libmango: types: add macro to define a kern_msg_handle_t

when locking a pair of objects, the object with the lesser memory address
is always locked first. the pair is unlocked in the opposite order.

arch/user/hwlock.c

@@ -1,5 +1,5 @@
-#include <kernel/machine/hwlock.h>
 #include <kernel/compiler.h>
+#include <kernel/machine/hwlock.h>
 
 void ml_hwlock_lock(ml_hwlock_t *lck)
 {

arch/x86_64/config.cmake

@@ -2,9 +2,3 @@ target_compile_options(${kernel_exe_name} PRIVATE
 	-z max-page-size=0x1000 -m64 -mcmodel=large -mno-red-zone -mno-mmx
 	-mno-sse -mno-sse2 -D_64BIT -DBYTE_ORDER=1234)
 target_link_libraries(${kernel_exe_name} "-z max-page-size=0x1000" "-T ${CMAKE_CURRENT_SOURCE_DIR}/arch/x86_64/layout.ld")
-
-add_custom_command(TARGET ${kernel_exe_name} POST_BUILD
-	COMMAND ${BUILD_TOOLS_DIR}/e64patch $<TARGET_FILE:${kernel_exe_name}>
-	WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}
-	COMMENT "Patching kernel elf64 image"
-)

arch/x86_64/hwlock.S

@@ -11,6 +11,41 @@ ml_hwlock_lock:
 
 	mov $1, %ecx
 
+	mfence
+
+1:	mov $0, %eax
+	lock cmpxchg %ecx, (%rdi)
+	jne 1b
+
+	pop %rbp
+	ret
+
+
+	.global ml_hwlock_unlock
+	.type ml_hwlock_unlock, @function
+
+	/* %rdi = pointer to ml_hwlock_t (int) */
+ml_hwlock_unlock:
+	push %rbp
+	mov %rsp, %rbp
+
+	movl $0, (%rdi)
+	mfence
+
+	pop %rbp
+	ret
+
+
+	.global ml_hwlock_lock_irq
+	.type ml_hwlock_lock_irq, @function
+
+	/* %rdi = pointer to ml_hwlock_t (int) */
+ml_hwlock_lock_irq:
+	push %rbp
+	mov %rsp, %rbp
+
+	mov $1, %ecx
+
 	cli
 	mfence
 
@@ -21,11 +56,12 @@ ml_hwlock_lock:
 	pop %rbp
 	ret
 
-	.global ml_hwlock_unlock
+
-	.type ml_hwlock_unlock, @function
+	.global ml_hwlock_unlock_irq
+	.type ml_hwlock_unlock_irq, @function
 
 	/* %rdi = pointer to ml_hwlock_t (int) */
-ml_hwlock_unlock:
+ml_hwlock_unlock_irq:
 	push %rbp
 	mov %rsp, %rbp
 
@@ -62,6 +98,7 @@ ml_hwlock_lock_irqsave:
 	pop %rbp
 	ret
 
+
 	.global ml_hwlock_unlock_irqrestore
 	.type ml_hwlock_unlock_irqrestore, @function
 

arch/x86_64/include/arch/msr.h

@@ -4,6 +4,7 @@
 #include <stdint.h>
 
 #define MSR_GS_BASE        0xC0000101
+#define MSR_KERNEL_GS_BASE 0xC0000102
 
 #ifdef __cplusplus
 extern "C" {

arch/x86_64/include/kernel/machine/hwlock.h

@@ -12,6 +12,9 @@ typedef int ml_hwlock_t;
 extern void ml_hwlock_lock(ml_hwlock_t *lck);
 extern void ml_hwlock_unlock(ml_hwlock_t *lck);
 
+extern void ml_hwlock_lock_irq(ml_hwlock_t *lck);
+extern void ml_hwlock_unlock_irq(ml_hwlock_t *lck);
+
 extern void ml_hwlock_lock_irqsave(ml_hwlock_t *lck, unsigned long *flags);
 extern void ml_hwlock_unlock_irqrestore(ml_hwlock_t *lck, unsigned long flags);
 

arch/x86_64/init.c

@@ -20,6 +20,19 @@
 
 #define PTR32(x)            ((void *)((uintptr_t)(x)))
 
+/* the physical address of the start of the memblock heap.
+ * this is an arbirary value; the heap can start anywhere in memory.
+ * any reserved areas of memory (the kernel, bsp, bios data, etc) are
+ * automatically taken into account.
+ * HOWEVER, this value will dictate how much physical memory is required for
+ * the kernel to boot successfully.
+ * the value of 16MiB (0x1000000) means that all heap allocations will be
+ * above 16MiB, leaving the area below free for DMA operations.
+ * this value CAN be reduced all the way to zero to minimise the amount of
+ * memory required to boot, but this may leave you with no DMA memory available.
+ */
+#define MEMBLOCK_HEAP_START 0x1000000
+
 static ml_cpu_block g_bootstrap_cpu = {0};
 
 /* start and end of kernel image (physical addresses) */
@@ -33,7 +46,7 @@ static void bootstrap_cpu_init(void)
 
 static void early_vm_init(uintptr_t reserve_end)
 {
-	uintptr_t alloc_start = VM_KERNEL_VOFFSET;
+	uintptr_t alloc_start = VM_KERNEL_VOFFSET + MEMBLOCK_HEAP_START;
 	/* boot code mapped 2 GiB of memory from
 	   VM_KERNEL_VOFFSET */
 	uintptr_t alloc_end = VM_KERNEL_VOFFSET + 0x7fffffff;

arch/x86_64/irqvec.S

@@ -333,11 +333,30 @@ IRQ 223, 255
 isr_common_stub:
 	PUSH_REGS
 
+	# When ISR occurs in Ring 3, CPU sets %ss (and other non-code selectors)
+	# to 0.
+	mov %ss, %ax
+	cmp $0, %ax
+	jne isr_skipgs1
+
+	mov $0x10, %ax
+	mov %ax, %ss
+	swapgs
+
+isr_skipgs1:
 	mov %rsp, %rdi
 	call isr_dispatch
 
 	POP_REGS
 	add $16, %rsp
+
+	cmpq $0x1b, 32(%rsp)
+	jne isr_skipgs2
+
+	swapgs
+
+isr_skipgs2:
+
 	iretq
 
 
@@ -347,11 +366,31 @@ isr_common_stub:
 irq_common_stub:
 	PUSH_REGS
 
+	# When IRQ occurs in Ring 3, CPU sets %ss (and other non-code selectors)
+	# to 0.
+	mov %ss, %ax
+	cmp $0, %ax
+	jne irq_skipgs1
+
+	mov $0x10, %ax
+	mov %ax, %ss
+	swapgs
+
+irq_skipgs1:
 	mov %rsp, %rdi
 	call irq_dispatch
 
+
 	POP_REGS
 	add $16, %rsp
+
+	cmpq $0x1b, 32(%rsp)
+	jne isr_skipgs2
+
+	swapgs
+
+irq_skipgs2:
+
 	iretq
 
 
@@ -363,12 +402,12 @@ irq_common_stub:
 
 syscall_gate:
 	swapgs
-	movq %rsp, %gs:20       # GS+20 = rsp2 in the current TSS block (user stack storage)
+	movq %rsp, %gs:94       # GS+20 = rsp2 in the current TSS block (user stack storage)
-	movq %gs:4, %rsp	# GS+4  = rsp0 in the current TSS block (per-thread kstack)
+	movq %gs:78, %rsp	# GS+4  = rsp0 in the current TSS block (per-thread kstack)
 
 	# start building a ml_cpu_context
 	pushq $0x1b
-	pushq %gs:20
+	pushq %gs:94
 	push %r11
 	push $0x23
 	push %rcx
@@ -380,10 +419,6 @@ syscall_gate:
 
 	mov %rsp, %rdi
 
-	# switch back to user gs while in syscall_dispatch. Interrupts are enabled in syscall_dispatch,
-	# and if the task gets pre-empted, the incoming task will expect %gs to have its usermode value.
-	swapgs
-
 	call syscall_dispatch
 
 	POP_REGS
@@ -394,8 +429,8 @@ syscall_gate:
 	pop %r11
 	add $16, %rsp
 
-	swapgs
+	movq %gs:94, %rsp       # GS+20 = rsp2 in the current TSS block
-	movq %gs:20, %rsp       # GS+20 = rsp2 in the current TSS block
+
 	swapgs
 
 	# back to usermode

arch/x86_64/pmap.c

@@ -364,7 +364,12 @@ kern_status_t pmap_handle_fault(
 	struct task *task = current_task();
 	struct vm_region *space = task->t_address_space;
 
-	return vm_region_demand_map(space, fault_addr, flags);
+	unsigned long lock_flags;
+	vm_region_lock_irqsave(space, &lock_flags);
+	kern_status_t status = vm_region_demand_map(space, fault_addr, flags);
+	vm_region_unlock_irqrestore(space, lock_flags);
+
+	return status;
 }
 
 kern_status_t pmap_add(

arch/x86_64/thread_switch.S

@@ -73,4 +73,5 @@ ml_thread_switch_user:
 	pop %rax
 
 	add $16, %rsp
+	swapgs
 	iretq

arch/x86_64/tss.c

@@ -1,5 +1,3 @@
-#include "arch/msr.h"
-
 #include <arch/gdt.h>
 #include <arch/tss.h>
 #include <kernel/libc/string.h>
@@ -22,9 +20,6 @@ void tss_init(struct tss *tss, struct tss_ptr *ptr)
 void tss_load(struct tss *tss)
 {
 	tss_flush(TSS_GDT_INDEX);
-
-	uintptr_t kernel_gs_base_reg = 0xC0000102;
-	wrmsr(kernel_gs_base_reg, (uintptr_t)tss);
 }
 
 virt_addr_t tss_get_kstack(struct tss *tss)

include/kernel/arg.h

@@ -1,8 +1,8 @@
 #ifndef KERNEL_ARG_H_
 #define KERNEL_ARG_H_
 
+#include <mango/types.h>
 #include <stdbool.h>
-#include <mango/status.h>
 
 #define CMDLINE_MAX 4096
 

include/kernel/channel.h

@@ -4,7 +4,7 @@
 #include <kernel/object.h>
 #include <kernel/sched.h>
 
-struct kmsg;
+struct msg;
 
 struct channel {
 	struct object c_base;
@@ -15,37 +15,39 @@ struct channel {
 };
 
 extern kern_status_t channel_type_init(void);
+extern struct channel *channel_cast(struct object *obj);
 
 extern struct channel *channel_create(void);
 
 extern kern_status_t channel_enqueue_msg(
 	struct channel *channel,
-	struct kmsg *msg);
+	struct msg *msg);
 
 extern kern_status_t channel_recv_msg(
 	struct channel *channel,
-	struct msg *out_msg,
+	kern_msg_t *out_msg,
-	msgid_t *out_id,
 	unsigned long *irq_flags);
 extern kern_status_t channel_reply_msg(
 	struct channel *channel,
 	msgid_t id,
-	const struct msg *resp,
+	const kern_msg_t *reply,
 	unsigned long *irq_flags);
 
 extern kern_status_t channel_read_msg(
 	struct channel *channel,
 	msgid_t msg,
 	size_t offset,
-	void *buf,
+	struct vm_region *dest_region,
-	size_t len,
+	const kern_iovec_t *dest_iov,
+	size_t dest_iov_count,
 	size_t *nr_read);
 extern kern_status_t channel_write_msg(
 	struct channel *channel,
 	msgid_t msg,
 	size_t offset,
-	const void *buf,
+	struct vm_region *src_region,
-	size_t len,
+	const kern_iovec_t *src_iov,
+	size_t src_iov_count,
 	size_t *nr_written);
 
 DEFINE_OBJECT_LOCK_FUNCTION(channel, c_base)

include/kernel/console.h

@@ -14,9 +14,10 @@
    representing a serial port may allow both sending AND receiving over the
    port.
 */
-#include <kernel/queue.h>
 #include <kernel/locks.h>
+#include <kernel/queue.h>
 #include <mango/status.h>
+#include <mango/types.h>
 
 #ifdef __cplusplus
 extern "C" {

include/kernel/handle.h

@@ -3,6 +3,7 @@
 
 #include <kernel/bitmap.h>
 #include <mango/status.h>
+#include <mango/types.h>
 #include <stddef.h>
 #include <stdint.h>
 
@@ -16,6 +17,7 @@ typedef uintptr_t handle_flags_t;
 
 struct task;
 struct object;
+struct vm_region;
 struct handle_list;
 
 struct handle {
@@ -54,12 +56,14 @@ extern struct handle *handle_table_get_handle(
 	struct handle_table *tab,
 	kern_handle_t handle);
 
-extern kern_status_t handle_list_transfer(
+extern kern_status_t handle_table_transfer(
-	struct handle_table *dest,
+	struct vm_region *dst_region,
-	struct handle_list *dest_list,
+	struct handle_table *dst,
-	size_t dest_list_count,
+	kern_msg_handle_t *dst_handles,
+	size_t dst_handles_max,
+	struct vm_region *src_region,
 	struct handle_table *src,
-	const struct handle_list *src_list,
+	kern_msg_handle_t *src_handles,
-	size_t src_list_count);
+	size_t src_handles_count);
 
 #endif

include/kernel/iovec.h

@@ -5,7 +5,10 @@
 #include <stddef.h>
 
 struct iovec_iterator {
-	const struct iovec *it_vecs;
+	/* if this is set, we are iterating over a list of iovecs stored in
+	 * userspace, and must go through this region to retrieve the data. */
+	struct vm_region *it_region;
+	const kern_iovec_t *it_vecs;
 	size_t it_nr_vecs;
 	size_t it_vec_ptr;
 
@@ -15,7 +18,12 @@ struct iovec_iterator {
 
 extern void iovec_iterator_begin(
 	struct iovec_iterator *it,
-	const struct iovec *vecs,
+	const kern_iovec_t *vecs,
+	size_t nr_vecs);
+extern void iovec_iterator_begin_user(
+	struct iovec_iterator *it,
+	struct vm_region *address_space,
+	const kern_iovec_t *vecs,
 	size_t nr_vecs);
 
 extern void iovec_iterator_seek(struct iovec_iterator *it, size_t nr_bytes);

include/kernel/locks.h

@@ -15,8 +15,44 @@ typedef __aligned(8) ml_hwlock_t spin_lock_t;
 #define spin_lock(lck)                ml_hwlock_lock(lck);
 #define spin_unlock(lck)              ml_hwlock_unlock(lck);
 
+#define spin_lock_irq(lck)            ml_hwlock_lock_irq(lck);
+#define spin_unlock_irq(lck)          ml_hwlock_unlock_irq(lck);
+
 #define spin_lock_irqsave(lck, flags) ml_hwlock_lock_irqsave(lck, flags);
-#define spin_unlock_irqrestore(lck, flags) ml_hwlock_unlock_irqrestore(lck, flags);
+#define spin_unlock_irqrestore(lck, flags)                                     \
+	ml_hwlock_unlock_irqrestore(lck, flags);
+
+static inline void spin_lock_pair_irqsave(
+	spin_lock_t *a,
+	spin_lock_t *b,
+	unsigned long *flags)
+{
+	if (a == b) {
+		spin_lock_irqsave(a, flags);
+	} else if (a < b) {
+		spin_lock_irqsave(a, flags);
+		spin_lock(b);
+	} else {
+		spin_lock_irqsave(b, flags);
+		spin_lock(a);
+	}
+}
+
+static inline void spin_unlock_pair_irqrestore(
+	spin_lock_t *a,
+	spin_lock_t *b,
+	unsigned long flags)
+{
+	if (a == b) {
+		spin_unlock_irqrestore(a, flags);
+	} else if (a < b) {
+		spin_unlock(b);
+		spin_unlock_irqrestore(a, flags);
+	} else {
+		spin_unlock(a);
+		spin_unlock_irqrestore(b, flags);
+	}
+}
 
 #ifdef __cplusplus
 }

include/kernel/memblock.h

@@ -22,8 +22,8 @@
 #ifndef KERNEL_MEMBLOCK_H_
 #define KERNEL_MEMBLOCK_H_
 
-#include <limits.h>
 #include <kernel/types.h>
+#include <limits.h>
 #include <stddef.h>
 
 #ifdef __cplusplus
@@ -338,6 +338,8 @@ extern void __next_memory_region(
 	phys_addr_t start,
 	phys_addr_t end);
 
+extern void memblock_dump(void);
+
 #ifdef __cplusplus
 }
 #endif

include/kernel/msg.h

@@ -15,7 +15,7 @@ enum kmsg_status {
 	KMSG_REPLY_SENT,
 };
 
-struct kmsg {
+struct msg {
 	spin_lock_t msg_lock;
 	enum kmsg_status msg_status;
 	struct btree_node msg_node;
@@ -23,8 +23,7 @@ struct kmsg {
 	kern_status_t msg_result;
 	struct port *msg_sender_port;
 	struct thread *msg_sender_thread;
-	const struct msg *msg_req;
+	kern_msg_t msg_req, msg_resp;
-	struct msg *msg_resp;
 };
 
 #endif

include/kernel/object.h

@@ -3,8 +3,8 @@
 
 #include <kernel/flags.h>
 #include <kernel/locks.h>
-#include <mango/status.h>
 #include <kernel/vm.h>
+#include <mango/status.h>
 #include <stddef.h>
 
 #ifdef __cplusplus
@@ -31,6 +31,20 @@ extern "C" {
 		unsigned long flags)                                           \
 	{                                                                      \
 		object_unlock_irqrestore(&p->base, flags);                     \
+	}                                                                      \
+	static inline void object_name##_lock_pair_irqsave(                    \
+		struct object_name *a,                                         \
+		struct object_name *b,                                         \
+		unsigned long *flags)                                          \
+	{                                                                      \
+		object_lock_pair_irqsave(&a->base, &b->base, flags);           \
+	}                                                                      \
+	static inline void object_name##_unlock_pair_irqrestore(               \
+		struct object_name *a,                                         \
+		struct object_name *b,                                         \
+		unsigned long flags)                                           \
+	{                                                                      \
+		object_unlock_pair_irqrestore(&a->base, &b->base, flags);      \
 	}
 
