vm: implement private and shared address space mappings

whether a mapping is private or shared determines how the mapping is handled
when a task is duplicated.

CMakeLists.txt

@@ -1,4 +1,4 @@
-cmake_minimum_required(VERSION 4.0)
+cmake_minimum_required(VERSION 3.31)
 project(mango C ASM)
 
 if (NOT BUILD_TOOLS_DIR)

arch/x86_64/include/arch/paging.h

@@ -1,8 +1,8 @@
 #ifndef ARCH_PAGING_H_
 #define ARCH_PAGING_H_
 
-#include <kernel/types.h>
 #include <kernel/compiler.h>
+#include <kernel/types.h>
 
 #ifdef __cplusplus
 extern "C" {
@@ -55,6 +55,7 @@ enum page_size {
    defined in pmap_ctrl.S */
 extern int gigabyte_pages(void);
 extern int enable_nx(void);
+extern int enable_wp(void);
 
 #ifdef __cplusplus
 }

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

@@ -31,6 +31,14 @@ extern kern_status_t ml_thread_prepare_user_context(
 	virt_addr_t *kernel_sp,
 	const uintptr_t *args,
 	size_t nr_args);
+/* prepare the stack so that ml_thread_switch_user can jump to usermode
+ * with the specified register context */
+extern kern_status_t ml_thread_clone_user_context(
+	const struct ml_cpu_context *src_regs,
+	const struct ml_thread *src_ml,
+	struct ml_thread *dest_ml,
+	uintptr_t return_value,
+	virt_addr_t *kernel_sp);
 
 extern kern_status_t ml_thread_config_get(
 	struct thread *thread,

arch/x86_64/init.c

@@ -20,7 +20,7 @@
 #include <kernel/util.h>
 #include <kernel/vm.h>
 
-#undef HARDWARE_RNG
+#define HARDWARE_RNG
 
 #define PTR32(x)            ((void *)((uintptr_t)(x)))
 
@@ -56,9 +56,7 @@ static void early_vm_init(uintptr_t reserve_end)
 	uintptr_t alloc_end = VM_KERNEL_VOFFSET + 0x7fffffff;
 
 	memblock_init(alloc_start, alloc_end, VM_KERNEL_VOFFSET);
-	printk("memblock: allocating from [0x%llx-0x%llx]",
-	       alloc_start,
-	       alloc_end);
+	printk("memblock: allocating from [0x%llx-0x%llx]", alloc_start, alloc_end);
 
 	memblock_reserve(0x00, reserve_end);
 	printk("memblock: reserved bios+kernel at [0x%016llx-0x%016llx]",

arch/x86_64/irq.c

@@ -8,6 +8,7 @@
 #include <kernel/panic.h>
 #include <kernel/sched.h>
 #include <kernel/syscall.h>
+#include <kernel/thread.h>
 #include <stddef.h>
 
 #define MAX_ISR_HANDLERS  16
@@ -166,6 +167,12 @@ int idt_load(struct idt_ptr *ptr)
 
 void isr_dispatch(struct ml_cpu_context *regs)
 {
+	struct thread *thr = get_current_thread();
+	if (thr) {
+		thr->tr_irqctx = regs;
+		put_current_thread(thr);
+	}
+
 	int_hook h = isr_handlers[regs->int_no];
 	if (h) {
 		h(regs);
@@ -188,6 +195,13 @@ void irq_dispatch(struct ml_cpu_context *regs)
 	end_charge_period();
 
 	irq_ack(regs->int_no);
+
+	struct thread *thr = get_current_thread();
+	if (thr) {
+		thr->tr_irqctx = regs;
+		put_current_thread(thr);
+	}
+
 	struct queue *hooks = &irq_hooks[regs->int_no - IRQ0];
 	queue_foreach(struct irq_hook, hook, hooks, irq_entry)
 	{
@@ -203,6 +217,12 @@ void irq_dispatch(struct ml_cpu_context *regs)
 
 void syscall_dispatch(struct ml_cpu_context *regs)
 {
+	struct thread *thr = get_current_thread();
+	if (thr) {
+		thr->tr_irqctx = regs;
+		put_current_thread(thr);
+	}
+
 	unsigned int sysid = regs->rax;
 	virt_addr_t syscall_impl = syscall_get_function(sysid);
 

arch/x86_64/panic.c

@@ -217,16 +217,18 @@ static void print_stack_trace(
 
 void ml_print_stack_trace(uintptr_t ip)
 {
-	struct task *task = current_task();
+	struct task *task = get_current_task();
 	struct address_space *space = task ? task->t_address_space : NULL;
 	uintptr_t bp;
 	asm volatile("mov %%rbp, %0" : "=r"(bp));
 	print_stack_trace(space, ip, bp);
+	put_current_task(task);
 }
 
 void ml_print_stack_trace_irq(struct ml_cpu_context *ctx)
 {
-	struct task *task = current_task();
+	struct task *task = get_current_task();
 	struct address_space *space = task ? task->t_address_space : NULL;
 	print_stack_trace(space, ctx->rip, ctx->rbp);
+	put_current_task(task);
 }

arch/x86_64/pmap.c

@@ -19,7 +19,7 @@
 #define PTR_TO_ENTRY(x) (((x) & ~VM_PAGE_MASK) | PTE_PRESENT | PTE_RW | PTE_USR)
 #define ENTRY_TO_PTR(x) ((x) & ~VM_PAGE_MASK)
 
-#define PFN(x)          ((x) >> VM_PAGE_SHIFT)
+#define PFN(x)          (((x) >> VM_PAGE_SHIFT) & 0xFFFFFFFFFF)
 
 static int can_use_gbpages = 0;
 static pmap_t kernel_pmap;
@@ -136,6 +136,81 @@ static void delete_pdir(phys_addr_t pd)
 	kfree(pdir);
 }
 
+kern_status_t pmap_get(
+	pmap_t pmap,
+	virt_addr_t pv,
+	pfn_t *out_pfn,
+	vm_prot_t *out_prot)
+{
+	unsigned int pml4t_index = BAD_INDEX, pdpt_index = BAD_INDEX,
+		     pd_index = BAD_INDEX, pt_index = BAD_INDEX;
+
+	pml4t_index = (pv >> 39) & 0x1FF;
+	pdpt_index = (pv >> 30) & 0x1FF;
+	pd_index = (pv >> 21) & 0x1FF;
+	pt_index = (pv >> 12) & 0x1FF;
+
+	/* 1. get PML4T (mandatory) */
+	struct pml4t *pml4t = vm_phys_to_virt(ENTRY_TO_PTR(pmap));
+	if (!pml4t) {
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	/* 2. traverse PML4T, get PDPT (mandatory) */
+	struct pdpt *pdpt = NULL;
+	if (!pml4t->p_entries[pml4t_index]) {
+		return KERN_NO_ENTRY;
+	} else {
+		pdpt = vm_phys_to_virt(
+			ENTRY_TO_PTR(pml4t->p_entries[pml4t_index]));
+	}
+
+	/* 3. traverse PDPT, get PDIR (optional, 4K and 2M only) */
+	struct pdir *pdir = NULL;
+	if (!pdpt->p_entries[pdpt_index]
+	    || pdpt->p_pages[pdpt_index] & PTE_PAGESIZE) {
+		return KERN_NO_ENTRY;
+	} else {
+		pdir = vm_phys_to_virt(
+			ENTRY_TO_PTR(pdpt->p_entries[pdpt_index]));
+	}
+
+	/* 4. traverse PDIR, get PTAB (optional, 4K only) */
+	struct ptab *ptab = NULL;
+	if (!pdir->p_entries[pd_index]
+	    || pdir->p_pages[pd_index] & PTE_PAGESIZE) {
+		/* entry is null, or points to a hugepage */
+		return KERN_NO_ENTRY;
+	} else {
+		ptab = vm_phys_to_virt(ENTRY_TO_PTR(pdir->p_entries[pd_index]));
+	}
+
+	uint64_t pte = ptab->p_pages[pt_index];
+	if (out_pfn) {
+		*out_pfn = PFN(pte);
+	}
+
+	if (out_prot) {
+		if (pte & PTE_PRESENT) {
+			*out_prot |= VM_PROT_USER;
+		}
+
+		if (pte & PTE_RW) {
+			*out_prot |= (VM_PROT_READ | VM_PROT_WRITE);
+		}
+
+		if (pte & PTE_USR) {
+			*out_prot |= VM_PROT_USER;
+		}
+
+		if (!(pte & PTE_NX)) {
+			*out_prot |= VM_PROT_EXEC;
+		}
+	}
+
+	return KERN_OK;
+}
+
 static kern_status_t do_pmap_add(
 	pmap_t pmap,
 	virt_addr_t pv,
@@ -351,6 +426,8 @@ void pmap_bootstrap(void)
 	       can_use_gbpages == 1 ? "en" : "dis");
 	enable_nx();
 	printk("pmap: NX protection enabled");
+	enable_wp();
+	printk("pmap: kernel-mode write protection enabled");
 
 	enum page_size hugepage = PS_2M;
 	if (can_use_gbpages) {
@@ -494,11 +571,7 @@ kern_status_t pmap_handle_fault(
 {
 	// log_fault(fault_addr, flags);
 
-	if (flags & PMAP_FAULT_PRESENT) {
-		return KERN_FATAL_ERROR;
-	}
-
-	struct task *task = current_task();
+	struct task *task = get_current_task();
 	if (!task) {
 		return KERN_FATAL_ERROR;
 	}
@@ -509,7 +582,11 @@ kern_status_t pmap_handle_fault(
 	}
 
 	/* this must be called with `space` unlocked. */
-	return address_space_demand_map(space, fault_addr, flags);
+	kern_status_t status
+		= address_space_demand_map(space, fault_addr, flags);
+
+	put_current_task(task);
+	return status;
 }
 
 kern_status_t pmap_add(

arch/x86_64/pmap_ctrl.S

@@ -30,6 +30,7 @@ gigabyte_pages:
 	pop %rbp
 	ret
 
+
 	.global enable_nx
 	.type enable_nx, @function
 
@@ -40,3 +41,14 @@ enable_nx:
 	wrmsr
 
 	ret
+
+
+	.global enable_wp
+	.type enable_wp, @function
+
+enable_wp:
+	mov %cr0, %rax
+	or $0x10000, %rax
+	mov %rax, %cr0
+
+	ret

arch/x86_64/serial.c

@@ -28,6 +28,10 @@ void serial_send_byte(int device, char out)
 
 	outportb(device, out);
 
+	if (device == COM1) {
+		outportb(0xe9, out);
+	}
+
 	while (!transmit_empty(device)) {
 		_count++;
 	}
@@ -138,6 +142,7 @@ static struct console serialcon = {
 
 static int serial_irq1(void)
 {
+#if 0
 	if (serial_received(COM1)) {
 		unsigned char c = serial_recv_byte(COM1);
 		printk("serial: COM1 received %c", c);
@@ -147,6 +152,7 @@ static int serial_irq1(void)
 		unsigned char c = serial_recv_byte(COM3);
 		printk("serial: COM3 received %c", c);
 	}
+#endif
 
 	return 0;
 }

arch/x86_64/thread.c

@@ -80,6 +80,26 @@ extern kern_status_t ml_thread_prepare_user_context(
 	return KERN_OK;
 }
 
+kern_status_t ml_thread_clone_user_context(
+	const struct ml_cpu_context *src_regs,
+	const struct ml_thread *src_ml,
+	struct ml_thread *dest_ml,
+	uintptr_t return_value,
+	virt_addr_t *kernel_sp)
+{
+	(*kernel_sp) -= sizeof(struct ml_cpu_context);
+
+	struct ml_cpu_context *regs = (struct ml_cpu_context *)(*kernel_sp);
+	memcpy(regs, src_regs, sizeof *regs);
+
+	regs->rax = return_value;
+
+	dest_ml->tr_fsbase = src_ml->tr_fsbase;
+	dest_ml->tr_gsbase = src_ml->tr_gsbase;
+
+	return KERN_OK;
+}
+
 kern_status_t ml_thread_config_get(
 	struct thread *thread,
 	kern_config_key_t key,
@@ -95,25 +115,30 @@ kern_status_t ml_thread_config_set(
 	const void *ptr,
 	size_t len)
 {
+	struct thread *self = get_current_thread();
+	kern_status_t status = KERN_OK;
+
 	switch (key) {
 	case THREAD_CFG_FSBASE:
 		if (len != sizeof(thread->tr_ml.tr_fsbase)) {
-			return KERN_INVALID_ARGUMENT;
+			status = KERN_INVALID_ARGUMENT;
+			break;
 		}
 
 		thread->tr_ml.tr_fsbase = *(virt_addr_t *)ptr;
-		if (thread == current_thread()) {
+		if (thread == self) {
 			wrmsr(MSR_FS_BASE, thread->tr_ml.tr_fsbase);
 		}
 
 		break;
 	case THREAD_CFG_GSBASE:
 		if (len != sizeof(thread->tr_ml.tr_gsbase)) {
-			return KERN_INVALID_ARGUMENT;
+			status = KERN_INVALID_ARGUMENT;
+			break;
 		}
 
 		thread->tr_ml.tr_gsbase = *(virt_addr_t *)ptr;
-		if (thread == current_thread()) {
+		if (thread == self) {
 			/* we're in the kernel right now, so the user and kernel
 			 * gs-base registers are swapped. when we return to
 			 * usermode, this value will be swapped back into
@@ -123,8 +148,10 @@ kern_status_t ml_thread_config_set(
 
 		break;
 	default:
-		return KERN_INVALID_ARGUMENT;
+		status = KERN_INVALID_ARGUMENT;
+		break;
 	}
 
-	return KERN_OK;
+	put_current_thread(self);
+	return status;
 }

ds/ringbuffer.c

@@ -1,189 +0,0 @@
-#include <kernel/ringbuffer.h>
-#include <kernel/sched.h>
-#include <kernel/vm.h>
-
-size_t ringbuffer_unread(struct ringbuffer *ring_buffer)
-{
-	if (ring_buffer->r_read_ptr == ring_buffer->r_write_ptr) {
-		return 0;
-	}
-
-	if (ring_buffer->r_read_ptr > ring_buffer->r_write_ptr) {
-		return (ring_buffer->r_size - ring_buffer->r_read_ptr)
-		     + ring_buffer->r_write_ptr;
-	} else {
-		return (ring_buffer->r_write_ptr - ring_buffer->r_read_ptr);
-	}
-}
-
-size_t ringbuffer_avail(struct ringbuffer *ring_buffer)
-{
-	if (ring_buffer->r_read_ptr == ring_buffer->r_write_ptr) {
-		return ring_buffer->r_size - 1;
-	}
-
-	if (ring_buffer->r_read_ptr > ring_buffer->r_write_ptr) {
-		return ring_buffer->r_read_ptr - ring_buffer->r_write_ptr - 1;
-	} else {
-		return (ring_buffer->r_size - ring_buffer->r_write_ptr)
-		     + ring_buffer->r_read_ptr - 1;
-	}
-}
-
-static inline void increment_read(struct ringbuffer *ring_buffer)
-{
-	ring_buffer->r_read_ptr++;
-	if (ring_buffer->r_read_ptr == ring_buffer->r_size) {
-		ring_buffer->r_read_ptr = 0;
-	}
-}
-
-static inline void increment_write(struct ringbuffer *ring_buffer)
-{
-	ring_buffer->r_write_ptr++;
-	if (ring_buffer->r_write_ptr == ring_buffer->r_size) {
-		ring_buffer->r_write_ptr = 0;
-	}
-}
-
-size_t ringbuffer_read(
-	struct ringbuffer *ring_buffer,
-	size_t size,
-	void *p,
-	mango_flags_t flags)
-{
-	if (!ring_buffer) {
-		return 0;
-	}
-
-	unsigned char *buffer = p;
-	unsigned long lock_flags;
-	size_t collected = 0;
-
-	while (collected < size) {
-		spin_lock_irqsave(&ring_buffer->r_lock, &lock_flags);
-		while (ringbuffer_unread(ring_buffer) > 0 && collected < size) {
-			buffer[collected]
-				= ring_buffer
-			                  ->r_buffer[ring_buffer->r_read_ptr];
-			increment_read(ring_buffer);
-			collected++;
-		}
-
-		wakeup_queue(&ring_buffer->r_wait_writers);
-
-		if (flags & S_NOBLOCK) {
-			spin_unlock_irqrestore(
-				&ring_buffer->r_lock,
-				lock_flags);
-			break;
-		}
-
-		struct wait_item waiter;
-		wait_item_init(&waiter, current_thread());
-		thread_wait_begin(&waiter, &ring_buffer->r_wait_readers);
-		spin_unlock_irqrestore(&ring_buffer->r_lock, lock_flags);
-
-		if (collected < size) {
-			schedule(SCHED_NORMAL);
-		}
-
-		thread_wait_end(&waiter, &ring_buffer->r_wait_readers);
-	}
-
-	wakeup_queue(&ring_buffer->r_wait_writers);
-	return collected;
-}
-
-size_t ringbuffer_write(
-	struct ringbuffer *ring_buffer,
-	size_t size,
-	const void *p,
-	mango_flags_t flags)
-{
-	if (!ring_buffer || !size) {
-		return 0;
-	}
-
-	const unsigned char *buffer = p;
-	unsigned long lock_flags;
-	size_t written = 0;
-
-	while (written < size) {
-		spin_lock_irqsave(&ring_buffer->r_lock, &lock_flags);
-
-		while (ringbuffer_avail(ring_buffer) > 0 && written < size) {
-			ring_buffer->r_buffer[ring_buffer->r_write_ptr]
-				= buffer[written];
-			increment_write(ring_buffer);
-			written++;
-		}
-
-		wakeup_queue(&ring_buffer->r_wait_readers);
-
-		if (flags & S_NOBLOCK) {
-			spin_unlock_irqrestore(
-				&ring_buffer->r_lock,
-				lock_flags);
-			break;
-		}
-
-		struct wait_item waiter;
-		wait_item_init(&waiter, current_thread());
-		thread_wait_begin(&waiter, &ring_buffer->r_wait_writers);
-		spin_unlock_irqrestore(&ring_buffer->r_lock, lock_flags);
-
-		if (written < size) {
-			schedule(SCHED_NORMAL);
-		}
-
-		thread_wait_end(&waiter, &ring_buffer->r_wait_writers);
-	}
-
-	wakeup_queue(&ring_buffer->r_wait_readers);
-	return written;
-}
-
-struct ringbuffer *ringbuffer_create(size_t size)
-{
-	struct ringbuffer *out = kzalloc(sizeof(struct ringbuffer), VM_NORMAL);
-	if (!out) {
-		return NULL;
-	}
-
-	if (ringbuffer_init(out, size) != KERN_OK) {
-		kfree(out);
-		return NULL;
-	}
-
-	return out;
-}
-
-void ringbuffer_destroy(struct ringbuffer *ring_buffer)
-{
-	ringbuffer_deinit(ring_buffer);
-	kfree(ring_buffer);
-}
-
-kern_status_t ringbuffer_init(struct ringbuffer *buf, size_t size)
-{
-	buf->r_buffer = kmalloc(size, VM_NORMAL);
-	if (!buf->r_buffer) {
-		return KERN_NO_MEMORY;
-	}
-
-	buf->r_write_ptr = 0;
-	buf->r_read_ptr = 0;
-	buf->r_size = size;
-	buf->r_lock = SPIN_LOCK_INIT;
-
-	return KERN_OK;
-}
-
-kern_status_t ringbuffer_deinit(struct ringbuffer *buf)
-{
-	kfree(buf->r_buffer);
-	buf->r_buffer = NULL;
-
-	return KERN_OK;
-}

