Re: Kernel panic (NULL ptr deref?) in find_ge_pid()/next_pidmap() (via sys_getdents or sys_readdir)

158 views
Skip to first unread message

Robert Święcki

unread,
Apr 13, 2011, 5:36:08 PM4/13/11
to linux-...@vger.kernel.org
On Wed, Apr 13, 2011 at 11:12 PM, Robert Święcki <rob...@swiecki.net> wrote:
> Oops: (kdb's dumpall attached)
>
> [18608.476700] general protection fault: 0000 [#1] PREEMPT SMP
> [18608.476704] last sysfs file:
> /sys/devices/platform/microcode/power/runtime_status
> [18608.477002] CPU 0
> [18608.477002] Pid: 31157, comm: iknowthis Not tainted 2.6.39-rc3 #4
> Dell Inc.                 Precision WorkStation 390    /0GH911
> [18608.477002] RIP: 0010:[<ffffffff810ac3cc>]  [<ffffffff810ac3cc>]
> next_pidmap+0x38/0x7f
> [18608.477002] RSP: 0000:ffff88007b40fd48  EFLAGS: 00010203
> [18608.477002] RAX: 0000000000000000 RBX: 001fffff82753988 RCX: 0000000000000034
> [18608.477002] RDX: 0000000000003b6e RSI: 001ffffffff2c980 RDI: ffffffff82827000
> [18608.477002] RBP: ffff88007b40fd68 R08: a000000000000000 R09: 5b68000000000000
> [18608.477002] R10: ffff88007b40e000 R11: ffff88007b40fdb8 R12: ffffffff82827000
> [18608.477002] R13: ffffffff82827808 R14: ffffffff81199146 R15: ffffffff81199146
> [18608.477002] FS:  0000000000000000(0000) GS:ffff88012bc00000(0063)
> knlGS:00000000f75406c0
> [18608.477002] CS:  0010 DS: 002b ES: 002b CR0: 000000008005003b
> [18608.477002] CR2: 000000000809601c CR3: 0000000118457000 CR4: 00000000000006f0
> [18608.477002] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> [18608.477002] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
> [18608.477002] Process iknowthis (pid: 31157, threadinfo
> ffff88007b40e000, task ffff880124f6c560)
> [18608.477002] Stack:
> [18608.477002]  ffff88011e2e5c80 00000000964c3b6d 0000000000000000
> ffffffff82827000
> [18608.477002]  ffff88007b40fd98 ffffffff810ac446 ffff88011e2e5c80
> ffffffff82827000
> [18608.477002]  00000000964c3b6d ffff880124f68000 ffff88007b40fdf8
> ffffffff811aedbb
> [18608.477002] Call Trace:
> [18608.477002]  [<ffffffff810ac446>] find_ge_pid+0x33/0x45
> [18608.477002]  [<ffffffff811aedbb>] next_tgid+0x3c/0x93
> [18608.477002]  [<ffffffff81199146>] ? compat_sys_io_getevents+0xae/0xae
> [18608.477002]  [<ffffffff810abf69>] ? get_pid_task+0x49/0x51
> [18608.477002]  [<ffffffff811aef6f>] proc_pid_readdir+0x15d/0x1b0
> [18608.477002]  [<ffffffff81199146>] ? compat_sys_io_getevents+0xae/0xae
> [18608.477002]  [<ffffffff81199146>] ? compat_sys_io_getevents+0xae/0xae
> [18608.477002]  [<ffffffff81199146>] ? compat_sys_io_getevents+0xae/0xae
> [18608.477002]  [<ffffffff811ab262>] proc_root_readdir+0x43/0x4b
> [18608.477002]  [<ffffffff8116bd57>] vfs_readdir+0x71/0xae
> [18608.477002]  [<ffffffff81197d18>] compat_sys_getdents+0x81/0xcc
> [18608.477002]  [<ffffffff81f26703>] ia32_do_call+0x13/0x13
> [18608.477002] Code: 0f 1f 44 00 00 ff c6 49 89 fc 89 f2 4c 8d af 08
> 08 00 00 48 63 f6 81 e2 ff 7f 00 00 48 c1 ee 0f 48 c1 e6 04 48 8d 5c
> 37 08 eb 36
> [18608.477002]  8b 7b 08 48 85 ff 74 27 48 63 d2 be 00 80 00 00 e8 66 ab 3e
> [18608.477002] RIP  [<ffffffff810ac3cc>] next_pidmap+0x38/0x7f
> [18608.477002]  RSP <ffff88007b40fd48>
> [18626.313012] ---[ end trace 947a53151707da51 ]---
>
> $ ./decodecode </tmp/ooops
> Code: 0f 1f 44 00 00 ff c6 49 89 fc 89 f2 4c 8d af 08 08 00 00 48 63
> f6 81 e2 ff 7f 00 00 48 c1 ee 0f 48 c1 e6 04 48 8d 5c 37 08 eb 36
>
> Code starting with the faulting instruction
> ===========================================
>   0:   0f 1f 44 00 00          nopl   0x0(%rax,%rax,1)
>   5:   ff c6                   inc    %esi
>   7:   49 89 fc                mov    %rdi,%r12
>   a:   89 f2                   mov    %esi,%edx
>   c:   4c 8d af 08 08 00 00    lea    0x808(%rdi),%r13
>  13:   48 63 f6                movslq %esi,%rsi
>  16:   81 e2 ff 7f 00 00       and    $0x7fff,%edx
>  1c:   48 c1 ee 0f             shr    $0xf,%rsi
>  20:   48 c1 e6 04             shl    $0x4,%rsi
>  24:   48 8d 5c 37 08          lea    0x8(%rdi,%rsi,1),%rbx
>  29:   eb 36                   jmp    0x61

