Re: [PATCH v2] gpib: fmh_gpib: Fix resource leaks in fmh_gpib_attach_impl

[Hongling Zeng <zhongling0719@126.com>]

   Hi Dominik Karol,

   Thanks for the analysis. You're right about the double-free issue.

   However, there's still a real leak : when dma_fifos resource
   lookup fails, the previously allocated gpib_iomem_res leaks because
   dma_port_res isn't assigned yet, so detach() won't clean it up.

   Minimal fix: cleanup only before field assignment:
```c
   if (!res) {
       dev_err(board->dev, "Unable to locate mmio resource for gpib dma 
port\n");
       release_mem_region(e_priv->gpib_iomem_res->start,
                         resource_size(e_priv->gpib_iomem_res));
       return -ENODEV;
   }

   This fixes the leak while avoiding double-free. Acceptable approach?

   Best regards,
   Hongling Zeng


在 2026年05月29日 00:46, Dominik Karol Piątkowski 写道:
> Hi Hongling Zeng,
>
> On Thursday, May 28th, 2026 at 03:50, Hongling Zeng <zenghongling@kylinos.cn> wrote:
>
>> The fmh_gpib_attach_impl() function has multiple resource leaks in its
>> error handling paths. When any initialization step fails, the function
>> returns early without properly releasing previously acquired resources.
>>
>> Fix by adding proper error handling labels and cleanup code that releases
>> resources in the reverse order they were acquired.
>>
>> Fixes: 8e4841a0888c7 ("staging: gpib: Add Frank Mori Hess FPGA PCI GPIB driver")
>> Signed-off-by: Hongling Zeng <zenghongling@kylinos.cn>
>> Cc: stable@vger.kernel.org
>> Suggested-by: Dominik Karol Piątkowski <dominik.karol.piatkowski@protonmail.com>
>>
>> ---
>> Changes in v2:
>>   - Fixed unnecessary retval assignments in early error returns
>>   - Removed extra newline
>>   - Kept e_priv->irq assignment after request_irq() succeeds,as suggested by Dominik Karol
>> ---
>>   drivers/gpib/fmh_gpib/fmh_gpib.c | 37 +++++++++++++++++++++++++-------
>>   1 file changed, 29 insertions(+), 8 deletions(-)
>>
>> diff --git a/drivers/gpib/fmh_gpib/fmh_gpib.c b/drivers/gpib/fmh_gpib/fmh_gpib.c
>> index fcafdc02ea2e..404379cd1565 100644
>> --- a/drivers/gpib/fmh_gpib/fmh_gpib.c
>> +++ b/drivers/gpib/fmh_gpib/fmh_gpib.c
>> @@ -1418,7 +1418,8 @@ static int fmh_gpib_attach_impl(struct gpib_board *board, const struct gpib_boar
>>   				     resource_size(e_priv->gpib_iomem_res));
>>   	if (!nec_priv->mmiobase) {
>>   		dev_err(board->dev, "Could not map I/O memory\n");
>> -		return -ENOMEM;
>> +		retval = -ENOMEM;
>> +		goto err_release_gpib_region;
>>   	}
>>   	dev_dbg(board->dev, "iobase %pr remapped to %p\n",
>>   		e_priv->gpib_iomem_res, nec_priv->mmiobase);
>> @@ -1426,34 +1427,39 @@ static int fmh_gpib_attach_impl(struct gpib_board *board, const struct gpib_boar
>>   	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "dma_fifos");
>>   	if (!res) {
>>   		dev_err(board->dev, "Unable to locate mmio resource for gpib dma port\n");
>> -		return -ENODEV;
>> +		retval = -ENODEV;
>> +		goto err_iounmap_gpib;
>>   	}
>>   	if (request_mem_region(res->start,
>>   			       resource_size(res),
>>   			       pdev->name) == NULL) {
>>   		dev_err(board->dev, "cannot claim registers\n");
