On Tue, 14 Dec 2021, David Vernet wrote: > Miroslav Benes <mbenes@xxxxxxx> wrote on Tue [2021-Dec-14 10:17:08 +0100]: > > It would help to share warning outputs (or whatever) from DEBUG_KOBJECTS. > > This is the output when running kpatch load livepatch-sample.ko if an extra > 'struct klp_obj' entry is added that has a name but no funcs: > > Without patch: > > [ 67.285762] livepatch_sample: tainting kernel with TAINT_LIVEPATCH > [ 67.286107] kobject: 'livepatch_sample' (00000000cf89f7b6): kobject_add_internal: parent: 'module', set: 'module' > [ 67.286113] kobject: 'holders' (00000000858b03bf): kobject_add_internal: parent: 'livepatch_sample', set: '<NULL>' > [ 67.286126] kobject: 'notes' (00000000f2a3a4ce): kobject_add_internal: parent: 'livepatch_sample', set: '<NULL>' > [ 67.297856] kobject: 'holders' (00000000858b03bf): kobject_cleanup, parent 00000000cf89f7b6 > [ 67.297859] kobject: 'holders' (00000000858b03bf): auto cleanup kobject_del > [ 67.297861] kobject: 'holders' (00000000858b03bf): calling ktype release > [ 67.297862] kobject: (00000000858b03bf): dynamic_kobj_release > [ 67.297863] kobject: 'holders': free name > [ 67.297865] kobject: 'notes' (00000000f2a3a4ce): kobject_cleanup, parent 00000000cf89f7b6 > [ 67.297866] kobject: 'notes' (00000000f2a3a4ce): auto cleanup kobject_del > [ 67.297867] kobject: 'notes' (00000000f2a3a4ce): calling ktype release > [ 67.297868] kobject: (00000000f2a3a4ce): dynamic_kobj_release > [ 67.297869] kobject: 'notes': free name > [ 67.297874] kobject: 'livepatch_sample' (00000000cf89f7b6): kobject_cleanup, parent 000000002555fa2d > [ 67.297876] kobject: 'livepatch_sample' (00000000cf89f7b6): auto cleanup kobject_del > [ 67.297877] kobject: 'livepatch_sample' (00000000cf89f7b6): calling ktype release > [ 67.297878] kobject: 'livepatch_sample': free name > [ 99.445938] kobject: '0:40' (000000002a98d11d): kobject_add_internal: parent: 'bdi', set: 'devices' > [ 99.445954] kobject: '0:40' (000000002a98d11d): kobject_uevent_env > [ 99.445957] kobject: '0:40' (000000002a98d11d): fill_kobj_path: path = '/devices/virtual/bdi/0:40' > > With patch: > > [ 162.275251] livepatch_sample: tainting kernel with TAINT_LIVEPATCH > [ 162.275985] kobject: 'livepatch_sample' (00000000e688ee30): kobject_add_internal: parent: 'module', set: 'module' > [ 162.275993] kobject: 'holders' (000000004eee7860): kobject_add_internal: parent: 'livepatch_sample', set: '<NULL>' > [ 162.276010] kobject: 'notes' (00000000c4f390ab): kobject_add_internal: parent: 'livepatch_sample', set: '<NULL>' > [ 162.276028] kobject: '(null)' (000000003acccf72): kobject_cleanup, parent 0000000000000000 > [ 162.276031] kobject: '(null)' (000000003acccf72): calling ktype release > [ 162.276033] kobject: '(null)' (00000000aeae6326): kobject_cleanup, parent 0000000000000000 > [ 162.276035] kobject: '(null)' (00000000aeae6326): calling ktype release > [ 162.276037] kobject: '(null)' (0000000093b68297): kobject_cleanup, parent 0000000000000000 > [ 162.276039] kobject: '(null)' (0000000093b68297): calling ktype release > [ 162.294063] kobject: 'holders' (000000004eee7860): kobject_cleanup, parent 00000000e688ee30 > [ 162.294070] kobject: 'holders' (000000004eee7860): auto cleanup kobject_del > [ 162.294074] kobject: 'holders' (000000004eee7860): calling ktype release > [ 162.294078] kobject: (000000004eee7860): dynamic_kobj_release > [ 162.294081] kobject: 'holders': free name > [ 162.294086] kobject: 'notes' (00000000c4f390ab): kobject_cleanup, parent 00000000e688ee30 > [ 162.294090] kobject: 'notes' (00000000c4f390ab): auto cleanup kobject_del > [ 162.294094] kobject: 'notes' (00000000c4f390ab): calling ktype release > [ 162.294097] kobject: (00000000c4f390ab): dynamic_kobj_release > [ 162.294100] kobject: 'notes': free name > [ 162.294114] kobject: 'livepatch_sample' (00000000e688ee30): kobject_cleanup, parent 00000000f9317c72 > [ 162.294118] kobject: 'livepatch_sample' (00000000e688ee30): auto cleanup kobject_del > [ 162.294123] kobject: 'livepatch_sample' (00000000e688ee30): calling ktype release > [ 162.294126] kobject: 'livepatch_sample': free name > > The extra lines are of course the kobject: '(null)' entries, for which we > *don't* see auto cleanup kobject_del being called. So it seems that what's > there now is probably not actually leaking memory, and the question is > really whether or not the documentation in kobject.c is the source of truth > (i.e. whether the code needs to be "fixed" to honor the API contract). Indeed. [...] > > And if it is not a false positive, we should implement some rollback for > > processed klp_funcs and klp_objects if an error happens. It is not only > > klp_patch kobject affected. > > The patch (though it needs to be corrected in its current form, as Josh > pointed out) does already address this for the klp_funcs and klp_objects. > The 'err' label invokes klp_free_patch_start(), which eventually invokes > __klp_free_objects(), which itself invokes __klp_free_funcs(). I was just thinking out loud whether it would make sense in the _init_early stage to actually clean up (call kobject_put()) immediately, but you are right that the code in klp_enable_patch() and elsewhere is already prepared for that and it would not make sense to change the approach just because. Thanks Miroslav