include/kernel/address-space.h

@@ -22,6 +22,8 @@ struct vm_area {
 	struct address_space *vma_space;
 	/* used to link to vm_object->vo_mappings */
 	struct queue_entry vma_object_entry;
+	/* the memory control flags applied to this area */
+	vm_flags_t vma_flags;
 	/* the memory protection flags applied to this area */
 	vm_prot_t vma_prot;
 	/* offset in bytes to the start of the object data that was mapped */
@@ -83,6 +85,7 @@ extern kern_status_t address_space_map(
 	struct vm_object *object,
 	off_t object_offset,
 	size_t length,
+	vm_flags_t flags,
 	vm_prot_t prot,
 	virt_addr_t *out);
 extern kern_status_t address_space_unmap(
@@ -105,6 +108,12 @@ extern kern_status_t address_space_release(
 	virt_addr_t base,
 	size_t length);
 
+/* duplicate all of the mappings in `src` within `dest. the duplication will use
+ * copy-on-write; page data will not be copied until it is written to. */
+extern kern_status_t address_space_duplicate(
+	struct address_space *dest,
+	struct address_space *src);
+
 extern bool address_space_validate_access(
 	struct address_space *region,
 	virt_addr_t base,

include/kernel/atomic.h

@@ -4,7 +4,7 @@
 #include <stdbool.h>
 #include <stdint.h>
 
-typedef int64_t atomic_t;
+typedef int32_t atomic_t;
 
 /* load and return the value pointed to by `v` */
 static inline atomic_t atomic_load(atomic_t *v)

include/kernel/fb.h

@@ -1,49 +0,0 @@
-#ifndef KERNEL_FB_H_
-#define KERNEL_FB_H_
-
-#include <stdint.h>
-
-enum framebuffer_flags {
-	FB_MODE_RGB = 0x01u,
-	FB_MODE_VGATEXT = 0x02u,
-	FB_MODE_PALETTE = 0x04u,
-};
-
-struct framebuffer_bitfield {
-	uint32_t b_offset;
-	uint16_t b_length;
-};
-
-struct framebuffer_varinfo {
-	enum framebuffer_flags fb_flags;
-
-	uint32_t fb_xres;
-	uint32_t fb_yres;
-	uint32_t fb_bpp;
-	uint32_t fb_stride;
-
-	union {
-		struct {
-			uint32_t fb_xcells;
-			uint32_t fb_ycells;
-		};
-
-		struct {
-			struct framebuffer_bitfield fb_red;
-			struct framebuffer_bitfield fb_green;
-			struct framebuffer_bitfield fb_blue;
-			struct framebuffer_bitfield fb_alpha;
-		};
-
-		struct {
-			uintptr_t fb_palette_addr;
-			uint16_t fb_palette_nr_colours;
-		};
-	};
-};
-
-struct framebuffer_fixedinfo {
-	uint64_t fb_baseptr;
-};
-
-#endif

include/kernel/flags.h

@@ -1,11 +0,0 @@
-#ifndef KERNEL_FLAGS_H_
-#define KERNEL_FLAGS_H_
-
-#include <stdint.h>
-
-typedef enum {
-	S_NORMAL = 0x00u,
-	S_NOBLOCK = 0x01u,
-} mango_flags_t;
-
-#endif

include/kernel/handle.h

@@ -44,9 +44,13 @@ struct handle_table {
 
 extern struct handle_table *handle_table_create(void);
 extern void handle_table_destroy(struct handle_table *tab);
+extern kern_status_t handle_table_duplicate(
+	struct handle_table *src,
+	struct handle_table **dest);
 
 extern kern_status_t handle_table_alloc_handle(
 	struct handle_table *tab,
+	kern_handle_t value,
 	struct handle **out_slot,
 	kern_handle_t *out_handle);
 extern kern_status_t handle_table_free_handle(

include/kernel/input.h

@@ -1,184 +0,0 @@
-#ifndef KERNEL_INPUT_H_
-#define KERNEL_INPUT_H_
-
-#include <stdint.h>
-#include <kernel/queue.h>
-#include <mango/status.h>
-
-enum input_event_hook_flags {
-	INPUT_HOOK_SQUASH_EVENT = 0x01u,
-};
-
-struct device;
-
-enum input_event_type {
-	INPUT_TYPE_UNKNOWN = 0x00u,
-	INPUT_TYPE_KEY     = 0x01u,
-	INPUT_TYPE_MOTION  = 0x02u,
-};
-
-enum input_event_motion_type {
-	INPUT_MOTION_TYPE_MOUSE  = 0x01u,
-	INPUT_MOTION_TYPE_SCROLL = 0x02u,
-};
-
-enum input_keycode {
-	KEY_UNKNOWN                = 0x00u,
-	KEY_A                      = 0x01u,
-	KEY_B                      = 0x02u,
-	KEY_C                      = 0x03u,
-	KEY_D                      = 0x04u,
-	KEY_E                      = 0x05u,
-	KEY_F                      = 0x06u,
-	KEY_G                      = 0x07u,
-	KEY_H                      = 0x08u,
-	KEY_I                      = 0x09u,
-	KEY_J                      = 0x0Au,
-	KEY_K                      = 0x0Bu,
-	KEY_L                      = 0x0Cu,
-	KEY_M                      = 0x0Du,
-	KEY_N                      = 0x0Eu,
-	KEY_O                      = 0x0Fu,
-	KEY_P                      = 0x10u,
-	KEY_Q                      = 0x11u,
-	KEY_R                      = 0x12u,
-	KEY_S                      = 0x13u,
-	KEY_T                      = 0x14u,
-	KEY_U                      = 0x15u,
-	KEY_V                      = 0x16u,
-	KEY_W                      = 0x17u,
-	KEY_X                      = 0x18u,
-	KEY_Y                      = 0x19u,
-	KEY_Z                      = 0x1Au,
-	KEY_KEY_1                  = 0x1Bu,
-	KEY_KEY_2                  = 0x1Cu,
-	KEY_KEY_3                  = 0x1Du,
-	KEY_KEY_4                  = 0x1Eu,
-	KEY_KEY_5                  = 0x1Fu,
-	KEY_KEY_6                  = 0x20u,
-	KEY_KEY_7                  = 0x21u,
-	KEY_KEY_8                  = 0x22u,
-	KEY_KEY_9                  = 0x23u,
-	KEY_KEY_0                  = 0x24u,
-	KEY_ENTER                  = 0x25u,
-	KEY_ESCAPE                 = 0x26u,
-	KEY_BACKSPACE              = 0x27u,
-	KEY_TAB                    = 0x28u,
-	KEY_SPACE                  = 0x29u,
-	KEY_MINUS                  = 0x2Au,
-	KEY_EQUALS                 = 0x2Bu,
-	KEY_LEFT_BRACE             = 0x2Cu,
-	KEY_RIGHT_BRACE            = 0x2Du,
-	KEY_BACKSLASH              = 0x2Eu,
-	KEY_NON_US_HASH            = 0x2Fu,
-	KEY_SEMICOLON              = 0x30u,
-	KEY_APOSTROPHE             = 0x31u,
-	KEY_GRAVE_ACCENT           = 0x32u,
-	KEY_COMMA                  = 0x33u,
-	KEY_DOT                    = 0x34u,
-	KEY_SLASH                  = 0x35u,
-	KEY_CAPS_LOCK              = 0x36u,
-	KEY_F1                     = 0x37u,
-	KEY_F2                     = 0x38u,
-	KEY_F3                     = 0x39u,
-	KEY_F4                     = 0x3Au,
-	KEY_F5                     = 0x3Bu,
-	KEY_F6                     = 0x3Cu,
-	KEY_F7                     = 0x3Du,
-	KEY_F8                     = 0x3Eu,
-	KEY_F9                     = 0x3Fu,
-	KEY_F10                    = 0x40u,
-	KEY_F11                    = 0x41u,
-	KEY_F12                    = 0x42u,
-	KEY_PRINT_SCREEN           = 0x43u,
-	KEY_SCROLL_LOCK            = 0x44u,
-	KEY_PAUSE                  = 0x45u,
-	KEY_INSERT                 = 0x46u,
-	KEY_HOME                   = 0x47u,
-	KEY_PAGE_UP                = 0x48u,
-	KEY_DELETE                 = 0x49u,
-	KEY_END                    = 0x4Au,
-	KEY_PAGE_DOWN              = 0x4Bu,
-	KEY_RIGHT                  = 0x4Cu,
-	KEY_LEFT                   = 0x4Du,
-	KEY_DOWN                   = 0x4Eu,
-	KEY_UP                     = 0x4Fu,
-	KEY_NUM_LOCK               = 0x50u,
-	KEY_KEYPAD_SLASH           = 0x51u,
-	KEY_KEYPAD_ASTERISK        = 0x52u,
-	KEY_KEYPAD_MINUS           = 0x53u,
-	KEY_KEYPAD_PLUS            = 0x54u,
-	KEY_KEYPAD_ENTER           = 0x55u,
-	KEY_KEYPAD_1               = 0x56u,
-	KEY_KEYPAD_2               = 0x57u,
-	KEY_KEYPAD_3               = 0x58u,
-	KEY_KEYPAD_4               = 0x59u,
-	KEY_KEYPAD_5               = 0x5Au,
-	KEY_KEYPAD_6               = 0x5Bu,
-	KEY_KEYPAD_7               = 0x5Cu,
-	KEY_KEYPAD_8               = 0x5Du,
-	KEY_KEYPAD_9               = 0x5Eu,
-	KEY_KEYPAD_0               = 0x5Fu,
-	KEY_KEYPAD_DOT             = 0x60u,
-	KEY_NON_US_BACKSLASH       = 0x61u,
-	KEY_KEYPAD_EQUALS          = 0x62u,
-	KEY_MENU                   = 0x63u,
-	KEY_LEFT_CTRL              = 0x64u,
-	KEY_LEFT_SHIFT             = 0x65u,
-	KEY_LEFT_ALT               = 0x66u,
-	KEY_LEFT_META              = 0x67u,
-	KEY_RIGHT_CTRL             = 0x68u,
-	KEY_RIGHT_SHIFT            = 0x69u,
-	KEY_RIGHT_ALT              = 0x6Au,
-	KEY_RIGHT_META             = 0x6Bu,
-	KEY_MEDIA_MUTE             = 0x6Cu,
-	KEY_MEDIA_VOLUME_INCREMENT = 0x6Du,
-	KEY_MEDIA_VOLUME_DECREMENT = 0x6Eu,
-};
-
-enum input_key_state {
-	INPUT_KEYSTATE_DOWN = 0x00u,
-	INPUT_KEYSTATE_UP   = 0x01u,
-};
-
-enum input_button {
-	INPUT_BUTTON_MOUSE_LEFT    = 0x00u,
-	INPUT_BUTTON_MOUSE_MIDDLE  = 0x01u,
-	INPUT_BUTTON_MOUSE_RIGHT   = 0x02u,
-	INPUT_BUTTON_MOUSE_BACK    = 0x03u,
-	INPUT_BUTTON_MOUSE_FORWARD = 0x04u,
-};
-
-enum input_button_state {
-	INPUT_BUTTON_DOWN = 0x00u,
-	INPUT_BUTTON_UP   = 0x01u,
-};
-
-struct input_event {
-	enum input_event_type ev_type;
-	union {
-		struct {
-			enum input_event_motion_type type;
-			int16_t movement_x;
-			int16_t movement_y;
-		} ev_motion;
-
-		struct {
-			enum input_button button;
-			enum input_button_state state;
-		} ev_button;
-
-		struct {
-			enum input_keycode key;
-			enum input_key_state state;
-		} ev_key;
-	};
-};
-
-struct input_event_hook {
-	void(*hook_callback)(struct device *, struct input_event *, enum input_event_hook_flags *, void *);
-	void *hook_arg;
-	struct queue_entry hook_head;
-};
-
-#endif

include/kernel/msg.h

@@ -21,8 +21,20 @@ struct msg {
 	enum kmsg_status msg_status;
 	struct btree_node msg_node;
 	msgid_t msg_id;
+	union {
+		/* only valid for asynchronous messages (msg_status ==
+		 * KMSG_ASYNC) */
+		struct {
+			koid_t msg_sender_port_id;
+			tid_t msg_sender_thread_id;
+		};
+		/* only valid for synchronous messages (msg_status !=
+		 * KMSG_ASYNC) */
+		struct {
 			struct port *msg_sender_port;
 			struct thread *msg_sender_thread;
+		};
+	};
 	kern_status_t msg_result;
 	kern_msg_type_t msg_type;
 

include/kernel/object.h

@@ -2,7 +2,6 @@
 #define KERNEL_OBJECT_H_
 
 #include <kernel/atomic.h>
-#include <kernel/flags.h>
 #include <kernel/locks.h>
 #include <kernel/vm.h>
 #include <kernel/wait.h>
@@ -66,7 +65,7 @@ extern "C" {
 #define OBJECT_PATH_MAX 256
 
 #define OBJECT_CAST(to_type, to_type_member, p)                                \
-	((to_type *)((uintptr_t)p) - offsetof(to_type, to_type_member))
+	((to_type *)(((uintptr_t)p) - offsetof(to_type, to_type_member)))
 #define OBJECT_C_CAST(c_type, c_type_member, obj_type, objp)                   \
 	OBJECT_IS_TYPE(objp, obj_type)                                         \
 	? OBJECT_CAST(c_type, c_type_member, (objp)) : NULL

include/kernel/pmap.h

@@ -9,7 +9,6 @@
 #include <stddef.h>
 
 #define PMAP_INVALID ML_PMAP_INVALID
-#define PFN(x)       ((x) >> VM_PAGE_SHIFT)
 
 #ifdef __cplusplus
 extern "C" {
@@ -57,6 +56,11 @@ extern kern_status_t pmap_handle_fault(
 	virt_addr_t fault_addr,
 	enum pmap_fault_flags flags);
 
