Re: [PATCH v10 9/9] fsdax: set a CoW flag when associate reflink mappings

[Shiyang Ruan <ruansy.fnst@fujitsu.com>]

在 2022/2/16 10:09, Dan Williams 写道:
> On Thu, Jan 27, 2022 at 4:41 AM Shiyang Ruan <ruansy.fnst@fujitsu.com> wrote:
>>
>> Introduce a PAGE_MAPPING_DAX_COW flag to support association with CoW file
>> mappings.  In this case, the dax-RMAP already takes the responsibility
>> to look up for shared files by given dax page.  The page->mapping is no
>> longer to used for rmap but for marking that this dax page is shared.
>> And to make sure disassociation works fine, we use page->index as
>> refcount, and clear page->mapping to the initial state when page->index
>> is decreased to 0.
>>
>> With the help of this new flag, it is able to distinguish normal case
>> and CoW case, and keep the warning in normal case.
>>
>> Signed-off-by: Shiyang Ruan <ruansy.fnst@fujitsu.com>
>> ---
>>   fs/dax.c                   | 65 ++++++++++++++++++++++++++++++++------
>>   include/linux/page-flags.h |  6 ++++
>>   2 files changed, 62 insertions(+), 9 deletions(-)
>>
>> diff --git a/fs/dax.c b/fs/dax.c
>> index 250794a5b789..88879c579c1f 100644
>> --- a/fs/dax.c
>> +++ b/fs/dax.c
>> @@ -334,13 +334,46 @@ static unsigned long dax_end_pfn(void *entry)
>>          for (pfn = dax_to_pfn(entry); \
>>                          pfn < dax_end_pfn(entry); pfn++)
>>
>> +static inline void dax_mapping_set_cow_flag(struct address_space *mapping)
>> +{
>> +       mapping = (struct address_space *)PAGE_MAPPING_DAX_COW;
>> +}
>> +
>> +static inline bool dax_mapping_is_cow(struct address_space *mapping)
>> +{
>> +       return (unsigned long)mapping == PAGE_MAPPING_DAX_COW;
>> +}
>> +
>>   /*
>> - * TODO: for reflink+dax we need a way to associate a single page with
>> - * multiple address_space instances at different linear_page_index()
>> - * offsets.
>> + * Set or Update the page->mapping with FS_DAX_MAPPING_COW flag.
>> + * Return true if it is an Update.
>> + */
>> +static inline bool dax_mapping_set_cow(struct page *page)
>> +{
>> +       if (page->mapping) {
>> +               /* flag already set */
>> +               if (dax_mapping_is_cow(page->mapping))
>> +                       return false;
>> +
>> +               /*
>> +                * This page has been mapped even before it is shared, just
>> +                * need to set this FS_DAX_MAPPING_COW flag.
>> +                */
>> +               dax_mapping_set_cow_flag(page->mapping);
>> +               return true;
>> +       }
>> +       /* Newly associate CoW mapping */
>> +       dax_mapping_set_cow_flag(page->mapping);
>> +       return false;
>> +}
>> +
>> +/*
>> + * When it is called in dax_insert_entry(), the cow flag will indicate that
>> + * whether this entry is shared by multiple files.  If so, set the page->mapping
>> + * to be FS_DAX_MAPPING_COW, and use page->index as refcount.
>>    */
>>   static void dax_associate_entry(void *entry, struct address_space *mapping,
>> -               struct vm_area_struct *vma, unsigned long address)
>> +               struct vm_area_struct *vma, unsigned long address, bool cow)
>>   {
>>          unsigned long size = dax_entry_size(entry), pfn, index;
>>          int i = 0;
>> @@ -352,9 +385,17 @@ static void dax_associate_entry(void *entry, struct address_space *mapping,
>>          for_each_mapped_pfn(entry, pfn) {
>>                  struct page *page = pfn_to_page(pfn);
>>
>> -               WARN_ON_ONCE(page->mapping);
>> -               page->mapping = mapping;
>> -               page->index = index + i++;
>> +               if (cow) {
>> +                       if (dax_mapping_set_cow(page)) {
>> +                               /* Was normal, now updated to CoW */
>> +                               page->index = 2;
>> +                       } else
>> +                               page->index++;
>> +               } else {
>> +                       WARN_ON_ONCE(page->mapping);
>> +                       page->mapping = mapping;
>> +                       page->index = index + i++;
>> +               }
>>          }
>>   }
>>
>> @@ -370,7 +411,12 @@ static void dax_disassociate_entry(void *entry, struct address_space *mapping,
>>                  struct page *page = pfn_to_page(pfn);
>>
>>                  WARN_ON_ONCE(trunc && page_ref_count(page) > 1);
>> -               WARN_ON_ONCE(page->mapping && page->mapping != mapping);
>> +               if (!dax_mapping_is_cow(page->mapping)) {
>> +                       /* keep the CoW flag if this page is still shared */
>> +                       if (page->index-- > 0)
>> +                               continue;
>> +               } else
>> +                       WARN_ON_ONCE(page->mapping && page->mapping != mapping);
>>                  page->mapping = NULL;
>>                  page->index = 0;
>>          }
>> @@ -810,7 +856,8 @@ static void *dax_insert_entry(struct xa_state *xas,
>>                  void *old;
>>
>>                  dax_disassociate_entry(entry, mapping, false);
>> -               dax_associate_entry(new_entry, mapping, vmf->vma, vmf->address);
>> +               dax_associate_entry(new_entry, mapping, vmf->vma, vmf->address,
>> +                               false);
> 
> Where is the caller that passes 'true'? Also when that caller arrives
> introduce a separate dax_associate_cow_entry() as that's easier to
> read than dax_associate_entry(..., true) in case someone does not
> remember what that boolean flag means.

This flag is supposed to be used when CoW support is introduced.  When 
it is a CoW operation, which is decided by iomap & srcmap's flag, this 
flag will be set true.

I think I should describe it in detail in the commit message.

> 
> However, it's not clear to me that this approach is a good idea given
> that the filesystem is the source of truth for how many address_spaces
> this page mapping might be duplicated. What about a iomap_page_ops for
> fsdax to ask the filesystem when it is ok to clear the mapping
> association for a page?

I'll think how to implement it in this way.


--
Thanks,
Ruan.

> 
>>                  /*
>>                   * Only swap our new entry into the page cache if the current
>>                   * entry is a zero page or an empty entry.  If a normal PTE or
>> diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
>> index 1c3b6e5c8bfd..6370d279795a 100644
>> --- a/include/linux/page-flags.h
>> +++ b/include/linux/page-flags.h
>> @@ -572,6 +572,12 @@ __PAGEFLAG(Reported, reported, PF_NO_COMPOUND)
>>   #define PAGE_MAPPING_KSM       (PAGE_MAPPING_ANON | PAGE_MAPPING_MOVABLE)
>>   #define PAGE_MAPPING_FLAGS     (PAGE_MAPPING_ANON | PAGE_MAPPING_MOVABLE)
>>
>> +/*
>> + * Different with flags above, this flag is used only for fsdax mode.  It
>> + * indicates that this page->mapping is now under reflink case.
>> + */
>> +#define PAGE_MAPPING_DAX_COW   0x1
>> +
>>   static __always_inline int PageMappingFlags(struct page *page)
>>   {
>>          return ((unsigned long)page->mapping & PAGE_MAPPING_FLAGS) != 0;
>> --
>> 2.34.1
>>
>>
>>