Skip to contentSkip to conversation

add missing page refcount overflow checks

PATCH Applied Vlastimil Babka 4 participants 9 messages Nov 29, 2019 – Dec 4, 2019

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 add missing page refcount overflow checks

This collection of patches add the missing overflow checks in arch-specific gup.c variants for x86 and s390. Those were missed in backport of 8fde12ca79af ("mm: prevent get_user_pages() from overflowing page refcount") as mainline had a single gup.c implementation at that point. See individual patches for details.

Vlastimil

Signature · 1 line
2.24.0
#1 lore raw

PATCH 1/1 mm, gup: add missing refcount overflow checks on s390

The mainline commit 8fde12ca79af ("mm: prevent get_user_pages() from overflowing page refcount") was backported to 4.14.y stable as commit 04198de24771. The backport however missed that in 4.14, there are several arch-specific gup.c versions with fast gup implementations, so these do not prevent refcount overflow.

This stable-only commit fixes the s390 version, and is based on the backport in SUSE SLES/openSUSE 4.12-based kernels.

The remaining architectures with own gup.c are sparc, mips, sh. It's unlikely the known overflow scenario based on FUSE, which needs 140GB of RAM, is a problem for those architectures, and I don't feel confident enough to patch them.

---
 arch/s390/mm/gup.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
arch/s390/mm/gup.c +6 -3
index 05c8abd864f1..9bce54eac0b0 100644
@@ -39,7 +39,8 @@ static inline int gup_pte_range(pmd_t *pmdp, pmd_t pmd, unsigned long addr,
 		VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
 		page = pte_page(pte);
 		head = compound_head(page);
-		if (!page_cache_get_speculative(head))
+		if (unlikely(WARN_ON_ONCE(page_ref_count(head) < 0)
+		    || !page_cache_get_speculative(head)))
 			return 0;
 		if (unlikely(pte_val(pte) != pte_val(*ptep))) {
 			put_page(head);
@@ -77,7 +78,8 @@ static inline int gup_huge_pmd(pmd_t *pmdp, pmd_t pmd, unsigned long addr,
 		refs++;
 	} while (addr += PAGE_SIZE, addr != end);
 
-	if (!page_cache_add_speculative(head, refs)) {
+	if (unlikely(WARN_ON_ONCE(page_ref_count(head) < 0)
+	    || !page_cache_add_speculative(head, refs))) {
 		*nr -= refs;
 		return 0;
 	}
@@ -151,7 +153,8 @@ static int gup_huge_pud(pud_t *pudp, pud_t pud, unsigned long addr,
 		refs++;
 	} while (addr += PAGE_SIZE, addr != end);
 
-	if (!page_cache_add_speculative(head, refs)) {
+	if (unlikely(WARN_ON_ONCE(page_ref_count(head) < 0)
+	    || !page_cache_add_speculative(head, refs))) {
 		*nr -= refs;
 		return 0;
 	}
Signature · 1 line
2.24.0
#2 lore raw

PATCH 1/1 mm, gup: add missing refcount overflow checks on s390

In reply to Vlastimil Babka

The mainline commit 8fde12ca79af ("mm: prevent get_user_pages() from overflowing page refcount") was backported to 4.19.y stable as commit d972ebbf42ba. The backport however missed that in 4.19, there are several arch-specific gup.c versions with fast gup implementations, so these do not prevent refcount overflow.

This stable-only commit fixes the s390 version, and is based on the backport in SUSE SLES/openSUSE 4.12-based kernels.

The remaining architectures with own gup.c are sparc, mips, sh. It's unlikely the known overflow scenario based on FUSE, which needs 140GB of RAM, is a problem for those architectures, and I don't feel confident enough to patch them.

