diff mbox series

mm: pagemap: restrict pagewalk to the requested range

Message ID 20230515172608.3558391-1-yuanchu@google.com
State New
Headers
Series mm: pagemap: restrict pagewalk to the requested range |

Commit Message

Yuanchu Xie May 15, 2023, 5:26 p.m. UTC 
  The pagewalk in pagemap_read reads one PTE past the end of the requested
range, and stops when the buffer runs out of space. While it produces
the right result, the extra read is unnecessary and less performant.

I timed the following command before and after this patch:
	dd count=100000 if=/proc/self/pagemap of=/dev/null
The results are consistently within 0.001s across 5 runs.

Before:
100000+0 records in
100000+0 records out
51200000 bytes (51 MB) copied, 0.0763159 s, 671 MB/s

real    0m0.078s
user    0m0.012s
sys     0m0.065s

After:
100000+0 records in
100000+0 records out
51200000 bytes (51 MB) copied, 0.0487928 s, 1.0 GB/s

real    0m0.050s
user    0m0.011s
sys     0m0.039s

Signed-off-by: Yuanchu Xie <yuanchu@google.com>
---
 fs/proc/task_mmu.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

Comments

Peter Xu May 15, 2023, 10:05 p.m. UTC | #1 
On Tue, May 16, 2023 at 01:26:08AM +0800, Yuanchu Xie wrote:
> The pagewalk in pagemap_read reads one PTE past the end of the requested
> range, and stops when the buffer runs out of space. While it produces
> the right result, the extra read is unnecessary and less performant.
> 
> I timed the following command before and after this patch:
> 	dd count=100000 if=/proc/self/pagemap of=/dev/null
> The results are consistently within 0.001s across 5 runs.
> 
> Before:
> 100000+0 records in
> 100000+0 records out
> 51200000 bytes (51 MB) copied, 0.0763159 s, 671 MB/s
> 
> real    0m0.078s
> user    0m0.012s
> sys     0m0.065s
> 
> After:
> 100000+0 records in
> 100000+0 records out
> 51200000 bytes (51 MB) copied, 0.0487928 s, 1.0 GB/s
> 
> real    0m0.050s
> user    0m0.011s
> sys     0m0.039s
> 
> Signed-off-by: Yuanchu Xie <yuanchu@google.com>

Acked-by: Peter Xu <peterx@redhat.com>
Mike Rapoport May 16, 2023, 7:45 a.m. UTC | #2 
On Tue, May 16, 2023 at 01:26:08AM +0800, Yuanchu Xie wrote:
> The pagewalk in pagemap_read reads one PTE past the end of the requested
> range, and stops when the buffer runs out of space. While it produces
> the right result, the extra read is unnecessary and less performant.
> 
> I timed the following command before and after this patch:
> 	dd count=100000 if=/proc/self/pagemap of=/dev/null
> The results are consistently within 0.001s across 5 runs.
> 
> Before:
> 100000+0 records in
> 100000+0 records out
> 51200000 bytes (51 MB) copied, 0.0763159 s, 671 MB/s
> 
> real    0m0.078s
> user    0m0.012s
> sys     0m0.065s
> 
> After:
> 100000+0 records in
> 100000+0 records out
> 51200000 bytes (51 MB) copied, 0.0487928 s, 1.0 GB/s
> 
> real    0m0.050s
> user    0m0.011s
> sys     0m0.039s
> 
> Signed-off-by: Yuanchu Xie <yuanchu@google.com>

Acked-by: Mike Rapoport (IBM) <rppt@kernel.org>

> ---
>  fs/proc/task_mmu.c | 12 ++++++------
>  1 file changed, 6 insertions(+), 6 deletions(-)
> 
> diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
> index 420510f6a545..6259dd432eeb 100644
> --- a/fs/proc/task_mmu.c
> +++ b/fs/proc/task_mmu.c
> @@ -1689,23 +1689,23 @@ static ssize_t pagemap_read(struct file *file, char __user *buf,
>  	/* watch out for wraparound */
>  	start_vaddr = end_vaddr;
>  	if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) {
> +		unsigned long end;
> +
>  		ret = mmap_read_lock_killable(mm);
>  		if (ret)
>  			goto out_free;
>  		start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT);
>  		mmap_read_unlock(mm);
> +
> +		end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT);
> +		if (end >= start_vaddr && end < mm->task_size)
> +			end_vaddr = end;
>  	}
>  
>  	/* Ensure the address is inside the task */
>  	if (start_vaddr > mm->task_size)
>  		start_vaddr = end_vaddr;
>  
> -	/*
> -	 * The odds are that this will stop walking way
> -	 * before end_vaddr, because the length of the
> -	 * user buffer is tracked in "pm", and the walk
> -	 * will stop when we hit the end of the buffer.
> -	 */
>  	ret = 0;
>  	while (count && (start_vaddr < end_vaddr)) {
>  		int len;
> -- 
> 2.40.1.606.ga4b1b128d6-goog
> 
>
Yang Shi May 16, 2023, 5:10 p.m. UTC | #3 
On Mon, May 15, 2023 at 10:26 AM Yuanchu Xie <yuanchu@google.com> wrote:
>
> The pagewalk in pagemap_read reads one PTE past the end of the requested
> range, and stops when the buffer runs out of space. While it produces
> the right result, the extra read is unnecessary and less performant.
>
> I timed the following command before and after this patch:
>         dd count=100000 if=/proc/self/pagemap of=/dev/null
> The results are consistently within 0.001s across 5 runs.
>
> Before:
> 100000+0 records in
> 100000+0 records out
> 51200000 bytes (51 MB) copied, 0.0763159 s, 671 MB/s
>
> real    0m0.078s
> user    0m0.012s
> sys     0m0.065s
>
> After:
> 100000+0 records in
> 100000+0 records out
> 51200000 bytes (51 MB) copied, 0.0487928 s, 1.0 GB/s
>
> real    0m0.050s
> user    0m0.011s
> sys     0m0.039s
>
> Signed-off-by: Yuanchu Xie <yuanchu@google.com>

