Re: [RFC PATCH v1 20/26] kvm: vmi: intercept live migration

[Dr. David Alan Gilbert]

* Adalbert Lazăr (address@hidden) wrote:
> On Mon, 27 Apr 2020 20:08:55 +0100, "Dr. David Alan Gilbert" <address@hidden> 
> wrote:
> > * Adalbert Lazăr (address@hidden) wrote:
> > > From: Marian Rotariu <address@hidden>
> > > 
> > > It is possible that the introspection tool has made some changes inside
> > > the introspected VM which can make the guest crash if the introspection
> > > connection is suddenly closed.
> > > 
> > > When the live migration starts, for now, the introspection tool is
> > > signaled to remove its hooks from the introspected VM.
> > > 
> > > CC: Juan Quintela <address@hidden>
> > > CC: "Dr. David Alan Gilbert" <address@hidden>
> > > Signed-off-by: Marian Rotariu <address@hidden>
> > > Signed-off-by: Adalbert Lazăr <address@hidden>
> > 
> > OK, so this isn't too intrusive to the migration code; and other than
> > renaming 'start_live_migration_thread' to
> > 'start_outgoing_migration_thread' I think I'd be OK with this,
> > 
> > but it might depend what your overall aim is.
> > 
> > For example, you might be better intercepting each migration_state
> > change in your notifier, that's much finer grain than just the start of
> > migration.
> 
> Thank you, Dave.
> 
> We want to intercept the live migration and 'block' it while the guest
> is running (some changes made to the guest by the introspection app has
> to be undone while the vCPUs are in certain states).
> 
> I'm not sure what is the best way to block these kind of events
> (including the pause/shutdown commands). If calling main_loop_wait()
> is enough (patch [22/26] kvm: vmi: add 'async_unhook' property [1])
> then we can drop a lot of code.
> 
> The use of a notifier will be nice, but from what I understand, we can't
> block the migration from a notification callback.

Oh, if your intention is *just* to block a migration starting then you
can use 'migrate_add_blocker' - see hw/9pfs/9p.c for an example where
it's used and then removed; they use it to stop migration while the fs
 is mounted.  That causes an attempt to start a migration to give an
error (of your choosing).

> > The other thing I worry about is that there doesn't seem to be much
> > guard against odd orderings of things - for example, what happens
> > if the introspection client was to issue the  INTERCEPT_MIGRATE command
> > twice while a migration was already running?  Or before an actual
> > incoming channel connetion had happened?
> > 
> > Dave
> 
> Sorry that I haven't described the interception. When we intercept
> an action that we want to 'block', we set a static variable first,
> regardless if the introspection channel is connected or not, and :
> 
>    - if the introspection channel is not connected we don't block the
>    action, but this (variable) will prevent the activation of this
>    channel until the action (ie. migrate) is completed (a). I assume
>    that there could be only one migrate (or suspend/pause) user command
>    at any given time (b).
> 
>    - if the introspection channel is connected, the introspection app
>    is signaled to start its unhook/undo process. We let the code flow
>    continue, but the action (migrate/suspend/pause) is delayed until
>    the introspection channel is closed. Meanwhile, any other intercepted
>    action will not be blocked/delayed (c), but the fact that these actions
>    are in progress is saved to static variables and the introspecton
>    channel won't be reactivated.
> 
> Indeed, there are cases that are not handled well:
> 
>   a) if the migration is started and canceled before the introspection
>   object is created (through QMP), the introspection channel will be
>   disabled until the next migration starts and finishes.
> 
>   b) if a migration command has been delayed, a following migrate command
>   (if this is possible) won't be delayed and we will have two migration
>   threads started.
> 
>   c) if a migration command has been delayed, a following suspend/pause
>   command won't be delayed and the introspection app might not have
>   enough time to finish its unhook/undo process.

Yeh that sounds a bit messy.