---
 arch/s390/mm/gup.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
arch/s390/mm/gup.c +6 -3
index 2809d11c7a28..9b5b866d8adf 100644
@@ -39,7 +39,8 @@ static inline int gup_pte_range(pmd_t *pmdp, pmd_t pmd, unsigned long addr,
 		VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
 		page = pte_page(pte);
 		head = compound_head(page);
-		if (!page_cache_get_speculative(head))
+		if (unlikely(WARN_ON_ONCE(page_ref_count(head) < 0)
+		    || !page_cache_get_speculative(head)))
 			return 0;
 		if (unlikely(pte_val(pte) != pte_val(*ptep))) {
 			put_page(head);
@@ -77,7 +78,8 @@ static inline int gup_huge_pmd(pmd_t *pmdp, pmd_t pmd, unsigned long addr,
 		refs++;
 	} while (addr += PAGE_SIZE, addr != end);
 
-	if (!page_cache_add_speculative(head, refs)) {
+	if (unlikely(WARN_ON_ONCE(page_ref_count(head) < 0)
+	    || !page_cache_add_speculative(head, refs))) {
 		*nr -= refs;
 		return 0;
 	}
@@ -151,7 +153,8 @@ static int gup_huge_pud(pud_t *pudp, pud_t pud, unsigned long addr,
 		refs++;
 	} while (addr += PAGE_SIZE, addr != end);
 
-	if (!page_cache_add_speculative(head, refs)) {
+	if (unlikely(WARN_ON_ONCE(page_ref_count(head) < 0)
+	    || !page_cache_add_speculative(head, refs))) {
 		*nr -= refs;
 		return 0;
 	}
Signature · 1 line
2.24.0
#3 lore raw

PATCH 1/1 mm, gup: add missing refcount overflow checks on x86 and s390

In reply to Vlastimil Babka

The mainline commit 8fde12ca79af ("mm: prevent get_user_pages() from overflowing page refcount") was backported to 4.9.y stable as commit 2ed768cfd895. The backport however missed that in 4.9, there are several arch-specific gup.c versions with fast gup implementations, so these do not prevent refcount overflow.

This is partially fixed for x86 in stable-only commit d73af79742e7 ("x86, mm,
gup: prevent get_page() race with munmap in paravirt guest"). This stable-only commit adds missing parts to x86 version, as well as s390 version, both taken from the SUSE SLES/openSUSE 4.12-based kernels.

The remaining architectures with own gup.c are sparc, mips, sh. It's unlikely the known overflow scenario based on FUSE, which needs 140GB of RAM, is a problem for those architectures, and I don't feel confident enough to patch them.

---
 arch/s390/mm/gup.c |  9 ++++++---
 arch/x86/mm/gup.c  | 10 ++++++++--
 2 files changed, 14 insertions(+), 5 deletions(-)
arch/s390/mm/gup.c +6 -3
index 97fc449a7470..33a940389a6d 100644
@@ -38,7 +38,8 @@ static inline int gup_pte_range(pmd_t *pmdp, pmd_t pmd, unsigned long addr,
 		VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
 		page = pte_page(pte);
 		head = compound_head(page);
-		if (!page_cache_get_speculative(head))
+		if (unlikely(WARN_ON_ONCE(page_ref_count(head) < 0)
+		    || !page_cache_get_speculative(head)))
 			return 0;
 		if (unlikely(pte_val(pte) != pte_val(*ptep))) {
 			put_page(head);
@@ -76,7 +77,8 @@ static inline int gup_huge_pmd(pmd_t *pmdp, pmd_t pmd, unsigned long addr,
 		refs++;
 	} while (addr += PAGE_SIZE, addr != end);
 
-	if (!page_cache_add_speculative(head, refs)) {
+	if (unlikely(WARN_ON_ONCE(page_ref_count(head) < 0)
+	    || !page_cache_add_speculative(head, refs))) {
 		*nr -= refs;
 		return 0;
 	}
@@ -150,7 +152,8 @@ static int gup_huge_pud(pud_t *pudp, pud_t pud, unsigned long addr,
 		refs++;
 	} while (addr += PAGE_SIZE, addr != end);
 
