Re: [PATCH v4 21/31] virtio-9p: Fix error_append_hint/error_prepend usage

[Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>]

02.10.2019 12:19, Greg Kurz wrote:
> On Tue,  1 Oct 2019 18:53:09 +0300
> Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> wrote:
> 
>> If we want to add some info to errp (by error_prepend() or
>> error_append_hint()), we must use the ERRP_AUTO_PROPAGATE macro.
>> Otherwise, this info will not be added when errp == &fatal_err
>> (the program will exit prior to the error_append_hint() or
>> error_prepend() call).  Fix such cases.
>>
> 
> Well... this patch doesn't really fix anything because...

I'm sure it fixes. But it also breaks some things you mention below..

> 
>> This commit (together with its neighbors) was generated by
>>
>> git grep -l 'error_\(append_hint\|prepend\)(errp' | while read f; do \
>> spatch --sp-file scripts/coccinelle/fix-error-add-info.cocci \
>> --in-place $f; done
>>
>> and then
>>
>> ./python/commit-per-subsystem.py MAINTAINERS "$(< auto-msg)"
>>
>> (auto-msg was a file with this commit message)
>>
>> and then by hand, for not maintained changed files:
>>
>> git commit -m "<SUB-SYSTEM>: $(< auto-msg)" <FILES>
>>
>> Still, for backporting it may be more comfortable to use only the first
>> command and then do one huge commit.
>>
>> Reported-by: Greg Kurz <groug@kaod.org>
>> Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
>> ---
>>   hw/9pfs/9p-local.c | 1 +
>>   hw/9pfs/9p-proxy.c | 1 +
>>   hw/9pfs/9p.c       | 1 +
>>   3 files changed, 3 insertions(+)
>>
>> diff --git a/hw/9pfs/9p-local.c b/hw/9pfs/9p-local.c
>> index 08e673a79c..fccbf758bd 100644
>> --- a/hw/9pfs/9p-local.c
>> +++ b/hw/9pfs/9p-local.c
>> @@ -1471,6 +1471,7 @@ static void local_cleanup(FsContext *ctx)
>>   
>>   static void error_append_security_model_hint(Error **errp)
>>   {
>> +    ERRP_AUTO_PROPAGATE();
>>       error_append_hint(errp, "Valid options are: security_model="
>>                         "[passthrough|mapped-xattr|mapped-file|none]\n");
>>   }
> 
> This function doesn't need auto propagation in the first place. It is
> simply a wrapper around error_append_hint(). ERRP_AUTO_PROPAGATE()
> should go to the caller local_parse_opts().

Hmm, that's bad. So, actually it's one more errp IN parameter.

> 
> Also some extra care is needed there to handle part [3.] of the
> cleanup. I understand this is out of the scope of that series,
> but I'd rather see all of this fixed in the same patch.
> 
>> diff --git a/hw/9pfs/9p-proxy.c b/hw/9pfs/9p-proxy.c
>> index 57a8c1c808..9291c8efa2 100644
>> --- a/hw/9pfs/9p-proxy.c
>> +++ b/hw/9pfs/9p-proxy.c
>> @@ -1116,6 +1116,7 @@ static int connect_namedsocket(const char *path, Error **errp)
>>   
>>   static void error_append_socket_sockfd_hint(Error **errp)
>>   {
>> +    ERRP_AUTO_PROPAGATE();
>>       error_append_hint(errp, "Either specify socket=/some/path where /some/path"
>>                         " points to a listening AF_UNIX socket or sock_fd=fd"
>>                         " where fd is a file descriptor to a connected AF_UNIX"
> 
> Same here. ERRP_AUTO_PROPAGATE() should go to proxy_parse_opts().

and this one.

> 
>> diff --git a/hw/9pfs/9p.c b/hw/9pfs/9p.c
>> index cce2366219..1df2749e03 100644
>> --- a/hw/9pfs/9p.c
>> +++ b/hw/9pfs/9p.c
>> @@ -3552,6 +3552,7 @@ void pdu_submit(V9fsPDU *pdu, P9MsgHeader *hdr)
>>   int v9fs_device_realize_common(V9fsState *s, const V9fsTransport *t,
>>                                  Error **errp)
>>   {
>> +    ERRP_AUTO_PROPAGATE();
> 
> This is correct since this function calls error_prepend() but I think
> errp is never &error_fatal or &error_abort on the realize path. Anyway,
> better safe than sorry.
> 
>>       int i, len;
>>       struct stat stat;
>>       FsDriverEntry *fse;
> 
> Please drop this patch from your series and I'll do the change once
> ERRP_AUTO_PROPAGATE() gets merged.
> 
> Great thanks for your time in finding a clever way to deal with error
> propagation. :)
> 

Hmm, actual question is: how many other errp IN parameters I've broken in
these series?

I can't simply drop one patch from atomatically genereated series: how people
will backport it than, or understand how to backport other crossing things?

Actually, for the cases you mean, ERRP_AUTO_PROPAGATE is no-op, as it uses original
errp if it's not NULL and don't point to error_fatal.

So, nothing is broken here, may be we can proceed as is? But I agree, it looks strange.

Eric, what do you think?

I tried to change cocci script to

   @rule0@
   // Add invocation to errp-functions
~ identifier fn, fn2;
   @@

    fn(..., Error **errp, ...)
    {
   +   ERRP_AUTO_PROPAGATE();
       <+...
+     fn2(..., errp, ...)
+     ...
   (
       error_append_hint(errp, ...);
   |
       error_prepend(errp, ...);
   )
       ...+>
    }


but it stubs on hw/vfio/pci.c... and it skips a lot of valid cases, so it's wrong.



-- 
Best regards,
Vladimir