[Xen-devel] [PATCH V2] xen/hvm: fix hypervisor crash with hvm_save_one()

[Razvan Cojocaru <rcojocaru@xxxxxxxxxxxxxxx>]

hvm_save_cpu_ctxt() returns success without writing any data into
hvm_domain_context_t when all VCPUs are offline. This can then crash
the hypervisor (with FATAL PAGE FAULT) in hvm_save_one() via the
"off < (ctxt.cur - sizeof(*desc))" for() test, where ctxt.cur remains 0,
causing an underflow which leads the hypervisor to go off the end of the
ctxt buffer.

This has been broken since Xen 4.4 (c/s e019c606f59).
It has happened in practice with an HVM Linux VM (Debian 8) queried around
shutdown:

(XEN) hvm.c:1595:d3v0 All CPUs offline -- powering off.
(XEN) ----[ Xen-4.9-rc  x86_64  debug=y   Not tainted ]----
(XEN) CPU:    5
(XEN) RIP:    e008:[<ffff82d0802496d2>] hvm_save_one+0x145/0x1fd
(XEN) RFLAGS: 0000000000010286   CONTEXT: hypervisor (d0v2)
(XEN) rax: ffff830492cbb445   rbx: 0000000000000000   rcx: ffff83039343b400
(XEN) rdx: 00000000ff88004d   rsi: fffffffffffffff8   rdi: 0000000000000000
(XEN) rbp: ffff8304103e7c88   rsp: ffff8304103e7c48   r8:  0000000000000001
(XEN) r9:  deadbeefdeadf00d   r10: 0000000000000000   r11: 0000000000000282
(XEN) r12: 00007f43a3b14004   r13: 00000000fffffffe   r14: 0000000000000000
(XEN) r15: ffff830400c41000   cr0: 0000000080050033   cr4: 00000000001526e0
(XEN) cr3: 0000000402e13000   cr2: ffff830492cbb447
(XEN) ds: 0000   es: 0000   fs: 0000   gs: 0000   ss: e010   cs: e008
(XEN) Xen code around <ffff82d0802496d2> (hvm_save_one+0x145/0x1fd):
(XEN)  00 00 48 01 c8 83 c2 08 <66> 39 58 02 75 64 eb 08 48 89 c8 ba 08 00 00 00
(XEN) Xen stack trace from rsp=ffff8304103e7c48:
(XEN)    0000041000000000 ffff83039343b400 ffff8304103e7c70 ffff8304103e7da8
(XEN)    ffff830400c41000 00007f43a3b13004 ffff8304103b7000 ffffffffffffffea
(XEN)    ffff8304103e7d48 ffff82d0802683d4 ffff8300d19fd000 ffff82d0802320d8
(XEN)    ffff830400c41000 0000000000000000 ffff8304103e7cd8 ffff82d08026ff3d
(XEN)    0000000000000000 ffff8300d19fd000 ffff8304103e7cf8 ffff82d080232142
(XEN)    0000000000000000 ffff8300d19fd000 ffff8304103e7d28 ffff82d080207051
(XEN)    ffff8304103e7d18 ffff830400c41000 0000000000000202 ffff830400c41000
(XEN)    0000000000000000 00007f43a3b13004 0000000000000000 deadbeefdeadf00d
(XEN)    ffff8304103e7e68 ffff82d080206c47 0700000000000000 ffff830410375bd0
(XEN)    0000000000000296 ffff830410375c78 ffff830410375c80 0000000000000003
(XEN)    ffff8304103e7e68 ffff8304103b67c0 ffff8304103b7000 ffff8304103b67c0
(XEN)    0000000d00000037 0000000000000003 0000000000000002 00007f43a3b14004
(XEN)    00007ffd5d925590 0000000000000000 0000000100000000 0000000000000000
(XEN)    00000000ea8f8000 0000000000000000 00007ffd00000000 0000000000000000
(XEN)    00007f43a276f557 0000000000000000 00000000ea8f8000 0000000000000000
(XEN)    00007ffd5d9255e0 00007f43a23280b2 00007ffd5d926058 ffff8304103e7f18
(XEN)    ffff8300d19fe000 0000000000000024 ffff82d0802053e5 deadbeefdeadf00d
(XEN)    ffff8304103e7f08 ffff82d080351565 010000003fffffff 00007f43a3b13004
(XEN)    deadbeefdeadf00d deadbeefdeadf00d deadbeefdeadf00d deadbeefdeadf00d
(XEN)    ffff8800781425c0 ffff88007ce94300 ffff8304103e7ed8 ffff82d0802719ec
(XEN) Xen call trace:
(XEN)    [<ffff82d0802496d2>] hvm_save_one+0x145/0x1fd
(XEN)    [<ffff82d0802683d4>] arch_do_domctl+0xa7a/0x259f
(XEN)    [<ffff82d080206c47>] do_domctl+0x1862/0x1b7b
(XEN)    [<ffff82d080351565>] pv_hypercall+0x1ef/0x42c
(XEN)    [<ffff82d080355106>] entry.o#test_all_events+0/0x30
(XEN)
(XEN) Pagetable walk from ffff830492cbb447:
(XEN)  L4[0x106] = 00000000dbc36063 ffffffffffffffff
(XEN)  L3[0x012] = 0000000000000000 ffffffffffffffff
(XEN)
(XEN) ****************************************
(XEN) Panic on CPU 5:
(XEN) FATAL PAGE FAULT
(XEN) [error_code=0000]
(XEN) Faulting linear address: ffff830492cbb447
(XEN) ****************************************

Reported-by: Razvan Cojocaru <rcojocaru@xxxxxxxxxxxxxxx>
Signed-off-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
Signed-off-by: Razvan Cojocaru <rcojocaru@xxxxxxxxxxxxxxx>
Tested-by: Razvan Cojocaru <rcojocaru@xxxxxxxxxxxxxxx>

---
Changes since V1:
 - Corrected patch description.
 - Now checking whether the function got back any data at all, prior to
   entering the for() loop.
---
 xen/common/hvm/save.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/xen/common/hvm/save.c b/xen/common/hvm/save.c
index 78706f5..3bdd124 100644
--- a/xen/common/hvm/save.c
+++ b/xen/common/hvm/save.c
@@ -113,6 +113,9 @@ int hvm_save_one(struct domain *d, uint16_t typecode, 
uint16_t instance,
         const struct hvm_save_descriptor *desc;
 
         rv = -ENOENT;
+        if ( ctxt.cur < sizeof(*desc) )
+            goto out;
+
         for ( off = 0; off < (ctxt.cur - sizeof(*desc)); off += desc->length )
         {
             desc = (void *)(ctxt.data + off);
@@ -132,6 +135,7 @@ int hvm_save_one(struct domain *d, uint16_t typecode, 
uint16_t instance,
         }
     }
 
+ out:
     xfree(ctxt.data);
     return rv;
 }
-- 
1.9.1


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel