[PATCH v2 5.15] io_uring: auto-removal for direct open/accept

[PATCH v2 5.15] io_uring: auto-removal for direct open/accept

[Pavel Begunkov]

It might be inconvenient that direct open/accept deviates from the
update semantics and fails if the slot is taken instead of removing a
file sitting there. Implement this auto-removal.

Note that removal might need to allocate and so may fail. However, if an
empty slot is specified, it's guaraneed to not fail on the fd
installation side for valid userspace programs. It's needed for users
who can't tolerate such failures, e.g. accept where the other end
never retries.

Suggested-by: Franz-B. Tuneke <franz-bernhard.tuneke@tu-dortmund.de>
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
---

v2: simplify io_rsrc_node_switch_start() handling

 fs/io_uring.c | 52 +++++++++++++++++++++++++++++++++------------------
 1 file changed, 34 insertions(+), 18 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index a864a94364c6..58c0cbfdd128 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -8287,11 +8287,27 @@ static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file,
 #endif
 }
 
+static int io_queue_rsrc_removal(struct io_rsrc_data *data, unsigned idx,
+				 struct io_rsrc_node *node, void *rsrc)
+{
+	struct io_rsrc_put *prsrc;
+
+	prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL);
+	if (!prsrc)
+		return -ENOMEM;
+
+	prsrc->tag = *io_get_tag_slot(data, idx);
+	prsrc->rsrc = rsrc;
+	list_add(&prsrc->list, &node->rsrc_list);
+	return 0;
+}
+
 static int io_install_fixed_file(struct io_kiocb *req, struct file *file,
 				 unsigned int issue_flags, u32 slot_index)
 {
 	struct io_ring_ctx *ctx = req->ctx;
 	bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
+	bool needs_switch = false;
 	struct io_fixed_file *file_slot;
 	int ret = -EBADF;
 
@@ -8307,9 +8323,22 @@ static int io_install_fixed_file(struct io_kiocb *req, struct file *file,
 
 	slot_index = array_index_nospec(slot_index, ctx->nr_user_files);
 	file_slot = io_fixed_file_slot(&ctx->file_table, slot_index);
-	ret = -EBADF;
-	if (file_slot->file_ptr)
-		goto err;
+
+	if (file_slot->file_ptr) {
+		struct file *old_file;
+
+		ret = io_rsrc_node_switch_start(ctx);
+		if (ret)
+			goto err;
+
+		old_file = (struct file *)(file_slot->file_ptr & FFS_MASK);
+		ret = io_queue_rsrc_removal(ctx->file_data, slot_index,
+					    ctx->rsrc_node, old_file);
+		if (ret)
+			goto err;
+		file_slot->file_ptr = 0;
+		needs_switch = true;
+	}
 
 	*io_get_tag_slot(ctx->file_data, slot_index) = 0;
 	io_fixed_file_set(file_slot, file);
@@ -8321,27 +8350,14 @@ static int io_install_fixed_file(struct io_kiocb *req, struct file *file,
 
 	ret = 0;
 err:
+	if (needs_switch)
+		io_rsrc_node_switch(ctx, ctx->file_data);
 	io_ring_submit_unlock(ctx, !force_nonblock);
 	if (ret)
 		fput(file);
 	return ret;
 }
 
-static int io_queue_rsrc_removal(struct io_rsrc_data *data, unsigned idx,
-				 struct io_rsrc_node *node, void *rsrc)
-{
-	struct io_rsrc_put *prsrc;
-
-	prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL);
-	if (!prsrc)
-		return -ENOMEM;
-
-	prsrc->tag = *io_get_tag_slot(data, idx);
-	prsrc->rsrc = rsrc;
-	list_add(&prsrc->list, &node->rsrc_list);
-	return 0;
-}
-
 static int __io_sqe_files_update(struct io_ring_ctx *ctx,
 				 struct io_uring_rsrc_update2 *up,
 				 unsigned nr_args)
-- 
2.33.0

Re: [PATCH v2 5.15] io_uring: auto-removal for direct open/accept

[Jens Axboe]

On 9/14/21 9:12 AM, Pavel Begunkov wrote:
> It might be inconvenient that direct open/accept deviates from the
> update semantics and fails if the slot is taken instead of removing a
> file sitting there. Implement this auto-removal.
> 
> Note that removal might need to allocate and so may fail. However, if an
> empty slot is specified, it's guaraneed to not fail on the fd
> installation side for valid userspace programs. It's needed for users
> who can't tolerate such failures, e.g. accept where the other end
> never retries.

Can you submit a liburing test case for this too?

-- 
Jens Axboe

Re: [PATCH v2 5.15] io_uring: auto-removal for direct open/accept

[Pavel Begunkov]

On 9/14/21 4:14 PM, Jens Axboe wrote:
> On 9/14/21 9:12 AM, Pavel Begunkov wrote:
>> It might be inconvenient that direct open/accept deviates from the
>> update semantics and fails if the slot is taken instead of removing a
>> file sitting there. Implement this auto-removal.
>>
>> Note that removal might need to allocate and so may fail. However, if an
>> empty slot is specified, it's guaraneed to not fail on the fd
>> installation side for valid userspace programs. It's needed for users
>> who can't tolerate such failures, e.g. accept where the other end
>> never retries.
> 
> Can you submit a liburing test case for this too?

Made some for v1, I just sent out them.

-- 
Pavel Begunkov

Re: [PATCH v2 5.15] io_uring: auto-removal for direct open/accept

[Jens Axboe]

On 9/14/21 9:12 AM, Pavel Begunkov wrote:
> It might be inconvenient that direct open/accept deviates from the
> update semantics and fails if the slot is taken instead of removing a
> file sitting there. Implement this auto-removal.
> 
> Note that removal might need to allocate and so may fail. However, if an
> empty slot is specified, it's guaraneed to not fail on the fd
> installation side for valid userspace programs. It's needed for users
> who can't tolerate such failures, e.g. accept where the other end
> never retries.
> 
> Suggested-by: Franz-B. Tuneke <franz-bernhard.tuneke@tu-dortmund.de>
> Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
> ---
> 
> v2: simplify io_rsrc_node_switch_start() handling
> 
>  fs/io_uring.c | 52 +++++++++++++++++++++++++++++++++------------------
>  1 file changed, 34 insertions(+), 18 deletions(-)
> 
> diff --git a/fs/io_uring.c b/fs/io_uring.c
> index a864a94364c6..58c0cbfdd128 100644
> --- a/fs/io_uring.c
> +++ b/fs/io_uring.c
> @@ -8287,11 +8287,27 @@ static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file,
>  #endif
>  }
>  
> +static int io_queue_rsrc_removal(struct io_rsrc_data *data, unsigned idx,
> +				 struct io_rsrc_node *node, void *rsrc)
> +{
> +	struct io_rsrc_put *prsrc;
> +
> +	prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL);
> +	if (!prsrc)
> +		return -ENOMEM;
> +
> +	prsrc->tag = *io_get_tag_slot(data, idx);
> +	prsrc->rsrc = rsrc;
> +	list_add(&prsrc->list, &node->rsrc_list);
> +	return 0;
> +}

I know this code is just being moved, but I tend to like making the
expected/fast path inline:

prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL);
if (prsrc) {
	prsrc->tag = *io_get_tag_slot(data, idx);
	prsrc->rsrc = rsrc;
	list_add(&prsrc->list, &node->rsrc_list);
	return 0;
}
return -ENOMEM;

-- 
Jens Axboe

Re: [PATCH v2 5.15] io_uring: auto-removal for direct open/accept

[Pavel Begunkov]