-	if (!page_cache_add_speculative(head, refs)) {
+	if (unlikely(WARN_ON_ONCE(page_ref_count(head) < 0)
+	    || !page_cache_add_speculative(head, refs))) {
 		*nr -= refs;
 		return 0;
 	}
arch/x86/mm/gup.c +8 -2
index d7db45bdfb3b..551fc7fea046 100644
@@ -202,10 +202,12 @@ static int __gup_device_huge_pmd(pmd_t pmd, unsigned long addr,
 			undo_dev_pagemap(nr, nr_start, pages);
 			return 0;
 		}
+		if (unlikely(!try_get_page(page))) {
+			put_dev_pagemap(pgmap);
+			return 0;
+		}
 		SetPageReferenced(page);
 		pages[*nr] = page;
-		get_page(page);
-		put_dev_pagemap(pgmap);
 		(*nr)++;
 		pfn++;
 	} while (addr += PAGE_SIZE, addr != end);
@@ -230,6 +232,8 @@ static noinline int gup_huge_pmd(pmd_t pmd, unsigned long addr,
 
 	refs = 0;
 	head = pmd_page(pmd);
+	if (WARN_ON_ONCE(page_ref_count(head) <= 0))
+		return 0;
 	page = head + ((addr & ~PMD_MASK) >> PAGE_SHIFT);
 	do {
 		VM_BUG_ON_PAGE(compound_head(page) != head, page);
@@ -289,6 +293,8 @@ static noinline int gup_huge_pud(pud_t pud, unsigned long addr,
 
 	refs = 0;
 	head = pud_page(pud);
+	if (WARN_ON_ONCE(page_ref_count(head) <= 0))
+		return 0;
 	page = head + ((addr & ~PUD_MASK) >> PAGE_SHIFT);
 	do {
 		VM_BUG_ON_PAGE(compound_head(page) != head, page);
Signature · 1 line
2.24.0
#4 lore raw

Re: [PATCH STABLE 4.9 1/1] mm, gup: add missing refcount overflow checks on x86 and s390

Quoted 27 lines · Vlastimil Babka

The mainline commit 8fde12ca79af ("mm: prevent get_user_pages() from overflowing page refcount") was backported to 4.9.y stable as commit 2ed768cfd895. The backport however missed that in 4.9, there are several arch-specific gup.c versions with fast gup implementations, so these do not prevent refcount overflow.

This is partially fixed for x86 in stable-only commit d73af79742e7 ("x86, mm,
gup: prevent get_page() race with munmap in paravirt guest"). This stable-only commit adds missing parts to x86 version, as well as s390 version, both taken from the SUSE SLES/openSUSE 4.12-based kernels.

The remaining architectures with own gup.c are sparc, mips, sh. It's unlikely the known overflow scenario based on FUSE, which needs 140GB of RAM, is a problem for those architectures, and I don't feel confident enough to patch them.

---
 arch/s390/mm/gup.c |  9 ++++++---
 arch/x86/mm/gup.c  | 10 ++++++++--
 2 files changed, 14 insertions(+), 5 deletions(-)
arch/s390/mm/gup.c +0 -0
index 97fc449a7470..33a940389a6d 100644

arch/s390/mm/gup.c static inline int gup_pte_range(pmd_t *pmdp, pmd_t pmd, unsigned long addr, · line 41

Quoted 5 lines
arch/s390/mm/gup.c +1 -1
@@ -38,7 +38,8 @@ static inline int gup_pte_range(pmd_t *pmdp, pmd_t pmd, unsigned long addr,
 		VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
 		page = pte_page(pte);
 		head = compound_head(page);
-		if (!page_cache_get_speculative(head))
+		if (unlikely(WARN_ON_ONCE(page_ref_count(head) < 0)

No need for unlikely(); WARN_ON() includes that.

Quoted 4 lines
arch/s390/mm/gup.c +1 -0
+		    || !page_cache_get_speculative(head)))
 			return 0;
 		if (unlikely(pte_val(pte) != pte_val(*ptep))) {
 			put_page(head);

[...]