 #define OBJECT_MAGIC    0xBADDCAFE
@@ -52,7 +66,10 @@ enum object_type_flags {
 };
 
 struct object_ops {
-	kern_status_t (*destroy)(struct object *obj);
+	kern_status_t (*destroy)(struct object *obj, struct queue *q);
+	kern_status_t (*destroy_recurse)(
+		struct queue_entry *entry,
+		struct object **out);
 };
 
 struct object_type {
@@ -67,6 +84,7 @@ struct object_type {
 
 struct object {
 	uint32_t ob_magic;
+	koid_t ob_id;
 	struct object_type *ob_type;
 	spin_lock_t ob_lock;
 	unsigned int ob_refcount;
@@ -88,6 +106,15 @@ extern void object_unlock(struct object *obj);
 extern void object_lock_irqsave(struct object *obj, unsigned long *flags);
 extern void object_unlock_irqrestore(struct object *obj, unsigned long flags);
 
+extern void object_lock_pair_irqsave(
+	struct object *a,
+	struct object *b,
+	unsigned long *flags);
+extern void object_unlock_pair_irqrestore(
+	struct object *a,
+	struct object *b,
+	unsigned long flags);
+
 #ifdef __cplusplus
 }
 #endif

include/kernel/port.h

@@ -29,10 +29,12 @@ extern struct port *port_cast(struct object *obj);
 extern struct port *port_create(void);
 
 extern kern_status_t port_connect(struct port *port, struct channel *remote);
+extern kern_status_t port_disconnect(struct port *port);
 extern kern_status_t port_send_msg(
 	struct port *port,
-	const struct msg *req,
+	const kern_msg_t *msg,
-	struct msg *resp);
+	kern_msg_t *out_response,
+	unsigned long *lock_flags);
 
 DEFINE_OBJECT_LOCK_FUNCTION(port, p_base)
 

include/kernel/sched.h

@@ -85,6 +85,7 @@ struct task {
 
 	pmap_t t_pmap;
 	struct vm_region *t_address_space;
+	spin_lock_t t_handles_lock;
 	struct handle_table *t_handles;
 	struct btree b_channels;
 
@@ -114,7 +115,7 @@ struct thread {
 	virt_addr_t tr_cpu_user_sp, tr_cpu_kernel_sp;
 
 	struct runqueue *tr_rq;
-	struct kmsg tr_msg;
+	struct msg tr_msg;
 
 	struct queue_entry tr_parent_entry;
 	struct queue_entry tr_rqentry;

include/kernel/syscall.h

@@ -2,16 +2,14 @@
 #define KERNEL_SYSCALL_H_
 
 #include <kernel/handle.h>
+#include <kernel/sched.h>
+#include <kernel/vm-region.h>
 #include <kernel/vm.h>
 #include <mango/status.h>
 #include <mango/syscall.h>
 
 #define validate_access(task, ptr, len, flags)                                 \
-	vm_region_validate_access(                                             \
+	__validate_access(task, (const void *)ptr, len, flags)
-		task->t_address_space,                                         \
-		(virt_addr_t)ptr,                                              \
-		len,                                                           \
-		flags | VM_PROT_USER)
 #define validate_access_r(task, ptr, len)                                      \
 	validate_access(task, ptr, len, VM_PROT_READ | VM_PROT_USER)
 #define validate_access_w(task, ptr, len)                                      \
@@ -23,7 +21,25 @@
 		len,                                                           \
 		VM_PROT_READ | VM_PROT_WRITE | VM_PROT_USER)
 
+static inline bool __validate_access(
+	struct task *task,
+	const void *ptr,
+	size_t len,
+	vm_prot_t flags)
+{
+	unsigned long irq_flags;
+	vm_region_lock_irqsave(task->t_address_space, &irq_flags);
+	bool result = vm_region_validate_access(
+		task->t_address_space,
+		(virt_addr_t)ptr,
+		len,
+		flags | VM_PROT_USER);
+	vm_region_unlock_irqrestore(task->t_address_space, irq_flags);
+	return result;
+}
+
 extern kern_status_t sys_task_exit(int status);
+extern kern_status_t sys_task_self(kern_handle_t *out);
 extern kern_status_t sys_task_create(
 	kern_handle_t parent_handle,
 	const char *name,
@@ -37,6 +53,9 @@ extern kern_status_t sys_task_create_thread(
 	uintptr_t *args,
 	size_t nr_args,
 	kern_handle_t *out_thread);
+extern kern_status_t sys_task_get_address_space(
+	kern_handle_t task,
+	kern_handle_t *out);
 
 extern kern_status_t sys_thread_start(kern_handle_t thread);
 
@@ -75,6 +94,7 @@ extern kern_status_t sys_vm_region_create(
 	vm_prot_t prot,
 	kern_handle_t *out,
 	virt_addr_t *out_base_address);
+extern kern_status_t sys_vm_region_kill(kern_handle_t region);
 extern kern_status_t sys_vm_region_read(
 	kern_handle_t region,
 	void *dst,
@@ -123,10 +143,7 @@ extern kern_status_t sys_kern_config_set(
 	const void *ptr,
 	size_t len);
 
-extern kern_status_t sys_channel_create(
+extern kern_status_t sys_channel_create(unsigned int id, kern_handle_t *out);
-	unsigned int id,
-	channel_flags_t flags,
-	kern_handle_t *out);
 extern kern_status_t sys_port_create(kern_handle_t *out);
 extern kern_status_t sys_port_connect(
 	kern_handle_t port,
@@ -136,47 +153,28 @@ extern kern_status_t sys_port_disconnect(kern_handle_t port);
 
 extern kern_status_t sys_msg_send(
 	kern_handle_t port,
-	msg_flags_t flags,
+	const kern_msg_t *msg,
-	const struct msg *req,
+	kern_msg_t *out_reply);
-	struct msg *resp);
+extern kern_status_t sys_msg_recv(kern_handle_t channel, kern_msg_t *out_msg);
-
-extern kern_status_t sys_msg_recv(
-	kern_handle_t channel,
-	msg_flags_t flags,
-	msgid_t *out_id,
-	struct msg *out_msg);
 
 extern kern_status_t sys_msg_reply(
 	kern_handle_t channel,
-	msg_flags_t flags,
 	msgid_t id,
-	const struct msg *reply);
+	const kern_msg_t *msg);
-
 extern kern_status_t sys_msg_read(
-	kern_handle_t channel,
+	kern_handle_t channel_handle,
 	msgid_t id,
 	size_t offset,
-	struct iovec *out,
+	const kern_iovec_t *iov,
-	size_t nr_out);
+	size_t iov_count,
-extern kern_status_t sys_msg_read_handles(
+	size_t *nr_read);
-	kern_handle_t channel,
-	msgid_t id,
-	size_t offset,
-	struct handle_list *out,
-	size_t nr_out);
-
 extern kern_status_t sys_msg_write(
 	kern_handle_t channel,
 	msgid_t id,
 	size_t offset,
-	const struct iovec *in,
+	const kern_iovec_t *in,
-	size_t nr_in);
+	size_t nr_in,
-extern kern_status_t sys_msg_write_handles(
+	size_t *nr_written);
-	kern_handle_t channel,
-	msgid_t id,
-	size_t offset,
-	const struct handle_list *in,
-	size_t nr_in);
 
 extern virt_addr_t syscall_get_function(unsigned int sysid);
 

include/kernel/vm-region.h

@@ -11,6 +11,11 @@
 struct vm_region;
 struct vm_object;
 
+enum vm_region_status {
+	VM_REGION_DEAD = 0,
+	VM_REGION_ONLINE,
+};
+
 enum vm_region_entry_type {
 	VM_REGION_ENTRY_NONE = 0,
 	VM_REGION_ENTRY_REGION,
@@ -18,9 +23,16 @@ enum vm_region_entry_type {
 };
 
 struct vm_region_entry {
+	union {
 		struct btree_node e_node;
+		/* this entry is only used to queue vm-region objects for
+		 * recursive cleanup */
+		struct queue_entry e_entry;
+	};
 	struct vm_region_entry *e_parent;
 	enum vm_region_entry_type e_type;
+	/* absolute address of this entry */
+	virt_addr_t e_address;
 	/* offset in bytes of this entry within its immediate parent. */
 	off_t e_offset;
 	/* size of the entry in bytes */
@@ -41,6 +53,7 @@ struct vm_region_mapping {
 
 struct vm_region {
 	struct object vr_base;
+	enum vm_region_status vr_status;
 	struct vm_region_entry vr_entry;
 
 	char vr_name[VM_REGION_NAME_MAX];
@@ -81,6 +94,19 @@ extern kern_status_t vm_region_create(
 	vm_prot_t prot,
 	struct vm_region **out);
 
+/* recursively kills a given region and all of its sub-regions.
+ * when a region is killed, all of its mappings are unmapped, and any further
+ * operations on the region are denied. however, all handles and references to
+ * the region (any any sub-region) remain valid, and no kernel memory is
+ * de-allocated.
+ * the memory used by the vm-region object itself is de-allocated when the last
+ * handle/reference to the object is released.
+ * this function should be called with `region` locked.
+ */
+extern kern_status_t vm_region_kill(
+	struct vm_region *region,
+	unsigned long *lock_flags);
+
 /* map a vm-object into a vm-region.
  * [region_offset,length] must fall within exactly one region, and cannot span
  * multiple sibling regions.
@@ -122,6 +148,24 @@ extern virt_addr_t vm_region_get_base_address(const struct vm_region *region);
 
 extern void vm_region_dump(struct vm_region *region);
 
+/* read data from the user-space area of a vm-region into a kernel-mode buffer
+ */
+extern kern_status_t vm_region_read_kernel(
+	struct vm_region *src_region,
+	virt_addr_t src_ptr,
+	size_t count,
+	void *dest,
+	size_t *nr_read);
+
+/* write data to the user-space area of a vm-region from a kernel-mode buffer
+ */
+extern kern_status_t vm_region_write_kernel(
+	struct vm_region *dst_region,
+	virt_addr_t dst_ptr,
+	size_t count,
+	const void *src,
+	size_t *nr_written);
+
 extern kern_status_t vm_region_memmove(
 	struct vm_region *dest_region,
 	virt_addr_t dest_ptr,
@@ -133,13 +177,14 @@ extern kern_status_t vm_region_memmove(
 extern kern_status_t vm_region_memmove_v(
 	struct vm_region *dest_region,
 	size_t dest_offset,
-	struct iovec *dest,
+	const kern_iovec_t *dest,
 	size_t nr_dest,
 	struct vm_region *src_region,
 	size_t src_offset,
-	const struct iovec *src,
+	const kern_iovec_t *src,
 	size_t nr_src,
-	size_t bytes_to_move);
+	size_t bytes_to_move,
+	size_t *nr_bytes_moved);
 
 DEFINE_OBJECT_LOCK_FUNCTION(vm_region, vr_base)
 

kernel/arg.c

@@ -1,6 +1,7 @@
 #include <kernel/arg.h>
-#include <kernel/libc/string.h>
 #include <kernel/libc/ctype.h>
+#include <kernel/libc/string.h>
+#include <mango/status.h>
 
 static char g_cmdline[CMDLINE_MAX + 1] = {0};
 
@@ -81,7 +82,6 @@ static char *advance_to_next_arg(char *s, char *max)
 	return s;
 }
 
-
 const char *arg_value(const char *arg_name)
 {
 	char *s = g_cmdline;

kernel/bsp.c

@@ -69,7 +69,7 @@ kern_status_t bsp_load(struct bsp *bsp, const struct boot_module *mod)
 	return KERN_OK;
 }
 
-static kern_status_t map_executable(
+static kern_status_t map_executable_dyn(
 	struct bsp *bsp,
 	struct task *task,
 	virt_addr_t *entry)
@@ -164,6 +164,82 @@ static kern_status_t map_executable(
 	return KERN_OK;
 }
 
+static kern_status_t map_executable_exec(
+	struct bsp *bsp,
+	struct task *task,
+	virt_addr_t *entry)
+{
+	kern_status_t status = KERN_OK;
+	struct vm_object *data = vm_object_create(
+		".data",
+		5,
+		bsp->bsp_trailer.bsp_data_size,
+		VM_PROT_READ | VM_PROT_WRITE | VM_PROT_USER);
+	if (!data) {
+		return KERN_NO_MEMORY;
+	}
+
+	virt_addr_t text_base = 0, data_base = 0;
+
+	off_t text_foffset = bsp->bsp_trailer.bsp_exec_offset
+	                   + bsp->bsp_trailer.bsp_text_faddr;
+	off_t data_foffset = 0;
+	off_t text_voffset = bsp->bsp_trailer.bsp_text_vaddr;
+	off_t data_voffset = bsp->bsp_trailer.bsp_data_vaddr;
+
+	text_voffset -= vm_region_get_base_address(task->t_address_space);
+	data_voffset -= vm_region_get_base_address(task->t_address_space);
+
+#if 0
+	size_t tmp = 0;
+	status = vm_object_copy(
+		data,
+		0,
+		bsp->bsp_vmo,
+		bsp->bsp_trailer.bsp_data_faddr,
+		bsp->bsp_trailer.bsp_data_size,
+		&tmp);
+
+	tracek("read %zuB of data from executable", tmp);
+#endif
+
+	tracek("text_foffset=%06llx, data_foffset=%06llx",
+	       text_foffset,
+	       data_foffset);
+	tracek("text_voffset=%08llx, data_voffset=%08llx",
+	       text_voffset,
+	       data_voffset);
+
+	status = vm_region_map_object(
+		task->t_address_space,
+		text_voffset,
+		bsp->bsp_vmo,
+		text_foffset,
+		bsp->bsp_trailer.bsp_text_size,
+		VM_PROT_READ | VM_PROT_EXEC | VM_PROT_USER,
+		&text_base);
+	if (status != KERN_OK) {
+		return status;
+	}
+
+	status = vm_region_map_object(
+		task->t_address_space,
+		data_voffset,
+		data,
+		data_foffset,
+		bsp->bsp_trailer.bsp_data_size,
+		VM_PROT_READ | VM_PROT_WRITE | VM_PROT_USER,
+		&data_base);
+	if (status != KERN_OK) {
+		return status;
+	}
+
+	tracek("text_base=%08llx, data_base=%08llx", text_base, data_base);
+
+	*entry = bsp->bsp_trailer.bsp_exec_entry;
+	return KERN_OK;
+}
+
 kern_status_t bsp_launch_async(struct bsp *bsp, struct task *task)
 {
 	virt_addr_t stack_buffer, bsp_data_base;
@@ -205,7 +281,7 @@ kern_status_t bsp_launch_async(struct bsp *bsp, struct task *task)
 		return status;
 	}
 
-	status = map_executable(bsp, task, &entry);
+	status = map_executable_exec(bsp, task, &entry);
 	if (status != KERN_OK) {
 		return status;
 	}

kernel/channel.c

@@ -1,5 +1,6 @@
 #include <kernel/channel.h>
 #include <kernel/msg.h>
+#include <kernel/port.h>
 #include <kernel/util.h>
 #include <kernel/vm-region.h>
 
@@ -11,13 +12,18 @@ static struct object_type channel_type = {
 	.ob_header_offset = offsetof(struct channel, c_base),
 };
 
-BTREE_DEFINE_SIMPLE_GET(struct kmsg, msgid_t, msg_node, msg_id, get_msg_with_id)
+BTREE_DEFINE_SIMPLE_GET(struct msg, msgid_t, msg_node, msg_id, get_msg_with_id)
 
 kern_status_t channel_type_init(void)
 {
 	return object_type_register(&channel_type);
 }
 
+struct channel *channel_cast(struct object *obj)
+{
+	return CHANNEL_CAST(obj);
+}
+
 extern struct channel *channel_create(void)
 {
 	struct object *channel_object = object_create(&channel_type);
@@ -30,7 +36,7 @@ extern struct channel *channel_create(void)
 	return channel;
 }
 
-static bool try_enqueue(struct btree *tree, struct kmsg *msg)
+static bool try_enqueue(struct btree *tree, struct msg *msg)
 {
 	if (!tree->b_root) {
 		tree->b_root = &msg->msg_node;
@@ -40,8 +46,8 @@ static bool try_enqueue(struct btree *tree, struct kmsg *msg)
 
 	struct btree_node *cur = tree->b_root;
 	while (1) {
-		struct kmsg *cur_node
+		struct msg *cur_node
-			= BTREE_CONTAINER(struct kmsg, msg_node, cur);
+			= BTREE_CONTAINER(struct msg, msg_node, cur);
 		struct btree_node *next = NULL;
 
 		if (msg->msg_id > cur_node->msg_id) {
@@ -69,26 +75,33 @@ static bool try_enqueue(struct btree *tree, struct kmsg *msg)
 	return true;
 }
 
-static void kmsg_reply_error(struct kmsg *msg, kern_status_t status)
+static void kmsg_reply_error(
+	struct msg *msg,
+	kern_status_t status,
+	unsigned long *lock_flags)
 {
 	msg->msg_status = KMSG_REPLY_SENT;
-	msg->msg_status = status;
+	msg->msg_sender_port->p_status = PORT_READY;
+	msg->msg_result = status;
 	thread_awaken(msg->msg_sender_thread);
+	spin_unlock_irqrestore(&msg->msg_lock, *lock_flags);
 }
 
-static struct kmsg *get_next_msg(struct channel *channel)
+static struct msg *get_next_msg(
+	struct channel *channel,
+	unsigned long *lock_flags)
 {
-	unsigned long flags;
 	struct btree_node *cur = btree_first(&channel->c_msg);
 	while (cur) {
-		struct kmsg *msg = BTREE_CONTAINER(struct kmsg, msg_node, cur);
+		struct msg *msg = BTREE_CONTAINER(struct msg, msg_node, cur);
-		spin_lock_irqsave(&msg->msg_lock, &flags);
+		spin_lock_irqsave(&msg->msg_lock, lock_flags);
 		if (msg->msg_status == KMSG_WAIT_RECEIVE) {
 			msg->msg_status = KMSG_WAIT_REPLY;
+			msg->msg_sender_port->p_status = PORT_REPLY_BLOCKED;
 			return msg;
 		}
 
-		spin_unlock_irqrestore(&msg->msg_lock, flags);
+		spin_unlock_irqrestore(&msg->msg_lock, *lock_flags);
 		cur = btree_next(cur);
 	}
 
