Skip to contentSkip to conversation

kernel/sysctl.c: fix out of bounds access in fs.file-max

PATCH In review Matteo Croce 4 participants 11 messages Mar 28, 2019 – Apr 4, 2019

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 kernel/sysctl.c: fix out of bounds access in fs.file-max

fs.file-max sysctl uses proc_doulongvec_minmax() as proc handler, which accesses *extra1 and *extra2 as unsigned long, but commit 32a5ad9c2285 ("sysctl: handle overflow for file-max") assigns &zero, which is an int, to extra1, generating the following KASAN report.
Fix this by changing 'zero' to long, which does not need to be duplicated like 'one' and 'one_ul' for two data types.

==================================================================

BUG: KASAN: global-out-of-bounds in __do_proc_doulongvec_minmax+0x2f9/0x600
Read of size 8 at addr ffffffff8233dc20 by task systemd/1

CPU: 0 PID: 1 Comm: systemd Not tainted 5.1.0-rc2-kvm+ #22
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS ?-20180724_192412-buildhw-07.phx2.fedoraproject.org-1.fc29 04/01/2014
Call Trace:
 print_address_description+0x67/0x23d
 kasan_report.cold.3+0x1c/0x36
 __do_proc_doulongvec_minmax+0x2f9/0x600
 proc_doulongvec_minmax+0x3a/0x50
 proc_sys_call_handler+0x11d/0x170
 vfs_write+0xd7/0x200
 ksys_write+0x93/0x110
 do_syscall_64+0x57/0x140
 entry_SYSCALL_64_after_hwframe+0x44/0xa9
RIP: 0033:0x7f67d33e8804
Code: 00 f7 d8 64 89 02 48 c7 c0 ff ff ff ff eb b3 0f 1f 80 00 00 00 00 48 8d 05 f9 5e 0d 00 8b 00 85 c0 75 13 b8 01 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 54 c3 0f 1f 00 41 54 49 89 d4 55 48 89 f5 53
RSP: 002b:00007fffd9992ed8 EFLAGS: 00000246 ORIG_RAX: 0000000000000001
RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f67d33e8804
RDX: 0000000000000015 RSI: 00005586ce2607b0 RDI: 0000000000000004
RBP: 00007fffd9992f30 R08: 000000000000c0c0 R09: ffffffffffff0000
R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000004
R13: 0000000000000015 R14: 00005586ce2607c4 R15: 00007fffd9992f70

The buggy address belongs to the variable:
 0xffffffff8233dc20

Memory state around the buggy address:
 ffffffff8233db00: 00 00 00 00 00 00 00 00 fa fa fa fa 04 fa fa fa
 ffffffff8233db80: fa fa fa fa 04 fa fa fa fa fa fa fa 04 fa fa fa
>ffffffff8233dc00: fa fa fa fa 04 fa fa fa fa fa fa fa 00 00 00 00
                               ^
 ffffffff8233dc80: fa fa fa fa fa fa fa fa 00 00 00 00 00 00 00 00
 ffffffff8233dd00: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
==================================================================
Fixes 32a5ad9c2285 ("sysctl: handle overflow for file-max")
---
 kernel/sysctl.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/sysctl.c +1 -1
index e5da394d1ca3..3e959d67d619 100644
@@ -124,7 +124,7 @@ static int sixty = 60;
 
 static int __maybe_unused neg_one = -1;
 
-static int zero;
+static long zero;
 static int __maybe_unused one = 1;
 static int __maybe_unused two = 2;
 static int __maybe_unused four = 4;
Signature · 1 line
2.20.1
#1 lore raw
Quoted 7 lines · Matteo Croce

fs.file-max sysctl uses proc_doulongvec_minmax() as proc handler, which accesses *extra1 and *extra2 as unsigned long, but commit 32a5ad9c2285 ("sysctl: handle overflow for file-max") assigns &zero, which is an int, to extra1, generating the following KASAN report.
Fix this by changing 'zero' to long, which does not need to be duplicated like 'one' and 'one_ul' for two data types.

Hi,

Anyone looked at this patch? Does my fix looks sane?

Regards,
--
Matteo Croce
per aspera ad upstream

#2 lore raw
In reply to Matteo Croce
Quoted 6 lines · Matteo Croce

fs.file-max sysctl uses proc_doulongvec_minmax() as proc handler, which accesses *extra1 and *extra2 as unsigned long, but commit 32a5ad9c2285 ("sysctl: handle overflow for file-max") assigns &zero, which is an int, to extra1, generating the following KASAN report.
Fix this by changing 'zero' to long, which does not need to be duplicated like 'one' and 'one_ul' for two data types.

