[PATCH] fs/pipe.c: simplify tmp_page handling

[PATCH] fs/pipe.c: simplify tmp_page handling

[Rasmus Villemoes]

Assigning the newly allocated page to pipe->tmp_page, only to
unconditionally clear ->tmp_page a little later, seems somewhat odd.

It made sense prior to commit a194dfe6e6f6 ("pipe: Rearrange sequence
in pipe_write() to preallocate slot"), when a user copy was done
between the allocation and the buf->page = page assignment, and a
failure there would then just leave the pipe's one-element page cache
populated. Now, the same purpose is served by the page being inserted
as a size-0 buffer, and the next write attempting to merge with that
buffer.

Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
---
 fs/pipe.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/fs/pipe.c b/fs/pipe.c
index 4d0799e4e719..097400cce241 100644
--- a/fs/pipe.c
+++ b/fs/pipe.c
@@ -508,13 +508,14 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from)
 			struct page *page = pipe->tmp_page;
 			int copied;
 
-			if (!page) {
+			if (page) {
+				pipe->tmp_page = NULL;
+			} else {
 				page = alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT);
 				if (unlikely(!page)) {
 					ret = ret ? : -ENOMEM;
 					break;
 				}
-				pipe->tmp_page = page;
 			}
 
 			/* Allocate a slot in the ring in advance and attach an
@@ -534,7 +535,6 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from)
 				buf->flags = PIPE_BUF_FLAG_PACKET;
 			else
 				buf->flags = PIPE_BUF_FLAG_CAN_MERGE;
-			pipe->tmp_page = NULL;
 
 			copied = copy_page_from_iter(page, 0, PAGE_SIZE, from);
 			if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) {
-- 
2.48.1

Re: [PATCH] fs/pipe.c: simplify tmp_page handling

[Mateusz Guzik]

On Fri, Mar 07, 2025 at 11:10:04PM +0100, Rasmus Villemoes wrote:
> Assigning the newly allocated page to pipe->tmp_page, only to
> unconditionally clear ->tmp_page a little later, seems somewhat odd.
> 
> It made sense prior to commit a194dfe6e6f6 ("pipe: Rearrange sequence
> in pipe_write() to preallocate slot"), when a user copy was done
> between the allocation and the buf->page = page assignment, and a
> failure there would then just leave the pipe's one-element page cache
> populated. Now, the same purpose is served by the page being inserted
> as a size-0 buffer, and the next write attempting to merge with that
> buffer.
> 
> Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
> ---
>  fs/pipe.c | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/fs/pipe.c b/fs/pipe.c
> index 4d0799e4e719..097400cce241 100644
> --- a/fs/pipe.c
> +++ b/fs/pipe.c
> @@ -508,13 +508,14 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from)
>  			struct page *page = pipe->tmp_page;
>  			int copied;
>  
> -			if (!page) {
> +			if (page) {
> +				pipe->tmp_page = NULL;
> +			} else {
>  				page = alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT);
>  				if (unlikely(!page)) {
>  					ret = ret ? : -ENOMEM;
>  					break;
>  				}
> -				pipe->tmp_page = page;
>  			}
>  
>  			/* Allocate a slot in the ring in advance and attach an
> @@ -534,7 +535,6 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from)
>  				buf->flags = PIPE_BUF_FLAG_PACKET;
>  			else
>  				buf->flags = PIPE_BUF_FLAG_CAN_MERGE;
> -			pipe->tmp_page = NULL;
>  
>  			copied = copy_page_from_iter(page, 0, PAGE_SIZE, from);
>  			if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) {

This bit got reworked in https://lore.kernel.org/all/20250303230409.452687-3-mjguzik@gmail.com/

I see it is in -next, merged from https://web.git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs.git/log/?h=vfs-6.15.pipe

Re: [PATCH] fs/pipe.c: simplify tmp_page handling

[Oleg Nesterov]

On 03/07, Rasmus Villemoes wrote:
>
> Assigning the newly allocated page to pipe->tmp_page, only to
> unconditionally clear ->tmp_page a little later, seems somewhat odd.

Oh yes, but could you look at

	[PATCH v2 1/1] pipe: change pipe_write() to never add a zero-sized buffer
	https://lore.kernel.org/all/20250210114039.GA3588@redhat.com/

already in Christian's tree although I will probably need to rebase
this patch on top of the recent changes.

Oleg.

> It made sense prior to commit a194dfe6e6f6 ("pipe: Rearrange sequence
> in pipe_write() to preallocate slot"), when a user copy was done
> between the allocation and the buf->page = page assignment, and a
> failure there would then just leave the pipe's one-element page cache
> populated. Now, the same purpose is served by the page being inserted
> as a size-0 buffer, and the next write attempting to merge with that
> buffer.
> 
> Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
> ---
>  fs/pipe.c | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/fs/pipe.c b/fs/pipe.c
> index 4d0799e4e719..097400cce241 100644
> --- a/fs/pipe.c
> +++ b/fs/pipe.c
> @@ -508,13 +508,14 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from)
>  			struct page *page = pipe->tmp_page;
>  			int copied;
>  
> -			if (!page) {
> +			if (page) {
> +				pipe->tmp_page = NULL;
> +			} else {
>  				page = alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT);
>  				if (unlikely(!page)) {
>  					ret = ret ? : -ENOMEM;
>  					break;
>  				}
> -				pipe->tmp_page = page;
>  			}
>  
>  			/* Allocate a slot in the ring in advance and attach an
> @@ -534,7 +535,6 @@ pipe_write(struct kiocb *iocb, struct iov_iter *from)
>  				buf->flags = PIPE_BUF_FLAG_PACKET;
>  			else
>  				buf->flags = PIPE_BUF_FLAG_CAN_MERGE;
> -			pipe->tmp_page = NULL;
>  
>  			copied = copy_page_from_iter(page, 0, PAGE_SIZE, from);
>  			if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) {
> -- 
> 2.48.1
>