syzbot detected a case where the page table counters were not properly
updated.
syzkaller login: ------------[ cut here ]------------
kernel BUG at mm/page_table_check.c:162!
invalid opcode: 0000 [#1] PREEMPT SMP KASAN
CPU: 0 PID: 3099 Comm: pasha Not tainted 5.16.0+ #48
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIO4
RIP: 0010:__page_table_check_zero+0x159/0x1a0
Code: 7d 3a b2 ff 45 39 f5 74 2a e8 43 38 b2 ff 4d 85 e4 01
RSP: 0018:ffff888010667418 EFLAGS: 00010293
RAX: 0000000000000000 RBX: 0000000000000001 RCX: 0000000000
RDX: ffff88800cea8680 RSI: ffffffff81becaf9 RDI: 0000000003
RBP: ffff888010667450 R08: 0000000000000001 R09: 0000000000
R10: ffffffff81becaab R11: 0000000000000001 R12: ffff888008
R13: 0000000000000001 R14: 0000000000000200 R15: dffffc0000
FS: 0000000000000000(0000) GS:ffff888035e00000(0000) knlG0
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00007ffd875cad00 CR3: 00000000094ce000 CR4: 0000000000
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000
Call Trace:
<TASK>
free_pcp_prepare+0x3be/0xaa0
free_unref_page+0x1c/0x650
? trace_hardirqs_on+0x6a/0x1d0
free_compound_page+0xec/0x130
free_transhuge_page+0x1be/0x260
__put_compound_page+0x90/0xd0
release_pages+0x54c/0x1060
? filemap_remove_folio+0x161/0x210
? lock_downgrade+0x720/0x720
? __put_page+0x150/0x150
? filemap_free_folio+0x164/0x350
__pagevec_release+0x7c/0x110
shmem_undo_range+0x85e/0x1250
...
The repro involved having a huge page that is split due to uprobe event
temporarily replacing one of the pages in the huge page. Later the huge
page was combined again, but the counters were off, as the PTE level
was not properly updated.
Make sure that not only huge page but also small pages are updated when
a new entry is set or cleared.
Fixes: df4e817b7108 ("mm: page table check")
Signed-off-by: Pasha Tatashin <[email protected]>
---
mm/page_table_check.c | 60 ++++++++++++++++++++++++++-----------------
1 file changed, 36 insertions(+), 24 deletions(-)
diff --git a/mm/page_table_check.c b/mm/page_table_check.c
index 7504e7caa2a1..877d967742bc 100644
--- a/mm/page_table_check.c
+++ b/mm/page_table_check.c
@@ -145,6 +145,30 @@ static void page_table_check_set(struct mm_struct *mm, unsigned long addr,
}
}
+static void pte_clear_level(struct mm_struct *mm, unsigned long addr,
+ pte_t *ptep)
+{
+ unsigned long i;
+
+ for (i = 0; i < PTRS_PER_PTE; i++) {
+ __page_table_check_pte_clear(mm, addr, *ptep);
+ addr += PAGE_SIZE;
+ ptep++;
+ }
+}
+
+static void pmd_clear_level(struct mm_struct *mm, unsigned long addr,
+ pmd_t *pmdp)
+{
+ unsigned long i;
+
+ for (i = 0; i < PTRS_PER_PMD; i++) {
+ __page_table_check_pmd_clear(mm, addr, *pmdp);
+ addr += PMD_PAGE_SIZE;
+ pmdp++;
+ }
+}
+
/*
* page is on free list, or is being allocated, verify that counters are zeroes
* crash if they are not.
