[PATCH] pci: add O_CLOEXEC when open uio device

[PATCH] pci: add O_CLOEXEC when open uio device

[Weifeng Su]

In this scenario, the DPDK process invokes a script which
inherits an open file descriptor (FD) for a UIO device.
After the script execution is complete, the UIO device's
close operation is called. However, in a new kernel version
(865a11f987ab5f03:uio/uio_pci_generic: Disable bus-mastering on release),
this close operation causes the PCI bus master bit to be cleared,
rendering the device unusable and leading to unexpected behavior.
This modification was made to prevent the UIO device's FD
from being inherited by the child process.
Cc: stable@dpdk.org

Signed-off-by: Weifeng Su <suweifeng1@huawei.com>
---
 drivers/bus/pci/linux/pci_uio.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/bus/pci/linux/pci_uio.c b/drivers/bus/pci/linux/pci_uio.c
index d52125e49b..7ac142c36e 100644
--- a/drivers/bus/pci/linux/pci_uio.c
+++ b/drivers/bus/pci/linux/pci_uio.c
@@ -246,7 +246,7 @@ pci_uio_alloc_resource(struct rte_pci_device *dev,
 	snprintf(devname, sizeof(devname), "/dev/uio%u", uio_num);
 
 	/* save fd if in primary process */
-	fd = open(devname, O_RDWR);
+	fd = open(devname, O_RDWR | O_CLOEXEC);
 	if (fd < 0) {
 		RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
 			devname, strerror(errno));
-- 
2.18.0.windows.1

Re: [PATCH] pci: add O_CLOEXEC when open uio device

[Stephen Hemminger]

On Thu, 25 May 2023 11:00:35 +0800
Weifeng Su <suweifeng1@huawei.com> wrote:

> In this scenario, the DPDK process invokes a script which
> inherits an open file descriptor (FD) for a UIO device.
> After the script execution is complete, the UIO device's
> close operation is called. However, in a new kernel version
> (865a11f987ab5f03:uio/uio_pci_generic: Disable bus-mastering on release),
> this close operation causes the PCI bus master bit to be cleared,
> rendering the device unusable and leading to unexpected behavior.
> This modification was made to prevent the UIO device's FD
> from being inherited by the child process.
> Cc: stable@dpdk.org
> 
> Signed-off-by: Weifeng Su <suweifeng1@huawei.com>

Makes sense that UIO fd's would be marked close on exec.
But should the interrupt fd, and all the other fd's which EAL leaves.

DPDK internal code doesn't invoke scripts. If your application is
going to do so then it needs to lots more cleanup. Probably some
variant of closing all fd's.