IPsec/crypto kernel panic in 2.6.[456]
I get a kernel panic that is very easy to reproduce. The panic goes away if I revert the following changeset ("fix in-place de/encryption bug with highmem"):
http://linux.bkbits.net:8080/linux-2.5/[email protected]
I have not analyzed the code closely, so I cannot say for certain that there is a problem with that specific changeset, and I'm certainly not out to blame anyone, but reverting it at least seems to fix the problem.
The panic occurs when I receive more than a little TCP data in an
IPsec ESP tunnel. I'm of course willing to test patches or provide further info if required.
[Please CC me because I'm not subscribed to the list.]
/Tobias
Unable to handle kernel paging request at virtual address 77d89bb3 printing eip: e38c230d *pde = 00000000 Oops: 0002 [#1] CPU: 0 EIP: 0060:[<e38c230d>] Not tainted EFLAGS: 00010216 (2.6.6)
EIP is at des_small_fips_decrypt+0x96d/0x9a0 [des]
eax: 0030f000 ebx: 00030f00 ecx: 1b24b797 edx: ba4d7f77
esi: dcf91438 edi: 77d89bb3 ebp: c040fbb0 esp: c040fb5c
ds: 007b es: 007b ss: 0068
Process swapper (pid: 0, threadinfo=c040e000 task=c0386a40)
Stack: 77d89bb3 dcf91330 da6e4458 e38c28ed dcf91438 77d89bb3 c040fbf0 77d89bb3
c040fbf0 c01e7623 dcf91330 77d89bb3 c040fbf0 c01e7373 c040fbf2
dec77000
00000006 c040fb94 c040fbf0 000004a0 00000008 c040fc50 c01e7513
dcf91304
Call Trace:
[<e38c28ed>] des3_ede_decrypt+0x2d/0x70 [des]
[<c01e7623>] cbc_process+0x63/0x170
[<c01e7373>] scatterwalk_copychunks+0x83/0xd0
[<c01e7513>] crypt+0x153/0x200
[<e38c28c0>] des3_ede_decrypt+0x0/0x70 [des]
[<c02d2000>] dst_output+0x0/0x30
[<c02d1b74>] ip_push_pending_frames+0x344/0x420
[<c02d2000>] dst_output+0x0/0x30
[<c01e7948>] cbc_decrypt+0x48/0x50
[<e38c28c0>] des3_ede_decrypt+0x0/0x70 [des]
[<c01e75c0>] cbc_process+0x0/0x170
[<e38ccb1b>] esp_input+0x1fb/0x500 [esp4]
[<e38cca3e>] esp_input+0x11e/0x500 [esp4]
[<c02b78f4>] kfree_skbmem+0x24/0x30
[<c030fdeb>] xfrm_state_lookup+0x5b/0x70
[<c030b58e>] xfrm4_rcv_encap+0xee/0x4a0
[<c02d20e2>] ip_finish_output2+0xb2/0x1d0
[<c0302484>] ipt_do_table+0x294/0x380
[<c02c5302>] nf_hook_slow+0xd2/0x120
[<c02ccfc0>] ip_local_deliver_finish+0x0/0x1d0
[<c0304b37>] ipt_hook+0x37/0x40
[<c030b3d7>] xfrm4_rcv+0x17/0x20
[<c02cd089>] ip_local_deliver_finish+0xc9/0x1d0
[<c02ccfc0>] ip_local_deliver_finish+0x0/0x1d0
[<c02c5302>] nf_hook_slow+0xd2/0x120
[<c02ccfc0>] ip_local_deliver_finish+0x0/0x1d0
[<c02ccb27>] ip_local_deliver+0x217/0x240
[<c02ccfc0>] ip_local_deliver_finish+0x0/0x1d0
[<c02cd350>] ip_rcv_finish+0x1c0/0x230
[<c02cd190>] ip_rcv_finish+0x0/0x230
[<c02c5302>] nf_hook_slow+0xd2/0x120
[<c02cd190>] ip_rcv_finish+0x0/0x230
[<c02ccf04>] ip_rcv+0x3b4/0x470
[<c02cd190>] ip_rcv_finish+0x0/0x230
[<c02bc430>] netif_receive_skb+0x150/0x190
[<c02bc4dd>] process_backlog+0x6d/0x100
[<c02bc5da>] net_rx_action+0x6a/0xf0
[<c011afc5>] __do_softirq+0x85/0x90
[<c011aff6>] do_softirq+0x26/0x30
[<c0106439>] do_IRQ+0xc9/0xf0
[<c0101ef0>] default_idle+0x0/0x30
[<c01047cc>] common_interrupt+0x18/0x20
[<c0101ef0>] default_idle+0x0/0x30
[<c0101f14>] default_idle+0x24/0x30
[<c0101f90>] cpu_idle+0x30/0x40
[<c041072d>] start_kernel+0x14d/0x170
[<c0410470>] unknown_bootoption+0x0/0x130
Code: 88 0f c1 e9 08 31 c2 88 4f 01 c1 e9 08 88 57 04 88 4f 02 c1
<0>Kernel panic: Fatal exception in interruptIn interrupt handler - not syncing
Quoted 5 lines · Tobias Ringström
I get a kernel panic that is very easy to reproduce. The panic goes away if I revert the following changeset ("fix in-place de/encryption bug with highmem"):
http://linux.bkbits.net:8080/linux-2.5/[email protected]
Or,
http://linux.bkbits.net:8080/linux-2.5/cset@40443186BowXpsBHgFbxClLNqkdCyw
(need to use "bookmarkable link"; revision numbers change.)
-andy
Am Mi, den 19.05.2004 um 21:48 Uhr +0200 schrieb Tobias Ringström:
Quoted 6 lines
I have not analyzed the code closely, so I cannot say for certain that there is a problem with that specific changeset, and I'm certainly not out to blame anyone, but reverting it at least seems to fix the problem. The panic occurs when I receive more than a little TCP data in an IPsec ESP tunnel. I'm of course willing to test patches or provide further info if required.
I've been using linux 2.6. IPSec myself for quite some time and never seen this. Hmm. What kind of tunnel is this? How large are the packets involved?
Can you send me your .config and tell me what compiler was used, and perhaps the object files in your kernel crypto/ directory?
Quoted 2 lines
eax:
0030f000ebx:00030f00ecx:1b24b797edx:ba4d7f77
esi:dcf91438edi:77d89bb3ebp:c040fbb0esp:c040fb5c
^^^^^^^^
What kind of address is this? Something random? The ESP code uses in-place decryption, so this should point on the stack.
Quoted 4 lines
ds: 007b es: 007b ss: 0068
Process swapper (pid: 0, threadinfo=c040e000 task=c0386a40)Stack:
77d89bb3dcf91330da6e4458e38c28eddcf9143877d89bb3c040fbf077d89bb3
^^^^^^^^ ^^^^^^^^ ^^^^^^^
Here it is again.