@@ -186,6 +210,11 @@ void __page_table_check_pmd_clear(struct mm_struct *mm, unsigned long addr,
if (pmd_user_accessible_page(pmd)) {
page_table_check_clear(mm, addr, pmd_pfn(pmd),
PMD_PAGE_SIZE >> PAGE_SHIFT);
+ } else if (!pmd_bad(pmd) && !pmd_leaf(pmd)) {
+ pte_t *ptep = pte_offset_map(&pmd, addr);
+
+ pte_clear_level(mm, addr, ptep);
+ pte_unmap(ptep);
}
}
EXPORT_SYMBOL(__page_table_check_pmd_clear);
@@ -199,6 +228,10 @@ void __page_table_check_pud_clear(struct mm_struct *mm, unsigned long addr,
if (pud_user_accessible_page(pud)) {
page_table_check_clear(mm, addr, pud_pfn(pud),
PUD_PAGE_SIZE >> PAGE_SHIFT);
+ } else if (!pud_bad(pud) && !pud_leaf(pud)) {
+ pmd_t *pmdp = pmd_offset(&pud, addr);
+
+ pmd_clear_level(mm, addr, pmdp);
}
}
EXPORT_SYMBOL(__page_table_check_pud_clear);
@@ -206,17 +239,10 @@ EXPORT_SYMBOL(__page_table_check_pud_clear);
void __page_table_check_pte_set(struct mm_struct *mm, unsigned long addr,
pte_t *ptep, pte_t pte)
{
- pte_t old_pte;
-
if (&init_mm == mm)
return;
- old_pte = *ptep;
- if (pte_user_accessible_page(old_pte)) {
- page_table_check_clear(mm, addr, pte_pfn(old_pte),
- PAGE_SIZE >> PAGE_SHIFT);
- }
-
+ __page_table_check_pte_clear(mm, addr, *ptep);
if (pte_user_accessible_page(pte)) {
page_table_check_set(mm, addr, pte_pfn(pte),
PAGE_SIZE >> PAGE_SHIFT,
@@ -228,17 +254,10 @@ EXPORT_SYMBOL(__page_table_check_pte_set);
void __page_table_check_pmd_set(struct mm_struct *mm, unsigned long addr,
pmd_t *pmdp, pmd_t pmd)
{
- pmd_t old_pmd;
-
if (&init_mm == mm)
return;
- old_pmd = *pmdp;
- if (pmd_user_accessible_page(old_pmd)) {
- page_table_check_clear(mm, addr, pmd_pfn(old_pmd),
- PMD_PAGE_SIZE >> PAGE_SHIFT);
- }
-
+ __page_table_check_pmd_clear(mm, addr, *pmdp);
if (pmd_user_accessible_page(pmd)) {
page_table_check_set(mm, addr, pmd_pfn(pmd),
PMD_PAGE_SIZE >> PAGE_SHIFT,
@@ -250,17 +269,10 @@ EXPORT_SYMBOL(__page_table_check_pmd_set);
void __page_table_check_pud_set(struct mm_struct *mm, unsigned long addr,
pud_t *pudp, pud_t pud)
{
- pud_t old_pud;
-
if (&init_mm == mm)
return;
- old_pud = *pudp;
- if (pud_user_accessible_page(old_pud)) {
- page_table_check_clear(mm, addr, pud_pfn(old_pud),
- PUD_PAGE_SIZE >> PAGE_SHIFT);
- }
-
+ __page_table_check_pud_clear(mm, addr, *pudp);
if (pud_user_accessible_page(pud)) {
page_table_check_set(mm, addr, pud_pfn(pud),
PUD_PAGE_SIZE >> PAGE_SHIFT,
--
2.34.1.703.g22d0c6ccf7-goog
On Thu, Jan 20, 2022 at 11:12 AM Pasha Tatashin
<[email protected]> wrote:
>
> syzbot detected a case where the page table counters were not properly
> updated.
>
> syzkaller login: ------------[ cut here ]------------
> kernel BUG at mm/page_table_check.c:162!
