Skip to contentSkip to conversation

mm: userfaultfd: refactor and add UFFDIO_CONTINUE_MODE_WP

PATCH Superseded Axel Rasmussen 4 participants 20 messages Mar 6, 2023 – Mar 8, 2023

Subsystems (7)

v3 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

v3 cover mm: userfaultfd: refactor and add UFFDIO_CONTINUE_MODE_WP

This series, currently based on 6.3-rc1, is divided into two parts:

- Commits 1-4 refactor userfaultfd ioctl code without behavior changes, with the main goal of improving consistency and reducing the number of function args.
- Commit 5 adds UFFDIO_CONTINUE_MODE_WP.

The refactors are sorted by increasing controversial-ness, the idea being we could drop some of the refactors if they are deemed not worth it.

Changelog:

v2->v3:
- rebase onto 6.3-rc1
- typedef a new type for mfill flags in patch 3/5 (suggested by Nadav)

v1->v2:
- refactor before adding the new flag, to avoid perpetuating messiness

Axel Rasmussen (5):
mm: userfaultfd: rename functions for clarity + consistency
mm: userfaultfd: don't pass around both mm and vma
mm: userfaultfd: combine 'mode' and 'wp_copy' arguments
mm: userfaultfd: don't separate addr + len arguments
mm: userfaultfd: add UFFDIO_CONTINUE_MODE_WP to install WP PTEs

 fs/userfaultfd.c                         | 120 +++++-------
 include/linux/hugetlb.h                  |  27 ++-
 include/linux/shmem_fs.h                 |   9 +-
 include/linux/userfaultfd_k.h            |  61 +++---
 include/uapi/linux/userfaultfd.h         |   7 +
 mm/hugetlb.c                             |  34 ++--
 mm/shmem.c                               |  14 +-
 mm/userfaultfd.c                         | 235 +++++++++++------------
 tools/testing/selftests/mm/userfaultfd.c |   4 +
 9 files changed, 247 insertions(+), 264 deletions(-)

--
2.40.0.rc0.216.gc4246ad0f0-goog

#1 lore raw

PATCH v3 1/5 mm: userfaultfd: rename functions for clarity + consistency

The basic problem is, over time we've added new userfaultfd ioctls, and we've refactored the code so functions which used to handle only one case are now re-used to deal with several cases. While this happened, we didn't bother to rename the functions.

Similarly, as we added new functions, we cargo-culted pieces of the now-inconsistent naming scheme, so those functions too ended up with names that don't make a lot of sense.

A key point here is, "copy" in most userfaultfd code refers specifically to UFFDIO_COPY, where we allocate a new page and copy its contents from userspace. There are many functions with "copy" in the name that don't actually do this (at least in some cases).

So, rename things into a consistent scheme. The high level idea is that the call stack for userfaultfd ioctls becomes:

userfaultfd_ioctl
-> userfaultfd_(particular ioctl)

    -> mfill_atomic_(particular kind of fill operation)
      -> mfill_atomic    /* loops over pages in range */
        -> mfill_atomic_pte    /* deals with single pages */
          -> mfill_atomic_pte_(particular kind of fill operation)
            -> mfill_atomic_install_pte

There are of course some special cases (shmem, hugetlb), but this is the general structure which all function names now adhere to.

---
 fs/userfaultfd.c              | 18 +++----
 include/linux/hugetlb.h       | 30 +++++------
 include/linux/userfaultfd_k.h | 18 +++----
 mm/hugetlb.c                  | 20 +++----
 mm/userfaultfd.c              | 98 +++++++++++++++++------------------
 5 files changed, 92 insertions(+), 92 deletions(-)
fs/userfaultfd.c +9 -9
index 44d1ee429eb0..365bf00dd8dd 100644
@@ -1741,9 +1741,9 @@ static int userfaultfd_copy(struct userfaultfd_ctx *ctx,
 	if (uffdio_copy.mode & ~(UFFDIO_COPY_MODE_DONTWAKE|UFFDIO_COPY_MODE_WP))
 		goto out;
 	if (mmget_not_zero(ctx->mm)) {
-		ret = mcopy_atomic(ctx->mm, uffdio_copy.dst, uffdio_copy.src,
-				   uffdio_copy.len, &ctx->mmap_changing,
-				   uffdio_copy.mode);
+		ret = mfill_atomic_copy(ctx->mm, uffdio_copy.dst, uffdio_copy.src,
+					uffdio_copy.len, &ctx->mmap_changing,
+					uffdio_copy.mode);
 		mmput(ctx->mm);
 	} else {
 		return -ESRCH;
@@ -1793,9 +1793,9 @@ static int userfaultfd_zeropage(struct userfaultfd_ctx *ctx,
 		goto out;
 
 	if (mmget_not_zero(ctx->mm)) {
-		ret = mfill_zeropage(ctx->mm, uffdio_zeropage.range.start,
-				     uffdio_zeropage.range.len,
-				     &ctx->mmap_changing);
+		ret = mfill_atomic_zeropage(ctx->mm, uffdio_zeropage.range.start,
+					   uffdio_zeropage.range.len,
+					   &ctx->mmap_changing);
 		mmput(ctx->mm);
 	} else {
 		return -ESRCH;
@@ -1903,9 +1903,9 @@ static int userfaultfd_continue(struct userfaultfd_ctx *ctx, unsigned long arg)
 		goto out;
 
 	if (mmget_not_zero(ctx->mm)) {
-		ret = mcopy_continue(ctx->mm, uffdio_continue.range.start,
-				     uffdio_continue.range.len,
-				     &ctx->mmap_changing);
+		ret = mfill_atomic_continue(ctx->mm, uffdio_continue.range.start,
+					    uffdio_continue.range.len,
+					    &ctx->mmap_changing);
 		mmput(ctx->mm);
 	} else {
 		return -ESRCH;
include/linux/hugetlb.h +15 -15
index 7c977d234aba..8f0467bf1cbd 100644
@@ -158,13 +158,13 @@ unsigned long hugetlb_total_pages(void);
 vm_fault_t hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
 			unsigned long address, unsigned int flags);
 #ifdef CONFIG_USERFAULTFD
-int hugetlb_mcopy_atomic_pte(struct mm_struct *dst_mm, pte_t *dst_pte,
-				struct vm_area_struct *dst_vma,
-				unsigned long dst_addr,
-				unsigned long src_addr,
-				enum mcopy_atomic_mode mode,
-				struct page **pagep,
-				bool wp_copy);
+int hugetlb_mfill_atomic_pte(struct mm_struct *dst_mm, pte_t *dst_pte,
+			     struct vm_area_struct *dst_vma,
+			     unsigned long dst_addr,
+			     unsigned long src_addr,
+			     enum mcopy_atomic_mode mode,
+			     struct page **pagep,
+			     bool wp_copy);
 #endif /* CONFIG_USERFAULTFD */
 bool hugetlb_reserve_pages(struct inode *inode, long from, long to,
 						struct vm_area_struct *vma,
@@ -393,14 +393,14 @@ static inline void hugetlb_free_pgd_range(struct mmu_gather *tlb,
 }
 
 #ifdef CONFIG_USERFAULTFD
-static inline int hugetlb_mcopy_atomic_pte(struct mm_struct *dst_mm,
-						pte_t *dst_pte,
-						struct vm_area_struct *dst_vma,
-						unsigned long dst_addr,
-						unsigned long src_addr,
-						enum mcopy_atomic_mode mode,
-						struct page **pagep,
-						bool wp_copy)
+static inline int hugetlb_mfill_atomic_pte(struct mm_struct *dst_mm,
+					   pte_t *dst_pte,
+					   struct vm_area_struct *dst_vma,
+					   unsigned long dst_addr,
+					   unsigned long src_addr,
+					   enum mcopy_atomic_mode mode,
+					   struct page **pagep,
+					   bool wp_copy)
 {
 	BUG();
 	return 0;
include/linux/userfaultfd_k.h +9 -9
index 3767f18114ef..468080125612 100644
@@ -61,15 +61,15 @@ extern int mfill_atomic_install_pte(struct mm_struct *dst_mm, pmd_t *dst_pmd,
 				    unsigned long dst_addr, struct page *page,
 				    bool newly_allocated, bool wp_copy);
 
-extern ssize_t mcopy_atomic(struct mm_struct *dst_mm, unsigned long dst_start,
-			    unsigned long src_start, unsigned long len,
-			    atomic_t *mmap_changing, __u64 mode);
-extern ssize_t mfill_zeropage(struct mm_struct *dst_mm,
-			      unsigned long dst_start,
-			      unsigned long len,
-			      atomic_t *mmap_changing);
-extern ssize_t mcopy_continue(struct mm_struct *dst_mm, unsigned long dst_start,
-			      unsigned long len, atomic_t *mmap_changing);
+extern ssize_t mfill_atomic_copy(struct mm_struct *dst_mm, unsigned long dst_start,
+				 unsigned long src_start, unsigned long len,
+				 atomic_t *mmap_changing, __u64 mode);
+extern ssize_t mfill_atomic_zeropage(struct mm_struct *dst_mm,
+				     unsigned long dst_start,
+				     unsigned long len,
+				     atomic_t *mmap_changing);
+extern ssize_t mfill_atomic_continue(struct mm_struct *dst_mm, unsigned long dst_start,
+				     unsigned long len, atomic_t *mmap_changing);
 extern int mwriteprotect_range(struct mm_struct *dst_mm,
 			       unsigned long start, unsigned long len,
 			       bool enable_wp, atomic_t *mmap_changing);
mm/hugetlb.c +10 -10
index 07abcb6eb203..4c9276549394 100644
@@ -6154,17 +6154,17 @@ vm_fault_t hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
 
 #ifdef CONFIG_USERFAULTFD
 /*
- * Used by userfaultfd UFFDIO_COPY.  Based on mcopy_atomic_pte with
- * modifications for huge pages.
+ * Used by userfaultfd UFFDIO_* ioctls. Based on userfaultfd's mfill_atomic_pte
+ * with modifications for hugetlb pages.
  */
-int hugetlb_mcopy_atomic_pte(struct mm_struct *dst_mm,
-			    pte_t *dst_pte,
-			    struct vm_area_struct *dst_vma,
-			    unsigned long dst_addr,
-			    unsigned long src_addr,
-			    enum mcopy_atomic_mode mode,
-			    struct page **pagep,
-			    bool wp_copy)
+int hugetlb_mfill_atomic_pte(struct mm_struct *dst_mm,
+			     pte_t *dst_pte,
+			     struct vm_area_struct *dst_vma,
+			     unsigned long dst_addr,
+			     unsigned long src_addr,
+			     enum mcopy_atomic_mode mode,
+			     struct page **pagep,
+			     bool wp_copy)
 {
 	bool is_continue = (mode == MCOPY_ATOMIC_CONTINUE);
 	struct hstate *h = hstate_vma(dst_vma);
mm/userfaultfd.c +49 -49

174 lines not on this page · raw message

Signature · 1 line
2.40.0.rc0.216.gc4246ad0f0-goog
reviewer Acked-by on v3 1/5
#2 lore raw

Re: [PATCH v3 1/5] mm: userfaultfd: rename functions for clarity + consistency

Quoted 25 lines · Axel Rasmussen

The basic problem is, over time we've added new userfaultfd ioctls, and we've refactored the code so functions which used to handle only one case are now re-used to deal with several cases. While this happened, we didn't bother to rename the functions.

Similarly, as we added new functions, we cargo-culted pieces of the now-inconsistent naming scheme, so those functions too ended up with names that don't make a lot of sense.

A key point here is, "copy" in most userfaultfd code refers specifically to UFFDIO_COPY, where we allocate a new page and copy its contents from userspace. There are many functions with "copy" in the name that don't actually do this (at least in some cases).

So, rename things into a consistent scheme. The high level idea is that the call stack for userfaultfd ioctls becomes:

userfaultfd_ioctl
-> userfaultfd_(particular ioctl)

    -> mfill_atomic_(particular kind of fill operation)
      -> mfill_atomic    /* loops over pages in range */
        -> mfill_atomic_pte    /* deals with single pages */
          -> mfill_atomic_pte_(particular kind of fill operation)
            -> mfill_atomic_install_pte
Quoted 4 lines

There are of course some special cases (shmem, hugetlb), but this is the general structure which all function names now adhere to.

Signature · 1 line
Peter Xu
#3 lore raw

PATCH v3 2/5 mm: userfaultfd: don't pass around both mm and vma

In reply to Axel Rasmussen

Quite a few userfaultfd functions took both mm and vma pointers as arguments. Since the mm is trivially accessible via vma->vm_mm, there's no reason to pass both; it just needlessly extends the already long argument list.

Get rid of the mm pointer, where possible, to shorten the argument list.