Src dump of the RIP address

/vmlinux: file format elf64-x86-64

Disassembly of section .text:

ffffffff810ac394 <next_pidmap>:
}
return -1;
}

int next_pidmap(struct pid_namespace *pid_ns, int last)
{
ffffffff810ac394: 55 push %rbp
ffffffff810ac395: 48 89 e5 mov %rsp,%rbp
ffffffff810ac398: 41 55 push %r13
ffffffff810ac39a: 41 54 push %r12
ffffffff810ac39c: 53 push %rbx
ffffffff810ac39d: 48 83 ec 08 sub $0x8,%rsp
ffffffff810ac3a1: e8 1a 88 e7 00 callq ffffffff81f24bc0 <mcount>
int offset;
struct pidmap *map, *end;

offset = (last + 1) & BITS_PER_PAGE_MASK;
ffffffff810ac3a6: ff c6 inc %esi
}
return -1;
}

int next_pidmap(struct pid_namespace *pid_ns, int last)
{
ffffffff810ac3a8: 49 89 fc mov %rdi,%r12
int offset;
struct pidmap *map, *end;

offset = (last + 1) & BITS_PER_PAGE_MASK;
ffffffff810ac3ab: 89 f2 mov %esi,%edx
map = &pid_ns->pidmap[(last + 1)/BITS_PER_PAGE];
end = &pid_ns->pidmap[PIDMAP_ENTRIES];
ffffffff810ac3ad: 4c 8d af 08 08 00 00 lea 0x808(%rdi),%r13
{
int offset;
struct pidmap *map, *end;

offset = (last + 1) & BITS_PER_PAGE_MASK;
map = &pid_ns->pidmap[(last + 1)/BITS_PER_PAGE];
ffffffff810ac3b4: 48 63 f6 movslq %esi,%rsi
int next_pidmap(struct pid_namespace *pid_ns, int last)
{
int offset;
struct pidmap *map, *end;

offset = (last + 1) & BITS_PER_PAGE_MASK;
ffffffff810ac3b7: 81 e2 ff 7f 00 00 and $0x7fff,%edx
map = &pid_ns->pidmap[(last + 1)/BITS_PER_PAGE];
ffffffff810ac3bd: 48 c1 ee 0f shr $0xf,%rsi
ffffffff810ac3c1: 48 c1 e6 04 shl $0x4,%rsi
ffffffff810ac3c5: 48 8d 5c 37 08 lea 0x8(%rdi,%rsi,1),%rbx
end = &pid_ns->pidmap[PIDMAP_ENTRIES];
for (; map < end; map++, offset = 0) {
ffffffff810ac3ca: eb 36 jmp
ffffffff810ac402 <next_pidmap+0x6e>
if (unlikely(!map->page))
ffffffff810ac3cc: 48 8b 7b 08 mov 0x8(%rbx),%rdi
ffffffff810ac3d0: 48 85 ff test %rdi,%rdi
ffffffff810ac3d3: 74 27 je
ffffffff810ac3fc <next_pidmap+0x68>
continue;
offset = find_next_bit((map)->page, BITS_PER_PAGE, offset);
ffffffff810ac3d5: 48 63 d2 movslq %edx,%rdx
ffffffff810ac3d8: be 00 80 00 00 mov $0x8000,%esi
ffffffff810ac3dd: e8 66 ab 3e 00 callq
ffffffff81496f48 <find_next_bit>
if (offset < BITS_PER_PAGE)
ffffffff810ac3e2: 3d ff 7f 00 00 cmp $0x7fff,%eax
ffffffff810ac3e7: 77 13 ja
ffffffff810ac3fc <next_pidmap+0x68>
#define BITS_PER_PAGE_MASK (BITS_PER_PAGE-1)