> invalid opcode: 0000 [#1] PREEMPT SMP KASAN
> CPU: 0 PID: 3099 Comm: pasha Not tainted 5.16.0+ #48
> Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIO4
> RIP: 0010:__page_table_check_zero+0x159/0x1a0
> Code: 7d 3a b2 ff 45 39 f5 74 2a e8 43 38 b2 ff 4d 85 e4 01
> RSP: 0018:ffff888010667418 EFLAGS: 00010293
> RAX: 0000000000000000 RBX: 0000000000000001 RCX: 0000000000
> RDX: ffff88800cea8680 RSI: ffffffff81becaf9 RDI: 0000000003
> RBP: ffff888010667450 R08: 0000000000000001 R09: 0000000000
> R10: ffffffff81becaab R11: 0000000000000001 R12: ffff888008
> R13: 0000000000000001 R14: 0000000000000200 R15: dffffc0000
> FS: 0000000000000000(0000) GS:ffff888035e00000(0000) knlG0
> CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> CR2: 00007ffd875cad00 CR3: 00000000094ce000 CR4: 0000000000
> DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000
> DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000
> Call Trace:
> <TASK>
> free_pcp_prepare+0x3be/0xaa0
> free_unref_page+0x1c/0x650
> ? trace_hardirqs_on+0x6a/0x1d0
> free_compound_page+0xec/0x130
> free_transhuge_page+0x1be/0x260
> __put_compound_page+0x90/0xd0
> release_pages+0x54c/0x1060
> ? filemap_remove_folio+0x161/0x210
> ? lock_downgrade+0x720/0x720
> ? __put_page+0x150/0x150
> ? filemap_free_folio+0x164/0x350
> __pagevec_release+0x7c/0x110
> shmem_undo_range+0x85e/0x1250
> ...
>
> The repro involved having a huge page that is split due to uprobe event
> temporarily replacing one of the pages in the huge page. Later the huge
> page was combined again, but the counters were off, as the PTE level
> was not properly updated.
>
> Make sure that not only huge page but also small pages are updated when
> a new entry is set or cleared.
>
> Fixes: df4e817b7108 ("mm: page table check")
>
> Signed-off-by: Pasha Tatashin <[email protected]>
> ---
> mm/page_table_check.c | 60 ++++++++++++++++++++++++++-----------------
> 1 file changed, 36 insertions(+), 24 deletions(-)
>
> diff --git a/mm/page_table_check.c b/mm/page_table_check.c
> index 7504e7caa2a1..877d967742bc 100644
> --- a/mm/page_table_check.c
> +++ b/mm/page_table_check.c
> @@ -145,6 +145,30 @@ static void page_table_check_set(struct mm_struct *mm, unsigned long addr,
> }
> }
>
> +static void pte_clear_level(struct mm_struct *mm, unsigned long addr,
> + pte_t *ptep)
> +{
> + unsigned long i;
> +
> + for (i = 0; i < PTRS_PER_PTE; i++) {
> + __page_table_check_pte_clear(mm, addr, *ptep);
> + addr += PAGE_SIZE;
> + ptep++;
> + }
> +}
> +
> +static void pmd_clear_level(struct mm_struct *mm, unsigned long addr,
> + pmd_t *pmdp)
> +{
> + unsigned long i;
> +
> + for (i = 0; i < PTRS_PER_PMD; i++) {
> + __page_table_check_pmd_clear(mm, addr, *pmdp);
> + addr += PMD_PAGE_SIZE;
> + pmdp++;
> + }
> +}
> +
> /*
> * page is on free list, or is being allocated, verify that counters are zeroes
> * crash if they are not.
