Re: [PATCH 1/1] arm64/sve: Fix wrong free for task->thread.sve_state
From: Masayoshi Mizuma
Date: Fri Sep 27 2019 - 10:38:15 EST
Hi Julien and Dave,
Thank you for your comments!
Dave's suggestion looks good for me, many thanks!
I'll post it as v2.
- Masa
On Fri, Sep 27, 2019 at 01:52:30PM +0100, Dave Martin wrote:
> On Thu, Sep 26, 2019 at 03:08:46PM -0400, Masayoshi Mizuma wrote:
> > From: Masayoshi Mizuma <m.mizuma@xxxxxxxxxxxxxx>
> >
> > The system which has SVE feature crashed because of
> > the memory pointed by task->thread.sve_state was destroyed
> > by someone.
> >
> > That is because sve_state is freed while the forking the
> > child process. The child process has the pointer of sve_state
> > which is same as the parent's because the child's task_struct
> > is copied from the parent's one. If the copy_process()
> > fails as an error on somewhere, for example, copy_creds(),
> > then the sve_state is freed even if the parent is alive.
> > The flow is as follows.
> >
> > copy_process
> > p = dup_task_struct
> > => arch_dup_task_struct
> > *dst = *src; // copy the entire region.
> > :
> > retval = copy_creds
> > if (retval < 0)
> > goto bad_fork_free;
> > :
> > bad_fork_free:
> > ...
> > delayed_free_task(p);
> > => free_task
> > => arch_release_task_struct
> > => fpsimd_release_task
> > => __sve_free
> > => kfree(task->thread.sve_state);
> > // free the parent's sve_state
> >
> > Add a flag in task->thread which shows the fork is in progress.
> > If the fork is in progress, that means the child has the pointer
> > to the parent's sve_state, doesn't free the sve_state.
> >
> > Signed-off-by: Masayoshi Mizuma <m.mizuma@xxxxxxxxxxxxxx>
> > Reported-by: Hidetoshi Seto <seto.hidetoshi@xxxxxxxxxxxxxx>
> > ---
> > arch/arm64/include/asm/processor.h | 1 +
> > arch/arm64/kernel/fpsimd.c | 6 ++++--
> > arch/arm64/kernel/process.c | 2 ++
> > 3 files changed, 7 insertions(+), 2 deletions(-)
> >
> > diff --git a/arch/arm64/include/asm/processor.h b/arch/arm64/include/asm/processor.h
> > index 5623685c7d13..3ca3e350145a 100644
> > --- a/arch/arm64/include/asm/processor.h
> > +++ b/arch/arm64/include/asm/processor.h
> > @@ -143,6 +143,7 @@ struct thread_struct {
> > unsigned long fault_address; /* fault info */
> > unsigned long fault_code; /* ESR_EL1 value */
> > struct debug_info debug; /* debugging */
> > + unsigned int fork_in_progress;
> > #ifdef CONFIG_ARM64_PTR_AUTH
> > struct ptrauth_keys keys_user;
> > #endif
> > diff --git a/arch/arm64/kernel/fpsimd.c b/arch/arm64/kernel/fpsimd.c
> > index 37d3912cfe06..8641db4cb062 100644
> > --- a/arch/arm64/kernel/fpsimd.c
> > +++ b/arch/arm64/kernel/fpsimd.c
> > @@ -202,8 +202,10 @@ static bool have_cpu_fpsimd_context(void)
> > */
> > static void __sve_free(struct task_struct *task)
> > {
> > - kfree(task->thread.sve_state);
> > - task->thread.sve_state = NULL;
> > + if (!task->thread.fork_in_progress) {
> > + kfree(task->thread.sve_state);
> > + task->thread.sve_state = NULL;
> > + }
> > }
> >
> > static void sve_free(struct task_struct *task)
> > diff --git a/arch/arm64/kernel/process.c b/arch/arm64/kernel/process.c
> > index a47462def04b..8ac0ee4e5f76 100644
> > --- a/arch/arm64/kernel/process.c
> > +++ b/arch/arm64/kernel/process.c
> > @@ -347,6 +347,7 @@ int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
> > if (current->mm)
> > fpsimd_preserve_current_state();
> > *dst = *src;
> > + dst->thread.fork_in_progress = 1;
> >
> > return 0;
> > }
> > @@ -365,6 +366,7 @@ int copy_thread(unsigned long clone_flags, unsigned long stack_start,
> > * and disable discard SVE state for p:
> > */
> > clear_tsk_thread_flag(p, TIF_SVE);
> > + p->thread.fork_in_progress = 0;
> > p->thread.sve_state = NULL;
>
> There's definitely a problem here, but a simpler fix is probably to
> NULL sve_state and clear TIF_SVE for dst at the same time.
>
> Once upon a time, I had to cope with the thread_flags not being copied
> as part of task_struct here, which is one reason why the code is the
> (broken) way it is, but this is ancient history...
>
> Commit c02433dd6de3 ("arm64: split thread_info from task stack") was
> merged in v4.10 and predates SVE support anyway.
>
>
> So can just do the following (untested) and delete the confusing
> comments?
>
> Cheers
> ---Dave
>
> --8<--
>
>
> diff --git a/arch/arm64/kernel/process.c b/arch/arm64/kernel/process.c
> index f674f28..6937f59 100644
> --- a/arch/arm64/kernel/process.c
> +++ b/arch/arm64/kernel/process.c
> @@ -323,22 +323,16 @@ void arch_release_task_struct(struct task_struct *tsk)
> fpsimd_release_task(tsk);
> }
>
> -/*
> - * src and dst may temporarily have aliased sve_state after task_struct
> - * is copied. We cannot fix this properly here, because src may have
> - * live SVE state and dst's thread_info may not exist yet, so tweaking
> - * either src's or dst's TIF_SVE is not safe.
> - *
> - * The unaliasing is done in copy_thread() instead. This works because
> - * dst is not schedulable or traceable until both of these functions
> - * have been called.
> - */
> int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
> {
> if (current->mm)
> fpsimd_preserve_current_state();
> *dst = *src;
>
> + BUILD_BUG_ON(!IS_ENABLED(CONFIG_THREAD_INFO_IN_TASK));
> + dst->thread.sve_state = NULL;
> + clear_tsk_thread_flag(dst, TIF_SVE);
> +
> return 0;
> }
>
> @@ -352,13 +346,6 @@ int copy_thread(unsigned long clone_flags, unsigned long stack_start,
> memset(&p->thread.cpu_context, 0, sizeof(struct cpu_context));
>
> /*
> - * Unalias p->thread.sve_state (if any) from the parent task
> - * and disable discard SVE state for p:
> - */
> - clear_tsk_thread_flag(p, TIF_SVE);
> - p->thread.sve_state = NULL;
> -
> - /*
> * In case p was allocated the same task_struct pointer as some
> * other recently-exited task, make sure p is disassociated from
> * any cpu that may have run that now-exited task recently.