Skip to contentSkip to conversation

mm/gup: some cleanups

PATCH Superseded John Hubbard 6 participants 19 messages Feb 3, 2022

Subsystems (5)

v3 of 5 Superseded Latest is v5 →

lore
Share & tools
lore

John Hubbard

author reviewer
#0 lore raw

v3 cover mm/gup: some cleanups

Hi Peter, Jason and all,

Changes since v2:

* Patch 2: Removed an unnecessary line that was causing a clang-analyzer complaint, as reported by Lukas Bulwahn
[1] (thanks!), and retested locally.

Jason: I've boldly left your Reviewed-by tag on this patch, because I am predicting you'll agree with it...

* Added Reviewed-by tags from Jan Kara, Christoph Hellwig, and Jason Gunthorpe that have collected since v2.

Changes since v1:
* Patch 4: changed from get_user_pages(), to get_user_pages_fast().

* Patch 4: Rewrote the commit description--thanks to Jan Kara for that feedback.

* Patch 1: Removed Jerome's Cc from patch 1, due to a stale email address.

* Added Reviewed-by's from David Hildenbrand and Jason Gunthorpe.

Original cover letter, updated as necessary:

I'm including Peter's patch as the first one in this tiny series. (The commit description has my r-b tag in place of my Cc, and removes Jerome's Cc because he is no longer at redhat.com) The second patch is what I had in mind for a follow-up to that, when we were discussing that fix [2].

Plus, a couple more small removals that I had queued up:

The third patch removes a completely unused routine:
pin_user_pages_locked().

The forth patch removes a similar routine, get_user_pages_locked(), that only has one caller. It now calls get_user_pages_fast(), instead.

v1 of this patchset is here:
https://lore.kernel.org/all/[email protected]/

v2 of this patchset is here:
https://lore.kernel.org/r/[email protected]

[1] https://lore.kernel.org/r/CAKXUXMxFK9bo8jDoRZbQ0r2j-JwAGg3Xc5cpAcLaHfwHddJ7ew@mail.gmail.com

[2] https://lore.kernel.org/all/[email protected]/

thanks,
John Hubbard

John Hubbard (3):
mm/gup: clean up follow_pfn_pte() slightly
mm/gup: remove unused pin_user_pages_locked()
mm/gup: remove get_user_pages_locked()

Peter Xu (1):
mm: Fix invalid page pointer returned with FOLL_PIN gups

 include/linux/mm.h |  4 --
 mm/gup.c           | 99 +++-------------------------------------------
 mm/mempolicy.c     | 21 +++++-----
 3 files changed, 15 insertions(+), 109 deletions(-)

base-commit: 88808fbbead481aedb46640a5ace69c58287f56a --
2.35.1

author reviewer
#1 lore raw

PATCH v3 1/4 mm: Fix invalid page pointer returned with FOLL_PIN gups

From: Peter Xu <[email protected]>

Alex reported invalid page pointer returned with pin_user_pages_remote() from vfio after upstream commit 4b6c33b32296 ("vfio/type1: Prepare for batched pinning with struct vfio_batch"). This problem breaks NVIDIA vfio mdev.

It turns out that it's not the fault of the vfio commit; however after vfio switches to a full page buffer to store the page pointers it starts to expose the problem easier.

The problem is for VM_PFNMAP vmas we should normally fail with an -EFAULT then vfio will carry on to handle the MMIO regions. However when the bug triggered, follow_page_mask() returned -EEXIST for such a page, which will jump over the current page, leaving that entry in **pages untouched. However the caller is not aware of it, hence the caller will reference the page as usual even if the pointer data can be anything.

We had that -EEXIST logic since commit 1027e4436b6a ("mm: make GUP handle pfn mapping unless FOLL_GET is requested") which seems very reasonable. It could be that when we reworked GUP with FOLL_PIN we could have overlooked that special path in commit 3faa52c03f44 ("mm/gup: track FOLL_PIN pages"), even if that commit rightfully touched up follow_devmap_pud() on checking FOLL_PIN when it needs to return an -EEXIST.

Attaching the Fixes to the FOLL_PIN rework commit, as it happened later than 1027e4436b6a.

Cc Andrea Arcangeli <[email protected]>
Cc Andrew Morton <[email protected]>
Cc Jan Kara <[email protected]>
Cc Kirill A. Shutemov <[email protected]>
Fixes 3faa52c03f44 ("mm/gup: track FOLL_PIN pages")
---
 mm/gup.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
mm/gup.c +1 -1
index f0af462ac1e2..65575ae3602f 100644
@@ -440,7 +440,7 @@ static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address,
 		pte_t *pte, unsigned int flags)
 {
 	/* No page to get reference */
-	if (flags & FOLL_GET)
+	if (flags & (FOLL_GET | FOLL_PIN))
 		return -EFAULT;
 
 	if (flags & FOLL_TOUCH) {
Signature · 1 line
2.35.1
Reviewed-by on v3 1/4
#2 lore raw

Re: [PATCH v3 1/4] mm: Fix invalid page pointer returned with FOLL_PIN gups

Quoted 27 lines · On Thu, 3 Feb 2022 01:32:29 -0800 John Hubbard <[email protected]> wrote:

From: Peter Xu <[email protected]>

Alex reported invalid page pointer returned with pin_user_pages_remote() from vfio after upstream commit 4b6c33b32296 ("vfio/type1: Prepare for batched pinning with struct vfio_batch"). This problem breaks NVIDIA vfio mdev.

