Andra Paraschiv <andraprs@xxxxxxxxxx> writes:
> After commit 5b78ed24e8ec (mm/pagemap: add mmap_assert_locked()
> annotations to find_vma*()), the call to get_user_pages() will trigger
> the mmap assert.
>
> static inline void mmap_assert_locked(struct mm_struct *mm)
> {
> lockdep_assert_held(&mm->mmap_lock);
> VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_lock), mm);
> }
>
> [ 62.521410] kernel BUG at include/linux/mmap_lock.h:156!
> ...........................................................
> [ 62.538938] RIP: 0010:find_vma+0x32/0x80
> ...........................................................
> [ 62.605889] Call Trace:
> [ 62.608502] <TASK>
> [ 62.610956] ? lock_timer_base+0x61/0x80
> [ 62.614106] find_extend_vma+0x19/0x80
> [ 62.617195] __get_user_pages+0x9b/0x6a0
> [ 62.620356] __gup_longterm_locked+0x42d/0x450
> [ 62.623721] ? finish_wait+0x41/0x80
> [ 62.626748] ? __kmalloc+0x178/0x2f0
> [ 62.629768] ne_set_user_memory_region_ioctl.isra.0+0x225/0x6a0 [nitro_enclaves]
> [ 62.635776] ne_enclave_ioctl+0x1cf/0x6d7 [nitro_enclaves]
> [ 62.639541] __x64_sys_ioctl+0x82/0xb0
> [ 62.642620] do_syscall_64+0x3b/0x90
> [ 62.645642] entry_SYSCALL_64_after_hwframe+0x44/0xae
>
> Add mmap_read_lock() for the get_user_pages() call when setting the
> enclave memory regions.
>
> Signed-off-by: Andra Paraschiv <andraprs@xxxxxxxxxx>
> Cc: stable@xxxxxxxxxxxxxxx
In case commit 5b78ed24e8ec broke Nitro Enclaves driver, we need to
explicitly state this:
Fixes: 5b78ed24e8ec ("mm/pagemap: add mmap_assert_locked() annotations to find_vma*()")
> ---
> drivers/virt/nitro_enclaves/ne_misc_dev.c | 5 +++++
> 1 file changed, 5 insertions(+)
>
> diff --git a/drivers/virt/nitro_enclaves/ne_misc_dev.c b/drivers/virt/nitro_enclaves/ne_misc_dev.c
> index 8939612ee0e0..6c51ff024036 100644
> --- a/drivers/virt/nitro_enclaves/ne_misc_dev.c
> +++ b/drivers/virt/nitro_enclaves/ne_misc_dev.c
> @@ -886,8 +886,13 @@ static int ne_set_user_memory_region_ioctl(struct ne_enclave *ne_enclave,
> goto put_pages;
> }
>
> + mmap_read_lock(current->mm);
> +
> gup_rc = get_user_pages(mem_region.userspace_addr + memory_size, 1, FOLL_GET,
> ne_mem_region->pages + i, NULL);
> +
> + mmap_read_unlock(current->mm);
> +
This looks very much like get_user_pages_unlocked(), I think we can use
it instead of open-coding it.
> if (gup_rc < 0) {
> rc = gup_rc;
--
Vitaly