Reviewed-by: Yang Shi <shy828301@gmail.com>

> ---
>  fs/proc/task_mmu.c | 12 ++++++------
>  1 file changed, 6 insertions(+), 6 deletions(-)
>
> diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
> index 420510f6a545..6259dd432eeb 100644
> --- a/fs/proc/task_mmu.c
> +++ b/fs/proc/task_mmu.c
> @@ -1689,23 +1689,23 @@ static ssize_t pagemap_read(struct file *file, char __user *buf,
>         /* watch out for wraparound */
>         start_vaddr = end_vaddr;
>         if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) {
> +               unsigned long end;
> +
>                 ret = mmap_read_lock_killable(mm);
>                 if (ret)
>                         goto out_free;
>                 start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT);
>                 mmap_read_unlock(mm);
> +
> +               end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT);
> +               if (end >= start_vaddr && end < mm->task_size)
> +                       end_vaddr = end;
>         }
>
>         /* Ensure the address is inside the task */
>         if (start_vaddr > mm->task_size)
>                 start_vaddr = end_vaddr;
>
> -       /*
> -        * The odds are that this will stop walking way
> -        * before end_vaddr, because the length of the
> -        * user buffer is tracked in "pm", and the walk
> -        * will stop when we hit the end of the buffer.
> -        */
>         ret = 0;
>         while (count && (start_vaddr < end_vaddr)) {
>                 int len;
> --
> 2.40.1.606.ga4b1b128d6-goog
>
David Rientjes May 21, 2023, 11:55 p.m. UTC | #4 
On Tue, 16 May 2023, Yuanchu Xie wrote:

> The pagewalk in pagemap_read reads one PTE past the end of the requested
> range, and stops when the buffer runs out of space. While it produces
> the right result, the extra read is unnecessary and less performant.
> 
> I timed the following command before and after this patch:
> 	dd count=100000 if=/proc/self/pagemap of=/dev/null
> The results are consistently within 0.001s across 5 runs.
> 
> Before:
> 100000+0 records in
> 100000+0 records out
> 51200000 bytes (51 MB) copied, 0.0763159 s, 671 MB/s
> 
> real    0m0.078s
> user    0m0.012s
> sys     0m0.065s
> 
> After:
> 100000+0 records in
> 100000+0 records out
> 51200000 bytes (51 MB) copied, 0.0487928 s, 1.0 GB/s
> 
> real    0m0.050s
> user    0m0.011s
> sys     0m0.039s
> 
> Signed-off-by: Yuanchu Xie <yuanchu@google.com>

Acked-by: David Rientjes <rientjes@google.com>
diff mbox series

Patch

diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
index 420510f6a545..6259dd432eeb 100644
--- a/fs/proc/task_mmu.c
+++ b/fs/proc/task_mmu.c
@@ -1689,23 +1689,23 @@  static ssize_t pagemap_read(struct file *file, char __user *buf,
 	/* watch out for wraparound */
 	start_vaddr = end_vaddr;
 	if (svpfn <= (ULONG_MAX >> PAGE_SHIFT)) {
+		unsigned long end;
+
 		ret = mmap_read_lock_killable(mm);
 		if (ret)
 			goto out_free;
 		start_vaddr = untagged_addr_remote(mm, svpfn << PAGE_SHIFT);
 		mmap_read_unlock(mm);
+
+		end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT);
+		if (end >= start_vaddr && end < mm->task_size)
+			end_vaddr = end;
 	}
 
 	/* Ensure the address is inside the task */
 	if (start_vaddr > mm->task_size)
 		start_vaddr = end_vaddr;
 
-	/*
-	 * The odds are that this will stop walking way
-	 * before end_vaddr, because the length of the
-	 * user buffer is tracked in "pm", and the walk
-	 * will stop when we hit the end of the buffer.
-	 */
 	ret = 0;
 	while (count && (start_vaddr < end_vaddr)) {
 		int len;