It turns out that it's not the fault of the vfio commit; however after vfio switches to a full page buffer to store the page pointers it starts to expose the problem easier.

The problem is for VM_PFNMAP vmas we should normally fail with an -EFAULT then vfio will carry on to handle the MMIO regions. However when the bug triggered, follow_page_mask() returned -EEXIST for such a page, which will jump over the current page, leaving that entry in **pages untouched. However the caller is not aware of it, hence the caller will reference the page as usual even if the pointer data can be anything.

We had that -EEXIST logic since commit 1027e4436b6a ("mm: make GUP handle pfn mapping unless FOLL_GET is requested") which seems very reasonable. It could be that when we reworked GUP with FOLL_PIN we could have overlooked that special path in commit 3faa52c03f44 ("mm/gup: track FOLL_PIN pages"), even if that commit rightfully touched up follow_devmap_pud() on checking FOLL_PIN when it needs to return an -EEXIST.

Attaching the Fixes to the FOLL_PIN rework commit, as it happened later than 1027e4436b6a.

Quoted 11 lines
Cc Andrea Arcangeli <[email protected]>
Cc Andrew Morton <[email protected]>
Cc Jan Kara <[email protected]>
Cc Kirill A. Shutemov <[email protected]>
Fixes 3faa52c03f44 ("mm/gup: track FOLL_PIN pages")

you can add

although maybe this would look better if it were squashed into the next patch, as others have also suggested

Quoted 17 lines
---
 mm/gup.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
mm/gup.c +1 -1
index f0af462ac1e2..65575ae3602f 100644
@@ -440,7 +440,7 @@ static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address,
 		pte_t *pte, unsigned int flags)
 {
 	/* No page to get reference */
-	if (flags & FOLL_GET)
+	if (flags & (FOLL_GET | FOLL_PIN))
 		return -EFAULT;
 
 	if (flags & FOLL_TOUCH) {
author reviewer
#3 lore raw

Re: [PATCH v3 1/4] mm: Fix invalid page pointer returned with FOLL_PIN gups

Quoted 42 lines · Claudio Imbrenda
Quoted 27 lines · On Thu, 3 Feb 2022 01:32:29 -0800 John Hubbard <[email protected]> wrote:

From: Peter Xu <[email protected]>

Alex reported invalid page pointer returned with pin_user_pages_remote() from vfio after upstream commit 4b6c33b32296 ("vfio/type1: Prepare for batched pinning with struct vfio_batch"). This problem breaks NVIDIA vfio mdev.

It turns out that it's not the fault of the vfio commit; however after vfio switches to a full page buffer to store the page pointers it starts to expose the problem easier.

The problem is for VM_PFNMAP vmas we should normally fail with an -EFAULT then vfio will carry on to handle the MMIO regions. However when the bug triggered, follow_page_mask() returned -EEXIST for such a page, which will jump over the current page, leaving that entry in **pages untouched. However the caller is not aware of it, hence the caller will reference the page as usual even if the pointer data can be anything.

We had that -EEXIST logic since commit 1027e4436b6a ("mm: make GUP handle pfn mapping unless FOLL_GET is requested") which seems very reasonable. It could be that when we reworked GUP with FOLL_PIN we could have overlooked that special path in commit 3faa52c03f44 ("mm/gup: track FOLL_PIN pages"), even if that commit rightfully touched up follow_devmap_pud() on checking FOLL_PIN when it needs to return an -EEXIST.

Attaching the Fixes to the FOLL_PIN rework commit, as it happened later than 1027e4436b6a.

Quoted 11 lines
Cc Andrea Arcangeli <[email protected]>
Cc Andrew Morton <[email protected]>
Cc Jan Kara <[email protected]>
Cc Kirill A. Shutemov <[email protected]>
Fixes 3faa52c03f44 ("mm/gup: track FOLL_PIN pages")
Quoted 4 lines

you can add

Thanks!

Quoted 3 lines

although maybe this would look better if it were squashed into the next patch, as others have also suggested

I was thinking about that. It seems like this patch here cleanly addresses an oversight, and it is tiny and reasonably suitable for backporting.

Patch 2, on the other hand is less of a fix, and more of a "let's improve things". And now it is expanding now to cover move_pages() too. So maybe it is better to leave them separate, after all.

thanks,

Signature · 2 lines
John Hubbard
NVIDIA
Quoted 19 lines
Quoted 17 lines
---
  mm/gup.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
mm/gup.c +1 -1
index f0af462ac1e2..65575ae3602f 100644
@@ -440,7 +440,7 @@ static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address,
  		pte_t *pte, unsigned int flags)
  {
  	/* No page to get reference */
-	if (flags & FOLL_GET)
+	if (flags & (FOLL_GET | FOLL_PIN))
  		return -EFAULT;
 
  	if (flags & FOLL_TOUCH) {
#4 lore raw

Re: [PATCH v3 1/4] mm: Fix invalid page pointer returned with FOLL_PIN gups

In reply to John Hubbard
Quoted 5 lines · John Hubbard

From: Peter Xu <[email protected]>

Alex reported invalid page pointer returned with pin_user_pages_remote() from vfio after upstream commit 4b6c33b32296 ("vfio/type1: Prepare for batched pinning with struct vfio_batch"). This problem breaks NVIDIA vfio mdev.

