Skip to contentSkip to conversation

fs/userfaultfd.c: remove a redundant check on end

PATCH In review Wei Yang 2 participants 5 messages Sep 12, 2019 – Nov 9, 2019 VFS

lore
Share & tools
lore

Wei Yang

author
#0 lore raw

PATCH 1/3 fs/userfaultfd.c: remove a redundant check on end

For the ending vma, there is a check to make sure the end is huge page aligned.

The *if* check makes sure vm_start < end <= vm_end. While the first half is not necessary, because the *for* clause makes sure vm_start < end.

This patch just removes it.

---
 fs/userfaultfd.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
fs/userfaultfd.c +1 -2
index 653d8f7c453c..9ce09ac619a2 100644
@@ -1402,8 +1402,7 @@ static int userfaultfd_register(struct userfaultfd_ctx *ctx,
 		 * If this vma contains ending address, and huge pages
 		 * check alignment.
 		 */
-		if (is_vm_hugetlb_page(cur) && end <= cur->vm_end &&
-		    end > cur->vm_start) {
+		if (is_vm_hugetlb_page(cur) && end <= cur->vm_end) {
 			unsigned long vma_hpagesize = vma_kernel_pagesize(cur);
 
 			ret = -EINVAL;
Signature · 1 line
2.17.1
author
#1 lore raw

PATCH 2/3 fs/userfaultfd.c: reorder the if check to reduce some computation

When there are several condition check in *if* clause, the check will stop at the first false one.

Since the for loop iterates vma list, we are sure only the last vma meets the condition "end <= vm_end". Reorder the check sequence to reduce some computation.

---
 fs/userfaultfd.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/userfaultfd.c +1 -1
index 9ce09ac619a2..70c0e0ef01d7 100644
@@ -1402,7 +1402,7 @@ static int userfaultfd_register(struct userfaultfd_ctx *ctx,
 		 * If this vma contains ending address, and huge pages
 		 * check alignment.
 		 */
-		if (is_vm_hugetlb_page(cur) && end <= cur->vm_end) {
+		if (end <= cur->vm_end && is_vm_hugetlb_page(cur)) {
 			unsigned long vma_hpagesize = vma_kernel_pagesize(cur);
 
 			ret = -EINVAL;
Signature · 1 line
2.17.1
author
#2 lore raw

PATCH 3/3 fs/userfaultfd.c: wrap cheching huge page alignment into a helper

In reply to Wei Yang

There are three places checking whether one address is huge page aligned.

This patch just makes a helper function to wrap it up.

---
 fs/userfaultfd.c | 30 +++++++++++++++---------------
 1 file changed, 15 insertions(+), 15 deletions(-)
fs/userfaultfd.c +15 -15
index 70c0e0ef01d7..d8665ffdd576 100644
@@ -1296,6 +1296,16 @@ static inline bool vma_can_userfault(struct vm_area_struct *vma)
 		vma_is_shmem(vma);
 }
 
+static inline bool addr_huge_page_aligned(unsigned long addr,
+					  struct vm_area_struct *vma)
+{
+	unsigned long vma_hpagesize = vma_kernel_pagesize(vma);
+
+	if (addr & (vma_hpagesize - 1))
+		return false;
+	return true;
+}
+
 static int userfaultfd_register(struct userfaultfd_ctx *ctx,
 				unsigned long arg)
 {
@@ -1363,12 +1373,8 @@ static int userfaultfd_register(struct userfaultfd_ctx *ctx,
 	 * If the first vma contains huge pages, make sure start address
 	 * is aligned to huge page size.
 	 */
-	if (is_vm_hugetlb_page(vma)) {
-		unsigned long vma_hpagesize = vma_kernel_pagesize(vma);
-
-		if (start & (vma_hpagesize - 1))
-			goto out_unlock;
-	}
+	if (is_vm_hugetlb_page(vma) && !addr_huge_page_aligned(start, vma))
+		goto out_unlock;
 
 	/*
 	 * Search for not compatible vmas.
@@ -1403,11 +1409,9 @@ static int userfaultfd_register(struct userfaultfd_ctx *ctx,
 		 * check alignment.
 		 */
 		if (end <= cur->vm_end && is_vm_hugetlb_page(cur)) {
-			unsigned long vma_hpagesize = vma_kernel_pagesize(cur);
-
 			ret = -EINVAL;
 
-			if (end & (vma_hpagesize - 1))
+			if (!addr_huge_page_aligned(end, cur))
 				goto out_unlock;
 		}
 
@@ -1551,12 +1555,8 @@ static int userfaultfd_unregister(struct userfaultfd_ctx *ctx,
 	 * If the first vma contains huge pages, make sure start address
 	 * is aligned to huge page size.
 	 */
-	if (is_vm_hugetlb_page(vma)) {
-		unsigned long vma_hpagesize = vma_kernel_pagesize(vma);
-
-		if (start & (vma_hpagesize - 1))
-			goto out_unlock;
-	}
+	if (is_vm_hugetlb_page(vma) && !addr_huge_page_aligned(start, vma))
+		goto out_unlock;
 
 	/*
 	 * Search for not compatible vmas.
Signature · 1 line
2.17.1
author
#3 lore raw
In reply to Wei Yang

Ping for comment :-)

Quoted 29 lines · Wei Yang

For the ending vma, there is a check to make sure the end is huge page aligned.

The *if* check makes sure vm_start < end <= vm_end. While the first half is not necessary, because the *for* clause makes sure vm_start < end.

This patch just removes it.

---
fs/userfaultfd.c | 3 +--

1 file changed, 1 insertion(+), 2 deletions(-)
fs/userfaultfd.c +1 -2
index 653d8f7c453c..9ce09ac619a2 100644
@@ -1402,8 +1402,7 @@ static int userfaultfd_register(struct userfaultfd_ctx *ctx,
 		 * If this vma contains ending address, and huge pages
 		 * check alignment.
 		 */
-		if (is_vm_hugetlb_page(cur) && end <= cur->vm_end &&
-		    end > cur->vm_start) {
+		if (is_vm_hugetlb_page(cur) && end <= cur->vm_end) {
 			unsigned long vma_hpagesize = vma_kernel_pagesize(cur);
 
 			ret = -EINVAL;
Signature · 1 line
2.17.1
Signature · 2 lines
Wei Yang
Help you, Help me
#4 lore raw
In reply to Wei Yang
Quoted 8 lines · Wei Yang

For the ending vma, there is a check to make sure the end is huge page aligned.

The *if* check makes sure vm_start < end <= vm_end. While the first half is not necessary, because the *for* clause makes sure vm_start < end.

This patch just removes it.

Does this one look good?

Quoted 21 lines

---
fs/userfaultfd.c | 3 +--

1 file changed, 1 insertion(+), 2 deletions(-)
fs/userfaultfd.c +1 -2
index 653d8f7c453c..9ce09ac619a2 100644
@@ -1402,8 +1402,7 @@ static int userfaultfd_register(struct userfaultfd_ctx *ctx,
 		 * If this vma contains ending address, and huge pages
 		 * check alignment.
 		 */
-		if (is_vm_hugetlb_page(cur) && end <= cur->vm_end &&
-		    end > cur->vm_start) {
+		if (is_vm_hugetlb_page(cur) && end <= cur->vm_end) {
 			unsigned long vma_hpagesize = vma_kernel_pagesize(cur);
 
 			ret = -EINVAL;
Signature · 1 line
2.17.1
Signature · 2 lines
Wei Yang
Help you, Help me