Dave


> [1]: https://lore.kernel.org/qemu-devel/address@hidden/
> 
> > > ---
> > >  accel/kvm/vmi.c                | 31 +++++++++++++++++++++++++++----
> > >  include/sysemu/vmi-intercept.h |  1 +
> > >  migration/migration.c          | 18 +++++++++++++++---
> > >  migration/migration.h          |  2 ++
> > >  4 files changed, 45 insertions(+), 7 deletions(-)
> > > 
> > > diff --git a/accel/kvm/vmi.c b/accel/kvm/vmi.c
> > > index 90906478b4..ea7191e48d 100644
> > > --- a/accel/kvm/vmi.c
> > > +++ b/accel/kvm/vmi.c
> > > @@ -21,6 +21,8 @@
> > >  #include "chardev/char.h"
> > >  #include "chardev/char-fe.h"
> > >  #include "migration/vmstate.h"
> > > +#include "migration/migration.h"
> > > +#include "migration/misc.h"
> > >  
> > >  #include "sysemu/vmi-intercept.h"
> > >  #include "sysemu/vmi-handshake.h"
> > > @@ -58,6 +60,7 @@ typedef struct VMIntrospection {
> > >      int64_t vm_start_time;
> > >  
> > >      Notifier machine_ready;
> > > +    Notifier migration_state_change;
> > >      bool created_from_command_line;
> > >  
> > >      bool kvmi_hooked;
> > > @@ -74,9 +77,11 @@ static const char *action_string[] = {
> > >      "suspend",
> > >      "resume",
> > >      "force-reset",
> > > +    "migrate",
> > >  };
> > >  
> > >  static bool suspend_pending;
> > > +static bool migrate_pending;
> > >  
> > >  #define TYPE_VM_INTROSPECTION "introspection"
> > >  
> > > @@ -88,6 +93,15 @@ static bool suspend_pending;
> > >  static Error *vm_introspection_init(VMIntrospection *i);
> > >  static void vm_introspection_reset(void *opaque);
> > >  
> > > +static void migration_state_notifier(Notifier *notifier, void *data)
> > > +{
> > > +    MigrationState *s = data;
> > > +
> > > +    if (migration_has_failed(s)) {
> > > +        migrate_pending = false;
> > > +    }
> > > +}
> > > +
> > >  static void machine_ready(Notifier *notifier, void *data)
> > >  {
> > >      VMIntrospection *i = container_of(notifier, VMIntrospection, 
> > > machine_ready);
> > > @@ -144,6 +158,9 @@ static void complete(UserCreatable *uc, Error **errp)
> > >  
> > >      ic->uniq = i;
> > >  
> > > +    i->migration_state_change.notify = migration_state_notifier;
> > > +    add_migration_state_change_notifier(&i->migration_state_change);
> > > +
> > >      qemu_register_reset(vm_introspection_reset, i);
> > >  }
> > >  
> > > @@ -478,6 +495,9 @@ static void 
> > > continue_with_the_intercepted_action(VMIntrospection *i)
> > >      case VMI_INTERCEPT_SUSPEND:
> > >          vm_stop(RUN_STATE_PAUSED);
> > >          break;
> > > +    case VMI_INTERCEPT_MIGRATE:
> > > +        start_live_migration_thread(migrate_get_current());
> > > +        break;
> > >      default:
> > >          error_report("VMI: %s: unexpected action %d",
> > >                       __func__, i->intercepted_action);
> > > @@ -571,9 +591,9 @@ static void chr_event_open(VMIntrospection *i)
> > >  {
> > >      Error *local_err = NULL;
> > >  
> > > -    if (suspend_pending) {
> > > -        info_report("VMI: %s: too soon (suspend=%d)",
> > > -                    __func__, suspend_pending);
> > > +    if (suspend_pending || migrate_pending) {
> > > +        info_report("VMI: %s: too soon (suspend=%d, migrate=%d)",
> > > +                    __func__, suspend_pending, migrate_pending);
> > >          maybe_disable_socket_reconnect(i);
> > >          qemu_chr_fe_disconnect(&i->sock);
> > >          return;