There still isn't any nvidia vfio mdev driver in the tree, so this changelog stilldoesn't make sense.

author reviewer
#5 lore raw

Re: [PATCH v3 1/4] mm: Fix invalid page pointer returned with FOLL_PIN gups

Quoted 9 lines · Christoph Hellwig
Quoted 5 lines · John Hubbard

From: Peter Xu <[email protected]>

Alex reported invalid page pointer returned with pin_user_pages_remote() from vfio after upstream commit 4b6c33b32296 ("vfio/type1: Prepare for batched pinning with struct vfio_batch"). This problem breaks NVIDIA vfio mdev.

There still isn't any nvidia vfio mdev driver in the tree, so this changelog stilldoesn't make sense.

I'll remove that last sentence (and put in a tiny note that I'm scribbling on Peter's commit description a little bit more), in order to avoid any references to out of tree things.

thanks,

Signature · 2 lines
John Hubbard
NVIDIA
author reviewer
#6 lore raw

PATCH v3 2/4 mm/gup: clean up follow_pfn_pte() slightly

In reply to John Hubbard

Regardless of any FOLL_* flags, get_user_pages() and its variants should handle PFN-only entries by stopping early, if the caller expected **pages to be filled in.

This makes for a more reliable API, as compared to the previous approach of skipping over such entries (and thus leaving them silently unwritten).

---
 mm/gup.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)
mm/gup.c +6 -5
index 65575ae3602f..cad3f28492e3 100644
@@ -439,10 +439,6 @@ static struct page *no_page_table(struct vm_area_struct *vma,
 static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address,
 		pte_t *pte, unsigned int flags)
 {
-	/* No page to get reference */
-	if (flags & (FOLL_GET | FOLL_PIN))
-		return -EFAULT;
-
 	if (flags & FOLL_TOUCH) {
 		pte_t entry = *pte;
 
@@ -1180,8 +1176,13 @@ static long __get_user_pages(struct mm_struct *mm,
 		} else if (PTR_ERR(page) == -EEXIST) {
 			/*
 			 * Proper page table entry exists, but no corresponding
-			 * struct page.
+			 * struct page. If the caller expects **pages to be
+			 * filled in, bail out now, because that can't be done
+			 * for this page.
 			 */
+			if (pages)
+				goto out;
+
 			goto next_page;
 		} else if (IS_ERR(page)) {
 			ret = PTR_ERR(page);
Signature · 1 line
2.35.1
#7 lore raw

Re: [PATCH v3 2/4] mm/gup: clean up follow_pfn_pte() slightly

Quoted 37 lines · On Thu, 3 Feb 2022 01:32:30 -0800 John Hubbard <[email protected]> wrote:

Regardless of any FOLL_* flags, get_user_pages() and its variants should handle PFN-only entries by stopping early, if the caller expected **pages to be filled in.

This makes for a more reliable API, as compared to the previous approach of skipping over such entries (and thus leaving them silently unwritten).

---
 mm/gup.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)
mm/gup.c +0 -5
index 65575ae3602f..cad3f28492e3 100644
@@ -439,10 +439,6 @@ static struct page *no_page_table(struct vm_area_struct *vma,
 static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address,
 		pte_t *pte, unsigned int flags)
 {
-	/* No page to get reference */
-	if (flags & (FOLL_GET | FOLL_PIN))
-		return -EFAULT;
-
 	if (flags & FOLL_TOUCH) {
 		pte_t entry = *pte;
 
@@ -1180,8 +1176,13 @@ static long __get_user_pages(struct mm_struct *mm,
 		} else if (PTR_ERR(page) == -EEXIST) {
 			/*
 			 * Proper page table entry exists, but no corresponding
-			 * struct page.

mm/gup.c static long __get_user_pages(struct mm_struct *mm, · line 1188

Quoted 10 lines
mm/gup.c +6 -0
+			 * struct page. If the caller expects **pages to be
+			 * filled in, bail out now, because that can't be done
+			 * for this page.
 			 */
+			if (pages)
+				goto out;
+
 			goto next_page;
 		} else if (IS_ERR(page)) {
 			ret = PTR_ERR(page);

I'm not an expert, can you explain why this is better, and why it does not cause new issues?

If I understand correctly, the problem you are trying to solve is that in some cases you might try to get n pages, but you only get m < n pages instead, because some don't have an associated struct page, and the missing pages might even be in the middle.

The pages array would contain the list of pages actually pinned (getted?), but this won't tell which of the requested pages have been pinned (e.g. if some pages in the middle of the run were skipped)

With your patch you will stop at the first page without a struct page, meaning that if the caller tries again, it will get 0 pages. Why won't this cause issues?

Why will this not cause problems when the pages parameter is NULL?

sorry for the dumb questions, but this seems a rather important change, and I think in these circumstances you can't have too much documentation.

author reviewer
#8 lore raw

Re: [PATCH v3 2/4] mm/gup: clean up follow_pfn_pte() slightly

On 2/3/22 05:31, Claudio Imbrenda wrote:
...

Quoted 19 lines
Quoted 5 lines
mm/gup.c +0 -1
@@ -1180,8 +1176,13 @@ static long __get_user_pages(struct mm_struct *mm,
  		} else if (PTR_ERR(page) == -EEXIST) {
  			/*
  			 * Proper page table entry exists, but no corresponding
-			 * struct page.

mm/gup.c static long __get_user_pages(struct mm_struct *mm, · line 1188

Quoted 10 lines
mm/gup.c +6 -0
+			 * struct page. If the caller expects **pages to be
+			 * filled in, bail out now, because that can't be done
+			 * for this page.
  			 */
+			if (pages)
+				goto out;
+
  			goto next_page;
  		} else if (IS_ERR(page)) {
  			ret = PTR_ERR(page);

I'm not an expert, can you explain why this is better, and why it does not cause new issues?