---
 fs/userfaultfd.c              |  2 +-
 include/linux/hugetlb.h       |  5 ++-
 include/linux/shmem_fs.h      |  4 +--
 include/linux/userfaultfd_k.h |  4 +--
 mm/hugetlb.c                  |  9 +++--
 mm/shmem.c                    |  7 ++--
 mm/userfaultfd.c              | 66 ++++++++++++++++-------------------
 7 files changed, 45 insertions(+), 52 deletions(-)
fs/userfaultfd.c +1 -1
index 365bf00dd8dd..84d5d402214a 100644
@@ -1629,7 +1629,7 @@ static int userfaultfd_unregister(struct userfaultfd_ctx *ctx,
 
 		/* Reset ptes for the whole vma range if wr-protected */
 		if (userfaultfd_wp(vma))
-			uffd_wp_range(mm, vma, start, vma_end - start, false);
+			uffd_wp_range(vma, start, vma_end - start, false);
 
 		new_flags = vma->vm_flags & ~__VM_UFFD_FLAGS;
 		prev = vma_merge(&vmi, mm, prev, start, vma_end, new_flags,
include/linux/hugetlb.h +2 -3
index 8f0467bf1cbd..8b9325f77ac3 100644
@@ -158,7 +158,7 @@ unsigned long hugetlb_total_pages(void);
 vm_fault_t hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
 			unsigned long address, unsigned int flags);
 #ifdef CONFIG_USERFAULTFD
-int hugetlb_mfill_atomic_pte(struct mm_struct *dst_mm, pte_t *dst_pte,
+int hugetlb_mfill_atomic_pte(pte_t *dst_pte,
 			     struct vm_area_struct *dst_vma,
 			     unsigned long dst_addr,
 			     unsigned long src_addr,
@@ -393,8 +393,7 @@ static inline void hugetlb_free_pgd_range(struct mmu_gather *tlb,
 }
 
 #ifdef CONFIG_USERFAULTFD
-static inline int hugetlb_mfill_atomic_pte(struct mm_struct *dst_mm,
-					   pte_t *dst_pte,
+static inline int hugetlb_mfill_atomic_pte(pte_t *dst_pte,
 					   struct vm_area_struct *dst_vma,
 					   unsigned long dst_addr,
 					   unsigned long src_addr,
include/linux/shmem_fs.h +2 -2
index 103d1000a5a2..b82916c25e61 100644
@@ -151,14 +151,14 @@ extern void shmem_uncharge(struct inode *inode, long pages);
 
 #ifdef CONFIG_USERFAULTFD
 #ifdef CONFIG_SHMEM
-extern int shmem_mfill_atomic_pte(struct mm_struct *dst_mm, pmd_t *dst_pmd,
+extern int shmem_mfill_atomic_pte(pmd_t *dst_pmd,
 				  struct vm_area_struct *dst_vma,
 				  unsigned long dst_addr,
 				  unsigned long src_addr,
 				  bool zeropage, bool wp_copy,
 				  struct page **pagep);
 #else /* !CONFIG_SHMEM */
-#define shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma, dst_addr, \
+#define shmem_mfill_atomic_pte(dst_pmd, dst_vma, dst_addr, \
 			       src_addr, zeropage, wp_copy, pagep) ({ BUG(); 0; })
 #endif /* CONFIG_SHMEM */
 #endif /* CONFIG_USERFAULTFD */
include/linux/userfaultfd_k.h +2 -2
index 468080125612..ba79e296fcc7 100644
@@ -56,7 +56,7 @@ enum mcopy_atomic_mode {
 	MCOPY_ATOMIC_CONTINUE,
 };
 
-extern int mfill_atomic_install_pte(struct mm_struct *dst_mm, pmd_t *dst_pmd,
+extern int mfill_atomic_install_pte(pmd_t *dst_pmd,
 				    struct vm_area_struct *dst_vma,
 				    unsigned long dst_addr, struct page *page,
 				    bool newly_allocated, bool wp_copy);
@@ -73,7 +73,7 @@ extern ssize_t mfill_atomic_continue(struct mm_struct *dst_mm, unsigned long dst
 extern int mwriteprotect_range(struct mm_struct *dst_mm,
 			       unsigned long start, unsigned long len,
 			       bool enable_wp, atomic_t *mmap_changing);
-extern long uffd_wp_range(struct mm_struct *dst_mm, struct vm_area_struct *vma,
+extern long uffd_wp_range(struct vm_area_struct *vma,
 			  unsigned long start, unsigned long len, bool enable_wp);
 
 /* mm helpers */
mm/hugetlb.c +4 -5
index 4c9276549394..b4bda5f7f29f 100644
@@ -6157,8 +6157,7 @@ vm_fault_t hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
  * Used by userfaultfd UFFDIO_* ioctls. Based on userfaultfd's mfill_atomic_pte
  * with modifications for hugetlb pages.
  */
-int hugetlb_mfill_atomic_pte(struct mm_struct *dst_mm,
-			     pte_t *dst_pte,
+int hugetlb_mfill_atomic_pte(pte_t *dst_pte,
 			     struct vm_area_struct *dst_vma,
 			     unsigned long dst_addr,
 			     unsigned long src_addr,
@@ -6277,7 +6276,7 @@ int hugetlb_mfill_atomic_pte(struct mm_struct *dst_mm,
 		folio_in_pagecache = true;
 	}
 
-	ptl = huge_pte_lock(h, dst_mm, dst_pte);
+	ptl = huge_pte_lock(h, dst_vma->vm_mm, dst_pte);
 
 	ret = -EIO;
 	if (folio_test_hwpoison(folio))
@@ -6319,9 +6318,9 @@ int hugetlb_mfill_atomic_pte(struct mm_struct *dst_mm,
 	if (wp_copy)
 		_dst_pte = huge_pte_mkuffd_wp(_dst_pte);
 
-	set_huge_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
+	set_huge_pte_at(dst_vma->vm_mm, dst_addr, dst_pte, _dst_pte);
 
-	hugetlb_count_add(pages_per_huge_page(h), dst_mm);
+	hugetlb_count_add(pages_per_huge_page(h), dst_vma->vm_mm);
 
 	/* No need to invalidate - it was non-present before */
 	update_mmu_cache(dst_vma, dst_addr, dst_pte);
mm/shmem.c +3 -4
index 448f393d8ab2..1d751b6cf1ac 100644
@@ -2415,8 +2415,7 @@ static struct inode *shmem_get_inode(struct mnt_idmap *idmap, struct super_block
 }
 
 #ifdef CONFIG_USERFAULTFD
-int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
-			   pmd_t *dst_pmd,
+int shmem_mfill_atomic_pte(pmd_t *dst_pmd,
 			   struct vm_area_struct *dst_vma,
 			   unsigned long dst_addr,
 			   unsigned long src_addr,
@@ -2506,11 +2505,11 @@ int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
 		goto out_release;
 
 	ret = shmem_add_to_page_cache(folio, mapping, pgoff, NULL,
-				      gfp & GFP_RECLAIM_MASK, dst_mm);
+				      gfp & GFP_RECLAIM_MASK, dst_vma->vm_mm);
 	if (ret)
 		goto out_release;
 
-	ret = mfill_atomic_install_pte(dst_mm, dst_pmd, dst_vma, dst_addr,
+	ret = mfill_atomic_install_pte(dst_pmd, dst_vma, dst_addr,
 				       &folio->page, true, wp_copy);
 	if (ret)
 		goto out_delete_from_cache;
mm/userfaultfd.c +31 -35

202 lines not on this page · raw message

Signature · 1 line
2.40.0.rc0.216.gc4246ad0f0-goog
reviewer Acked-by on v3 2/5
#4 lore raw

Re: [PATCH v3 2/5] mm: userfaultfd: don't pass around both mm and vma

Quoted 8 lines · Axel Rasmussen

Quite a few userfaultfd functions took both mm and vma pointers as arguments. Since the mm is trivially accessible via vma->vm_mm, there's no reason to pass both; it just needlessly extends the already long argument list.

Get rid of the mm pointer, where possible, to shorten the argument list.

One nit below:

Quoted 10 lines
mm/hugetlb.c +1 -1
@@ -6277,7 +6276,7 @@ int hugetlb_mfill_atomic_pte(struct mm_struct *dst_mm,
 		folio_in_pagecache = true;
 	}
 
-	ptl = huge_pte_lock(h, dst_mm, dst_pte);
+	ptl = huge_pte_lock(h, dst_vma->vm_mm, dst_pte);
 
 	ret = -EIO;
 	if (folio_test_hwpoison(folio))

mm/hugetlb.c int hugetlb_mfill_atomic_pte(struct mm_struct *dst_mm, · line 6323

Quoted 8 lines
mm/hugetlb.c +2 -2
@@ -6319,9 +6318,9 @@ int hugetlb_mfill_atomic_pte(struct mm_struct *dst_mm,
 	if (wp_copy)
 		_dst_pte = huge_pte_mkuffd_wp(_dst_pte);
 
-	set_huge_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
+	set_huge_pte_at(dst_vma->vm_mm, dst_addr, dst_pte, _dst_pte);
 
-	hugetlb_count_add(pages_per_huge_page(h), dst_mm);
+	hugetlb_count_add(pages_per_huge_page(h), dst_vma->vm_mm);

When vm_mm referenced multiple times (say, >=3?), let's still cache it in a temp var?

I'm not sure whether compiler is smart enough to already do that with a reg, even if so it may slightly improve readability too, imho, by avoiding the multiple but same indirection for the reader.

Thanks,

Signature · 1 line
Peter Xu
#5 lore raw

Re: [PATCH v3 2/5] mm: userfaultfd: don't pass around both mm and vma

Quoted 42 lines · Peter Xu

!! External Email

Quoted 8 lines · Axel Rasmussen

Quite a few userfaultfd functions took both mm and vma pointers as arguments. Since the mm is trivially accessible via vma->vm_mm, there's no reason to pass both; it just needlessly extends the already long argument list.

Get rid of the mm pointer, where possible, to shorten the argument list.

One nit below:

Quoted 10 lines
mm/hugetlb.c +1 -1
@@ -6277,7 +6276,7 @@ int hugetlb_mfill_atomic_pte(struct mm_struct *dst_mm,
             folio_in_pagecache = true;
     }
 
-     ptl = huge_pte_lock(h, dst_mm, dst_pte);
+     ptl = huge_pte_lock(h, dst_vma->vm_mm, dst_pte);
 
     ret = -EIO;
     if (folio_test_hwpoison(folio))

mm/hugetlb.c int hugetlb_mfill_atomic_pte(struct mm_struct *dst_mm, · line 6323

Quoted 8 lines
mm/hugetlb.c +2 -2
@@ -6319,9 +6318,9 @@ int hugetlb_mfill_atomic_pte(struct mm_struct *dst_mm,
     if (wp_copy)
             _dst_pte = huge_pte_mkuffd_wp(_dst_pte);
 
-     set_huge_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
+     set_huge_pte_at(dst_vma->vm_mm, dst_addr, dst_pte, _dst_pte);
 
-     hugetlb_count_add(pages_per_huge_page(h), dst_mm);
+     hugetlb_count_add(pages_per_huge_page(h), dst_vma->vm_mm);

When vm_mm referenced multiple times (say, >=3?), let's still cache it in a temp var?

Quoted 3 lines

I'm not sure whether compiler is smart enough to already do that with a reg, even if so it may slightly improve readability too, imho, by avoiding the multiple but same indirection for the reader.

I am not sure if you referred to this code specifically or in general. I once looked into it, and the compiler is really stupid in this regard and super conservative when it comes to aliasing. Even if you use “restrict” keyword or “__pure” or “__const” function attributes, in certain cases (function calls to other compilation units, or inline assembly - I don’t remember) the compiler might ignore them. Worse, llvm and gcc are inconsistent.

From code-generated perspective, I did not see a clear cut that benefits caching over not. From performance perspective the impact is negligible. I mention all of that because I thought it matters too, but it mostly does not.

That’s all to say that in most cases, I think that whatever makes the code more readable should be preferred. I think that you are correct in saying that “caching” it will make the code more readable, but performance-wise it is probably meaningless.

reviewer
#6 lore raw

Re: [PATCH v3 2/5] mm: userfaultfd: don't pass around both mm and vma

Quoted 42 lines · Nadav Amit
Quoted 42 lines · Peter Xu

!! External Email

Quoted 8 lines · Axel Rasmussen

Quite a few userfaultfd functions took both mm and vma pointers as arguments. Since the mm is trivially accessible via vma->vm_mm, there's no reason to pass both; it just needlessly extends the already long argument list.

Get rid of the mm pointer, where possible, to shorten the argument list.