@@ -97,7 +110,7 @@ static struct kmsg *get_next_msg(struct channel *channel)
 
 extern kern_status_t channel_enqueue_msg(
 	struct channel *channel,
-	struct kmsg *msg)
+	struct msg *msg)
 {
 	fill_random(&msg->msg_id, sizeof msg->msg_id);
 	while (!try_enqueue(&channel->c_msg, msg)) {
@@ -111,18 +124,18 @@ extern kern_status_t channel_enqueue_msg(
 
 extern kern_status_t channel_recv_msg(
 	struct channel *channel,
-	struct msg *out_msg,
+	kern_msg_t *out_msg,
-	msgid_t *out_id,
 	unsigned long *irq_flags)
 {
 	struct wait_item waiter;
 	struct thread *self = current_thread();
-	struct kmsg *msg = NULL;
+	struct msg *msg = NULL;
+	unsigned long msg_lock_flags;
 
 	wait_item_init(&waiter, self);
 	for (;;) {
 		thread_wait_begin(&waiter, &channel->c_wq);
-		msg = get_next_msg(channel);
+		msg = get_next_msg(channel, &msg_lock_flags);
 		if (msg) {
 			break;
 		}
@@ -138,103 +151,229 @@ extern kern_status_t channel_recv_msg(
 	struct task *sender = msg->msg_sender_thread->tr_parent;
 	struct task *receiver = self->tr_parent;
 
+	struct vm_region *src = sender->t_address_space,
+			 *dst = receiver->t_address_space;
+
+	unsigned long f;
+	vm_region_lock_pair_irqsave(src, dst, &f);
+
 	kern_status_t status = vm_region_memmove_v(
-		receiver->t_address_space,
+		dst,
 		0,
 		out_msg->msg_data,
 		out_msg->msg_data_count,
-		sender->t_address_space,
+		src,
 		0,
-		msg->msg_req->msg_data,
+		msg->msg_req.msg_data,
-		msg->msg_req->msg_data_count,
+		msg->msg_req.msg_data_count,
-		VM_REGION_COPY_ALL);
+		VM_REGION_COPY_ALL,
+		NULL);
+
 	if (status != KERN_OK) {
-		kmsg_reply_error(msg, status);
+		kmsg_reply_error(msg, status, &msg_lock_flags);
 		return status;
 	}
 
-	status = handle_list_transfer(
+	struct handle_table *src_table = sender->t_handles,
-		receiver->t_handles,
+			    *dst_table = receiver->t_handles;
+
+	spin_lock_pair_irqsave(
+		&sender->t_handles_lock,
+		&receiver->t_handles_lock,
+		&f);
+	status = handle_table_transfer(
+		dst,
+		dst_table,
 		out_msg->msg_handles,
 		out_msg->msg_handles_count,
-		sender->t_handles,
+		src,
-		msg->msg_req->msg_handles,
+		src_table,
-		msg->msg_req->msg_handles_count);
+		msg->msg_req.msg_handles,
+		msg->msg_req.msg_handles_count);
+	spin_unlock_pair_irqrestore(
+		&sender->t_handles_lock,
+		&receiver->t_handles_lock,
+		f);
+	vm_region_unlock_pair_irqrestore(src, dst, f);
+
 	if (status != KERN_OK) {
-		kmsg_reply_error(msg, status);
+		kmsg_reply_error(msg, status, &msg_lock_flags);
 		return status;
 	}
 
-	kmsg_reply_error(msg, KERN_OK);
+	out_msg->msg_id = msg->msg_id;
+	out_msg->msg_sender = msg->msg_sender_thread->tr_parent->t_id;
+	out_msg->msg_endpoint = msg->msg_sender_port->p_base.ob_id;
+
+	spin_unlock_irqrestore(&msg->msg_lock, msg_lock_flags);
+
 	return KERN_OK;
 }
 
 extern kern_status_t channel_reply_msg(
 	struct channel *channel,
 	msgid_t id,
-	const struct msg *resp,
+	const kern_msg_t *reply,
 	unsigned long *irq_flags)
 {
-	struct kmsg *msg = get_msg_with_id(&channel->c_msg, id);
+	unsigned long msg_lock_flags;
-	if (!msg || msg->msg_status != KMSG_WAIT_REPLY) {
+	struct msg *msg = get_msg_with_id(&channel->c_msg, id);
+	if (!msg) {
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	spin_lock_irqsave(&msg->msg_lock, &msg_lock_flags);
+	if (msg->msg_status != KMSG_WAIT_REPLY) {
+		spin_unlock_irqrestore(&msg->msg_lock, msg_lock_flags);
 		return KERN_INVALID_ARGUMENT;
 	}
 
 	struct thread *self = current_thread();
-	struct task *sender = msg->msg_sender_thread->tr_parent;
+	/* the task that is about to receive the response */
-	struct task *receiver = self->tr_parent;
+	struct task *receiver = msg->msg_sender_thread->tr_parent;
+	/* the task that is about to send the response */
+	struct task *sender = self->tr_parent;
+
+	struct vm_region *src = sender->t_address_space,
+			 *dst = receiver->t_address_space;
+	unsigned long f;
+	vm_region_lock_pair_irqsave(src, dst, &f);
 
 	kern_status_t status = vm_region_memmove_v(
-		receiver->t_address_space,
+		dst,
 		0,
-		msg->msg_resp->msg_data,
+		msg->msg_resp.msg_data,
-		msg->msg_resp->msg_data_count,
+		msg->msg_resp.msg_data_count,
-		sender->t_address_space,
+		src,
 		0,
-		resp->msg_data,
+		reply->msg_data,
-		resp->msg_data_count,
+		reply->msg_data_count,
-		VM_REGION_COPY_ALL);
+		VM_REGION_COPY_ALL,
+		NULL);
+
 	if (status != KERN_OK) {
-		kmsg_reply_error(msg, status);
+		kmsg_reply_error(msg, status, &msg_lock_flags);
 		return status;
 	}
 
-	status = handle_list_transfer(
+	struct handle_table *src_table = sender->t_handles,
-		receiver->t_handles,
+			    *dst_table = receiver->t_handles;
-		msg->msg_resp->msg_handles,
+
-		msg->msg_resp->msg_handles_count,
+	spin_lock_pair_irqsave(
-		sender->t_handles,
+		&sender->t_handles_lock,
-		resp->msg_handles,
+		&receiver->t_handles_lock,
-		resp->msg_handles_count);
+		&f);
+	status = handle_table_transfer(
+		dst,
+		dst_table,
+		msg->msg_resp.msg_handles,
+		msg->msg_resp.msg_handles_count,
+		src,
+		src_table,
+		reply->msg_handles,
+		reply->msg_handles_count);
+	spin_unlock_pair_irqrestore(
+		&sender->t_handles_lock,
+		&receiver->t_handles_lock,
+		f);
+	vm_region_unlock_pair_irqrestore(src, dst, f);
+
 	if (status != KERN_OK) {
-		kmsg_reply_error(msg, status);
+		kmsg_reply_error(msg, status, &msg_lock_flags);
 		return status;
 	}
 
-	msg->msg_status = KERN_OK;
+	kmsg_reply_error(msg, KERN_OK, &msg_lock_flags);
-	msg->msg_status = KMSG_REPLY_SENT;
 
-	return KERN_UNIMPLEMENTED;
+	return KERN_OK;
 }
 
 extern kern_status_t channel_read_msg(
 	struct channel *channel,
-	msgid_t msg,
+	msgid_t id,
 	size_t offset,
-	void *buf,
+	struct vm_region *dest_region,
-	size_t len,
+	const kern_iovec_t *dest_iov,
+	size_t dest_iov_count,
 	size_t *nr_read)
 {
-	return KERN_UNIMPLEMENTED;
+	unsigned long msg_lock_flags;
+	struct msg *msg = get_msg_with_id(&channel->c_msg, id);
+	if (!msg) {
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	spin_lock_irqsave(&msg->msg_lock, &msg_lock_flags);
+	if (msg->msg_status != KMSG_WAIT_REPLY) {
+		spin_unlock_irqrestore(&msg->msg_lock, msg_lock_flags);
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	struct vm_region *src_region
+		= msg->msg_sender_thread->tr_parent->t_address_space;
+
+	unsigned long f;
+	vm_region_lock_pair_irqsave(src_region, dest_region, &f);
+
+	kern_status_t status = vm_region_memmove_v(
+		dest_region,
+		0,
+		dest_iov,
+		dest_iov_count,
+		src_region,
+		offset,
+		msg->msg_req.msg_data,
+		msg->msg_req.msg_data_count,
+		VM_REGION_COPY_ALL,
+		nr_read);
+	vm_region_unlock_pair_irqrestore(src_region, dest_region, f);
+
+	spin_unlock_irqrestore(&msg->msg_lock, msg_lock_flags);
+
+	return status;
 }
 
 extern kern_status_t channel_write_msg(
 	struct channel *channel,
-	msgid_t msg,
+	msgid_t id,
 	size_t offset,
-	const void *buf,
+	struct vm_region *src_region,
-	size_t len,
+	const kern_iovec_t *src_iov,
+	size_t src_iov_count,
 	size_t *nr_written)
 {
-	return KERN_UNIMPLEMENTED;
+	unsigned long msg_lock_flags;
+	struct msg *msg = get_msg_with_id(&channel->c_msg, id);
+	if (!msg) {
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	spin_lock_irqsave(&msg->msg_lock, &msg_lock_flags);
+	if (msg->msg_status != KMSG_WAIT_REPLY) {
+		spin_unlock_irqrestore(&msg->msg_lock, msg_lock_flags);
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	struct vm_region *dest_region
+		= msg->msg_sender_thread->tr_parent->t_address_space;
+
+	unsigned long f;
+	vm_region_lock_pair_irqsave(src_region, dest_region, &f);
+
+	kern_status_t status = vm_region_memmove_v(
+		dest_region,
+		offset,
+		msg->msg_resp.msg_data,
+		msg->msg_resp.msg_data_count,
+		src_region,
+		0,
+		src_iov,
+		src_iov_count,
+		VM_REGION_COPY_ALL,
+		nr_written);
+	vm_region_unlock_pair_irqrestore(src_region, dest_region, f);
+
+	spin_unlock_irqrestore(&msg->msg_lock, msg_lock_flags);
+
+	return status;
 }

kernel/handle.c

@@ -3,7 +3,9 @@
 #include <kernel/object.h>
 #include <kernel/sched.h>
 #include <kernel/util.h>
+#include <kernel/vm-region.h>
 #include <kernel/vm.h>
+#include <mango/types.h>
 
 /* depth=3 gives a maximum of ~66.6 million handles */
 #define MAX_TABLE_DEPTH  3
@@ -192,122 +194,143 @@ struct handle *handle_table_get_handle(
 	return &tab->t_handles.t_handle_list[handle_index];
 }
 
-struct handle_list_iterator {
+kern_status_t handle_table_transfer(
-	struct handle_list *it_list;
+	struct vm_region *dst_region,
-	size_t it_list_count;
+	struct handle_table *dst,
-	size_t it_list_ptr;
+	kern_msg_handle_t *dst_handles,
-
+	size_t dst_handles_max,
-	kern_handle_t *it_handles;
+	struct vm_region *src_region,
-	size_t it_nr_handles;
+	struct handle_table *src,
-};
+	kern_msg_handle_t *src_handles,
-
+	size_t src_handles_count)
-static void handle_list_iterator_begin(
-	struct handle_list_iterator *it,
-	struct handle_list *list,
-	size_t list_count)
 {
-	memset(it, 0x0, sizeof *it);
+	kern_status_t status = KERN_OK;
-	it->it_list = list;
+	size_t to_transfer = MIN(dst_handles_max, src_handles_count);
-	it->it_list_count = list_count;
 
-	while (it->it_list_ptr < list_count) {
+	size_t i = 0;
-		if (list[it->it_list_ptr].l_nr_handles > 0) {
+	for (size_t i = 0; i < to_transfer; i++) {
-			break;
+		kern_msg_handle_t src_handle = {0}, dst_handle = {0};
-		}
+		virt_addr_t src_handle_addr
+			= (virt_addr_t)src_handles + (i * sizeof src_handle);
+		virt_addr_t dst_handle_addr
+			= (virt_addr_t)dst_handles + (i * sizeof dst_handle);
+		status = vm_region_read_kernel(
+			src_region,
+			src_handle_addr,
+			sizeof src_handle,
+			&src_handle,
+			NULL);
 
-		it->it_list_ptr++;
-	}
-
-	if (it->it_list_ptr >= list_count) {
-		return;
-	}
-
-	it->it_handles = list[it->it_list_ptr].l_handles;
-	it->it_nr_handles = list[it->it_list_ptr].l_nr_handles;
-}
-
-static void handle_list_iterator_seek(
-	struct handle_list_iterator *it,
-	size_t nr_handles)
-{
-	if (nr_handles > it->it_nr_handles) {
-		nr_handles = it->it_nr_handles;
-	}
-
-	if (nr_handles < it->it_nr_handles) {
-		it->it_handles += nr_handles;
-		it->it_nr_handles -= nr_handles;
-		return;
-	}
-
-	it->it_list_ptr++;
-	while (it->it_list_ptr < it->it_list_count) {
-		if (it->it_list[it->it_list_ptr].l_nr_handles > 0) {
-			break;
-		}
-
-		it->it_list_ptr++;
-	}
-
-	if (it->it_list_ptr >= it->it_list_count) {
-		return;
-	}
-
-	it->it_handles = it->it_list[it->it_list_ptr].l_handles;
-	it->it_nr_handles = it->it_list[it->it_list_ptr].l_nr_handles;
-}
-
-kern_status_t handle_list_transfer(
-	struct handle_table *dest_table,
-	struct handle_list *dest_list,
-	size_t dest_list_count,
-	struct handle_table *src_table,
-	const struct handle_list *src_list,
-	size_t src_list_count)
-{
-	struct handle_list_iterator src, dest;
-	handle_list_iterator_begin(
-		&src,
-		(struct handle_list *)src_list,
-		src_list_count);
-	handle_list_iterator_begin(&dest, dest_list, dest_list_count);
-
-	while (src.it_nr_handles && dest.it_nr_handles) {
-		size_t to_copy = MIN(src.it_nr_handles, dest.it_nr_handles);
-		for (size_t i = 0; i < to_copy; i++) {
-			kern_handle_t handle_v = src.it_handles[i];
-			struct handle *handle
-				= handle_table_get_handle(src_table, handle_v);
-			if (!handle) {
-				return KERN_HANDLE_INVALID;
-			}
-
-			struct object *obj = object_ref(handle->h_object);
-			handle_flags_t flags = handle->h_flags;
-
-			handle_table_free_handle(src_table, handle_v);
-
-			struct handle *dest_slot = NULL;
-			kern_status_t status = handle_table_alloc_handle(
-				dest_table,
-				&dest_slot,
-				&handle_v);
 		if (status != KERN_OK) {
+			src_handle.hnd_result = KERN_OK;
+			vm_region_write_kernel(
+				src_region,
+				src_handle_addr,
+				sizeof src_handle,
+				&src_handle,
+				NULL);
+			break;
+		}
+
+		struct handle *src_entry
+			= handle_table_get_handle(src, src_handle.hnd_value);
+		struct handle *dst_entry = NULL;
+		kern_handle_t dst_value = KERN_HANDLE_INVALID;
+
+		if (!src_entry) {
+			status = KERN_INVALID_ARGUMENT;
+			src_handle.hnd_result = KERN_OK;
+			vm_region_write_kernel(
+				src_region,
+				src_handle_addr,
+				sizeof src_handle,
+				&src_handle,
+				NULL);
+			break;
+		}
+
+		switch (src_handle.hnd_mode) {
+		case KERN_MSG_HANDLE_IGNORE:
+			break;
+		case KERN_MSG_HANDLE_MOVE:
+			status = handle_table_alloc_handle(
+				dst,
+				&dst_entry,
+				&dst_value);
+			if (status != KERN_OK) {
+				break;
+			}
+
+			dst_entry->h_object = src_entry->h_object;
+			dst_entry->h_flags = src_entry->h_flags;
+			object_add_handle(dst_entry->h_object);
+
+			handle_table_free_handle(src, src_handles[i].hnd_value);
+
+			dst_handle.hnd_mode = src_handles[i].hnd_mode;
+			dst_handle.hnd_value = dst_value;
+			dst_handle.hnd_result = KERN_OK;
+			break;
+		case KERN_MSG_HANDLE_COPY:
+			status = handle_table_alloc_handle(
+				dst,
+				&dst_entry,
+				&dst_value);
+			if (status != KERN_OK) {
+				break;
+			}
+
+			dst_entry->h_object = src_entry->h_object;
+			dst_entry->h_flags = src_entry->h_flags;
+			object_add_handle(dst_entry->h_object);
+
+			dst_handle.hnd_mode = src_handles[i].hnd_mode;
+			dst_handle.hnd_value = dst_value;
+			dst_handle.hnd_result = KERN_OK;
+			break;
+		default:
+			status = KERN_INVALID_ARGUMENT;
+			break;
+		}
+
+		src_handle.hnd_result = status;
+
+		vm_region_write_kernel(
+			src_region,
+			src_handle_addr,
+			sizeof src_handle,
+			&src_handle,
+			NULL);
+		vm_region_write_kernel(
+			dst_region,
+			dst_handle_addr,
+			sizeof dst_handle,
+			&dst_handle,
+			NULL);
+	}
+
+	for (; i < src_handles_count; i++) {
+		kern_msg_handle_t handle = {0};
+		virt_addr_t handle_addr
+			= (virt_addr_t)src_handles + (i * sizeof handle);
+		vm_region_read_kernel(
+			src_region,
+			handle_addr,
+			sizeof handle,
+			&handle,
+			NULL);
+
+		if (handle.hnd_mode != KERN_MSG_HANDLE_MOVE) {
+			continue;
+		}
+
+		struct handle *src_entry
+			= handle_table_get_handle(src, handle.hnd_value);
+		if (src_entry) {
+			object_remove_handle(src_entry->h_object);
+			handle_table_free_handle(src, handle.hnd_value);
+		}
+	}
+
 	return status;
-			}
-
-			dest_slot->h_object = obj;
-			dest_slot->h_flags = flags;
-
-			object_add_handle(obj);
-			object_unref(obj);
-
-			dest.it_handles[i] = handle_v;
-		}
-
-		handle_list_iterator_seek(&src, to_copy);
-		handle_list_iterator_seek(&dest, to_copy);
-	}
-
-	return KERN_OK;
 }

kernel/iovec.c

@@ -1,10 +1,66 @@
 #include <kernel/iovec.h>
 #include <kernel/libc/string.h>
 #include <kernel/util.h>
+#include <kernel/vm-region.h>
+
+static bool read_iovec(
+	struct iovec_iterator *it,
+	size_t index,
+	kern_iovec_t *out)
+{
+	if (index >= it->it_nr_vecs) {
+		return false;
+	}
+
+	if (!it->it_region) {
+		memcpy(out, &it->it_vecs[index], sizeof *out);
+		return true;
+	}
+
+	size_t nr_read = 0;
+	kern_status_t status = vm_region_read_kernel(
+		it->it_region,
+		(virt_addr_t)it->it_vecs + (index * sizeof(kern_iovec_t)),
+		sizeof(kern_iovec_t),
+		out,
+		&nr_read);
+
+	return (status == KERN_OK && nr_read != sizeof(kern_iovec_t));
+}
+
+void iovec_iterator_begin_user(
+	struct iovec_iterator *it,
+	struct vm_region *region,
+	const kern_iovec_t *vecs,
+	size_t nr_vecs)
+{
+	memset(it, 0x0, sizeof *it);
+	it->it_region = region;
+	it->it_vecs = vecs;
+	it->it_nr_vecs = nr_vecs;
+
+	kern_iovec_t iov;
+
+	while (it->it_vec_ptr < nr_vecs) {
+		read_iovec(it, it->it_vec_ptr, &iov);
+		if (iov.io_len > 0) {
+			break;
+		}
+
+		it->it_vec_ptr++;
+	}
+
+	if (it->it_vec_ptr >= nr_vecs) {
+		return;
+	}
+
+	it->it_base = iov.io_base;
+	it->it_len = iov.io_len;
+}
 
 void iovec_iterator_begin(
 	struct iovec_iterator *it,
-	const struct iovec *vecs,
+	const kern_iovec_t *vecs,
 	size_t nr_vecs)
 {
 	memset(it, 0x0, sizeof *it);
@@ -20,6 +76,8 @@ void iovec_iterator_begin(
 	}
 
 	if (it->it_vec_ptr >= nr_vecs) {
+		it->it_len = 0;
+		it->it_base = 0;
 		return;
 	}
 