> > > @@ -608,7 +628,7 @@ static void chr_event_close(VMIntrospection *i)
> > >      cancel_unhook_timer(i);
> > >      cancel_handshake_timer(i);
> > >  
> > > -    if (suspend_pending) {
> > > +    if (suspend_pending || migrate_pending) {
> > >          maybe_disable_socket_reconnect(i);
> > >  
> > >          if (i->intercepted_action != VMI_INTERCEPT_NONE) {
> > > @@ -680,6 +700,9 @@ static bool 
> > > record_intercept_action(VMI_intercept_command action)
> > >          break;
> > >      case VMI_INTERCEPT_FORCE_RESET:
> > >          break;
> > > +    case VMI_INTERCEPT_MIGRATE:
> > > +        migrate_pending = true;
> > > +        break;
> > >      default:
> > >          return false;
> > >      }
> > > diff --git a/include/sysemu/vmi-intercept.h 
> > > b/include/sysemu/vmi-intercept.h
> > > index ef591b49e7..b4a9a3faa7 100644
> > > --- a/include/sysemu/vmi-intercept.h
> > > +++ b/include/sysemu/vmi-intercept.h
> > > @@ -15,6 +15,7 @@ typedef enum {
> > >      VMI_INTERCEPT_SUSPEND,
> > >      VMI_INTERCEPT_RESUME,
> > >      VMI_INTERCEPT_FORCE_RESET,
> > > +    VMI_INTERCEPT_MIGRATE,
> > >  } VMI_intercept_command;
> > >  
> > >  bool vm_introspection_intercept(VMI_intercept_command ic, Error **errp);
> > > diff --git a/migration/migration.c b/migration/migration.c
> > > index 187ac0410c..222037d739 100644
> > > --- a/migration/migration.c
> > > +++ b/migration/migration.c
> > > @@ -55,6 +55,8 @@
> > >  #include "qemu/queue.h"
> > >  #include "multifd.h"
> > >  
> > > +#include "sysemu/vmi-intercept.h"
> > > +
> > >  #define MAX_THROTTLE  (32 << 20)      /* Migration transfer speed 
> > > throttling */
> > >  
> > >  /* Amount of time to allocate to each "chunk" of bandwidth-throttled
> > > @@ -3471,6 +3473,13 @@ static void *migration_thread(void *opaque)
> > >      return NULL;
> > >  }
> > >  
> > > +void start_live_migration_thread(MigrationState *s)
> > > +{
> > > +    qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
> > > +                    QEMU_THREAD_JOINABLE);
> > > +    s->migration_thread_running = true;
> > > +}
> > > +
> > >  void migrate_fd_connect(MigrationState *s, Error *error_in)
> > >  {
> > >      Error *local_err = NULL;
> > > @@ -3534,9 +3543,12 @@ void migrate_fd_connect(MigrationState *s, Error 
> > > *error_in)
> > >          migrate_fd_cleanup(s);
> > >          return;
> > >      }
> > > -    qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
> > > -                       QEMU_THREAD_JOINABLE);
> > > -    s->migration_thread_running = true;
> > > +
> > > +    if (vm_introspection_intercept(VMI_INTERCEPT_MIGRATE, &error_in)) {
> > > +        return;
> > > +    }
> > > +
> > > +    start_live_migration_thread(s);
> > >  }
> > >  
> > >  void migration_global_dump(Monitor *mon)
> > > diff --git a/migration/migration.h b/migration/migration.h
> > > index 507284e563..eb5668e1f2 100644
> > > --- a/migration/migration.h
> > > +++ b/migration/migration.h
> > > @@ -263,6 +263,8 @@ struct MigrationState
> > >      uint8_t clear_bitmap_shift;
> > >  };
> > >  
> > > +void start_live_migration_thread(MigrationState *s);
> > > +
> > >  void migrate_set_state(int *state, int old_state, int new_state);
> > >  
> > >  void migration_fd_process_incoming(QEMUFile *f, Error **errp);
> > > 
> > --
> > Dr. David Alan Gilbert / address@hidden / Manchester, UK
> > 
> 
--
Dr. David Alan Gilbert / address@hidden / Manchester, UK