One nit below:

Quoted 10 lines
mm/hugetlb.c +1 -1
@@ -6277,7 +6276,7 @@ int hugetlb_mfill_atomic_pte(struct mm_struct *dst_mm,
             folio_in_pagecache = true;
     }
 
-     ptl = huge_pte_lock(h, dst_mm, dst_pte);
+     ptl = huge_pte_lock(h, dst_vma->vm_mm, dst_pte);
 
     ret = -EIO;
     if (folio_test_hwpoison(folio))

mm/hugetlb.c int hugetlb_mfill_atomic_pte(struct mm_struct *dst_mm, · line 6323

Quoted 8 lines
mm/hugetlb.c +2 -2
@@ -6319,9 +6318,9 @@ int hugetlb_mfill_atomic_pte(struct mm_struct *dst_mm,
     if (wp_copy)
             _dst_pte = huge_pte_mkuffd_wp(_dst_pte);
 
-     set_huge_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
+     set_huge_pte_at(dst_vma->vm_mm, dst_addr, dst_pte, _dst_pte);
 
-     hugetlb_count_add(pages_per_huge_page(h), dst_mm);
+     hugetlb_count_add(pages_per_huge_page(h), dst_vma->vm_mm);

When vm_mm referenced multiple times (say, >=3?), let's still cache it in a temp var?

Quoted 5 lines
Quoted 3 lines

I'm not sure whether compiler is smart enough to already do that with a reg, even if so it may slightly improve readability too, imho, by avoiding the multiple but same indirection for the reader.

I am not sure if you referred to this code specifically or in general.

In general. IIRC there're more than one such case in this patch.

Quoted 12 lines

I once looked into it, and the compiler is really stupid in this regard and super conservative when it comes to aliasing. Even if you use “restrict” keyword or “__pure” or “__const” function attributes, in certain cases (function calls to other compilation units, or inline assembly - I don’t remember) the compiler might ignore them. Worse, llvm and gcc are inconsistent.

From code-generated perspective, I did not see a clear cut that benefits caching over not. From performance perspective the impact is negligible. I mention all of that because I thought it matters too, but it mostly does not.

Quoted 4 lines

That’s all to say that in most cases, I think that whatever makes the code more readable should be preferred. I think that you are correct in saying that “caching” it will make the code more readable, but performance-wise it is probably meaningless.

Thanks for the knowledge. I would suspect no matter how the output layout of the compilers will be the difference will be small. I prefer it more for readability as you said but not strongly either way.

Signature · 1 line
Peter Xu
#7 lore raw

PATCH v3 3/5 mm: userfaultfd: combine 'mode' and 'wp_copy' arguments

In reply to Axel Rasmussen

Many userfaultfd ioctl functions take both a 'mode' and a 'wp_copy' argument. In future commits we plan to plumb the flags through to more places, so we'd be proliferating the very long argument list even further.

Let's take the time to simplify the argument list. Combine the two arguments into one - and generalize, so when we add more flags in the future, it doesn't imply more function arguments.

Since the modes (copy, zeropage, continue) are mutually exclusive, store them as an integer value (0, 1, 2) in the low bits. Place combine-able flag bits in the high bits.

This is quite similar to an earlier patch proposed by Nadav Amit ("userfaultfd: introduce uffd_flags" - for some reason Lore no longer has a copy of the patch). The main difference is that patch only handled flags, whereas this patch *also* combines the "mode" argument into the same type to shorten the argument list.

---
 fs/userfaultfd.c              |  5 ++-
 include/linux/hugetlb.h       | 10 ++---
 include/linux/shmem_fs.h      |  5 ++-
 include/linux/userfaultfd_k.h | 34 ++++++++--------
 mm/hugetlb.c                  | 13 +++---
 mm/shmem.c                    |  7 ++--
 mm/userfaultfd.c              | 76 ++++++++++++++++-------------------
 7 files changed, 74 insertions(+), 76 deletions(-)
fs/userfaultfd.c +4 -1
index 84d5d402214a..b8e328123b71 100644
@@ -1714,6 +1714,7 @@ static int userfaultfd_copy(struct userfaultfd_ctx *ctx,
 	struct uffdio_copy uffdio_copy;
 	struct uffdio_copy __user *user_uffdio_copy;
 	struct userfaultfd_wake_range range;
+	int flags = 0;
 
 	user_uffdio_copy = (struct uffdio_copy __user *) arg;
 
@@ -1740,10 +1741,12 @@ static int userfaultfd_copy(struct userfaultfd_ctx *ctx,
 		goto out;
 	if (uffdio_copy.mode & ~(UFFDIO_COPY_MODE_DONTWAKE|UFFDIO_COPY_MODE_WP))
 		goto out;
+	if (uffdio_copy.mode & UFFDIO_COPY_MODE_WP)
+		flags |= MFILL_ATOMIC_WP;
 	if (mmget_not_zero(ctx->mm)) {
 		ret = mfill_atomic_copy(ctx->mm, uffdio_copy.dst, uffdio_copy.src,
 					uffdio_copy.len, &ctx->mmap_changing,
-					uffdio_copy.mode);
+					flags);
 		mmput(ctx->mm);
 	} else {
 		return -ESRCH;
include/linux/hugetlb.h +4 -6
index 8b9325f77ac3..6270a4786584 100644
@@ -162,9 +162,8 @@ int hugetlb_mfill_atomic_pte(pte_t *dst_pte,
 			     struct vm_area_struct *dst_vma,
 			     unsigned long dst_addr,
 			     unsigned long src_addr,
-			     enum mcopy_atomic_mode mode,
-			     struct page **pagep,
-			     bool wp_copy);
+			     uffd_flags_t flags,
+			     struct page **pagep);
 #endif /* CONFIG_USERFAULTFD */
 bool hugetlb_reserve_pages(struct inode *inode, long from, long to,
 						struct vm_area_struct *vma,
@@ -397,9 +396,8 @@ static inline int hugetlb_mfill_atomic_pte(pte_t *dst_pte,
 					   struct vm_area_struct *dst_vma,
 					   unsigned long dst_addr,
 					   unsigned long src_addr,
-					   enum mcopy_atomic_mode mode,
-					   struct page **pagep,
-					   bool wp_copy)
+					   uffd_flags_t flags,
+					   struct page **pagep)
 {
 	BUG();
 	return 0;
include/linux/shmem_fs.h +3 -2
index b82916c25e61..b7048bd88a8d 100644
@@ -9,6 +9,7 @@
 #include <linux/percpu_counter.h>
 #include <linux/xattr.h>
 #include <linux/fs_parser.h>
+#include <linux/userfaultfd_k.h>
 
 /* inode in-kernel data */
 
@@ -155,11 +156,11 @@ extern int shmem_mfill_atomic_pte(pmd_t *dst_pmd,
 				  struct vm_area_struct *dst_vma,
 				  unsigned long dst_addr,
 				  unsigned long src_addr,
-				  bool zeropage, bool wp_copy,
+				  uffd_flags_t flags,
 				  struct page **pagep);
 #else /* !CONFIG_SHMEM */
 #define shmem_mfill_atomic_pte(dst_pmd, dst_vma, dst_addr, \
-			       src_addr, zeropage, wp_copy, pagep) ({ BUG(); 0; })
+			       src_addr, flags, pagep) ({ BUG(); 0; })
 #endif /* CONFIG_SHMEM */
 #endif /* CONFIG_USERFAULTFD */
 
include/linux/userfaultfd_k.h +18 -16
index ba79e296fcc7..a45c1b42e500 100644
@@ -40,30 +40,32 @@ extern int sysctl_unprivileged_userfaultfd;
 
 extern vm_fault_t handle_userfault(struct vm_fault *vmf, unsigned long reason);
 
-/*
- * The mode of operation for __mcopy_atomic and its helpers.
- *
- * This is almost an implementation detail (mcopy_atomic below doesn't take this
- * as a parameter), but it's exposed here because memory-kind-specific
- * implementations (e.g. hugetlbfs) need to know the mode of operation.
- */
-enum mcopy_atomic_mode {
-	/* A normal copy_from_user into the destination range. */
-	MCOPY_ATOMIC_NORMAL,
-	/* Don't copy; map the destination range to the zero page. */
-	MCOPY_ATOMIC_ZEROPAGE,
-	/* Just install pte(s) with the existing page(s) in the page cache. */
-	MCOPY_ATOMIC_CONTINUE,
+/* A combined operation mode + behavior flags. */
+typedef unsigned int __bitwise uffd_flags_t;
+
+/* Mutually exclusive modes of operation. */
+enum mfill_atomic_mode {
+	MFILL_ATOMIC_COPY = (__force uffd_flags_t) 0,
+	MFILL_ATOMIC_ZEROPAGE = (__force uffd_flags_t) 1,
+	MFILL_ATOMIC_CONTINUE = (__force uffd_flags_t) 2,
+	NR_MFILL_ATOMIC_MODES,
 };
 
+#define MFILL_ATOMIC_MODE_BITS (const_ilog2(NR_MFILL_ATOMIC_MODES - 1) + 1)
+#define MFILL_ATOMIC_BIT(nr) ((__force uffd_flags_t) BIT(MFILL_ATOMIC_MODE_BITS + (nr)))
+#define MFILL_ATOMIC_MODE_MASK (MFILL_ATOMIC_BIT(0) - 1)
+
+/* Flags controlling behavior. */
+#define MFILL_ATOMIC_WP MFILL_ATOMIC_BIT(0)
+
 extern int mfill_atomic_install_pte(pmd_t *dst_pmd,
 				    struct vm_area_struct *dst_vma,
 				    unsigned long dst_addr, struct page *page,
-				    bool newly_allocated, bool wp_copy);
+				    bool newly_allocated, uffd_flags_t flags);
 
 extern ssize_t mfill_atomic_copy(struct mm_struct *dst_mm, unsigned long dst_start,
 				 unsigned long src_start, unsigned long len,
-				 atomic_t *mmap_changing, __u64 mode);
+				 atomic_t *mmap_changing, uffd_flags_t flags);
 extern ssize_t mfill_atomic_zeropage(struct mm_struct *dst_mm,
 				     unsigned long dst_start,
 				     unsigned long len,
mm/hugetlb.c +7 -6
index b4bda5f7f29f..1339f527b540 100644
@@ -6161,11 +6161,12 @@ int hugetlb_mfill_atomic_pte(pte_t *dst_pte,
 			     struct vm_area_struct *dst_vma,
 			     unsigned long dst_addr,
 			     unsigned long src_addr,
-			     enum mcopy_atomic_mode mode,
-			     struct page **pagep,
-			     bool wp_copy)
+			     uffd_flags_t flags,
+			     struct page **pagep)
 {
-	bool is_continue = (mode == MCOPY_ATOMIC_CONTINUE);
+	int mode = flags & MFILL_ATOMIC_MODE_MASK;
+	bool is_continue = (mode == MFILL_ATOMIC_CONTINUE);
+	bool wp_enabled = (flags & MFILL_ATOMIC_WP);
 	struct hstate *h = hstate_vma(dst_vma);
 	struct address_space *mapping = dst_vma->vm_file->f_mapping;
 	pgoff_t idx = vma_hugecache_offset(h, dst_vma, dst_addr);
@@ -6300,7 +6301,7 @@ int hugetlb_mfill_atomic_pte(pte_t *dst_pte,
 	 * For either: (1) CONTINUE on a non-shared VMA, or (2) UFFDIO_COPY
 	 * with wp flag set, don't set pte write bit.
 	 */
-	if (wp_copy || (is_continue && !vm_shared))
+	if (wp_enabled || (is_continue && !vm_shared))
 		writable = 0;
 	else
 		writable = dst_vma->vm_flags & VM_WRITE;
@@ -6315,7 +6316,7 @@ int hugetlb_mfill_atomic_pte(pte_t *dst_pte,
 	_dst_pte = huge_pte_mkdirty(_dst_pte);
 	_dst_pte = pte_mkyoung(_dst_pte);
 
-	if (wp_copy)
+	if (wp_enabled)
 		_dst_pte = huge_pte_mkuffd_wp(_dst_pte);
 
 	set_huge_pte_at(dst_vma->vm_mm, dst_addr, dst_pte, _dst_pte);
mm/shmem.c +3 -4
index 1d751b6cf1ac..0258054a0270 100644
@@ -76,7 +76,6 @@ static struct vfsmount *shm_mnt;
 #include <linux/syscalls.h>
 #include <linux/fcntl.h>
 #include <uapi/linux/memfd.h>
-#include <linux/userfaultfd_k.h>
 #include <linux/rmap.h>
 #include <linux/uuid.h>
 
@@ -2419,7 +2418,7 @@ int shmem_mfill_atomic_pte(pmd_t *dst_pmd,
 			   struct vm_area_struct *dst_vma,
 			   unsigned long dst_addr,
 			   unsigned long src_addr,
-			   bool zeropage, bool wp_copy,
+			   uffd_flags_t flags,
 			   struct page **pagep)
 {
 	struct inode *inode = file_inode(dst_vma->vm_file);
@@ -2451,7 +2450,7 @@ int shmem_mfill_atomic_pte(pmd_t *dst_pmd,
 		if (!folio)
 			goto out_unacct_blocks;
 
-		if (!zeropage) {	/* COPY */
+		if ((flags & MFILL_ATOMIC_MODE_MASK) == MFILL_ATOMIC_COPY) {
 			page_kaddr = kmap_local_folio(folio, 0);
 			/*
 			 * The read mmap_lock is held here.  Despite the
@@ -2510,7 +2509,7 @@ int shmem_mfill_atomic_pte(pmd_t *dst_pmd,
 		goto out_release;
 
 	ret = mfill_atomic_install_pte(dst_pmd, dst_vma, dst_addr,
-				       &folio->page, true, wp_copy);
+				       &folio->page, true, flags);
 	if (ret)
 		goto out_delete_from_cache;
 
mm/userfaultfd.c +35 -41

218 lines not on this page · raw message

Signature · 1 line
2.40.0.rc0.216.gc4246ad0f0-goog
reviewer
#8 lore raw

Re: [PATCH v3 3/5] mm: userfaultfd: combine 'mode' and 'wp_copy' arguments

Quoted 9 lines · Axel Rasmussen

Many userfaultfd ioctl functions take both a 'mode' and a 'wp_copy' argument. In future commits we plan to plumb the flags through to more places, so we'd be proliferating the very long argument list even further.