@@ -39,10 +97,12 @@ void iovec_iterator_seek(struct iovec_iterator *it, size_t nr_bytes)
 		}
 
 		nr_bytes -= to_seek;
+		kern_iovec_t iov;
 
 		it->it_vec_ptr++;
 		while (it->it_vec_ptr < it->it_nr_vecs) {
-			if (it->it_vecs[it->it_vec_ptr].io_len > 0) {
+			read_iovec(it, it->it_vec_ptr, &iov);
+			if (iov.io_len > 0) {
 				break;
 			}
 
@@ -50,10 +110,12 @@ void iovec_iterator_seek(struct iovec_iterator *it, size_t nr_bytes)
 		}
 
 		if (it->it_vec_ptr >= it->it_nr_vecs) {
+			it->it_len = 0;
+			it->it_base = 0;
 			return;
 		}
 
-		it->it_base = it->it_vecs[it->it_vec_ptr].io_base;
+		it->it_base = iov.io_base;
-		it->it_len = it->it_vecs[it->it_vec_ptr].io_len;
+		it->it_len = iov.io_len;
 	}
 }

kernel/object.c

@@ -7,6 +7,20 @@
 static struct queue object_types;
 static spin_lock_t object_types_lock = SPIN_LOCK_INIT;
 
+static koid_t koid_alloc(void)
+{
+	static koid_t counter = 0;
+	static spin_lock_t lock = SPIN_LOCK_INIT;
+
+	unsigned long flags;
+	spin_lock_irqsave(&lock, &flags);
+	koid_t result = counter;
+	counter++;
+	spin_unlock_irqrestore(&lock, flags);
+
+	return result;
+}
+
 kern_status_t object_bootstrap(void)
 {
 	return KERN_OK;
@@ -50,11 +64,10 @@ struct object *object_create(struct object_type *type)
 		return NULL;
 	}
 
-	memset(obj_buf, 0x00, type->ob_size);
-
 	struct object *obj = (struct object *)((unsigned char *)obj_buf
 	                                       + type->ob_header_offset);
 
+	obj->ob_id = koid_alloc();
 	obj->ob_type = type;
 	obj->ob_lock = SPIN_LOCK_INIT;
 	obj->ob_magic = OBJECT_MAGIC;
@@ -70,6 +83,15 @@ struct object *object_ref(struct object *obj)
 	return obj;
 }
 
+static void __cleanup(struct object *obj, struct queue *queue)
+{
+	if (HAS_OP(obj, destroy)) {
+		obj->ob_type->ob_ops.destroy(obj, queue);
+	}
+
+	vm_cache_free(&obj->ob_type->ob_cache, obj);
+}
+
 static void object_cleanup(struct object *obj, unsigned long flags)
 {
 	if (obj->ob_refcount > 0 || obj->ob_handles > 0) {
@@ -77,11 +99,30 @@ static void object_cleanup(struct object *obj, unsigned long flags)
 		return;
 	}
 
-	if (HAS_OP(obj, destroy)) {
+	struct queue queue = QUEUE_INIT;
-		obj->ob_type->ob_ops.destroy(obj);
+	__cleanup(obj, &queue);
+
+	if (!HAS_OP(obj, destroy_recurse)) {
+		return;
 	}
 
-	vm_cache_free(&obj->ob_type->ob_cache, obj);
+	while (!queue_empty(&queue)) {
+		struct queue_entry *entry = queue_pop_front(&queue);
+		struct object *child = NULL;
+		obj->ob_type->ob_ops.destroy_recurse(entry, &child);
+		if (!child) {
+			continue;
+		}
+
+		if (child->ob_refcount > 1) {
+			child->ob_refcount--;
+			continue;
+		}
+
+		if (child->ob_refcount == 0 && child->ob_handles == 0) {
+			__cleanup(child, &queue);
+		}
+	}
 }
 
 void object_unref(struct object *obj)
@@ -137,6 +178,38 @@ void object_unlock_irqrestore(struct object *obj, unsigned long flags)
 	spin_unlock_irqrestore(&obj->ob_lock, flags);
 }
 
+void object_lock_pair_irqsave(
+	struct object *a,
+	struct object *b,
+	unsigned long *flags)
+{
+	if (a == b) {
+		object_lock_irqsave(a, flags);
+	} else if (a < b) {
+		object_lock_irqsave(a, flags);
+		object_lock(b);
+	} else {
+		object_lock_irqsave(b, flags);
+		object_lock(a);
+	}
+}
+
+void object_unlock_pair_irqrestore(
+	struct object *a,
+	struct object *b,
+	unsigned long flags)
+{
+	if (a == b) {
+		object_unlock_irqrestore(a, flags);
+	} else if (a < b) {
+		object_unlock(b);
+		object_unlock_irqrestore(a, flags);
+	} else {
+		object_unlock(a);
+		object_unlock_irqrestore(b, flags);
+	}
+}
+
 void *object_data(struct object *obj)
 {
 	return (char *)obj + sizeof *obj;

kernel/port.c

@@ -20,8 +20,24 @@ struct port *port_cast(struct object *obj)
 	return PORT_CAST(obj);
 }
 
-static void wait_for_reply(struct port *port)
+static void wait_for_reply(struct msg *msg, unsigned long *lock_flags)
 {
+	struct wait_item waiter;
+	struct thread *self = current_thread();
+
+	wait_item_init(&waiter, self);
+	for (;;) {
+		self->tr_state = THREAD_SLEEPING;
+		if (msg->msg_status == KMSG_REPLY_SENT) {
+			break;
+		}
+
+		port_unlock_irqrestore(msg->msg_sender_port, *lock_flags);
+		schedule(SCHED_NORMAL);
+		port_lock_irqsave(msg->msg_sender_port, lock_flags);
+	}
+
+	self->tr_state = THREAD_READY;
 }
 
 struct port *port_create(void)
@@ -49,30 +65,47 @@ kern_status_t port_connect(struct port *port, struct channel *remote)
 	return KERN_OK;
 }
 
+kern_status_t port_disconnect(struct port *port)
+{
+	if (port->p_status != PORT_READY) {
+		return KERN_BAD_STATE;
+	}
+
+	port->p_remote = NULL;
+	port->p_status = PORT_OFFLINE;
+	return KERN_OK;
+}
+
 kern_status_t port_send_msg(
 	struct port *port,
-	const struct msg *req,
+	const kern_msg_t *in_msg,
-	struct msg *resp)
+	kern_msg_t *out_reply,
+	unsigned long *lock_flags)
 {
 	if (port->p_status != PORT_READY) {
 		return KERN_BAD_STATE;
 	}
 
 	struct thread *self = current_thread();
-	struct kmsg *msg = &self->tr_msg;
+	struct msg *msg = &self->tr_msg;
+	memset(msg, 0x0, sizeof *msg);
+	msg->msg_status = KMSG_WAIT_RECEIVE;
 	msg->msg_sender_thread = self;
 	msg->msg_sender_port = port;
-	msg->msg_req = req;
+	memcpy(&msg->msg_req, in_msg, sizeof msg->msg_req);
-	msg->msg_resp = resp;
+	memcpy(&msg->msg_resp, out_reply, sizeof msg->msg_req);
 
 	unsigned long flags;
 	channel_lock_irqsave(port->p_remote, &flags);
+	port->p_status = PORT_SEND_BLOCKED;
 	channel_enqueue_msg(port->p_remote, msg);
 	channel_unlock_irqrestore(port->p_remote, flags);
 
-	port->p_status = PORT_SEND_BLOCKED;
+	wait_for_reply(msg, lock_flags);
 
-	wait_for_reply(port);
+	channel_lock_irqsave(port->p_remote, &flags);
+	btree_delete(&port->p_remote->c_msg, &msg->msg_node);
+	channel_unlock_irqrestore(port->p_remote, flags);
 
 	return msg->msg_result;
 }

kernel/status.c

@@ -1,4 +1,5 @@
 #include <mango/status.h>
+#include <mango/types.h>
 
 #define ERROR_STRING_CASE(code)                                                \
 	case code:                                                             \

libmango/arch/x86_64/syscall.S

@@ -56,8 +56,10 @@
 .endm
 
 SYSCALL_GATE task_exit SYS_TASK_EXIT 1
+SYSCALL_GATE task_self SYS_TASK_SELF 0
 SYSCALL_GATE task_create SYS_TASK_CREATE 5
 SYSCALL_GATE task_create_thread SYS_TASK_CREATE_THREAD 6
+SYSCALL_GATE task_get_address_space SYS_TASK_GET_ADDRESS_SPACE 1
 
 SYSCALL_GATE thread_start SYS_THREAD_START 1
 
@@ -67,6 +69,7 @@ SYSCALL_GATE vm_object_write SYS_VM_OBJECT_WRITE 5
 SYSCALL_GATE vm_object_copy SYS_VM_OBJECT_COPY 6
 
 SYSCALL_GATE vm_region_create SYS_VM_REGION_CREATE 8
+SYSCALL_GATE vm_region_kill SYS_VM_REGION_KILL 1
 SYSCALL_GATE vm_region_read SYS_VM_REGION_READ 5
 SYSCALL_GATE vm_region_write SYS_VM_REGION_WRITE 5
 SYSCALL_GATE vm_region_map_absolute SYS_VM_REGION_MAP_ABSOLUTE 7
@@ -79,15 +82,13 @@ SYSCALL_GATE kern_handle_close SYS_KERN_HANDLE_CLOSE 1
 SYSCALL_GATE kern_config_get SYS_KERN_CONFIG_GET 3
 SYSCALL_GATE kern_config_set SYS_KERN_CONFIG_SET 3
 
-SYSCALL_GATE channel_create SYS_CHANNEL_CREATE 3
+SYSCALL_GATE channel_create SYS_CHANNEL_CREATE 2
 SYSCALL_GATE port_create SYS_PORT_CREATE 1
 SYSCALL_GATE port_connect SYS_PORT_CONNECT 3
 SYSCALL_GATE port_disconnect SYS_PORT_DISCONNECT 1
-SYSCALL_GATE msg_send SYS_MSG_SEND 4
+SYSCALL_GATE msg_send SYS_MSG_SEND 5
 SYSCALL_GATE msg_recv SYS_MSG_RECV 4
 SYSCALL_GATE msg_reply SYS_MSG_REPLY 4
-SYSCALL_GATE msg_read SYS_MSG_READ 5
+SYSCALL_GATE msg_read SYS_MSG_READ 6
-SYSCALL_GATE msg_read_handles SYS_MSG_READ_HANDLES 5
+SYSCALL_GATE msg_write SYS_MSG_WRITE 6
-SYSCALL_GATE msg_write SYS_MSG_WRITE 5
-SYSCALL_GATE msg_write_handles SYS_MSG_WRITE_HANDLES 5
 

libmango/include-user/mango/log.h

@@ -2,16 +2,26 @@
 #define MANGO_LOG_H_
 
 #include <mango/status.h>
+#include <mango/types.h>
+
+#undef TRACE
 
 extern kern_status_t kern_log(const char *s);
 
+#define kern_logf(...)                                                         \
+	do {                                                                   \
+		char __logbuf[128];                                            \
+		snprintf(__logbuf, sizeof __logbuf, __VA_ARGS__);              \
+		kern_log(__logbuf);                                            \
+	} while (0)
+
 #ifdef TRACE
 #define kern_trace(...) kern_log(__VA_ARGS__)
 #define kern_tracef(...)                                                       \
 	do {                                                                   \
-		char s[128];                                                   \
+		char __logbuf[128];                                            \
-		snprintf(s, sizeof s, __VA_ARGS__);                            \
+		snprintf(__logbuf, sizeof __logbuf, __VA_ARGS__);              \
-		kern_log(s);                                                   \
+		kern_log(__logbuf);                                            \
 	} while (0)
 #else
 #define kern_trace(...)

libmango/include-user/mango/msg.h

@@ -4,10 +4,7 @@
 #include <mango/status.h>
 #include <mango/types.h>
 
-extern kern_status_t channel_create(
+extern kern_status_t channel_create(unsigned int id, kern_handle_t *out);
-	unsigned int id,
-	channel_flags_t flags,
-	kern_handle_t *out);
 extern kern_status_t port_create(kern_handle_t *out);
 extern kern_status_t port_connect(
 	kern_handle_t port,
@@ -17,46 +14,30 @@ extern kern_status_t port_disconnect(kern_handle_t port);
 
 extern kern_status_t msg_send(
 	kern_handle_t port,
-	msg_flags_t flags,
+	const kern_msg_t *msg,
-	const struct msg *req,
+	kern_msg_t *out_response);
-	struct msg *resp);
 
-extern kern_status_t msg_recv(
+extern kern_status_t msg_recv(kern_handle_t channel, kern_msg_t *out);
-	kern_handle_t channel,
-	msg_flags_t flags,
-	msgid_t *out_id,
-	struct msg *out_msg);
 
 extern kern_status_t msg_reply(
 	kern_handle_t channel,
-	msg_flags_t flags,
 	msgid_t id,
-	const struct msg *reply);
+	const kern_msg_t *response);
 
 extern kern_status_t msg_read(
 	kern_handle_t channel,
 	msgid_t id,
 	size_t offset,
-	struct iovec *out,
+	kern_iovec_t *out,
-	size_t nr_out);
+	size_t out_count,
-extern kern_status_t msg_read_handles(
+	size_t *nr_read);
-	kern_handle_t channel,
-	msgid_t id,
-	size_t offset,
-	struct handle_list *out,
-	size_t nr_out);
 
 extern kern_status_t msg_write(
 	kern_handle_t channel,
 	msgid_t id,
 	size_t offset,
-	const struct iovec *in,
+	const kern_iovec_t *in,
-	size_t nr_in);
+	size_t nr_in,
-extern kern_status_t msg_write_handles(
+	size_t *nr_written);
-	kern_handle_t channel,
-	msgid_t id,
-	size_t offset,
-	const struct handle_list *in,
-	size_t nr_in);
 
 #endif

libmango/include-user/mango/task.h

@@ -5,6 +5,7 @@
 #include <mango/types.h>
 
 extern kern_status_t task_exit(int status);
+extern kern_status_t task_self(kern_handle_t *out);
 
 extern kern_status_t task_create(
 	kern_handle_t parent,
@@ -19,6 +20,9 @@ extern kern_status_t task_create_thread(
 	uintptr_t *args,
 	size_t nr_args,
 	kern_handle_t *out_thread);
+extern kern_status_t task_get_address_space(
+	kern_handle_t task,
+	kern_handle_t *out);
 
 extern kern_status_t thread_start(kern_handle_t thread);
 

libmango/include-user/mango/vm.h

@@ -39,6 +39,7 @@ extern kern_status_t vm_region_create(
 	vm_prot_t prot,
 	kern_handle_t *out,
 	virt_addr_t *out_base_address);
+extern kern_status_t vm_region_kill(kern_handle_t region);
 extern kern_status_t vm_region_read(
 	kern_handle_t region,
 	void *dst,

libmango/include/mango/status.h

@@ -1,8 +1,6 @@
 #ifndef MANGO_STATUS_H_
 #define MANGO_STATUS_H_
 
-typedef unsigned int kern_status_t;
-
 #define KERN_OK               (0)
 #define KERN_UNIMPLEMENTED    (1)
 #define KERN_NAME_EXISTS      (2)

libmango/include/mango/syscall.h

@@ -2,14 +2,17 @@
 #define MANGO_SYSCALL_H_
 
 #define SYS_TASK_EXIT                1
+#define SYS_TASK_SELF                31
 #define SYS_TASK_CREATE              2
 #define SYS_TASK_CREATE_THREAD       3
+#define SYS_TASK_GET_ADDRESS_SPACE   33
 #define SYS_THREAD_START             30
 #define SYS_VM_OBJECT_CREATE         4
 #define SYS_VM_OBJECT_READ           5
 #define SYS_VM_OBJECT_WRITE          6
 #define SYS_VM_OBJECT_COPY           29
 #define SYS_VM_REGION_CREATE         7
+#define SYS_VM_REGION_KILL           34
 #define SYS_VM_REGION_READ           8
 #define SYS_VM_REGION_WRITE          9
 #define SYS_VM_REGION_MAP_ABSOLUTE   10
@@ -24,9 +27,7 @@
 #define SYS_MSG_RECV                 19
 #define SYS_MSG_REPLY                20
 #define SYS_MSG_READ                 21
-#define SYS_MSG_READ_HANDLES         22
 #define SYS_MSG_WRITE                23
-#define SYS_MSG_WRITE_HANDLES        24
 #define SYS_CHANNEL_CREATE           25
 #define SYS_PORT_CREATE              26
 #define SYS_PORT_CONNECT             27

libmango/include/mango/types.h

@@ -12,51 +12,76 @@
 #define VM_PROT_NOCACHE        0x10u
 #define VM_PROT_MAP_SPECIFIC   0x40u
 
