Re: [PATCH 3/6] io_uring: allow allocated fixed files for openat/openat2

[Jens Axboe <axboe@kernel.dk>]

On 5/12/22 10:38 PM, Hao Xu wrote:
> ? 2022/5/9 ??11:50, Jens Axboe ??:
>> If the application passes in IORING_FILE_INDEX_ALLOC as the file_slot,
>> then that's a hint to allocate a fixed file descriptor rather than have
>> one be passed in directly.
>>
>> This can be useful for having io_uring manage the direct descriptor space.
>>
>> Normal open direct requests will complete with 0 for success, and < 0
>> in case of error. If io_uring is asked to allocated the direct descriptor,
>> then the direct descriptor is returned in case of success.
>>
>> Signed-off-by: Jens Axboe <axboe@kernel.dk>
>> ---
>>   fs/io_uring.c                 | 32 +++++++++++++++++++++++++++++---
>>   include/uapi/linux/io_uring.h |  9 +++++++++
>>   2 files changed, 38 insertions(+), 3 deletions(-)
>>
>> diff --git a/fs/io_uring.c b/fs/io_uring.c
>> index 8c40411a7e78..ef999d0e09de 100644
>> --- a/fs/io_uring.c
>> +++ b/fs/io_uring.c
>> @@ -4697,7 +4697,7 @@ static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
>>       return __io_openat_prep(req, sqe);
>>   }
>>   -static int __maybe_unused io_file_bitmap_get(struct io_ring_ctx *ctx)
>> +static int io_file_bitmap_get(struct io_ring_ctx *ctx)
>>   {
>>       struct io_file_table *table = &ctx->file_table;
>>       unsigned long nr = ctx->nr_user_files;
>> @@ -4722,6 +4722,32 @@ static int __maybe_unused io_file_bitmap_get(struct io_ring_ctx *ctx)
>>       return -ENFILE;
>>   }
>>   +static int io_fixed_fd_install(struct io_kiocb *req, unsigned int issue_flags,
>> +                   struct file *file, unsigned int file_slot)
>> +{
>> +    int alloc_slot = file_slot == IORING_FILE_INDEX_ALLOC;
>> +    struct io_ring_ctx *ctx = req->ctx;
>> +    int ret;
>> +
>> +    if (alloc_slot) {
>> +        io_ring_submit_lock(ctx, issue_flags);
>> +        file_slot = io_file_bitmap_get(ctx);
>> +        if (unlikely(file_slot < 0)) {
>> +            io_ring_submit_unlock(ctx, issue_flags);
>> +            return file_slot;
>> +        }
>> +    }
>> +
>> +    ret = io_install_fixed_file(req, file, issue_flags, file_slot);
>> +    if (alloc_slot) {
>> +        io_ring_submit_unlock(ctx, issue_flags);
>> +        if (!ret)
>> +            return file_slot;
> 
> Sorry, I missed onething, looks like this should be file_slot+1, as this
> is returned to the userspace. I refer to the previous open/accept direct
> feature, they see the file_index from userspace as number counted from
> one, so it'd better to keep it consistent.

We need to return the actual slot, not slot + 1. It's kind of an ugly
API, but what we had left to deal with. The actual slot number is what
most other things would fill into sqe->fd and set IOSQE_FIXED_FILE with,
not slot + 1.

-- 
Jens Axboe