Yeah, maybe but it still feels cleaner and more obvious to just add:

static long long_zero;

given that most callers actually seem to want an (unsigned) int.

I don't have a strong opinion though so if others feel that it's just a waste of space consider it acked.

Quoted 38 lines

==================================================================

BUG: KASAN: global-out-of-bounds in __do_proc_doulongvec_minmax+0x2f9/0x600
Read of size 8 at addr ffffffff8233dc20 by task systemd/1

CPU: 0 PID: 1 Comm: systemd Not tainted 5.1.0-rc2-kvm+ #22
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS ?-20180724_192412-buildhw-07.phx2.fedoraproject.org-1.fc29 04/01/2014
Call Trace:
 print_address_description+0x67/0x23d
 kasan_report.cold.3+0x1c/0x36
 __do_proc_doulongvec_minmax+0x2f9/0x600
 proc_doulongvec_minmax+0x3a/0x50
 proc_sys_call_handler+0x11d/0x170
 vfs_write+0xd7/0x200
 ksys_write+0x93/0x110
 do_syscall_64+0x57/0x140
 entry_SYSCALL_64_after_hwframe+0x44/0xa9
RIP: 0033:0x7f67d33e8804
Code: 00 f7 d8 64 89 02 48 c7 c0 ff ff ff ff eb b3 0f 1f 80 00 00 00 00 48 8d 05 f9 5e 0d 00 8b 00 85 c0 75 13 b8 01 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 54 c3 0f 1f 00 41 54 49 89 d4 55 48 89 f5 53
RSP: 002b:00007fffd9992ed8 EFLAGS: 00000246 ORIG_RAX: 0000000000000001
RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f67d33e8804
RDX: 0000000000000015 RSI: 00005586ce2607b0 RDI: 0000000000000004
RBP: 00007fffd9992f30 R08: 000000000000c0c0 R09: ffffffffffff0000
R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000004
R13: 0000000000000015 R14: 00005586ce2607c4 R15: 00007fffd9992f70

The buggy address belongs to the variable:
 0xffffffff8233dc20

Memory state around the buggy address:
 ffffffff8233db00: 00 00 00 00 00 00 00 00 fa fa fa fa 04 fa fa fa
 ffffffff8233db80: fa fa fa fa 04 fa fa fa fa fa fa fa 04 fa fa fa
>ffffffff8233dc00: fa fa fa fa 04 fa fa fa fa fa fa fa 00 00 00 00
                               ^
 ffffffff8233dc80: fa fa fa fa fa fa fa fa 00 00 00 00 00 00 00 00
 ffffffff8233dd00: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
==================================================================
Quoted 1 line
Fixes 32a5ad9c2285 ("sysctl: handle overflow for file-max")

Next time, please take the time to Cc the author of the Fixes patch as well whose commit this is fixing right away.

Quoted 22 lines
---
 kernel/sysctl.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/sysctl.c +1 -1
index e5da394d1ca3..3e959d67d619 100644
@@ -124,7 +124,7 @@ static int sixty = 60;
 
 static int __maybe_unused neg_one = -1;
 
-static int zero;
+static long zero;
 
 static int __maybe_unused one = 1;
 static int __maybe_unused two = 2;

static int __maybe_unused four = 4;

Signature · 1 line
2.20.1
#3 lore raw
Quoted 9 lines · Christian Brauner
Quoted 6 lines · Matteo Croce

fs.file-max sysctl uses proc_doulongvec_minmax() as proc handler, which accesses *extra1 and *extra2 as unsigned long, but commit 32a5ad9c2285 ("sysctl: handle overflow for file-max") assigns &zero, which is an int, to extra1, generating the following KASAN report.
Fix this by changing 'zero' to long, which does not need to be duplicated like 'one' and 'one_ul' for two data types.

Quoted 9 lines

Yeah, maybe but it still feels cleaner and more obvious to just add:

static long long_zero;

given that most callers actually seem to want an (unsigned) int.

I don't have a strong opinion though so if others feel that it's just a waste of space consider it acked.

Well, given that the value is zero, in this expectional case we could avoid duplicating the symbol and save 4 bytes.
What the maintainers think?

Cheers,

Signature · 2 lines
Matteo Croce
per aspera ad upstream
#4 lore raw
Quoted 7 lines · Matteo Croce
Quoted 6 lines · Christian Brauner

Yeah, maybe but it still feels cleaner and more obvious to just add:

static long long_zero;

given that most callers actually seem to want an (unsigned) int.

Quoted 7 lines
Quoted 3 lines

I don't have a strong opinion though so if others feel that it's just a waste of space consider it acked.

Well, given that the value is zero, in this expectional case we could avoid duplicating the symbol and save 4 bytes.
What the maintainers think?

If we care about saving four bytes, we could just pass the address of ZERO_PAGE(0).

#5 lore raw
Quoted 13 lines · Matthew Wilcox
Quoted 11 lines · Matteo Croce
Quoted 9 lines · Christian Brauner

Yeah, maybe but it still feels cleaner and more obvious to just add:

static long long_zero;

given that most callers actually seem to want an (unsigned) int.

I don't have a strong opinion though so if others feel that it's just a waste of space consider it acked.

Quoted 6 lines
Quoted 3 lines

Well, given that the value is zero, in this expectional case we could avoid duplicating the symbol and save 4 bytes.
What the maintainers think?

If we care about saving four bytes, we could just pass the address of ZERO_PAGE(0).

That would work, work too, maybe it's a bit overkill.
int zero is always there and it's static, so enlarging it to long should be a straightforward fix.
Obviously we can't do it for other numbers, but we can alias it just for the zero case..

Regards,

--
Matteo Croce
per aspera ad upstream

#6 lore raw
Quoted 28 lines · Matteo Croce
Quoted 13 lines · Matthew Wilcox
Quoted 11 lines · Matteo Croce
Quoted 9 lines · Christian Brauner

Yeah, maybe but it still feels cleaner and more obvious to just add:

static long long_zero;

given that most callers actually seem to want an (unsigned) int.

I don't have a strong opinion though so if others feel that it's just a waste of space consider it acked.

Quoted 6 lines
Quoted 3 lines

Well, given that the value is zero, in this expectional case we could avoid duplicating the symbol and save 4 bytes.
What the maintainers think?

If we care about saving four bytes, we could just pass the address of ZERO_PAGE(0).

That would work, work too, maybe it's a bit overkill.
int zero is always there and it's static, so enlarging it to long should be a straightforward fix.
Obviously we can't do it for other numbers, but we can alias it just for the zero case..

Quoted 5 lines

Regards,

--
Matteo Croce
per aspera ad upstream

Anyway, I'm fine with both solutions, as I have other patches in the queue which depends on this fix.

Regards,

Signature · 2 lines
Matteo Croce
per aspera ad upstream
#7 lore raw
Quoted 29 lines · Matteo Croce
Quoted 28 lines · Matteo Croce
Quoted 13 lines · Matthew Wilcox
Quoted 11 lines · Matteo Croce
Quoted 9 lines · Christian Brauner

Yeah, maybe but it still feels cleaner and more obvious to just add:

static long long_zero;

given that most callers actually seem to want an (unsigned) int.

I don't have a strong opinion though so if others feel that it's just a waste of space consider it acked.

Quoted 6 lines
Quoted 3 lines

Well, given that the value is zero, in this expectional case we could avoid duplicating the symbol and save 4 bytes.
What the maintainers think?

If we care about saving four bytes, we could just pass the address of ZERO_PAGE(0).

That would work, work too, maybe it's a bit overkill.
int zero is always there and it's static, so enlarging it to long should be a straightforward fix.
Obviously we can't do it for other numbers, but we can alias it just for the zero case..

Quoted 7 lines
Quoted 5 lines

Regards,

--
Matteo Croce
per aspera ad upstream

Anyway, I'm fine with both solutions, as I have other patches in the

I think Matthew's idea gets us best of both worlds so I'd suggest to use it and resend the patch. You likely want to Cc [email protected] since the original patch this fixes got backported by Greg quite a bit since this was a rather long-standing issue. Please also Cc Andrew this time since he's likely going to pick it up.

Thanks for the patch!
Christian

#8 lore raw
Quoted 39 lines · Christian Brauner
Quoted 29 lines · Matteo Croce
Quoted 28 lines · Matteo Croce
Quoted 13 lines · Matthew Wilcox
Quoted 11 lines · Matteo Croce
Quoted 9 lines · Christian Brauner

Yeah, maybe but it still feels cleaner and more obvious to just add:

static long long_zero;

given that most callers actually seem to want an (unsigned) int.

I don't have a strong opinion though so if others feel that it's just a waste of space consider it acked.

Quoted 6 lines
Quoted 3 lines

Well, given that the value is zero, in this expectional case we could avoid duplicating the symbol and save 4 bytes.
What the maintainers think?

If we care about saving four bytes, we could just pass the address of ZERO_PAGE(0).

That would work, work too, maybe it's a bit overkill.
int zero is always there and it's static, so enlarging it to long should be a straightforward fix.
Obviously we can't do it for other numbers, but we can alias it just for the zero case..

Quoted 7 lines
Quoted 5 lines

Regards,

--
Matteo Croce
per aspera ad upstream

Anyway, I'm fine with both solutions, as I have other patches in the

Quoted 8 lines

I think Matthew's idea gets us best of both worlds so I'd suggest to use it and resend the patch. You likely want to Cc [email protected] since the original patch this fixes got backported by Greg quite a bit since this was a rather long-standing issue. Please also Cc Andrew this time since he's likely going to pick it up.

Thanks for the patch!
Christian

So you mean using page_address(ZERO_PAGE(0)) ?
The idea is nice, but since struct ctl_table kern_table is declared as global variable, how can I assign it to the structure?
GCC complains about 'initializer element is not constant', and ZERO_PAGE(0)->virtual only works if WANT_PAGE_VIRTUAL.

Anyway, I'm preparing a treewide patch to move all "zero", "one" and "int_max" to three single, const variables in fs/proc/proc_sysctl.c as there are 200+ occourrences of them, so I'd rather keep this simple to have it easily backported to stable.

Signature · 2 lines
Matteo Croce
per aspera ad upstream
Acked-by on v1 1/1
#9 lore raw
In reply to Matteo Croce
Quoted 56 lines · Matteo Croce

fs.file-max sysctl uses proc_doulongvec_minmax() as proc handler, which accesses *extra1 and *extra2 as unsigned long, but commit 32a5ad9c2285 ("sysctl: handle overflow for file-max") assigns &zero, which is an int, to extra1, generating the following KASAN report.
Fix this by changing 'zero' to long, which does not need to be duplicated like 'one' and 'one_ul' for two data types.

==================================================================

BUG: KASAN: global-out-of-bounds in __do_proc_doulongvec_minmax+0x2f9/0x600
Read of size 8 at addr ffffffff8233dc20 by task systemd/1

CPU: 0 PID: 1 Comm: systemd Not tainted 5.1.0-rc2-kvm+ #22
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS ?-20180724_192412-buildhw-07.phx2.fedoraproject.org-1.fc29 04/01/2014
Call Trace:
 print_address_description+0x67/0x23d
 kasan_report.cold.3+0x1c/0x36
 __do_proc_doulongvec_minmax+0x2f9/0x600
 proc_doulongvec_minmax+0x3a/0x50
 proc_sys_call_handler+0x11d/0x170
 vfs_write+0xd7/0x200
 ksys_write+0x93/0x110
 do_syscall_64+0x57/0x140
 entry_SYSCALL_64_after_hwframe+0x44/0xa9
RIP: 0033:0x7f67d33e8804
Code: 00 f7 d8 64 89 02 48 c7 c0 ff ff ff ff eb b3 0f 1f 80 00 00 00 00 48 8d 05 f9 5e 0d 00 8b 00 85 c0 75 13 b8 01 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 54 c3 0f 1f 00 41 54 49 89 d4 55 48 89 f5 53
RSP: 002b:00007fffd9992ed8 EFLAGS: 00000246 ORIG_RAX: 0000000000000001
RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f67d33e8804
RDX: 0000000000000015 RSI: 00005586ce2607b0 RDI: 0000000000000004
RBP: 00007fffd9992f30 R08: 000000000000c0c0 R09: ffffffffffff0000
R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000004
R13: 0000000000000015 R14: 00005586ce2607c4 R15: 00007fffd9992f70

The buggy address belongs to the variable:
 0xffffffff8233dc20

Memory state around the buggy address:
 ffffffff8233db00: 00 00 00 00 00 00 00 00 fa fa fa fa 04 fa fa fa
 ffffffff8233db80: fa fa fa fa 04 fa fa fa fa fa fa fa 04 fa fa fa
>ffffffff8233dc00: fa fa fa fa 04 fa fa fa fa fa fa fa 00 00 00 00
                               ^
 ffffffff8233dc80: fa fa fa fa fa fa fa fa 00 00 00 00 00 00 00 00
 ffffffff8233dd00: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
==================================================================
Fixes 32a5ad9c2285 ("sysctl: handle overflow for file-max")
---
 kernel/sysctl.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/sysctl.c +0 -0
index e5da394d1ca3..3e959d67d619 100644

kernel/sysctl.c static int sixty = 60; · line 130

Quoted 8 lines
kernel/sysctl.c +1 -1
@@ -124,7 +124,7 @@ static int sixty = 60;
 
 static int __maybe_unused neg_one = -1;
 
-static int zero;
+static long zero;
 static int __maybe_unused one = 1;
 static int __maybe_unused two = 2;
 static int __maybe_unused four = 4;

This seems okay to me; thanks for the fix! (I think it's fine to keep this merged instead of a distinct long_zero, as long as we're not seeing type warnings during the build.)

