Re: [PATCH] nfs: Fix misuses of folio_shift() and folio_order()

[Trond Myklebust <trondmy@hammerspace.com>]

On Tue, 2024-05-28 at 22:03 +0100, Matthew Wilcox (Oracle) wrote:
> Page cache indices are in units of PAGE_SIZE, not in units of
> the folio size.  Revert the change in nfs_grow_file(), and
> pass the inode to nfs_folio_length() so it can be reimplemented
> in terms of folio_mkwrite_check_truncate() which handles this
> correctly.

For the record, the code being replaced here is not assuming that page
cache indices are in units of the folio size. It is assuming that folio
boundaries will lie on offsets that are multiples of the folio size and
that the current page attributes (page lock, uptodate, etc) are
expected to apply to the data that lies within those folio boundaries.
The way the folio code is written today, that assumption appears to be
correct.

I'm fine with replacing NFS-specific code with generic code when
obviously correct, but AFAICS this would be a cleanup, and not a bug
fix.

> 
> Fixes: 0c493b5cf16e ("NFS: Convert buffered writes to use folios")
> Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
> Cc: Trond Myklebust <trond.myklebust@hammerspace.com>
> Cc: Anna Schumaker <Anna.Schumaker@Netapp.com>
> Cc: Christoph Hellwig <hch@infradead.org>
> ---
>  fs/nfs/file.c           |  6 +++---
>  fs/nfs/internal.h       | 16 +++++-----------
>  fs/nfs/read.c           |  2 +-
>  fs/nfs/write.c          |  9 +++++----
>  include/linux/pagemap.h |  4 ++--
>  5 files changed, 16 insertions(+), 21 deletions(-)
> 
> diff --git a/fs/nfs/file.c b/fs/nfs/file.c
> index 6bd127e6683d..723d78bbfe3f 100644
> --- a/fs/nfs/file.c
> +++ b/fs/nfs/file.c
> @@ -301,7 +301,7 @@ EXPORT_SYMBOL_GPL(nfs_file_fsync);
>  static bool nfs_folio_is_full_write(struct folio *folio, loff_t pos,
>  				    unsigned int len)
>  {
> -	unsigned int pglen = nfs_folio_length(folio);
> +	unsigned int pglen = nfs_folio_length(folio, folio->mapping-
> >host);
>  	unsigned int offset = offset_in_folio(folio, pos);
>  	unsigned int end = offset + len;
>  
> @@ -386,7 +386,7 @@ static int nfs_write_end(struct file *file,
> struct address_space *mapping,
>  	 */
>  	if (!folio_test_uptodate(folio)) {
>  		size_t fsize = folio_size(folio);
> -		unsigned pglen = nfs_folio_length(folio);
> +		unsigned pglen = nfs_folio_length(folio, mapping-
> >host);
>  		unsigned end = offset + copied;
>  
>  		if (pglen == 0) {
> @@ -610,7 +610,7 @@ static vm_fault_t nfs_vm_page_mkwrite(struct
> vm_fault *vmf)
>  
>  	folio_wait_writeback(folio);
>  
> -	pagelen = nfs_folio_length(folio);
> +	pagelen = nfs_folio_length(folio, inode);
>  	if (pagelen == 0)
>  		goto out_unlock;
>  
> diff --git a/fs/nfs/internal.h b/fs/nfs/internal.h
> index 9f0f4534744b..3b0236e67257 100644
> --- a/fs/nfs/internal.h
> +++ b/fs/nfs/internal.h
> @@ -819,19 +819,13 @@ unsigned int nfs_page_length(struct page *page)
>  /*
>   * Determine the number of bytes of data the page contains
>   */
> -static inline size_t nfs_folio_length(struct folio *folio)
> +static inline size_t nfs_folio_length(struct folio *folio, struct
> inode *inode)
