Re: [PATCHSET v2 0/6] Allow allocated direct descriptors

[Hao Xu <hao.xu@linux.dev>]

Hi Stefan,
On 6/9/22 16:57, Stefan Metzmacher wrote:
> 
> Hi Jens,
> 
> this looks very useful, thanks!
> 
> I have an additional feature request to make this even more useful...
> 
> IO_OP_ACCEPT allows a fixed descriptor for the listen socket
> and then can generate a fixed descriptor for the accepted connection,
> correct?

Yes.

> 
> It would be extremely useful to also allow that pattern
> for IO_OP_OPENAT[2], which currently is not able to get
> a fixed descriptor for the dirfd argument (this also applies to
> IO_OP_STATX, IO_OP_UNLINK and all others taking a dirfd).
> 
> Being able use such a sequence:
> 
> OPENTAT2(AT_FDCWD, "directory") => 1 (fixed)
> STATX(1 (fixed))
> FGETXATTR(1 (fixed)
> OPENAT2(1 (fixed), "file") => 2 (fixed)
> STATX(2 (fixed))
> FGETXATTR(2 (fixed))
> CLOSE(1 (fixed)
> DUP( 2 (fixed)) => per-process fd for ("file")
> 
> I looked briefly how to implement that.
> But set_nameidata() takes 'int dfd' to store the value
> and it's used later somewhere deep down the stack.
> And makes it too complex for me to create patches :-(
> 

Indeed.. dirfd is used in path_init() etc. For me, no idea how to tackle
it for now.We surely can register a fixed descriptor to the process
fdtable but that is against the purpose of fixed file..

> It would great if someone could have a look how to make this work.
> 
> 
>> Currently using direct descriptors with open or accept requires the
>> application to manage the descriptor space, picking which slot to use
>> for any given file. However, there are cases where it's useful to just
>> get a direct descriptor and not care about which value it is, instead
>> just return it like a normal open or accept would.
>>
>> This will also be useful for multishot accept support, where allocated
>> direct descriptors are a requirement to make that feature work with
>> these kinds of files.
>>
>> This adds support for allocating a new fixed descriptor. This is chosen
>> by passing in IORING_FILE_INDEX_ALLOC as the fixed slot, which is beyond
>> any valid value for the file_index.
> 
> I guess that would not work for linked request, as we don't know the
> actual id before submitting the requests.
> 
> Thanks!
> metze