-/* if this flag is set, other tasks can connect to this channel using
- * the port_connect_* syscalls.
- * if this flag is NOT set, only threads in the task that owns the channel
- * can create ports connecting to it. */
-#define CHANNEL_F_ALLOW_DIRECT_CONNECTIONS 0x01u
-
-/* msg_reply: once the reply has been sent, disconnect the port that sent the
- * original message */
-#define MSG_F_DISCONNECT_AFTER_REPLY       0x01u
-
 #define VM_REGION_ANY_OFFSET   ((off_t) - 1)
 #define KERN_HANDLE_INVALID    ((kern_handle_t)0xFFFFFFFF)
 
 #define KERN_CFG_INVALID       0x00u
 #define KERN_CFG_PAGE_SIZE     0x01u
 
+#define KERN_MSG_MAX_HANDLES   64
+#define KERN_MSG_HANDLE_IGNORE 0
+#define KERN_MSG_HANDLE_MOVE   1
+#define KERN_MSG_HANDLE_COPY   2
+
+#define IOVEC(p, len)                                                          \
+	{                                                                      \
+		.io_base = (virt_addr_t)(p),                                   \
+		.io_len = (len),                                               \
+	}
+#define MSG_HANDLE(mode, value)                                                \
+	{                                                                      \
+		.hnd_mode = (mode),                                            \
+		.hnd_value = (value),                                          \
+	}
+#define MSG(data, data_count, handles, handles_len)                            \
+	{                                                                      \
+		.msg_data = (data),                                            \
+		.msg_data_count = (data_count),                                \
+		.msg_handles = (handles),                                      \
+		.msg_handles_count = (handles_len),                            \
+	}
+
 typedef uintptr_t phys_addr_t;
 typedef uintptr_t virt_addr_t;
 typedef uint64_t msgid_t;
 typedef uint64_t off_t;
+typedef uint64_t koid_t;
 typedef unsigned int tid_t;
+typedef unsigned int kern_status_t;
 typedef uint32_t kern_handle_t;
 typedef uint32_t kern_config_key_t;
 typedef uint32_t vm_prot_t;
-typedef uint32_t channel_flags_t;
+typedef int64_t ssize_t;
-typedef uint32_t msg_flags_t;
 
 typedef unsigned int umode_t;
 
-struct iovec {
+typedef struct {
 	virt_addr_t io_base;
 	size_t io_len;
-};
+} kern_iovec_t;
 
-struct handle_list {
+typedef struct {
-	kern_handle_t *l_handles;
+	unsigned int hnd_mode;
-	size_t l_nr_handles;
+	kern_handle_t hnd_value;
-};
+	kern_status_t hnd_result;
+} kern_msg_handle_t;
 
-struct msg {
+typedef struct {
-	struct iovec *msg_data;
+	/* transaction id. identifies a particular request/response exchange.
+	 * used when replying to a particular message. */
+	msgid_t msg_id;
+	/* the id of the task that sent a particular message. */
+	tid_t msg_sender;
+	/* the id of the port or channel used to send a particular message. */
+	koid_t msg_endpoint;
+	/* a list of iovecs that point to the buffers that make up the main
+	 * message data. */
+	kern_iovec_t *msg_data;
 	size_t msg_data_count;
-
+	/* a list of handle entries that contain the kernel handles included
-	struct handle_list *msg_handles;
+	 * in a message. */
+	kern_msg_handle_t *msg_handles;
 	size_t msg_handles_count;
-};
+} kern_msg_t;
 
 #endif

sched/core.c

@@ -91,7 +91,9 @@ void context_switch(struct thread *old, struct thread *new)
 
 void __schedule(enum sched_mode mode)
 {
+	if (mode != SCHED_IRQ) {
 		ml_int_disable();
+	}
 
 	struct cpu_data *this_cpu = get_this_cpu();
 	struct runqueue *rq = &this_cpu->c_rq;
@@ -142,7 +144,9 @@ void __schedule(enum sched_mode mode)
 		context_switch(prev, next);
 	}
 
+	if (mode != SCHED_IRQ) {
 		ml_int_enable();
+	}
 }
 
 void schedule(enum sched_mode mode)

syscall/dispatch.c

@@ -6,14 +6,17 @@
 
 static const virt_addr_t syscall_table[] = {
 	SYSCALL_TABLE_ENTRY(TASK_EXIT, task_exit),
+	SYSCALL_TABLE_ENTRY(TASK_SELF, task_self),
 	SYSCALL_TABLE_ENTRY(TASK_CREATE, task_create),
 	SYSCALL_TABLE_ENTRY(TASK_CREATE_THREAD, task_create_thread),
+	SYSCALL_TABLE_ENTRY(TASK_GET_ADDRESS_SPACE, task_get_address_space),
 	SYSCALL_TABLE_ENTRY(THREAD_START, thread_start),
 	SYSCALL_TABLE_ENTRY(VM_OBJECT_CREATE, vm_object_create),
 	SYSCALL_TABLE_ENTRY(VM_OBJECT_READ, vm_object_read),
 	SYSCALL_TABLE_ENTRY(VM_OBJECT_WRITE, vm_object_write),
 	SYSCALL_TABLE_ENTRY(VM_OBJECT_COPY, vm_object_copy),
 	SYSCALL_TABLE_ENTRY(VM_REGION_CREATE, vm_region_create),
+	SYSCALL_TABLE_ENTRY(VM_REGION_KILL, vm_region_kill),
 	SYSCALL_TABLE_ENTRY(VM_REGION_READ, vm_region_read),
 	SYSCALL_TABLE_ENTRY(VM_REGION_WRITE, vm_region_write),
 	SYSCALL_TABLE_ENTRY(VM_REGION_MAP_ABSOLUTE, vm_region_map_absolute),
@@ -32,9 +35,7 @@ static const virt_addr_t syscall_table[] = {
 	SYSCALL_TABLE_ENTRY(MSG_RECV, msg_recv),
 	SYSCALL_TABLE_ENTRY(MSG_REPLY, msg_reply),
 	SYSCALL_TABLE_ENTRY(MSG_READ, msg_read),
-	SYSCALL_TABLE_ENTRY(MSG_READ_HANDLES, msg_read_handles),
 	SYSCALL_TABLE_ENTRY(MSG_WRITE, msg_write),
-	SYSCALL_TABLE_ENTRY(MSG_WRITE_HANDLES, msg_write_handles),
 };
 static const size_t syscall_table_count
 	= sizeof syscall_table / sizeof syscall_table[0];

syscall/log.c

@@ -4,6 +4,6 @@
 kern_status_t sys_kern_log(const char *s)
 {
 	struct task *task = current_task();
-	printk("%s: %s", task->t_name, s);
+	printk("%s[%d]: %s", task->t_name, task->t_id, s);
 	return KERN_OK;
 }

syscall/msg.c

@@ -5,10 +5,7 @@
 #include <kernel/syscall.h>
 #include <kernel/vm-region.h>
 
-kern_status_t sys_channel_create(
+kern_status_t sys_channel_create(unsigned int id, kern_handle_t *out)
-	unsigned int id,
-	channel_flags_t flags,
-	kern_handle_t *out)
 {
 	struct task *self = current_task();
 	if (!validate_access_w(self, out, sizeof *out)) {
@@ -62,14 +59,13 @@ kern_status_t sys_port_create(kern_handle_t *out)
 	kern_handle_t handle;
 	kern_status_t status
 		= task_open_handle(self, &port->p_base, 0, &handle);
-	if (status != KERN_OK) {
 	task_unlock_irqrestore(self, irq_flags);
+
+	if (status != KERN_OK) {
 		object_unref(&port->p_base);
 		return status;
 	}
 
-	task_unlock_irqrestore(self, irq_flags);
-
 	*out = handle;
 	return KERN_OK;
 }
@@ -98,6 +94,7 @@ kern_status_t sys_port_connect(
 	/* add a reference to the port object to make sure it isn't deleted
 	 * while we're using it */
 	object_ref(port_obj);
+	struct port *port = port_cast(port_obj);
 	task_unlock_irqrestore(self, flags);
 
 	struct task *remote_task = task_from_tid(task_id);
@@ -116,81 +113,364 @@ kern_status_t sys_port_connect(
 	object_ref(&remote->c_base);
 	task_unlock_irqrestore(remote_task, flags);
 
-	status = port_connect(port_cast(port_obj), remote);
+	port_lock_irqsave(port, &flags);
+	status = port_connect(port, remote);
+	port_unlock_irqrestore(port, flags);
 	object_unref(port_obj);
 	object_unref(&remote->c_base);
 
 	return KERN_OK;
 }
 
-kern_status_t sys_port_disconnect(kern_handle_t port)
+kern_status_t sys_port_disconnect(kern_handle_t port_handle)
 {
-	return KERN_UNIMPLEMENTED;
+	unsigned long flags;
+
+	struct task *self = current_task();
+	task_lock_irqsave(self, &flags);
+
+	struct object *port_obj = NULL;
+	handle_flags_t port_handle_flags = 0;
+	kern_status_t status = task_resolve_handle(
+		self,
+		port_handle,
+		&port_obj,
+		&port_handle_flags);
+	if (status != KERN_OK) {
+		return status;
+	}
+
+	/* add a reference to the port object to make sure it isn't deleted
+	 * while we're using it */
+	object_ref(port_obj);
+	task_unlock_irqrestore(self, flags);
+
+	struct port *port = port_cast(port_obj);
+	if (!port) {
+		object_unref(port_obj);
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	object_unref(port_obj);
+	port_lock_irqsave(port, &flags);
+	status = port_disconnect(port);
+	port_unlock_irqrestore(port, flags);
+
+	return status;
+}
+
+static bool validate_iovec(
+	struct task *task,
+	const kern_iovec_t *iov,
+	size_t count,
+	bool rw)
+{
+	if (!validate_access_r(task, iov, count * sizeof(*iov))) {
+		return false;
+	}
+
+	for (size_t i = 0; i < count; i++) {
+		bool ok = false;
+		const kern_iovec_t *vec = &iov[i];
+		if (rw) {
+			ok = validate_access_w(task, vec->io_base, vec->io_len);
+		} else {
+			ok = validate_access_r(task, vec->io_base, vec->io_len);
+		}
+
+		if (!ok) {
+			return false;
+		}
+	}
+
+	return true;
+}
+
+static bool validate_msg(struct task *task, const kern_msg_t *msg, bool rw)
+{
+	if (!msg) {
+		return false;
+	}
+
+	vm_prot_t flags;
+	if (rw) {
+		flags = VM_PROT_WRITE | VM_PROT_USER;
+	} else {
+		flags = VM_PROT_READ | VM_PROT_USER;
+	}
+
+	if (!validate_access(task, msg, sizeof *msg, flags)) {
+		return false;
+	}
+
+	if (!validate_iovec(task, msg->msg_data, msg->msg_data_count, rw)) {
+		return false;
+	}
+
+	size_t handle_buffer_len
+		= msg->msg_handles_count * sizeof(*msg->msg_handles);
+	if (!validate_access(
+		    task,
+		    msg->msg_handles,
+		    handle_buffer_len,
+		    flags)) {
+		return false;
+	}
+
+	return true;
 }
 
 kern_status_t sys_msg_send(
-	kern_handle_t port,
+	kern_handle_t port_handle,
-	msg_flags_t flags,
+	const kern_msg_t *msg,
-	const struct msg *req,
+	kern_msg_t *out_reply)
-	struct msg *resp)
 {
-	return KERN_UNIMPLEMENTED;
+	struct task *self = current_task();
+
+	if (!validate_msg(self, msg, false)) {
+		return KERN_MEMORY_FAULT;
+	}
+
+	if (!validate_msg(self, out_reply, true)) {
+		return KERN_MEMORY_FAULT;
+	}
+
+	unsigned long flags;
+
+	task_lock_irqsave(self, &flags);
+
+	struct object *port_obj = NULL;
+	handle_flags_t port_handle_flags = 0;
+	kern_status_t status = task_resolve_handle(
+		self,
+		port_handle,
+		&port_obj,
+		&port_handle_flags);
+	if (status != KERN_OK) {
+		return status;
+	}
+
+	/* add a reference to the port object to make sure it isn't deleted
+	 * while we're using it */
+	object_ref(port_obj);
+	task_unlock_irqrestore(self, flags);
+
+	struct port *port = port_cast(port_obj);
+	if (!port) {
+		object_unref(port_obj);
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	port_lock_irqsave(port, &flags);
+	status = port_send_msg(port, msg, out_reply, &flags);
+	port_unlock_irqrestore(port, flags);
+	object_unref(port_obj);
+
+	return status;
 }
 
-kern_status_t sys_msg_recv(
+kern_status_t sys_msg_recv(kern_handle_t channel_handle, kern_msg_t *out_msg)
-	kern_handle_t channel,
-	msg_flags_t flags,
-	msgid_t *out_id,
-	struct msg *out_msg)
 {
-	return KERN_UNIMPLEMENTED;
+	struct task *self = current_task();
+
+	if (!validate_msg(self, out_msg, true)) {
+		return KERN_MEMORY_FAULT;
+	}
+
+	unsigned long flags;
+
+	task_lock_irqsave(self, &flags);
+
+	struct object *channel_obj = NULL;
+	handle_flags_t channel_handle_flags = 0;
+	kern_status_t status = task_resolve_handle(
+		self,
+		channel_handle,
+		&channel_obj,
+		&channel_handle_flags);
+	if (status != KERN_OK) {
+		return status;
+	}
+
+	/* add a reference to the port object to make sure it isn't deleted
+	 * while we're using it */
+	object_ref(channel_obj);
+	task_unlock_irqrestore(self, flags);
+
+	struct channel *channel = channel_cast(channel_obj);
+	if (!channel) {
+		object_unref(channel_obj);
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	channel_lock_irqsave(channel, &flags);
+	status = channel_recv_msg(channel, out_msg, &flags);
+	channel_unlock_irqrestore(channel, flags);
+	object_unref(channel_obj);
+
+	return status;
 }
 
 kern_status_t sys_msg_reply(
-	kern_handle_t channel,
+	kern_handle_t channel_handle,
-	msg_flags_t flags,
 	msgid_t id,
-	const struct msg *reply)
+	const kern_msg_t *reply)
 {
-	return KERN_UNIMPLEMENTED;
+	struct task *self = current_task();
+
+	if (!validate_msg(self, reply, true)) {
+		return KERN_MEMORY_FAULT;
+	}
+
+	unsigned long flags;
+
+	task_lock_irqsave(self, &flags);
+
+	struct object *channel_obj = NULL;
+	handle_flags_t channel_handle_flags = 0;
+	kern_status_t status = task_resolve_handle(
+		self,
+		channel_handle,
+		&channel_obj,
+		&channel_handle_flags);
+	if (status != KERN_OK) {
+		return status;
+	}
+
+	/* add a reference to the port object to make sure it isn't deleted
+	 * while we're using it */
+	object_ref(channel_obj);
+	task_unlock_irqrestore(self, flags);
+
+	struct channel *channel = channel_cast(channel_obj);
+	if (!channel) {
+		object_unref(channel_obj);
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	channel_lock_irqsave(channel, &flags);
+	status = channel_reply_msg(channel, id, reply, &flags);
+	channel_unlock_irqrestore(channel, flags);
+	object_unref(channel_obj);
+
+	return status;
 }
 
 kern_status_t sys_msg_read(
-	kern_handle_t channel,
+	kern_handle_t channel_handle,
 	msgid_t id,
 	size_t offset,
-	struct iovec *out,
+	const kern_iovec_t *iov,
-	size_t nr_out)
+	size_t iov_count,
+	size_t *nr_read)
 {
-	return KERN_UNIMPLEMENTED;
+	struct task *self = current_task();
-}
 
-kern_status_t sys_msg_read_handles(
+	if (nr_read && !validate_access_w(self, nr_read, sizeof *nr_read)) {
-	kern_handle_t channel,
+		return KERN_MEMORY_FAULT;
-	msgid_t id,
+	}
-	size_t offset,
+
-	struct handle_list *out,
+	if (!validate_iovec(self, iov, iov_count, true)) {
-	size_t nr_out)
+		return KERN_MEMORY_FAULT;
-{
+	}
-	return KERN_UNIMPLEMENTED;
+
+	unsigned long flags;
+
+	task_lock_irqsave(self, &flags);
+
+	struct object *channel_obj = NULL;
+	handle_flags_t channel_handle_flags = 0;
+	kern_status_t status = task_resolve_handle(
+		self,
+		channel_handle,
+		&channel_obj,
+		&channel_handle_flags);
+	if (status != KERN_OK) {
+		return status;
+	}
+
+	/* add a reference to the port object to make sure it isn't deleted
+	 * while we're using it */
+	object_ref(channel_obj);
+	task_unlock_irqrestore(self, flags);
+
+	struct channel *channel = channel_cast(channel_obj);
+	if (!channel) {
+		object_unref(channel_obj);
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	channel_lock_irqsave(channel, &flags);
+	status = channel_read_msg(
+		channel,
+		id,
+		offset,
+		self->t_address_space,
+		iov,
+		iov_count,
+		nr_read);
+	channel_unlock_irqrestore(channel, flags);
+	object_unref(channel_obj);
+
+	return status;
 }
 
 kern_status_t sys_msg_write(
-	kern_handle_t channel,
+	kern_handle_t channel_handle,
 	msgid_t id,
 	size_t offset,
-	const struct iovec *in,
+	const kern_iovec_t *iov,
-	size_t nr_in)
+	size_t iov_count,
+	size_t *nr_written)
 {
-	return KERN_UNIMPLEMENTED;
+	struct task *self = current_task();
-}
 
