Re: [PATCH v2] arm: pgtable: fix NULL pointer dereference issue

[David Hildenbrand <david@redhat.com>]

On 14.02.25 04:03, Qi Zheng wrote:
> When update_mmu_cache_range() is called by update_mmu_cache(), the vmf
> parameter is NULL, which will cause a NULL pointer dereference issue in
> adjust_pte():
> 
> Unable to handle kernel NULL pointer dereference at virtual address 00000030 when read
> Hardware name: Atmel AT91SAM9
> PC is at update_mmu_cache_range+0x1e0/0x278
> LR is at pte_offset_map_rw_nolock+0x18/0x2c
> Call trace:
>   update_mmu_cache_range from remove_migration_pte+0x29c/0x2ec
>   remove_migration_pte from rmap_walk_file+0xcc/0x130
>   rmap_walk_file from remove_migration_ptes+0x90/0xa4
>   remove_migration_ptes from migrate_pages_batch+0x6d4/0x858
>   migrate_pages_batch from migrate_pages+0x188/0x488
>   migrate_pages from compact_zone+0x56c/0x954
>   compact_zone from compact_node+0x90/0xf0
>   compact_node from kcompactd+0x1d4/0x204
>   kcompactd from kthread+0x120/0x12c
>   kthread from ret_from_fork+0x14/0x38
> Exception stack(0xc0d8bfb0 to 0xc0d8bff8)
> 
> To fix it, do not rely on whether 'ptl' is equal to decide whether to hold
> the pte lock, but decide it by whether CONFIG_SPLIT_PTE_PTLOCKS is
> enabled. In addition, if two vmas map to the same PTE page, there is no
> need to hold the pte lock again, otherwise a deadlock will occur. Just add
> the need_lock parameter to let adjust_pte() know this information.
> 
> Reported-by: Ezra Buehler <ezra.buehler@husqvarnagroup.com>
> Closes: https://lore.kernel.org/lkml/CAM1KZSmZ2T_riHvay+7cKEFxoPgeVpHkVFTzVVEQ1BO0cLkHEQ@mail.gmail.com/
> Fixes: fc9c45b71f43 ("arm: adjust_pte() use pte_offset_map_rw_nolock()")
> Cc: stable@vger.kernel.org
> Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
> ---
> Changes in v2:
>   - change Ezra's email address (Ezra Buehler)
>   - some cleanups (David Hildenbrand)
> 
>   arch/arm/mm/fault-armv.c | 38 ++++++++++++++++++++++++++------------
>   1 file changed, 26 insertions(+), 12 deletions(-)
> 
> diff --git a/arch/arm/mm/fault-armv.c b/arch/arm/mm/fault-armv.c
> index 2bec87c3327d2..ea4c4e15f0d31 100644
> --- a/arch/arm/mm/fault-armv.c
> +++ b/arch/arm/mm/fault-armv.c
> @@ -62,7 +62,7 @@ static int do_adjust_pte(struct vm_area_struct *vma, unsigned long address,
>   }
>   
>   static int adjust_pte(struct vm_area_struct *vma, unsigned long address,
> -		      unsigned long pfn, struct vm_fault *vmf)
> +		      unsigned long pfn, bool need_lock)
>   {
>   	spinlock_t *ptl;
>   	pgd_t *pgd;
> @@ -99,12 +99,11 @@ static int adjust_pte(struct vm_area_struct *vma, unsigned long address,
>   	if (!pte)
>   		return 0;
>   
> -	/*
> -	 * If we are using split PTE locks, then we need to take the page
> -	 * lock here.  Otherwise we are using shared mm->page_table_lock
> -	 * which is already locked, thus cannot take it.
> -	 */
> -	if (ptl != vmf->ptl) {
> +	if (need_lock) {
> +		/*
> +		 * Use nested version here to indicate that we are already
> +		 * holding one similar spinlock.
> +		 */
>   		spin_lock_nested(ptl, SINGLE_DEPTH_NESTING);
>   		if (unlikely(!pmd_same(pmdval, pmdp_get_lockless(pmd)))) {
>   			pte_unmap_unlock(pte, ptl);
> @@ -114,7 +113,7 @@ static int adjust_pte(struct vm_area_struct *vma, unsigned long address,
>   
>   	ret = do_adjust_pte(vma, address, pfn, pte);
>   
> -	if (ptl != vmf->ptl)
> +	if (need_lock)
>   		spin_unlock(ptl);
>   	pte_unmap(pte);
>   
> @@ -123,16 +122,18 @@ static int adjust_pte(struct vm_area_struct *vma, unsigned long address,
>   
>   static void
>   make_coherent(struct address_space *mapping, struct vm_area_struct *vma,
> -	      unsigned long addr, pte_t *ptep, unsigned long pfn,
> -	      struct vm_fault *vmf)
> +	      unsigned long addr, pte_t *ptep, unsigned long pfn)
>   {
>   	struct mm_struct *mm = vma->vm_mm;
>   	struct vm_area_struct *mpnt;
>   	unsigned long offset;
> +	unsigned long pmd_start_addr, pmd_end_addr;