> @@ -186,6 +210,11 @@ void __page_table_check_pmd_clear(struct mm_struct *mm, unsigned long addr,
> if (pmd_user_accessible_page(pmd)) {
> page_table_check_clear(mm, addr, pmd_pfn(pmd),
> PMD_PAGE_SIZE >> PAGE_SHIFT);
> + } else if (!pmd_bad(pmd) && !pmd_leaf(pmd)) {
> + pte_t *ptep = pte_offset_map(&pmd, addr);
> +
> + pte_clear_level(mm, addr, ptep);
> + pte_unmap(ptep);
> }
> }
> EXPORT_SYMBOL(__page_table_check_pmd_clear);
> @@ -199,6 +228,10 @@ void __page_table_check_pud_clear(struct mm_struct *mm, unsigned long addr,
> if (pud_user_accessible_page(pud)) {
> page_table_check_clear(mm, addr, pud_pfn(pud),
> PUD_PAGE_SIZE >> PAGE_SHIFT);
> + } else if (!pud_bad(pud) && !pud_leaf(pud)) {
> + pmd_t *pmdp = pmd_offset(&pud, addr);
> +
> + pmd_clear_level(mm, addr, pmdp);
> }
> }
> EXPORT_SYMBOL(__page_table_check_pud_clear);
> @@ -206,17 +239,10 @@ EXPORT_SYMBOL(__page_table_check_pud_clear);
> void __page_table_check_pte_set(struct mm_struct *mm, unsigned long addr,
> pte_t *ptep, pte_t pte)
> {
> - pte_t old_pte;
> -
> if (&init_mm == mm)
> return;
>
> - old_pte = *ptep;
> - if (pte_user_accessible_page(old_pte)) {
> - page_table_check_clear(mm, addr, pte_pfn(old_pte),
> - PAGE_SIZE >> PAGE_SHIFT);
> - }
> -
> + __page_table_check_pte_clear(mm, addr, *ptep);
> if (pte_user_accessible_page(pte)) {
> page_table_check_set(mm, addr, pte_pfn(pte),
> PAGE_SIZE >> PAGE_SHIFT,
> @@ -228,17 +254,10 @@ EXPORT_SYMBOL(__page_table_check_pte_set);
> void __page_table_check_pmd_set(struct mm_struct *mm, unsigned long addr,
> pmd_t *pmdp, pmd_t pmd)
> {
> - pmd_t old_pmd;
> -
> if (&init_mm == mm)
> return;
>
> - old_pmd = *pmdp;
> - if (pmd_user_accessible_page(old_pmd)) {
> - page_table_check_clear(mm, addr, pmd_pfn(old_pmd),
> - PMD_PAGE_SIZE >> PAGE_SHIFT);
> - }
> -
> + __page_table_check_pmd_clear(mm, addr, *pmdp);
> if (pmd_user_accessible_page(pmd)) {
> page_table_check_set(mm, addr, pmd_pfn(pmd),
> PMD_PAGE_SIZE >> PAGE_SHIFT,
> @@ -250,17 +269,10 @@ EXPORT_SYMBOL(__page_table_check_pmd_set);
> void __page_table_check_pud_set(struct mm_struct *mm, unsigned long addr,
> pud_t *pudp, pud_t pud)
> {
> - pud_t old_pud;
> -
> if (&init_mm == mm)
> return;
>
> - old_pud = *pudp;
> - if (pud_user_accessible_page(old_pud)) {
> - page_table_check_clear(mm, addr, pud_pfn(old_pud),
> - PUD_PAGE_SIZE >> PAGE_SHIFT);
> - }
> -
> + __page_table_check_pud_clear(mm, addr, *pudp);
> if (pud_user_accessible_page(pud)) {
> page_table_check_set(mm, addr, pud_pfn(pud),
> PUD_PAGE_SIZE >> PAGE_SHIFT,
> --
> 2.34.1.703.g22d0c6ccf7-goog
>
Reviewed-by: Wei Xu <[email protected]>
On Thu, Jan 20, 2022 at 2:19 PM Wei Xu <[email protected]> wrote:
>
> On Thu, Jan 20, 2022 at 11:12 AM Pasha Tatashin
> <[email protected]> wrote:
> >
> > syzbot detected a case where the page table counters were not properly
> > updated.
> >
> > syzkaller login: ------------[ cut here ]------------
> > kernel BUG at mm/page_table_check.c:162!
