Re: [PATCH] software node: replace -EEXIST with -EBUSY

[Daniel Gomez <da.gomez@kernel.org>]

On 06/01/2026 15.24, Lucas De Marchi wrote:
> On Mon, Dec 22, 2025 at 12:56:38PM +0100, Greg Kroah-Hartman wrote:
>> On Mon, Dec 22, 2025 at 09:48:54AM +0100, Daniel Gomez wrote:
>>> On 22/12/2025 09.19, Greg Kroah-Hartman wrote:
>>> > On Sat, Dec 20, 2025 at 04:55:00AM +0100, Daniel Gomez wrote:
>>> >> From: Daniel Gomez <da.gomez@samsung.com>
>>> >>
>>> >> The -EEXIST error code is reserved by the module loading infrastructure
>>> >> to indicate that a module is already loaded. When a module's init
>>> >> function returns -EEXIST, userspace tools like kmod interpret this as
>>> >> "module already loaded" and treat the operation as successful, returning
>>> >> 0 to the user even though the module initialization actually failed.
>>> >>
>>> >> This follows the precedent set by commit 54416fd76770 ("netfilter:
>>> >> conntrack: helper: Replace -EEXIST by -EBUSY") which fixed the same
>>> >> issue in nf_conntrack_helper_register().
>>> >>
>>> >> Affected modules:
>>> >>   * meraki_mx100 pcengines_apuv2
>>> >>
>>> >> Signed-off-by: Daniel Gomez <da.gomez@samsung.com>
>>> >> ---
>>> >> The error code -EEXIST is reserved by the kernel module loader to
>>> >> indicate that a module with the same name is already loaded. When a
>>> >> module's init function returns -EEXIST, kmod interprets this as "module
>>> >> already loaded" and reports success instead of failure [1].
>>> >>
>>> >> The kernel module loader will include a safety net that provides -EEXIST
>>> >> to -EBUSY with a warning [2], and a documentation patch has been sent to
>>> >> prevent future occurrences [3].
>>> >>
>>> >> These affected code paths were identified using a static analysis tool
>>> >> [4] that traces -EEXIST returns to module_init(). The tool was developed
>>> >> with AI assistance and all findings were manually validated.
>>> >>
>>> >> Link: https://lore.kernel.org/all/aKEVQhJpRdiZSliu@orbyte.nwl.cc/ [1]
>>> >> Link: https://lore.kernel.org/all/20251013-module-warn-ret-v1-0-ab65b41af01f@intel.com/ [2]
>>> >> Link: https://lore.kernel.org/all/20251218-dev-module-init-eexists-modules-docs-v1-0-361569aa782a@samsung.com/ [3]
>>> >> Link: https://gitlab.com/-/snippets/4913469 [4]
>>> >> ---
>>> >>  drivers/base/swnode.c | 2 +-
>>> >>  1 file changed, 1 insertion(+), 1 deletion(-)
>>> >>
>>> >> diff --git a/drivers/base/swnode.c b/drivers/base/swnode.c
>>> >> index 16a8301c25d6..083593d99a18 100644
>>> >> --- a/drivers/base/swnode.c
>>> >> +++ b/drivers/base/swnode.c
>>> >> @@ -919,7 +919,7 @@ int software_node_register(const struct software_node *node)
>>> >>      struct swnode *parent = software_node_to_swnode(node->parent);
>>> >>
>>> >>      if (software_node_to_swnode(node))
>>> >> -        return -EEXIST;
>>> >> +        return -EBUSY;
>>> >
>>> > While I understand the want for the module loader to be returning
>>> > -EBUSY, that doesn't really make sense down here in this layer of the
>>> > kernel.
>>> >
>>> > So why doesn't the module loader turn -EEXIST return values into -EBUSY
>>> > if it wishes to pass that value on to userspace?  Otherwise you are
>>>
>>> Indeed, we are planning to do that as well with "[PATCH 0/2] module: Tweak
>>> return and warning":
>>>
>>> https://lore.kernel.org/all/20251013-module-warn-ret-v1-0-ab65b41af01f@intel.com/#t
>>>
>>> However, we don't consider that as the right fix.
>>>
>>> > going to be constantly playing "whack-a-mole" here and have really
>>> > set things up so that NO api can ever return EEXIST as an error value in
>>> > the future.
>>>
>>> 100%.
>>>
>>> For that reason, on top of the series from Lucas, we are documenting this to
>>> make it clear:
>>>
>>> https://lore.kernel.org/linux-modules/20251218-dev-module-init-eexists-modules-docs-v1-0-361569aa782a@samsung.com/T/#m2ed6fbffb3f78b9bff53840f6492a198c389cb50
>>
>> Wait, no, that's not what I mean at all :)
>>
>>> And sending patches where we see modules need fixing. I have already sent 6 out
>>> of 20-ish series (that include a total of 40+ fixes):
>>>
>>> https://lore.kernel.org/all/20251220-dev-module-init-eexists-linux-scsi-v1-0-5379db749d54@samsung.com
>>> https://lore.kernel.org/all/20251219-dev-module-init-eexists-netfilter-v1-1-efd3f62412dc@samsung.com
>>> https://lore.kernel.org/all/20251220-dev-module-init-eexists-bpf-v1-1-7f186663dbe7@samsung.com
>>> https://lore.kernel.org/all/20251220-dev-module-init-eexists-keyring-v1-1-a2f23248c300@samsung.com
>>> https://lore.kernel.org/all/20251220-dev-module-init-eexists-dm-devel-v1-1-90ed00444ea0@samsung.com
>>
>> Please no, let us keep using -EEXIST in the kernel source, and if your
>> usage is going to map this to userspace somehow, do the translation
>> there, in the module code, as your original patch above said.
>>
>> Otherwise, again, this is never going to work, let the subsystems use
>> this error code how ever they feel they need to.
> 
> Ok. When I added the warning I was more following what the other error
> handling was doing for positive values. Happy to change that to simply
> map the error code before returning from do_init_module().
> 
> Daniel, do you want me to resend that with the warning removed?

Yes please, I think we should do that and explain the agreement in this thread
in the commit message so others can understand the why.