On 9/14/21 4:36 PM, Jens Axboe wrote:
> On 9/14/21 9:12 AM, Pavel Begunkov wrote:
>> It might be inconvenient that direct open/accept deviates from the
>> update semantics and fails if the slot is taken instead of removing a
>> file sitting there. Implement this auto-removal.
>>
>> Note that removal might need to allocate and so may fail. However, if an
>> empty slot is specified, it's guaraneed to not fail on the fd
>> installation side for valid userspace programs. It's needed for users
>> who can't tolerate such failures, e.g. accept where the other end
>> never retries.
>>
>> Suggested-by: Franz-B. Tuneke <franz-bernhard.tuneke@tu-dortmund.de>
>> Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
>> ---
>>
>> v2: simplify io_rsrc_node_switch_start() handling
>>
>>  fs/io_uring.c | 52 +++++++++++++++++++++++++++++++++------------------
>>  1 file changed, 34 insertions(+), 18 deletions(-)
>>
>> diff --git a/fs/io_uring.c b/fs/io_uring.c
>> index a864a94364c6..58c0cbfdd128 100644
>> --- a/fs/io_uring.c
>> +++ b/fs/io_uring.c
>> @@ -8287,11 +8287,27 @@ static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file,
>>  #endif
>>  }
>>  
>> +static int io_queue_rsrc_removal(struct io_rsrc_data *data, unsigned idx,
>> +				 struct io_rsrc_node *node, void *rsrc)
>> +{
>> +	struct io_rsrc_put *prsrc;
>> +
>> +	prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL);
>> +	if (!prsrc)
>> +		return -ENOMEM;
>> +
>> +	prsrc->tag = *io_get_tag_slot(data, idx);
>> +	prsrc->rsrc = rsrc;
>> +	list_add(&prsrc->list, &node->rsrc_list);
>> +	return 0;
>> +}
> 
> I know this code is just being moved, but I tend to like making the
> expected/fast path inline:

I think it's more natural as now, as we always have

ret = do_something();
if (ret)
    return ret;
ret = do_something2();
if (ret)
    goto err;


And I remember you telling once "... I tend to like to do the error
path like that unless it's a hot path ...". So maybe just add
unlikely()?


> prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL);
> if (prsrc) {
> 	prsrc->tag = *io_get_tag_slot(data, idx);
> 	prsrc->rsrc = rsrc;
> 	list_add(&prsrc->list, &node->rsrc_list);
> 	return 0;
> }
> return -ENOMEM;
> 

-- 
Pavel Begunkov

Re: [PATCH v2 5.15] io_uring: auto-removal for direct open/accept

[Jens Axboe]

On 9/14/21 9:48 AM, Pavel Begunkov wrote:
> On 9/14/21 4:36 PM, Jens Axboe wrote:
>> On 9/14/21 9:12 AM, Pavel Begunkov wrote:
>>> It might be inconvenient that direct open/accept deviates from the
>>> update semantics and fails if the slot is taken instead of removing a
>>> file sitting there. Implement this auto-removal.
>>>
>>> Note that removal might need to allocate and so may fail. However, if an
>>> empty slot is specified, it's guaraneed to not fail on the fd
>>> installation side for valid userspace programs. It's needed for users
>>> who can't tolerate such failures, e.g. accept where the other end
>>> never retries.
>>>
>>> Suggested-by: Franz-B. Tuneke <franz-bernhard.tuneke@tu-dortmund.de>
>>> Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
>>> ---
>>>
>>> v2: simplify io_rsrc_node_switch_start() handling
>>>
>>>  fs/io_uring.c | 52 +++++++++++++++++++++++++++++++++------------------
>>>  1 file changed, 34 insertions(+), 18 deletions(-)
>>>
>>> diff --git a/fs/io_uring.c b/fs/io_uring.c
>>> index a864a94364c6..58c0cbfdd128 100644
>>> --- a/fs/io_uring.c
>>> +++ b/fs/io_uring.c
>>> @@ -8287,11 +8287,27 @@ static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file,
>>>  #endif
>>>  }
>>>  
>>> +static int io_queue_rsrc_removal(struct io_rsrc_data *data, unsigned idx,
>>> +				 struct io_rsrc_node *node, void *rsrc)
>>> +{
>>> +	struct io_rsrc_put *prsrc;
>>> +
>>> +	prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL);
>>> +	if (!prsrc)
>>> +		return -ENOMEM;
>>> +
>>> +	prsrc->tag = *io_get_tag_slot(data, idx);
>>> +	prsrc->rsrc = rsrc;
>>> +	list_add(&prsrc->list, &node->rsrc_list);
>>> +	return 0;
>>> +}
>>
>> I know this code is just being moved, but I tend to like making the
>> expected/fast path inline:
> 
> I think it's more natural as now, as we always have
> 
> ret = do_something();
> if (ret)
>     return ret;
> ret = do_something2();
> if (ret)
>     goto err;
> 
> 
> And I remember you telling once "... I tend to like to do the error
> path like that unless it's a hot path ...". So maybe just add
> unlikely()?

Yeah doesn't matter too much for a non-fast path, and unlikely() here
won't matter as well. Let's just use it as-is.

-- 
Jens Axboe