vm: implement lazy-attach cow-duplication of vm-objects attached to a controller

include/kernel/syscall.h

@@ -211,7 +211,7 @@ extern kern_status_t sys_kern_object_query(
 extern kern_status_t sys_vm_controller_create(kern_handle_t *out);
 extern kern_status_t sys_vm_controller_recv(
 	kern_handle_t ctrl,
-	equeue_packet_page_request_t *out);
+	equeue_packet_vm_request_t *out);
 extern kern_status_t sys_vm_controller_recv_async(
 	kern_handle_t ctrl,
 	kern_handle_t eq,
@@ -224,6 +224,14 @@ extern kern_status_t sys_vm_controller_create_object(
 	size_t data_len,
 	vm_prot_t prot,
 	kern_handle_t *out);
+extern kern_status_t sys_vm_controller_prepare_attach(
+	kern_handle_t ctrl,
+	uint64_t req_id,
+	kern_handle_t *out_vmo);
+extern kern_status_t sys_vm_controller_finish_attach(
+	kern_handle_t ctrl,
+	uint64_t req_id,
+	equeue_key_t new_key);
 extern kern_status_t sys_vm_controller_detach_object(
 	kern_handle_t ctrl,
 	kern_handle_t vmo);

include/kernel/vm-controller.h

@@ -9,34 +9,35 @@ struct thread;
 struct equeue;
 struct vm_object;
 
-enum page_request_status {
-	PAGE_REQUEST_PENDING = 0,
-	PAGE_REQUEST_IN_PROGRESS,
-	PAGE_REQUEST_COMPLETE,
-	PAGE_REQUEST_ASYNC,
+enum vm_request_status {
+	VM_REQUEST_PENDING = 0,
+	VM_REQUEST_IN_PROGRESS,
+	VM_REQUEST_COMPLETE,
+	VM_REQUEST_ASYNC,
 };
 
 struct vm_controller {
 	struct object vc_base;
-	/* tree of pending page requests */
+	/* tree of pending vm requests */
 	struct btree vc_requests;
-	/* the equeue to send async page requests to */
 	struct equeue *vc_eq;
 	equeue_key_t vc_eq_key;
-	/* the number of page requests queued with status PAGE_REQUEST_PENDING.
+	/* the number of page requests queued with status VM_REQUEST_PENDING.
 	 * used to assert/clear VM_CONTROLLER_SIGNAL_REQUEST_RECEIVED */
 	size_t vc_requests_waiting;
 };
 
-struct page_request {
+struct vm_request {
 	uint64_t req_id;
 	unsigned int req_type;
-	enum page_request_status req_status;
+	enum vm_request_status req_status;
 	kern_status_t req_result;
 	spin_lock_t req_lock;
-	equeue_key_t req_object;
+	struct vm_object *req_object;
 	struct thread *req_sender;
+	/* this node is added to vm-controller vc_requests list */
 	struct btree_node req_node;
+	/* these values are used for VM_REQUEST_READ and VM_REQUEST_DIRTY */
 	off_t req_offset;
 	size_t req_length;
 };
@@ -48,7 +49,7 @@ extern struct vm_controller *vm_controller_create(void);
 
 extern kern_status_t vm_controller_recv(
 	struct vm_controller *ctrl,
-	equeue_packet_page_request_t *out);
+	equeue_packet_vm_request_t *out);
 extern kern_status_t vm_controller_recv_async(
 	struct vm_controller *ctrl,
 	struct equeue *eq,
@@ -62,6 +63,14 @@ extern kern_status_t vm_controller_create_object(
 	size_t data_len,
 	vm_prot_t prot,
 	struct vm_object **out);
+extern kern_status_t vm_controller_prepare_attach(
+	struct vm_controller *ctrl,
+	uint64_t req_id,
+	struct vm_object **out_vmo);
+extern kern_status_t vm_controller_finish_attach(
+	struct vm_controller *ctrl,
+	uint64_t req_id,
+	equeue_key_t new_key);
 extern kern_status_t vm_controller_detach_object(
 	struct vm_controller *ctrl,
 	struct vm_object *vmo);
@@ -72,10 +81,16 @@ extern kern_status_t vm_controller_supply_pages(
 	struct vm_object *src,
 	off_t src_offset,
 	size_t count);
+extern void vm_controller_fulfill_requests(
+	struct vm_controller *ctrl,
+	equeue_key_t object,
+	off_t offset,
+	size_t length,
+	kern_status_t result);
 
 extern kern_status_t vm_controller_send_request(
 	struct vm_controller *ctrl,
-	struct page_request *req,
+	struct vm_request *req,
 	unsigned long *irq_flags);
 
 DEFINE_OBJECT_LOCK_FUNCTION(vm_controller, vc_base)

include/kernel/vm-object.h

@@ -20,19 +20,25 @@ enum vm_object_flags {
 	 * be detached, allowing the server to close the last handle to the
 	 * object and dispose of it. */
 	VMO_AUTO_DETACH = 0x04u,
+	/* this vmo is a duplicate of a vmo that is attached to a vm-controller.
+	 * the duplicate vmo is scheduled to be attached to the same controller,
+	 * but this won't actually happen until the controller is needed to
+	 * fulfill a page request. once the duplicate vmo has been attached to
+	 * the controller, this flag will be cleared. */
+	VMO_LAZY_ATTACH = 0x08u,
 
 	/* these flags are for use with vm_object_get_page */
 	/**************************************************/
 
 	/* if the relevant page hasn't been allocated yet, it will be allocated
 	 * and returned. if this flag isn't specified, NULL will be returned. */
-	VMO_ALLOCATE_MISSING_PAGE = 0x08u,
+	VMO_ALLOCATE_MISSING_PAGE = 0x0100u,
 	/* if the vm-object is attached to a vm-controller, and the relevant
 	 * page is uncommitted, send a request to the vm-controller to provide
 	 * the missing page. will result in the vm-object being unlocked and
 	 * the current thread sleeping until the request is fulfilled. the
 	 * vm-object will be re-locked before the function returns. */
-	VMO_REQUEST_MISSING_PAGE = 0x10u,
+	VMO_REQUEST_MISSING_PAGE = 0x0200u,
 };
 
 struct vm_object {
@@ -45,7 +51,7 @@ struct vm_object {
 	struct queue vo_mappings;
 
 	struct vm_controller *vo_ctrl;
-	equeue_key_t vo_key;
+	equeue_key_t vo_key, vo_src_key;
 	struct btree_node vo_ctrl_node;
 
 	/* memory protection flags. mappings of this vm_object can only
@@ -86,6 +92,11 @@ extern struct vm_object *vm_object_create_in_place(
 
 /* create a copy-on-write duplicate of a vm-object */
 extern struct vm_object *vm_object_duplicate_cow(struct vm_object *vmo);
+/* attach a copy-on-write duplicate of a vm-object to the vm-controller that
+ * controlled the original vm-object */
+extern kern_status_t vm_object_attach_cow(
+	struct vm_object *vmo,
+	unsigned long *irq_flags);
 
 extern struct vm_page *vm_object_get_page(
 	struct vm_object *vo,

syscall/dispatch.c

@@ -47,6 +47,12 @@ static const virt_addr_t syscall_table[] = {
 	SYSCALL_TABLE_ENTRY(
 		VM_CONTROLLER_CREATE_OBJECT,
 		vm_controller_create_object),
+	SYSCALL_TABLE_ENTRY(
+		VM_CONTROLLER_PREPARE_ATTACH,
+		vm_controller_prepare_attach),
+	SYSCALL_TABLE_ENTRY(
+		VM_CONTROLLER_FINISH_ATTACH,
+		vm_controller_finish_attach),
 	SYSCALL_TABLE_ENTRY(
 		VM_CONTROLLER_DETACH_OBJECT,
 		vm_controller_detach_object),

syscall/task.c

@@ -506,8 +506,9 @@ kern_status_t sys_task_duplicate(
 	struct task *new_task = task_create(
 		self->t_name,
 		strlen(self->t_name),
-		NULL);
+		child_handle_table);
 	if (!new_task) {
+		handle_table_destroy(child_handle_table);
 		put_current_task(self);
 		return KERN_NO_MEMORY;
 	}

syscall/vm-controller.c

@@ -32,7 +32,7 @@ kern_status_t sys_vm_controller_create(kern_handle_t *out)
 
 kern_status_t sys_vm_controller_recv(
 	kern_handle_t ctrl_handle,
-	equeue_packet_page_request_t *out)
+	equeue_packet_vm_request_t *out)
 {
 	struct task *self = get_current_task();
 
