Re: [PATCH RFC v6 05/16] fuse: make args->in_args[0] to be always the header

[Bernd Schubert <bernd.schubert@fastmail.fm>]

On 11/23/24 10:01, Miklos Szeredi wrote:
> On Fri, 22 Nov 2024 at 00:44, Bernd Schubert <bschubert@ddn.com> wrote:
> 
>> diff --git a/fs/fuse/dax.c b/fs/fuse/dax.c
>> index 12ef91d170bb3091ac35a33d2b9dc38330b00948..e459b8134ccb089f971bebf8da1f7fc5199c1271 100644
>> --- a/fs/fuse/dax.c
>> +++ b/fs/fuse/dax.c
>> @@ -237,14 +237,17 @@ static int fuse_send_removemapping(struct inode *inode,
>>         struct fuse_inode *fi = get_fuse_inode(inode);
>>         struct fuse_mount *fm = get_fuse_mount(inode);
>>         FUSE_ARGS(args);
>> +       struct fuse_zero_in zero_arg;
> 
> I'd move this to global scope (i.e. just a single instance for all
> uses) and rename to zero_header.
> 
>> diff --git a/fs/fuse/dev.c b/fs/fuse/dev.c
>> index fd8898b0c1cca4d117982d5208d78078472b0dfb..6cb45b5332c45f322e9163469ffd114cbc07dc4f 100644
>> --- a/fs/fuse/dev.c
>> +++ b/fs/fuse/dev.c
>> @@ -1053,6 +1053,19 @@ static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
>>
>>         for (i = 0; !err && i < numargs; i++)  {
>>                 struct fuse_arg *arg = &args[i];
>> +
>> +               /* zero headers */
>> +               if (arg->size == 0) {
>> +                       if (WARN_ON_ONCE(i != 0)) {
>> +                               if (cs->req)
>> +                                       pr_err_once(
>> +                                               "fuse: zero size header in opcode %d\n",
>> +                                               cs->req->in.h.opcode);
>> +                               return -EINVAL;
>> +                       }
> 
> Just keep the WARN_ON_ONCE() and drop everything else, including
> return -EINVAL.  The same thing should happen without the arg->size ==
> 0 check

I have to remove the WARN_ON_ONCE condition altogether, gets triggered by
/dev/fuse read (i.e. with io-uring being disabled), in generic/062, 
op code=39 (FUSE_IOCTL).
Without the pr_err_once() and printing the op code it would have been impossible
to see which op code that is - the trace does not help here.


Thanks,
Bernd