-kern_status_t sys_msg_write_handles(
+	if (nr_written
-	kern_handle_t channel,
+	    && !validate_access_w(self, nr_written, sizeof *nr_written)) {
-	msgid_t id,
+		return KERN_MEMORY_FAULT;
-	size_t offset,
+	}
-	const struct handle_list *in,
+
-	size_t nr_in)
+	if (!validate_iovec(self, iov, iov_count, false)) {
-{
+		return KERN_MEMORY_FAULT;
-	return KERN_UNIMPLEMENTED;
+	}
+
+	unsigned long flags;
+
+	task_lock_irqsave(self, &flags);
+
+	struct object *channel_obj = NULL;
+	handle_flags_t channel_handle_flags = 0;
+	kern_status_t status = task_resolve_handle(
+		self,
+		channel_handle,
+		&channel_obj,
+		&channel_handle_flags);
+	if (status != KERN_OK) {
+		return status;
+	}
+
+	/* add a reference to the port object to make sure it isn't deleted
+	 * while we're using it */
+	object_ref(channel_obj);
+	task_unlock_irqrestore(self, flags);
+
+	struct channel *channel = channel_cast(channel_obj);
+	if (!channel) {
+		object_unref(channel_obj);
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	channel_lock_irqsave(channel, &flags);
+	status = channel_write_msg(
+		channel,
+		id,
+		offset,
+		self->t_address_space,
+		iov,
+		iov_count,
+		nr_written);
+	channel_unlock_irqrestore(channel, flags);
+	object_unref(channel_obj);
+
+	return status;
 }

syscall/task.c

@@ -6,14 +6,44 @@
 
 extern kern_status_t sys_task_exit(int status)
 {
+	struct task *self = current_task();
+	printk("%s[%d]: task_exit(%d)", self->t_name, self->t_id, status);
 	while (1) {
-		printk("sys_exit(%d)", status);
+		milli_sleep(5000);
-		milli_sleep(1000);
 	}
 
 	return KERN_UNIMPLEMENTED;
 }
 
+kern_status_t sys_task_self(kern_handle_t *out)
+{
+	struct task *self = current_task();
+	if (!validate_access_w(self, out, sizeof *out)) {
+		return KERN_MEMORY_FAULT;
+	}
+
+	unsigned long flags;
+	task_lock_irqsave(self, &flags);
+
+	struct handle *handle_slot = NULL;
+	kern_handle_t handle;
+	kern_status_t status = handle_table_alloc_handle(
+		self->t_handles,
+		&handle_slot,
+		&handle);
+	task_unlock_irqrestore(self, flags);
+
+	if (status != KERN_OK) {
+		return status;
+	}
+
+	object_add_handle(&self->t_base);
+	handle_slot->h_object = &self->t_base;
+
+	*out = handle;
+	return KERN_OK;
+}
+
 kern_status_t sys_task_create(
 	kern_handle_t parent_handle,
 	const char *name,
@@ -175,6 +205,57 @@ kern_status_t sys_task_create_thread(
 	return KERN_OK;
 }
 
+kern_status_t sys_task_get_address_space(
+	kern_handle_t task_handle,
+	kern_handle_t *out)
+{
+	struct task *self = current_task();
+	if (!validate_access_w(self, out, sizeof *out)) {
+		return KERN_MEMORY_FAULT;
+	}
+
+	unsigned long flags;
+	task_lock_irqsave(self, &flags);
+
+	struct handle *handle_slot = NULL;
+	kern_handle_t handle;
+	struct object *task_obj = NULL;
+	handle_flags_t handle_flags = 0;
+	kern_status_t status = task_resolve_handle(
+		self,
+		task_handle,
+		&task_obj,
+		&handle_flags);
+	if (status != KERN_OK) {
+		task_unlock_irqrestore(self, flags);
+		return status;
+	}
+
+	status = handle_table_alloc_handle(
+		self->t_handles,
+		&handle_slot,
+		&handle);
+	if (status != KERN_OK) {
+		task_unlock_irqrestore(self, flags);
+		return status;
+	}
+
+	struct task *task = task_cast(task_obj);
+
+	if (!task) {
+		handle_table_free_handle(self->t_handles, handle);
+		task_unlock_irqrestore(self, flags);
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	handle_slot->h_object = &task->t_address_space->vr_base;
+	object_add_handle(&task->t_address_space->vr_base);
+	task_unlock_irqrestore(self, flags);
+
+	*out = handle;
+	return KERN_OK;
+}
+
 kern_status_t sys_thread_start(kern_handle_t thread_handle)
 {
 	unsigned long flags;

syscall/vm-object.c

@@ -111,6 +111,13 @@ kern_status_t sys_vm_object_copy(
 	size_t count,
 	size_t *nr_copied)
 {
+	tracek("vm_object_copy(%x, %zx, %x, %zx, %zx, %p)",
+	       dst,
+	       dst_offset,
+	       src,
+	       src_offset,
+	       count,
+	       nr_copied);
 	struct task *self = current_task();
 
 	if (nr_copied

syscall/vm-region.c

@@ -63,6 +63,7 @@ kern_status_t sys_vm_region_create(
 
 	object_ref(obj);
 	task_unlock_irqrestore(self, flags);
+	vm_region_lock_irqsave(parent_region, &flags);
 
 	struct vm_region *child = NULL;
 	status = vm_region_create(
@@ -73,6 +74,7 @@ kern_status_t sys_vm_region_create(
 		region_len,
 		prot,
 		&child);
+	vm_region_unlock_irqrestore(parent_region, flags);
 	object_unref(obj);
 
 	if (status != KERN_OK) {
@@ -92,6 +94,39 @@ kern_status_t sys_vm_region_create(
 	return KERN_OK;
 }
 
+kern_status_t sys_vm_region_kill(kern_handle_t region_handle)
+{
+	struct task *self = current_task();
+
+	unsigned long flags;
+	task_lock_irqsave(self, &flags);
+
+	struct object *obj = NULL;
+	handle_flags_t handle_flags = 0;
+	kern_status_t status
+		= task_resolve_handle(self, region_handle, &obj, &handle_flags);
+	if (status != KERN_OK) {
+		task_unlock_irqrestore(self, flags);
+		return status;
+	}
+
+	struct vm_region *region = vm_region_cast(obj);
+	if (!region) {
+		task_unlock_irqrestore(self, flags);
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	object_ref(obj);
+	task_unlock_irqrestore(self, flags);
+
+	vm_region_lock_irqsave(region, &flags);
+	status = vm_region_kill(region, &flags);
+	vm_region_unlock_irqrestore(region, flags);
+	object_unref(obj);
+
+	return status;
+}
+
 kern_status_t sys_vm_region_read(
 	kern_handle_t region_handle,
 	void *dst,

vm/cache.c

@@ -198,6 +198,8 @@ void *vm_cache_alloc(struct vm_cache *cache, enum vm_flags flags)
 	}
 
 	spin_unlock_irqrestore(&cache->c_lock, irq_flags);
+
+	memset(p, 0x0, cache->c_obj_size);
 	return p;
 }
 

vm/memblock.c

@@ -19,10 +19,11 @@
     contributors may be used to endorse or promote products derived from this
     software without specific prior written permission.
  */
-#include <limits.h>
 #include <kernel/libc/string.h>
 #include <kernel/memblock.h>
+#include <kernel/printk.h>
 #include <kernel/types.h>
+#include <limits.h>
 #include <stdbool.h>
 
 #define MIN(a, b)  ((a) < (b) ? (a) : (b))
@@ -447,10 +448,10 @@ void __next_memory_region(
 			}
 
 			/* we want the area that is overlapped by both
-			        region M (m_start - m_end) : The region defined
+			    - region M (m_start - m_end) : The region defined
-			   as system memory. region R (r_start - r_end) : The
+			      as system memory.
-			   region defined as free / outside of any reserved
+			    - region R (r_start - r_end) : The region defined as
-			   regions.
+			      free / outside of any reserved regions.
 			 */
 			it->it_base = MAX(m_start, r_start);
 			it->it_limit = MIN(m_end, r_end);
@@ -497,3 +498,28 @@ void *memblock_phys_to_virt(phys_addr_t p)
 {
 	return (void *)(p + memblock.m_voffset);
 }
+
+#ifdef TRACE
+static void memblock_type_dump(struct memblock_type *type)
+{
+	tracek("%s:", type->name);
+	for (size_t i = 0; i < type->count; i++) {
+		tracek("  [%zx-%zx]",
+		       type->regions[i].base,
+		       type->regions[i].limit);
+	}
+}
+
+extern void memblock_dump(void)
+{
+	memblock_type_dump(&memblock.memory);
+	memblock_type_dump(&memblock.reserved);
+
+	tracek("free:");
+	struct memblock_iter it;
+	for_each_free_mem_range(&it, 0, ADDR_MAX)
+	{
+		tracek("  [%zx-%zx]", it.it_base, it.it_limit);
+	}
+}
+#endif

vm/vm-object.c

@@ -96,8 +96,12 @@ static kern_status_t object_iterator_seek(
 	}
 
 	if (it->it_pg) {
-		it->it_buf = vm_page_get_vaddr(it->it_pg);
+		virt_addr_t vaddr = (virt_addr_t)vm_page_get_vaddr(it->it_pg);
+		vaddr += (it->it_offset & VM_PAGE_MASK);
+
+		it->it_buf = (void *)vaddr;
 		it->it_max = vm_page_get_size_bytes(it->it_pg);
+		it->it_max -= (it->it_offset & VM_PAGE_MASK);
 	} else {
 		struct btree_node *n = btree_first(&it->it_obj->vo_pages);
 		struct vm_page *pg
@@ -112,7 +116,8 @@ static kern_status_t object_iterator_seek(
 		}
 
 		it->it_buf = NULL;
-		it->it_max = pg ? pg->p_vmo_offset
+		it->it_max
+			= pg ? pg->p_vmo_offset - (it->it_offset & VM_PAGE_MASK)
 		             : it->it_obj->vo_size - it->it_offset;
 	}
 
@@ -281,6 +286,12 @@ extern struct vm_page *vm_object_alloc_page(
 			return NULL;
 		}
 
+		void *page_buf = vm_page_get_vaddr(page);
+		memset(page_buf, 0x0, vm_page_get_size_bytes(page));
+		tracek("vm-object: [%s] alloc offset %zx -> page %zx",
+		       vo->vo_name,
+		       offset,
+		       vm_page_get_paddr(page));
 		page->p_vmo_offset = offset;
 		vo->vo_pages.b_root = &page->p_bnode;
 		btree_insert_fixup(&vo->vo_pages, &page->p_bnode);

vm/vm-region.c

@@ -8,6 +8,20 @@
 #include <kernel/vm-region.h>
 #include <mango/status.h>
 
+/* NOTE Locking Rules
+ * To avoid deadlocks and crashes, the following locking rules should be
+ * followed:
+ *   1. Do NOT lock more than one region at a time IF the regions are siblings.
+ *   2. When locking a region and it's child(ren) or ancestors, always lock
+ *      the parent region BEFORE the child region.
+ *   3. When locking a region and a vm-object mapped into that region, always
+ *      lock the region BEFORE the vm-object.
+ *   3. An entry MUST be locked before any of its data can be read/written,
+ *      including its children (if it's a region) and its e_parent pointer.
+ *   4. vm_region_mapping has no lock. Instead, its immediate parent region must
+ *      be locked before any child mappings can be accessed.
+ */
+
 /*** STATIC DATA + MACROS *****************************************************/
 
 #undef ASLR
@@ -22,6 +36,15 @@
 	region_find_free_area_linear(region, length)
 #endif
 
+#define unlock_mapping_parent(p, root)                                         \
+	do {                                                                   \
+		struct vm_region *parent                                       \
+			= region_from_entry(p->m_entry.e_parent);              \
+		if (parent != root) {                                          \
+			vm_region_unlock(parent);                              \
+		}                                                              \
+	} while (0)
+
 /* iterates over a range of mapped virtual memory in a region, and provides
  * a moving buffer through which the memory can be accessed */
 struct vm_iterator {
@@ -49,10 +72,21 @@ enum search_direction {
 #define VM_REGION_CAST(p)                                                      \
 	OBJECT_C_CAST(struct vm_region, vr_base, &vm_region_type, p)
 
+static kern_status_t vm_region_object_destroy(struct object *obj);
+
+static kern_status_t region_object_destroy(struct object *obj, struct queue *q);
+static kern_status_t region_object_destroy_recurse(
+	struct queue_entry *entry,
+	struct object **out);
+
 static struct object_type vm_region_type = {
 	.ob_name = "vm-region",
 	.ob_size = sizeof(struct vm_region),
 	.ob_header_offset = offsetof(struct vm_region, vr_base),
+	.ob_ops = {
+		.destroy = region_object_destroy,
+		.destroy_recurse = region_object_destroy_recurse,
+	},
 };
 
 static struct vm_cache mapping_cache = {
@@ -81,17 +115,53 @@ static struct vm_region_mapping *mapping_from_entry(
 	return BTREE_CONTAINER(struct vm_region_mapping, m_entry, entry);
 }
 
-static virt_addr_t entry_absolute_address(const struct vm_region_entry *entry)
+kern_status_t region_object_destroy(struct object *obj, struct queue *q)
 {
-	virt_addr_t result = 0;
+	struct vm_region *region = VM_REGION_CAST(obj);
-	while (entry) {
+	if (region->vr_status == VM_REGION_ONLINE) {
-		result += entry->e_offset;
+		panic("last reference closed on an online vm-region");
-		entry = entry->e_parent;
 	}
 
-	return result;
+	struct btree_node *node = btree_first(&region->vr_entries);
+	while (node) {
+		struct btree_node *next = btree_next(node);
+		btree_delete(&region->vr_entries, node);
+
+		struct vm_region_entry *entry
+			= BTREE_CONTAINER(struct vm_region_entry, e_node, node);
+		if (entry->e_type != VM_REGION_ENTRY_REGION) {
+			panic("offline vm-region still contains non-region "
+			      "children.");
+		}
+
+		queue_push_back(q, &entry->e_entry);
+		node = next;
+	}
+
+	return KERN_OK;
 }
 
+kern_status_t region_object_destroy_recurse(
+	struct queue_entry *entry,
+	struct object **out)
+{
+	struct vm_region_entry *region_entry
+		= BTREE_CONTAINER(struct vm_region_entry, e_entry, entry);
+	if (region_entry->e_type != VM_REGION_ENTRY_REGION) {
+		panic("offline vm-region still contains non-region "
+		      "children.");
+	}
+	struct vm_region *region = region_from_entry(region_entry);
+	*out = &region->vr_base;
+	return KERN_OK;
+}
+
+static virt_addr_t entry_absolute_address(const struct vm_region_entry *entry)
+{
+	return entry->e_address;
+}
+
+/* this function must be called with `parent` locked */
 static void region_put_entry(
 	struct vm_region *parent,
 	struct vm_region_entry *child)
@@ -119,7 +189,15 @@ static void region_put_entry(
 		} else if (child_base > cur_limit) {
 			next = btree_right(cur);
 		} else {
-			panic("tried to add an overlapping entry to vm-region");
+#ifdef TRACE
+			vm_region_dump(parent);
+#endif
+			panic("tried to add an overlapping entry [%zx-%zx] to "
+			      "vm-region (overlaps [%zx-%zx])",
+			      child_base,
+			      child_limit,
+			      cur_base,
+			      cur_limit);
 		}
 
 		if (next) {
@@ -179,12 +257,17 @@ static struct vm_region_entry *region_get_entry(
 
 /* find the child region that covers the area [*offp,len]. searches recursively
  * the value in `offp` is updated to the offset of the returned entry relative
- * to its parent */
+ * to its parent.
+ * this function should be called with `region` locked.
+ * the region returned by this function will also be locked. any intermediary
+ * regions traversed by this function will be locked temporarily, but will
+ * be unlocked by the time the function returns. */
 static struct vm_region *region_get_child_region_recursive(
 	struct vm_region *region,
 	off_t *offp,
 	size_t len)
 {
+	struct vm_region *root = region;
 	off_t offset = *offp;
 	if (offset >= region->vr_entry.e_size) {
 		return NULL;
@@ -197,6 +280,14 @@ static struct vm_region *region_get_child_region_recursive(
 		struct vm_region *next_region = region_from_entry(next);
 		if (next_region) {
 			offset -= next->e_offset;
+			/* since `region` is locked, interrupts are already
+			 * disabled, so don't use lock_irq() here */
+			vm_region_lock(next_region);
+
+			if (region != root) {
+				vm_region_unlock(region);
+			}
+
 			region = next_region;
 		} else {
 			break;
@@ -207,20 +298,38 @@ static struct vm_region *region_get_child_region_recursive(
 	return region;
 }
 
+/* find the vm_region_mapping that contains a given memory area.
+ * `offp` should be a pointer to an off_t value that contains the offset
+ * of the area relative to the start of `region`. this value will be updated
+ * to the offset of the mapping relative to its immediate parent.
+ * this function should be called with `region` locked. if a mapping is found,
+ * it will be returned with its immediate parent locked. */
 static struct vm_region_mapping *region_get_mapping_recursive(
-	struct vm_region *region,
+	struct vm_region *root,
 	off_t *offp,
 	size_t len)
 {
 	off_t offset = *offp;
-	region = region_get_child_region_recursive(region, &offset, len);
+	struct vm_region *region
+		= region_get_child_region_recursive(root, &offset, len);
 	if (!region) {
 		return NULL;
 	}
 
+	/* if `region` is a different region than what was originally passed to
+	 * us, it has now been locked, and its children can be accessed. */
 	struct vm_region_entry *entry = region_get_entry(region, offset, len);
 	*offp = offset;
 
+	if (!entry) {
+		if (region != root) {
+			vm_region_unlock(region);
+		}
+
+		return NULL;
+	}
+
+	/* return the mapping with the parent region still locked */
 	return mapping_from_entry(entry);
 }
 
@@ -488,6 +597,7 @@ static kern_status_t region_validate_allocation(
 	return KERN_OK;
 }
 
+/* this function should be called with `region` locked */
 static void vm_iterator_begin(
 	struct vm_iterator *it,
 	struct vm_region *region,
@@ -501,7 +611,12 @@ static void vm_iterator_begin(
 
 	off_t offset = base - vm_region_get_base_address(region);
 	it->it_mapping = region_get_mapping_recursive(region, &offset, 1);
-	if (!it->it_mapping || (it->it_mapping->m_prot & prot) != prot) {
+	if (!it->it_mapping) {
+		return;
+	}
+
+	if ((it->it_mapping->m_prot & prot) != prot) {
+		unlock_mapping_parent(it->it_mapping, region);
 		return;
 	}
 
@@ -520,6 +635,7 @@ static void vm_iterator_begin(
 	}
 
 	if (!pg) {
+		unlock_mapping_parent(it->it_mapping, region);
 		return;
 	}
 
@@ -545,8 +661,8 @@ static void vm_iterator_begin(
 		buffer_size += vm_page_get_size_bytes(next);
 	}
 
