migration/vmstate: replace NULL pointer assert with proper error
PATCH 1/1 migration/vmstate: replace NULL pointer assert with proper error
From: Trieu Huynh <[email protected]>
vmstate_load_state() contains an assert added in commit ("migration/vmstate: split up vmstate_base_addr" [1]) to catch NULL pointer dereferences early:
assert(first_elem || !n_elems || !size);
However, the assert fires as an unrecoverable crash in a valid user scenario, for instance: loading a snapshot saved with -icount auto into a QEMU started without -icount. The "timer/icount/warp_timer" subsection is present in the stream, but icount_warp_timer is NULL because icount was never initialized. QEMU aborts with:
qemu-system-x86_64: ../migration/vmstate.c:185: vmstate_load_state: Assertion `first_elem || !n_elems || !size' failed.
Replace the assert with a proper error_setg() and -EINVAL return so QEMU reports a clear diagnostic instead of crashing.
Also remove the error_prepend() in vmstate_subsection_load() that wrapped the error at each subsection level, producing a redundant and noisy error chain. The original error message is now propagated as-is.
[1] https://lore.kernel.org/qemu-devel/[email protected]/
Resolves: https://gitlab.com/qemu-project/qemu/-/work_items/2736
---
Steps to reproduce:
* 1. Start src with icount config:
./qemu-system-x86_64 \
-icount auto \
-<other_options>
* 2. Save snapshot
(qemu) savevm snap1
* 3. Start dest without icount config:
./qemu-system-x86_64 \
-<other_options>
* 4. Load snapshot
* As-is:
~ # QEMU 10.2.94 monitor - type 'help' for more information (qemu) loadvm snap1
qemu-system-x86_64: ../migration/vmstate.c:185: vmstate_load_state:
Assertion `first_elem || !n_elems || !size' failed. Aborted (core dumped)
* To-be:
** Before removing error_prepend()
~ # QEMU 10.2.94 monitor - type 'help' for more information (qemu) loadvm snap1
Error: error while loading state for instance 0x0 of device 'timer':
Loading VM subsection 'timer/icount' in 'timer' failed: -22:
Loading VM subsection 'timer/icount/warp_timer' in 'timer/icount' failed: -22:
VM load of device 'timer/icount/warp_timer': field 'icount_warp_timer' is a NULL pointer but snapshot contains data for it.
** After removing error_prepend()
~ # QEMU 10.2.94 monitor - type 'help' for more information (qemu) loadvm snap1
Error: error while loading state for instance 0x0 of device 'timer':
VM load of device 'timer/icount/warp_timer': field 'icount_warp_timer' is a NULL pointer but snapshot contains data for it.
migration/vmstate.c | 11 +++++++---- 1 file changed, 7 insertions(+), 4 deletions(-)
migration/vmstate.c +7 -4
@@ -182,7 +182,13 @@ int vmstate_load_state(QEMUFile *f, const VMStateDescription *vmsd,vmstate_handle_alloc (first_elem, field, opaque);if (field->flags &VMS_POINTER ) { first_elem = *(void **)first_elem; -assert (first_elem || !n_elems || !size); +if (!first_elem && n_elems && size) { +error_setg (errp, +"VM load of device '%s': field '%s' is a NULL " +"pointer but snapshot contains data for it." , + vmsd->name, field->name); +return -EINVAL ; + } }for (i =0 ; i < n_elems; i++) {void *curr_elem = first_elem + size * i;
@@ -651,9 +657,6 @@ static int vmstate_subsection_load(QEMUFile *f, const VMStateDescription *vmsd, ret =vmstate_load_state (f, sub_vmsd, opaque, version_id, errp);if (ret) {trace_vmstate_subsection_load_bad (vmsd->name, idstr,"(child)" ); -error_prepend (errp, -"Loading VM subsection '%s' in '%s' failed: %d: " , - idstr, vmsd->name, ret);return ret; } }
Signature · 1 line
2.43.0
Quoted 61 lines · Trieu Huynh
From: Trieu Huynh <[email protected]>
vmstate_load_state()contains an assert added in commit ("migration/vmstate: split upvmstate_base_addr" [1]) to catch NULL pointer dereferences early:assert(first_elem || !n_elems || !size);However, the assert fires as an unrecoverable crash in a valid user scenario, for instance: loading a snapshot saved with -icount auto into a QEMU started without -icount. The "timer/icount/warp_timer" subsection is present in the stream, but
icount_warp_timeris NULL because icount was never initialized. QEMU aborts with:qemu-system-x86_64: ../migration/vmstate.c:185: vmstate_load_state: Assertion `first_elem || !n_elems || !size' failed.Replace the assert with a proper
error_setg()and-EINVALreturn so QEMU reports a clear diagnostic instead of crashing.Also remove the
error_prepend()invmstate_subsection_load()that wrapped the error at each subsection level, producing a redundant and noisy error chain. The original error message is now propagated as-is.[1] https://lore.kernel.org/qemu-devel/[email protected]/
Resolves: https://gitlab.com/qemu-project/qemu/-/work_items/2736Signed-off-by Trieu Huynh <[email protected]>---
Steps to reproduce:
* 1. Start src with icount config:
./qemu-system-x86_64 \
-icount auto \
-<other_options>* 2. Save snapshot
(qemu) savevm snap1* 3. Start dest without icount config:
./qemu-system-x86_64 \
-<other_options>* 4. Load snapshot
* As-is:
~ # QEMU 10.2.94 monitor - type 'help' for more information (qemu) loadvm snap1
qemu-system-x86_64:../migration/vmstate.c:185:vmstate_load_state:
Assertion `first_elem || !n_elems || !size' failed. Aborted (core dumped)* To-be:
** Before removingerror_prepend()
~ # QEMU 10.2.94 monitor - type 'help' for more information (qemu) loadvm snap1
Error: error while loading state for instance0x0of device 'timer':
Loading VM subsection 'timer/icount' in 'timer' failed: -22:
Loading VM subsection 'timer/icount/warp_timer' in 'timer/icount' failed: -22:
VM load of device 'timer/icount/warp_timer': field 'icount_warp_timer' is a NULL pointer but snapshot contains data for it.
Quoted 6 lines
** After removing
error_prepend()
~ # QEMU 10.2.94 monitor - type 'help' for more information (qemu) loadvm snap1
Error: error while loading state for instance0x0of device 'timer':
VM load of device 'timer/icount/warp_timer': field 'icount_warp_timer' is a NULL pointer but snapshot contains data for it.
Thanks again for help looking into QEMU opened issues, Trieu.
I had a feeling the user was requesting for successful migrations with modified -icount parameters, rather than a prettier failure. In that case this patch won't be enough.
I've added a comment in the issue to check with reporter. If we need that feature, we may want to check with TCG developers on whether it'll work out.
PS: we shouldn't mention "snapshot" in the error message, because it can also be a live migration that triggers this path.
Thanks,
Signature · 1 line
Peter Xu