static inline int mk_pid(struct pid_namespace *pid_ns,
struct pidmap *map, int off)
{
return (map - pid_ns->pidmap)*BITS_PER_PAGE + off;
ffffffff810ac3e9: 49 83 c4 08 add $0x8,%r12
ffffffff810ac3ed: 4c 29 e3 sub %r12,%rbx
ffffffff810ac3f0: 48 c1 fb 04 sar $0x4,%rbx
ffffffff810ac3f4: c1 e3 0f shl $0xf,%ebx
ffffffff810ac3f7: 8d 04 03 lea (%rbx,%rax,1),%eax
for (; map < end; map++, offset = 0) {
if (unlikely(!map->page))
continue;
offset = find_next_bit((map)->page, BITS_PER_PAGE, offset);
if (offset < BITS_PER_PAGE)
return mk_pid(pid_ns, map, offset);
ffffffff810ac3fa: eb 0e jmp
ffffffff810ac40a <next_pidmap+0x76>
struct pidmap *map, *end;

offset = (last + 1) & BITS_PER_PAGE_MASK;
map = &pid_ns->pidmap[(last + 1)/BITS_PER_PAGE];
end = &pid_ns->pidmap[PIDMAP_ENTRIES];
for (; map < end; map++, offset = 0) {
ffffffff810ac3fc: 48 83 c3 10 add $0x10,%rbx
ffffffff810ac400: 31 d2 xor %edx,%edx
ffffffff810ac402: 4c 39 eb cmp %r13,%rbx
ffffffff810ac405: 72 c5 jb
ffffffff810ac3cc <next_pidmap+0x38>
ffffffff810ac407: 83 c8 ff or $0xffffffffffffffff,%eax
offset = find_next_bit((map)->page, BITS_PER_PAGE, offset);
if (offset < BITS_PER_PAGE)
return mk_pid(pid_ns, map, offset);
}
return -1;
}
ffffffff810ac40a: 41 5b pop %r11
ffffffff810ac40c: 5b pop %rbx
ffffffff810ac40d: 41 5c pop %r12
ffffffff810ac40f: 41 5d pop %r13
ffffffff810ac411: c9 leaveq

--
Robert Święcki
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majo...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Robert Święcki

unread,
Apr 14, 2011, 12:40:06 PM4/14/11
to linux-...@vger.kernel.org

I'm surprised by the discrepancy between this oops, and kdb memory dump:

From oops/kdb (from address 0xffffffff810ac3a1 which is start of next_pidmap)
Code: 0f 1f 44 00 00 ff c6 49 89 fc 89 f2 4c 8d af 08 08 00 00 48....

From objdump vmlinux (the same address)
Code: e8 1a 88 e7 00 ff c6 49 89 fc 89 f2 4c 8d af 08 08 00 00 48...

Which decodes to

kdb:


0: 0f 1f 44 00 00 nopl 0x0(%rax,%rax,1)
5: ff c6 inc %esi
7: 49 89 fc mov %rdi,%r12
a: 89 f2 mov %esi,%edx
c: 4c 8d af 08 08 00 00 lea 0x808(%rdi),%r13

vmlinux:


ffffffff810ac3a1: e8 1a 88 e7 00 callq ffffffff81f24bc0 <mcount>

ffffffff810ac3a6: ff c6 inc %esi

ffffffff810ac3a8: 49 89 fc mov %rdi,%r12

ffffffff810ac3ab: 89 f2 mov %esi,%edx

ffffffff810ac3ad: 4c 8d af 08 08 00 00 lea 0x808(%rdi),%r13


Might this difference (nopl 0x0(%rax,%rax,1) vs callq
ffffffff81f24bc0 <mcount> at 0xffffffff810ac3a1) be some kind of
kernel instrumentation (ftrace, perf or so), or a symptom of a bug
(overwritten memory)?