Let's take the time to simplify the argument list. Combine the two arguments into one - and generalize, so when we add more flags in the future, it doesn't imply more function arguments.

Quoted 7 lines

Since the modes (copy, zeropage, continue) are mutually exclusive, store them as an integer value (0, 1, 2) in the low bits. Place combine-able flag bits in the high bits.

This is quite similar to an earlier patch proposed by Nadav Amit ("userfaultfd: introduce uffd_flags" - for some reason Lore no longer has a copy of the patch). The main difference is that patch only handled

Lore has. :)

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

And btw sorry to review late.

Quoted 5 lines

flags, whereas this patch *also* combines the "mode" argument into the same type to shorten the argument list.

Mostly good to me, a few nitpicks below.

[...]

include/linux/userfaultfd_k.h extern int sysctl_unprivileged_userfaultfd; · line 52

Quoted 10 lines
include/linux/userfaultfd_k.h +9 -0
+/* A combined operation mode + behavior flags. */
+typedef unsigned int __bitwise uffd_flags_t;
+
+/* Mutually exclusive modes of operation. */
+enum mfill_atomic_mode {
+	MFILL_ATOMIC_COPY = (__force uffd_flags_t) 0,
+	MFILL_ATOMIC_ZEROPAGE = (__force uffd_flags_t) 1,
+	MFILL_ATOMIC_CONTINUE = (__force uffd_flags_t) 2,
+	NR_MFILL_ATOMIC_MODES,
 };

I never used enum like this. I had a feeling that this will enforce setting the enum entries but would the enforce applied to later assignments? I'm not sure.

I had a quick test and actually I found sparse already complains about calculating the last enum entry:

---8<---

$ cat a.c
typedef unsigned int __attribute__((bitwise)) flags_t;
enum {
    FLAG1 = (__attribute__((force)) flags_t) 0,
    FLAG_NUM,
};

void main(void)

{
    uffd_flags_t flags = FLAG1;
}
$ sparse a.c
a.c:5:5: error: can't increment the last enum member
---8<---

Maybe just use the simple "#define"s?

Quoted 2 lines

+#define MFILL_ATOMIC_MODE_BITS (const_ilog2(NR_MFILL_ATOMIC_MODES - 1) + 1)

Here IIUC it should be "const_ilog2(NR_MFILL_ATOMIC_MODES) + 1", but maybe.. we don't bother and define every bit explicitly?

Quoted 5 lines
include/linux/userfaultfd_k.h +5 -0
+#define MFILL_ATOMIC_BIT(nr) ((__force uffd_flags_t) BIT(MFILL_ATOMIC_MODE_BITS + (nr)))
+#define MFILL_ATOMIC_MODE_MASK (MFILL_ATOMIC_BIT(0) - 1)
+
+/* Flags controlling behavior. */
+#define MFILL_ATOMIC_WP MFILL_ATOMIC_BIT(0)

[...]

Quoted 13 lines
mm/userfaultfd.c +2 -2
@@ -312,9 +312,9 @@ static __always_inline ssize_t mfill_atomic_hugetlb(
 					      unsigned long dst_start,
 					      unsigned long src_start,
 					      unsigned long len,
-					      enum mcopy_atomic_mode mode,
-					      bool wp_copy)
+					      uffd_flags_t flags)
 {
+	int mode = flags & MFILL_ATOMIC_MODE_MASK;
 	struct mm_struct *dst_mm = dst_vma->vm_mm;
 	int vm_shared = dst_vma->vm_flags & VM_SHARED;
 	ssize_t err;

mm/userfaultfd.c static __always_inline ssize_t mfill_atomic_hugetlb( · line 336

Quoted 5 lines
mm/userfaultfd.c +1 -1
@@ -333,7 +333,7 @@ static __always_inline ssize_t mfill_atomic_hugetlb(
 	 * by THP.  Since we can not reliably insert a zero page, this
 	 * feature is not supported.
 	 */
-	if (mode == MCOPY_ATOMIC_ZEROPAGE) {
+	if (mode == MFILL_ATOMIC_ZEROPAGE) {

The mode comes from "& MFILL_ATOMIC_MODE_MASK" but it doesn't quickly tell whether there's a shift for the mask.

Would it look better we just have a helper to fetch the mode? The function tells that whatever it returns must be the mode:

       if (uffd_flags_get_mode(flags) == MFILL_ATOMIC_ZEROPAGE)

We also avoid quite a few "mode" variables. All the rest bits will be fine to use "flags & FLAG1" if it's a boolean (so only this "mode" is slightly tricky).

What do you think?

Thanks,

Signature · 1 line
Peter Xu
#9 lore raw

Re: [PATCH v3 3/5] mm: userfaultfd: combine 'mode' and 'wp_copy' arguments

Quoted 63 lines · Peter Xu
Quoted 9 lines · Axel Rasmussen

Many userfaultfd ioctl functions take both a 'mode' and a 'wp_copy' argument. In future commits we plan to plumb the flags through to more places, so we'd be proliferating the very long argument list even further.

Let's take the time to simplify the argument list. Combine the two arguments into one - and generalize, so when we add more flags in the future, it doesn't imply more function arguments.

Quoted 7 lines

Since the modes (copy, zeropage, continue) are mutually exclusive, store them as an integer value (0, 1, 2) in the low bits. Place combine-able flag bits in the high bits.

This is quite similar to an earlier patch proposed by Nadav Amit ("userfaultfd: introduce uffd_flags" - for some reason Lore no longer has a copy of the patch). The main difference is that patch only handled

Lore has. :)

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

And btw sorry to review late.

Quoted 5 lines

flags, whereas this patch *also* combines the "mode" argument into the same type to shorten the argument list.

Mostly good to me, a few nitpicks below.

[...]

include/linux/userfaultfd_k.h extern int sysctl_unprivileged_userfaultfd; · line 52

Quoted 10 lines
include/linux/userfaultfd_k.h +9 -0
+/* A combined operation mode + behavior flags. */
+typedef unsigned int __bitwise uffd_flags_t;
+
+/* Mutually exclusive modes of operation. */
+enum mfill_atomic_mode {
+     MFILL_ATOMIC_COPY = (__force uffd_flags_t) 0,
+     MFILL_ATOMIC_ZEROPAGE = (__force uffd_flags_t) 1,
+     MFILL_ATOMIC_CONTINUE = (__force uffd_flags_t) 2,
+     NR_MFILL_ATOMIC_MODES,
 };

I never used enum like this. I had a feeling that this will enforce setting the enum entries but would the enforce applied to later assignments? I'm not sure.

I had a quick test and actually I found sparse already complains about calculating the last enum entry:

---8<---

$ cat a.c
typedef unsigned int __attribute__((bitwise)) flags_t;
enum {
    FLAG1 = (__attribute__((force)) flags_t) 0,
    FLAG_NUM,
};

void main(void)

Quoted 8 lines
{
    uffd_flags_t flags = FLAG1;
}
$ sparse a.c
a.c:5:5: error: can't increment the last enum member
---8<---

Maybe just use the simple "#define"s?

Agreed, if sparse isn't happy with this then using the force macros is pointless.

The enum is valuable because it lets us get the # of modes; assuming we agree that's useful below ...

Quoted 6 lines
Quoted 3 lines

+#define MFILL_ATOMIC_MODE_BITS (const_ilog2(NR_MFILL_ATOMIC_MODES - 1) + 1)

Here IIUC it should be "const_ilog2(NR_MFILL_ATOMIC_MODES) + 1", but maybe.. we don't bother and define every bit explicitly?

If my reading of const_ilog2's definition is correct, then:

const_ilog2(4) = 2
const_ilog2(3) = 1
const_ilog2(2) = 1

For either 3 or 4 modes, we need 2 bits to represent them (0, 1, 2, 3), i.e. we want MFILL_ATOMIC_MODE_BITS = 2. I think this is correct as is, because const_ilog2(4 - 1) + 1 = 2, and const_ilog2(3 - 1) + 1 = 2.

In other words, I think const_ilog2 is defined as floor(log2()), whereas what we want is ceil(log2()).

The benefit of doing this vs. just doing defines with fixed values is, if we ever added a new mode, we wouldn't have to do bit twiddling and update the mask, flag bits, etc. - it would happen "automatically". I prefer it this way, but I agree it is a matter of opinion / taste. :) If you or others feel strongly this is overcomplicated, I can take the other approach.

Quoted 31 lines
Quoted 6 lines
include/linux/userfaultfd_k.h +5 -0
+#define MFILL_ATOMIC_BIT(nr) ((__force uffd_flags_t) BIT(MFILL_ATOMIC_MODE_BITS + (nr)))
+#define MFILL_ATOMIC_MODE_MASK (MFILL_ATOMIC_BIT(0) - 1)
+
+/* Flags controlling behavior. */
+#define MFILL_ATOMIC_WP MFILL_ATOMIC_BIT(0)

[...]

Quoted 13 lines
mm/userfaultfd.c +2 -2
@@ -312,9 +312,9 @@ static __always_inline ssize_t mfill_atomic_hugetlb(
                                            unsigned long dst_start,
                                            unsigned long src_start,
                                            unsigned long len,
-                                           enum mcopy_atomic_mode mode,
-                                           bool wp_copy)
+                                           uffd_flags_t flags)
 {
+     int mode = flags & MFILL_ATOMIC_MODE_MASK;
      struct mm_struct *dst_mm = dst_vma->vm_mm;
      int vm_shared = dst_vma->vm_flags & VM_SHARED;
      ssize_t err;

mm/userfaultfd.c static __always_inline ssize_t mfill_atomic_hugetlb( · line 336

Quoted 5 lines
mm/userfaultfd.c +1 -1
@@ -333,7 +333,7 @@ static __always_inline ssize_t mfill_atomic_hugetlb(
       * by THP.  Since we can not reliably insert a zero page, this
       * feature is not supported.
       */
-     if (mode == MCOPY_ATOMIC_ZEROPAGE) {
+     if (mode == MFILL_ATOMIC_ZEROPAGE) {

The mode comes from "& MFILL_ATOMIC_MODE_MASK" but it doesn't quickly tell whether there's a shift for the mask.

Quoted 8 lines

Would it look better we just have a helper to fetch the mode? The function tells that whatever it returns must be the mode:

       if (uffd_flags_get_mode(flags) == MFILL_ATOMIC_ZEROPAGE)

We also avoid quite a few "mode" variables. All the rest bits will be fine to use "flags & FLAG1" if it's a boolean (so only this "mode" is slightly tricky).

Agreed, this is simpler. I'll make this change.

Quoted 8 lines

What do you think?

Thanks,

--
Peter Xu

reviewer
#10 lore raw

Re: [PATCH v3 3/5] mm: userfaultfd: combine 'mode' and 'wp_copy' arguments

Quoted 13 lines · Axel Rasmussen
Quoted 6 lines
Quoted 3 lines

+#define MFILL_ATOMIC_MODE_BITS (const_ilog2(NR_MFILL_ATOMIC_MODES - 1) + 1)

Here IIUC it should be "const_ilog2(NR_MFILL_ATOMIC_MODES) + 1", but maybe.. we don't bother and define every bit explicitly?

If my reading of const_ilog2's definition is correct, then:

const_ilog2(4) = 2
const_ilog2(3) = 1
const_ilog2(2) = 1

Quoted 7 lines

For either 3 or 4 modes, we need 2 bits to represent them (0, 1, 2, 3), i.e. we want MFILL_ATOMIC_MODE_BITS = 2. I think this is correct as is, because const_ilog2(4 - 1) + 1 = 2, and const_ilog2(3 - 1) + 1 = 2.

In other words, I think const_ilog2 is defined as floor(log2()), whereas what we want is ceil(log2()).

You're right.

Quoted 7 lines

The benefit of doing this vs. just doing defines with fixed values is, if we ever added a new mode, we wouldn't have to do bit twiddling and update the mask, flag bits, etc. - it would happen "automatically". I prefer it this way, but I agree it is a matter of opinion / taste. :) If you or others feel strongly this is overcomplicated, I can take the other approach.