Quoted 4 lines
arch/x86/mm/gup.c +0 -0
@@ -202,10 +202,12 @@ static int __gup_device_huge_pmd(pmd_t pmd, unsigned long addr,
 			undo_dev_pagemap(nr, nr_start, pages);

arch/x86/mm/gup.c static int __gup_device_huge_pmd(pmd_t pmd, unsigned long addr, · line 208

Quoted 10 lines
arch/x86/mm/gup.c +4 -2
 			return 0;
 		}
+		if (unlikely(!try_get_page(page))) {
+			put_dev_pagemap(pgmap);
+			return 0;
+		}
 		SetPageReferenced(page);
 		pages[*nr] = page;
-		get_page(page);
-		put_dev_pagemap(pgmap);

This leaks a pgmap reference on success!

arch/x86/mm/gup.c static noinline int gup_huge_pmd(pmd_t pmd, unsigned long addr, · line 235

Quoted 8 lines
arch/x86/mm/gup.c +1 -0
 		(*nr)++;
 		pfn++;
 	} while (addr += PAGE_SIZE, addr != end);
@@ -230,6 +232,8 @@ static noinline int gup_huge_pmd(pmd_t pmd, unsigned long addr,
 
 	refs = 0;
 	head = pmd_page(pmd);
+	if (WARN_ON_ONCE(page_ref_count(head) <= 0))

Why <= 0, given we use < 0 elsewhere?

arch/x86/mm/gup.c static noinline int gup_huge_pud(pud_t pud, unsigned long addr, · line 296

Quoted 9 lines
arch/x86/mm/gup.c +2 -0
+		return 0;
 	page = head + ((addr & ~PMD_MASK) >> PAGE_SHIFT);
 	do {
 		VM_BUG_ON_PAGE(compound_head(page) != head, page);
@@ -289,6 +293,8 @@ static noinline int gup_huge_pud(pud_t pud, unsigned long addr,
 
 	refs = 0;
 	head = pud_page(pud);
+	if (WARN_ON_ONCE(page_ref_count(head) <= 0))

Same question here.

Ben.

Quoted 4 lines
arch/x86/mm/gup.c +1 -0
+		return 0;
 	page = head + ((addr & ~PUD_MASK) >> PAGE_SHIFT);
 	do {
 		VM_BUG_ON_PAGE(compound_head(page) != head, page);
Signature · 3 lines
Ben Hutchings, Software Developer                         Codethink Ltd
https://www.codethink.co.uk/                 Dale House, 35 Dale Street
                                     Manchester, M1 2HF, United Kingdom
#5 lore raw

Re: [PATCH STABLE 4.9 1/1] mm, gup: add missing refcount overflow checks on x86 and s390

Quoted 5 lines · Ben Hutchings
Quoted 4 lines
arch/s390/mm/gup.c +1 -0
+		    || !page_cache_get_speculative(head)))
 			return 0;
 		if (unlikely(pte_val(pte) != pte_val(*ptep))) {
 			put_page(head);

[...]

Quoted 16 lines
Quoted 4 lines
arch/x86/mm/gup.c +0 -0
@@ -202,10 +202,12 @@ static int __gup_device_huge_pmd(pmd_t pmd, unsigned long addr,
 			undo_dev_pagemap(nr, nr_start, pages);

arch/x86/mm/gup.c static int __gup_device_huge_pmd(pmd_t pmd, unsigned long addr, · line 208

Quoted 10 lines
arch/x86/mm/gup.c +4 -2
 			return 0;
 		}
+		if (unlikely(!try_get_page(page))) {
+			put_dev_pagemap(pgmap);
+			return 0;
+		}
 		SetPageReferenced(page);
 		pages[*nr] = page;
-		get_page(page);
-		put_dev_pagemap(pgmap);

This leaks a pgmap reference on success!

Good catch, deleted one line too many!

Quoted 10 lines

arch/x86/mm/gup.c static noinline int gup_huge_pmd(pmd_t pmd, unsigned long addr, · line 235

Quoted 8 lines
arch/x86/mm/gup.c +1 -0
 		(*nr)++;
 		pfn++;
 	} while (addr += PAGE_SIZE, addr != end);
@@ -230,6 +232,8 @@ static noinline int gup_huge_pmd(pmd_t pmd, unsigned long addr,
 
 	refs = 0;
 	head = pmd_page(pmd);
+	if (WARN_ON_ONCE(page_ref_count(head) <= 0))

Why <= 0, given we use < 0 elsewhere?