> > invalid opcode: 0000 [#1] PREEMPT SMP KASAN
> > CPU: 0 PID: 3099 Comm: pasha Not tainted 5.16.0+ #48
> > Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIO4
> > RIP: 0010:__page_table_check_zero+0x159/0x1a0
> > Code: 7d 3a b2 ff 45 39 f5 74 2a e8 43 38 b2 ff 4d 85 e4 01
> > RSP: 0018:ffff888010667418 EFLAGS: 00010293
> > RAX: 0000000000000000 RBX: 0000000000000001 RCX: 0000000000
> > RDX: ffff88800cea8680 RSI: ffffffff81becaf9 RDI: 0000000003
> > RBP: ffff888010667450 R08: 0000000000000001 R09: 0000000000
> > R10: ffffffff81becaab R11: 0000000000000001 R12: ffff888008
> > R13: 0000000000000001 R14: 0000000000000200 R15: dffffc0000
> > FS: 0000000000000000(0000) GS:ffff888035e00000(0000) knlG0
> > CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > CR2: 00007ffd875cad00 CR3: 00000000094ce000 CR4: 0000000000
> > DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000
> > DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000
> > Call Trace:
> > <TASK>
> > free_pcp_prepare+0x3be/0xaa0
> > free_unref_page+0x1c/0x650
> > ? trace_hardirqs_on+0x6a/0x1d0
> > free_compound_page+0xec/0x130
> > free_transhuge_page+0x1be/0x260
> > __put_compound_page+0x90/0xd0
> > release_pages+0x54c/0x1060
> > ? filemap_remove_folio+0x161/0x210
> > ? lock_downgrade+0x720/0x720
> > ? __put_page+0x150/0x150
> > ? filemap_free_folio+0x164/0x350
> > __pagevec_release+0x7c/0x110
> > shmem_undo_range+0x85e/0x1250
> > ...
> >
> > The repro involved having a huge page that is split due to uprobe event
> > temporarily replacing one of the pages in the huge page. Later the huge
> > page was combined again, but the counters were off, as the PTE level
> > was not properly updated.
> >
> > Make sure that not only huge page but also small pages are updated when
> > a new entry is set or cleared.
> >
> > Fixes: df4e817b7108 ("mm: page table check")
> >
> > Signed-off-by: Pasha Tatashin <[email protected]>
> > ---
> > mm/page_table_check.c | 60 ++++++++++++++++++++++++++-----------------
> > 1 file changed, 36 insertions(+), 24 deletions(-)
> >
> > diff --git a/mm/page_table_check.c b/mm/page_table_check.c
> > index 7504e7caa2a1..877d967742bc 100644
> > --- a/mm/page_table_check.c
> > +++ b/mm/page_table_check.c
> > @@ -145,6 +145,30 @@ static void page_table_check_set(struct mm_struct *mm, unsigned long addr,
> > }
> > }
> >
> > +static void pte_clear_level(struct mm_struct *mm, unsigned long addr,
> > + pte_t *ptep)
> > +{
> > + unsigned long i;
> > +
> > + for (i = 0; i < PTRS_PER_PTE; i++) {
> > + __page_table_check_pte_clear(mm, addr, *ptep);
> > + addr += PAGE_SIZE;
> > + ptep++;
> > + }
> > +}
> > +
> > +static void pmd_clear_level(struct mm_struct *mm, unsigned long addr,
> > + pmd_t *pmdp)
> > +{
> > + unsigned long i;
> > +
> > + for (i = 0; i < PTRS_PER_PMD; i++) {
> > + __page_table_check_pmd_clear(mm, addr, *pmdp);
> > + addr += PMD_PAGE_SIZE;
> > + pmdp++;
> > + }
> > +}
> > +
> > /*
> > * page is on free list, or is being allocated, verify that counters are zeroes
> > * crash if they are not.