-	it->it_buf = buffer_base;
+	it->it_buf = (char *)buffer_base + (object_offset & VM_PAGE_MASK);
-	it->it_max = buffer_size;
+	it->it_max = buffer_size - (object_offset & VM_PAGE_MASK);
 }
 
 static kern_status_t vm_iterator_seek(struct vm_iterator *it, size_t nr_bytes)
@@ -558,6 +674,10 @@ static kern_status_t vm_iterator_seek(struct vm_iterator *it, size_t nr_bytes)
 		return KERN_OK;
 	}
 
+	/* the parent region of it->it_mapping is locked here. if it is
+	 * different from it->it_region, it must be unlocked */
+	unlock_mapping_parent(it->it_mapping, it->it_region);
+
 	it->it_base += nr_bytes;
 	off_t offset = it->it_base - vm_region_get_base_address(it->it_region);
 
@@ -569,9 +689,13 @@ static kern_status_t vm_iterator_seek(struct vm_iterator *it, size_t nr_bytes)
 		return KERN_MEMORY_FAULT;
 	}
 
+	/* past this point, if we encounter an error, must remember to
+	 * unlock the parent region of next_mapping */
+
 	if ((next_mapping->m_prot & it->it_prot) != it->it_prot) {
 		it->it_buf = NULL;
 		it->it_max = 0;
+		unlock_mapping_parent(next_mapping, it->it_region);
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -590,6 +714,7 @@ static kern_status_t vm_iterator_seek(struct vm_iterator *it, size_t nr_bytes)
 	}
 
 	if (!pg) {
+		unlock_mapping_parent(next_mapping, it->it_region);
 		return KERN_NO_MEMORY;
 	}
 
@@ -615,11 +740,25 @@ static kern_status_t vm_iterator_seek(struct vm_iterator *it, size_t nr_bytes)
 		buffer_size += vm_page_get_size_bytes(next);
 	}
 
-	it->it_buf = buffer_base;
+	it->it_buf = (char *)buffer_base + (object_offset & VM_PAGE_MASK);
 	it->it_max = buffer_size;
 	return KERN_OK;
 }
 
+/* this function must be called when you are finished with a
+ * vm_iterator, to ensure that all held locks are released. */
+static void vm_iterator_finish(struct vm_iterator *it)
+{
+	if (it->it_mapping) {
+		unlock_mapping_parent(it->it_mapping, it->it_region);
+	}
+
+	memset(it, 0x0, sizeof *it);
+}
+
+/* this function must be called with `root` locked. `root` will be the
+ * first entry visited by the iterator. from there, child entries are
+ * visited in depth-first order. */
 static void entry_iterator_begin(
 	struct entry_iterator *it,
 	struct vm_region *root)
@@ -629,8 +768,43 @@ static void entry_iterator_begin(
 	it->it_entry = &root->vr_entry;
 }
 
+/* this function must be called when you are finished with an
+ * entry_iterator, to ensure that all held locks are released. */
+static void entry_iterator_finish(struct entry_iterator *it)
+{
+	struct vm_region_entry *cur = it->it_entry;
+	if (!cur) {
+		return;
+	}
+
+	struct vm_region *region = NULL;
+	if (cur->e_type == VM_REGION_ENTRY_MAPPING) {
+		region = region_from_entry(cur->e_parent);
+	} else {
+		region = region_from_entry(cur);
+	}
+
+	while (region && region != it->it_root) {
+		struct vm_region *parent
+			= region_from_entry(region->vr_entry.e_parent);
+		vm_region_unlock(region);
+		region = parent;
+	}
+
+	memset(it, 0x0, sizeof *it);
+}
+
+/* move to the next entry in the traversal order.
+ * when this function returns:
+ *   1. if the visited entry is a region, it will be locked.
+ *   2. if the visited entry is a mapping, its parent region will be
+ * locked. a region will remain locked until all of its children and
+ * n-grand-children have been visited. once iteration is finished, only
+ * `it->it_root` will be locked.
+ */
 static void entry_iterator_move_next(struct entry_iterator *it)
 {
+	/* `region` is locked */
 	struct vm_region *region = region_from_entry(it->it_entry);
 	bool has_children = (region && !btree_empty(&region->vr_entries));
 
@@ -639,6 +813,16 @@ static void entry_iterator_move_next(struct entry_iterator *it)
 		struct btree_node *node = btree_first(&region->vr_entries);
 		struct vm_region_entry *entry
 			= BTREE_CONTAINER(struct vm_region_entry, e_node, node);
+
+		if (entry->e_type == VM_REGION_ENTRY_REGION) {
+			struct vm_region *child_region
+				= region_from_entry(entry);
+			/* since `region` is locked, interrupts are
+			 * already disabled, so don't use lock_irq()
+			 * here */
+			vm_region_lock(child_region);
+		}
+
 		it->it_depth++;
 		it->it_entry = entry;
 		return;
@@ -670,6 +854,73 @@ static void entry_iterator_move_next(struct entry_iterator *it)
 			return;
 		}
 
+		if (cur->e_type == VM_REGION_ENTRY_REGION) {
+			struct vm_region *child_region = region_from_entry(cur);
+			if (child_region != it->it_root) {
+				vm_region_unlock(child_region);
+			}
+		}
+
+		it->it_depth--;
+		cur = parent_entry;
+	}
+}
+
+/* erase the current entry and move to the next entry in the traversal
+ * order. the current entry MUST be a mapping, otherwise nothing will
+ * happen.
+ */
+static void entry_iterator_erase(struct entry_iterator *it)
+{
+	/* the parent region of `mapping` is locked */
+	struct vm_region_mapping *mapping = mapping_from_entry(it->it_entry);
+	if (!mapping) {
+		return;
+	}
+
+	struct vm_region *parent = region_from_entry(mapping->m_entry.e_parent);
+
+	/* go back up until we find a right sibling. */
+	struct vm_region_entry *cur = it->it_entry;
+
+	while (1) {
+		struct btree_node *sibling = btree_next(&cur->e_node);
+		if (mapping) {
+			btree_delete(
+				&parent->vr_entries,
+				&mapping->m_entry.e_node);
+			vm_cache_free(&mapping_cache, mapping);
+			mapping = NULL;
+		}
+
+		if (sibling) {
+			it->it_entry = BTREE_CONTAINER(
+				struct vm_region_entry,
+				e_node,
+				sibling);
+			return;
+		}
+
+		if (cur == &it->it_root->vr_entry) {
+			it->it_entry = NULL;
+			return;
+		}
+
+		struct vm_region_entry *parent_entry = cur->e_parent;
+		struct vm_region *parent = region_from_entry(parent_entry);
+
+		if (!parent) {
+			it->it_entry = NULL;
+			return;
+		}
+
+		if (cur->e_type == VM_REGION_ENTRY_REGION) {
+			struct vm_region *child_region = region_from_entry(cur);
+			if (child_region != it->it_root) {
+				vm_region_unlock(child_region);
+			}
+		}
+
 		it->it_depth--;
 		cur = parent_entry;
 	}
@@ -705,6 +956,11 @@ static void mapping_iterator_begin(
 	}
 }
 
+static void mapping_iterator_finish(struct entry_iterator *it)
+{
+	entry_iterator_finish(it);
+}
+
 static void mapping_iterator_move_next(
 	struct entry_iterator *it,
 	off_t offset,
@@ -730,6 +986,34 @@ static void mapping_iterator_move_next(
 	}
 }
 
+static void mapping_iterator_erase(
+	struct entry_iterator *it,
+	off_t offset,
+	size_t length,
+	off_t *offp)
+{
+	entry_iterator_erase(it);
+
+	while (it->it_entry
+	       && it->it_entry->e_type != VM_REGION_ENTRY_MAPPING) {
+
+		entry_iterator_move_next(it);
+	}
+
+	if (!it->it_entry) {
+		return;
+	}
+
+	off_t base = entry_absolute_address(it->it_entry)
+	           - it->it_root->vr_entry.e_offset;
+
+	if (base >= offset + length) {
+		it->it_entry = NULL;
+	} else {
+		*offp = base;
+	}
+}
+
 /*** PUBLIC API ***************************************************************/
 
 kern_status_t vm_region_type_init(void)
@@ -743,6 +1027,9 @@ struct vm_region *vm_region_cast(struct object *obj)
 	return VM_REGION_CAST(obj);
 }
 
+/* this function should be called with `parent` locked (if parent is
+ * non-NULL)
+ */
 kern_status_t vm_region_create(
 	struct vm_region *parent,
 	const char *name,
@@ -752,6 +1039,10 @@ kern_status_t vm_region_create(
 	vm_prot_t prot,
 	struct vm_region **out)
 {
+	if (parent && parent->vr_status != VM_REGION_ONLINE) {
+		return KERN_BAD_STATE;
+	}
+
 	if (!offset || !region_len) {
 		return KERN_INVALID_ARGUMENT;
 	}
@@ -781,8 +1072,10 @@ kern_status_t vm_region_create(
 
 	struct vm_region *region = VM_REGION_CAST(region_object);
 
+	region->vr_status = VM_REGION_ONLINE;
 	region->vr_prot = prot;
 	region->vr_entry.e_type = VM_REGION_ENTRY_REGION;
+	region->vr_entry.e_address = offset;
 	region->vr_entry.e_offset = offset;
 	region->vr_entry.e_size = region_len;
 
@@ -794,8 +1087,11 @@ kern_status_t vm_region_create(
 
 	if (parent) {
 		region->vr_entry.e_parent = &parent->vr_entry;
+		region->vr_entry.e_address += parent->vr_entry.e_address;
 		region->vr_pmap = parent->vr_pmap;
 		region_put_entry(parent, &region->vr_entry);
+		/* `parent` holds a reference to child `region` */
+		object_ref(&region->vr_base);
 	}
 
 	if (name && name_len) {
@@ -808,8 +1104,67 @@ kern_status_t vm_region_create(
 	return KERN_OK;
 }
 
-kern_status_t vm_region_map_object(
+kern_status_t vm_region_kill(
 	struct vm_region *region,
+	unsigned long *lock_flags)
+{
+	if (region->vr_status != VM_REGION_ONLINE) {
+		return KERN_BAD_STATE;
+	}
+
+	if (region->vr_entry.e_parent) {
+		struct vm_region *parent
+			= region_from_entry(region->vr_entry.e_parent);
+		region->vr_entry.e_parent = NULL;
+
+		/* locks must be acquired in parent->child order. since
+		 * we're going backwards here, unlock `region` before
+		 * locking its parent */
+		vm_region_unlock_irqrestore(region, *lock_flags);
+		vm_region_lock_irqsave(parent, lock_flags);
+		btree_delete(&parent->vr_entries, &region->vr_entry.e_node);
+		vm_region_unlock_irqrestore(parent, *lock_flags);
+		vm_region_lock_irqsave(region, lock_flags);
+		/* `region` lock is held, and e_parent is NULL */
+	}
+
+	struct entry_iterator it;
+	entry_iterator_begin(&it, region);
+
+	while (it.it_entry) {
+		if (it.it_entry->e_type == VM_REGION_ENTRY_REGION) {
+			struct vm_region *region
+				= region_from_entry(it.it_entry);
+			region->vr_status = VM_REGION_DEAD;
+			entry_iterator_move_next(&it);
+			continue;
+		}
+
+		struct vm_region_mapping *mapping
+			= mapping_from_entry(it.it_entry);
+
+		virt_addr_t base = entry_absolute_address(it.it_entry);
+
+		for (size_t i = 0; i < mapping->m_entry.e_size;
+		     i += VM_PAGE_SIZE) {
+			pmap_remove(region->vr_pmap, base + i);
+		}
+
+		unsigned long flags;
+		vm_object_lock_irqsave(mapping->m_object, &flags);
+		queue_delete(
+			&mapping->m_object->vo_mappings,
+			&mapping->m_object_entry);
+		vm_object_unlock_irqrestore(mapping->m_object, flags);
+
+		entry_iterator_erase(&it);
+	}
+
+	return KERN_OK;
+}
+
+kern_status_t vm_region_map_object(
+	struct vm_region *root,
 	off_t region_offset,
 	struct vm_object *object,
 	off_t object_offset,
@@ -839,10 +1194,24 @@ kern_status_t vm_region_map_object(
 		length += VM_PAGE_SIZE;
 	}
 
-	if (!region || !object) {
+	if (!root || !object) {
 		return KERN_INVALID_ARGUMENT;
 	}
 
+	struct vm_region *region = root;
+	if (region_offset != VM_REGION_ANY_OFFSET) {
+		region = region_get_child_region_recursive(
+			root,
+			&region_offset,
+			length);
+		/* if `region` != `root`, it will need to be unlocked at
+		 * the end of the function */
+	}
+
+	if (region->vr_status != VM_REGION_ONLINE) {
+		return KERN_BAD_STATE;
+	}
+
 	if ((prot & region->vr_prot) != prot) {
 		return KERN_INVALID_ARGUMENT;
 	}
@@ -855,13 +1224,6 @@ kern_status_t vm_region_map_object(
 		return KERN_INVALID_ARGUMENT;
 	}
 
-	if (region_offset != VM_REGION_ANY_OFFSET) {
-		region = region_get_child_region_recursive(
-			region,
-			&region_offset,
-			length);
-	}
-
 	if (!region) {
 		return KERN_INVALID_ARGUMENT;
 	}
@@ -876,6 +1238,15 @@ kern_status_t vm_region_map_object(
 		return KERN_INVALID_ARGUMENT;
 	}
 
+	tracek("vm_region_map_object(%s, %zx, %s, %zx, %zx, %x, %p)",
+	       region->vr_name,
+	       region_offset,
+	       object->vo_name,
+	       object_offset,
+	       length,
+	       prot,
+	       out);
+
 	struct vm_region_mapping *mapping
 		= vm_cache_alloc(&mapping_cache, VM_NORMAL);
 	if (!mapping) {
@@ -887,6 +1258,7 @@ kern_status_t vm_region_map_object(
 	mapping->m_object_offset = object_offset;
 	mapping->m_entry.e_type = VM_REGION_ENTRY_MAPPING;
 	mapping->m_entry.e_parent = &region->vr_entry;
+	mapping->m_entry.e_address = region->vr_entry.e_address + region_offset;
 	mapping->m_entry.e_offset = region_offset;
 	mapping->m_entry.e_size = length;
 
@@ -898,7 +1270,14 @@ kern_status_t vm_region_map_object(
 	       abs_base + length);
 #endif
 	region_put_entry(region, &mapping->m_entry);
+	if (region != root) {
+		vm_region_unlock(region);
+	}
+
+	unsigned long lock_flags;
+	vm_object_lock_irqsave(object, &lock_flags);
 	queue_push_back(&object->vo_mappings, &mapping->m_object_entry);
+	vm_object_unlock_irqrestore(object, lock_flags);
 
 	if (out) {
 		*out = entry_absolute_address(&mapping->m_entry);
@@ -907,8 +1286,8 @@ kern_status_t vm_region_map_object(
 	return KERN_OK;
 }
 
-/* unmap some pages in the middle of a mapping, splitting it into two separate
+/* unmap some pages in the middle of a mapping, splitting it into two
- * mappings */
+ * separate mappings */
 static kern_status_t split_mapping(
 	struct vm_region_mapping *mapping,
 	struct vm_region *root,
@@ -1059,12 +1438,13 @@ static kern_status_t delete_mapping(
 		pmap_remove(root->vr_pmap, base + i);
 	}
 
-	struct vm_region *parent = region_from_entry(mapping->m_entry.e_parent);
+	unsigned long flags;
-
+	vm_object_lock_irqsave(mapping->m_object, &flags);
 	queue_delete(&mapping->m_object->vo_mappings, &mapping->m_object_entry);
-	btree_delete(&parent->vr_entries, &mapping->m_entry.e_node);
+	vm_object_unlock_irqrestore(mapping->m_object, flags);
 
-	vm_cache_free(&mapping_cache, mapping);
+	/* don't actually delete the mapping yet. that will be done by
+	 * vm_region_unmap */
 
 	return KERN_OK;
 }
@@ -1074,6 +1454,10 @@ kern_status_t vm_region_unmap(
 	off_t unmap_area_offset,
 	size_t unmap_area_length)
 {
+	if (region->vr_status != VM_REGION_ONLINE) {
+		return KERN_BAD_STATE;
+	}
+
 	kern_status_t status = KERN_OK;
 	struct entry_iterator it;
 	off_t unmap_area_limit = unmap_area_offset + unmap_area_length;
@@ -1092,12 +1476,6 @@ kern_status_t vm_region_unmap(
 		off_t mapping_offset = tmp;
 		off_t mapping_limit = mapping_offset + it.it_entry->e_size;
 
-		mapping_iterator_move_next(
-			&it,
-			unmap_area_offset,
-			unmap_area_length,
-			&tmp);
-
 		bool split
 			= (unmap_area_offset > mapping_offset
 		           && unmap_area_limit < mapping_limit);
@@ -1118,6 +1496,7 @@ kern_status_t vm_region_unmap(
 				mapping_offset,
 				unmap_area_offset,
 				unmap_area_limit);
+			delete = true;
 		} else if (delete) {
 			status = delete_mapping(
 				mapping,
@@ -1138,7 +1517,22 @@ kern_status_t vm_region_unmap(
 				unmap_area_offset,
 				unmap_area_limit);
 		} else {
-			panic("don't know what to do with this mapping");
+			panic("don't know what to do with this "
+			      "mapping");
+		}
+
+		if (delete) {
+			mapping_iterator_erase(
+				&it,
+				unmap_area_offset,
+				unmap_area_length,
+				&tmp);
+		} else {
+			mapping_iterator_move_next(
+				&it,
+				unmap_area_offset,
+				unmap_area_length,
+				&tmp);
 		}
 
 		if (status != KERN_OK) {
@@ -1146,6 +1540,8 @@ kern_status_t vm_region_unmap(
 		}
 	}
 
+	mapping_iterator_finish(&it);
+
 	return status;
 }
 
@@ -1155,6 +1551,10 @@ bool vm_region_validate_access(
 	size_t len,
 	vm_prot_t prot)
 {
+	if (region->vr_status != VM_REGION_ONLINE) {
+		return false;
+	}
+
 	if (len == 0) {
 		return true;
 	}
@@ -1187,7 +1587,16 @@ bool vm_region_validate_access(
 			return false;
 		}
 