Robert Święcki

unread,
Apr 18, 2011, 2:54:16 AM4/18/11
to linux-...@vger.kernel.org

And a quick kgdb session:

(gdb) target remote /dev/ttyS0
Remote debugging using /dev/ttyS0
next_pidmap (pid_ns=0xffffffff82827000, last=<value optimized out>) at
kernel/pid.c:229
229 if (unlikely(!map->page))

(gdb) p map
$1 = (struct pidmap *) 0x1fffff82745c68
(gdb) p map->page
Cannot access memory at address 0x1fffff82745c70

(gdb) bt
#0 next_pidmap (pid_ns=0xffffffff82827000, last=<value optimized
out>) at kernel/pid.c:229
#1 0xffffffff810ac446 in find_ge_pid (nr=-1889315394,
ns=0xffffffff82827000) at kernel/pid.c:527
#2 0xffffffff811aedbb in next_tgid (ns=0xffffffff82827000, iter=...)
at fs/proc/base.c:3087
#3 0xffffffff811aef6f in proc_pid_readdir (filp=0xffff880020155840,
dirent=<value optimized out>, filldir=0xffffffff811992c0
<compat_fillonedir>) at fs/proc/base.c:3146
#4 0xffffffff811ab262 in proc_root_readdir (filp=0xffff880020155840,
dirent=0xffff8800c82d1f48, filldir=0xffffffff811992c0
<compat_fillonedir>) at fs/proc/root.c:159
#5 0xffffffff8116bd57 in vfs_readdir (file=0xffff880020155840,
filler=0xffffffff811992c0 <compat_fillonedir>, buf=0xffff8800c82d1f48)
at fs/readdir.c:40
#6 0xffffffff81197da9 in compat_sys_old_readdir (fd=<value optimized
out>, dirent=0xf777a000, count=<value optimized out>) at
fs/compat.c:901
#7 <signal handler called>
#8 0x00000000080658d2 in ?? ()
Cannot access memory at address 0x29d58137

(gdb) up
#1 0xffffffff810ac446 in find_ge_pid (nr=-1889315394,
ns=0xffffffff82827000) at kernel/pid.c:527
527 nr = next_pidmap(ns, nr);

(gdb) p nr
$5 = -1889315394

(gdb) up
#2 0xffffffff811aedbb in next_tgid (ns=0xffffffff82827000, iter=...)
at fs/proc/base.c:3087
3087 pid = find_ge_pid(iter.tgid, ns);

(gdb) p iter
$8 = {tgid = 2405651902, task = 0x0}

Might it be unsigned (2405651902) vs signed (-1889315394) problem? The
map is counted as

map = &pid_ns->pidmap[(last + 1)/BITS_PER_PAGE];

so

Robert Święcki

unread,
Apr 18, 2011, 8:58:12 AM4/18/11
to linux-...@vger.kernel.org, tav...@cmpxchg8b.com, secu...@kernel.org

and a repro - should be quite obvious for FS folks, I guess
filp->f_pos needs better checking in proc's readdir (or in llseek).
Works with at least 2.6.32 - 2.6.39-rc3

=====================================================
// Found by Tavis Ormandy's (tav...@cmpxchg8b.com):
// http://code.google.com/p/iknowthis/
// Analyzed by Robert Swiecki <rob...@swiecki.net>
#define _GNU_SOURCE 1
#define _LARGEFILE64_SOURCE
#include <sys/stat.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <fcntl.h>
#include <unistd.h>
#include <sys/syscall.h>

int main(void)
{
int fd = open("/proc", O_DIRECTORY | O_RDONLY);
if (fd == -1) {
perror("open");
return -1;
}
struct linux_dirent {
long d_ino;
off_t d_off;
unsigned short d_reclen;
char d_name[];
};
lseek64(fd, 4000000000ULL, SEEK_SET);
struct linux_dirent b[100];
syscall(__NR_getdents, fd, b, sizeof(b));
}
=====================================================

Kees Cook

unread,
Apr 18, 2011, 3:37:57 PM4/18/11
to Robert Święcki, linux-...@vger.kernel.org, tav...@cmpxchg8b.com, secu...@kernel.org
Hi Robert,

On Mon, Apr 18, 2011 at 02:57:55PM +0200, Robert Święcki wrote:
> and a repro - should be quite obvious for FS folks, I guess
> filp->f_pos needs better checking in proc's readdir (or in llseek).

Yup:

int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
{
unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
..
for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
..
}

ns = filp->f_dentry->d_sb->s_fs_info;
iter.task = NULL;
iter.tgid = filp->f_pos - TGID_OFFSET;
..

There's no test to validate f_pos. If it's out of bounds, the "for"
just doesn't run.

-Kees

--
Kees Cook
Ubuntu Security Team

Reply all
Reply to author
Forward
0 new messages