The code uses get_head_page_multiple() which boils down to atomic_add and not add_unless_zero(), so it assumes a pre-existing pin that must not go away or it's a bug (one that I've been hunting recently in this area). The check makes it explicit.

Quoted 7 lines
Quoted 7 lines
arch/x86/mm/gup.c +1 -0
+		return 0;
 	page = head + ((addr & ~PMD_MASK) >> PAGE_SHIFT);
 	do {
 		VM_BUG_ON_PAGE(compound_head(page) != head, page);
@@ -289,6 +293,8 @@ static noinline int gup_huge_pud(pud_t pud, unsigned long addr,
 
Quoted 5 lines

arch/x86/mm/gup.c static noinline int gup_huge_pud(pud_t pud, unsigned long addr, · line 296

Quoted 3 lines
arch/x86/mm/gup.c +1 -0
 	refs = 0;
 	head = pud_page(pud);
+	if (WARN_ON_ONCE(page_ref_count(head) <= 0))

Same question here.

Same as above.

Quoted 6 lines

Ben.

Quoted 4 lines
arch/x86/mm/gup.c +1 -0
+		return 0;
 	page = head + ((addr & ~PUD_MASK) >> PAGE_SHIFT);
 	do {
 		VM_BUG_ON_PAGE(compound_head(page) != head, page);
maintainer Applied
#6 lore raw
In reply to Vlastimil Babka
Quoted 5 lines · Vlastimil Babka

This collection of patches add the missing overflow checks in arch-specific gup.c variants for x86 and s390. Those were missed in backport of 8fde12ca79af ("mm: prevent get_user_pages() from overflowing page refcount") as mainline had a single gup.c implementation at that point. See individual patches for details.

Queued up, thanks!

Signature · 2 lines
Thanks,
Sasha
#7 lore raw
Quoted 8 lines · Sasha Levin
Quoted 5 lines · Vlastimil Babka

This collection of patches add the missing overflow checks in arch-specific gup.c variants for x86 and s390. Those were missed in backport of 8fde12ca79af ("mm: prevent get_user_pages() from overflowing page refcount") as mainline had a single gup.c implementation at that point. See individual patches for details.

Queued up, thanks!

Please replace the 4.9 version with the following fixed one, thanks to Ben:

----8<----
From fe7f18bd152094f8516d79e847fcb5453a6f8368 Mon Sep 17 00:00:00 2001
From: Vlastimil Babka <[email protected]>
Date: Wed, 6 Nov 2019 16:32:57 +0100
Subject: [PATCH] mm, gup: add missing refcount overflow checks on x86 and s390

The mainline commit 8fde12ca79af ("mm: prevent get_user_pages() from overflowing page refcount") was backported to 4.9.y stable as commit 2ed768cfd895. The backport however missed that in 4.9, there are several arch-specific gup.c versions with fast gup implementations, so these do not prevent refcount overflow.

This is partially fixed for x86 in stable-only commit d73af79742e7 ("x86, mm,
gup: prevent get_page() race with munmap in paravirt guest"). This stable-only commit adds missing parts to x86 version, as well as s390 version, both taken from the SUSE SLES/openSUSE 4.12-based kernels.

The remaining architectures with own gup.c are sparc, mips, sh. It's unlikely the known overflow scenario based on FUSE, which needs 140GB of RAM, is a problem for those architectures, and I don't feel confident enough to patch them.

