firewire: Replace ENOSYS with appropriate error codes
PATCH 1/1 firewire: Replace ENOSYS with appropriate error codes
ENOSYS is reserved for "invalid syscall number" and should not be used for other error conditions. Replace incorrect usages with more appropriate error codes:
- In sbp2.c: Use -EOPNOTSUPP for unsupported operation (re-adding logical units via SCSI stack).
- In ohci.c: Use -EINVAL for invalid ISO context types in switch statements, and -EOPNOTSUPP for unsupported Pinnacle MovieBoard hardware.
- In core-cdev.c: Use -EACCES for access policy violations when operations are restricted to local nodes' device files.
--- drivers/firewire/core-cdev.c | 6 +++--- drivers/firewire/ohci.c | 8 ++++---- drivers/firewire/sbp2.c | 2 +- 3 files changed, 8 insertions(+), 8 deletions(-)
drivers/firewire/core-cdev.c +3 -3
@@ -935,7 +935,7 @@ static int ioctl_add_descriptor(struct client *client, union ioctl_arg *arg)/* Access policy: Allow this ioctl only on local nodes' device files. */ if (!client->device->is_local) -return -ENOSYS ; +return -EACCES ;if (a->length >256 )return -EINVAL ;
@@ -1628,7 +1628,7 @@ static int ioctl_send_phy_packet(struct client *client, union ioctl_arg *arg)/* Access policy: Allow this ioctl only on local nodes' device files. */ if (!client->device->is_local) -return -ENOSYS ; +return -EACCES ; e =kzalloc (sizeof (*e) +sizeof (a->data),GFP_KERNEL );if (e ==NULL )
@@ -1676,7 +1676,7 @@ static int ioctl_receive_phy_packets(struct client *client, union ioctl_arg *arg/* Access policy: Allow this ioctl only on local nodes' device files. */ if (!client->device->is_local) -return -ENOSYS ; +return -EACCES ;// NOTE: This can be without irq when we can guarantee that __fw_send_request() for local // destination never runs in any type of IRQ context.
drivers/firewire/ohci.c +4 -4
@@ -2918,7 +2918,7 @@ static struct fw_iso_context *ohci_allocate_iso_context(struct fw_card *card,default : index =-1 ; - ret = -ENOSYS ; + ret = -EINVAL ; }if (index <0 )
@@ -3370,7 +3370,7 @@ static int ohci_queue_iso(struct fw_iso_context *base,case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL :return queue_iso_buffer_fill (ctx, packet, buffer, payload);default : -return -ENOSYS ; +return -EINVAL ; } }
@@ -3401,7 +3401,7 @@ static int ohci_flush_iso_completions(struct fw_iso_context *base)flush_ir_buffer_fill (ctx);break ;default : - ret = -ENOSYS ; + ret = -EINVAL ; }clear_bit_unlock (0 , &ctx->flushing_completions);
@@ -3485,7 +3485,7 @@ static int pci_probe(struct pci_dev *dev,if (dev->vendor ==PCI_VENDOR_ID_PINNACLE_SYSTEMS ) {dev_err (&dev->dev,"Pinnacle MovieBoard is not yet supported\n" ); -return -ENOSYS ; +return -EOPNOTSUPP ; } ohci =devres_alloc (release_ohci,sizeof (*ohci),GFP_KERNEL );
drivers/firewire/sbp2.c +1 -1
@@ -1496,7 +1496,7 @@ static int sbp2_scsi_sdev_init(struct scsi_device *sdev)/* (Re-)Adding logical units via the SCSI stack is not supported. */ if (!lu) -return -ENOSYS ; +return -EOPNOTSUPP ; sdev->allow_restart =1 ;
Signature · 1 line
2.48.1
Takashi Sakamoto
maintainerHi,
Quoted 3 lines · Nirbhay Sharma
ENOSYS is reserved for "invalid syscall number" and should not be used for other error conditions. Replace incorrect usages with more appropriate error codes:
Yes. The newly-written code should not use ENOSYS for cadual use, indeed.
Quoted 10 lines
- In
sbp2.c: Use-EOPNOTSUPPfor unsupported operation (re-adding logical units via SCSI stack).- In
ohci.c: Use-EINVALfor invalid ISO context types in switch statements, and-EOPNOTSUPPfor unsupported Pinnacle MovieBoard hardware.- In
core-cdev.c: Use-EACCESfor access policy violations when operations are restricted to local nodes' device files.
Quoted 6 lines
Signed-off-by Nirbhay Sharma <[email protected]>--- drivers/firewire/core-cdev.c | 6 +++--- drivers/firewire/ohci.c | 8 ++++---- drivers/firewire/sbp2.c | 2 +- 3 files changed, 8 insertions(+), 8 deletions(-)
There is a rest to discuss when changing existing code in respect to this topic, since it brings loss of backward-compatibility to userspace software. In this reason, I've left them as is.
If there are any strong and specific reasons to correct them, let us change them. Do you have such reasons? For example, Linux kernel developer have shared the consensus and decision to ostracize such codes?
Thanks
Takashi Sakamoto
Quoted 9 lines · Takashi Sakamoto
Hi,
Yes. The newly-written code should not use ENOSYS for cadual use, indeed.
There is a rest to discuss when changing existing code in respect to this topic, since it brings loss of backward-compatibility to userspace software. In this reason, I've left them as is.
Quoted 8 lines
If there are any strong and specific reasons to correct them, let us change them. Do you have such reasons? For example, Linux kernel developer have shared the consensus and decision to ostracize such codes?
Thanks
Takashi Sakamoto
Hi Takashi,
Thank you for your detailed review and explanation.
You are absolutely right about the backward compatibility concern. I realize now that changing error codes in existing code paths could break userspace applications that might be checking for specific error values.
My patch was motivated by the checkpatch.pl warning and the general kernel policy that ENOSYS should only mean "invalid syscall number." BUt, I didn't fully consider the userspace ABI implications of changing error codes in code that's already been released.
I do not have a strong technical reason beyond code correctness to justify breaking backward compatibility in this case. Since these interfaces are already in use and userspace software may depend on the current behavior, the risk of breaking existing applications outweighs the benefit of this cleanup.
I withdraw this patch. Thank you for taking the time to explain this, its an important lesson about the difference between fixing issues in new code versus maintaining stability in existing interfaces.
Thanks,
Nirbhay