Re: find_get_entries_tag regression bisected

[Dan Williams <dan.j.williams@intel.com>]

On Sat, Feb 16, 2019 at 9:29 AM Matthew Wilcox <willy@infradead.org> wrote:
>
> On Sat, Feb 16, 2019 at 07:35:11AM -0800, Matthew Wilcox wrote:
> > Another way to fix this would be to mask the address in dax_entry_mkclean(),
> > but I think this is cleaner.
>
> That's clearly rubbish, dax_entry_mkclean() can't possibly mask the
> address.  It might be mis-aligned in another process.  But ... if it's
> misaligned in another process, dax_entry_mkclean() will only clean the first
> PTE associated with the PMD; it won't clean the whole thing.  I think we need
> something like this:
>
> (I'll have to split it apart to give us something to backport)

Looks good to me, care to send a formal patch?

Tested-by: Dan Williams <dan.j.williams@intel.com>


>
> diff --git a/fs/dax.c b/fs/dax.c
> index 6959837cc465..09680aa0481f 100644
> --- a/fs/dax.c
> +++ b/fs/dax.c
> @@ -768,7 +768,7 @@ unsigned long pgoff_address(pgoff_t pgoff, struct vm_area_struct *vma)
>
>  /* Walk all mappings of a given index of a file and writeprotect them */
>  static void dax_entry_mkclean(struct address_space *mapping, pgoff_t index,
> -               unsigned long pfn)
> +               pgoff_t end, unsigned long pfn)
>  {
>         struct vm_area_struct *vma;
>         pte_t pte, *ptep = NULL;
> @@ -776,7 +776,7 @@ static void dax_entry_mkclean(struct address_space *mapping, pgoff_t index,
>         spinlock_t *ptl;
>
>         i_mmap_lock_read(mapping);
> -       vma_interval_tree_foreach(vma, &mapping->i_mmap, index, index) {
> +       vma_interval_tree_foreach(vma, &mapping->i_mmap, index, end) {
>                 struct mmu_notifier_range range;
>                 unsigned long address;
>
> @@ -843,9 +843,9 @@ static void dax_entry_mkclean(struct address_space *mapping, pgoff_t index,
>  static int dax_writeback_one(struct xa_state *xas, struct dax_device *dax_dev,
>                 struct address_space *mapping, void *entry)
>  {
> -       unsigned long pfn;
> +       unsigned long pfn, index;
>         long ret = 0;
> -       size_t size;
> +       unsigned long count;
>
>         /*
>          * A page got tagged dirty in DAX mapping? Something is seriously
> @@ -894,17 +894,18 @@ static int dax_writeback_one(struct xa_state *xas, struct dax_device *dax_dev,
>         xas_unlock_irq(xas);
>
>         /*
> -        * Even if dax_writeback_mapping_range() was given a wbc->range_start
> -        * in the middle of a PMD, the 'index' we are given will be aligned to
> -        * the start index of the PMD, as will the pfn we pull from 'entry'.
> +        * If dax_writeback_mapping_range() was given a wbc->range_start
> +        * in the middle of a PMD, the 'index' we are given needs to be
> +        * aligned to the start index of the PMD.
>          * This allows us to flush for PMD_SIZE and not have to worry about
>          * partial PMD writebacks.
>          */
>         pfn = dax_to_pfn(entry);
> -       size = PAGE_SIZE << dax_entry_order(entry);
> +       count = 1UL << dax_entry_order(entry);
> +       index = xas->xa_index &~ (count - 1);
>
> -       dax_entry_mkclean(mapping, xas->xa_index, pfn);
> -       dax_flush(dax_dev, page_address(pfn_to_page(pfn)), size);
> +       dax_entry_mkclean(mapping, index, index + count - 1, pfn);
> +       dax_flush(dax_dev, page_address(pfn_to_page(pfn)), count * PAGE_SIZE);
>         /*
>          * After we have flushed the cache, we can clear the dirty tag. There
>          * cannot be new dirty data in the pfn after the flush has completed as
> @@ -917,8 +918,7 @@ static int dax_writeback_one(struct xa_state *xas, struct dax_device *dax_dev,
>         xas_clear_mark(xas, PAGECACHE_TAG_DIRTY);
>         dax_wake_entry(xas, entry, false);
>
> -       trace_dax_writeback_one(mapping->host, xas->xa_index,
> -                       size >> PAGE_SHIFT);
> +       trace_dax_writeback_one(mapping->host, xas->xa_index, count);
>         return ret;
>
>   put_unlocked:
>