Nit: reverse christmas tree would make us put this line at the very top.

Maybe do the initialization directly:

const unsigned long pmd_start_addr = ALIGN_DOWN(addr, PMD_SIZE);
const unsigned long pmd_end_addr = pmd_start_addr + PMD_SIZE;

>   	pgoff_t pgoff;
>   	int aliases = 0;
>   
>   	pgoff = vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT);
> +	pmd_start_addr = ALIGN_DOWN(addr, PMD_SIZE);
> +	pmd_end_addr = pmd_start_addr + PMD_SIZE;
>   
>   	/*
>   	 * If we have any shared mappings that are in the same mm
> @@ -141,6 +142,14 @@ make_coherent(struct address_space *mapping, struct vm_area_struct *vma,
>   	 */
>   	flush_dcache_mmap_lock(mapping);
>   	vma_interval_tree_foreach(mpnt, &mapping->i_mmap, pgoff, pgoff) {
> +		/*
> +		 * If we are using split PTE locks, then we need to take the pte
> +		 * lock. Otherwise we are using shared mm->page_table_lock which
> +		 * is already locked, thus cannot take it.
> +		 */
> +		bool need_lock = IS_ENABLED(CONFIG_SPLIT_PTE_PTLOCKS);
> +		unsigned long mpnt_addr;
> +
>   		/*
>   		 * If this VMA is not in our MM, we can ignore it.
>   		 * Note that we intentionally mask out the VMA
> @@ -151,7 +160,12 @@ make_coherent(struct address_space *mapping, struct vm_area_struct *vma,
>   		if (!(mpnt->vm_flags & VM_MAYSHARE))
>   			continue;
>   		offset = (pgoff - mpnt->vm_pgoff) << PAGE_SHIFT;
> -		aliases += adjust_pte(mpnt, mpnt->vm_start + offset, pfn, vmf);
> +		mpnt_addr = mpnt->vm_start + offset;
> +
> +		/* Avoid deadlocks by not grabbing the same PTE lock again. */
> +		if (mpnt_addr >= pmd_start_addr && mpnt_addr < pmd_end_addr)
> +			need_lock = false;
> +		aliases += adjust_pte(mpnt, mpnt_addr, pfn, need_lock);
>   	}
>   	flush_dcache_mmap_unlock(mapping);
>   	if (aliases)
> @@ -194,7 +208,7 @@ void update_mmu_cache_range(struct vm_fault *vmf, struct vm_area_struct *vma,
>   		__flush_dcache_folio(mapping, folio);
>   	if (mapping) {
>   		if (cache_is_vivt())
> -			make_coherent(mapping, vma, addr, ptep, pfn, vmf);
> +			make_coherent(mapping, vma, addr, ptep, pfn);
>   		else if (vma->vm_flags & VM_EXEC)
>   			__flush_icache_all();
>   	}


Apart from that LGTM. Hoping it will work :)

Acked-by: David Hildenbrand <david@redhat.com>

-- 
Cheers,

David / dhildenb