event with parent/name info to
sb/mount/non-dir marks

So, yes, the check could be removed.
It is a leftover from a previous revision, but even though it is a leftover
I kind of like the code better this way.

In principle, an event on sb/mnt that is not associated with any inode
(for example
FS_UNMOUNT) could be added in the future.
And then we will have to fix documentation and the inode dereference above.

In any case, thank you for the report, but I don't see a reason to make any
changes right now.

Thanks,
Amir.