-Kees

Signature · 1 line
Kees Cook
#10 lore raw
Quoted 71 lines · Kees Cook
Quoted 56 lines · Matteo Croce

fs.file-max sysctl uses proc_doulongvec_minmax() as proc handler, which accesses *extra1 and *extra2 as unsigned long, but commit 32a5ad9c2285 ("sysctl: handle overflow for file-max") assigns &zero, which is an int, to extra1, generating the following KASAN report.
Fix this by changing 'zero' to long, which does not need to be duplicated like 'one' and 'one_ul' for two data types.

==================================================================

BUG: KASAN: global-out-of-bounds in __do_proc_doulongvec_minmax+0x2f9/0x600
Read of size 8 at addr ffffffff8233dc20 by task systemd/1

CPU: 0 PID: 1 Comm: systemd Not tainted 5.1.0-rc2-kvm+ #22
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS ?-20180724_192412-buildhw-07.phx2.fedoraproject.org-1.fc29 04/01/2014
Call Trace:
 print_address_description+0x67/0x23d
 kasan_report.cold.3+0x1c/0x36
 __do_proc_doulongvec_minmax+0x2f9/0x600
 proc_doulongvec_minmax+0x3a/0x50
 proc_sys_call_handler+0x11d/0x170
 vfs_write+0xd7/0x200
 ksys_write+0x93/0x110
 do_syscall_64+0x57/0x140
 entry_SYSCALL_64_after_hwframe+0x44/0xa9
RIP: 0033:0x7f67d33e8804
Code: 00 f7 d8 64 89 02 48 c7 c0 ff ff ff ff eb b3 0f 1f 80 00 00 00 00 48 8d 05 f9 5e 0d 00 8b 00 85 c0 75 13 b8 01 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 54 c3 0f 1f 00 41 54 49 89 d4 55 48 89 f5 53
RSP: 002b:00007fffd9992ed8 EFLAGS: 00000246 ORIG_RAX: 0000000000000001
RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f67d33e8804
RDX: 0000000000000015 RSI: 00005586ce2607b0 RDI: 0000000000000004
RBP: 00007fffd9992f30 R08: 000000000000c0c0 R09: ffffffffffff0000
R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000004
R13: 0000000000000015 R14: 00005586ce2607c4 R15: 00007fffd9992f70

The buggy address belongs to the variable:
 0xffffffff8233dc20

Memory state around the buggy address:
 ffffffff8233db00: 00 00 00 00 00 00 00 00 fa fa fa fa 04 fa fa fa
 ffffffff8233db80: fa fa fa fa 04 fa fa fa fa fa fa fa 04 fa fa fa
>ffffffff8233dc00: fa fa fa fa 04 fa fa fa fa fa fa fa 00 00 00 00
                               ^
 ffffffff8233dc80: fa fa fa fa fa fa fa fa 00 00 00 00 00 00 00 00
 ffffffff8233dd00: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
==================================================================
Fixes 32a5ad9c2285 ("sysctl: handle overflow for file-max")
---
 kernel/sysctl.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/sysctl.c +0 -0
index e5da394d1ca3..3e959d67d619 100644

kernel/sysctl.c static int sixty = 60; · line 130

Quoted 8 lines
kernel/sysctl.c +1 -1
@@ -124,7 +124,7 @@ static int sixty = 60;
 
 static int __maybe_unused neg_one = -1;
 
-static int zero;
+static long zero;
 static int __maybe_unused one = 1;
 static int __maybe_unused two = 2;
 static int __maybe_unused four = 4;

This seems okay to me; thanks for the fix! (I think it's fine to keep this merged instead of a distinct long_zero, as long as we're not seeing type warnings during the build.)

Quoted 6 lines

-Kees

--
Kees Cook

No warnings with gcc version 8.3.1 20190223

CC kernel/sysctl.o
AR kernel/built-in.a
...

Signature · 2 lines
Matteo Croce
per aspera ad upstream