I don't know what this will look like at last. The thing is if you plan to define MFILL_ATOMIC_* with __bitwise I think it'll stop working with any calculations upon it.

I don't worry on growing modes, as I don't expect it to happen a lot.

No strong opinion here, as long as sparse won't complain.

Thanks,

Signature · 1 line
Peter Xu
#11 lore raw

Re: [PATCH v3 3/5] mm: userfaultfd: combine 'mode' and 'wp_copy' arguments

In reply to Axel Rasmussen

Excluding Peter’s comments, LGTM.

include/linux/hugetlb.h static int mfill_atomic_pte_copy(pmd_t *dst_pmd, · line 135

Quoted 10 lines · Axel Rasmussen
include/linux/hugetlb.h +2 -2
@@ -131,8 +131,8 @@ static int mfill_atomic_pte_copy(pmd_t *dst_pmd,
 				 struct vm_area_struct *dst_vma,
 				 unsigned long dst_addr,
 				 unsigned long src_addr,
-				 struct page **pagep,
-				 bool wp_copy)
+				 uffd_flags_t flags,
+				 struct page **pagep)

Yet, it would be nice if we can be consistent on whether pagep precedes flags or not (it’s the other way around in shmem_mfill_atomic_pte()).

#12 lore raw

PATCH v3 4/5 mm: userfaultfd: don't separate addr + len arguments

In reply to Axel Rasmussen

We have a lot of functions which take an address + length pair, currently passed as separate arguments. However, in our userspace API we already have struct uffdio_range, which is exactly this pair, and this is what we get from userspace when ioctls are called.

Instead of splitting the struct up into two separate arguments, just plumb the struct through to the functions which use it (once we get to the mfill_atomic_pte level, we're dealing with single (huge)pages, so we don't need both parts).

Relatedly, for waking, just re-use this existing structure instead of defining a new "struct uffdio_wake_range".

---
 fs/userfaultfd.c              | 107 +++++++++++++---------------------
 include/linux/userfaultfd_k.h |  17 +++---
 mm/userfaultfd.c              |  92 ++++++++++++++---------------
 3 files changed, 96 insertions(+), 120 deletions(-)
fs/userfaultfd.c +42 -65
index b8e328123b71..984b63b0fc75 100644
@@ -95,11 +95,6 @@ struct userfaultfd_wait_queue {
 	bool waken;
 };
 
-struct userfaultfd_wake_range {
-	unsigned long start;
-	unsigned long len;
-};
-
 /* internal indication that UFFD_API ioctl was successfully executed */
 #define UFFD_FEATURE_INITIALIZED		(1u << 31)
 
@@ -126,7 +121,7 @@ static void userfaultfd_set_vm_flags(struct vm_area_struct *vma,
 static int userfaultfd_wake_function(wait_queue_entry_t *wq, unsigned mode,
 				     int wake_flags, void *key)
 {
-	struct userfaultfd_wake_range *range = key;
+	struct uffdio_range *range = key;
 	int ret;
 	struct userfaultfd_wait_queue *uwq;
 	unsigned long start, len;
@@ -881,7 +876,7 @@ static int userfaultfd_release(struct inode *inode, struct file *file)
 	struct mm_struct *mm = ctx->mm;
 	struct vm_area_struct *vma, *prev;
 	/* len == 0 means wake all */
-	struct userfaultfd_wake_range range = { .len = 0, };
+	struct uffdio_range range = {0};
 	unsigned long new_flags;
 	VMA_ITERATOR(vmi, mm, 0);
 
@@ -1226,7 +1221,7 @@ static ssize_t userfaultfd_read(struct file *file, char __user *buf,
 }
 
 static void __wake_userfault(struct userfaultfd_ctx *ctx,
-			     struct userfaultfd_wake_range *range)
+			     struct uffdio_range *range)
 {
 	spin_lock_irq(&ctx->fault_pending_wqh.lock);
 	/* wake all in the range and autoremove */
@@ -1239,7 +1234,7 @@ static void __wake_userfault(struct userfaultfd_ctx *ctx,
 }
 
 static __always_inline void wake_userfault(struct userfaultfd_ctx *ctx,
-					   struct userfaultfd_wake_range *range)
+					   struct uffdio_range *range)
 {
 	unsigned seq;
 	bool need_wakeup;
@@ -1270,21 +1265,21 @@ static __always_inline void wake_userfault(struct userfaultfd_ctx *ctx,
 }
 
 static __always_inline int validate_range(struct mm_struct *mm,
-					  __u64 start, __u64 len)
+					  const struct uffdio_range *range)
 {
 	__u64 task_size = mm->task_size;
 
-	if (start & ~PAGE_MASK)
+	if (range->start & ~PAGE_MASK)
 		return -EINVAL;
-	if (len & ~PAGE_MASK)
+	if (range->len & ~PAGE_MASK)
 		return -EINVAL;
-	if (!len)
+	if (!range->len)
 		return -EINVAL;
-	if (start < mmap_min_addr)
+	if (range->start < mmap_min_addr)
 		return -EINVAL;
-	if (start >= task_size)
+	if (range->start >= task_size)
 		return -EINVAL;
-	if (len > task_size - start)
+	if (range->len > task_size - range->start)
 		return -EINVAL;
 	return 0;
 }
@@ -1331,8 +1326,7 @@ static int userfaultfd_register(struct userfaultfd_ctx *ctx,
 		vm_flags |= VM_UFFD_MINOR;
 	}
 
-	ret = validate_range(mm, uffdio_register.range.start,
-			     uffdio_register.range.len);
+	ret = validate_range(mm, &uffdio_register.range);
 	if (ret)
 		goto out;
 
@@ -1538,11 +1532,11 @@ static int userfaultfd_unregister(struct userfaultfd_ctx *ctx,
 	if (copy_from_user(&uffdio_unregister, buf, sizeof(uffdio_unregister)))
 		goto out;
 
-	ret = validate_range(mm, uffdio_unregister.start,
-			     uffdio_unregister.len);
+	ret = validate_range(mm, &uffdio_unregister);
 	if (ret)
 		goto out;
 
+	/* Get rid of start + end in favor of range *? */
 	start = uffdio_unregister.start;
 	end = start + uffdio_unregister.len;
 
@@ -1597,6 +1591,7 @@ static int userfaultfd_unregister(struct userfaultfd_ctx *ctx,
 	prev = vma_prev(&vmi);
 	ret = 0;
 	for_each_vma_range(vmi, vma, end) {
+		struct uffdio_range range;
 		cond_resched();
 
 		BUG_ON(!vma_can_userfault(vma, vma->vm_flags));
@@ -1614,6 +1609,8 @@ static int userfaultfd_unregister(struct userfaultfd_ctx *ctx,
 			start = vma->vm_start;
 		vma_end = min(end, vma->vm_end);
 
+		range.start = start;
+		range.len = vma_end - start;
 		if (userfaultfd_missing(vma)) {
 			/*
 			 * Wake any concurrent pending userfault while
@@ -1621,15 +1618,12 @@ static int userfaultfd_unregister(struct userfaultfd_ctx *ctx,
 			 * permanently and it avoids userland to call
 			 * UFFDIO_WAKE explicitly.
 			 */
-			struct userfaultfd_wake_range range;
-			range.start = start;
-			range.len = vma_end - start;
 			wake_userfault(vma->vm_userfaultfd_ctx.ctx, &range);
 		}
 
 		/* Reset ptes for the whole vma range if wr-protected */
 		if (userfaultfd_wp(vma))
-			uffd_wp_range(vma, start, vma_end - start, false);
+			uffd_wp_range(vma, &range, false);
 
 		new_flags = vma->vm_flags & ~__VM_UFFD_FLAGS;
 		prev = vma_merge(&vmi, mm, prev, start, vma_end, new_flags,
@@ -1680,27 +1674,23 @@ static int userfaultfd_wake(struct userfaultfd_ctx *ctx,
 {
 	int ret;
 	struct uffdio_range uffdio_wake;
-	struct userfaultfd_wake_range range;
 	const void __user *buf = (void __user *)arg;
 
 	ret = -EFAULT;
 	if (copy_from_user(&uffdio_wake, buf, sizeof(uffdio_wake)))
 		goto out;
 
-	ret = validate_range(ctx->mm, uffdio_wake.start, uffdio_wake.len);
+	ret = validate_range(ctx->mm, &uffdio_wake);
 	if (ret)
 		goto out;
 
-	range.start = uffdio_wake.start;
-	range.len = uffdio_wake.len;
-
 	/*
 	 * len == 0 means wake all and we don't want to wake all here,
 	 * so check it again to be sure.
 	 */
-	VM_BUG_ON(!range.len);
+	VM_BUG_ON(!uffdio_wake.len);
 
-	wake_userfault(ctx, &range);
+	wake_userfault(ctx, &uffdio_wake);
 	ret = 0;
 
 out:
@@ -1713,7 +1703,7 @@ static int userfaultfd_copy(struct userfaultfd_ctx *ctx,
 	__s64 ret;
 	struct uffdio_copy uffdio_copy;
 	struct uffdio_copy __user *user_uffdio_copy;
-	struct userfaultfd_wake_range range;
+	struct uffdio_range range;
 	int flags = 0;
 
 	user_uffdio_copy = (struct uffdio_copy __user *) arg;
@@ -1728,7 +1718,9 @@ static int userfaultfd_copy(struct userfaultfd_ctx *ctx,
 			   sizeof(uffdio_copy)-sizeof(__s64)))
 		goto out;
 
-	ret = validate_range(ctx->mm, uffdio_copy.dst, uffdio_copy.len);
+	range.start = uffdio_copy.dst;
+	range.len = uffdio_copy.len;
+	ret = validate_range(ctx->mm, &range);
 	if (ret)
 		goto out;
 	/*
@@ -1744,9 +1736,8 @@ static int userfaultfd_copy(struct userfaultfd_ctx *ctx,
 	if (uffdio_copy.mode & UFFDIO_COPY_MODE_WP)
 		flags |= MFILL_ATOMIC_WP;
 	if (mmget_not_zero(ctx->mm)) {
-		ret = mfill_atomic_copy(ctx->mm, uffdio_copy.dst, uffdio_copy.src,
-					uffdio_copy.len, &ctx->mmap_changing,
-					flags);
+		ret = mfill_atomic_copy(ctx->mm, uffdio_copy.src, &range,
+					&ctx->mmap_changing, flags);
 		mmput(ctx->mm);
 	} else {
 		return -ESRCH;
@@ -1758,10 +1749,8 @@ static int userfaultfd_copy(struct userfaultfd_ctx *ctx,
 	BUG_ON(!ret);
 	/* len == 0 would wake all */
 	range.len = ret;
-	if (!(uffdio_copy.mode & UFFDIO_COPY_MODE_DONTWAKE)) {
-		range.start = uffdio_copy.dst;
+	if (!(uffdio_copy.mode & UFFDIO_COPY_MODE_DONTWAKE))
 		wake_userfault(ctx, &range);
-	}
 	ret = range.len == uffdio_copy.len ? 0 : -EAGAIN;
 out:
 	return ret;
@@ -1773,7 +1762,7 @@ static int userfaultfd_zeropage(struct userfaultfd_ctx *ctx,
 	__s64 ret;
 	struct uffdio_zeropage uffdio_zeropage;
 	struct uffdio_zeropage __user *user_uffdio_zeropage;
-	struct userfaultfd_wake_range range;
+	struct uffdio_range range;
 
 	user_uffdio_zeropage = (struct uffdio_zeropage __user *) arg;
 
@@ -1787,8 +1776,8 @@ static int userfaultfd_zeropage(struct userfaultfd_ctx *ctx,
 			   sizeof(uffdio_zeropage)-sizeof(__s64)))
 		goto out;
 
-	ret = validate_range(ctx->mm, uffdio_zeropage.range.start,
-			     uffdio_zeropage.range.len);
+	range = uffdio_zeropage.range;
+	ret = validate_range(ctx->mm, &range);
 	if (ret)
 		goto out;
 	ret = -EINVAL;
@@ -1796,8 +1785,7 @@ static int userfaultfd_zeropage(struct userfaultfd_ctx *ctx,
 		goto out;
 
 	if (mmget_not_zero(ctx->mm)) {
-		ret = mfill_atomic_zeropage(ctx->mm, uffdio_zeropage.range.start,
-					   uffdio_zeropage.range.len,
+		ret = mfill_atomic_zeropage(ctx->mm, &uffdio_zeropage.range,
 					   &ctx->mmap_changing);
 		mmput(ctx->mm);
 	} else {
@@ -1811,7 +1799,6 @@ static int userfaultfd_zeropage(struct userfaultfd_ctx *ctx,
 	BUG_ON(!ret);
 	range.len = ret;
 	if (!(uffdio_zeropage.mode & UFFDIO_ZEROPAGE_MODE_DONTWAKE)) {
-		range.start = uffdio_zeropage.range.start;
 		wake_userfault(ctx, &range);
 	}
 	ret = range.len == uffdio_zeropage.range.len ? 0 : -EAGAIN;
@@ -1825,7 +1812,6 @@ static int userfaultfd_writeprotect(struct userfaultfd_ctx *ctx,
 	int ret;
 	struct uffdio_writeprotect uffdio_wp;
 	struct uffdio_writeprotect __user *user_uffdio_wp;
-	struct userfaultfd_wake_range range;
 	bool mode_wp, mode_dontwake;
 
 	if (atomic_read(&ctx->mmap_changing))
@@ -1837,8 +1823,7 @@ static int userfaultfd_writeprotect(struct userfaultfd_ctx *ctx,
 			   sizeof(struct uffdio_writeprotect)))
 		return -EFAULT;
 