Quoted 9 lines

If I understand correctly, the problem you are trying to solve is that in some cases you might try to get n pages, but you only get m < n pages instead, because some don't have an associated struct page, and the missing pages might even be in the middle.

The pages array would contain the list of pages actually pinned (getted?), but this won't tell which of the requested pages have been pinned (e.g. if some pages in the middle of the run were skipped)

The get_user_pages() API doesn't leave pages in the middle, ever. Instead, it stops at the first error, and reports the number of page that were successfully pinned. And the caller is responsible for unpinning.

From __get_user_pages()'s kerneldoc documentation:

* Returns either number of pages pinned (which may be less than the
* number requested), or an error. Details about the return value:
*
* -- If nr_pages is 0, returns 0.
* -- If nr_pages is >0, but no pages were pinned, returns -errno.
* -- If nr_pages is >0, and some pages were pinned, returns the number of
* pages pinned. Again, this may be less than nr_pages.
* -- 0 return value is possible when the fault would need to be retried. *
* The caller is responsible for releasing returned @pages, via put_page().

So the **pages array doesn't have holes, and the caller just counts up from the beginning of **pages and stops at nr_pages.

Quoted 3 lines

With your patch you will stop at the first page without a struct page, meaning that if the caller tries again, it will get 0 pages. Why won't this cause issues?

Callers are already written to deal with this case.

Quoted 2 lines

Why will this not cause problems when the pages parameter is NULL?

The behavior is unchanged here if pages == NULL. But maybe you meant, if pages != NULL. And in that case, the new behavior is to stop early and return n < m, which is (I am claiming) better than just leaving garbage values in **pages.

Another approach would be to file in PTR_ERR(page) values, but GUP is a well-established and widely used API, and that would be a large change that would require changing a lot of caller code.

Quoted 6 lines

sorry for the dumb questions, but this seems a rather important change, and I think in these circumstances you can't have too much documentation.

Thanks for reviewing this!

thanks,

Signature · 2 lines
John Hubbard
NVIDIA
#9 lore raw

Re: [PATCH v3 2/4] mm/gup: clean up follow_pfn_pte() slightly

In reply to John Hubbard
Quoted 22 lines · John Hubbard

Regardless of any FOLL_* flags, get_user_pages() and its variants should handle PFN-only entries by stopping early, if the caller expected **pages to be filled in.

This makes for a more reliable API, as compared to the previous approach of skipping over such entries (and thus leaving them silently unwritten).

---
 mm/gup.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)
mm/gup.c +0 -0
index 65575ae3602f..cad3f28492e3 100644

mm/gup.c static struct page *no_page_table(struct vm_area_struct *vma, · line 444

Quoted 10 lines
mm/gup.c +0 -4
@@ -439,10 +439,6 @@ static struct page *no_page_table(struct vm_area_struct *vma,
 static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address,
 		pte_t *pte, unsigned int flags)
 {
-	/* No page to get reference */
-	if (flags & (FOLL_GET | FOLL_PIN))
-		return -EFAULT;
-
 	if (flags & FOLL_TOUCH) {
 		pte_t entry = *pte;
 

This will also modify the error code returned from follow_page(). A quick audit shows that at least the user in mm/migrate.c will propagate this error code to userspace and I'm not sure the change in error code will not break something... EEXIST is a bit strange error code to get from move_pages(2).

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
#10 lore raw

Re: [PATCH v3 2/4] mm/gup: clean up follow_pfn_pte() slightly

Quoted 17 lines · Jan Kara
Quoted 16 lines · John Hubbard

Regardless of any FOLL_* flags, get_user_pages() and its variants should handle PFN-only entries by stopping early, if the caller expected **pages to be filled in.

This makes for a more reliable API, as compared to the previous approach of skipping over such entries (and thus leaving them silently unwritten).

 mm/gup.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)
Quoted 16 lines
Quoted 4 lines
mm/gup.c +1 -0
index 65575ae3602f..cad3f28492e3 100644
+++ b/mm/gup.c

mm/gup.c static struct page *no_page_table(struct vm_area_struct *vma, · line 444

Quoted 10 lines
mm/gup.c +0 -4
@@ -439,10 +439,6 @@ static struct page *no_page_table(struct vm_area_struct *vma,
 static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address,
 		pte_t *pte, unsigned int flags)
 {
-	/* No page to get reference */
-	if (flags & (FOLL_GET | FOLL_PIN))
-		return -EFAULT;
-
 	if (flags & FOLL_TOUCH) {
 		pte_t entry = *pte;
 

This will also modify the error code returned from follow_page().

Er, but isn't that the whole point of this entire design? It is what the commit that added it says:

commit 1027e4436b6a5c413c95d95e50d0f26348a602ac

Author: Kirill A. Shutemov <[email protected]>
Date:   Fri Sep 4 15:47:55 2015 -0700

mm: make GUP handle pfn mapping unless FOLL_GET is requested

With DAX, pfn mapping becoming more common. The patch adjusts GUP code to cover pfn mapping for cases when we don't need struct page to proceed.

To make it possible, let's change follow_page() code to return -EEXIST error code if proper page table entry exists, but no corresponding struct page. __get_user_page() would ignore the error code and move to the next page frame.

The immediate effect of the change is working MAP_POPULATE and mlock() on DAX mappings.

