When module relocation is done, process_accumulated_relocations()
frees all dynamic allocated memory. rel_head_iter->rel_entry is
missed to free that kmemleak might report:

unreferenced object 0xffffffd880c5fc40 (size 16):
  comm "insmod", pid 1101, jiffies 4295045138
  hex dump (first 16 bytes):
    e0 c0 f5 87 d8 ff ff ff 60 c5 f5 87 d8 ff ff ff  ........`.......
  backtrace (crc d2ecb20c):
    [<00000000b01655f6>] kmalloc_trace_noprof+0x268/0x2f6
    [<000000006dc0067a>] add_relocation_to_accumulate.constprop.0+0xf2/0x1aa
    [<00000000e1b29a36>] apply_relocate_add+0x13c/0x36e
    [<000000007543f1fb>] load_module+0x5c6/0x83e
    [<00000000abce12e8>] init_module_from_file+0x74/0xaa
    [<0000000049413e3d>] idempotent_init_module+0x116/0x22e
    [<00000000f9b98b85>] __riscv_sys_finit_module+0x62/0xae

Signed-off-by: Kai Zhang <zhangkai@iscas.ac.cn>
---
 arch/riscv/kernel/module.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/arch/riscv/kernel/module.c b/arch/riscv/kernel/module.c
index 1cd461f3d87..f8c3c4b47dc 100644
--- a/arch/riscv/kernel/module.c
+++ b/arch/riscv/kernel/module.c
@@ -643,6 +643,7 @@ process_accumulated_relocations(struct module *me,
                        }
                        reloc_handlers[curr_type].accumulate_handler(
                                me, location, buffer);
+                       kfree(rel_head_iter-&gt;rel_entry);
                        kfree(rel_head_iter);
                }
                kfree(bucket_iter);
-- 
2.39.5

</zhangkai@iscas.ac.cn>

On 16/11/2024 11:42, zhangkai@iscas.ac.cn wrote:
> When module relocation is done, process_accumulated_relocations()
> frees all dynamic allocated memory. rel_head_iter->rel_entry is
> missed to free that kmemleak might report:
> 
> unreferenced object 0xffffffd880c5fc40 (size 16):
>   comm "insmod", pid 1101, jiffies 4295045138
>   hex dump (first 16 bytes):
>     e0 c0 f5 87 d8 ff ff ff 60 c5 f5 87 d8 ff ff ff  ........`.......
>   backtrace (crc d2ecb20c):
>     [<00000000b01655f6>] kmalloc_trace_noprof+0x268/0x2f6
>     [<000000006dc0067a>] add_relocation_to_accumulate.constprop.0+0xf2/0x1aa
>     [<00000000e1b29a36>] apply_relocate_add+0x13c/0x36e
>     [<000000007543f1fb>] load_module+0x5c6/0x83e
>     [<00000000abce12e8>] init_module_from_file+0x74/0xaa
>     [<0000000049413e3d>] idempotent_init_module+0x116/0x22e
>     [<00000000f9b98b85>] __riscv_sys_finit_module+0x62/0xae
> 
> Signed-off-by: Kai Zhang <zhangkai@iscas.ac.cn>
> ---
>  arch/riscv/kernel/module.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/arch/riscv/kernel/module.c b/arch/riscv/kernel/module.c
> index 1cd461f3d87..f8c3c4b47dc 100644
> --- a/arch/riscv/kernel/module.c
> +++ b/arch/riscv/kernel/module.c
> @@ -643,6 +643,7 @@ process_accumulated_relocations(struct module *me,
>                         }
>                         reloc_handlers[curr_type].accumulate_handler(
>                                 me, location, buffer);
> +                       kfree(rel_head_iter-&gt;rel_entry);

Hey Kai,

Your patch output is messed up probably due to some encoding issue.

But looknig at it and the module code, it seems like rel_entry does not
need to be a pointer. It can be a plain "struct list_head rel_entry" and
then simply pass the list_head pointer rather than allocating/freeing
it. That remove an allocation as well as some memory leak.

BTW, for fixes, you should add a Fixes: tag with the original commit
sha1 that introduced the bug.

Thanks,

Clément

>                         kfree(rel_head_iter);
>                 }
>                 kfree(bucket_iter);

On Tue, 2024-11-19 at 17:57 +0100, Clément Léger wrote:
> 
> 
> On 16/11/2024 11:42, zhangkai@iscas.ac.cn wrote:
> > When module relocation is done, process_accumulated_relocations()
> > frees all dynamic allocated memory. rel_head_iter->rel_entry is
> > missed to free that kmemleak might report:
> > 
> > unreferenced object 0xffffffd880c5fc40 (size 16):
> >   comm "insmod", pid 1101, jiffies 4295045138
> >   hex dump (first 16 bytes):
> >     e0 c0 f5 87 d8 ff ff ff 60 c5 f5 87 d8 ff ff ff 
> > ........`.......
> >   backtrace (crc d2ecb20c):
> >     [<00000000b01655f6>] kmalloc_trace_noprof+0x268/0x2f6
> >     [<000000006dc0067a>]
> > add_relocation_to_accumulate.constprop.0+0xf2/0x1aa
> >     [<00000000e1b29a36>] apply_relocate_add+0x13c/0x36e
> >     [<000000007543f1fb>] load_module+0x5c6/0x83e
> >     [<00000000abce12e8>] init_module_from_file+0x74/0xaa
> >     [<0000000049413e3d>] idempotent_init_module+0x116/0x22e
> >     [<00000000f9b98b85>] __riscv_sys_finit_module+0x62/0xae
> > 
> > Signed-off-by: Kai Zhang <zhangkai@iscas.ac.cn>
> > ---
> >  arch/riscv/kernel/module.c | 1 +
> >  1 file changed, 1 insertion(+)
> > 
> > diff --git a/arch/riscv/kernel/module.c
> > b/arch/riscv/kernel/module.c
> > index 1cd461f3d87..f8c3c4b47dc 100644
> > --- a/arch/riscv/kernel/module.c
> > +++ b/arch/riscv/kernel/module.c
> > @@ -643,6 +643,7 @@ process_accumulated_relocations(struct module
> > *me,
> >                         }
> >                        
> > reloc_handlers[curr_type].accumulate_handler(
> >                                 me, location, buffer);
> > +                       kfree(rel_head_iter-&gt;rel_entry);
> 
> Hey Kai,
> 
> Your patch output is messed up probably due to some encoding issue.

Sorry, I've already resent one.

> But looknig at it and the module code, it seems like rel_entry does
> not
> need to be a pointer. It can be a plain "struct list_head rel_entry"
> and

I didn't find the plain(none pointer) usage. Where please?

> then simply pass the list_head pointer rather than allocating/freeing
> it. That remove an allocation as well as some memory leak.
> 
> BTW, for fixes, you should add a Fixes: tag with the original commit
> sha1 that introduced the bug.
Thank you. I'll try.

laokz

> Thanks,
> 
> Clément
> 
> >                         kfree(rel_head_iter);
> >                 }
> >                 kfree(bucket_iter);