+extern kern_status_t pmap_get(
+	pmap_t pmap,
+	virt_addr_t p,
+	pfn_t *out_pfn,
+	vm_prot_t *out_prot);
 extern kern_status_t pmap_add(
 	pmap_t pmap,
 	virt_addr_t p,

include/kernel/port.h

@@ -9,12 +9,6 @@ enum port_status {
 	PORT_OFFLINE = 0,
 	/* port is connected and ready to send messages */
 	PORT_READY,
-	/* port has sent a message, and is waiting for the remote to receive it
-	 */
-	PORT_SEND_BLOCKED,
-	/* port has sent a message, and the remote has received it. waiting for
-	 * the remote to reply. */
-	PORT_REPLY_BLOCKED,
 };
 
 struct port {

include/kernel/ringbuffer.h

@@ -1,45 +0,0 @@
-#ifndef KERNEL_RINGBUFFER_H_
-#define KERNEL_RINGBUFFER_H_
-
-#include <kernel/flags.h>
-#include <kernel/locks.h>
-#include <kernel/types.h>
-#include <kernel/wait.h>
-
-struct ringbuffer {
-	unsigned char *r_buffer;
-	size_t r_write_ptr;
-	size_t r_read_ptr;
-	size_t r_size;
-	spin_lock_t r_lock;
-	struct waitqueue r_wait_readers;
-	struct waitqueue r_wait_writers;
-};
-
-extern struct ringbuffer *ringbuffer_create(size_t size);
-extern void ringbuffer_destroy(struct ringbuffer *buf);
-
-extern kern_status_t ringbuffer_init(struct ringbuffer *buf, size_t size);
-extern kern_status_t ringbuffer_deinit(struct ringbuffer *buf);
-
-extern size_t ringbuffer_unread(struct ringbuffer *buf);
-extern size_t ringbuffer_avail(struct ringbuffer *buf);
-extern size_t ringbuffer_read(
-	struct ringbuffer *buf,
-	size_t size,
-	void *buffer,
-	mango_flags_t flags);
-extern size_t ringbuffer_write(
-	struct ringbuffer *buf,
-	size_t size,
-	const void *buffer,
-	mango_flags_t flags);
-
-/* TODO */
-// extern size_t ringbuffer_peek(struct ringbuffer *buf, size_t at, size_t size,
-// void *buffer); extern size_t ringbuffer_skip(struct ringbuffer *buf, size_t
-// count);
-
-extern int ringbuffer_write_would_block(struct ringbuffer *buf);
-
-#endif

include/kernel/sched.h

@@ -76,8 +76,10 @@ extern struct runqueue *cpu_rq(unsigned int cpu);
 extern cycles_t default_quantum(void);
 
 extern bool need_resched(void);
-extern struct task *current_task(void);
-extern struct thread *current_thread(void);
+extern struct task *get_current_task(void);
+extern struct thread *get_current_thread(void);
+extern void put_current_task(struct task *task);
+extern void put_current_thread(struct thread *thread);
 
 extern struct runqueue *select_rq_for_thread(struct thread *thr);
 extern void schedule_thread_on_cpu(struct thread *thr);

include/kernel/syscall.h

@@ -66,6 +66,9 @@ extern kern_status_t sys_task_config_set(
 	kern_config_key_t key,
 	const void *ptr,
 	size_t len);
+extern kern_status_t sys_task_duplicate(
+	kern_handle_t *out_task,
+	kern_handle_t *out_address_space);
 
 extern kern_status_t sys_thread_self(kern_handle_t *out);
 extern kern_status_t sys_thread_start(kern_handle_t thread);
@@ -125,6 +128,7 @@ extern kern_status_t sys_address_space_map(
 	kern_handle_t object,
 	off_t object_offset,
 	size_t length,
+	vm_flags_t flags,
 	vm_prot_t prot,
 	virt_addr_t *out_base_address);
 extern kern_status_t sys_address_space_unmap(
@@ -143,9 +147,19 @@ extern kern_status_t sys_address_space_release(
 
 extern kern_status_t sys_kern_log(const char *s);
 extern kern_status_t sys_kern_handle_close(kern_handle_t handle);
-extern kern_status_t sys_kern_handle_duplicate(
+extern kern_status_t sys_kern_handle_transfer(
+	kern_handle_t src_task_handle,
+	kern_handle_t src_handle,
+	kern_handle_t dest_task_handle,
+	kern_handle_t dest_handle,
+	unsigned int mode,
+	kern_handle_t *out_handle);
+extern kern_status_t sys_kern_handle_control(
+	kern_handle_t task,
 	kern_handle_t handle,
-	kern_handle_t *out);
+	uint32_t set_mask,
+	uint32_t clear_mask,
+	uint32_t *out_flags);
 extern kern_status_t sys_kern_config_get(
 	kern_config_key_t key,
 	void *ptr,
@@ -191,11 +205,14 @@ extern kern_status_t sys_msg_write(
 extern kern_status_t sys_kern_object_wait(
 	kern_wait_item_t *items,
 	size_t nr_items);
+extern kern_status_t sys_kern_object_query(
+	kern_handle_t object,
+	kern_object_info_t *out);
 
 extern kern_status_t sys_vm_controller_create(kern_handle_t *out);
 extern kern_status_t sys_vm_controller_recv(
 	kern_handle_t ctrl,
-	equeue_packet_page_request_t *out);
+	equeue_packet_vm_request_t *out);
 extern kern_status_t sys_vm_controller_recv_async(
 	kern_handle_t ctrl,
 	kern_handle_t eq,
@@ -208,6 +225,14 @@ extern kern_status_t sys_vm_controller_create_object(
 	size_t data_len,
 	vm_prot_t prot,
 	kern_handle_t *out);
+extern kern_status_t sys_vm_controller_prepare_attach(
+	kern_handle_t ctrl,
+	uint64_t req_id,
+	kern_handle_t *out_vmo);
+extern kern_status_t sys_vm_controller_finish_attach(
+	kern_handle_t ctrl,
+	uint64_t req_id,
+	equeue_key_t new_key);
 extern kern_status_t sys_vm_controller_detach_object(
 	kern_handle_t ctrl,
 	kern_handle_t vmo);

include/kernel/task.h

@@ -39,7 +39,10 @@ struct task {
 
 extern struct task *task_alloc(void);
 extern struct task *task_cast(struct object *obj);
-extern struct task *task_create(const char *name, size_t name_len);
+extern struct task *task_create(
+	const char *name,
+	size_t name_len,
+	struct handle_table *handles);
 static inline struct task *task_ref(struct task *task)
 {
 	return OBJECT_CAST(struct task, t_base, object_ref(&task->t_base));

include/kernel/test.h

@@ -1,14 +0,0 @@
-#ifndef KERNEL_TEST_H_
-#define KERNEL_TEST_H_
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-extern int run_all_tests(void);
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif

include/kernel/thread.h

@@ -8,6 +8,8 @@
 
 #define THREAD_KSTACK_ORDER VM_PAGE_4K
 
+struct ml_cpu_context;
+
 enum thread_state {
 	THREAD_READY = 1,
 	THREAD_SLEEPING = 2,
@@ -39,6 +41,9 @@ struct thread {
 	virt_addr_t tr_ip, tr_sp, tr_bp;
 	virt_addr_t tr_cpu_user_sp, tr_cpu_kernel_sp;
 
+	/* only valid within an interrupt or syscall context */
+	struct ml_cpu_context *tr_irqctx;
+
 	struct ml_thread tr_ml;
 	struct runqueue *tr_rq;
 
@@ -57,6 +62,10 @@ extern kern_status_t thread_init_user(
 	virt_addr_t sp,
 	const uintptr_t *args,
 	size_t nr_args);
+extern kern_status_t thread_init_user_clone(
+	struct thread *thr,
+	const struct thread *src,
+	uintptr_t return_value);
 extern int thread_priority(struct thread *thr);
 extern void thread_awaken(struct thread *thr);
 extern void thread_exit(void);

include/kernel/vm-controller.h

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

include/kernel/vm-object.h

@@ -20,19 +20,25 @@ enum vm_object_flags {
 	 * be detached, allowing the server to close the last handle to the
 	 * object and dispose of it. */
 	VMO_AUTO_DETACH = 0x04u,
+	/* this vmo is a duplicate of a vmo that is attached to a vm-controller.
+	 * the duplicate vmo is scheduled to be attached to the same controller,
+	 * but this won't actually happen until the controller is needed to
+	 * fulfill a page request. once the duplicate vmo has been attached to
+	 * the controller, this flag will be cleared. */
+	VMO_LAZY_ATTACH = 0x08u,
 
 	/* these flags are for use with vm_object_get_page */
 	/**************************************************/
 
 	/* if the relevant page hasn't been allocated yet, it will be allocated
 	 * and returned. if this flag isn't specified, NULL will be returned. */
-	VMO_ALLOCATE_MISSING_PAGE = 0x08u,
+	VMO_ALLOCATE_MISSING_PAGE = 0x0100u,
 	/* if the vm-object is attached to a vm-controller, and the relevant
 	 * page is uncommitted, send a request to the vm-controller to provide
 	 * the missing page. will result in the vm-object being unlocked and
 	 * the current thread sleeping until the request is fulfilled. the
 	 * vm-object will be re-locked before the function returns. */
-	VMO_REQUEST_MISSING_PAGE = 0x10u,
+	VMO_REQUEST_MISSING_PAGE = 0x0200u,
 };
 
 struct vm_object {
@@ -84,11 +90,23 @@ extern struct vm_object *vm_object_create_in_place(
 	size_t data_len,
 	vm_prot_t prot);
 
+/* create a copy-on-write duplicate of a vm-object */
+extern struct vm_object *vm_object_duplicate_cow(struct vm_object *vmo);
+/* attach a copy-on-write duplicate of a vm-object to the vm-controller that
+ * controlled the original vm-object */
+extern kern_status_t vm_object_attach_cow(
+	struct vm_object *vmo,
+	unsigned long *irq_flags);
+
 extern struct vm_page *vm_object_get_page(
 	struct vm_object *vo,
 	off_t offset,
 	enum vm_object_flags flags,
 	unsigned long *irq_flags);
+extern kern_status_t vm_object_put_page(
+	struct vm_object *vo,
+	off_t offset,
+	struct vm_page *pg);
 
 extern kern_status_t vm_object_read(
 	struct vm_object *vo,

include/kernel/vm.h

@@ -1,6 +1,7 @@
 #ifndef KERNEL_VM_H_
 #define KERNEL_VM_H_
 
+#include <kernel/atomic.h>
 #include <kernel/bitmap.h>
 #include <kernel/btree.h>
 #include <kernel/locks.h>
@@ -14,8 +15,6 @@
 extern "C" {
 #endif
 
-struct bcache;
-
 /* maximum number of NUMA nodes */
 #define VM_MAX_NODES            64
 /* maximum number of memory zones per node */
@@ -25,11 +24,6 @@ struct bcache;
 /* maximum number of sparse memory sectors */
 #define VM_MAX_SECTORS          8192
 
-/* maximum number of disk sectors that can be stored in a single
-   page. AKA the number of bits in the sector bitmap.
-   used by the block cache */
-#define VM_MAX_SECTORS_PER_PAGE 32
-
 #define VM_CHECK_ALIGN(p, mask) ((((p) & (mask)) == (p)) ? 1 : 0)
 
 #define VM_CACHE_INITIALISED(c) ((c)->c_obj_count != 0)
@@ -207,7 +201,6 @@ struct vm_page {
 	/* owner-specific pointer */
 	union {
 		struct vm_slab *p_slab;
-		struct bcache *p_bcache;
 		void *p_priv0;
 	};
 
@@ -218,8 +211,6 @@ struct vm_page {
 	       lists.
 	     - vm_object uses it to maintain a btree of allocated pages keyed
 	       by offset/size.
-	     - the block cache uses this to maintain a tree of pages keyed by
-	       block number.
 	*/
 	union {
 		struct queue_entry p_list;
@@ -231,15 +222,13 @@ struct vm_page {
 	};
 
 	union {
-		/* used by bcache when sector size is < page size. bitmap of
-		 * present/missing sectors */
-		DECLARE_BITMAP(p_blockbits, VM_MAX_SECTORS_PER_PAGE);
+		/* how many vm_areas reference this vm_page. if >0, the page is
+		 * subject to copy-on-write. */
+		atomic_t p_cow_ref;
 		uint32_t p_priv2;
 	};
 
 	union {
-		/* sector address, used by bcache */
-		sectors_t p_blockid;
 		/* offset of this page within the vm_object it is a part of */
 		off_t p_vmo_offset;
 

init/main.c

@@ -5,7 +5,6 @@
 #include <kernel/cpu.h>
 #include <kernel/handle.h>
 #include <kernel/init.h>
-#include <kernel/input.h>
 #include <kernel/libc/stdio.h>
 #include <kernel/machine/init.h>
 #include <kernel/object.h>
@@ -14,7 +13,6 @@
 #include <kernel/printk.h>
 #include <kernel/sched.h>
 #include <kernel/task.h>
-#include <kernel/test.h>
 #include <kernel/thread.h>
 #include <kernel/vm-object.h>
 #include <stdint.h>
@@ -46,8 +44,8 @@ static void hang(void)
 
 void background_thread(void)
 {
-	struct task *self = current_task();
-	struct thread *thread = current_thread();
+	struct task *self = get_current_task();
+	struct thread *thread = get_current_thread();
 	printk("background_thread() running on processor %u", this_cpu());
 	milli_sleep(1000);
 
@@ -110,7 +108,7 @@ void kernel_init(uintptr_t arg)
 	       bsp.bsp_trailer.bsp_exec_entry,
 	       bsp.bsp_vmo);
 
-	struct task *bootstrap_task = task_create("bootstrap", 9);
+	struct task *bootstrap_task = task_create("bootstrap", 9, NULL);
 	tracek("created bootstrap task (pid=%u)", bootstrap_task->t_id);
 
 	status = bsp_launch_async(&bsp, bootstrap_task);

kernel/bsp.c

@@ -120,6 +120,7 @@ static kern_status_t map_executable_exec(
 		bsp->bsp_vmo,
 		text_foffset,
 		bsp->bsp_trailer.bsp_text_size,
+		VM_SHARED,
 		VM_PROT_READ | VM_PROT_EXEC | VM_PROT_USER,
 		&text_base);
 	if (status != KERN_OK) {
@@ -132,6 +133,7 @@ static kern_status_t map_executable_exec(
 		data,
 		data_foffset,
 		bsp->bsp_trailer.bsp_data_size,
+		VM_PRIVATE,
 		VM_PROT_READ | VM_PROT_WRITE | VM_PROT_USER,
 		&data_base);
 	if (status != KERN_OK) {
@@ -165,6 +167,7 @@ kern_status_t bsp_launch_async(struct bsp *bsp, struct task *task)
 		user_stack,
 		0,
 		BOOTSTRAP_STACK_SIZE,
+		VM_PRIVATE,
 		VM_PROT_READ | VM_PROT_WRITE | VM_PROT_USER,
 		&stack_buffer);
 
@@ -178,6 +181,7 @@ kern_status_t bsp_launch_async(struct bsp *bsp, struct task *task)
 		bsp->bsp_vmo,
 		0,
 		bsp->bsp_trailer.bsp_exec_offset,
+		VM_PRIVATE,
 		VM_PROT_READ | VM_PROT_USER,
 		&bsp_data_base);
 

kernel/channel.c

@@ -2,6 +2,7 @@
 #include <kernel/channel.h>
 #include <kernel/msg.h>
 #include <kernel/port.h>
+#include <kernel/printk.h>
 #include <kernel/task.h>
 #include <kernel/thread.h>
 #include <kernel/util.h>
@@ -101,7 +102,6 @@ static struct msg *get_next_msg(
 		switch (msg->msg_status) {
 		case KMSG_WAIT_RECEIVE:
 			msg->msg_status = KMSG_WAIT_REPLY;
-			msg->msg_sender_port->p_status = PORT_REPLY_BLOCKED;
 			channel->c_msg_waiting--;
 			return msg;
 		case KMSG_ASYNC:
@@ -139,7 +139,6 @@ extern kern_status_t channel_recv_msg(
 	kern_msg_t *out_msg,
 	unsigned long *irq_flags)
 {
-	struct thread *self = current_thread();
 	struct msg *msg = NULL;
 	unsigned long msg_lock_flags;
 
@@ -157,12 +156,12 @@ extern kern_status_t channel_recv_msg(
 	/* msg is now set to the next message to process */
 
 	if (msg->msg_type != KERN_MSG_TYPE_DATA) {
-		/* event messages as asynchronous */
+		/* event messages are asynchronous */
 		out_msg->msg_id = msg->msg_id;
 		out_msg->msg_type = msg->msg_type;
 		out_msg->msg_event = msg->msg_event;
-		out_msg->msg_sender = msg->msg_sender_thread->tr_parent->t_id;
-		out_msg->msg_endpoint = msg->msg_sender_port->p_base.ob_id;
+		out_msg->msg_sender = msg->msg_sender_thread_id;
+		out_msg->msg_endpoint = msg->msg_sender_port_id;
 		spin_unlock_irqrestore(&msg->msg_lock, msg_lock_flags);
 		msg_free(msg);
 
@@ -170,7 +169,7 @@ extern kern_status_t channel_recv_msg(
 	}
 
 	struct task *sender = msg->msg_sender_thread->tr_parent;
-	struct task *receiver = self->tr_parent;
+	struct task *receiver = get_current_task();
 
 	struct address_space *src = sender->t_address_space,
 			     *dst = receiver->t_address_space;
@@ -192,6 +191,7 @@ extern kern_status_t channel_recv_msg(
 
 	if (status != KERN_OK) {
 		kmsg_reply_error(msg, status, &msg_lock_flags);
+		put_current_task(receiver);
 		return status;
 	}
 
@@ -216,6 +216,7 @@ extern kern_status_t channel_recv_msg(
 		&receiver->t_handles_lock,
 		f);
 	address_space_unlock_pair_irqrestore(src, dst, f);
+	put_current_task(receiver);
 
 	if (status != KERN_OK) {
 		kmsg_reply_error(msg, status, &msg_lock_flags);
@@ -250,11 +251,10 @@ extern kern_status_t channel_reply_msg(
 		return KERN_INVALID_ARGUMENT;
 	}
 
-	struct thread *self = current_thread();
 	/* the task that is about to receive the response */
 	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 task *sender = get_current_task();
 
 	struct address_space *src = sender->t_address_space,
 			     *dst = receiver->t_address_space;
@@ -275,6 +275,7 @@ extern kern_status_t channel_reply_msg(
 
 	if (status != KERN_OK) {
 		kmsg_reply_error(msg, status, &msg_lock_flags);
+		put_current_task(sender);
 		return status;
 	}
 
