Re: [PATCH 04/10] mm, fsdax: Refactor memory-failure handler for dax mapping

[Dan Williams]

On Wed, Dec 30, 2020 at 8:59 AM Shiyang Ruan <ruansy.fnst@xxxxxxxxxxxxxx> wrote:
>
> The current memory_failure_dev_pagemap() can only handle single-mapped
> dax page for fsdax mode.  The dax page could be mapped by multiple files
> and offsets if we let reflink feature & fsdax mode work together.  So,
> we refactor current implementation to support handle memory failure on
> each file and offset.
>
> Signed-off-by: Shiyang Ruan <ruansy.fnst@xxxxxxxxxxxxxx>
> ---
>  fs/dax.c            | 21 +++++++++++
>  include/linux/dax.h |  1 +
>  include/linux/mm.h  |  9 +++++
>  mm/memory-failure.c | 91 ++++++++++++++++++++++++++++++++++-----------
>  4 files changed, 100 insertions(+), 22 deletions(-)
>
> diff --git a/fs/dax.c b/fs/dax.c
> index 5b47834f2e1b..799210cfa687 100644
> --- a/fs/dax.c
> +++ b/fs/dax.c
> @@ -378,6 +378,27 @@ static struct page *dax_busy_page(void *entry)
>         return NULL;
>  }
>
> +/*
> + * dax_load_pfn - Load pfn of the DAX entry corresponding to a page
> + * @mapping: The file whose entry we want to load
> + * @index:   The offset where the DAX entry located in
> + *
> + * Return:   pfn of the DAX entry
> + */
> +unsigned long dax_load_pfn(struct address_space *mapping, unsigned long index)
> +{
> +       XA_STATE(xas, &mapping->i_pages, index);
> +       void *entry;
> +       unsigned long pfn;
> +
> +       xas_lock_irq(&xas);
> +       entry = xas_load(&xas);
> +       pfn = dax_to_pfn(entry);
> +       xas_unlock_irq(&xas);
> +
> +       return pfn;
> +}
> +
>  /*
>   * dax_lock_mapping_entry - Lock the DAX entry corresponding to a page
>   * @page: The page whose entry we want to lock
> diff --git a/include/linux/dax.h b/include/linux/dax.h
> index b52f084aa643..89e56ceeffc7 100644
> --- a/include/linux/dax.h
> +++ b/include/linux/dax.h
> @@ -150,6 +150,7 @@ int dax_writeback_mapping_range(struct address_space *mapping,
>
>  struct page *dax_layout_busy_page(struct address_space *mapping);
>  struct page *dax_layout_busy_page_range(struct address_space *mapping, loff_t start, loff_t end);
> +unsigned long dax_load_pfn(struct address_space *mapping, unsigned long index);
>  dax_entry_t dax_lock_page(struct page *page);
>  void dax_unlock_page(struct page *page, dax_entry_t cookie);
>  #else
> diff --git a/include/linux/mm.h b/include/linux/mm.h
> index db6ae4d3fb4e..db3059a1853e 100644
> --- a/include/linux/mm.h
> +++ b/include/linux/mm.h
> @@ -1141,6 +1141,14 @@ static inline bool is_device_private_page(const struct page *page)
>                 page->pgmap->type == MEMORY_DEVICE_PRIVATE;
>  }
>
> +static inline bool is_device_fsdax_page(const struct page *page)
> +{
> +       return IS_ENABLED(CONFIG_DEV_PAGEMAP_OPS) &&
> +               IS_ENABLED(CONFIG_DEVICE_PRIVATE) &&
> +               is_zone_device_page(page) &&
> +               page->pgmap->type == MEMORY_DEVICE_FS_DAX;
> +}
> +

Have a look at the recent fixes to pfn_to_online_page() vs DAX pages [1].

This above page type check is racy given that the pfn could stop being
pfn_valid() while this check is running. I think hwpoison_filter()
needs an explicit check for whether the page is already referenced or
not. For example the current call to hwpoison_filter() from
memory_failure_dev_pagemap() is safe because the page has already been
validated as ZONE_DEVICE and is safe to de-reference page->pgmap.

[1]: http://lore.kernel.org/r/161058499000.1840162.702316708443239771.stgit@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx