Call resume fast if an exception occurs during migration

If an exception is raised when performing the migration of a VM we call
resume fast to resume the VM on the original host if the VM is in
suspended state.

Signed-off-by: Guillaume <[email protected]>

ocaml/idl/datamodel_vm.ml

@@ -827,6 +827,22 @@ let unpause =
       ]
     ~allowed_roles:_R_VM_OP ()
 
+(* VM.ResumeFast *)
+
+let resumeFast =
+  call ~in_product_since:rel_rio ~name:"resume_fast"
+    ~doc:
+      "Resume the specified VM cooperatively. This can only be called when the \
+       specified VM is in the Suspended state."
+    ~params:[(Ref _vm, "vm", "The VM to resume fast")]
+    ~errs:
+      [
+        Api_errors.vm_bad_power_state
+      ; Api_errors.operation_not_allowed
+      ; Api_errors.vm_is_template
+      ]
+    ~allowed_roles:_R_VM_OP ()
+
 (* VM.CleanShutdown *)
 
 let cleanShutdown =

ocaml/xapi-idl/xen/xenops_interface.ml

@@ -737,6 +737,10 @@ module XenopsAPI (R : RPC) = struct
       declare "VM.unpause" []
         (debug_info_p @-> vm_id_p @-> returning task_id_p err)
 
+    let resume_fast =
+      declare "VM.resume_fast" []
+        (debug_info_p @-> vm_id_p @-> returning task_id_p err)
+
     let request_rdp =
       let enabled_p = Param.mk ~name:"enabled" Types.bool in
       declare "VM.request_rdp" []

ocaml/xapi/api_server.ml

@@ -197,6 +197,7 @@ let callback1 ?(json_rpc_version = Jsonrpc.V1) is_json req fd call =
   (* if we're a master or slave and whether the call came in on the unix domain socket or the tcp socket *)
   (* If we're a slave, and the call is from the unix domain socket or from the HIMN, and the call *isn't* *)
   (* in the whitelist, then forward *)
+  let _ = D.info "call.Rpc.name is %s" call.Rpc.name in
   let whitelisted = List.mem call.Rpc.name whitelist in
   let emergency_call = List.mem call.Rpc.name emergency_call_list in
   let is_slave = not (Pool_role.is_master ()) in

ocaml/xenopsd/lib/xenops_server.ml

@@ -138,6 +138,7 @@ type atomic =
   | VM_pause of Vm.id
   | VM_softreboot of Vm.id
   | VM_unpause of Vm.id
+  | VM_resume_fast of Vm.id
   | VM_request_rdp of (Vm.id * bool)
   | VM_run_script of (Vm.id * string)
   | VM_set_domain_action_request of (Vm.id * domain_action_request option)
@@ -231,6 +232,8 @@ let rec name_of_atomic = function
       "VM_softreboot"
   | VM_unpause _ ->
       "VM_unpause"
+  | VM_resume_fast _ ->
+      "VM_resume_fast"
   | VM_request_rdp _ ->
       "VM_request_rdp"
   | VM_run_script _ ->
@@ -302,6 +305,7 @@ type operation =
   | VM_restore_vifs of Vm.id
   | VM_restore_devices of (Vm.id * bool)
   | VM_migrate of vm_migrate_op
+  | VM_resume_fast of Vm.id
   | VM_receive_memory of vm_receive_op
   | VBD_hotplug of Vbd.id
   | VBD_hotunplug of Vbd.id * bool
@@ -336,6 +340,8 @@ let name_of_operation = function
       "VM_restore_devices"
   | VM_migrate _ ->
       "VM_migrate"
+  | VM_resume_fast _ ->
+      "VM_resumte_fast"
   | VM_receive_memory _ ->
       "VM_receive_memory"
   | VBD_hotplug _ ->
@@ -2195,6 +2201,16 @@ let rec perform_atomic ~progress_callback ?subtask:_ ?result (op : atomic)
       | _ ->
           info "VM %s is not paused" id
     )
+  | VM_resume_fast id -> (
+      debug "VM.resume_fast %s" id ;
+      let vm_t = VM_DB.read_exn id in
+      let power = (B.VM.get_state vm_t).Vm.power_state in
+      match power with
+      | Suspended ->
+          B.VM.resume_fast t vm_t ; VM_DB.signal id
+      | _ ->
+          info "VM %s is not suspended, fast resum is not allowed" id
+    )
   | VM_request_rdp (id, enabled) ->
       debug "VM.request_rdp %s %b" id enabled ;
       B.VM.request_rdp (VM_DB.read_exn id) enabled
@@ -2419,6 +2435,7 @@ let rec immediate_operation dbg _id op =
 and trigger_cleanup_after_failure op t =
   let dbg = (Xenops_task.to_interface_task t).Task.dbg in
   match op with
+  | VM_resume_fast _
   | VM_check_state _
   | PCI_check_state _
   | VBD_check_state _
@@ -2439,7 +2456,8 @@ and trigger_cleanup_after_failure op t =
       immediate_operation dbg final_id (VM_check_state final_id)
   | VM_migrate {vmm_id; vmm_tmp_src_id; _} ->
       immediate_operation dbg vmm_id (VM_check_state vmm_id) ;
-      immediate_operation dbg vmm_tmp_src_id (VM_check_state vmm_tmp_src_id)
+      immediate_operation dbg vmm_tmp_src_id (VM_check_state vmm_tmp_src_id) ;
+      immediate_operation dbg vmm_id (VM_resume_fast vmm_id)
   | VBD_hotplug id | VBD_hotunplug (id, _) ->
       immediate_operation dbg (fst id) (VBD_check_state id)
   | VIF_hotplug id | VIF_hotunplug (id, _) ->
@@ -2489,6 +2507,7 @@ and trigger_cleanup_after_failure_atom op t =
   | VM_set_memory_dynamic_range (id, _, _)
   | VM_pause id
   | VM_unpause id
+  | VM_resume_fast id
   | VM_request_rdp (id, _)
   | VM_run_script (id, _)
   | VM_set_domain_action_request (id, _)
@@ -2569,6 +2588,10 @@ and perform_exn ?subtask ?result (op : operation) (t : Xenops_task.task_handle)
   | VIF_hotunplug (id, force) ->
       debug "VIF_hotplug %s.%s %b" (fst id) (snd id) force ;
       perform_atomics (atomics_of_operation op) t
+  | VM_resume_fast id ->
+      debug "VM.resume_fast %s" id ;
+      perform_atomics (atomics_of_operation op) t ;
+      VM_DB.signal id
   | VM_migrate vmm ->
       debug "VM.migrate %s -> %s" vmm.vmm_id vmm.vmm_url ;
       let id = vmm.vmm_id in
@@ -3554,6 +3577,8 @@ module VM = struct
 
   let unpause _ dbg id = queue_operation dbg id (Atomic (VM_unpause id))
 
+  let resume_fast _ dbg id = queue_operation dbg id (Atomic (VM_resume_fast id))
+
   let request_rdp _ dbg id enabled =
     queue_operation dbg id (Atomic (VM_request_rdp (id, enabled)))
 
@@ -4137,6 +4162,7 @@ let _ =
   Server.VM.destroy (VM.destroy ()) ;
   Server.VM.pause (VM.pause ()) ;
   Server.VM.unpause (VM.unpause ()) ;
+  Server.VM.resume_fast (VM.resume_fast ()) ;
   Server.VM.request_rdp (VM.request_rdp ()) ;
   Server.VM.run_script (VM.run_script ()) ;
   Server.VM.set_xsdata (VM.set_xsdata ()) ;

ocaml/xenopsd/lib/xenops_server_plugin.ml

@@ -118,6 +118,8 @@ module type S = sig
 
     val unpause : Xenops_task.task_handle -> Vm.t -> unit
 
+    val resume_fast : Xenops_task.task_handle -> Vm.t -> unit
+
     val set_xsdata :
       Xenops_task.task_handle -> Vm.t -> (string * string) list -> unit
 

ocaml/xenopsd/lib/xenops_server_skeleton.ml

@@ -77,6 +77,8 @@ module VM = struct
 
   let unpause _ _ = unimplemented "VM.unpause"
 
+  let resume_fast _ _ = unimplemented "VM.resume_fast"
+
   let set_xsdata _ _ _ = unimplemented "VM.set_xsdata"
 
   let set_vcpus _ _ _ = unimplemented "VM.set_vcpus"

ocaml/xenopsd/xc/domain.ml

@@ -770,6 +770,11 @@ let pause ~xc domid = Xenctrl.domain_pause xc domid
 
 let unpause ~xc domid = Xenctrl.domain_unpause xc domid
 
+let resume_fast ~xc domid =
+  let uuid = get_uuid ~xc domid in
+  debug "VM = %s; domid = %d; calling resume fast" (Uuidx.to_string uuid) domid ;
+  Xenctrl.domain_resume_fast xc domid
+
 let set_action_request ~xs domid x =
   let path = xs.Xs.getdomainpath domid ^ "/action-request" in
   match x with None -> xs.Xs.rm path | Some v -> xs.Xs.write path v

ocaml/xenopsd/xc/domain.mli

@@ -209,6 +209,9 @@ val pause : xc:Xenctrl.handle -> domid -> unit
 val unpause : xc:Xenctrl.handle -> domid -> unit
 (** Unpause a domain *)
 
+val resume_fast : xc:Xenctrl.handle -> domid -> unit
+(** Resume fast a domain *)
+
 val set_action_request : xs:Xenstore.Xs.xsh -> domid -> string option -> unit
 (** [set_action_request xs domid None] declares this domain is fully intact. Any
     other string is a hint to the toolstack that the domain is still broken. *)

ocaml/xenopsd/xc/xenops_server_xen.ml

@@ -1786,6 +1786,15 @@ module VM = struct
         Domain.unpause ~xc di.Xenctrl.domid
     )
 
+  (** To be able to call resume fast the domain should have been suspended so we
+      need to ensure that it is the case. Fast resume is calling resume cooperative.
+      This can only be used for guests which can handle the special return code. *)
+  let resume_fast t vm =
+    on_domain t vm (fun xc _ _ _ di ->
+        (* TODO: check if we can get info in xenstore that the VM supports fast resume *)
+        Domain.resume_fast ~xc di.Xenctrl.domid
+    )
+
   let set_xsdata task vm xsdata =
     on_domain task vm (fun _ xs _ _ di ->
         Domain.set_xsdata ~xs di.Xenctrl.domid xsdata