@@ -209,6 +209,116 @@ kern_status_t sys_vm_controller_create_object(
 	return KERN_OK;
 }
 
+kern_status_t sys_vm_controller_prepare_attach(
+	kern_handle_t ctrl_handle,
+	uint64_t req_id,
+	kern_handle_t *out_vmo)
+{
+	struct task *self = get_current_task();
+	if (!out_vmo || !validate_access_w(self, out_vmo, sizeof *out_vmo)) {
+		return KERN_MEMORY_FAULT;
+	}
+
+	kern_status_t status = KERN_OK;
+	unsigned long flags;
+	task_lock_irqsave(self, &flags);
+
+	struct object *ctrl_obj = NULL;
+	handle_flags_t ctrl_flags = 0;
+	status = task_resolve_handle(self, ctrl_handle, &ctrl_obj, &ctrl_flags);
+	if (status != KERN_OK) {
+		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
+		return status;
+	}
+
+	struct handle *out_slot = NULL;
+	kern_handle_t out_handle = KERN_HANDLE_INVALID;
+	status = handle_table_alloc_handle(
+		self->t_handles,
+		KERN_HANDLE_INVALID,
+		&out_slot,
+		&out_handle);
+	if (status != KERN_OK) {
+		object_unref(ctrl_obj);
+		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
+		return status;
+	}
+
+	struct vm_controller *ctrl = vm_controller_cast(ctrl_obj);
+	task_unlock_irqrestore(self, flags);
+
+	if (!ctrl) {
+		object_unref(ctrl_obj);
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	vm_controller_lock_irqsave(ctrl, &flags);
+
+	struct vm_object *vmo = NULL;
+	status = vm_controller_prepare_attach(ctrl, req_id, &vmo);
+	vm_controller_unlock_irqrestore(ctrl, flags);
+
+	object_unref(ctrl_obj);
+	if (status != KERN_OK) {
+		task_lock_irqsave(self, &flags);
+		handle_table_free_handle(self->t_handles, out_handle);
+		task_unlock_irqrestore(self, flags);
+
+		return status;
+	}
+
+	out_slot->h_object = &vmo->vo_base;
+	put_current_task(self);
+
+	*out_vmo = out_handle;
+
+	return KERN_OK;
+}
+
+kern_status_t sys_vm_controller_finish_attach(
+	kern_handle_t ctrl_handle,
+	uint64_t req_id,
+	equeue_key_t new_key)
+{
+	struct task *self = get_current_task();
+
+	kern_status_t status = KERN_OK;
+	unsigned long flags;
+	task_lock_irqsave(self, &flags);
+
+	struct object *ctrl_obj = NULL;
+	handle_flags_t ctrl_flags = 0;
+	status = task_resolve_handle(self, ctrl_handle, &ctrl_obj, &ctrl_flags);
+	if (status != KERN_OK) {
+		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
+		return status;
+	}
+
+	struct vm_controller *ctrl = vm_controller_cast(ctrl_obj);
+	task_unlock_irqrestore(self, flags);
+	put_current_task(self);
+
+	if (!ctrl) {
+		object_unref(ctrl_obj);
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	vm_controller_lock_irqsave(ctrl, &flags);
+
+	status = vm_controller_finish_attach(ctrl, req_id, new_key);
+	vm_controller_unlock_irqrestore(ctrl, flags);
+
+	object_unref(ctrl_obj);
+	if (status != KERN_OK) {
+		return status;
+	}
+
+	return KERN_OK;
+}
+
 kern_status_t sys_vm_controller_detach_object(
 	kern_handle_t ctrl_handle,
 	kern_handle_t vmo_handle)
@@ -315,6 +425,8 @@ kern_status_t sys_vm_controller_supply_pages(
 