-	ret = validate_range(ctx->mm, uffdio_wp.range.start,
-			     uffdio_wp.range.len);
+	ret = validate_range(ctx->mm, &uffdio_wp.range);
 	if (ret)
 		return ret;
 
@@ -1853,9 +1838,8 @@ static int userfaultfd_writeprotect(struct userfaultfd_ctx *ctx,
 		return -EINVAL;
 
 	if (mmget_not_zero(ctx->mm)) {
-		ret = mwriteprotect_range(ctx->mm, uffdio_wp.range.start,
-					  uffdio_wp.range.len, mode_wp,
-					  &ctx->mmap_changing);
+		ret = mwriteprotect_range(ctx->mm, &uffdio_wp.range,
+					  mode_wp, &ctx->mmap_changing);
 		mmput(ctx->mm);
 	} else {
 		return -ESRCH;
@@ -1864,11 +1848,8 @@ static int userfaultfd_writeprotect(struct userfaultfd_ctx *ctx,
 	if (ret)
 		return ret;
 
-	if (!mode_wp && !mode_dontwake) {
-		range.start = uffdio_wp.range.start;
-		range.len = uffdio_wp.range.len;
-		wake_userfault(ctx, &range);
-	}
+	if (!mode_wp && !mode_dontwake)
+		wake_userfault(ctx, &uffdio_wp.range);
 	return ret;
 }
 
@@ -1877,7 +1858,7 @@ static int userfaultfd_continue(struct userfaultfd_ctx *ctx, unsigned long arg)
 	__s64 ret;
 	struct uffdio_continue uffdio_continue;
 	struct uffdio_continue __user *user_uffdio_continue;
-	struct userfaultfd_wake_range range;
+	struct uffdio_range range;
 
 	user_uffdio_continue = (struct uffdio_continue __user *)arg;
 
@@ -1891,23 +1872,20 @@ static int userfaultfd_continue(struct userfaultfd_ctx *ctx, unsigned long arg)
 			   sizeof(uffdio_continue) - (sizeof(__s64))))
 		goto out;
 
-	ret = validate_range(ctx->mm, uffdio_continue.range.start,
-			     uffdio_continue.range.len);
+	range = uffdio_continue.range;
+	ret = validate_range(ctx->mm, &range);
 	if (ret)
 		goto out;
 
 	ret = -EINVAL;
 	/* double check for wraparound just in case. */
-	if (uffdio_continue.range.start + uffdio_continue.range.len <=
-	    uffdio_continue.range.start) {
+	if (range.start + range.len <= range.start)
 		goto out;
-	}
 	if (uffdio_continue.mode & ~UFFDIO_CONTINUE_MODE_DONTWAKE)
 		goto out;
 
 	if (mmget_not_zero(ctx->mm)) {
-		ret = mfill_atomic_continue(ctx->mm, uffdio_continue.range.start,
-					    uffdio_continue.range.len,
+		ret = mfill_atomic_continue(ctx->mm, &range,
 					    &ctx->mmap_changing);
 		mmput(ctx->mm);
 	} else {
@@ -1923,7 +1901,6 @@ static int userfaultfd_continue(struct userfaultfd_ctx *ctx, unsigned long arg)
 	BUG_ON(!ret);
 	range.len = ret;
 	if (!(uffdio_continue.mode & UFFDIO_CONTINUE_MODE_DONTWAKE)) {
-		range.start = uffdio_continue.range.start;
 		wake_userfault(ctx, &range);
 	}
 	ret = range.len == uffdio_continue.range.len ? 0 : -EAGAIN;
include/linux/userfaultfd_k.h +9 -8
index a45c1b42e500..fcd95e3d3dcd 100644
@@ -63,20 +63,21 @@ extern int mfill_atomic_install_pte(pmd_t *dst_pmd,
 				    unsigned long dst_addr, struct page *page,
 				    bool newly_allocated, uffd_flags_t flags);
 
-extern ssize_t mfill_atomic_copy(struct mm_struct *dst_mm, unsigned long dst_start,
-				 unsigned long src_start, unsigned long len,
+extern ssize_t mfill_atomic_copy(struct mm_struct *dst_mm, unsigned long src_start,
+				 const struct uffdio_range *dst,
 				 atomic_t *mmap_changing, uffd_flags_t flags);
 extern ssize_t mfill_atomic_zeropage(struct mm_struct *dst_mm,
-				     unsigned long dst_start,
-				     unsigned long len,
+				     const struct uffdio_range *dst,
 				     atomic_t *mmap_changing);
-extern ssize_t mfill_atomic_continue(struct mm_struct *dst_mm, unsigned long dst_start,
-				     unsigned long len, atomic_t *mmap_changing);
+extern ssize_t mfill_atomic_continue(struct mm_struct *dst_mm,
+				     const struct uffdio_range *dst,
+				    atomic_t *mmap_changing);
 extern int mwriteprotect_range(struct mm_struct *dst_mm,
-			       unsigned long start, unsigned long len,
+			       const struct uffdio_range *range,
 			       bool enable_wp, atomic_t *mmap_changing);
 extern long uffd_wp_range(struct vm_area_struct *vma,
-			  unsigned long start, unsigned long len, bool enable_wp);
+			  const struct uffdio_range *range,
+			  bool enable_wp);
 
 /* mm helpers */
 static inline bool is_mergeable_vm_userfaultfd_ctx(struct vm_area_struct *vma,
mm/userfaultfd.c +45 -47

236 lines not on this page · raw message

Signature · 1 line
2.40.0.rc0.216.gc4246ad0f0-goog
reviewer
#13 lore raw

Re: [PATCH v3 4/5] mm: userfaultfd: don't separate addr + len arguments

Quoted 25 lines · Axel Rasmussen

We have a lot of functions which take an address + length pair, currently passed as separate arguments. However, in our userspace API we already have struct uffdio_range, which is exactly this pair, and this is what we get from userspace when ioctls are called.

Instead of splitting the struct up into two separate arguments, just plumb the struct through to the functions which use it (once we get to the mfill_atomic_pte level, we're dealing with single (huge)pages, so we don't need both parts).

Relatedly, for waking, just re-use this existing structure instead of defining a new "struct uffdio_wake_range".

---
 fs/userfaultfd.c              | 107 +++++++++++++---------------------
 include/linux/userfaultfd_k.h |  17 +++---
 mm/userfaultfd.c              |  92 ++++++++++++++---------------
 3 files changed, 96 insertions(+), 120 deletions(-)
fs/userfaultfd.c +0 -0
index b8e328123b71..984b63b0fc75 100644

fs/userfaultfd.c struct userfaultfd_wait_queue { · line 101

Quoted 7 lines
fs/userfaultfd.c +0 -4
@@ -95,11 +95,6 @@ struct userfaultfd_wait_queue {
 	bool waken;
 };
 
-struct userfaultfd_wake_range {
-	unsigned long start;
-	unsigned long len;
-};

Would there still be a difference on e.g. 32 bits systems?

[...]

Quoted 16 lines
fs/userfaultfd.c +4 -5
 static __always_inline int validate_range(struct mm_struct *mm,
-					  __u64 start, __u64 len)
+					  const struct uffdio_range *range)
 {
 	__u64 task_size = mm->task_size;
 
-	if (start & ~PAGE_MASK)
+	if (range->start & ~PAGE_MASK)
 		return -EINVAL;
-	if (len & ~PAGE_MASK)
+	if (range->len & ~PAGE_MASK)
 		return -EINVAL;
-	if (!len)
+	if (!range->len)
 		return -EINVAL;
-	if (start < mmap_min_addr)

fs/userfaultfd.c static __always_inline void wake_userfault(struct userfaultfd_ctx *ctx, · line 1285

Quoted 10 lines
fs/userfaultfd.c +3 -2
+	if (range->start < mmap_min_addr)
 		return -EINVAL;
-	if (start >= task_size)
+	if (range->start >= task_size)
 		return -EINVAL;
-	if (len > task_size - start)
+	if (range->len > task_size - range->start)
 		return -EINVAL;
 	return 0;
 }

Personally I don't like a lot on such a change. :( It avoids one parameter being passed over but it can add a lot indirections.

Do you strongly suggest this? Shall we move on without this so to not block the last patch (which I assume is the one you're looking for)?

Thanks,

Signature · 1 line
Peter Xu
#14 lore raw

Re: [PATCH v3 4/5] mm: userfaultfd: don't separate addr + len arguments

Quoted 71 lines · Peter Xu

!! External Email

Quoted 25 lines · Axel Rasmussen

We have a lot of functions which take an address + length pair, currently passed as separate arguments. However, in our userspace API we already have struct uffdio_range, which is exactly this pair, and this is what we get from userspace when ioctls are called.

Instead of splitting the struct up into two separate arguments, just plumb the struct through to the functions which use it (once we get to the mfill_atomic_pte level, we're dealing with single (huge)pages, so we don't need both parts).

Relatedly, for waking, just re-use this existing structure instead of defining a new "struct uffdio_wake_range".

---

fs/userfaultfd.c              | 107 +++++++++++++---------------------
include/linux/userfaultfd_k.h |  17 +++---
mm/userfaultfd.c              |  92 ++++++++++++++---------------
3 files changed, 96 insertions(+), 120 deletions(-)
fs/userfaultfd.c +0 -0
index b8e328123b71..984b63b0fc75 100644

fs/userfaultfd.c struct userfaultfd_wait_queue { · line 101

Quoted 7 lines
fs/userfaultfd.c +0 -4
@@ -95,11 +95,6 @@ struct userfaultfd_wait_queue {
     bool waken;
 };
 
-struct userfaultfd_wake_range {
-     unsigned long start;
-     unsigned long len;
-};

Would there still be a difference on e.g. 32 bits systems?

[...]

Quoted 25 lines
fs/userfaultfd.c +7 -7
 static __always_inline int validate_range(struct mm_struct *mm,
-                                       __u64 start, __u64 len)
+                                       const struct uffdio_range *range)
 {
     __u64 task_size = mm->task_size;
 
-     if (start & ~PAGE_MASK)
+     if (range->start & ~PAGE_MASK)
             return -EINVAL;
-     if (len & ~PAGE_MASK)
+     if (range->len & ~PAGE_MASK)
             return -EINVAL;
-     if (!len)
+     if (!range->len)
             return -EINVAL;
-     if (start < mmap_min_addr)
+     if (range->start < mmap_min_addr)
             return -EINVAL;
-     if (start >= task_size)
+     if (range->start >= task_size)
             return -EINVAL;
-     if (len > task_size - start)
+     if (range->len > task_size - range->start)
             return -EINVAL;
     return 0;
Quoted 1 line

}

Quoted 5 lines

Personally I don't like a lot on such a change. :( It avoids one parameter being passed over but it can add a lot indirections.

Do you strongly suggest this? Shall we move on without this so to not block the last patch (which I assume is the one you're looking for)?

Just in case you missed, it is __always_inline, so I presume that from a generated code point-of-view it is the same.

Having said that, small assignments to local start, let and range variables would make the code easier to read and reduce the change-set.

#15 lore raw

Re: [PATCH v3 4/5] mm: userfaultfd: don't separate addr + len arguments

Quoted 35 lines · Nadav Amit
Quoted 35 lines · Peter Xu

!! External Email

Quoted 28 lines · Axel Rasmussen

We have a lot of functions which take an address + length pair, currently passed as separate arguments. However, in our userspace API we already have struct uffdio_range, which is exactly this pair, and this is what we get from userspace when ioctls are called.

Instead of splitting the struct up into two separate arguments, just plumb the struct through to the functions which use it (once we get to the mfill_atomic_pte level, we're dealing with single (huge)pages, so we don't need both parts).

Relatedly, for waking, just re-use this existing structure instead of defining a new "struct uffdio_wake_range".

---

fs/userfaultfd.c              | 107 +++++++++++++---------------------
include/linux/userfaultfd_k.h |  17 +++---
mm/userfaultfd.c              |  92 ++++++++++++++---------------
3 files changed, 96 insertions(+), 120 deletions(-)
fs/userfaultfd.c +0 -0
index b8e328123b71..984b63b0fc75 100644
@@ -95,11 +95,6 @@ struct userfaultfd_wait_queue {
     bool waken;
 };
 
Quoted 7 lines
Quoted 6 lines

fs/userfaultfd.c struct userfaultfd_wait_queue { · line 101

Quoted 4 lines
fs/userfaultfd.c +0 -4
-struct userfaultfd_wake_range {
-     unsigned long start;
-     unsigned long len;
-};

Would there still be a difference on e.g. 32 bits systems?