Quoted 4 lines

A quick audit shows that at least the user in mm/migrate.c will propagate this error code to userspace and I'm not sure the change in error code will not break something... EEXIST is a bit strange error code to get from move_pages(2).

That makes sense, maybe move_pages should squash the return codes to EEXIST?

Jason

#11 lore raw

Re: [PATCH v3 2/4] mm/gup: clean up follow_pfn_pte() slightly

Quoted 55 lines · Jason Gunthorpe
Quoted 17 lines · Jan Kara
Quoted 16 lines · John Hubbard

Regardless of any FOLL_* flags, get_user_pages() and its variants should handle PFN-only entries by stopping early, if the caller expected **pages to be filled in.

This makes for a more reliable API, as compared to the previous approach of skipping over such entries (and thus leaving them silently unwritten).

 mm/gup.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)
Quoted 16 lines
Quoted 4 lines
mm/gup.c +1 -0
index 65575ae3602f..cad3f28492e3 100644
+++ b/mm/gup.c

mm/gup.c static struct page *no_page_table(struct vm_area_struct *vma, · line 444

Quoted 10 lines
mm/gup.c +0 -4
@@ -439,10 +439,6 @@ static struct page *no_page_table(struct vm_area_struct *vma,
 static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address,
 		pte_t *pte, unsigned int flags)
 {
-	/* No page to get reference */
-	if (flags & (FOLL_GET | FOLL_PIN))
-		return -EFAULT;
-
 	if (flags & FOLL_TOUCH) {
 		pte_t entry = *pte;
 

This will also modify the error code returned from follow_page().

Er, but isn't that the whole point of this entire design? It is what the commit that added it says:

commit 1027e4436b6a5c413c95d95e50d0f26348a602ac

Author: Kirill A. Shutemov <[email protected]>
Date:   Fri Sep 4 15:47:55 2015 -0700

mm: make GUP handle pfn mapping unless FOLL_GET is requested

With DAX, pfn mapping becoming more common. The patch adjusts GUP code to cover pfn mapping for cases when we don't need struct page to proceed.

To make it possible, let's change follow_page() code to return -EEXIST error code if proper page table entry exists, but no corresponding struct page. __get_user_page() would ignore the error code and move to the next page frame.

The immediate effect of the change is working MAP_POPULATE and mlock() on DAX mappings.

Quoted 7 lines
Quoted 4 lines

A quick audit shows that at least the user in mm/migrate.c will propagate this error code to userspace and I'm not sure the change in error code will not break something... EEXIST is a bit strange error code to get from move_pages(2).

That makes sense, maybe move_pages should squash the return codes to EEXIST?

I think EFAULT is the closest:
This is a zero page or the memory area is not mapped by the process.

EBUSY implies it can be tried again later.

author reviewer
#12 lore raw

Re: [PATCH v3 2/4] mm/gup: clean up follow_pfn_pte() slightly

On 2/3/22 07:18, Matthew Wilcox wrote:
...

Quoted 27 lines
Quoted 27 lines
Quoted 1 line

This will also modify the error code returned from follow_page().

Er, but isn't that the whole point of this entire design? It is what the commit that added it says:

commit 1027e4436b6a5c413c95d95e50d0f26348a602ac

Author: Kirill A. Shutemov <[email protected]>
Date:   Fri Sep 4 15:47:55 2015 -0700

mm: make GUP handle pfn mapping unless FOLL_GET is requested

With DAX, pfn mapping becoming more common. The patch adjusts GUP code to cover pfn mapping for cases when we don't need struct page to proceed.

To make it possible, let's change follow_page() code to return -EEXIST error code if proper page table entry exists, but no corresponding struct page. __get_user_page() would ignore the error code and move to the next page frame.

The immediate effect of the change is working MAP_POPULATE and mlock() on DAX mappings.

Quoted 4 lines

A quick audit shows that at least the user in mm/migrate.c will propagate this error code to userspace and I'm not sure the change in error code will not break something... EEXIST is a bit strange error code to get from move_pages(2).

Quoted 9 lines
Quoted 2 lines

That makes sense, maybe move_pages should squash the return codes to EEXIST?

I think EFAULT is the closest:
This is a zero page or the memory area is not mapped by the process.

EBUSY implies it can be tried again later.

OK. I definitely need to rework the commit description now, but the diffs are looking like this:

mm/gup.c +6 -5
index 65575ae3602f..cad3f28492e3 100644
@@ -439,10 +439,6 @@ static struct page *no_page_table(struct vm_area_struct *vma,
 static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address,
 		pte_t *pte, unsigned int flags)
 {
-	/* No page to get reference */
-	if (flags & (FOLL_GET | FOLL_PIN))
-		return -EFAULT;
-
 	if (flags & FOLL_TOUCH) {
 		pte_t entry = *pte;
 
@@ -1180,8 +1176,13 @@ static long __get_user_pages(struct mm_struct *mm,
 		} else if (PTR_ERR(page) == -EEXIST) {
 			/*
 			 * Proper page table entry exists, but no corresponding
-			 * struct page.
+			 * struct page. If the caller expects **pages to be
+			 * filled in, bail out now, because that can't be done
+			 * for this page.
 			 */
+			if (pages)
+				goto out;
+
 			goto next_page;
 		} else if (IS_ERR(page)) {
 			ret = PTR_ERR(page);
mm/migrate.c +7 -0
index c7da064b4781..be0d5ae36dc1 100644
@@ -1761,6 +1761,13 @@ static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes,
 			continue;
 		}
 
+		/*
+		 * The move_pages() man page does not have an -EEXIST choice, so
+		 * use -EFAULT instead.
+		 */
+		if (err == -EEXIST)
+			err = -EFAULT;
+
 		/*
 		 * If the page is already on the target node (!err), store the
 		 * node, otherwise, store the err.

thanks,

Signature · 2 lines
John Hubbard
NVIDIA
author reviewer
#13 lore raw

PATCH v3 3/4 mm/gup: remove unused pin_user_pages_locked()

In reply to John Hubbard

This routine was used for a short while, but then the calling code was refactored and the only caller was removed.