---
 arch/s390/mm/gup.c | 9 ++++++---
 arch/x86/mm/gup.c  | 9 ++++++++-
 2 files changed, 14 insertions(+), 4 deletions(-)
arch/s390/mm/gup.c +6 -3
index 97fc449a7470..cf045f56581e 100644
@@ -38,7 +38,8 @@ static inline int gup_pte_range(pmd_t *pmdp, pmd_t pmd, unsigned long addr,
 		VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
 		page = pte_page(pte);
 		head = compound_head(page);
-		if (!page_cache_get_speculative(head))
+		if (WARN_ON_ONCE(page_ref_count(head) < 0)
+		    || !page_cache_get_speculative(head))
 			return 0;
 		if (unlikely(pte_val(pte) != pte_val(*ptep))) {
 			put_page(head);
@@ -76,7 +77,8 @@ static inline int gup_huge_pmd(pmd_t *pmdp, pmd_t pmd, unsigned long addr,
 		refs++;
 	} while (addr += PAGE_SIZE, addr != end);
 
-	if (!page_cache_add_speculative(head, refs)) {
+	if (WARN_ON_ONCE(page_ref_count(head) < 0)
+	    || !page_cache_add_speculative(head, refs)) {
 		*nr -= refs;
 		return 0;
 	}
@@ -150,7 +152,8 @@ static int gup_huge_pud(pud_t *pudp, pud_t pud, unsigned long addr,
 		refs++;
 	} while (addr += PAGE_SIZE, addr != end);
 
-	if (!page_cache_add_speculative(head, refs)) {
+	if (WARN_ON_ONCE(page_ref_count(head) < 0)
+	    || !page_cache_add_speculative(head, refs)) {
 		*nr -= refs;
 		return 0;
 	}
arch/x86/mm/gup.c +8 -1
index d7db45bdfb3b..82f727fbbbd2 100644
@@ -202,9 +202,12 @@ static int __gup_device_huge_pmd(pmd_t pmd, unsigned long addr,
 			undo_dev_pagemap(nr, nr_start, pages);
 			return 0;
 		}
+		if (unlikely(!try_get_page(page))) {
+			put_dev_pagemap(pgmap);
+			return 0;
+		}
 		SetPageReferenced(page);
 		pages[*nr] = page;
-		get_page(page);
 		put_dev_pagemap(pgmap);
 		(*nr)++;
 		pfn++;
@@ -230,6 +233,8 @@ static noinline int gup_huge_pmd(pmd_t pmd, unsigned long addr,
 
 	refs = 0;
 	head = pmd_page(pmd);
+	if (WARN_ON_ONCE(page_ref_count(head) <= 0))
+		return 0;
 	page = head + ((addr & ~PMD_MASK) >> PAGE_SHIFT);
 	do {
 		VM_BUG_ON_PAGE(compound_head(page) != head, page);
@@ -289,6 +294,8 @@ static noinline int gup_huge_pud(pud_t pud, unsigned long addr,
 
 	refs = 0;
 	head = pud_page(pud);
+	if (WARN_ON_ONCE(page_ref_count(head) <= 0))
+		return 0;
 	page = head + ((addr & ~PUD_MASK) >> PAGE_SHIFT);
 	do {
 		VM_BUG_ON_PAGE(compound_head(page) != head, page);
Signature · 1 line
2.24.0
maintainer
#8 lore raw
Quoted 115 lines · Vlastimil Babka
Quoted 8 lines · Sasha Levin
Quoted 5 lines · Vlastimil Babka

This collection of patches add the missing overflow checks in arch-specific gup.c variants for x86 and s390. Those were missed in backport of 8fde12ca79af ("mm: prevent get_user_pages() from overflowing page refcount") as mainline had a single gup.c implementation at that point. See individual patches for details.

Queued up, thanks!

Please replace the 4.9 version with the following fixed one, thanks to Ben:

----8<----

Quoted 1 line

From fe7f18bd152094f8516d79e847fcb5453a6f8368 Mon Sep 17 00:00:00 2001

From: Vlastimil Babka <[email protected]>
Date: Wed, 6 Nov 2019 16:32:57 +0100
Subject: [PATCH] mm, gup: add missing refcount overflow checks on x86 and s390

The mainline commit 8fde12ca79af ("mm: prevent get_user_pages() from overflowing page refcount") was backported to 4.9.y stable as commit 2ed768cfd895. The backport however missed that in 4.9, there are several arch-specific gup.c versions with fast gup implementations, so these do not prevent refcount overflow.