My assumption is that __u64 is at least 64 bits wide on all platforms, and so it is sufficient. I believe the standard allows unsigned long to be 32-bits, so __u64 may be overkill on some platforms, but to me the cost is small enough I'd prefer to avoid defining a second almost-identical structure.

Then again though as I say below, I don't feel strongly about this refactor.

Quoted 31 lines
Quoted 6 lines

[...]

fs/userfaultfd.c

Quoted 3 lines

static __always_inline int validate_range(struct mm_struct *mm,

fs/userfaultfd.c +1 -1
-                                       __u64 start, __u64 len)
+                                       const struct uffdio_range

*range)

Quoted 24 lines
Quoted 23 lines
{
     __u64 task_size = mm->task_size;
fs/userfaultfd.c +6 -6
-     if (start & ~PAGE_MASK)
+     if (range->start & ~PAGE_MASK)
             return -EINVAL;
-     if (len & ~PAGE_MASK)
+     if (range->len & ~PAGE_MASK)
             return -EINVAL;
-     if (!len)
+     if (!range->len)
             return -EINVAL;
-     if (start < mmap_min_addr)
+     if (range->start < mmap_min_addr)
             return -EINVAL;
-     if (start >= task_size)
+     if (range->start >= task_size)
             return -EINVAL;
-     if (len > task_size - start)
+     if (range->len > task_size - range->start)
             return -EINVAL;
     return 0;

}

Quoted 7 lines
Quoted 1 line

Personally I don't like a lot on such a change. :( It avoids one

parameter

Quoted 4 lines

being passed over but it can add a lot indirections.

Do you strongly suggest this? Shall we move on without this so to not block the last patch (which I assume is the one you're looking for)?

I don't feel strongly, I'm fine with dropping this patch. I'll make that change in a v4 (I think there will be some conflicts to resolve in the patches after this one, so I'll post a new version to avoid troubling anyone).

Quoted 8 lines

Just in case you missed, it is __always_inline, so I presume that from a generated code point-of-view it is the same.

Having said that, small assignments to local start, let and range variables would make the code easier to read and reduce the change-set.

#16 lore raw

Re: [PATCH v3 4/5] mm: userfaultfd: don't separate addr + len arguments

In reply to Axel Rasmussen
Thank you for the patch! Yet something to improve:

Hi Axel,

Thank you for the patch! Yet something to improve:

[auto build test ERROR on linus/master]
[also build test ERROR on v6.3-rc1]
[cannot apply to akpm-mm/mm-everything next-20230308]
[If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/Axel-Rasmussen/mm-userfaultfd-rename-functions-for-clarity-consistency/20230307-065203
patch link:    https://lore.kernel.org/r/20230306225024.264858-5-axelrasmussen%40google.com
patch subject: [PATCH v3 4/5] mm: userfaultfd: don't separate addr + len arguments
config: x86_64-randconfig-a011-20230306 (https://download.01.org/0day-ci/archive/20230308/[email protected]/config)
compiler: clang version 14.0.6 (https://github.com/llvm/llvm-project f28c006a5895fc0e329fe15fead81e37457cb1d1)

reproduce (this is a W=1 build):
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# https://github.com/intel-lab-lkp/linux/commit/cee642b93be3ae01c7cc737c0176cbc16074a25a
git remote add linux-review https://github.com/intel-lab-lkp/linux
git fetch --no-tags linux-review Axel-Rasmussen/mm-userfaultfd-rename-functions-for-clarity-consistency/20230307-065203 git checkout cee642b93be3ae01c7cc737c0176cbc16074a25a
# save the config file
mkdir build_dir && cp config build_dir/.config
COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 O=build_dir ARCH=x86_64 olddefconfig COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 O=build_dir ARCH=x86_64 SHELL=/bin/bash

If you fix the issue, kindly add following tag where applicable
| Reported-by: kernel test robot <[email protected]>
| Link: https://lore.kernel.org/oe-kbuild-all/[email protected]/

All errors (new ones prefixed by >>):

>> mm/userfaultfd.c:577:52: error: passing 'const struct uffdio_range *' to parameter of incompatible type 'struct uffdio_range'
                   return  mfill_atomic_hugetlb(dst_vma, src_start, dst, flags);
                                                                    ^~~

mm/userfaultfd.c:463:29: note: passing argument to parameter 'dst' here
struct uffdio_range dst,
^
1 error generated.

vim +577 mm/userfaultfd.c

508
509 static __always_inline ssize_t mfill_atomic(struct mm_struct *dst_mm,

   510						    unsigned long src_start,
   511						    const struct uffdio_range *dst,
   512						    atomic_t *mmap_changing,
   513						    uffd_flags_t flags)
   514	{
   515		struct vm_area_struct *dst_vma;
   516		ssize_t err;
   517		pmd_t *dst_pmd;
   518		unsigned long src_addr, dst_addr;
   519		long copied;
   520		struct page *page;
   521	
   522		/*
   523		 * Sanitize the command parameters:
   524		 */
   525		BUG_ON(dst->start & ~PAGE_MASK);
   526		BUG_ON(dst->len & ~PAGE_MASK);
   527	
   528		/* Does the address range wrap, or is the span zero-sized? */
   529		BUG_ON(src_start + dst->len <= src_start);
   530		BUG_ON(dst->start + dst->len <= dst->start);
   531	
   532		src_addr = src_start;
   533		dst_addr = dst->start;
   534		copied = 0;
   535		page = NULL;
   536	retry:
   537		mmap_read_lock(dst_mm);
   538	
   539		/*
   540		 * If memory mappings are changing because of non-cooperative
   541		 * operation (e.g. mremap) running in parallel, bail out and
   542		 * request the user to retry later
   543		 */
   544		err = -EAGAIN;
   545		if (mmap_changing && atomic_read(mmap_changing))
   546			goto out_unlock;
   547	
   548		/*
   549		 * Make sure the vma is not shared, that the dst range is
   550		 * both valid and fully within a single existing vma.
   551		 */
   552		err = -ENOENT;
   553		dst_vma = find_dst_vma(dst_mm, dst);
   554		if (!dst_vma)
   555			goto out_unlock;
   556	
   557		err = -EINVAL;
   558		/*
   559		 * shmem_zero_setup is invoked in mmap for MAP_ANONYMOUS|MAP_SHARED but
   560		 * it will overwrite vm_ops, so vma_is_anonymous must return false.
   561		 */
   562		if (WARN_ON_ONCE(vma_is_anonymous(dst_vma) &&
   563		    dst_vma->vm_flags & VM_SHARED))
   564			goto out_unlock;
   565	
   566		/*
   567		 * validate 'mode' now that we know the dst_vma: don't allow
   568		 * a wrprotect copy if the userfaultfd didn't register as WP.
   569		 */
   570		if ((flags & MFILL_ATOMIC_WP) && !(dst_vma->vm_flags & VM_UFFD_WP))
   571			goto out_unlock;
   572	
   573		/*
   574		 * If this is a HUGETLB vma, pass off to appropriate routine
   575		 */
   576		if (is_vm_hugetlb_page(dst_vma))
 > 577			return  mfill_atomic_hugetlb(dst_vma, src_start, dst, flags);

578
579 if (!vma_is_anonymous(dst_vma) && !vma_is_shmem(dst_vma))
580 goto out_unlock;
581 if (!vma_is_shmem(dst_vma) &&
582 (flags & MFILL_ATOMIC_MODE_MASK) == MFILL_ATOMIC_CONTINUE)
583 goto out_unlock;
584
585 /*
586 * Ensure the dst_vma has a anon_vma or this page
587 * would get a NULL anon_vma when moved in the
588 * dst_vma.
589 */
590 err = -ENOMEM;
591 if (!(dst_vma->vm_flags & VM_SHARED) &&
592 unlikely(anon_vma_prepare(dst_vma)))
593 goto out_unlock;
594
595 while (src_addr < src_start + dst->len) {
596 pmd_t dst_pmdval;
597
598 BUG_ON(dst_addr >= dst->start + dst->len);
599
600 dst_pmd = mm_alloc_pmd(dst_mm, dst_addr);
601 if (unlikely(!dst_pmd)) {
602 err = -ENOMEM;
603 break;
604 }
605
606 dst_pmdval = pmdp_get_lockless(dst_pmd);
607 /*
608 * If the dst_pmd is mapped as THP don't
609 * override it and just be strict.
610 */
611 if (unlikely(pmd_trans_huge(dst_pmdval))) {
612 err = -EEXIST;
613 break;
614 }
615 if (unlikely(pmd_none(dst_pmdval)) &&
616 unlikely(__pte_alloc(dst_mm, dst_pmd))) {
617 err = -ENOMEM;
618 break;
619 }
620 /* If an huge pmd materialized from under us fail */
621 if (unlikely(pmd_trans_huge(*dst_pmd))) {
622 err = -EFAULT;
623 break;
624 }
625
626 BUG_ON(pmd_none(*dst_pmd));
627 BUG_ON(pmd_trans_huge(*dst_pmd));
628
629 err = mfill_atomic_pte(dst_pmd, dst_vma, dst_addr,
630 src_addr, &page, flags);
631 cond_resched();
632
633 if (unlikely(err == -ENOENT)) {
634 void *page_kaddr;
635
636 mmap_read_unlock(dst_mm);
637 BUG_ON(!page);
638
639 page_kaddr = kmap_local_page(page);
640 err = copy_from_user(page_kaddr,
641 (const void __user *) src_addr,
642 PAGE_SIZE);
643 kunmap_local(page_kaddr);
644 if (unlikely(err)) {
645 err = -EFAULT;
646 goto out;
647 }
648 flush_dcache_page(page);
649 goto retry;
650 } else
651 BUG_ON(page);
652
653 if (!err) {
654 dst_addr += PAGE_SIZE;
655 src_addr += PAGE_SIZE;
656 copied += PAGE_SIZE;
657
658 if (fatal_signal_pending(current))
659 err = -EINTR;
660 }
661 if (err)
662 break;
663 }
664
665 out_unlock:
666 mmap_read_unlock(dst_mm);
667 out:
668 if (page)
669 put_page(page);
670 BUG_ON(copied < 0);
671 BUG_ON(err > 0);
672 BUG_ON(!copied && !err);
673 return copied ? copied : err;
674 }
675

Signature · 2 lines
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests
#17 lore raw

Re: [PATCH v3 4/5] mm: userfaultfd: don't separate addr + len arguments

Quoted 35 lines · kernel test robot

Hi Axel,

Thank you for the patch! Yet something to improve:

[auto build test ERROR on linus/master]
[also build test ERROR on v6.3-rc1]
[cannot apply to akpm-mm/mm-everything next-20230308]
[If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/Axel-Rasmussen/mm-userfaultfd-rename-functions-for-clarity-consistency/20230307-065203
patch link:    https://lore.kernel.org/r/20230306225024.264858-5-axelrasmussen%40google.com
patch subject: [PATCH v3 4/5] mm: userfaultfd: don't separate addr + len arguments
config: x86_64-randconfig-a011-20230306 (https://download.01.org/0day-ci/archive/20230308/[email protected]/config)
compiler: clang version 14.0.6 (https://github.com/llvm/llvm-project f28c006a5895fc0e329fe15fead81e37457cb1d1)

reproduce (this is a W=1 build):
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# https://github.com/intel-lab-lkp/linux/commit/cee642b93be3ae01c7cc737c0176cbc16074a25a
git remote add linux-review https://github.com/intel-lab-lkp/linux
git fetch --no-tags linux-review Axel-Rasmussen/mm-userfaultfd-rename-functions-for-clarity-consistency/20230307-065203 git checkout cee642b93be3ae01c7cc737c0176cbc16074a25a
# save the config file
mkdir build_dir && cp config build_dir/.config
COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 O=build_dir ARCH=x86_64 olddefconfig COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 O=build_dir ARCH=x86_64 SHELL=/bin/bash

If you fix the issue, kindly add following tag where applicable
| Reported-by: kernel test robot <[email protected]>
| Link: https://lore.kernel.org/oe-kbuild-all/[email protected]/

All errors (new ones prefixed by >>):

Quoted 7 lines
Quoted 1 line
Quoted 1 line

mm/userfaultfd.c:577:52: error: passing 'const struct uffdio_range *' to parameter of incompatible type 'struct uffdio_range'

                   return  mfill_atomic_hugetlb(dst_vma, src_start, dst, flags);
                                                                    ^~~

mm/userfaultfd.c:463:29: note: passing argument to parameter 'dst' here
struct uffdio_range dst,
^
1 error generated.

Whoops. :) I admittedly didn't test with !CONFIG_HUGETLB_PAGE.