---
 include/linux/mm.h |  2 --
 mm/gup.c           | 29 -----------------------------
 2 files changed, 31 deletions(-)
include/linux/mm.h +0 -2
index 213cc569b192..80c540c17d83 100644
@@ -1918,8 +1918,6 @@ long pin_user_pages(unsigned long start, unsigned long nr_pages,
 		    struct vm_area_struct **vmas);
 long get_user_pages_locked(unsigned long start, unsigned long nr_pages,
 		    unsigned int gup_flags, struct page **pages, int *locked);
-long pin_user_pages_locked(unsigned long start, unsigned long nr_pages,
-		    unsigned int gup_flags, struct page **pages, int *locked);
 long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
 		    struct page **pages, unsigned int gup_flags);
 long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
mm/gup.c +0 -29
index cad3f28492e3..b0ecbfe03928 100644
@@ -3119,32 +3119,3 @@ long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
 	return get_user_pages_unlocked(start, nr_pages, pages, gup_flags);
 }
 EXPORT_SYMBOL(pin_user_pages_unlocked);
-
-/*
- * pin_user_pages_locked() is the FOLL_PIN variant of get_user_pages_locked().
- * Behavior is the same, except that this one sets FOLL_PIN and rejects
- * FOLL_GET.
- */
-long pin_user_pages_locked(unsigned long start, unsigned long nr_pages,
-			   unsigned int gup_flags, struct page **pages,
-			   int *locked)
-{
-	/*
-	 * FIXME: Current FOLL_LONGTERM behavior is incompatible with
-	 * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on
-	 * vmas.  As there are no users of this flag in this call we simply
-	 * disallow this option for now.
-	 */
-	if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM))
-		return -EINVAL;
-
-	/* FOLL_GET and FOLL_PIN are mutually exclusive. */
-	if (WARN_ON_ONCE(gup_flags & FOLL_GET))
-		return -EINVAL;
-
-	gup_flags |= FOLL_PIN;
-	return __get_user_pages_locked(current->mm, start, nr_pages,
-				       pages, NULL, locked,
-				       gup_flags | FOLL_TOUCH);
-}
-EXPORT_SYMBOL(pin_user_pages_locked);
Signature · 1 line
2.35.1
Reviewed-by on v3 3/4
#14 lore raw

Re: [PATCH v3 3/4] mm/gup: remove unused pin_user_pages_locked()

Quoted 8 lines · On Thu, 3 Feb 2022 01:32:31 -0800 John Hubbard <[email protected]> wrote:

This routine was used for a short while, but then the calling code was refactored and the only caller was removed.

if it's not used anymore, good riddance

Quoted 55 lines
---
 include/linux/mm.h |  2 --
 mm/gup.c           | 29 -----------------------------
 2 files changed, 31 deletions(-)
include/linux/mm.h +0 -2
index 213cc569b192..80c540c17d83 100644
@@ -1918,8 +1918,6 @@ long pin_user_pages(unsigned long start, unsigned long nr_pages,
 		    struct vm_area_struct **vmas);
 long get_user_pages_locked(unsigned long start, unsigned long nr_pages,
 		    unsigned int gup_flags, struct page **pages, int *locked);
-long pin_user_pages_locked(unsigned long start, unsigned long nr_pages,
-		    unsigned int gup_flags, struct page **pages, int *locked);
 long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
 		    struct page **pages, unsigned int gup_flags);
 long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
mm/gup.c +0 -29
index cad3f28492e3..b0ecbfe03928 100644
@@ -3119,32 +3119,3 @@ long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
 	return get_user_pages_unlocked(start, nr_pages, pages, gup_flags);
 }
 EXPORT_SYMBOL(pin_user_pages_unlocked);
-
-/*
- * pin_user_pages_locked() is the FOLL_PIN variant of get_user_pages_locked().
- * Behavior is the same, except that this one sets FOLL_PIN and rejects
- * FOLL_GET.
- */
-long pin_user_pages_locked(unsigned long start, unsigned long nr_pages,
-			   unsigned int gup_flags, struct page **pages,
-			   int *locked)
-{
-	/*
-	 * FIXME: Current FOLL_LONGTERM behavior is incompatible with
-	 * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on
-	 * vmas.  As there are no users of this flag in this call we simply
-	 * disallow this option for now.
-	 */
-	if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM))
-		return -EINVAL;
-
-	/* FOLL_GET and FOLL_PIN are mutually exclusive. */
-	if (WARN_ON_ONCE(gup_flags & FOLL_GET))
-		return -EINVAL;
-
-	gup_flags |= FOLL_PIN;
-	return __get_user_pages_locked(current->mm, start, nr_pages,
-				       pages, NULL, locked,
-				       gup_flags | FOLL_TOUCH);
-}
-EXPORT_SYMBOL(pin_user_pages_locked);
author reviewer
#15 lore raw

PATCH v3 4/4 mm/gup: remove get_user_pages_locked()

In reply to John Hubbard

There is only one caller of get_user_pages_locked(). The purpose of get_user_pages_locked() is to allow for unlocking the mmap_lock when reading a page from the disk during a page fault (hidden behind VM_FAULT_RETRY). The idea is to reduce contention on the heavily-used mmap_lock. (Thanks to Jan Kara for clearly pointing that out, and in fact I've used some of his wording here.)