This is partially fixed for x86 in stable-only commit d73af79742e7 ("x86, mm,
gup: prevent get_page() race with munmap in paravirt guest"). This stable-only commit adds missing parts to x86 version, as well as s390 version, both taken from the SUSE SLES/openSUSE 4.12-based kernels.

The remaining architectures with own gup.c are sparc, mips, sh. It's unlikely the known overflow scenario based on FUSE, which needs 140GB of RAM, is a problem for those architectures, and I don't feel confident enough to patch them.

---
 arch/s390/mm/gup.c | 9 ++++++---
 arch/x86/mm/gup.c  | 9 ++++++++-
 2 files changed, 14 insertions(+), 4 deletions(-)
arch/s390/mm/gup.c +6 -3
index 97fc449a7470..cf045f56581e 100644
@@ -38,7 +38,8 @@ static inline int gup_pte_range(pmd_t *pmdp, pmd_t pmd, unsigned long addr,
 		VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
 		page = pte_page(pte);
 		head = compound_head(page);
-		if (!page_cache_get_speculative(head))
+		if (WARN_ON_ONCE(page_ref_count(head) < 0)
+		    || !page_cache_get_speculative(head))
 			return 0;
 		if (unlikely(pte_val(pte) != pte_val(*ptep))) {
 			put_page(head);
@@ -76,7 +77,8 @@ static inline int gup_huge_pmd(pmd_t *pmdp, pmd_t pmd, unsigned long addr,
 		refs++;
 	} while (addr += PAGE_SIZE, addr != end);
 
-	if (!page_cache_add_speculative(head, refs)) {
+	if (WARN_ON_ONCE(page_ref_count(head) < 0)
+	    || !page_cache_add_speculative(head, refs)) {
 		*nr -= refs;
 		return 0;
 	}
@@ -150,7 +152,8 @@ static int gup_huge_pud(pud_t *pudp, pud_t pud, unsigned long addr,
 		refs++;
 	} while (addr += PAGE_SIZE, addr != end);
 
-	if (!page_cache_add_speculative(head, refs)) {
+	if (WARN_ON_ONCE(page_ref_count(head) < 0)
+	    || !page_cache_add_speculative(head, refs)) {
 		*nr -= refs;
 		return 0;
 	}
arch/x86/mm/gup.c +8 -1
index d7db45bdfb3b..82f727fbbbd2 100644
@@ -202,9 +202,12 @@ static int __gup_device_huge_pmd(pmd_t pmd, unsigned long addr,
 			undo_dev_pagemap(nr, nr_start, pages);
 			return 0;
 		}
+		if (unlikely(!try_get_page(page))) {
+			put_dev_pagemap(pgmap);
+			return 0;
+		}
 		SetPageReferenced(page);
 		pages[*nr] = page;
-		get_page(page);
 		put_dev_pagemap(pgmap);
 		(*nr)++;
 		pfn++;
@@ -230,6 +233,8 @@ static noinline int gup_huge_pmd(pmd_t pmd, unsigned long addr,
 
 	refs = 0;
 	head = pmd_page(pmd);
+	if (WARN_ON_ONCE(page_ref_count(head) <= 0))
+		return 0;
 	page = head + ((addr & ~PMD_MASK) >> PAGE_SHIFT);
 	do {
 		VM_BUG_ON_PAGE(compound_head(page) != head, page);
@@ -289,6 +294,8 @@ static noinline int gup_huge_pud(pud_t pud, unsigned long addr,
 
 	refs = 0;
 	head = pud_page(pud);
+	if (WARN_ON_ONCE(page_ref_count(head) <= 0))
+		return 0;
 	page = head + ((addr & ~PUD_MASK) >> PAGE_SHIFT);
 	do {
 		VM_BUG_ON_PAGE(compound_head(page) != head, page);
Signature · 1 line
2.24.0

Now updated, sorry I missed this earlier.

greg k-h