diff mbox series

[09/10] mm/mmap: start distinguishing if vma can be removed in mergeability test

Message ID 20230309111258.24079-10-vbabka@suse.cz (mailing list archive)
State New
Headers show
Series cleanup vma_merge() and improve mergeability tests | expand

Commit Message

Vlastimil Babka March 9, 2023, 11:12 a.m. UTC
Since pre-git times, is_mergeable_vma() returns false for a vma with
vm_ops->close, so that no owner assumptions are violated in case the vma
is removed as part of the merge.

This check is currently very conservative and can prevent merging even
situations where vma can't be removed, such as simple expansion of
previous vma, as evidenced by commit d014cd7c1c35 ("mm, mremap: fix
mremap() expanding for vma's with vm_ops->close()")

In order to allow more merging when appropriate and simplify the code
that was made more complex by commit d014cd7c1c35, start distinguishing
cases where the vma can be really removed, and allow merging with
vm_ops->close otherwise.

As a first step, add a may_remove_vma parameter to is_mergeable_vma().
can_vma_merge_before() sets it to true, because when called from
vma_merge(), a removal of the vma is possible.

In can_vma_merge_after(), pass the parameter as false, because no
removal can occur in each of its callers:
- vma_merge() calls it on the 'prev' vma, which is never removed
- mmap_region() and do_brk_flags() call it to determine if it can expand
  a vma, which is not removed

As a result, vma's with vm_ops->close may now merge with compatible
ranges in more situations than previously. We can also revert commit
d014cd7c1c35 as the next step to simplify mremap code again.

Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
---
 mm/mmap.c | 15 ++++++++++-----
 1 file changed, 10 insertions(+), 5 deletions(-)

Comments

Lorenzo Stoakes March 15, 2023, 10:05 p.m. UTC | #1
On Thu, Mar 09, 2023 at 12:12:57PM +0100, Vlastimil Babka wrote:
> Since pre-git times, is_mergeable_vma() returns false for a vma with
> vm_ops->close, so that no owner assumptions are violated in case the vma
> is removed as part of the merge.
>
> This check is currently very conservative and can prevent merging even
> situations where vma can't be removed, such as simple expansion of
> previous vma, as evidenced by commit d014cd7c1c35 ("mm, mremap: fix
> mremap() expanding for vma's with vm_ops->close()")
>
> In order to allow more merging when appropriate and simplify the code
> that was made more complex by commit d014cd7c1c35, start distinguishing
> cases where the vma can be really removed, and allow merging with
> vm_ops->close otherwise.
>
> As a first step, add a may_remove_vma parameter to is_mergeable_vma().
> can_vma_merge_before() sets it to true, because when called from
> vma_merge(), a removal of the vma is possible.
>
> In can_vma_merge_after(), pass the parameter as false, because no
> removal can occur in each of its callers:
> - vma_merge() calls it on the 'prev' vma, which is never removed
> - mmap_region() and do_brk_flags() call it to determine if it can expand
>   a vma, which is not removed
>
> As a result, vma's with vm_ops->close may now merge with compatible
> ranges in more situations than previously. We can also revert commit
> d014cd7c1c35 as the next step to simplify mremap code again.
>
> Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
> ---
>  mm/mmap.c | 15 ++++++++++-----
>  1 file changed, 10 insertions(+), 5 deletions(-)
>
> diff --git a/mm/mmap.c b/mm/mmap.c
> index d20bbe9ec613..65503ea07f32 100644
> --- a/mm/mmap.c
> +++ b/mm/mmap.c
> @@ -742,12 +742,13 @@ int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma,
>
>  /*
>   * If the vma has a ->close operation then the driver probably needs to release
> - * per-vma resources, so we don't attempt to merge those.
> + * per-vma resources, so we don't attempt to merge those in case the caller
> + * indicates the current vma may be removed as part of the merge.

Nit: 'in case the caller indicates' -> 'if the caller indicates'

>   */
>  static inline bool is_mergeable_vma(struct vm_area_struct *vma,
>  		struct file *file, unsigned long vm_flags,
>  		struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
> -		struct anon_vma_name *anon_name)
> +		struct anon_vma_name *anon_name, bool may_remove_vma)
>  {
>  	/*
>  	 * VM_SOFTDIRTY should not prevent from VMA merging, if we
> @@ -761,7 +762,7 @@ static inline bool is_mergeable_vma(struct vm_area_struct *vma,
>  		return false;
>  	if (vma->vm_file != file)
>  		return false;
> -	if (vma->vm_ops && vma->vm_ops->close)
> +	if (may_remove_vma && vma->vm_ops && vma->vm_ops->close)
>  		return false;
>  	if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
>  		return false;
> @@ -793,6 +794,8 @@ static inline bool is_mergeable_anon_vma(struct anon_vma *anon_vma1,
>   * We don't check here for the merged mmap wrapping around the end of pagecache
>   * indices (16TB on ia32) because do_mmap() does not permit mmap's which
>   * wrap, nor mmaps which cover the final page at index -1UL.
> + *
> + * We assume the vma may be removed as part of the merge.
>   */
>  static bool
>  can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
> @@ -800,7 +803,7 @@ can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
>  		pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
>  		struct anon_vma_name *anon_name)
>  {
> -	if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) &&
> +	if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name, true) &&
>  	    is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
>  		if (vma->vm_pgoff == vm_pgoff)
>  			return true;
> @@ -814,6 +817,8 @@ can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
>   *
>   * We cannot merge two vmas if they have differently assigned (non-NULL)
>   * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
> + *
> + * We assume that vma is not removed as part of the merge.
>   */
>  static bool
>  can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
> @@ -821,7 +826,7 @@ can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
>  		pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
>  		struct anon_vma_name *anon_name)
>  {
> -	if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) &&
> +	if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name, false) &&
>  	    is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
>  		pgoff_t vm_pglen;
>  		vm_pglen = vma_pages(vma);
> --
> 2.39.2
>

I wonder whether this might be moved later into the actual vma_merge() logic so
we only abort a merge at the point a VMA with ->close() would otherwise be
removed? But obviously that would probably need some further clean up to make it
not add yet more complexity to an already very complicated bit of logic.