@@ -299,6 +300,7 @@ extern kern_status_t channel_reply_msg(
 		&receiver->t_handles_lock,
 		f);
 	address_space_unlock_pair_irqrestore(src, dst, f);
+	put_current_task(sender);
 
 	if (status != KERN_OK) {
 		kmsg_reply_error(msg, status, &msg_lock_flags);

kernel/futex.c

@@ -32,8 +32,10 @@ static kern_status_t get_data(
 {
 	spin_lock_t *lock = NULL;
 	struct btree *futex_list = NULL;
+	struct task *self = NULL;
+
 	if (flags & FUTEX_PRIVATE) {
-		struct task *self = current_task();
+		self = get_current_task();
 		lock = &self->t_base.ob_lock;
 		futex_list = &self->t_futex;
 	} else if (flags & FUTEX_SHARED) {
@@ -48,12 +50,20 @@ static kern_status_t get_data(
 
 	if (!futex && !(flags & FUTEX_CREATE)) {
 		spin_unlock_irqrestore(lock, *irq_flags);
+		if (self) {
+			put_current_task(self);
+		}
+
 		return KERN_NO_ENTRY;
 	}
 
 	futex = vm_cache_alloc(&futex_cache, VM_NORMAL);
 	if (!futex) {
 		spin_unlock_irqrestore(lock, *irq_flags);
+		if (self) {
+			put_current_task(self);
+		}
+
 		return KERN_NO_MEMORY;
 	}
 
@@ -61,6 +71,10 @@ static kern_status_t get_data(
 
 	put_futex(futex_list, futex);
 
+	if (self) {
+		put_current_task(self);
+	}
+
 	*out = futex;
 	*out_lock = lock;
 	return KERN_OK;
@@ -68,9 +82,10 @@ static kern_status_t get_data(
 
 static kern_status_t cleanup_data(struct futex *futex, unsigned int flags)
 {
+	struct task *self = NULL;
 	struct btree *futex_list = NULL;
 	if (flags & FUTEX_PRIVATE) {
-		struct task *self = current_task();
+		self = get_current_task();
 		futex_list = &self->t_futex;
 	} else if (flags & FUTEX_SHARED) {
 		futex_list = &shared_futex_list;
@@ -81,12 +96,16 @@ static kern_status_t cleanup_data(struct futex *futex, unsigned int flags)
 	btree_delete(futex_list, &futex->f_node);
 	vm_cache_free(&futex_cache, futex);
 
+	if (self) {
+		put_current_task(self);
+	}
+
 	return KERN_OK;
 }
 
 static kern_status_t futex_get_shared(kern_futex_t *futex, futex_key_t *out)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 	struct address_space *space = self->t_address_space;
 
 	unsigned long flags;
@@ -97,6 +116,7 @@ static kern_status_t futex_get_shared(kern_futex_t *futex, futex_key_t *out)
 		out,
 		&flags);
 	address_space_unlock_irqrestore(space, flags);
+	put_current_task(self);
 	return status;
 }
 

kernel/handle.c

@@ -77,6 +77,81 @@ void handle_table_destroy(struct handle_table *tab)
 	do_handle_table_destroy(tab, 0);
 }
 
+static kern_status_t do_handle_table_duplicate_leaf(
+	struct handle_table *src,
+	struct handle_table **dest)
+{
+	struct handle_table *out
+		= vm_cache_alloc(&handle_table_cache, VM_NORMAL);
+	if (!out) {
+		return KERN_NO_MEMORY;
+	}
+
+	memcpy(out, src, sizeof *out);
+
+	for (size_t i = 0; i < HANDLES_PER_TABLE; i++) {
+		struct object *obj = src->t_handles.t_handle_list[i].h_object;
+		if (obj) {
+			object_ref(obj);
+		}
+	}
+
+	*dest = out;
+	return KERN_OK;
+}
+
+static kern_status_t do_handle_table_duplicate(
+	struct handle_table *src,
+	struct handle_table **dest,
+	unsigned int depth)
+{
+	if (depth == MAX_TABLE_DEPTH - 1) {
+		return do_handle_table_duplicate_leaf(src, dest);
+	}
+
+	struct handle_table *out
+		= vm_cache_alloc(&handle_table_cache, VM_NORMAL);
+	if (!out) {
+		return KERN_NO_MEMORY;
+	}
+
+	memcpy(out->t_subtables.t_subtable_map,
+	       src->t_subtables.t_subtable_map,
+	       sizeof out->t_subtables.t_subtable_map);
+	memset(out->t_subtables.t_subtable_list,
+	       0x0,
+	       sizeof out->t_subtables.t_subtable_list);
+
+	for (size_t i = 0; i < REFS_PER_TABLE; i++) {
+		struct handle_table *child
+			= src->t_subtables.t_subtable_list[i];
+		struct handle_table *dup = NULL;
+		kern_status_t status = KERN_OK;
+		if (child) {
+			status = do_handle_table_duplicate(
+				child,
+				&dup,
+				depth + 1);
+		}
+
+		if (status == KERN_OK) {
+			out->t_subtables.t_subtable_list[i] = dup;
+		} else {
+			return status;
+		}
+	}
+
+	*dest = out;
+	return KERN_OK;
+}
+
+kern_status_t handle_table_duplicate(
+	struct handle_table *src,
+	struct handle_table **dest)
+{
+	return do_handle_table_duplicate(src, dest, 0);
+}
+
 static kern_status_t decode_handle_indices(
 	kern_handle_t handle,
 	unsigned int indices[MAX_TABLE_DEPTH])
@@ -109,7 +184,7 @@ static kern_status_t encode_handle_indices(
 	return KERN_OK;
 }
 
-kern_status_t handle_table_alloc_handle(
+static kern_status_t alloc_handle(
 	struct handle_table *tab,
 	struct handle **out_slot,
 	kern_handle_t *out_handle)
@@ -164,6 +239,53 @@ kern_status_t handle_table_alloc_handle(
 	return encode_handle_indices(indices, out_handle);
 }
 
+kern_status_t handle_table_alloc_handle(
+	struct handle_table *tab,
+	kern_handle_t value,
+	struct handle **out_slot,
+	kern_handle_t *out_handle)
+{
+	if (value == KERN_HANDLE_INVALID) {
+		return alloc_handle(tab, out_slot, out_handle);
+	}
+
+	unsigned int indices[MAX_TABLE_DEPTH];
+	if (decode_handle_indices(value, indices) != KERN_OK) {
+		return KERN_HANDLE_INVALID;
+	}
+
+	int i;
+	for (i = 0; i < MAX_TABLE_DEPTH - 1; i++) {
+		struct handle_table *next
+			= tab->t_subtables.t_subtable_list[indices[i]];
+		if (!next) {
+			next = handle_table_create();
+			tab->t_subtables.t_subtable_list[indices[i]] = next;
+		}
+
+		if (!next) {
+			return KERN_NO_MEMORY;
+		}
+
+		tab = next;
+	}
+
+	unsigned int handle_index = indices[i];
+	bitmap_set(tab->t_handles.t_handle_map, handle_index);
+
+	struct handle *out = &tab->t_handles.t_handle_list[handle_index];
+
+	if (out->h_object) {
+		object_unref(out->h_object);
+		out->h_object = NULL;
+		out->h_flags = 0;
+	}
+
+	*out_slot = out;
+	*out_handle = value;
+	return KERN_OK;
+}
+
 kern_status_t handle_table_free_handle(
 	struct handle_table *tab,
 	kern_handle_t handle)
@@ -294,11 +416,12 @@ kern_status_t handle_table_transfer(
 		}
 
 		switch (src_handle.hnd_mode) {
-		case KERN_MSG_HANDLE_IGNORE:
+		case HANDLE_TRANSFER_IGNORE:
 			break;
-		case KERN_MSG_HANDLE_MOVE:
+		case HANDLE_TRANSFER_MOVE:
 			status = handle_table_alloc_handle(
 				dst,
+				KERN_HANDLE_INVALID,
 				&dst_entry,
 				&dst_value);
 			if (status != KERN_OK) {
@@ -315,9 +438,10 @@ kern_status_t handle_table_transfer(
 			dst_handle.hnd_value = dst_value;
 			dst_handle.hnd_result = KERN_OK;
 			break;
-		case KERN_MSG_HANDLE_COPY:
+		case HANDLE_TRANSFER_COPY:
 			status = handle_table_alloc_handle(
 				dst,
+				KERN_HANDLE_INVALID,
 				&dst_entry,
 				&dst_value);
 			if (status != KERN_OK) {
@@ -364,7 +488,7 @@ kern_status_t handle_table_transfer(
 			&handle,
 			NULL);
 
-		if (handle.hnd_mode != KERN_MSG_HANDLE_MOVE) {
+		if (handle.hnd_mode != HANDLE_TRANSFER_MOVE) {
 			continue;
 		}
 

kernel/object.c

@@ -175,7 +175,7 @@ void object_wait_signal(
 	uint32_t signals,
 	unsigned long *irq_flags)
 {
-	struct thread *self = current_thread();
+	struct thread *self = get_current_thread();
 	struct wait_item waiter;
 	wait_item_init(&waiter, self);
 	for (;;) {
@@ -189,4 +189,5 @@ void object_wait_signal(
 		object_lock_irqsave(obj, irq_flags);
 	}
 	thread_wait_end(&waiter, &obj->ob_wq);
+	put_current_thread(self);
 }

kernel/panic.c

@@ -19,8 +19,8 @@ void panic_irq(struct ml_cpu_context *ctx, const char *fmt, ...)
 	printk("---[ kernel panic: %s", buf);
 	printk("kernel: " BUILD_ID ", compiler version: " __VERSION__);
 
-	struct task *task = current_task();
-	struct thread *thr = current_thread();
+	struct task *task = get_current_task();
+	struct thread *thr = get_current_thread();
 
 	if (task && thr) {
 		printk("task: %s (id: %d, thread: %d)",
@@ -31,6 +31,9 @@ void panic_irq(struct ml_cpu_context *ctx, const char *fmt, ...)
 		printk("task: [bootstrap]");
 	}
 
+	put_current_thread(thr);
+	put_current_task(task);
+
 	printk("cpu: %u", this_cpu());
 
 	ml_print_cpu_state(ctx);

kernel/port.c

@@ -37,7 +37,7 @@ struct port *port_cast(struct object *obj)
 static void wait_for_reply(struct msg *msg, unsigned long *lock_flags)
 {
 	struct wait_item waiter;
-	struct thread *self = current_thread();
+	struct thread *self = get_current_thread();
 
 	wait_item_init(&waiter, self);
 	for (;;) {
@@ -52,6 +52,7 @@ static void wait_for_reply(struct msg *msg, unsigned long *lock_flags)
 	}
 
 	self->tr_state = THREAD_READY;
+	put_current_thread(self);
 }
 
 struct port *port_create(void)
@@ -83,8 +84,11 @@ kern_status_t port_connect(struct port *port, struct channel *remote)
 	msg->msg_status = KMSG_ASYNC;
 	msg->msg_type = KERN_MSG_TYPE_EVENT;
 	msg->msg_event = KERN_MSG_EVENT_CONNECTION;
-	msg->msg_sender_thread = current_thread();
-	msg->msg_sender_port = port;
+	msg->msg_sender_port_id = port->p_base.ob_id;
+
+	struct thread *self = get_current_thread();
+	msg->msg_sender_thread_id = self->tr_id;
+	put_current_thread(self);
 
 	unsigned long flags;
 	channel_lock_irqsave(remote, &flags);
@@ -112,8 +116,11 @@ kern_status_t port_disconnect(struct port *port)
 	msg->msg_status = KMSG_ASYNC;
 	msg->msg_type = KERN_MSG_TYPE_EVENT;
 	msg->msg_event = KERN_MSG_EVENT_DISCONNECTION;
-	msg->msg_sender_thread = current_thread();
-	msg->msg_sender_port = port;
+	msg->msg_sender_port_id = port->p_base.ob_id;
+
+	struct thread *self = get_current_thread();
+	msg->msg_sender_thread_id = self->tr_id;
+	put_current_thread(self);
 
 	unsigned long flags;
 	channel_lock_irqsave(port->p_remote, &flags);
@@ -136,7 +143,7 @@ kern_status_t port_send_msg(
 		return KERN_BAD_STATE;
 	}
 
-	struct thread *self = current_thread();
+	struct thread *self = get_current_thread();
 	struct msg msg;
 	memset(&msg, 0x0, sizeof msg);
 	msg.msg_type = KERN_MSG_TYPE_DATA;
@@ -148,7 +155,6 @@ kern_status_t port_send_msg(
 
 	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);
 
@@ -157,6 +163,7 @@ kern_status_t port_send_msg(
 	channel_lock_irqsave(port->p_remote, &flags);
 	btree_delete(&port->p_remote->c_msg, &msg.msg_node);
 	channel_unlock_irqrestore(port->p_remote, flags);
+	put_current_thread(self);
 
 	return msg.msg_result;
 }

kernel/printk.c

@@ -98,9 +98,8 @@ int printk(const char *format, ...)
 	unsigned long flags;
 	spin_lock_irqsave(&log_buffer_lock, &flags);
 	save_log_message(msg);
-	spin_unlock_irqrestore(&log_buffer_lock, flags);
-
 	flush_log_buffer();
+	spin_unlock_irqrestore(&log_buffer_lock, flags);
 
 	return 0;
 }

libc/string/memcmp.c

@@ -3,6 +3,7 @@
 int memcmp(const void *vl, const void *vr, size_t n)
 {
 	const unsigned char *l = vl, *r = vr;
-    for (; n && *l == *r; n--, l++, r++);
+	for (; n && *l == *r; n--, l++, r++)
+		;
 	return n ? *l - *r : 0;
 }

libc/string/memmove.c

@@ -1,5 +1,5 @@
-#include <stdint.h>
 #include <kernel/libc/string.h>
+#include <stdint.h>
 
 static void *memcpy_r(void *dest, const void *src, size_t sz)
 {

libmango/arch/x86_64/syscall.S

@@ -58,6 +58,7 @@
 SYSCALL_GATE task_exit SYS_TASK_EXIT 1
 SYSCALL_GATE task_self SYS_TASK_SELF 1
 SYSCALL_GATE task_create SYS_TASK_CREATE 5
+SYSCALL_GATE task_duplicate SYS_TASK_DUPLICATE 2
 SYSCALL_GATE task_create_thread SYS_TASK_CREATE_THREAD 6
 SYSCALL_GATE task_get_address_space SYS_TASK_GET_ADDRESS_SPACE 1
 SYSCALL_GATE task_config_get SYS_TASK_CONFIG_GET 4
@@ -76,14 +77,14 @@ SYSCALL_GATE vm_object_copy SYS_VM_OBJECT_COPY 6
 
 SYSCALL_GATE address_space_read SYS_ADDRESS_SPACE_READ 5
 SYSCALL_GATE address_space_write SYS_ADDRESS_SPACE_WRITE 5
-SYSCALL_GATE address_space_map SYS_ADDRESS_SPACE_MAP 7
+SYSCALL_GATE address_space_map SYS_ADDRESS_SPACE_MAP 8
 SYSCALL_GATE address_space_unmap SYS_ADDRESS_SPACE_UNMAP 3
 SYSCALL_GATE address_space_reserve SYS_ADDRESS_SPACE_RESERVE 4
 SYSCALL_GATE address_space_release SYS_ADDRESS_SPACE_RELEASE 3
 
 SYSCALL_GATE kern_log SYS_KERN_LOG 1
 SYSCALL_GATE kern_handle_close SYS_KERN_HANDLE_CLOSE 1
-SYSCALL_GATE kern_handle_duplicate SYS_KERN_HANDLE_DUPLICATE 2
+SYSCALL_GATE kern_handle_transfer SYS_KERN_HANDLE_TRANSFER 6
 SYSCALL_GATE kern_config_get SYS_KERN_CONFIG_GET 3
 SYSCALL_GATE kern_config_set SYS_KERN_CONFIG_SET 3
 
@@ -101,10 +102,13 @@ SYSCALL_GATE vm_controller_create SYS_VM_CONTROLLER_CREATE 1
 SYSCALL_GATE vm_controller_recv SYS_VM_CONTROLLER_RECV 2
 SYSCALL_GATE vm_controller_recv_async SYS_VM_CONTROLLER_RECV_ASYNC 3
 SYSCALL_GATE vm_controller_create_object SYS_VM_CONTROLLER_CREATE_OBJECT 7
+SYSCALL_GATE vm_controller_prepare_attach SYS_VM_CONTROLLER_PREPARE_ATTACH 3
+SYSCALL_GATE vm_controller_finish_attach SYS_VM_CONTROLLER_FINISH_ATTACH 3
 SYSCALL_GATE vm_controller_detach_object SYS_VM_CONTROLLER_DETACH_OBJECT 2
 SYSCALL_GATE vm_controller_supply_pages SYS_VM_CONTROLLER_SUPPLY_PAGES 6
 
 SYSCALL_GATE kern_object_wait SYS_KERN_OBJECT_WAIT 2
+SYSCALL_GATE kern_object_query SYS_KERN_OBJECT_QUERY 2
 