 	vm_controller_lock_irqsave(ctrl, &flags);
 	vm_object_lock_pair(src, dst);
+	equeue_key_t requester_key = dst->vo_key;
+
 	status = vm_controller_supply_pages(
 		ctrl,
 		dst,
@@ -323,6 +435,12 @@ kern_status_t sys_vm_controller_supply_pages(
 		src_offset,
 		count);
 	vm_object_unlock_pair(src, dst);
+	vm_controller_fulfill_requests(
+		ctrl,
+		requester_key,
+		dst_offset,
+		count,
+		status);
 	vm_controller_unlock_irqrestore(ctrl, flags);
 
 	object_unref(ctrl_obj);

vm/address-space.c

@@ -6,6 +6,7 @@
 #include <kernel/printk.h>
 #include <kernel/sched.h>
 #include <kernel/task.h>
+#include <kernel/thread.h>
 #include <kernel/util.h>
 #include <kernel/vm-controller.h>
 #include <kernel/vm-object.h>
@@ -1502,6 +1503,17 @@ static kern_status_t request_missing_page(
 	vm_object_lock(object);
 	address_space_unlock(region);
 
+	kern_status_t status = KERN_OK;
+
+	if (object->vo_flags & VMO_LAZY_ATTACH) {
+		status = vm_object_attach_cow(object, irq_flags);
+	}
+
+	if (status != KERN_OK) {
+		vm_object_unlock_irqrestore(object, *irq_flags);
+		return status;
+	}
+
 	struct vm_page *pg = vm_object_get_page(
 		object,
 		object_offset,
@@ -1515,7 +1527,7 @@ static kern_status_t request_missing_page(
 
 	/* now: `region` is unlocked, and `object` is locked */
 
-	kern_status_t status = pmap_add(
+	status = pmap_add(
 		region->s_pmap,
 		addr,
 		vm_page_get_pfn(pg),
@@ -1550,9 +1562,6 @@ static kern_status_t handle_cow_access(
 	}
 
 	tracek("cow access %zx", addr);
-	if (area->vma_object->vo_ctrl) {
-		panic("COW on controlled vm-object");
-	}
 
 	off_t object_offset = addr - area->vma_base + area->vma_object_offset;
 	vm_object_lock(area->vma_object);
@@ -1663,7 +1672,7 @@ kern_status_t address_space_demand_map(
 	address_space_lock_irqsave(region, &irq_flags);
 
 	const enum pmap_fault_flags cow_flags
-		= PMAP_FAULT_WRITE | PMAP_FAULT_PRESENT | PMAP_FAULT_USER;
+		= PMAP_FAULT_WRITE | PMAP_FAULT_PRESENT;
 
 	if ((flags & cow_flags) == cow_flags) {
 		return handle_cow_access(region, addr, flags, &irq_flags);

vm/vm-controller.c

@@ -9,13 +9,12 @@
 #define VM_CONTROLLER_CAST(p)                                                  \
 	OBJECT_C_CAST(struct vm_controller, vc_base, &vm_controller_type, p)
 
-BTREE_DEFINE_SIMPLE_INSERT(struct vm_object, vo_ctrl_node, vo_key, put_object)
 BTREE_DEFINE_SIMPLE_GET(
-	struct vm_object,
-	equeue_key_t,
-	vo_ctrl_node,
-	vo_key,
-	get_object)
+	struct vm_request,
+	uint64_t,
+	req_node,
+	req_id,
+	get_request)
 
 static struct object_type vm_controller_type = {
 	.ob_name = "vm-controller",
@@ -23,14 +22,14 @@ static struct object_type vm_controller_type = {
 	.ob_header_offset = offsetof(struct vm_controller, vc_base),
 };
 
-static struct vm_cache page_request_cache = {
-	.c_name = "page-request",
-	.c_obj_size = sizeof(struct page_request),
+static struct vm_cache vm_request_cache = {
+	.c_name = "vm-request",
+	.c_obj_size = sizeof(struct vm_request),
 };
 
 kern_status_t vm_controller_type_init(void)
 {
-	vm_cache_init(&page_request_cache);
+	vm_cache_init(&vm_request_cache);
 	return object_type_register(&vm_controller_type);
 }
 
@@ -51,19 +50,19 @@ struct vm_controller *vm_controller_create(void)
 	return ctrl;
 }
 