>  {
> -	loff_t i_size = i_size_read(folio_file_mapping(folio)-
> >host);
> +	ssize_t ret = folio_mkwrite_check_truncate(folio, inode);
>  
> -	if (i_size > 0) {
> -		pgoff_t index = folio_index(folio) >>
> folio_order(folio);
> -		pgoff_t end_index = (i_size - 1) >>
> folio_shift(folio);
> -		if (index < end_index)
> -			return folio_size(folio);
> -		if (index == end_index)
> -			return offset_in_folio(folio, i_size - 1) +
> 1;
> -	}
> -	return 0;
> +	if (ret < 0)
> +		ret = 0;
> +	return ret;
>  }
>  
>  /*
> diff --git a/fs/nfs/read.c b/fs/nfs/read.c
> index a142287d86f6..ba3bb496f832 100644
> --- a/fs/nfs/read.c
> +++ b/fs/nfs/read.c
> @@ -296,7 +296,7 @@ int nfs_read_add_folio(struct
> nfs_pageio_descriptor *pgio,
>  	unsigned int len, aligned_len;
>  	int error;
>  
> -	len = nfs_folio_length(folio);
> +	len = nfs_folio_length(folio, inode);
>  	if (len == 0)
>  		return nfs_return_empty_folio(folio);
>  
> diff --git a/fs/nfs/write.c b/fs/nfs/write.c
> index 2329cbb0e446..7713ce7c5b3a 100644
> --- a/fs/nfs/write.c
> +++ b/fs/nfs/write.c
> @@ -278,8 +278,8 @@ static void nfs_grow_file(struct folio *folio,
> unsigned int offset,
>  
>  	spin_lock(&inode->i_lock);
>  	i_size = i_size_read(inode);
> -	end_index = ((i_size - 1) >> folio_shift(folio)) <<
> folio_order(folio);
> -	if (i_size > 0 && folio_index(folio) < end_index)
> +	end_index = (i_size - 1) >> PAGE_SHIFT;
> +	if (i_size > 0 && folio->index < end_index)
>  		goto out;
>  	end = folio_file_pos(folio) + (loff_t)offset +
> (loff_t)count;
>  	if (i_size >= end)
> @@ -358,7 +358,8 @@ nfs_page_group_search_locked(struct nfs_page
> *head, unsigned int page_offset)
>   */
>  static bool nfs_page_group_covers_page(struct nfs_page *req)
>  {
> -	unsigned int len = nfs_folio_length(nfs_page_to_folio(req));
> +	struct folio *folio = nfs_page_to_folio(req);
> +	unsigned int len = nfs_folio_length(folio, folio->mapping-
> >host);
>  	struct nfs_page *tmp;
>  	unsigned int pos = 0;
>  
> @@ -1356,7 +1357,7 @@ int nfs_update_folio(struct file *file, struct
> folio *folio,
>  	struct nfs_open_context *ctx = nfs_file_open_context(file);
>  	struct address_space *mapping = folio_file_mapping(folio);
>  	struct inode *inode = mapping->host;
> -	unsigned int pagelen = nfs_folio_length(folio);
> +	unsigned int pagelen = nfs_folio_length(folio, inode);
>  	int		status = 0;
>  
>  	nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
> diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
> index c6aaceed0de6..df57d7361a9a 100644
> --- a/include/linux/pagemap.h
> +++ b/include/linux/pagemap.h
> @@ -212,8 +212,8 @@ enum mapping_flags {
>  	AS_FOLIO_ORDER_MAX = 21, /* Bits 16-25 are used for
> FOLIO_ORDER */
>  };
>  
> -#define AS_FOLIO_ORDER_MIN_MASK 0x001f0000
> -#define AS_FOLIO_ORDER_MAX_MASK 0x03e00000
> +#define AS_FOLIO_ORDER_MIN_MASK (31 << AS_FOLIO_ORDER_MIN)
> +#define AS_FOLIO_ORDER_MAX_MASK (31 << AS_FOLIO_ORDER_MAX)
>  #define AS_FOLIO_ORDER_MASK (AS_FOLIO_ORDER_MIN_MASK |
> AS_FOLIO_ORDER_MAX_MASK)
>  
>  /**

-- 
Trond Myklebust
Linux NFS client maintainer, Hammerspace
trond.myklebust@hammerspace.com