On Thu, 25 Nov 2010, Corentin Chary wrote:
> On Thu, Nov 25, 2010 at 7:01 AM, Julia Lawall <julia@xxxxxxx> wrote:
> > On Thu, 25 Nov 2010, Corentin Chary wrote:
> >
> >> Hi,
> >>
> >> I was checking debugfs code in platform/x86, because I want to add
> >> some files to eeepc-wmi. And I found something disturbing.
> >>
> >> The documentation says:
> >>
> >> > This call, if successful, will make a directory called name underneath the
> >> > indicated parent directory. If parent is NULL, the directory will be
> >> > created in the debugfs root. On success, the return value is a struct
> >> > dentry pointer which can be used to create files in the directory (and to
> >> > clean it up at the end). A NULL return value indicates that something went
> >> > wrong. If ERR_PTR(-ENODEV) is returned, that is an indication that the
> >> > kernel has been built without debugfs support and none of the functions
> >> > described below will work.
> >>
> >> But then, here is the code in acer-wmi:
> >>
> >> > static void remove_debugfs(void)
> >> > {
> >> > debugfs_remove(interface->debug.devices);
> >> > debugfs_remove(interface->debug.root);
> >> > }
> >> >
> >> > static int create_debugfs(void)
> >> > {
> >> > interface->debug.root = debugfs_create_dir("acer-wmi", NULL);
> >> > if (!interface->debug.root) {
> >> > printk(ACER_ERR "Failed to create debugfs directory");
> >> > return -ENOMEM;
> >> > }
> >>
> >> this code is *not* inside #ifndef CONFIG_DEBUG_FS, so debugfs_create_dir
> >> can return ERR_PTR(-ENODEV) right ?
> >>
> >> Then, remove_debug() will call debugfs_remove(ERR_PTR(-ENODEV)) right ?
> >>
> >> So, acpi-wmi seems to have an issue when debugfs is disabled, that's "ok".
> >>
> >> But then I took a look at intel_ips :
> >>
> >> > ips->debug_root = debugfs_create_dir("ips", NULL);
> >> > if (!ips->debug_root) {
> >> > dev_err(&ips->dev->dev,
> >> > "failed to create debugfs entries: %ld\n",
> >> > PTR_ERR(ips->debug_root));
> >> > return;
> >> > }
> >>
> >> Then PTR_ERR thing is strange, because ips->debug_root can only be NULL
> >> here...
> >> But here, it's ok to only check NULL, because it's inside #ifndef
> >> CONFIG_DEBUG_FS.
> >>
> >> So, two drivers checked, to weird error handling code. I did a quick grep and
> >> opened
> >> the first result: ec_sys.c.
> >>
> >> ec_sys.c depends on CONFIG_ACPI_EC_DEBUGFS but doesn't depend on
> >> CONFIG_DEBUG_FS.
> >>
> >> Here, again, the code only check for != NULL while it could be ERR_PTR(-
> >> ENODEV):
> >>
> >> > if (ec_device_count == 0) {
> >> > acpi_ec_debugfs_dir = debugfs_create_dir("ec", NULL);
> >> > if (!acpi_ec_debugfs_dir)
> >> > return -ENOMEM;
> >> > }
> >> >
> >> > sprintf(name, "ec%u", ec_device_count);
> >> > dev_dir = debugfs_create_dir(name, acpi_ec_debugfs_dir);
> >>
> >> Here, acpi_ec_debugfs_dir (that can be an invalid pointer) is used as
> >> a parent dentry, and will be dereferenced without checks.
> >>
> >> I am missing something obvious, or are most of debugfs implementation
> >> broken when debugfs is disabled ?
>
> Answer to myself, when debugfs is disabled, it's ok to give broken
> dentry pointers to debugfs functions since they won't do anything.
>
> >> Julia, if I am right, coccinelle could help us right ? Can the tool check
> >> if the code is between #ifdef CONFIG_DEBUGS_FS ? That would help a lot.
> >
> > Unfortunately, at the moment, it can't; there is no matching on #ifdefs.
> > Perhaps it could be added.
>
> Or better, something to check if a macro is defined in a particular contact ?
Actually, Daniel Lohmann's group has been working on analyzing #ifdef's.
Perhaps they have a solution to this problem? I have added them to the CC
list.
julia
> > I wonder though if sometimes returning NULL and sometimes returning
> > ERR_PTR is something that should be encouraged? Would one rather convert
> > the NULL case to a specific ERR_PTR case?
>
> But yeah, I found debugfs API disturbing, but it seems to be done like that to
> ease the "debugfs is disabled case".
>
> Thanks,
> --
> Corentin Chary
> http://xf.iksaif.net
>