 SYSCALL_GATE futex_wait SYS_FUTEX_WAIT 3
 SYSCALL_GATE futex_wake SYS_FUTEX_WAKE 3

libmango/include-user/mango/handle.h

@@ -5,8 +5,18 @@
 #include <mango/types.h>
 
 extern kern_status_t kern_handle_close(kern_handle_t handle);
-extern kern_status_t kern_handle_duplicate(
+extern kern_status_t kern_handle_transfer(
+	kern_handle_t src_task,
+	kern_handle_t src_handle,
+	kern_handle_t dest_task,
+	kern_handle_t dest_handle,
+	unsigned int mode,
+	kern_handle_t *out_handle);
+extern kern_status_t kern_handle_control(
+	kern_handle_t task,
 	kern_handle_t handle,
-	kern_handle_t *out);
+	uint32_t set_mask,
+	uint32_t clear_mask,
+	uint32_t *out_flags);
 
 #endif

libmango/include-user/mango/object.h

@@ -4,5 +4,8 @@
 #include <mango/types.h>
 
 extern kern_status_t kern_object_wait(kern_wait_item_t *items, size_t nr_items);
+extern kern_status_t kern_object_query(
+	kern_handle_t handle,
+	kern_object_info_t *out);
 
 #endif

libmango/include-user/mango/task.h

@@ -33,6 +33,9 @@ extern kern_status_t task_config_set(
 	kern_config_key_t key,
 	const void *ptr,
 	size_t len);
+extern kern_status_t task_duplicate(
+	kern_handle_t *out_task,
+	kern_handle_t *out_address_space);
 
 extern kern_status_t thread_self(kern_handle_t *out);
 extern kern_status_t thread_start(kern_handle_t thread);

libmango/include-user/mango/vm.h

@@ -48,6 +48,7 @@ extern kern_status_t address_space_map(
 	kern_handle_t object,
 	off_t object_offset,
 	size_t length,
+	vm_flags_t flags,
 	vm_prot_t prot,
 	virt_addr_t *out_base_address);
 extern kern_status_t address_space_unmap(
@@ -67,7 +68,7 @@ extern kern_status_t address_space_release(
 extern kern_status_t vm_controller_create(kern_handle_t *out);
 extern kern_status_t vm_controller_recv(
 	kern_handle_t ctrl,
-	equeue_packet_page_request_t *out);
+	equeue_packet_vm_request_t *out);
 extern kern_status_t vm_controller_recv_async(
 	kern_handle_t ctrl,
 	kern_handle_t eq,
@@ -80,6 +81,14 @@ extern kern_status_t vm_controller_create_object(
 	size_t data_len,
 	vm_prot_t prot,
 	kern_handle_t *out);
+extern kern_status_t vm_controller_prepare_attach(
+	kern_handle_t ctrl,
+	uint64_t req_id,
+	kern_handle_t *out_vmo);
+extern kern_status_t vm_controller_finish_attach(
+	kern_handle_t ctrl,
+	uint64_t req_id,
+	equeue_key_t new_key);
 extern kern_status_t vm_controller_detach_object(
 	kern_handle_t ctrl,
 	kern_handle_t vmo);

libmango/include/mango/syscall.h

@@ -3,51 +3,56 @@
 
 #define SYS_KERN_LOG                     1
 #define SYS_KERN_HANDLE_CLOSE            2
-#define SYS_KERN_HANDLE_DUPLICATE       3
-#define SYS_KERN_CONFIG_GET             4
-#define SYS_KERN_CONFIG_SET             5
-#define SYS_KERN_OBJECT_WAIT            6
-#define SYS_KERN_OBJECT_WAIT_ASYNC      7
-#define SYS_TASK_EXIT                   8
-#define SYS_TASK_SELF                   9
-#define SYS_TASK_CREATE                 10
-#define SYS_TASK_CREATE_THREAD          11
-#define SYS_TASK_GET_ADDRESS_SPACE      12
-#define SYS_TASK_CONFIG_GET             13
-#define SYS_TASK_CONFIG_SET             14
-#define SYS_THREAD_SELF                 15
-#define SYS_THREAD_START                16
-#define SYS_THREAD_EXIT                 17
-#define SYS_THREAD_CONFIG_GET           18
-#define SYS_THREAD_CONFIG_SET           19
-#define SYS_VM_OBJECT_CREATE            20
-#define SYS_VM_OBJECT_READ              21
-#define SYS_VM_OBJECT_WRITE             22
-#define SYS_VM_OBJECT_COPY              23
-#define SYS_ADDRESS_SPACE_READ          24
-#define SYS_ADDRESS_SPACE_WRITE         25
-#define SYS_ADDRESS_SPACE_MAP           26
-#define SYS_ADDRESS_SPACE_UNMAP         27
-#define SYS_ADDRESS_SPACE_RESERVE       28
-#define SYS_ADDRESS_SPACE_RELEASE       29
-#define SYS_MSG_SEND                    30
-#define SYS_MSG_RECV                    31
-#define SYS_MSG_REPLY                   32
-#define SYS_MSG_READ                    33
-#define SYS_MSG_WRITE                   34
-#define SYS_CHANNEL_CREATE              35
-#define SYS_PORT_CREATE                 36
-#define SYS_PORT_CONNECT                37
-#define SYS_PORT_DISCONNECT             38
-#define SYS_EQUEUE_CREATE               39
-#define SYS_EQUEUE_DEQUEUE              40
-#define SYS_VM_CONTROLLER_CREATE        41
-#define SYS_VM_CONTROLLER_RECV          42
-#define SYS_VM_CONTROLLER_RECV_ASYNC    43
-#define SYS_VM_CONTROLLER_CREATE_OBJECT 44
-#define SYS_VM_CONTROLLER_DETACH_OBJECT 45
-#define SYS_VM_CONTROLLER_SUPPLY_PAGES  46
-#define SYS_FUTEX_WAIT                  47
-#define SYS_FUTEX_WAKE                  48
+#define SYS_KERN_HANDLE_TRANSFER         3
+#define SYS_KERN_HANDLE_CONTROL          4
+#define SYS_KERN_CONFIG_GET              5
+#define SYS_KERN_CONFIG_SET              6
+#define SYS_KERN_OBJECT_WAIT             7
+#define SYS_KERN_OBJECT_WAIT_ASYNC       8
+#define SYS_TASK_EXIT                    9
+#define SYS_TASK_SELF                    10
+#define SYS_TASK_CREATE                  11
+#define SYS_TASK_CREATE_THREAD           12
+#define SYS_TASK_GET_ADDRESS_SPACE       13
+#define SYS_TASK_CONFIG_GET              14
+#define SYS_TASK_CONFIG_SET              15
+#define SYS_TASK_DUPLICATE               16
+#define SYS_THREAD_SELF                  17
+#define SYS_THREAD_START                 18
+#define SYS_THREAD_EXIT                  19
+#define SYS_THREAD_CONFIG_GET            20
+#define SYS_THREAD_CONFIG_SET            21
+#define SYS_VM_OBJECT_CREATE             22
+#define SYS_VM_OBJECT_READ               23
+#define SYS_VM_OBJECT_WRITE              24
+#define SYS_VM_OBJECT_COPY               25
+#define SYS_ADDRESS_SPACE_READ           26
+#define SYS_ADDRESS_SPACE_WRITE          27
+#define SYS_ADDRESS_SPACE_MAP            28
+#define SYS_ADDRESS_SPACE_UNMAP          29
+#define SYS_ADDRESS_SPACE_RESERVE        30
+#define SYS_ADDRESS_SPACE_RELEASE        31
+#define SYS_MSG_SEND                     32
+#define SYS_MSG_RECV                     33
+#define SYS_MSG_REPLY                    34
+#define SYS_MSG_READ                     35
+#define SYS_MSG_WRITE                    36
+#define SYS_CHANNEL_CREATE               37
+#define SYS_PORT_CREATE                  38
+#define SYS_PORT_CONNECT                 39
+#define SYS_PORT_DISCONNECT              40
+#define SYS_EQUEUE_CREATE                41
+#define SYS_EQUEUE_DEQUEUE               42
+#define SYS_VM_CONTROLLER_CREATE         43
+#define SYS_VM_CONTROLLER_RECV           44
+#define SYS_VM_CONTROLLER_RECV_ASYNC     45
+#define SYS_VM_CONTROLLER_CREATE_OBJECT  46
+#define SYS_VM_CONTROLLER_PREPARE_ATTACH 47
+#define SYS_VM_CONTROLLER_FINISH_ATTACH  48
+#define SYS_VM_CONTROLLER_DETACH_OBJECT  49
+#define SYS_VM_CONTROLLER_SUPPLY_PAGES   50
+#define SYS_FUTEX_WAIT                   51
+#define SYS_FUTEX_WAKE                   52
+#define SYS_KERN_OBJECT_QUERY            53
 
 #endif

libmango/include/mango/types.h

@@ -28,17 +28,33 @@
 #define THREAD_CFG_FSBASE            0x20001u
 #define THREAD_CFG_GSBASE            0x20002u
 
+/* user-defined flags that can be set on handles */
+#define KERN_HANDLE_FLAG0            0x10000000UL
+#define KERN_HANDLE_FLAG1            0x20000000UL
+#define KERN_HANDLE_FLAG2            0x40000000UL
+#define KERN_HANDLE_FLAG3            0x80000000UL
+
+/* flags to specify when creating address-space mappings */
+/* this mapping is private. if a task with this mapping is duplicated,
+   the duplicate task will receive a copy-on-write mapping. changes to one
+   mapping will not be visible to the other. */
+#define VM_PRIVATE                   0x0000u
+/* this mapping is shared. if a task with this mapping is duplicated,
+ * the duplicate will receive a mapping of the same data. changes to one mapping
+ * will be visibile to the other */
+#define VM_SHARED                    0x0001u
+
 /* maximum number of handles that can be sent in a single message */
 #define KERN_MSG_MAX_HANDLES         64
 
 /* the corresponding handle should be ignored */
-#define KERN_MSG_HANDLE_IGNORE       0
+#define HANDLE_TRANSFER_IGNORE       0
 /* the corresponding handle should be moved to the recipient task. the handle
  * will be closed. */
-#define KERN_MSG_HANDLE_MOVE         1
+#define HANDLE_TRANSFER_MOVE         1
 /* the corresponding handle should be copied to the recipient task. the handle
  * will remain valid for the sending task. */
-#define KERN_MSG_HANDLE_COPY         2
+#define HANDLE_TRANSFER_COPY         2
 
 /* maximum number of objects that can be waited on in a single call to
  * kern_object_wait */
@@ -55,13 +71,14 @@
 #define KERN_MSG_EVENT_DISCONNECTION 2
 
 /* equeue packet types */
-#define EQUEUE_PKT_PAGE_REQUEST      0x01u
+#define EQUEUE_PKT_VM_REQUEST        0x01u
 #define EQUEUE_PKT_ASYNC_SIGNAL      0x02u
 
-/* page request types */
-#define PAGE_REQUEST_READ            0x01u
-#define PAGE_REQUEST_DIRTY           0x02u
-#define PAGE_REQUEST_DETACH          0x03u
+/* vm request types */
+#define VM_REQUEST_READ              0x01u
+#define VM_REQUEST_DIRTY             0x02u
+#define VM_REQUEST_ATTACH            0x03u
+#define VM_REQUEST_DETACH            0x04u
 
 /* futex special values */
 #define FUTEX_WAKE_ALL               ((size_t)-1)
@@ -100,6 +117,7 @@ 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 vm_flags_t;
 typedef int64_t ssize_t;
 typedef uint32_t kern_futex_t;
 typedef uint32_t kern_msg_type_t;
@@ -109,6 +127,10 @@ typedef unsigned short equeue_packet_type_t;
 
 typedef unsigned int umode_t;
 
+typedef struct {
+	koid_t obj_id;
+} kern_object_info_t;
+
 typedef struct {
 	virt_addr_t io_base;
 	size_t io_len;
@@ -165,14 +187,33 @@ typedef struct {
 	/* the key of the vm-object for which the page request relates, as
 	 * specified when the vm-object was created */
 	equeue_key_t req_vmo;
-	/* page request type. one of PAGE_REQUEST_* */
+	/* page request type. one of VM_REQUEST_* */
 	unsigned short req_type;
-	/* of the offset into the vm-object for which pages are being requested
+	/* the offset into the vm-object for which pages are being requested */
+	union {
+		/* used for:
+		 * 	VM_REQUEST_READ
+		 * 	VM_REQUEST_DIRTY
 		 */
+		struct {
 			off_t req_offset;
 			/* the length in bytes of the region being requested */
 			size_t req_length;
-} equeue_packet_page_request_t;
+		};
+
+		/* used for:
+		 * 	VM_REQUEST_ATTACH
+		 */
+		struct {
+			/* the key of the original/source vmo. */
+			equeue_key_t req_src_vmo;
+			/* a request ID. used to retrieve information about
+			 * the newly-attached object, as the server won't know
+			 * about it yet, and won't have a handle to it. */
+			uint64_t req_id;
+		};
+	};
+} equeue_packet_vm_request_t;
 
 typedef struct {
 	/* the type of packet. one of EQUEUE_PKT_* */
@@ -182,8 +223,8 @@ typedef struct {
 	equeue_key_t p_key;
 
 	union {
-		/* p_type = EQUEUE_PKT_PAGE_REQUEST */
-		equeue_packet_page_request_t page_request;
+		/* p_type = EQUEUE_PKT_VM_REQUEST */
+		equeue_packet_vm_request_t vm_request;
 		/* p_type = EQUEUE_PKT_ASYNC_SIGNAL */
 		equeue_packet_async_signal_t async_signal;
 	};

sched/core.c

@@ -215,18 +215,19 @@ void schedule_thread_on_cpu(struct thread *thr)
 
 void start_charge_period(void)
 {
-	struct thread *self = current_thread();
+	struct thread *self = get_current_thread();
 	if (!self) {
 		return;
 	}
 
 	self->tr_charge_period_start = get_cycles();
+	put_current_thread(self);
 }
 
 void end_charge_period(void)
 {
 	preempt_disable();
-	struct thread *self = current_thread();
+	struct thread *self = get_current_thread();
 	if (!self) {
 		return;
 	}
@@ -244,6 +245,7 @@ void end_charge_period(void)
 	}
 
 	self->tr_charge_period_start = 0;
+	put_current_thread(self);
 
 	// printk("%llu cycles charged to %s/%u", charge,
 	// self->tr_parent->t_name, self->tr_parent->t_id);

sched/task.c

@@ -171,7 +171,10 @@ struct task *task_alloc(void)
 	return t;
 }
 
-struct task *task_create(const char *name, size_t name_len)
+struct task *task_create(
+	const char *name,
+	size_t name_len,
+	struct handle_table *handles)
 {
 	struct task *task = task_alloc();
 	if (!task) {
@@ -191,9 +194,13 @@ struct task *task_create(const char *name, size_t name_len)
 		VM_USER_LIMIT,
 		&task->t_address_space);
 
+	if (!handles) {
+		handles = handle_table_create();
+	}
+
 	task->t_address_space->s_pmap = pmap;
 	task->t_state = TASK_RUNNING;
-	task->t_handles = handle_table_create();
+	task->t_handles = handles;
 
 	if (name) {
 		name_len = MIN(name_len, sizeof task->t_name - 1);
@@ -282,7 +289,7 @@ struct task *task_from_tid(tid_t id)
 
 void task_exit(int status)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 	unsigned long flags;
 	task_lock_irqsave(self, &flags);
 	struct task *parent = self->t_parent;
@@ -295,7 +302,7 @@ void task_exit(int status)
 		task_unlock(parent);
 	}
 
-	struct thread *cur_thread = current_thread();
+	struct thread *cur_thread = get_current_thread();
 
 	self->t_state = TASK_STOPPED;
 	cur_thread->tr_state = THREAD_STOPPED;
@@ -340,6 +347,9 @@ void task_exit(int status)
 	       self->t_base.ob_refcount);
 	spin_unlock_irqrestore(handles_lock, flags);
 
+	put_current_thread(cur_thread);
+	put_current_task(self);
+
 	while (1) {
 		schedule(SCHED_NORMAL);
 	}
@@ -352,8 +362,11 @@ kern_status_t task_open_handle(
 	kern_handle_t *out)
 {
 	struct handle *handle_data = NULL;
-	kern_status_t status
-		= handle_table_alloc_handle(task->t_handles, &handle_data, out);
+	kern_status_t status = handle_table_alloc_handle(
+		task->t_handles,
+		KERN_HANDLE_INVALID,
+		&handle_data,
+		out);
 	if (status != KERN_OK) {
 		return status;
 	}
@@ -407,8 +420,19 @@ struct thread *task_create_thread(struct task *parent)
 	return thread;
 }
 