The next version of this series will drop this patch as per discussion though, so the issue is moot.

Quoted 176 lines

vim +577 mm/userfaultfd.c

   508
   509  static __always_inline ssize_t mfill_atomic(struct mm_struct *dst_mm,
   510                                              unsigned long src_start,
   511                                              const struct uffdio_range *dst,
   512                                              atomic_t *mmap_changing,
   513                                              uffd_flags_t flags)
   514  {
   515          struct vm_area_struct *dst_vma;
   516          ssize_t err;
   517          pmd_t *dst_pmd;
   518          unsigned long src_addr, dst_addr;
   519          long copied;
   520          struct page *page;
   521
   522          /*
   523           * Sanitize the command parameters:
   524           */
   525          BUG_ON(dst->start & ~PAGE_MASK);
   526          BUG_ON(dst->len & ~PAGE_MASK);
   527
   528          /* Does the address range wrap, or is the span zero-sized? */
   529          BUG_ON(src_start + dst->len <= src_start);
   530          BUG_ON(dst->start + dst->len <= dst->start);
   531
   532          src_addr = src_start;
   533          dst_addr = dst->start;
   534          copied = 0;
   535          page = NULL;
   536  retry:
   537          mmap_read_lock(dst_mm);
   538
   539          /*
   540           * If memory mappings are changing because of non-cooperative
   541           * operation (e.g. mremap) running in parallel, bail out and
   542           * request the user to retry later
   543           */
   544          err = -EAGAIN;
   545          if (mmap_changing && atomic_read(mmap_changing))
   546                  goto out_unlock;
   547
   548          /*
   549           * Make sure the vma is not shared, that the dst range is
   550           * both valid and fully within a single existing vma.
   551           */
   552          err = -ENOENT;
   553          dst_vma = find_dst_vma(dst_mm, dst);
   554          if (!dst_vma)
   555                  goto out_unlock;
   556
   557          err = -EINVAL;
   558          /*
   559           * shmem_zero_setup is invoked in mmap for MAP_ANONYMOUS|MAP_SHARED but
   560           * it will overwrite vm_ops, so vma_is_anonymous must return false.
   561           */
   562          if (WARN_ON_ONCE(vma_is_anonymous(dst_vma) &&
   563              dst_vma->vm_flags & VM_SHARED))
   564                  goto out_unlock;
   565
   566          /*
   567           * validate 'mode' now that we know the dst_vma: don't allow
   568           * a wrprotect copy if the userfaultfd didn't register as WP.
   569           */
   570          if ((flags & MFILL_ATOMIC_WP) && !(dst_vma->vm_flags & VM_UFFD_WP))
   571                  goto out_unlock;
   572
   573          /*
   574           * If this is a HUGETLB vma, pass off to appropriate routine
   575           */
   576          if (is_vm_hugetlb_page(dst_vma))
Quoted 1 line

577 return mfill_atomic_hugetlb(dst_vma, src_start, dst, flags);

578
579 if (!vma_is_anonymous(dst_vma) && !vma_is_shmem(dst_vma))

   580                  goto out_unlock;
   581          if (!vma_is_shmem(dst_vma) &&
   582              (flags & MFILL_ATOMIC_MODE_MASK) == MFILL_ATOMIC_CONTINUE)
   583                  goto out_unlock;
   584
   585          /*
   586           * Ensure the dst_vma has a anon_vma or this page
   587           * would get a NULL anon_vma when moved in the
   588           * dst_vma.
   589           */
   590          err = -ENOMEM;
   591          if (!(dst_vma->vm_flags & VM_SHARED) &&
   592              unlikely(anon_vma_prepare(dst_vma)))
   593                  goto out_unlock;
   594
   595          while (src_addr < src_start + dst->len) {
   596                  pmd_t dst_pmdval;
   597
   598                  BUG_ON(dst_addr >= dst->start + dst->len);
   599
   600                  dst_pmd = mm_alloc_pmd(dst_mm, dst_addr);
   601                  if (unlikely(!dst_pmd)) {
   602                          err = -ENOMEM;
   603                          break;
   604                  }
   605
   606                  dst_pmdval = pmdp_get_lockless(dst_pmd);
   607                  /*
   608                   * If the dst_pmd is mapped as THP don't
   609                   * override it and just be strict.
   610                   */
   611                  if (unlikely(pmd_trans_huge(dst_pmdval))) {
   612                          err = -EEXIST;
   613                          break;
   614                  }
   615                  if (unlikely(pmd_none(dst_pmdval)) &&
   616                      unlikely(__pte_alloc(dst_mm, dst_pmd))) {
   617                          err = -ENOMEM;
   618                          break;
   619                  }
   620                  /* If an huge pmd materialized from under us fail */
   621                  if (unlikely(pmd_trans_huge(*dst_pmd))) {
   622                          err = -EFAULT;
   623                          break;
   624                  }
   625
   626                  BUG_ON(pmd_none(*dst_pmd));
   627                  BUG_ON(pmd_trans_huge(*dst_pmd));
   628
   629                  err = mfill_atomic_pte(dst_pmd, dst_vma, dst_addr,
   630                                         src_addr, &page, flags);
   631                  cond_resched();
   632
   633                  if (unlikely(err == -ENOENT)) {
   634                          void *page_kaddr;
   635
   636                          mmap_read_unlock(dst_mm);
   637                          BUG_ON(!page);
   638
   639                          page_kaddr = kmap_local_page(page);
   640                          err = copy_from_user(page_kaddr,
   641                                               (const void __user *) src_addr,
   642                                               PAGE_SIZE);
   643                          kunmap_local(page_kaddr);
   644                          if (unlikely(err)) {
   645                                  err = -EFAULT;
   646                                  goto out;
   647                          }
   648                          flush_dcache_page(page);
   649                          goto retry;
   650                  } else
   651                          BUG_ON(page);
   652
   653                  if (!err) {
   654                          dst_addr += PAGE_SIZE;
   655                          src_addr += PAGE_SIZE;
   656                          copied += PAGE_SIZE;
   657
   658                          if (fatal_signal_pending(current))
   659                                  err = -EINTR;
   660                  }
   661                  if (err)
   662                          break;
   663          }
   664
   665  out_unlock:
   666          mmap_read_unlock(dst_mm);
   667  out:
   668          if (page)
   669                  put_page(page);
   670          BUG_ON(copied < 0);
   671          BUG_ON(err > 0);
   672          BUG_ON(!copied && !err);
   673          return copied ? copied : err;
   674  }
   675

--
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests

#18 lore raw

PATCH v3 5/5 mm: userfaultfd: add UFFDIO_CONTINUE_MODE_WP to install WP PTEs

In reply to Axel Rasmussen

UFFDIO_COPY already has UFFDIO_COPY_MODE_WP, so when installing a new PTE to resolve a missing fault, one can install a write-protected one. This is useful when using UFFDIO_REGISTER_MODE_{MISSING,WP} in combination.

So, add an analogous UFFDIO_CONTINUE_MODE_WP, which does the same thing but for *minor* faults.

Update the selftest to do some very basic exercising of the new flag.

---
 fs/userfaultfd.c                         | 8 ++++++--
 include/linux/userfaultfd_k.h            | 2 +-
 include/uapi/linux/userfaultfd.h         | 7 +++++++
 mm/userfaultfd.c                         | 5 +++--
 tools/testing/selftests/mm/userfaultfd.c | 4 ++++
 5 files changed, 21 insertions(+), 5 deletions(-)
fs/userfaultfd.c +6 -2
index 984b63b0fc75..b5750e20ae00 100644
@@ -1859,6 +1859,7 @@ static int userfaultfd_continue(struct userfaultfd_ctx *ctx, unsigned long arg)
 	struct uffdio_continue uffdio_continue;
 	struct uffdio_continue __user *user_uffdio_continue;
 	struct uffdio_range range;
+	int flags = 0;
 
 	user_uffdio_continue = (struct uffdio_continue __user *)arg;
 
@@ -1881,12 +1882,15 @@ static int userfaultfd_continue(struct userfaultfd_ctx *ctx, unsigned long arg)
 	/* double check for wraparound just in case. */
 	if (range.start + range.len <= range.start)
 		goto out;
-	if (uffdio_continue.mode & ~UFFDIO_CONTINUE_MODE_DONTWAKE)
+	if (uffdio_continue.mode & ~(UFFDIO_CONTINUE_MODE_DONTWAKE |
+				     UFFDIO_CONTINUE_MODE_WP))
 		goto out;
+	if (uffdio_continue.mode & UFFDIO_CONTINUE_MODE_WP)
+		flags |= MFILL_ATOMIC_WP;
 
 	if (mmget_not_zero(ctx->mm)) {
 		ret = mfill_atomic_continue(ctx->mm, &range,
-					    &ctx->mmap_changing);
+					    &ctx->mmap_changing, flags);
 		mmput(ctx->mm);
 	} else {
 		return -ESRCH;
include/linux/userfaultfd_k.h +1 -1
index fcd95e3d3dcd..d691f898bae2 100644
@@ -71,7 +71,7 @@ extern ssize_t mfill_atomic_zeropage(struct mm_struct *dst_mm,
 				     atomic_t *mmap_changing);
 extern ssize_t mfill_atomic_continue(struct mm_struct *dst_mm,
 				     const struct uffdio_range *dst,
-				    atomic_t *mmap_changing);
+				    atomic_t *mmap_changing, int flags);
 extern int mwriteprotect_range(struct mm_struct *dst_mm,
 			       const struct uffdio_range *range,
 			       bool enable_wp, atomic_t *mmap_changing);
include/uapi/linux/userfaultfd.h +7 -0
index 005e5e306266..14059a0861bf 100644
@@ -297,6 +297,13 @@ struct uffdio_writeprotect {
 struct uffdio_continue {
 	struct uffdio_range range;
 #define UFFDIO_CONTINUE_MODE_DONTWAKE		((__u64)1<<0)
+	/*
+	 * UFFDIO_CONTINUE_MODE_WP will map the page write protected on
+	 * the fly.  UFFDIO_CONTINUE_MODE_WP is available only if the
+	 * write protected ioctl is implemented for the range
+	 * according to the uffdio_register.ioctls.
+	 */
+#define UFFDIO_CONTINUE_MODE_WP			((__u64)1<<1)
 	__u64 mode;
 
 	/*
mm/userfaultfd.c +3 -2
index 870e7489e8d1..6adbfc8dc277 100644
@@ -691,10 +691,11 @@ ssize_t mfill_atomic_zeropage(struct mm_struct *dst_mm,
 
 ssize_t mfill_atomic_continue(struct mm_struct *dst_mm,
 			      const struct uffdio_range *dst,
-			      atomic_t *mmap_changing)
+			      atomic_t *mmap_changing,
+			      int flags)
 {
 	return mfill_atomic(dst_mm, 0, dst,
-			    mmap_changing, MFILL_ATOMIC_CONTINUE);
+			    mmap_changing, flags | MFILL_ATOMIC_CONTINUE);
 }
 
 long uffd_wp_range(struct vm_area_struct *dst_vma,
tools/testing/selftests/mm/userfaultfd.c +4 -0
index 7f22844ed704..41c1f9abc481 100644
@@ -585,6 +585,8 @@ static void continue_range(int ufd, __u64 start, __u64 len)
 	req.range.start = start;
 	req.range.len = len;
 	req.mode = 0;
+	if (test_uffdio_wp)
+		req.mode |= UFFDIO_CONTINUE_MODE_WP;
 
 	if (ioctl(ufd, UFFDIO_CONTINUE, &req))
 		err("UFFDIO_CONTINUE failed for address 0x%" PRIx64,
@@ -1332,6 +1334,8 @@ static int userfaultfd_minor_test(void)
 	uffdio_register.range.start = (unsigned long)area_dst_alias;
 	uffdio_register.range.len = nr_pages * page_size;
 	uffdio_register.mode = UFFDIO_REGISTER_MODE_MINOR;
+	if (test_uffdio_wp)
+		uffdio_register.mode |= UFFDIO_REGISTER_MODE_WP;
 	if (ioctl(uffd, UFFDIO_REGISTER, &uffdio_register))
 		err("register failure");
 
Signature · 1 line
2.40.0.rc0.216.gc4246ad0f0-goog
reviewer Acked-by on v3 5/5
#19 lore raw

Re: [PATCH v3 5/5] mm: userfaultfd: add UFFDIO_CONTINUE_MODE_WP to install WP PTEs

Quoted 11 lines · Axel Rasmussen

UFFDIO_COPY already has UFFDIO_COPY_MODE_WP, so when installing a new PTE to resolve a missing fault, one can install a write-protected one. This is useful when using UFFDIO_REGISTER_MODE_{MISSING,WP} in combination.

So, add an analogous UFFDIO_CONTINUE_MODE_WP, which does the same thing but for *minor* faults.

Update the selftest to do some very basic exercising of the new flag.

Some mentioning on the use case would be nice. :) No objection having it.

Thanks,

Signature · 1 line
Peter Xu