>> -		return -ENXIO;
>> +		retval = -ENXIO;
>> +		goto err_iounmap_gpib;
>>   	}
>>   	e_priv->dma_port_res = res;
>>   	e_priv->fifo_base = ioremap(e_priv->dma_port_res->start,
>>   				    resource_size(e_priv->dma_port_res));
>>   	if (!e_priv->fifo_base) {
>>   		dev_err(board->dev, "Could not map I/O memory for fifos\n");
>> -		return -ENOMEM;
>> +		retval = -ENOMEM;
>> +		goto err_release_dma_region;
>>   	}
>>   	dev_dbg(board->dev, "dma fifos 0x%lx remapped to %p, length=%ld\n",
>>   		(unsigned long)e_priv->dma_port_res->start, e_priv->fifo_base,
>>   		(unsigned long)resource_size(e_priv->dma_port_res));
>>
>>   	irq = platform_get_irq(pdev, 0);
>> -	if (irq < 0)
>> -		return -EBUSY;
>> +	if (irq < 0) {
>> +		retval = -EBUSY;
>> +		goto err_iounmap_fifo;
>> +	}
>>   	retval = request_irq(irq, fmh_gpib_interrupt, IRQF_SHARED, pdev->name, board);
>>   	if (retval) {
>>   		dev_err(board->dev,
>>   			"cannot register interrupt handler err=%d\n",
>>   			retval);
>> -		return retval;
>> +		goto err_iounmap_fifo;
>>   	}
>>   	e_priv->irq = irq;
>>
>> @@ -1461,7 +1467,8 @@ static int fmh_gpib_attach_impl(struct gpib_board *board, const struct gpib_boar
>>   		e_priv->dma_channel = dma_request_slave_channel(board->dev, "rxtx");
>>   		if (!e_priv->dma_channel) {
>>   			dev_err(board->dev, "failed to acquire dma channel \"rxtx\".\n");
>> -			return -EIO;
>> +			retval = -EIO;
>> +			goto err_free_irq;
>>   		}
>>   	}
>>   	/*
>> @@ -1473,6 +1480,20 @@ static int fmh_gpib_attach_impl(struct gpib_board *board, const struct gpib_boar
>>   		fifo_max_burst_length_mask;
>>
>>   	return fmh_gpib_init(e_priv, board, handshake_mode);
>> +
>> +err_free_irq:
>> +	free_irq(e_priv->irq, board);
>> +err_iounmap_fifo:
>> +	iounmap(e_priv->fifo_base);
>> +err_release_dma_region:
>> +	release_mem_region(e_priv->dma_port_res->start,
>> +				resource_size(e_priv->dma_port_res));
>> +err_iounmap_gpib:
>> +	iounmap(nec_priv->mmiobase);
>> +err_release_gpib_region:
>> +	release_mem_region(e_priv->gpib_iomem_res->start,
>> +				resource_size(e_priv->gpib_iomem_res));
>> +	return retval;
> I see a problem with this patch.
>
> fmh_gpib_attach_impl is called from fmh_gpib_attach_holdoff_(all|end), and these
> are passed as attach through fmh_gpib_(unaccel_)interface. The only place where
> I see attach is called, is in common/iblib.c, in ibonline function. If attach
> fails (that is, returns with -errno), detach is called (fmh_gpib_detach for
> this case), where a proper cleanup is performed.
>
> If we release the resources in attach and not clear corresponding e_priv fields,
> we will fail badly in detach, as we do double free/unmap and use after free.
> Clearing e_priv fields will result in having two competing cleanup routines,
> and I don't like that idea - let's have one proper cleanup in one place. The
> only way to keep this approach would be to rewrite the core gpib logic for
> attach/detach, and I don't know if it's worth the effort - probably not.
>
> Thanks,
> Dominik Karol
>
>>   }
>>
>>   int fmh_gpib_attach_holdoff_all(struct gpib_board *board, const struct gpib_board_config *config)
>> --
>> 2.25.1
>>
>>