> > @@ -186,6 +210,11 @@ void __page_table_check_pmd_clear(struct mm_struct *mm, unsigned long addr,
> > if (pmd_user_accessible_page(pmd)) {
> > page_table_check_clear(mm, addr, pmd_pfn(pmd),
> > PMD_PAGE_SIZE >> PAGE_SHIFT);
> > + } else if (!pmd_bad(pmd) && !pmd_leaf(pmd)) {
> > + pte_t *ptep = pte_offset_map(&pmd, addr);
> > +
> > + pte_clear_level(mm, addr, ptep);
> > + pte_unmap(ptep);
> > }
> > }
> > EXPORT_SYMBOL(__page_table_check_pmd_clear);
> > @@ -199,6 +228,10 @@ void __page_table_check_pud_clear(struct mm_struct *mm, unsigned long addr,
> > if (pud_user_accessible_page(pud)) {
> > page_table_check_clear(mm, addr, pud_pfn(pud),
> > PUD_PAGE_SIZE >> PAGE_SHIFT);
> > + } else if (!pud_bad(pud) && !pud_leaf(pud)) {
> > + pmd_t *pmdp = pmd_offset(&pud, addr);
> > +
> > + pmd_clear_level(mm, addr, pmdp);
> > }
> > }
> > EXPORT_SYMBOL(__page_table_check_pud_clear);
> > @@ -206,17 +239,10 @@ EXPORT_SYMBOL(__page_table_check_pud_clear);
> > void __page_table_check_pte_set(struct mm_struct *mm, unsigned long addr,
> > pte_t *ptep, pte_t pte)
> > {
> > - pte_t old_pte;
> > -
> > if (&init_mm == mm)
> > return;
> >
> > - old_pte = *ptep;
> > - if (pte_user_accessible_page(old_pte)) {
> > - page_table_check_clear(mm, addr, pte_pfn(old_pte),
> > - PAGE_SIZE >> PAGE_SHIFT);
> > - }
> > -
> > + __page_table_check_pte_clear(mm, addr, *ptep);
> > if (pte_user_accessible_page(pte)) {
> > page_table_check_set(mm, addr, pte_pfn(pte),
> > PAGE_SIZE >> PAGE_SHIFT,
> > @@ -228,17 +254,10 @@ EXPORT_SYMBOL(__page_table_check_pte_set);
> > void __page_table_check_pmd_set(struct mm_struct *mm, unsigned long addr,
> > pmd_t *pmdp, pmd_t pmd)
> > {
> > - pmd_t old_pmd;
> > -
> > if (&init_mm == mm)
> > return;
> >
> > - old_pmd = *pmdp;
> > - if (pmd_user_accessible_page(old_pmd)) {
> > - page_table_check_clear(mm, addr, pmd_pfn(old_pmd),
> > - PMD_PAGE_SIZE >> PAGE_SHIFT);
> > - }
> > -
> > + __page_table_check_pmd_clear(mm, addr, *pmdp);
> > if (pmd_user_accessible_page(pmd)) {
> > page_table_check_set(mm, addr, pmd_pfn(pmd),
> > PMD_PAGE_SIZE >> PAGE_SHIFT,
> > @@ -250,17 +269,10 @@ EXPORT_SYMBOL(__page_table_check_pmd_set);
> > void __page_table_check_pud_set(struct mm_struct *mm, unsigned long addr,
> > pud_t *pudp, pud_t pud)
> > {
> > - pud_t old_pud;
> > -
> > if (&init_mm == mm)
> > return;
> >
> > - old_pud = *pudp;
> > - if (pud_user_accessible_page(old_pud)) {
> > - page_table_check_clear(mm, addr, pud_pfn(old_pud),
> > - PUD_PAGE_SIZE >> PAGE_SHIFT);
> > - }
> > -
> > + __page_table_check_pud_clear(mm, addr, *pudp);
> > if (pud_user_accessible_page(pud)) {
> > page_table_check_set(mm, addr, pud_pfn(pud),
> > PUD_PAGE_SIZE >> PAGE_SHIFT,
> > --
> > 2.34.1.703.g22d0c6ccf7-goog
> >
>
> Reviewed-by: Wei Xu <[email protected]>
Andrew, please hold off on taking this patch. During testing, I have
found that the proposed fix does not work with khugepaged: we have
double clearing of PTE: first remove PMD entry and clear PTEs, but
later clear PTEs again in __collapse_huge_page_copy(). I am looking
for a more complete fix.
Thanks,
Pasha