-		if ((mapping->m_prot & prot) != prot) {
+		bool ok = (mapping->m_prot & prot) == prot;
+
+		struct vm_region *parent
+			= region_from_entry(mapping->m_entry.e_parent);
+
+		if (parent != region) {
+			vm_region_unlock(parent);
+		}
+
+		if (!ok) {
 			return false;
 		}
 	}
@@ -1195,11 +1604,16 @@ bool vm_region_validate_access(
 	return true;
 }
 
+/* this function must be called with `region` locked */
 kern_status_t vm_region_demand_map(
 	struct vm_region *region,
 	virt_addr_t addr,
 	enum pmap_fault_flags flags)
 {
+	if (region->vr_status != VM_REGION_ONLINE) {
+		return KERN_BAD_STATE;
+	}
+
 	addr &= ~VM_PAGE_MASK;
 	if (addr < region->vr_entry.e_offset
 	    || addr > region->vr_entry.e_offset + region->vr_entry.e_size) {
@@ -1221,24 +1635,130 @@ kern_status_t vm_region_demand_map(
 	       mapping->m_object->vo_name,
 	       object_offset);
 
+	unsigned long lock_flags;
+	vm_object_lock_irqsave(mapping->m_object, &lock_flags);
 	struct vm_page *pg = vm_object_alloc_page(
 		mapping->m_object,
 		object_offset,
 		VM_PAGE_4K);
+	vm_object_unlock_irqrestore(mapping->m_object, lock_flags);
 	tracek("vm: mapping %07llx -> %10llx", vm_page_get_paddr(pg), addr);
-	return pmap_add(
+	kern_status_t status = pmap_add(
 		region->vr_pmap,
 		addr,
 		vm_page_get_pfn(pg),
 		mapping->m_prot,
 		PMAP_NORMAL);
+
+	struct vm_region *parent = region_from_entry(mapping->m_entry.e_parent);
+	if (parent != region) {
+		vm_region_unlock(parent);
+	}
+
+	return status;
 }
 
 virt_addr_t vm_region_get_base_address(const struct vm_region *region)
 {
+	if (region->vr_status != VM_REGION_ONLINE) {
+		return 0;
+	}
+
 	return entry_absolute_address(&region->vr_entry);
 }
 
+kern_status_t vm_region_read_kernel(
+	struct vm_region *src_region,
+	virt_addr_t src_ptr,
+	size_t count,
+	void *destp,
+	size_t *nr_read)
+{
+	if (src_region->vr_status != VM_REGION_ONLINE) {
+		return KERN_BAD_STATE;
+	}
+
+	struct vm_iterator src;
+	char *dest = destp;
+
+	vm_iterator_begin(
+		&src,
+		src_region,
+		src_ptr,
+		VM_PROT_READ | VM_PROT_USER);
+
+	kern_status_t status = KERN_OK;
+	size_t r = 0;
+
+	while (r < count && src.it_max) {
+		size_t remaining = count - r;
+		size_t to_move = MIN(src.it_max, remaining);
+		memmove(dest, src.it_buf, to_move);
+
+		status = vm_iterator_seek(&src, to_move);
+		if (status != KERN_OK) {
+			break;
+		}
+
+		r += to_move;
+		dest += to_move;
+	}
+
+	vm_iterator_finish(&src);
+
+	if (nr_read) {
+		*nr_read = r;
+	}
+
+	return status;
+}
+
+kern_status_t vm_region_write_kernel(
+	struct vm_region *dst_region,
+	virt_addr_t dst_ptr,
+	size_t count,
+	const void *srcp,
+	size_t *nr_written)
+{
+	if (dst_region->vr_status != VM_REGION_ONLINE) {
+		return KERN_BAD_STATE;
+	}
+
+	struct vm_iterator dst;
+	const char *src = srcp;
+
+	vm_iterator_begin(
+		&dst,
+		dst_region,
+		dst_ptr,
+		VM_PROT_WRITE | VM_PROT_USER);
+
+	kern_status_t status = KERN_OK;
+	size_t r = 0;
+
+	while (r < count && dst.it_max) {
+		size_t remaining = count - r;
+		size_t to_move = MIN(dst.it_max, remaining);
+		memmove(dst.it_buf, src, to_move);
+
+		status = vm_iterator_seek(&dst, to_move);
+		if (status != KERN_OK) {
+			break;
+		}
+
+		r += to_move;
+		src += to_move;
+	}
+
+	vm_iterator_finish(&dst);
+
+	if (nr_written) {
+		*nr_written = r;
+	}
+
+	return status;
+}
+
 kern_status_t vm_region_memmove(
 	struct vm_region *dest_region,
 	virt_addr_t dest_ptr,
@@ -1247,6 +1767,14 @@ kern_status_t vm_region_memmove(
 	size_t count,
 	size_t *nr_moved)
 {
+	if (src_region->vr_status != VM_REGION_ONLINE) {
+		return KERN_BAD_STATE;
+	}
+
+	if (dest_region->vr_status != VM_REGION_ONLINE) {
+		return KERN_BAD_STATE;
+	}
+
 	struct vm_iterator src, dest;
 	vm_iterator_begin(
 		&src,
@@ -1280,6 +1808,9 @@ kern_status_t vm_region_memmove(
 		r += to_move;
 	}
 
+	vm_iterator_finish(&src);
+	vm_iterator_finish(&dest);
+
 	if (nr_moved) {
 		*nr_moved = r;
 	}
@@ -1290,21 +1821,31 @@ kern_status_t vm_region_memmove(
 extern kern_status_t vm_region_memmove_v(
 	struct vm_region *dest_region,
 	size_t dest_offset,
-	struct iovec *dest_vecs,
+	const kern_iovec_t *dest_vecs,
 	size_t nr_dest_vecs,
 	struct vm_region *src_region,
 	size_t src_offset,
-	const struct iovec *src_vecs,
+	const kern_iovec_t *src_vecs,
 	size_t nr_src_vecs,
-	size_t bytes_to_move)
+	size_t bytes_to_move,
+	size_t *nr_bytes_moved)
 {
+	if (src_region->vr_status != VM_REGION_ONLINE) {
+		return KERN_BAD_STATE;
+	}
+
+	if (dest_region->vr_status != VM_REGION_ONLINE) {
+		return KERN_BAD_STATE;
+	}
+
 	struct iovec_iterator src, dest;
-	iovec_iterator_begin(&src, src_vecs, nr_src_vecs);
+	iovec_iterator_begin_user(&src, src_region, src_vecs, nr_src_vecs);
-	iovec_iterator_begin(&dest, dest_vecs, nr_dest_vecs);
+	iovec_iterator_begin_user(&dest, dest_region, dest_vecs, nr_dest_vecs);
 
 	iovec_iterator_seek(&src, src_offset);
 	iovec_iterator_seek(&dest, dest_offset);
 
+	size_t moved = 0;
 	while (bytes_to_move && src.it_len && dest.it_len) {
 		size_t to_move
 			= MIN(MIN(src.it_len, dest.it_len), bytes_to_move);
@@ -1323,6 +1864,11 @@ extern kern_status_t vm_region_memmove_v(
 		iovec_iterator_seek(&src, to_move);
 		iovec_iterator_seek(&dest, to_move);
 		bytes_to_move -= to_move;
+		moved += to_move;
+	}
+
+	if (nr_bytes_moved) {
+		*nr_bytes_moved = moved;
 	}
 
 	return KERN_OK;

vm/zone.c

@@ -1,14 +1,19 @@
-#include <kernel/locks.h>
-#include <kernel/util.h>
-#include <kernel/queue.h>
-#include <kernel/memblock.h>
-#include <kernel/types.h>
-#include <kernel/vm.h>
-#include <kernel/printk.h>
 #include <kernel/libc/string.h>
+#include <kernel/locks.h>
 #include <kernel/machine/cpu.h>
+#include <kernel/memblock.h>
+#include <kernel/panic.h>
+#include <kernel/printk.h>
+#include <kernel/queue.h>
+#include <kernel/types.h>
+#include <kernel/util.h>
+#include <kernel/vm.h>
 
-static struct vm_page *group_pages_into_block(struct vm_zone *z, phys_addr_t base, phys_addr_t limit, int order)
+static struct vm_page *group_pages_into_block(
+	struct vm_zone *z,
+	phys_addr_t base,
+	phys_addr_t limit,
+	int order)
 {
 	struct vm_page *first_page = NULL;
 	for (phys_addr_t i = base; i < limit; i += VM_PAGE_SIZE) {
@@ -37,16 +42,23 @@ static struct vm_page *group_pages_into_block(struct vm_zone *z, phys_addr_t bas
 	return first_page;
 }
 
-static void convert_region_to_blocks(struct vm_zone *zone,
+static void convert_region_to_blocks(
-				     phys_addr_t base, phys_addr_t limit,
+	struct vm_zone *zone,
+	phys_addr_t base,
+	phys_addr_t limit,
 	int reserved)
 {
+	if (base & VM_PAGE_MASK || (limit + 1) & VM_PAGE_MASK) {
+		panic("convert_region_to_blocks: region must be page-aligned");
+	}
+
 	size_t block_frames = vm_bytes_to_pages(limit - base + 1);
 	int reset_order = 0;
 
-	for (int order = VM_PAGE_MAX_ORDER; order >= VM_PAGE_MIN_ORDER; ) {
+	for (int order = VM_PAGE_MAX_ORDER; order >= VM_PAGE_MIN_ORDER;) {
 		size_t order_frames = vm_page_order_to_pages(order);
-		vm_alignment_t order_alignment = vm_page_order_to_alignment(order);
+		vm_alignment_t order_alignment
+			= vm_page_order_to_alignment(order);
 
 		if (order_frames > block_frames) {
 			order--;
@@ -59,11 +71,18 @@ static void convert_region_to_blocks(struct vm_zone *zone,
 			continue;
 		}
 
-		phys_addr_t block_limit = base + (order_frames * VM_PAGE_SIZE) - 1;
+		phys_addr_t block_limit
-		struct vm_page *block_page = group_pages_into_block(zone, base, block_limit, order);
+			= base + (order_frames * VM_PAGE_SIZE) - 1;
+		struct vm_page *block_page = group_pages_into_block(
+			zone,
+			base,
+			block_limit,
+			order);
 
 		if (reserved == 0) {
-			queue_push_back(&zone->z_free_pages[order], &block_page->p_list);
+			queue_push_back(
+				&zone->z_free_pages[order],
+				&block_page->p_list);
 		}
 
 		base = block_limit + 1;
@@ -83,10 +102,12 @@ static void convert_region_to_blocks(struct vm_zone *zone,
 static size_t zone_free_bytes(struct vm_zone *z)
 {
 	size_t free_bytes = 0;
-	for (enum vm_page_order i = VM_PAGE_MIN_ORDER; i <= VM_PAGE_MAX_ORDER; i++) {
+	for (enum vm_page_order i = VM_PAGE_MIN_ORDER; i <= VM_PAGE_MAX_ORDER;
+	     i++) {
 		size_t page_bytes = vm_page_order_to_bytes(i);
 		size_t nr_pages = 0;
-		queue_foreach (struct vm_page, pg, &z->z_free_pages[i], p_list) {
+		queue_foreach(struct vm_page, pg, &z->z_free_pages[i], p_list)
+		{
 			free_bytes += page_bytes;
 			nr_pages++;
 		}
@@ -106,24 +127,37 @@ void vm_zone_init(struct vm_zone *z, const struct vm_zone_descriptor *zone_info)
 
 	struct memblock_iter it;
 
-	/* TODO this only creates page blocks for free memory regions, not reserved memory regions.
+	/* TODO this only creates page blocks for free memory regions, not
-	 * this is faster for systems that have huge amounts of reserved memory, but it means
+	 * reserved memory regions. this is faster for systems that have huge
-	 * that a call to vm_page_get() for a reserved memory region will return null
+	 * amounts of reserved memory, but it means that a call to vm_page_get()
-	 * rather than a reserved page.
+	 * for a reserved memory region will return null rather than a reserved
+	 * page.
 	 *
-	 * vm_page_get() should probably create reserved pages on-demand for these regions. */
+	 * vm_page_get() should probably create reserved pages on-demand for
+	 * these regions. */
 
 	size_t nr_pages_found = 0;
-	for_each_free_mem_range(&it, z->z_info.zd_base, z->z_info.zd_limit) {
+	for_each_free_mem_range(&it, z->z_info.zd_base, z->z_info.zd_limit)
+	{
+		it.it_base &= ~VM_PAGE_MASK;
+		if (it.it_limit & VM_PAGE_MASK) {
+			it.it_limit &= ~VM_PAGE_MASK;
+			it.it_limit += VM_PAGE_SIZE;
+		}
+
 		phys_addr_t block_start = it.it_base, block_end = it.it_limit;
+
 		int this_page_reserved = 0, last_page_reserved = -1;
 
-		for (uintptr_t i = it.it_base; i < it.it_limit; i += VM_PAGE_SIZE) {
+		for (uintptr_t i = it.it_base; i < it.it_limit;
+		     i += VM_PAGE_SIZE) {
 			struct vm_page *pg = vm_page_get(i);
 
 			if (pg) {
 				nr_pages_found++;
-				this_page_reserved = (pg->p_flags & VM_PAGE_RESERVED) ? 1 : 0;
+				this_page_reserved
+					= (pg->p_flags & VM_PAGE_RESERVED) ? 1
+				                                           : 0;
 			} else {
 				this_page_reserved = 1;
 			}
@@ -139,20 +173,28 @@ void vm_zone_init(struct vm_zone *z, const struct vm_zone_descriptor *zone_info)
 
 			convert_region_to_blocks(
 				z,
-					block_start, block_end + VM_PAGE_SIZE - 1,
+				block_start,
+				block_end + VM_PAGE_SIZE - 1,
 				last_page_reserved);
 
 			block_start = i;
+			if (block_start & VM_PAGE_MASK) {
+				block_start &= ~VM_PAGE_MASK;
+				block_start += VM_PAGE_SIZE;
+			}
 			last_page_reserved = this_page_reserved;
+			nr_pages_found = 0;
 		}
 
 		if (block_start != block_end) {
-			/* either the entire zone is homogeneous (all free/all reserved) or the entire zone is empty. */
+			/* either the entire zone is homogeneous (all free/all
+			 * reserved) or the entire zone is empty. */
 			if (nr_pages_found > 0) {
 				/* the entire zone is homogeneous :) */
 				convert_region_to_blocks(
 					z,
-						block_start, block_end + VM_PAGE_SIZE - 1,
+					block_start,
+					block_end + VM_PAGE_SIZE - 1,
 					this_page_reserved);
 			}
 		}
@@ -201,7 +243,10 @@ void vm_zone_init(struct vm_zone *z, const struct vm_zone_descriptor *zone_info)
 	char free_bytes_str[64];
 	data_size_to_string(free_bytes, free_bytes_str, sizeof free_bytes_str);
 
-	printk("vm: zone %u/%s: %s of memory online.", z->z_info.zd_node, z->z_info.zd_name, free_bytes_str);
+	printk("vm: zone %u/%s: %s of memory online.",
+	       z->z_info.zd_node,
+	       z->z_info.zd_name,
+	       free_bytes_str);
 }
 
 static int replenish_free_page_list(struct vm_zone *z, enum vm_page_order order)
@@ -216,7 +261,8 @@ static int replenish_free_page_list(struct vm_zone *z, enum vm_page_order order)
 		return -1;
 	}
 
-	/* the lowest page order that is >= `order` and still has pages available */
+	/* the lowest page order that is >= `order` and still has pages
+	 * available */
 	enum vm_page_order first_order_with_free = VM_MAX_PAGE_ORDERS;
 
 	for (enum vm_page_order i = order; i <= VM_PAGE_MAX_ORDER; i++) {
@@ -232,7 +278,8 @@ static int replenish_free_page_list(struct vm_zone *z, enum vm_page_order order)
 	}
 
 	if (first_order_with_free == order) {
-		/* there are free pages of the requested order, so nothing needs to be done */
+		/* there are free pages of the requested order, so nothing needs
+		 * to be done */
 		return 0;
 	}
 
@@ -240,8 +287,10 @@ static int replenish_free_page_list(struct vm_zone *z, enum vm_page_order order)
 	   take a page, split it in half, and add the sub-pages
 	   to the next order's free list. */
 	for (enum vm_page_order i = first_order_with_free; i > order; i--) {
-		struct queue_entry *pg_entry = queue_pop_front(&z->z_free_pages[i]);
+		struct queue_entry *pg_entry
-		struct vm_page *pg = QUEUE_CONTAINER(struct vm_page, p_list, pg_entry);
+			= queue_pop_front(&z->z_free_pages[i]);
+		struct vm_page *pg
+			= QUEUE_CONTAINER(struct vm_page, p_list, pg_entry);
 
 		struct vm_page *a, *b;
 		vm_page_split(pg, &a, &b);
@@ -253,7 +302,10 @@ static int replenish_free_page_list(struct vm_zone *z, enum vm_page_order order)
 	return 0;
 }
 
-struct vm_page *vm_zone_alloc_page(struct vm_zone *z, enum vm_page_order order, enum vm_flags flags)
+struct vm_page *vm_zone_alloc_page(
+	struct vm_zone *z,
+	enum vm_page_order order,
+	enum vm_flags flags)
 {
 	unsigned long irq_flags;
 	spin_lock_irqsave(&z->z_lock, &irq_flags);
@@ -266,7 +318,8 @@ struct vm_page *vm_zone_alloc_page(struct vm_zone *z, enum vm_page_order order,
 
 	struct queue_entry *pg_entry = queue_pop_front(&z->z_free_pages[order]);
 	struct vm_page *pg = QUEUE_CONTAINER(struct vm_page, p_list, pg_entry);
-	vm_page_foreach (pg, i) {
+	vm_page_foreach(pg, i)
+	{
 		i->p_flags |= VM_PAGE_ALLOC;
 	}
 
@@ -289,7 +342,9 @@ void vm_zone_free_page(struct vm_zone *z, struct vm_page *pg)
 			break;
 		}
 
-		queue_delete(&z->z_free_pages[buddy->p_order - 1], &buddy->p_list);
+		queue_delete(
+			&z->z_free_pages[buddy->p_order - 1],
+			&buddy->p_list);
 		queue_delete(&z->z_free_pages[buddy->p_order - 1], &pg->p_list);
 		queue_push_back(&z->z_free_pages[huge->p_order], &huge->p_list);