However, it is unlikely for lookup_node() to take a page fault. With that in mind, change over to calling get_user_pages_fast(). This simplifies the code, runs a little faster in the expected case, and allows removing get_user_pages_locked() entirely.

---
 include/linux/mm.h |  2 --
 mm/gup.c           | 59 ----------------------------------------------
 mm/mempolicy.c     | 21 +++++++----------
 3 files changed, 9 insertions(+), 73 deletions(-)
include/linux/mm.h +0 -2
index 80c540c17d83..528ef1cb4f3a 100644
@@ -1916,8 +1916,6 @@ long get_user_pages(unsigned long start, unsigned long nr_pages,
 long pin_user_pages(unsigned long start, unsigned long nr_pages,
 		    unsigned int gup_flags, struct page **pages,
 		    struct vm_area_struct **vmas);
-long get_user_pages_locked(unsigned long start, unsigned long nr_pages,
-		    unsigned int gup_flags, struct page **pages, int *locked);
 long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
 		    struct page **pages, unsigned int gup_flags);
 long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
mm/gup.c +0 -59

65 lines not on this page · raw message

mm/mempolicy.c +9 -12
index 028e8dd82b44..3f8dc58da3e8 100644
@@ -907,17 +907,14 @@ static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes)
 static int lookup_node(struct mm_struct *mm, unsigned long addr)
 {
 	struct page *p = NULL;
-	int err;
+	int ret;
 
-	int locked = 1;
-	err = get_user_pages_locked(addr & PAGE_MASK, 1, 0, &p, &locked);
-	if (err > 0) {
-		err = page_to_nid(p);
+	ret = get_user_pages_fast(addr & PAGE_MASK, 1, 0, &p);
+	if (ret > 0) {
+		ret = page_to_nid(p);
 		put_page(p);
 	}
-	if (locked)
-		mmap_read_unlock(mm);
-	return err;
+	return ret;
 }
 
 /* Retrieve NUMA policy */
@@ -968,14 +965,14 @@ static long do_get_mempolicy(int *policy, nodemask_t *nmask,
 	if (flags & MPOL_F_NODE) {
 		if (flags & MPOL_F_ADDR) {
 			/*
-			 * Take a refcount on the mpol, lookup_node()
-			 * will drop the mmap_lock, so after calling
-			 * lookup_node() only "pol" remains valid, "vma"
-			 * is stale.
+			 * Take a refcount on the mpol, because we are about to
+			 * drop the mmap_lock, after which only "pol" remains
+			 * valid, "vma" is stale.
 			 */
 			pol_refcount = pol;
 			vma = NULL;
 			mpol_get(pol);
+			mmap_read_unlock(mm);
 			err = lookup_node(mm, addr);
 			if (err < 0)
 				goto out;
Signature · 1 line
2.35.1
Reviewed-by on v3 4/4
#16 lore raw

Re: [PATCH v3 4/4] mm/gup: remove get_user_pages_locked()

Quoted 7 lines · On Thu, 3 Feb 2022 01:32:32 -0800 John Hubbard <[email protected]> wrote:

There is only one caller of get_user_pages_locked(). The purpose of get_user_pages_locked() is to allow for unlocking the mmap_lock when reading a page from the disk during a page fault (hidden behind VM_FAULT_RETRY). The idea is to reduce contention on the heavily-used mmap_lock. (Thanks to Jan Kara for clearly pointing that out, and in fact I've used some of his wording here.)

Quoted 8 lines

However, it is unlikely for lookup_node() to take a page fault. With that in mind, change over to calling get_user_pages_fast(). This simplifies the code, runs a little faster in the expected case, and allows removing get_user_pages_locked() entirely.

I have always disliked these functions that might or might not unlock the lock under the hood. I'm happy to see one more go.

Quoted 135 lines
---
 include/linux/mm.h |  2 --
 mm/gup.c           | 59 ----------------------------------------------
 mm/mempolicy.c     | 21 +++++++----------
 3 files changed, 9 insertions(+), 73 deletions(-)
include/linux/mm.h +0 -2
index 80c540c17d83..528ef1cb4f3a 100644
@@ -1916,8 +1916,6 @@ long get_user_pages(unsigned long start, unsigned long nr_pages,
 long pin_user_pages(unsigned long start, unsigned long nr_pages,
 		    unsigned int gup_flags, struct page **pages,
 		    struct vm_area_struct **vmas);
-long get_user_pages_locked(unsigned long start, unsigned long nr_pages,
-		    unsigned int gup_flags, struct page **pages, int *locked);
 long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
 		    struct page **pages, unsigned int gup_flags);
 long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
mm/gup.c +0 -59
index b0ecbfe03928..7da49df59110 100644
@@ -2118,65 +2118,6 @@ long get_user_pages(unsigned long start, unsigned long nr_pages,
 }
 EXPORT_SYMBOL(get_user_pages);
 
