Re: [PATCH v5 07/11] mm/vmalloc: undo what __vmap_pages_range_noflush() mapped on failure

From: Suren Baghdasaryan

Date: Sat Nov 21 2026 - 22:58:43 EST


On Sun, Nov 15, 2026 at 11:14 PM Hao Ge <hao.ge@xxxxxxxxx> wrote:
>
> __vmap_pages_range_noflush() maps a struct page array and may fail
> half way. The ptes installed so far stay behind, so a retry of the
> same range, the __GFP_NOFAIL loop in __vmalloc_area_node() or the
> module tags populate, trips the WARN_ON(!pte_none()) in
> vmap_pages_pte_range() and fails with -EBUSY.
>
> vmap_pages_range_noflush_walk() now reports how many pages it
> installed through its running index into the pages array, and
> __vmap_pages_range_noflush() undoes [addr, addr + mapped_pages *
> PAGE_SIZE) on failure. vmap_pages_pte_range() fails with -EBUSY
> instead of installing over a pte that is already set, so the ptes
> below that point are all ours, while a pte above it can only have
> been put there by somebody else, for example when the caller's range
> overlaps an existing mapping. Going further would clear ptes that
> are not ours.
>
> This path can run with a non-blocking gfp mask, the undo passes
> gfpflags_allow_blocking() as can_resched.
>
> Suggested-by: Suren Baghdasaryan <surenb@xxxxxxxxxx>
> Suggested-by: Uladzislau Rezki <urezki@xxxxxxxxx>
> Signed-off-by: Hao Ge <hao.ge@xxxxxxxxx>
> ---
> mm/kmsan/hooks.c | 4 ++--
> mm/kmsan/shadow.c | 4 ++--
> mm/vmalloc.c | 36 +++++++++++++++++++++++++++++-------
> mm/vmalloc.h | 3 ++-
> 4 files changed, 35 insertions(+), 12 deletions(-)
>
> diff --git a/mm/kmsan/hooks.c b/mm/kmsan/hooks.c
> index 69008b085658..d830a4e649f1 100644
> --- a/mm/kmsan/hooks.c
> +++ b/mm/kmsan/hooks.c
> @@ -208,7 +208,7 @@ int kmsan_ioremap_page_range(unsigned long start, unsigned long end,
> mapped = __vmap_pages_range_noflush(
> vmalloc_shadow(start + off),
> vmalloc_shadow(start + off + PAGE_SIZE), prot, &shadow,
> - PAGE_SHIFT);
> + PAGE_SHIFT, gfp_mask);
> if (mapped) {
> err = mapped;
> goto ret;
> @@ -218,7 +218,7 @@ int kmsan_ioremap_page_range(unsigned long start, unsigned long end,
> mapped = __vmap_pages_range_noflush(
> vmalloc_origin(start + off),
> vmalloc_origin(start + off + PAGE_SIZE), prot, &origin,
> - PAGE_SHIFT);
> + PAGE_SHIFT, gfp_mask);
> if (mapped) {
> err = mapped;
> goto ret;
> diff --git a/mm/kmsan/shadow.c b/mm/kmsan/shadow.c
> index 212449db34a1..452760a333e4 100644
> --- a/mm/kmsan/shadow.c
> +++ b/mm/kmsan/shadow.c
> @@ -247,7 +247,7 @@ int kmsan_vmap_pages_range_noflush(unsigned long start, unsigned long end,
> origin_end = vmalloc_meta((void *)end, KMSAN_META_ORIGIN);
> kmsan_enter_runtime();
> mapped = __vmap_pages_range_noflush(shadow_start, shadow_end, prot,
> - s_pages, page_shift);
> + s_pages, page_shift, gfp_mask);
> kmsan_leave_runtime();
> if (mapped) {
> err = mapped;
> @@ -255,7 +255,7 @@ int kmsan_vmap_pages_range_noflush(unsigned long start, unsigned long end,
> }
> kmsan_enter_runtime();
> mapped = __vmap_pages_range_noflush(origin_start, origin_end, prot,
> - o_pages, page_shift);
> + o_pages, page_shift, gfp_mask);
> kmsan_leave_runtime();
> if (mapped) {
> /* Undo the shadow mapping set up above. */
> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> index 5c1eacc2a0ac..4f3c320a4849 100644
> --- a/mm/vmalloc.c
> +++ b/mm/vmalloc.c
> @@ -725,7 +725,8 @@ static int vmap_pages_p4d_range(pgd_t *pgd, unsigned long addr,
> * may have contiguous chunks, as hinted by @shift.
> */
> static int vmap_pages_range_noflush_walk(unsigned long addr, unsigned long end,
> - pgprot_t prot, struct page **pages, unsigned int shift)
> + pgprot_t prot, struct page **pages, unsigned int shift,
> + unsigned int *mapped_pages)
> {
> unsigned long start = addr;
> pgd_t *pgd;
> @@ -748,6 +749,7 @@ static int vmap_pages_range_noflush_walk(unsigned long addr, unsigned long end,