-static struct page_request *get_next_request(struct vm_controller *ctrl)
+static struct vm_request *get_next_request(struct vm_controller *ctrl)
 {
 	struct btree_node *cur = btree_first(&ctrl->vc_requests);
 	while (cur) {
-		struct page_request *req
-			= BTREE_CONTAINER(struct page_request, req_node, cur);
+		struct vm_request *req
+			= BTREE_CONTAINER(struct vm_request, req_node, cur);
 		spin_lock(&req->req_lock);
 		switch (req->req_status) {
-		case PAGE_REQUEST_PENDING:
-			req->req_status = PAGE_REQUEST_IN_PROGRESS;
+		case VM_REQUEST_PENDING:
+			req->req_status = VM_REQUEST_IN_PROGRESS;
 			ctrl->vc_requests_waiting--;
 			return req;
-		case PAGE_REQUEST_ASYNC:
+		case VM_REQUEST_ASYNC:
 			btree_delete(&ctrl->vc_requests, &req->req_node);
 			ctrl->vc_requests_waiting--;
 			return req;
@@ -78,7 +77,7 @@ static struct page_request *get_next_request(struct vm_controller *ctrl)
 	return NULL;
 }
 
-static kern_status_t try_enqueue(struct btree *tree, struct page_request *req)
+static kern_status_t try_enqueue(struct btree *tree, struct vm_request *req)
 {
 	if (!tree->b_root) {
 		tree->b_root = &req->req_node;
@@ -88,8 +87,8 @@ static kern_status_t try_enqueue(struct btree *tree, struct page_request *req)
 
 	struct btree_node *cur = tree->b_root;
 	while (1) {
-		struct page_request *cur_node
-			= BTREE_CONTAINER(struct page_request, req_node, cur);
+		struct vm_request *cur_node
+			= BTREE_CONTAINER(struct vm_request, req_node, cur);
 		struct btree_node *next = NULL;
 
 		if (req->req_id > cur_node->req_id) {
@@ -119,7 +118,7 @@ static kern_status_t try_enqueue(struct btree *tree, struct page_request *req)
 
 static kern_status_t send_request_async(
 	struct vm_controller *ctrl,
-	struct page_request *req)
+	struct vm_request *req)
 {
 	fill_random(&req->req_id, sizeof req->req_id);
 	while (!try_enqueue(&ctrl->vc_requests, req)) {
@@ -136,9 +135,9 @@ static kern_status_t send_request_async(
 
 kern_status_t vm_controller_recv(
 	struct vm_controller *ctrl,
-	equeue_packet_page_request_t *out)
+	equeue_packet_vm_request_t *out)
 {
-	struct page_request *req = NULL;
+	struct vm_request *req = NULL;
 
 	req = get_next_request(ctrl);
 	if (!req) {
@@ -151,16 +150,30 @@ kern_status_t vm_controller_recv(
 			VM_CONTROLLER_SIGNAL_REQUEST_RECEIVED);
 	}
 
-	out->req_vmo = req->req_object;
+	vm_object_lock(req->req_object);
+	out->req_id = req->req_id;
+	out->req_vmo = req->req_object->vo_key;
 	out->req_type = req->req_type;
-	out->req_offset = req->req_offset;
-	out->req_length = req->req_length;
 
+	switch (req->req_type) {
+	case VM_REQUEST_READ:
+	case VM_REQUEST_DIRTY:
+		out->req_offset = req->req_offset;
+		out->req_length = req->req_length;
+		break;
+	case VM_REQUEST_ATTACH:
+		out->req_src_vmo = req->req_object->vo_src_key;
+		break;
+	default:
+		break;
+	}
+
+	vm_object_unlock(req->req_object);
 	spin_unlock(&req->req_lock);
 
-	if (req->req_status == PAGE_REQUEST_ASYNC) {
+	if (req->req_status == VM_REQUEST_ASYNC) {
 		put_current_thread(req->req_sender);
-		vm_cache_free(&page_request_cache, req);
+		vm_cache_free(&vm_request_cache, req);
 	}
 
 	return KERN_OK;
@@ -208,6 +221,53 @@ kern_status_t vm_controller_create_object(
 	return KERN_OK;
 }
 