-struct task *current_task(void)
+struct task *get_current_task(void)
 {
-	struct thread *thr = current_thread();
-	return thr ? thr->tr_parent : NULL;
+	struct thread *thr = get_current_thread();
+	if (!thr) {
+		return NULL;
+	}
+
+	struct task *out = task_ref(thr->tr_parent);
+	put_current_thread(thr);
+	return out;
+}
+
+void put_current_task(struct task *task)
+{
+	task_unref(task);
 }

sched/thread.c

@@ -101,11 +101,57 @@ kern_status_t thread_init_user(
 	return KERN_OK;
 }
 
+kern_status_t thread_init_user_clone(
+	struct thread *thr,
+	const struct thread *src,
+	uintptr_t return_value)
+{
+	thr->tr_state = THREAD_READY;
+	thr->tr_quantum_target = default_quantum();
+
+	thr->tr_kstack = vm_page_alloc(THREAD_KSTACK_ORDER, VM_NORMAL);
+	if (!thr->tr_kstack) {
+		return KERN_NO_MEMORY;
+	}
+
+	thr->tr_sp = (uintptr_t)vm_page_get_vaddr(thr->tr_kstack)
+	           + vm_page_order_to_bytes(THREAD_KSTACK_ORDER);
+	thr->tr_bp = thr->tr_sp;
+	thr->tr_cpu_kernel_sp = thr->tr_sp;
+
+	/* the new thread needs two contextx:
+	 *   1) to get the thread running in kernel mode, so that it can
+	 *   execute ml_thread_switch_user
+	 *   2) to allow ml_thread_switch_user to jump to the correct place
+	 *   in usermode (and with the correct stack).
+	 *
+	 * these two contexts are constructed on the thread's kernel stack
+	 * in reverse order.
+	 */
+
+	/* this context will be used by ml_user_return to jump to userspace
+	 * with the specified instruction pointer and user stack */
+	ml_thread_clone_user_context(
+		src->tr_irqctx,
+		&src->tr_ml,
+		&thr->tr_ml,
+		return_value,
+		&thr->tr_sp);
+	/* this context will be used by the scheduler and ml_thread_switch to
+	 * jump to ml_user_return in kernel mode with the thread's kernel stack.
+	 */
+	ml_thread_prepare_kernel_context(
+		(uintptr_t)ml_thread_switch_user,
+		&thr->tr_sp);
+
+	return KERN_OK;
+}
+
 void thread_free(struct thread *thr)
 {
 }
 
-struct thread *current_thread(void)
+struct thread *get_current_thread(void)
 {
 	struct cpu_data *cpu = get_this_cpu();
 	if (!cpu) {
@@ -113,13 +159,24 @@ struct thread *current_thread(void)
 	}
 
 	struct thread *out = cpu->c_rq.rq_cur;
+	object_ref(&out->tr_base);
+
 	put_cpu(cpu);
 	return out;
 }
 
+void put_current_thread(struct thread *thr)
+{
+	object_unref(&thr->tr_base);
+}
+
 bool need_resched(void)
 {
-	return (current_thread()->tr_flags & THREAD_F_NEED_RESCHED) != 0;
+	struct thread *thr = get_current_thread();
+	bool result = (thr->tr_flags & THREAD_F_NEED_RESCHED) != 0;
+	put_current_thread(thr);
+
+	return result;
 }
 
 int thread_priority(struct thread *thr)
@@ -143,7 +200,7 @@ void thread_awaken(struct thread *thr)
 
 void thread_exit(void)
 {
-	struct thread *self = current_thread();
+	struct thread *self = get_current_thread();
 	unsigned long flags;
 	thread_lock_irqsave(self, &flags);
 	self->tr_state = THREAD_STOPPED;
@@ -153,6 +210,7 @@ void thread_exit(void)
 	       self->tr_parent->t_id,
 	       self->tr_id);
 	thread_unlock_irqrestore(self, flags);
+	put_current_thread(self);
 
 	while (1) {
 		schedule(SCHED_NORMAL);

sched/timer.c

@@ -44,7 +44,7 @@ unsigned long schedule_timeout(unsigned long ticks)
 {
 	struct timer timer;
 
-	struct thread *self = current_thread();
+	struct thread *self = get_current_thread();
 
 	timer.t_entry = QUEUE_ENTRY_INIT;
 	timer.t_expiry = clock_ticks + ticks;
@@ -58,6 +58,7 @@ unsigned long schedule_timeout(unsigned long ticks)
 	schedule(SCHED_NORMAL);
 
 	remove_timer(&timer);
+	put_current_thread(self);
 
 	return 0;
 }

sched/wait.c

@@ -110,7 +110,7 @@ void wakeup_one(struct waitqueue *q)
 
 void sleep_forever(void)
 {
-	struct thread *thr = current_thread();
+	struct thread *thr = get_current_thread();
 	struct runqueue *rq = thr->tr_rq;
 
 	unsigned long flags;
@@ -121,7 +121,17 @@ void sleep_forever(void)
 
 	rq_unlock(rq, flags);
 
-	while (thr->tr_state == THREAD_SLEEPING) {
+	put_current_thread(thr);
+
+	while (1) {
+		thr = get_current_thread();
+		bool sleep = (thr->tr_state == THREAD_SLEEPING);
+		put_current_thread(thr);
+
+		if (!sleep) {
+			break;
+		}
+
 		schedule(SCHED_NORMAL);
 	}
 }

syscall/address-space.c

@@ -11,13 +11,15 @@ kern_status_t sys_address_space_read(
 	size_t count,
 	size_t *nr_read)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	if (!validate_access_w(self, dst, count)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
 	if (nr_read && !validate_access_w(self, nr_read, sizeof *nr_read)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -30,12 +32,14 @@ kern_status_t sys_address_space_read(
 		= task_resolve_handle(self, region_handle, &obj, &handle_flags);
 	if (status != KERN_OK) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
 	struct address_space *region = address_space_cast(obj);
 	if (!region) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return KERN_INVALID_ARGUMENT;
 	}
 
@@ -52,6 +56,7 @@ kern_status_t sys_address_space_read(
 	address_space_unlock_irqrestore(region, flags);
 
 	object_unref(obj);
+	put_current_task(self);
 
 	return status;
 }
@@ -63,14 +68,16 @@ kern_status_t sys_address_space_write(
 	size_t count,
 	size_t *nr_written)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	if (!validate_access_r(self, src, count)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
 	if (nr_written
 	    && !validate_access_w(self, nr_written, sizeof *nr_written)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -83,12 +90,14 @@ kern_status_t sys_address_space_write(
 		= task_resolve_handle(self, region_handle, &obj, &handle_flags);
 	if (status != KERN_OK) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
 	struct address_space *region = address_space_cast(obj);
 	if (!region) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return KERN_INVALID_ARGUMENT;
 	}
 
@@ -105,6 +114,7 @@ kern_status_t sys_address_space_write(
 	address_space_unlock_irqrestore(region, flags);
 
 	object_unref(obj);
+	put_current_task(self);
 
 	return status;
 }
@@ -115,22 +125,24 @@ kern_status_t sys_address_space_map(
 	kern_handle_t object_handle,
 	off_t object_offset,
 	size_t length,
+	vm_flags_t flags,
 	vm_prot_t prot,
 	virt_addr_t *out_base_address)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	if (out_base_address
 	    && !validate_access_r(
 		    self,
 		    out_base_address,
 		    sizeof *out_base_address)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
 	kern_status_t status = KERN_OK;
-	unsigned long flags;
-	task_lock_irqsave(self, &flags);
+	unsigned long irq_flags;
+	task_lock_irqsave(self, &irq_flags);
 
 	struct object *region_obj = NULL, *vmo_obj = NULL;
 	handle_flags_t region_flags = 0, vmo_flags = 0;
@@ -140,30 +152,34 @@ kern_status_t sys_address_space_map(
 		&region_obj,
 		&region_flags);
 	if (status != KERN_OK) {
-		task_unlock_irqrestore(self, flags);
+		task_unlock_irqrestore(self, irq_flags);
+		put_current_task(self);
 		return status;
 	}
 
 	status = task_resolve_handle(self, object_handle, &vmo_obj, &vmo_flags);
 	if (status != KERN_OK) {
-		task_unlock_irqrestore(self, flags);
+		task_unlock_irqrestore(self, irq_flags);
+		put_current_task(self);
 		return status;
 	}
 
 	struct address_space *region = address_space_cast(region_obj);
 	if (!region) {
-		task_unlock_irqrestore(self, flags);
+		task_unlock_irqrestore(self, irq_flags);
+		put_current_task(self);
 		return KERN_INVALID_ARGUMENT;
 	}
 
 	struct vm_object *vmo = vm_object_cast(vmo_obj);
 	if (!vmo) {
-		task_unlock_irqrestore(self, flags);
+		task_unlock_irqrestore(self, irq_flags);
+		put_current_task(self);
 		return KERN_INVALID_ARGUMENT;
 	}
 
-	task_unlock_irqrestore(self, flags);
-	address_space_lock_irqsave(region, &flags);
+	task_unlock_irqrestore(self, irq_flags);
+	address_space_lock_irqsave(region, &irq_flags);
 	/* address_space_map will take care of locking `vmo` */
 	status = address_space_map(
 		region,
@@ -171,12 +187,14 @@ kern_status_t sys_address_space_map(
 		vmo,
 		object_offset,
 		length,
+		flags,
 		prot,
 		out_base_address);
-	address_space_unlock_irqrestore(region, flags);
+	address_space_unlock_irqrestore(region, irq_flags);
 
 	object_unref(vmo_obj);
 	object_unref(region_obj);
+	put_current_task(self);
 
 	return status;
 }
@@ -186,7 +204,7 @@ kern_status_t sys_address_space_unmap(
 	virt_addr_t base,
 	size_t length)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	kern_status_t status = KERN_OK;
 	unsigned long flags;
@@ -201,12 +219,14 @@ kern_status_t sys_address_space_unmap(
 		&region_flags);
 	if (status != KERN_OK) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
 	struct address_space *region = address_space_cast(region_obj);
 	if (!region) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return KERN_INVALID_ARGUMENT;
 	}
 
@@ -215,6 +235,7 @@ kern_status_t sys_address_space_unmap(
 	status = address_space_unmap(region, base, length);
 
 	object_unref(region_obj);
+	put_current_task(self);
 
 	return status;
 }
@@ -225,13 +246,14 @@ kern_status_t sys_address_space_reserve(
 	size_t length,
 	virt_addr_t *out_base_address)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	if (out_base_address
 	    && !validate_access_r(
 		    self,
 		    out_base_address,
 		    sizeof *out_base_address)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -248,12 +270,14 @@ kern_status_t sys_address_space_reserve(
 		&region_flags);
 	if (status != KERN_OK) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
 	struct address_space *region = address_space_cast(region_obj);
 	if (!region) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return KERN_INVALID_ARGUMENT;
 	}
 
@@ -268,6 +292,7 @@ kern_status_t sys_address_space_reserve(
 	address_space_unlock_irqrestore(region, flags);
 
 	object_unref(region_obj);
+	put_current_task(self);
 
 	return status;
 }
@@ -277,7 +302,7 @@ kern_status_t sys_address_space_release(
 	virt_addr_t base,
 	size_t length)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	kern_status_t status = KERN_OK;
 	unsigned long flags;
@@ -292,12 +317,14 @@ kern_status_t sys_address_space_release(
 		&region_flags);
 	if (status != KERN_OK) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
 	struct address_space *region = address_space_cast(region_obj);
 	if (!region) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return KERN_INVALID_ARGUMENT;
 	}
 
@@ -308,6 +335,7 @@ kern_status_t sys_address_space_release(
 	address_space_unlock_irqrestore(region, flags);
 
 	object_unref(region_obj);
+	put_current_task(self);
 
 	return status;
 }

syscall/config.c

@@ -3,21 +3,26 @@
 
 kern_status_t sys_kern_config_get(kern_config_key_t key, void *ptr, size_t len)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
+	kern_status_t status = KERN_OK;
 
 	switch (key) {
 	case KERN_CFG_PAGE_SIZE:
 		if (!validate_access_w(self, ptr, sizeof(uintptr_t))) {
-			return KERN_MEMORY_FAULT;
+			status = KERN_MEMORY_FAULT;
+			break;
 		}
 
 		*(uint32_t *)ptr = VM_PAGE_SIZE;
-		return KERN_OK;
+		status = KERN_OK;
+		break;
 	default:
-		return KERN_INVALID_ARGUMENT;
+		status = KERN_INVALID_ARGUMENT;
+		break;
 	}
 
-	return KERN_UNSUPPORTED;
+	put_current_task(self);
+	return status;
 }
 
 kern_status_t sys_kern_config_set(

syscall/dispatch.c

@@ -10,6 +10,7 @@ static const virt_addr_t syscall_table[] = {
 	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(TASK_DUPLICATE, task_duplicate),
 	SYSCALL_TABLE_ENTRY(THREAD_SELF, thread_self),
 	SYSCALL_TABLE_ENTRY(THREAD_START, thread_start),
 	SYSCALL_TABLE_ENTRY(THREAD_EXIT, thread_exit),
@@ -27,7 +28,8 @@ static const virt_addr_t syscall_table[] = {
 	SYSCALL_TABLE_ENTRY(ADDRESS_SPACE_RELEASE, address_space_release),
 	SYSCALL_TABLE_ENTRY(KERN_LOG, kern_log),
 	SYSCALL_TABLE_ENTRY(KERN_HANDLE_CLOSE, kern_handle_close),
-	SYSCALL_TABLE_ENTRY(KERN_HANDLE_DUPLICATE, kern_handle_duplicate),
+	SYSCALL_TABLE_ENTRY(KERN_HANDLE_TRANSFER, kern_handle_transfer),
+	SYSCALL_TABLE_ENTRY(KERN_HANDLE_CONTROL, kern_handle_control),
 	SYSCALL_TABLE_ENTRY(KERN_CONFIG_GET, kern_config_get),
 	SYSCALL_TABLE_ENTRY(KERN_CONFIG_SET, kern_config_set),
 	SYSCALL_TABLE_ENTRY(CHANNEL_CREATE, channel_create),
@@ -45,6 +47,12 @@ static const virt_addr_t syscall_table[] = {
 	SYSCALL_TABLE_ENTRY(
 		VM_CONTROLLER_CREATE_OBJECT,
 		vm_controller_create_object),
+	SYSCALL_TABLE_ENTRY(
+		VM_CONTROLLER_PREPARE_ATTACH,
+		vm_controller_prepare_attach),
+	SYSCALL_TABLE_ENTRY(
+		VM_CONTROLLER_FINISH_ATTACH,
+		vm_controller_finish_attach),
 	SYSCALL_TABLE_ENTRY(
 		VM_CONTROLLER_DETACH_OBJECT,
 		vm_controller_detach_object),
@@ -52,6 +60,7 @@ static const virt_addr_t syscall_table[] = {
 		VM_CONTROLLER_SUPPLY_PAGES,
 		vm_controller_supply_pages),
 	SYSCALL_TABLE_ENTRY(KERN_OBJECT_WAIT, kern_object_wait),
+	SYSCALL_TABLE_ENTRY(KERN_OBJECT_QUERY, kern_object_query),
 	SYSCALL_TABLE_ENTRY(FUTEX_WAIT, futex_wait),
 	SYSCALL_TABLE_ENTRY(FUTEX_WAKE, futex_wake),
 };

syscall/futex.c

@@ -8,18 +8,22 @@ kern_status_t sys_futex_wait(
 	kern_futex_t new_val,
 	unsigned int flags)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 	if (!validate_access_r(self, futex, sizeof *futex)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
 	futex_key_t key;
 	kern_status_t status = futex_get(futex, &key, flags);
 	if (status != KERN_OK) {
+		put_current_task(self);
 		return status;
 	}
 
-	return futex_wait(key, new_val, flags);
+	status = futex_wait(key, new_val, flags);
+	put_current_task(self);
+	return status;
 }
 
 kern_status_t sys_futex_wake(

syscall/handle.c

@@ -3,37 +3,156 @@
 
 kern_status_t sys_kern_handle_close(kern_handle_t handle)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
-	return task_close_handle(self, handle);
+	kern_status_t status = task_close_handle(self, handle);
+
+	put_current_task(self);
+
+	return status;
 }
 
-kern_status_t sys_kern_handle_duplicate(
-	kern_handle_t handle,
-	kern_handle_t *out)
+kern_status_t sys_kern_handle_transfer(
+	kern_handle_t src_task_handle,
+	kern_handle_t src_handle,
+	kern_handle_t dest_task_handle,
+	kern_handle_t dest_handle,
+	unsigned int mode,
+	kern_handle_t *out_handle)
 {
-	struct task *self = current_task();
+	switch (mode) {
+	case HANDLE_TRANSFER_MOVE:
+	case HANDLE_TRANSFER_COPY:
+		break;
+	default:
+		return KERN_INVALID_ARGUMENT;
+	}
 
-	if (!validate_access_w(self, out, sizeof *out)) {
+	struct task *self = get_current_task();
+
+	if (out_handle
+	    && !validate_access_w(self, out_handle, sizeof *out_handle)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
 	unsigned long flags;
 	task_lock_irqsave(self, &flags);
 
+	struct object *src_object = NULL;
+	struct task *src_task = NULL;
+	struct task *dest_task = NULL;
+	kern_status_t status = KERN_OK;
+
 	struct object *obj = NULL;
 	handle_flags_t handle_flags = 0;
-	kern_status_t status
-		= task_resolve_handle(self, handle, &obj, &handle_flags);
+
+	if (src_task_handle != KERN_HANDLE_INVALID) {
+		status = task_resolve_handle(
+			self,
+			src_task_handle,
+			&obj,
+			&handle_flags);
 		if (status != KERN_OK) {
 			task_unlock_irqrestore(self, flags);
-		return status;
+			goto cleanup;
 		}
 
-	status = task_open_handle(self, obj, handle_flags, out);
-	object_unref(obj);
+		src_task = task_cast(obj);
+		if (!src_task) {
+			status = KERN_INVALID_ARGUMENT;
+			goto cleanup;
+		}
+	} else {
+		src_task = self;
+	}
+
+	if (dest_task_handle != KERN_HANDLE_INVALID) {
+		status = task_resolve_handle(
+			self,
+			dest_task_handle,
+			&obj,
+			&handle_flags);
+		if (status != KERN_OK) {
+			task_unlock_irqrestore(self, flags);
+			goto cleanup;
+		}
+
+		dest_task = task_cast(obj);
+		if (!dest_task) {
+			status = KERN_INVALID_ARGUMENT;
+			goto cleanup;
+		}
+	} else {
+		dest_task = self;
+	}
+
+	status = task_resolve_handle(
+		self,
+		src_handle,
+		&src_object,
+		&handle_flags);
+	if (status != KERN_OK) {
+		goto cleanup;
+	}
 
 	task_unlock_irqrestore(self, flags);
 
