On 2020/2/18 18:21, Ilya Dryomov wrote:
On Tue, Feb 18, 2020 at 4:30 AM <xiubli@xxxxxxxxxx> wrote:
From: Xiubo Li <xiubli@xxxxxxxxxx>
When there has new osdmap comes, it will replace and release
the old osdmap memory, but if the mount is still on the way
without the osdc->lock wrappers, we will hit use after free
bug, like:
<3>[ 3797.775970] BUG: KASAN: use-after-free in __ceph_open_session+0x2a9/0x370 [libceph]
<3>[ 3797.775974] Read of size 4 at addr ffff8883d8b8a110 by task mount.ceph/64782
<3>[ 3797.775975]
<4>[ 3797.775980] CPU: 0 PID: 64782 Comm: mount.ceph Tainted: G E 5.5.0+ #16
<4>[ 3797.775982] Hardware name: VMware, Inc. VMware Virtual Platform/440BX Desktop Reference Platform, BIOS 6.00 05/19/2017
<4>[ 3797.775984] Call Trace:
<4>[ 3797.775992] dump_stack+0x8c/0xc0
<4>[ 3797.775997] print_address_description.constprop.0+0x1b/0x210
<4>[ 3797.776029] ? __ceph_open_session+0x2a9/0x370 [libceph]
<4>[ 3797.776062] ? __ceph_open_session+0x2a9/0x370 [libceph]
<4>[ 3797.776065] __kasan_report.cold+0x1a/0x33
<4>[ 3797.776098] ? __ceph_open_session+0x2a9/0x370 [libceph]
<4>[ 3797.776101] kasan_report+0xe/0x20
<4>[ 3797.776133] __ceph_open_session+0x2a9/0x370 [libceph]
<4>[ 3797.776170] ? ceph_reset_client_addr+0x30/0x30 [libceph]
<4>[ 3797.776173] ? _raw_spin_lock+0x7a/0xd0
<4>[ 3797.776178] ? finish_wait+0x100/0x100
<4>[ 3797.776182] ? __mutex_lock_slowpath+0x10/0x10
<4>[ 3797.776227] ceph_get_tree+0x65b/0xa40 [ceph]
<4>[ 3797.776236] vfs_get_tree+0x46/0x120
<4>[ 3797.776240] do_mount+0xa2c/0xd70
<4>[ 3797.776245] ? __list_add_valid+0x2f/0x60
<4>[ 3797.776249] ? copy_mount_string+0x20/0x20
<4>[ 3797.776254] ? __kasan_kmalloc.constprop.0+0xc2/0xd0
<4>[ 3797.776258] __x64_sys_mount+0xbe/0x100
<4>[ 3797.776263] do_syscall_64+0x73/0x210
<4>[ 3797.776268] entry_SYSCALL_64_after_hwframe+0x44/0xa9
<4>[ 3797.776271] RIP: 0033:0x7f8f026e5b8e
<4>[ 3797.776275] Code: 48 8b 0d fd 42 0c 00 f7 d8 64 89 01 48 83 c8 ff c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa 49 89 ca b8 a5 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d ca 42 0c 00 f7 d8 64 89 01 48
<4>[ 3797.776277] RSP: 002b:00007ffc2d7cccd8 EFLAGS: 00000206 ORIG_RAX: 00000000000000a5
<4>[ 3797.776281] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f8f026e5b8e
<4>[ 3797.776283] RDX: 00005582afb2a558 RSI: 00007ffc2d7cef0d RDI: 00005582b01707a0
<4>[ 3797.776285] RBP: 00007ffc2d7ccda0 R08: 00005582b0173250 R09: 0000000000000067
<4>[ 3797.776287] R10: 0000000000000000 R11: 0000000000000206 R12: 00005582afb043a0
<4>[ 3797.776289] R13: 00007ffc2d7cce80 R14: 0000000000000000 R15: 0000000000000000
<3>[ 3797.776293]
<3>[ 3797.776295] Allocated by task 64782:
<4>[ 3797.776299] save_stack+0x1b/0x80
<4>[ 3797.776302] __kasan_kmalloc.constprop.0+0xc2/0xd0
<4>[ 3797.776336] ceph_osdmap_alloc+0x29/0xd0 [libceph]
<4>[ 3797.776368] ceph_osdc_init+0x1ff/0x490 [libceph]
<4>[ 3797.776399] ceph_create_client+0x154/0x1b0 [libceph]
<4>[ 3797.776427] ceph_get_tree+0x97/0xa40 [ceph]
<4>[ 3797.776430] vfs_get_tree+0x46/0x120
<4>[ 3797.776433] do_mount+0xa2c/0xd70