+kern_status_t vm_controller_prepare_attach(
+	struct vm_controller *ctrl,
+	uint64_t req_id,
+	struct vm_object **out_vmo)
+{
+	struct vm_request *req = get_request(&ctrl->vc_requests, req_id);
+	if (!req) {
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	spin_lock(&req->req_lock);
+	req->req_status = VM_REQUEST_IN_PROGRESS;
+	*out_vmo = req->req_object;
+	spin_unlock(&req->req_lock);
+
+	return KERN_OK;
+}
+
+kern_status_t vm_controller_finish_attach(
+	struct vm_controller *ctrl,
+	uint64_t req_id,
+	equeue_key_t new_key)
+{
+	struct vm_request *req = get_request(&ctrl->vc_requests, req_id);
+	if (!req) {
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	spin_lock(&req->req_lock);
+	struct vm_object *vmo = req->req_object;
+	spin_unlock(&req->req_lock);
+
+	vm_object_lock(vmo);
+	vmo->vo_key = new_key;
+	vmo->vo_src_key = 0;
+	vmo->vo_flags &= ~VMO_LAZY_ATTACH;
+	vm_object_unlock(vmo);
+
+	spin_lock(&req->req_lock);
+	req->req_status = VM_REQUEST_COMPLETE;
+	req->req_result = KERN_OK;
+	thread_awaken(req->req_sender);
+	spin_unlock(&req->req_lock);
+
+	return KERN_OK;
+}
+
 kern_status_t vm_controller_detach_object(
 	struct vm_controller *ctrl,
 	struct vm_object *vmo)
@@ -216,7 +276,7 @@ kern_status_t vm_controller_detach_object(
 		return KERN_INVALID_ARGUMENT;
 	}
 
-	if (vmo->vo_key == 0) {
+	if (vmo->vo_flags & VMO_LAZY_ATTACH) {
 		/* this vmo isn't actually attached to this controller yet.
 		 * this can happen if a controller-attached vmo was duplicated
 		 * via copy-on-write, and the duplicate vmo has not yet been
@@ -225,16 +285,14 @@ kern_status_t vm_controller_detach_object(
 		return KERN_OK;
 	}
 
-	struct page_request *req
-		= vm_cache_alloc(&page_request_cache, VM_NORMAL);
-	req->req_type = PAGE_REQUEST_DETACH;
-	req->req_status = PAGE_REQUEST_ASYNC;
-	req->req_object = vmo->vo_key;
+	struct vm_request *req = vm_cache_alloc(&vm_request_cache, VM_NORMAL);
+	req->req_type = VM_REQUEST_DETACH;
+	req->req_status = VM_REQUEST_ASYNC;
+	req->req_object = vmo;
 	req->req_sender = get_current_thread();
 	send_request_async(ctrl, req);
 
 	vmo->vo_ctrl = NULL;
-	vmo->vo_key = 0;
 	object_unref(&ctrl->vc_base);
 
 	return KERN_OK;
@@ -242,7 +300,7 @@ kern_status_t vm_controller_detach_object(
 
 static void wait_for_reply(
 	struct vm_controller *ctrl,
-	struct page_request *req,
+	struct vm_request *req,
 	unsigned long *lock_flags)
 {
 	struct wait_item waiter;
@@ -251,7 +309,7 @@ static void wait_for_reply(
 	wait_item_init(&waiter, self);
 	for (;;) {
 		self->tr_state = THREAD_SLEEPING;
-		if (req->req_status == PAGE_REQUEST_COMPLETE) {
+		if (req->req_status == VM_REQUEST_COMPLETE) {
 			break;
 		}
 
@@ -264,7 +322,7 @@ static void wait_for_reply(
 	put_current_thread(self);
 }
 
-static void fulfill_requests(
+void vm_controller_fulfill_requests(
 	struct vm_controller *ctrl,
 	equeue_key_t object,
 	off_t offset,
@@ -274,8 +332,8 @@ static void fulfill_requests(
 	off_t limit = offset + length - 1;
 	struct btree_node *cur = btree_first(&ctrl->vc_requests);
 	while (cur) {
-		struct page_request *req
-			= BTREE_CONTAINER(struct page_request, req_node, cur);
+		struct vm_request *req
+			= BTREE_CONTAINER(struct vm_request, req_node, cur);
 		spin_lock(&req->req_lock);
 		bool match = false;
 		off_t req_base = req->req_offset;
@@ -287,12 +345,14 @@ static void fulfill_requests(
 			match = true;
 		}
 