+	struct handle *dest = NULL;
+	task_lock_irqsave(dest_task, &flags);
+	status = handle_table_alloc_handle(
+		dest_task->t_handles,
+		dest_handle,
+		&dest,
+		&dest_handle);
+	task_unlock_irqrestore(dest_task, flags);
+
+	if (status != KERN_OK) {
+		goto cleanup;
+	}
+
+	if (mode == HANDLE_TRANSFER_MOVE) {
+		object_ref(src_object);
+		task_lock_irqsave(src_task, &flags);
+		handle_table_free_handle(src_task->t_handles, src_handle);
+		task_unlock_irqrestore(src_task, flags);
+	}
+
+	dest->h_object = src_object;
+	dest->h_flags = handle_flags;
+
+	if (out_handle) {
+		*out_handle = dest_handle;
+	}
+
+	put_current_task(self);
+
+	return KERN_OK;
+
+cleanup:
+	if (src_task && src_task_handle != KERN_HANDLE_INVALID) {
+		object_unref(&src_task->t_base);
+	}
+
+	if (dest_task && dest_task_handle != KERN_HANDLE_INVALID) {
+		object_unref(&dest_task->t_base);
+	}
+
+	if (src_object) {
+		object_unref(src_object);
+	}
+
+	put_current_task(self);
+
 	return status;
 }
+
+kern_status_t sys_kern_handle_control(
+	kern_handle_t task,
+	kern_handle_t handle,
+	uint32_t set_mask,
+	uint32_t clear_mask,
+	uint32_t *out_flags)
+{
+	return KERN_UNIMPLEMENTED;
+}

syscall/log.c

@@ -5,8 +5,14 @@
 
 kern_status_t sys_kern_log(const char *s)
 {
-	struct task *task = current_task();
-	struct thread *thread = current_thread();
+#ifdef TRACE
+	struct task *task = get_current_task();
+	struct thread *thread = get_current_thread();
 	printk("%s[%d.%d]: %s", task->t_name, task->t_id, thread->tr_id, s);
+	put_current_thread(thread);
+	put_current_task(task);
+#else
+	printk("%s", s);
+#endif
 	return KERN_OK;
 }

syscall/msg.c

@@ -7,13 +7,15 @@
 
 kern_status_t sys_channel_create(unsigned int id, kern_handle_t *out)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 	if (!validate_access_w(self, out, sizeof *out)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
 	struct channel *channel = channel_create();
 	if (!channel) {
+		put_current_task(self);
 		return KERN_NO_MEMORY;
 	}
 
@@ -22,6 +24,7 @@ kern_status_t sys_channel_create(unsigned int id, kern_handle_t *out)
 
 	if (task_get_channel(self, id)) {
 		task_unlock_irqrestore(self, irq_flags);
+		put_current_task(self);
 		return KERN_NAME_EXISTS;
 	}
 
@@ -31,11 +34,13 @@ kern_status_t sys_channel_create(unsigned int id, kern_handle_t *out)
 	if (status != KERN_OK) {
 		task_unlock_irqrestore(self, irq_flags);
 		object_unref(&channel->c_base);
+		put_current_task(self);
 		return status;
 	}
 
 	task_add_channel(self, channel, id);
 	task_unlock_irqrestore(self, irq_flags);
+	put_current_task(self);
 
 	*out = handle;
 	return KERN_OK;
@@ -43,13 +48,15 @@ kern_status_t sys_channel_create(unsigned int id, kern_handle_t *out)
 
 kern_status_t sys_port_create(kern_handle_t *out)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 	if (!validate_access_w(self, out, sizeof *out)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
 	struct port *port = port_create();
 	if (!port) {
+		put_current_task(self);
 		return KERN_NO_MEMORY;
 	}
 
@@ -61,6 +68,7 @@ kern_status_t sys_port_create(kern_handle_t *out)
 		= task_open_handle(self, &port->p_base, 0, &handle);
 	task_unlock_irqrestore(self, irq_flags);
 	object_unref(&port->p_base);
+	put_current_task(self);
 
 	if (status != KERN_OK) {
 		return status;
@@ -77,7 +85,7 @@ kern_status_t sys_port_connect(
 {
 	unsigned long flags;
 
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 	task_lock_irqsave(self, &flags);
 
 	struct object *port_obj = NULL;
@@ -88,6 +96,7 @@ kern_status_t sys_port_connect(
 		&port_obj,
 		&port_handle_flags);
 	if (status != KERN_OK) {
+		put_current_task(self);
 		return status;
 	}
 
@@ -96,6 +105,7 @@ kern_status_t sys_port_connect(
 
 	struct task *remote_task = task_from_tid(task_id);
 	if (!remote_task) {
+		put_current_task(self);
 		return KERN_NO_ENTRY;
 	}
 
@@ -104,6 +114,7 @@ kern_status_t sys_port_connect(
 	struct channel *remote = task_get_channel(remote_task, channel_id);
 	if (!remote) {
 		task_unlock_irqrestore(remote_task, flags);
+		put_current_task(self);
 		return KERN_NO_ENTRY;
 	}
 
@@ -115,6 +126,7 @@ kern_status_t sys_port_connect(
 	port_unlock_irqrestore(port, flags);
 	object_unref(&remote->c_base);
 	object_unref(port_obj);
+	put_current_task(self);
 
 	return KERN_OK;
 }
@@ -123,7 +135,7 @@ kern_status_t sys_port_disconnect(kern_handle_t port_handle)
 {
 	unsigned long flags;
 
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 	task_lock_irqsave(self, &flags);
 
 	struct object *port_obj = NULL;
@@ -134,6 +146,7 @@ kern_status_t sys_port_disconnect(kern_handle_t port_handle)
 		&port_obj,
 		&port_handle_flags);
 	if (status != KERN_OK) {
+		put_current_task(self);
 		return status;
 	}
 
@@ -142,6 +155,7 @@ kern_status_t sys_port_disconnect(kern_handle_t port_handle)
 	struct port *port = port_cast(port_obj);
 	if (!port) {
 		object_unref(port_obj);
+		put_current_task(self);
 		return KERN_INVALID_ARGUMENT;
 	}
 
@@ -149,6 +163,7 @@ kern_status_t sys_port_disconnect(kern_handle_t port_handle)
 	port_lock_irqsave(port, &flags);
 	status = port_disconnect(port);
 	port_unlock_irqrestore(port, flags);
+	put_current_task(self);
 
 	return status;
 }
@@ -219,13 +234,15 @@ kern_status_t sys_msg_send(
 	const kern_msg_t *msg,
 	kern_msg_t *out_reply)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	if (!validate_msg(self, msg, false)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
 	if (!validate_msg(self, out_reply, true)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -240,15 +257,17 @@ kern_status_t sys_msg_send(
 		port_handle,
 		&port_obj,
 		&port_handle_flags);
+	task_unlock_irqrestore(self, flags);
+
 	if (status != KERN_OK) {
+		put_current_task(self);
 		return status;
 	}
 
-	task_unlock_irqrestore(self, flags);
-
 	struct port *port = port_cast(port_obj);
 	if (!port) {
 		object_unref(port_obj);
+		put_current_task(self);
 		return KERN_INVALID_ARGUMENT;
 	}
 
@@ -256,15 +275,17 @@ kern_status_t sys_msg_send(
 	status = port_send_msg(port, msg, out_reply, &flags);
 	port_unlock_irqrestore(port, flags);
 	object_unref(port_obj);
+	put_current_task(self);
 
 	return status;
 }
 
 kern_status_t sys_msg_recv(kern_handle_t channel_handle, kern_msg_t *out_msg)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	if (!validate_msg(self, out_msg, true)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -280,6 +301,7 @@ kern_status_t sys_msg_recv(kern_handle_t channel_handle, kern_msg_t *out_msg)
 		&channel_obj,
 		&channel_handle_flags);
 	if (status != KERN_OK) {
+		put_current_task(self);
 		return status;
 	}
 
@@ -288,6 +310,7 @@ kern_status_t sys_msg_recv(kern_handle_t channel_handle, kern_msg_t *out_msg)
 	struct channel *channel = channel_cast(channel_obj);
 	if (!channel) {
 		object_unref(channel_obj);
+		put_current_task(self);
 		return KERN_INVALID_ARGUMENT;
 	}
 
@@ -295,6 +318,7 @@ kern_status_t sys_msg_recv(kern_handle_t channel_handle, kern_msg_t *out_msg)
 	status = channel_recv_msg(channel, out_msg, &flags);
 	channel_unlock_irqrestore(channel, flags);
 	object_unref(channel_obj);
+	put_current_task(self);
 
 	return status;
 }
@@ -304,9 +328,10 @@ kern_status_t sys_msg_reply(
 	msgid_t id,
 	const kern_msg_t *reply)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	if (!validate_msg(self, reply, true)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -322,6 +347,7 @@ kern_status_t sys_msg_reply(
 		&channel_obj,
 		&channel_handle_flags);
 	if (status != KERN_OK) {
+		put_current_task(self);
 		return status;
 	}
 
@@ -330,6 +356,7 @@ kern_status_t sys_msg_reply(
 	struct channel *channel = channel_cast(channel_obj);
 	if (!channel) {
 		object_unref(channel_obj);
+		put_current_task(self);
 		return KERN_INVALID_ARGUMENT;
 	}
 
@@ -337,6 +364,7 @@ kern_status_t sys_msg_reply(
 	status = channel_reply_msg(channel, id, reply, &flags);
 	channel_unlock_irqrestore(channel, flags);
 	object_unref(channel_obj);
+	put_current_task(self);
 
 	return status;
 }
@@ -349,13 +377,15 @@ kern_status_t sys_msg_read(
 	size_t iov_count,
 	size_t *nr_read)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	if (nr_read && !validate_access_w(self, nr_read, sizeof *nr_read)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
 	if (!validate_iovec(self, iov, iov_count, true)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -371,6 +401,7 @@ kern_status_t sys_msg_read(
 		&channel_obj,
 		&channel_handle_flags);
 	if (status != KERN_OK) {
+		put_current_task(self);
 		return status;
 	}
 
@@ -379,6 +410,7 @@ kern_status_t sys_msg_read(
 	struct channel *channel = channel_cast(channel_obj);
 	if (!channel) {
 		object_unref(channel_obj);
+		put_current_task(self);
 		return KERN_INVALID_ARGUMENT;
 	}
 
@@ -393,6 +425,7 @@ kern_status_t sys_msg_read(
 		nr_read);
 	channel_unlock_irqrestore(channel, flags);
 	object_unref(channel_obj);
+	put_current_task(self);
 
 	return status;
 }
@@ -405,14 +438,16 @@ kern_status_t sys_msg_write(
 	size_t iov_count,
 	size_t *nr_written)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	if (nr_written
 	    && !validate_access_w(self, nr_written, sizeof *nr_written)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
 	if (!validate_iovec(self, iov, iov_count, false)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -428,6 +463,7 @@ kern_status_t sys_msg_write(
 		&channel_obj,
 		&channel_handle_flags);
 	if (status != KERN_OK) {
+		put_current_task(self);
 		return status;
 	}
 
@@ -436,6 +472,7 @@ kern_status_t sys_msg_write(
 	struct channel *channel = channel_cast(channel_obj);
 	if (!channel) {
 		object_unref(channel_obj);
+		put_current_task(self);
 		return KERN_INVALID_ARGUMENT;
 	}
 
@@ -450,6 +487,7 @@ kern_status_t sys_msg_write(
 		nr_written);
 	channel_unlock_irqrestore(channel, flags);
 	object_unref(channel_obj);
+	put_current_task(self);
 
 	return status;
 }

syscall/object.c

@@ -13,12 +13,13 @@ kern_status_t sys_kern_object_wait(kern_wait_item_t *items, size_t nr_items)
 		return KERN_INVALID_ARGUMENT;
 	}
 
-	struct task *self = current_task();
-	struct thread *self_thread = current_thread();
+	struct task *self = get_current_task();
 	if (!validate_access_rw(self, items, nr_items * sizeof *items)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
+	struct thread *self_thread = get_current_thread();
 	self_thread->tr_state = THREAD_SLEEPING;
 
 	kern_status_t status = KERN_OK;
@@ -78,5 +79,41 @@ cleanup:
 	}
 
 	self_thread->tr_state = THREAD_READY;
+	put_current_thread(self_thread);
+	put_current_task(self);
+
 	return status;
 }
+
+kern_status_t sys_kern_object_query(
+	kern_handle_t object_handle,
+	kern_object_info_t *out)
+{
+	struct task *self = get_current_task();
+
+	if (!out) {
+		put_current_task(self);
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	if (!validate_access_w(self, out, sizeof *out)) {
+		put_current_task(self);
+		return KERN_MEMORY_FAULT;
+	}
+
+	struct object *obj = NULL;
+	handle_flags_t flags = 0;
+	kern_status_t status
+		= task_resolve_handle(self, object_handle, &obj, &flags);
+	if (status != KERN_OK) {
+		put_current_task(self);
+		return status;
+	}
+
+	out->obj_id = obj->ob_id;
+
+	object_unref(obj);
+	put_current_task(self);
+
+	return KERN_OK;
+}

syscall/task.c

@@ -1,5 +1,6 @@
 #include <kernel/address-space.h>
 #include <kernel/machine/cpu.h>
+#include <kernel/panic.h>
 #include <kernel/printk.h>
 #include <kernel/sched.h>
 #include <kernel/syscall.h>
@@ -9,8 +10,9 @@
 extern kern_status_t sys_task_exit(int status)
 {
 #if defined(TRACE)
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 	printk("%s[%d]: task_exit(%d)", self->t_name, self->t_id, status);
+	put_current_task(self);
 #endif
 	task_exit(status);
 	return KERN_FATAL_ERROR;
@@ -18,8 +20,9 @@ extern kern_status_t sys_task_exit(int status)
 
 kern_status_t sys_task_self(kern_handle_t *out)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 	if (!validate_access_w(self, out, sizeof *out)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -30,16 +33,19 @@ kern_status_t sys_task_self(kern_handle_t *out)
 	kern_handle_t handle;
 	kern_status_t status = handle_table_alloc_handle(
 		self->t_handles,
+		KERN_HANDLE_INVALID,
 		&handle_slot,
 		&handle);
 	task_unlock_irqrestore(self, flags);
 
 	if (status != KERN_OK) {
+		put_current_task(self);
 		return status;
 	}
 
 	object_ref(&self->t_base);
 	handle_slot->h_object = &self->t_base;
+	put_current_task(self);
 
 	*out = handle;
 	return KERN_OK;
@@ -53,13 +59,15 @@ kern_status_t sys_task_create(
 	kern_handle_t *out_address_space)
 {
 	unsigned long flags;
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	if (name_len && !validate_access_r(self, name, name_len)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
 	if (!validate_access_w(self, out_task, sizeof *out_task)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -67,6 +75,7 @@ kern_status_t sys_task_create(
 		    self,
 		    out_address_space,
 		    sizeof *out_address_space)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -80,6 +89,7 @@ kern_status_t sys_task_create(
 		&parent_flags);
 	if (status != KERN_OK) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
@@ -89,28 +99,32 @@ kern_status_t sys_task_create(
 	kern_handle_t child_handle, space_handle;
 	status = handle_table_alloc_handle(
 		self->t_handles,
+		KERN_HANDLE_INVALID,
 		&child_handle_slot,
 		&child_handle);
 	if (status != KERN_OK) {
 		object_unref(parent_obj);
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
 	status = handle_table_alloc_handle(
 		self->t_handles,
+		KERN_HANDLE_INVALID,
 		&space_handle_slot,
 		&space_handle);
 	if (status != KERN_OK) {
 		object_unref(parent_obj);
 		handle_table_free_handle(self->t_handles, child_handle);
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
 	task_unlock_irqrestore(self, flags);
 
-	struct task *child = task_create(name, name_len);
+	struct task *child = task_create(name, name_len, NULL);
 	if (!child) {
 		object_unref(parent_obj);
 
@@ -118,6 +132,7 @@ kern_status_t sys_task_create(
 		handle_table_free_handle(self->t_handles, child_handle);
 		handle_table_free_handle(self->t_handles, space_handle);
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 
 		return KERN_NO_MEMORY;
 	}
@@ -133,6 +148,7 @@ kern_status_t sys_task_create(
 	object_ref(&child->t_address_space->s_base);
 
 	object_unref(parent_obj);
+	put_current_task(self);
 
 	*out_task = child_handle;
 	*out_address_space = space_handle;
@@ -149,13 +165,15 @@ kern_status_t sys_task_create_thread(
 	kern_handle_t *out_thread)
 {
 	unsigned long flags;
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	if (!validate_access_r(self, args, nr_args * sizeof(uintptr_t))) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
 	if (!validate_access_w(self, out_thread, sizeof *out_thread)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -166,6 +184,7 @@ kern_status_t sys_task_create_thread(
 		= task_resolve_handle(self, task, &target_obj, &target_flags);
 	if (status != KERN_OK) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
@@ -175,11 +194,13 @@ kern_status_t sys_task_create_thread(
 	kern_handle_t out_handle;
 	status = handle_table_alloc_handle(
 		self->t_handles,
+		KERN_HANDLE_INVALID,
 		&target_handle,
 		&out_handle);
 	if (status != KERN_OK) {
 		object_unref(target_obj);
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
@@ -194,6 +215,7 @@ kern_status_t sys_task_create_thread(
 		task_lock_irqsave(self, &flags);
 		handle_table_free_handle(self->t_handles, out_handle);
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return KERN_NO_MEMORY;
 	}
 