<4>[ 3797.776436] __x64_sys_mount+0xbe/0x100
<4>[ 3797.776439] do_syscall_64+0x73/0x210
<4>[ 3797.776443] entry_SYSCALL_64_after_hwframe+0x44/0xa9
<3>[ 3797.776443]
<3>[ 3797.776445] Freed by task 55184:
<4>[ 3797.776461] save_stack+0x1b/0x80
<4>[ 3797.776464] __kasan_slab_free+0x12c/0x170
<4>[ 3797.776467] kfree+0xa3/0x290
<4>[ 3797.776500] handle_one_map+0x1f4/0x3c0 [libceph]
<4>[ 3797.776533] ceph_osdc_handle_map+0x910/0xa90 [libceph]
<4>[ 3797.776567] dispatch+0x826/0xde0 [libceph]
<4>[ 3797.776599] ceph_con_workfn+0x18c1/0x3b30 [libceph]
<4>[ 3797.776603] process_one_work+0x3b1/0x6a0
<4>[ 3797.776606] worker_thread+0x78/0x5d0
<4>[ 3797.776609] kthread+0x191/0x1e0
<4>[ 3797.776612] ret_from_fork+0x35/0x40
<3>[ 3797.776613]
<3>[ 3797.776616] The buggy address belongs to the object at ffff8883d8b8a100
<3>[ 3797.776616] which belongs to the cache kmalloc-192 of size 192
<3>[ 3797.776836] The buggy address is located 16 bytes inside of
<3>[ 3797.776836] 192-byte region [ffff8883d8b8a100, ffff8883d8b8a1c0)
<3>[ 3797.776838] The buggy address belongs to the page:
<4>[ 3797.776842] page:ffffea000f62e280 refcount:1 mapcount:0 mapping:ffff8883ec80f000 index:0xffff8883d8b8bf00 compound_mapcount: 0
<4>[ 3797.776847] raw: 0017ffe000010200 ffffea000f6c6780 0000000200000002 ffff8883ec80f000
<4>[ 3797.776851] raw: ffff8883d8b8bf00 000000008020001b 00000001ffffffff 0000000000000000
<4>[ 3797.776852] page dumped because: kasan: bad access detected
<3>[ 3797.776853]
<3>[ 3797.776854] Memory state around the buggy address:
<3>[ 3797.776857] ffff8883d8b8a000: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
<3>[ 3797.776859] ffff8883d8b8a080: 00 00 00 00 00 00 00 00 fc fc fc fc fc fc fc fc
<3>[ 3797.776862] >ffff8883d8b8a100: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
<3>[ 3797.776863] ^
<3>[ 3797.776866] ffff8883d8b8a180: fb fb fb fb fb fb fb fb fc fc fc fc fc fc fc fc
<3>[ 3797.776868] ffff8883d8b8a200: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
<3>[ 3797.776869] ==================================================================
URL: https://tracker.ceph.com/issues/44177
Signed-off-by: Xiubo Li <xiubli@xxxxxxxxxx>
---
net/ceph/ceph_common.c | 8 +++++++-
1 file changed, 7 insertions(+), 1 deletion(-)
diff --git a/net/ceph/ceph_common.c b/net/ceph/ceph_common.c
index a0e97f6c1072..5c6230b56e02 100644
--- a/net/ceph/ceph_common.c
+++ b/net/ceph/ceph_common.c
@@ -682,8 +682,14 @@ EXPORT_SYMBOL(ceph_reset_client_addr);
*/
static bool have_mon_and_osd_map(struct ceph_client *client)
{
+ bool have_osd_map = false;
+
+ down_read(&client->osdc.lock);
+ have_osd_map = !!(client->osdc.osdmap && client->osdc.osdmap->epoch);
+ up_read(&client->osdc.lock);
+
return client->monc.monmap && client->monc.monmap->epoch &&
- client->osdc.osdmap && client->osdc.osdmap->epoch;
+ have_osd_map;
Hi Xiubo,
The monmap pointer is reset the same way, so it needs some locking
as well. And a quick grep shows more places that need to be fixed,
some between libceph and rbd.
I'll treat this patch as a bug report and fix them myself.
Yeah, sure.
Thanks,
Thanks,
Ilya