Otherwise, very nice, clear + conservative so,

Reviewed-By: Lorenzo Stoakes <lstoakes@gmail.com>
Vlastimil Babka March 16, 2023, 10:57 a.m. UTC | #2
On 3/15/23 23:05, Lorenzo Stoakes wrote:
> On Thu, Mar 09, 2023 at 12:12:57PM +0100, Vlastimil Babka wrote:
>> --- a/mm/mmap.c
>> +++ b/mm/mmap.c
>> @@ -742,12 +742,13 @@ int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma,
>>
>>  /*
>>   * If the vma has a ->close operation then the driver probably needs to release
>> - * per-vma resources, so we don't attempt to merge those.
>> + * per-vma resources, so we don't attempt to merge those in case the caller
>> + * indicates the current vma may be removed as part of the merge.
> 
> Nit: 'in case the caller indicates' -> 'if the caller indicates'

Ok, -fix patch below.
 
...

>> -	if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) &&
>> +	if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name, false) &&
>>  	    is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
>>  		pgoff_t vm_pglen;
>>  		vm_pglen = vma_pages(vma);
>> --
>> 2.39.2
>>
> 
> I wonder whether this might be moved later into the actual vma_merge() logic so
> we only abort a merge at the point a VMA with ->close() would otherwise be
> removed? But obviously that would probably need some further clean up to make it
> not add yet more complexity to an already very complicated bit of logic.

Yeah, can try that later to cover the remaining cases where
->close prevents merging unnecessarily.

> Otherwise, very nice, clear + conservative so,
> 
> Reviewed-By: Lorenzo Stoakes <lstoakes@gmail.com>

Thanks!

----8<----
From b8072720288491bdc887ebebbb099babcfb108a5 Mon Sep 17 00:00:00 2001
From: Vlastimil Babka <vbabka@suse.cz>
Date: Thu, 16 Mar 2023 11:54:27 +0100
Subject: [PATCH] mm/mmap: start distinguishing if vma can be removed in
 mergeability test-fix

Adjust comment as suggested by Lorenzo.

Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
---
 mm/mmap.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/mm/mmap.c b/mm/mmap.c
index 4259095a2982..b255e6352710 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -742,8 +742,8 @@ int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma,
 
 /*
  * If the vma has a ->close operation then the driver probably needs to release
- * per-vma resources, so we don't attempt to merge those in case the caller
- * indicates the current vma may be removed as part of the merge.
+ * per-vma resources, so we don't attempt to merge those if the caller indicates
+ * the current vma may be removed as part of the merge.
  */
 static inline bool is_mergeable_vma(struct vm_area_struct *vma,
 		struct file *file, unsigned long vm_flags,
diff mbox series

Patch

diff --git a/mm/mmap.c b/mm/mmap.c
index d20bbe9ec613..65503ea07f32 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -742,12 +742,13 @@  int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma,
 
 /*
  * If the vma has a ->close operation then the driver probably needs to release
- * per-vma resources, so we don't attempt to merge those.
+ * per-vma resources, so we don't attempt to merge those in case the caller
+ * indicates the current vma may be removed as part of the merge.
  */
 static inline bool is_mergeable_vma(struct vm_area_struct *vma,
 		struct file *file, unsigned long vm_flags,
 		struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
-		struct anon_vma_name *anon_name)
+		struct anon_vma_name *anon_name, bool may_remove_vma)
 {
 	/*
 	 * VM_SOFTDIRTY should not prevent from VMA merging, if we
@@ -761,7 +762,7 @@  static inline bool is_mergeable_vma(struct vm_area_struct *vma,
 		return false;
 	if (vma->vm_file != file)
 		return false;
-	if (vma->vm_ops && vma->vm_ops->close)
+	if (may_remove_vma && vma->vm_ops && vma->vm_ops->close)
 		return false;
 	if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
 		return false;
@@ -793,6 +794,8 @@  static inline bool is_mergeable_anon_vma(struct anon_vma *anon_vma1,
  * We don't check here for the merged mmap wrapping around the end of pagecache
  * indices (16TB on ia32) because do_mmap() does not permit mmap's which
  * wrap, nor mmaps which cover the final page at index -1UL.
+ *
+ * We assume the vma may be removed as part of the merge.
  */
 static bool
 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
@@ -800,7 +803,7 @@  can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
 		pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
 		struct anon_vma_name *anon_name)
 {
-	if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) &&
+	if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name, true) &&
 	    is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
 		if (vma->vm_pgoff == vm_pgoff)
 			return true;
@@ -814,6 +817,8 @@  can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
  *
  * We cannot merge two vmas if they have differently assigned (non-NULL)
  * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
+ *
+ * We assume that vma is not removed as part of the merge.
  */
 static bool
 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
@@ -821,7 +826,7 @@  can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
 		pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
 		struct anon_vma_name *anon_name)
 {
-	if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) &&
+	if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name, false) &&
 	    is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
 		pgoff_t vm_pglen;
 		vm_pglen = vma_pages(vma);