-		if (req->req_object != object) {
+		vm_object_lock(req->req_object);
+		if (req->req_object->vo_key != object) {
 			match = false;
 		}
+		vm_object_unlock(req->req_object);
 
 		if (match) {
-			req->req_status = PAGE_REQUEST_COMPLETE;
+			req->req_status = VM_REQUEST_COMPLETE;
 			req->req_result = result;
 			thread_awaken(req->req_sender);
 		}
@@ -325,14 +385,13 @@ kern_status_t vm_controller_supply_pages(
 		src_offset,
 		count,
 		NULL);
-	fulfill_requests(ctrl, dst->vo_key, dst_offset, count, status);
 
 	return status;
 }
 
 kern_status_t vm_controller_send_request(
 	struct vm_controller *ctrl,
-	struct page_request *req,
+	struct vm_request *req,
 	unsigned long *irq_flags)
 {
 	fill_random(&req->req_id, sizeof req->req_id);

vm/vm-object.c

@@ -288,8 +288,10 @@ struct vm_object *vm_object_duplicate_cow(struct vm_object *vmo)
 	struct vm_object *out = VM_OBJECT_CAST(obj);
 
 	memcpy(out->vo_name, vmo->vo_name, sizeof out->vo_name);
-	out->vo_flags = vmo->vo_flags;
+	out->vo_flags = vmo->vo_flags | VMO_LAZY_ATTACH;
 	out->vo_ctrl = vmo->vo_ctrl;
+	out->vo_key = 0;
+	out->vo_src_key = vmo->vo_key;
 	out->vo_prot = vmo->vo_prot;
 	out->vo_size = vmo->vo_size;
 	memcpy(out->vo_name, vmo->vo_name, sizeof vmo->vo_name);
@@ -315,6 +317,37 @@ struct vm_object *vm_object_duplicate_cow(struct vm_object *vmo)
 	return out;
 }
 
+kern_status_t vm_object_attach_cow(
+	struct vm_object *vmo,
+	unsigned long *irq_flags)
+{
+	struct vm_controller *ctrl = vmo->vo_ctrl;
+	struct vm_request req = {0};
+	req.req_status = VM_REQUEST_PENDING;
+	req.req_type = VM_REQUEST_ATTACH;
+	req.req_length = vm_page_order_to_bytes(VM_PAGE_4K);
+	req.req_sender = get_current_thread();
+
+	object_ref(&vmo->vo_base);
+	req.req_object = vmo;
+
+	vm_object_unlock_irqrestore(vmo, *irq_flags);
+	vm_controller_lock_irqsave(ctrl, irq_flags);
+
+	spin_lock(&req.req_lock);
+
+	kern_status_t status
+		= vm_controller_send_request(ctrl, &req, irq_flags);
+
+	put_current_thread(req.req_sender);
+	spin_unlock(&req.req_lock);
+	vm_controller_unlock_irqrestore(ctrl, *irq_flags);
+	object_unref(&vmo->vo_base);
+	vm_object_lock_irqsave(vmo, irq_flags);
+
+	return status;
+}
+
 static struct vm_page *alloc_page(struct vm_object *vo, off_t offset)
 {
 	struct vm_page *page = NULL;
@@ -458,15 +491,15 @@ static kern_status_t request_page(
 	unsigned long *irq_flags)
 {
 	struct vm_controller *ctrl = vo->vo_ctrl;
-	struct page_request req = {0};
-	req.req_status = PAGE_REQUEST_PENDING;
-	req.req_type = PAGE_REQUEST_READ;
+	struct vm_request req = {0};
+	req.req_status = VM_REQUEST_PENDING;
+	req.req_type = VM_REQUEST_READ;
 	req.req_offset = offset;
 	req.req_length = vm_page_order_to_bytes(VM_PAGE_4K);
 	req.req_sender = get_current_thread();
 
 	object_ref(&vo->vo_base);
-	req.req_object = vo->vo_key;
+	req.req_object = vo;
 
 	vm_object_unlock_irqrestore(vo, *irq_flags);
 	vm_controller_lock_irqsave(ctrl, irq_flags);