Re: [PATCH 06/11] moduleparam: Add seq_buf-based .get callback alongside .get_str
Petr Pavlu <[email protected]> Mon, 25 May 2026 18:19:19 +0200
| Newsgroups | gmane.comp.freedesktop.xorg.drivers.intel,gmane.comp.emulators.kvm.devel,gmane.linux.kernel.modules,gmane.linux.kernel.mm,gmane.comp.security.apparmor,gmane.linux.kernel.lsm,gmane.linux.uml.devel,gmane.linux.acpi.devel,gmane.linux.kernel.openipmi,gmane.comp.emulators.qemu,gmane.comp.video.dri.devel,gmane.linux.drivers.rdma,gmane.linux.drivers.video-input-infrastructure,gmane.linux.kernel.pci,gmane.linux.scsi,gmane.linux.power-management.general,gmane.linux.ports.ppc64.devel,gmane.linux.serial,gmane.linux.usb.general,gmane.linux.kernel,gmane.linux.kernel.cross-arch,gmane.linux.network,gmane.linux.file-systems |
|---|---|
| Message-ID | <[email protected]> |
On 5/21/26 3:33 PM, Kees Cook wrote: > Add a new struct kernel_param_ops::get callback whose signature > takes a struct seq_buf instead of a raw char buffer: > > int (*get)(struct seq_buf *sb, const struct kernel_param *kp); > > The previously-legacy .get field is now .get_str (char *buffer); > .get is the new seq_buf-aware form. param_attr_show() prefers .get > when set, otherwise falls back to .get_str. WARN_ON_ONCE() if both > are set. Return contract for .get: > > < 0 : errno propagated to userspace; seq_buf contents discarded > = 0 : success; length derived from seq_buf_used() > > 0 : forbidden; the dispatcher WARN_ON_ONCE()s and treats as 0 > > The default policy on seq_buf_has_overflowed() is silent truncation, > matching scnprintf()/sysfs_emit() behaviour. Callbacks that want a > specific overflow errno can check seq_buf_has_overflowed() and > return their preferred error. > > No callbacks use .get yet; the legacy path is still the only one in use > after this commit. A subsequent commit teaches DEFINE_KERNEL_PARAM_OPS > to route initializers by type. > > Signed-off-by: Kees Cook <[email protected]> Reviewed-by: Petr Pavlu <[email protected]> -- Petr