-/**
- * get_user_pages_locked() - variant of get_user_pages()
- *
- * @start:      starting user address
- * @nr_pages:   number of pages from start to pin
- * @gup_flags:  flags modifying lookup behaviour
- * @pages:      array that receives pointers to the pages pinned.
- *              Should be at least nr_pages long. Or NULL, if caller
- *              only intends to ensure the pages are faulted in.
- * @locked:     pointer to lock flag indicating whether lock is held and
- *              subsequently whether VM_FAULT_RETRY functionality can be
- *              utilised. Lock must initially be held.
- *
- * It is suitable to replace the form:
- *
- *      mmap_read_lock(mm);
- *      do_something()
- *      get_user_pages(mm, ..., pages, NULL);
- *      mmap_read_unlock(mm);
- *
- *  to:
- *
- *      int locked = 1;
- *      mmap_read_lock(mm);
- *      do_something()
- *      get_user_pages_locked(mm, ..., pages, &locked);
- *      if (locked)
- *          mmap_read_unlock(mm);
- *
- * We can leverage the VM_FAULT_RETRY functionality in the page fault
- * paths better by using either get_user_pages_locked() or
- * get_user_pages_unlocked().
- *
- */
-long get_user_pages_locked(unsigned long start, unsigned long nr_pages,
-			   unsigned int gup_flags, struct page **pages,
-			   int *locked)
-{
-	/*
-	 * FIXME: Current FOLL_LONGTERM behavior is incompatible with
-	 * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on
-	 * vmas.  As there are no users of this flag in this call we simply
-	 * disallow this option for now.
-	 */
-	if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM))
-		return -EINVAL;
-	/*
-	 * FOLL_PIN must only be set internally by the pin_user_pages*() APIs,
-	 * never directly by the caller, so enforce that:
-	 */
-	if (WARN_ON_ONCE(gup_flags & FOLL_PIN))
-		return -EINVAL;
-
-	return __get_user_pages_locked(current->mm, start, nr_pages,
-				       pages, NULL, locked,
-				       gup_flags | FOLL_TOUCH);
-}
-EXPORT_SYMBOL(get_user_pages_locked);
-
 /*
  * get_user_pages_unlocked() is suitable to replace the form:
  *
mm/mempolicy.c +9 -12
index 028e8dd82b44..3f8dc58da3e8 100644
@@ -907,17 +907,14 @@ static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes)
 static int lookup_node(struct mm_struct *mm, unsigned long addr)
 {
 	struct page *p = NULL;
-	int err;
+	int ret;
 
-	int locked = 1;
-	err = get_user_pages_locked(addr & PAGE_MASK, 1, 0, &p, &locked);
-	if (err > 0) {
-		err = page_to_nid(p);
+	ret = get_user_pages_fast(addr & PAGE_MASK, 1, 0, &p);
+	if (ret > 0) {
+		ret = page_to_nid(p);
 		put_page(p);
 	}
-	if (locked)
-		mmap_read_unlock(mm);
-	return err;
+	return ret;
 }
 
 /* Retrieve NUMA policy */
@@ -968,14 +965,14 @@ static long do_get_mempolicy(int *policy, nodemask_t *nmask,
 	if (flags & MPOL_F_NODE) {
 		if (flags & MPOL_F_ADDR) {
 			/*
-			 * Take a refcount on the mpol, lookup_node()
-			 * will drop the mmap_lock, so after calling
-			 * lookup_node() only "pol" remains valid, "vma"
-			 * is stale.
+			 * Take a refcount on the mpol, because we are about to
+			 * drop the mmap_lock, after which only "pol" remains
+			 * valid, "vma" is stale.
 			 */
 			pol_refcount = pol;
 			vma = NULL;
 			mpol_get(pol);
+			mmap_read_unlock(mm);
 			err = lookup_node(mm, addr);
 			if (err < 0)
 				goto out;
#17 lore raw

Re: [PATCH v3 4/4] mm/gup: remove get_user_pages_locked()

In reply to John Hubbard
Quoted 11 lines · John Hubbard

There is only one caller of get_user_pages_locked(). The purpose of get_user_pages_locked() is to allow for unlocking the mmap_lock when reading a page from the disk during a page fault (hidden behind VM_FAULT_RETRY). The idea is to reduce contention on the heavily-used mmap_lock. (Thanks to Jan Kara for clearly pointing that out, and in fact I've used some of his wording here.)

However, it is unlikely for lookup_node() to take a page fault. With that in mind, change over to calling get_user_pages_fast(). This simplifies the code, runs a little faster in the expected case, and allows removing get_user_pages_locked() entirely.

Maybe split the lookup_node changes into a separate patch, as that allows to document that change even better.

author reviewer
#18 lore raw

Re: [PATCH v3 4/4] mm/gup: remove get_user_pages_locked()

Quoted 8 lines · Christoph Hellwig
Quoted 7 lines · John Hubbard

There is only one caller of get_user_pages_locked(). The purpose of get_user_pages_locked() is to allow for unlocking the mmap_lock when reading a page from the disk during a page fault (hidden behind VM_FAULT_RETRY). The idea is to reduce contention on the heavily-used mmap_lock. (Thanks to Jan Kara for clearly pointing that out, and in fact I've used some of his wording here.)

Quoted 7 lines
Quoted 4 lines

However, it is unlikely for lookup_node() to take a page fault. With that in mind, change over to calling get_user_pages_fast(). This simplifies the code, runs a little faster in the expected case, and allows removing get_user_pages_locked() entirely.

Maybe split the lookup_node changes into a separate patch, as that allows to document that change even better.

OK, I'll do that.

thanks,

Signature · 2 lines
John Hubbard
NVIDIA