> if (mask & ARCH_PAGE_TABLE_SYNC_MASK)
> arch_sync_kernel_mappings(start, end);
>
> + *mapped_pages = nr;
> return err;
> }
>
> @@ -758,18 +760,36 @@ static int vmap_pages_range_noflush_walk(unsigned long addr, unsigned long end,
> * The caller is responsible for calling flush_cache_vmap() after this
> * function returns successfully and before the addresses are accessed.
> *
> + * On failure the PTEs installed by this call are torn down before
> + * returning, so a retry sees a clean range; PTEs that were already
> + * there when the walk failed are left alone. When @gfp_mask cannot
> + * block, the teardown skips cond_resched(), it may run in atomic
> + * context.
> + *
> * This is an internal function only. Do not use outside mm/.
> */
> int __vmap_pages_range_noflush(unsigned long addr, unsigned long end,
> - pgprot_t prot, struct page **pages, unsigned int page_shift)
> + pgprot_t prot, struct page **pages, unsigned int page_shift,
> + gfp_t gfp_mask)
> {
> + int err;
> + unsigned int mapped_pages = 0;
> +
> if (WARN_ON_ONCE(page_shift < PAGE_SHIFT))
> return -EINVAL;
>
> if (!IS_ENABLED(CONFIG_HAVE_ARCH_HUGE_VMALLOC))
> page_shift = PAGE_SHIFT;
>
> - return vmap_pages_range_noflush_walk(addr, end, prot, pages, page_shift);
> + err = vmap_pages_range_noflush_walk(addr, end, prot, pages, page_shift,
> + &mapped_pages);
> +
> + /* Undo the PTEs installed before the failure. */
> + if (err && mapped_pages)
> + __vunmap_range_noflush(addr, addr + (mapped_pages * PAGE_SIZE),
> + gfpflags_allow_blocking(gfp_mask));
> +
> + return err;
> }
>
> int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
> @@ -781,7 +801,8 @@ int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
>
> if (ret)
> return ret;
> - return __vmap_pages_range_noflush(addr, end, prot, pages, page_shift);
> + return __vmap_pages_range_noflush(addr, end, prot, pages, page_shift,
> + gfp_mask);
> }
>
> static int __vmap_pages_range(unsigned long addr, unsigned long end,
> @@ -3715,7 +3736,7 @@ static int vmap_pages_range_batched(unsigned long addr, unsigned long end,
> pgprot_t prot, struct page **pages)
> {
> const unsigned int nr_pages = (end - addr) >> PAGE_SHIFT;
> - unsigned int prev_shift = 0, batch_idx = 0;
> + unsigned int prev_shift = 0, batch_idx = 0, nr_mapped;
> unsigned long batch_start = addr, batch_end = addr;
> int err;
>
> @@ -3734,7 +3755,8 @@ static int vmap_pages_range_batched(unsigned long addr, unsigned long end,
>
> if (shift != prev_shift) {
> err = vmap_pages_range_noflush_walk(batch_start, batch_end,
> - prot, pages + batch_idx, prev_shift);
> + prot, pages + batch_idx, prev_shift,
> + &nr_mapped);
> if (err)
> goto out;
> prev_shift = shift;
> @@ -3756,7 +3778,7 @@ static int vmap_pages_range_batched(unsigned long addr, unsigned long end,
> /* Remaining */
> if (batch_start < end)
> err = vmap_pages_range_noflush_walk(batch_start, end, prot,
> - pages + batch_idx, prev_shift);
> + pages + batch_idx, prev_shift, &nr_mapped);
>
> out:
> if (!err)

So, you are recording nr_mapped but not undoing
vmap_pages_range_batched() partial mappings yet. You do that in patch
#9. It's weird to fix one place but leave the other one until later. I
would suggest merging patches 7 and 9 to have a complete fix.

> diff --git a/mm/vmalloc.h b/mm/vmalloc.h
> index 4ec114a75f04..9ea28845f713 100644
> --- a/mm/vmalloc.h
> +++ b/mm/vmalloc.h
> @@ -40,7 +40,8 @@ void clear_vm_uninitialized_flag(struct vm_struct *vm);
>
> int __must_check __vmap_pages_range_noflush(unsigned long addr,
> unsigned long end, pgprot_t prot,
> - struct page **pages, unsigned int page_shift);
> + struct page **pages, unsigned int page_shift,
> + gfp_t gfp_mask);
>
> void vunmap_range_noflush(unsigned long start, unsigned long end);
>
> --
> 2.25.1
>