@@ -203,6 +225,7 @@ kern_status_t sys_task_create_thread(
 
 	task_unlock_irqrestore(target, flags);
 	object_unref(target_obj);
+	put_current_task(self);
 
 	*out_thread = out_handle;
 	return KERN_OK;
@@ -212,8 +235,9 @@ kern_status_t sys_task_get_address_space(
 	kern_handle_t task_handle,
 	kern_handle_t *out)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 	if (!validate_access_w(self, out, sizeof *out)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -231,16 +255,19 @@ kern_status_t sys_task_get_address_space(
 		&handle_flags);
 	if (status != KERN_OK) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
 	status = handle_table_alloc_handle(
 		self->t_handles,
+		KERN_HANDLE_INVALID,
 		&handle_slot,
 		&handle);
 	if (status != KERN_OK) {
 		object_unref(task_obj);
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
@@ -250,6 +277,7 @@ kern_status_t sys_task_get_address_space(
 		object_unref(task_obj);
 		handle_table_free_handle(self->t_handles, handle);
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return KERN_INVALID_ARGUMENT;
 	}
 
@@ -257,6 +285,7 @@ kern_status_t sys_task_get_address_space(
 	object_ref(&task->t_address_space->s_base);
 	task_unlock_irqrestore(self, flags);
 	object_unref(task_obj);
+	put_current_task(self);
 
 	*out = handle;
 	return KERN_OK;
@@ -264,12 +293,13 @@ kern_status_t sys_task_get_address_space(
 
 kern_status_t sys_thread_self(kern_handle_t *out)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 	if (!validate_access_w(self, out, sizeof *out)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
-	struct thread *self_thread = current_thread();
+	struct thread *self_thread = get_current_thread();
 
 	unsigned long flags;
 	task_lock_irqsave(self, &flags);
@@ -278,16 +308,21 @@ kern_status_t sys_thread_self(kern_handle_t *out)
 	kern_handle_t handle;
 	kern_status_t status = handle_table_alloc_handle(
 		self->t_handles,
+		KERN_HANDLE_INVALID,
 		&handle_slot,
 		&handle);
 	task_unlock_irqrestore(self, flags);
 
 	if (status != KERN_OK) {
+		put_current_thread(self_thread);
+		put_current_task(self);
 		return status;
 	}
 
 	object_ref(&self_thread->tr_base);
 	handle_slot->h_object = &self_thread->tr_base;
+	put_current_thread(self_thread);
+	put_current_task(self);
 
 	*out = handle;
 	return KERN_OK;
@@ -296,7 +331,7 @@ kern_status_t sys_thread_self(kern_handle_t *out)
 kern_status_t sys_thread_start(kern_handle_t thread_handle)
 {
 	unsigned long flags;
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	struct object *thread_obj;
 	handle_flags_t thread_flags;
@@ -308,6 +343,7 @@ kern_status_t sys_thread_start(kern_handle_t thread_handle)
 		&thread_flags);
 	if (status != KERN_OK) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
@@ -316,6 +352,7 @@ kern_status_t sys_thread_start(kern_handle_t thread_handle)
 
 	schedule_thread_on_cpu(thread);
 	object_unref(thread_obj);
+	put_current_task(self);
 
 	return KERN_OK;
 }
@@ -334,9 +371,10 @@ kern_status_t sys_thread_config_get(
 	size_t len)
 {
 	unsigned long flags;
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	if (!validate_access_w(self, ptr, len)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -350,6 +388,7 @@ kern_status_t sys_thread_config_get(
 		&thread_flags);
 	if (status != KERN_OK) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
@@ -359,6 +398,7 @@ kern_status_t sys_thread_config_get(
 	status = thread_config_get(thread, key, ptr, len);
 
 	object_unref(thread_obj);
+	put_current_task(self);
 
 	return status;
 }
@@ -370,9 +410,10 @@ kern_status_t sys_thread_config_set(
 	size_t len)
 {
 	unsigned long flags;
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	if (!validate_access_w(self, ptr, len)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -386,6 +427,7 @@ kern_status_t sys_thread_config_set(
 		&thread_flags);
 	if (status != KERN_OK) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
@@ -395,6 +437,122 @@ kern_status_t sys_thread_config_set(
 	status = thread_config_set(thread, key, ptr, len);
 
 	object_unref(thread_obj);
+	put_current_task(self);
 
 	return status;
 }
+
+kern_status_t sys_task_duplicate(
+	kern_handle_t *out_task,
+	kern_handle_t *out_address_space)
+{
+	struct task *self = get_current_task();
+
+	if (!validate_access_w(self, out_task, sizeof *out_task)) {
+		put_current_task(self);
+		return KERN_MEMORY_FAULT;
+	}
+
+	if (!validate_access_w(
+		    self,
+		    out_address_space,
+		    sizeof *out_address_space)) {
+		put_current_task(self);
+		return KERN_MEMORY_FAULT;
+	}
+
+	*out_task = KERN_HANDLE_INVALID;
+	*out_address_space = KERN_HANDLE_INVALID;
+
+	kern_status_t status = KERN_OK;
+	unsigned long flags;
+	task_lock_irqsave(self, &flags);
+
+	struct handle_table *child_handle_table = NULL;
+	status = handle_table_duplicate(self->t_handles, &child_handle_table);
+	if (status != KERN_OK) {
+		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
+		return status;
+	}
+
+	struct handle *child_handle_slot = NULL, *space_handle_slot = NULL;
+	kern_handle_t child_handle, space_handle;
+	status = handle_table_alloc_handle(
+		self->t_handles,
+		KERN_HANDLE_INVALID,
+		&child_handle_slot,
+		&child_handle);
+	if (status != KERN_OK) {
+		handle_table_destroy(child_handle_table);
+		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
+		return status;
+	}
+
+	status = handle_table_alloc_handle(
+		self->t_handles,
+		KERN_HANDLE_INVALID,
+		&space_handle_slot,
+		&space_handle);
+	if (status != KERN_OK) {
+		handle_table_destroy(child_handle_table);
+		handle_table_free_handle(self->t_handles, child_handle);
+		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
+		return status;
+	}
+
+	struct task *new_task = task_create(
+		self->t_name,
+		strlen(self->t_name),
+		child_handle_table);
+	if (!new_task) {
+		handle_table_destroy(child_handle_table);
+		put_current_task(self);
+		return KERN_NO_MEMORY;
+	}
+
+	struct thread *new_thread = task_create_thread(new_task);
+	if (!new_thread) {
+		handle_table_free_handle(self->t_handles, child_handle);
+		handle_table_free_handle(self->t_handles, space_handle);
+		task_unlock_irqrestore(self, flags);
+		object_unref(&new_task->t_base);
+		put_current_task(self);
+		return KERN_NO_MEMORY;
+	}
+
+	struct thread *self_thread = get_current_thread();
+	thread_init_user_clone(new_thread, self_thread, KERN_OK);
+	put_current_thread(self_thread);
+
+	status = address_space_duplicate(
+		new_task->t_address_space,
+		self->t_address_space);
+	if (status != KERN_OK) {
+		handle_table_free_handle(self->t_handles, child_handle);
+		handle_table_free_handle(self->t_handles, space_handle);
+		task_unlock_irqrestore(self, flags);
+		object_unref(&new_thread->tr_base);
+		object_unref(&new_task->t_base);
+		put_current_task(self);
+		return status;
+	}
+
+	child_handle_slot->h_object = &new_task->t_base;
+	space_handle_slot->h_object
+		= object_ref(&new_task->t_address_space->s_base);
+	task_unlock_irqrestore(self, flags);
+
+	/* clear TLB */
+	pmap_switch(self->t_pmap);
+	put_current_task(self);
+
+	*out_task = child_handle;
+	*out_address_space = space_handle;
+
+	schedule_thread_on_cpu(new_thread);
+
+	return KERN_OK;
+}

syscall/vm-controller.c

@@ -7,18 +7,21 @@
 
 kern_status_t sys_vm_controller_create(kern_handle_t *out)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	if (!validate_access_w(self, out, sizeof *out)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
 	struct vm_controller *ctrl = vm_controller_create();
 	if (!ctrl) {
+		put_current_task(self);
 		return KERN_NO_MEMORY;
 	}
 
 	kern_status_t status = task_open_handle(self, &ctrl->vc_base, 0, out);
+	put_current_task(self);
 	if (status != KERN_OK) {
 		object_unref(&ctrl->vc_base);
 		return status;
@@ -29,11 +32,12 @@ kern_status_t sys_vm_controller_create(kern_handle_t *out)
 
 kern_status_t sys_vm_controller_recv(
 	kern_handle_t ctrl_handle,
-	equeue_packet_page_request_t *out)
+	equeue_packet_vm_request_t *out)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	if (!validate_access_w(self, out, sizeof *out)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -50,6 +54,7 @@ kern_status_t sys_vm_controller_recv(
 		&handle_flags);
 	if (status != KERN_OK) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
@@ -57,6 +62,7 @@ kern_status_t sys_vm_controller_recv(
 	task_unlock_irqrestore(self, flags);
 	if (!ctrl) {
 		object_unref(ctrl_obj);
+		put_current_task(self);
 		return KERN_INVALID_ARGUMENT;
 	}
 
@@ -65,6 +71,7 @@ kern_status_t sys_vm_controller_recv(
 	vm_controller_unlock_irqrestore(ctrl, flags);
 
 	object_unref(ctrl_obj);
+	put_current_task(self);
 
 	return status;
 }
@@ -74,7 +81,7 @@ kern_status_t sys_vm_controller_recv_async(
 	kern_handle_t eq_handle,
 	equeue_key_t key)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	kern_status_t status = KERN_OK;
 	unsigned long flags;
@@ -85,6 +92,7 @@ kern_status_t sys_vm_controller_recv_async(
 	status = task_resolve_handle(self, ctrl_handle, &ctrl_obj, &ctrl_flags);
 	if (status != KERN_OK) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
@@ -92,6 +100,7 @@ kern_status_t sys_vm_controller_recv_async(
 	if (status != KERN_OK) {
 		object_unref(ctrl_obj);
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
@@ -102,6 +111,7 @@ kern_status_t sys_vm_controller_recv_async(
 	if (!ctrl || !eq) {
 		object_unref(ctrl_obj);
 		object_unref(eq_obj);
+		put_current_task(self);
 		return KERN_INVALID_ARGUMENT;
 	}
 
@@ -111,6 +121,7 @@ kern_status_t sys_vm_controller_recv_async(
 
 	object_unref(ctrl_obj);
 	object_unref(eq_obj);
+	put_current_task(self);
 
 	return status;
 }
@@ -124,19 +135,21 @@ kern_status_t sys_vm_controller_create_object(
 	vm_prot_t prot,
 	kern_handle_t *out)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	if (!validate_access_r(self, name, name_len)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
 	if (!validate_access_w(self, out, sizeof *out)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
 	kern_status_t status = KERN_OK;
-	unsigned long flags;
-	task_lock_irqsave(self, &flags);
+	unsigned long irq_flags;
+	task_lock_irqsave(self, &irq_flags);
 
 	struct object *ctrl_obj = NULL;
 	handle_flags_t handle_flags = 0;
@@ -146,7 +159,8 @@ kern_status_t sys_vm_controller_create_object(
 		&ctrl_obj,
 		&handle_flags);
 	if (status != KERN_OK) {
-		task_unlock_irqrestore(self, flags);
+		task_unlock_irqrestore(self, irq_flags);
+		put_current_task(self);
 		return status;
 	}
 
@@ -154,17 +168,19 @@ kern_status_t sys_vm_controller_create_object(
 	kern_handle_t out_handle = KERN_HANDLE_INVALID;
 	status = handle_table_alloc_handle(
 		self->t_handles,
+		KERN_HANDLE_INVALID,
 		&out_slot,
 		&out_handle);
 
 	struct vm_controller *ctrl = vm_controller_cast(ctrl_obj);
-	task_unlock_irqrestore(self, flags);
+	task_unlock_irqrestore(self, irq_flags);
 	if (!ctrl) {
 		object_unref(ctrl_obj);
+		put_current_task(self);
 		return KERN_INVALID_ARGUMENT;
 	}
 
-	vm_controller_lock_irqsave(ctrl, &flags);
+	vm_controller_lock_irqsave(ctrl, &irq_flags);
 	struct vm_object *out_vmo = NULL;
 	status = vm_controller_create_object(
 		ctrl,
@@ -174,28 +190,140 @@ kern_status_t sys_vm_controller_create_object(
 		data_len,
 		prot,
 		&out_vmo);
-	vm_controller_unlock_irqrestore(ctrl, flags);
+	vm_controller_unlock_irqrestore(ctrl, irq_flags);
 
 	object_unref(ctrl_obj);
 
 	if (status != KERN_OK) {
-		task_lock_irqsave(self, &flags);
+		task_lock_irqsave(self, &irq_flags);
 		handle_table_free_handle(self->t_handles, out_handle);
-		task_unlock_irqrestore(self, flags);
+		task_unlock_irqrestore(self, irq_flags);
+		put_current_task(self);
 		return status;
 	}
 
 	out_slot->h_object = &out_vmo->vo_base;
+	put_current_task(self);
 
 	*out = out_handle;
 	return KERN_OK;
 }
 
+kern_status_t sys_vm_controller_prepare_attach(
+	kern_handle_t ctrl_handle,
+	uint64_t req_id,
+	kern_handle_t *out_vmo)
+{
+	struct task *self = get_current_task();
+	if (!out_vmo || !validate_access_w(self, out_vmo, sizeof *out_vmo)) {
+		return KERN_MEMORY_FAULT;
+	}
+
+	kern_status_t status = KERN_OK;
+	unsigned long flags;
+	task_lock_irqsave(self, &flags);
+
+	struct object *ctrl_obj = NULL;
+	handle_flags_t ctrl_flags = 0;
+	status = task_resolve_handle(self, ctrl_handle, &ctrl_obj, &ctrl_flags);
+	if (status != KERN_OK) {
+		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
+		return status;
+	}
+
+	struct handle *out_slot = NULL;
+	kern_handle_t out_handle = KERN_HANDLE_INVALID;
+	status = handle_table_alloc_handle(
+		self->t_handles,
+		KERN_HANDLE_INVALID,
+		&out_slot,
+		&out_handle);
+	if (status != KERN_OK) {
+		object_unref(ctrl_obj);
+		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
+		return status;
+	}
+
+	struct vm_controller *ctrl = vm_controller_cast(ctrl_obj);
+	task_unlock_irqrestore(self, flags);
+
+	if (!ctrl) {
+		object_unref(ctrl_obj);
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	vm_controller_lock_irqsave(ctrl, &flags);
+
+	struct vm_object *vmo = NULL;
+	status = vm_controller_prepare_attach(ctrl, req_id, &vmo);
+	vm_controller_unlock_irqrestore(ctrl, flags);
+
+	object_unref(ctrl_obj);
+	if (status != KERN_OK) {
+		task_lock_irqsave(self, &flags);
+		handle_table_free_handle(self->t_handles, out_handle);
+		task_unlock_irqrestore(self, flags);
+
+		return status;
+	}
+
+	out_slot->h_object = &vmo->vo_base;
+	put_current_task(self);
+
+	*out_vmo = out_handle;
+
+	return KERN_OK;
+}
+
+kern_status_t sys_vm_controller_finish_attach(
+	kern_handle_t ctrl_handle,
+	uint64_t req_id,
+	equeue_key_t new_key)
+{
+	struct task *self = get_current_task();
+
+	kern_status_t status = KERN_OK;
+	unsigned long flags;
+	task_lock_irqsave(self, &flags);
+
+	struct object *ctrl_obj = NULL;
+	handle_flags_t ctrl_flags = 0;
+	status = task_resolve_handle(self, ctrl_handle, &ctrl_obj, &ctrl_flags);
+	if (status != KERN_OK) {
+		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
+		return status;
+	}
+
+	struct vm_controller *ctrl = vm_controller_cast(ctrl_obj);
+	task_unlock_irqrestore(self, flags);
+	put_current_task(self);
+
+	if (!ctrl) {
+		object_unref(ctrl_obj);
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	vm_controller_lock_irqsave(ctrl, &flags);
+
+	status = vm_controller_finish_attach(ctrl, req_id, new_key);
+	vm_controller_unlock_irqrestore(ctrl, flags);
+
+	object_unref(ctrl_obj);
+	if (status != KERN_OK) {
+		return status;
+	}
+
+	return KERN_OK;
+}
+
 kern_status_t sys_vm_controller_detach_object(
 	kern_handle_t ctrl_handle,
 	kern_handle_t vmo_handle)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	kern_status_t status = KERN_OK;
 	unsigned long flags;
@@ -206,6 +334,7 @@ kern_status_t sys_vm_controller_detach_object(
 	status = task_resolve_handle(self, ctrl_handle, &ctrl_obj, &ctrl_flags);
 	if (status != KERN_OK) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
@@ -213,6 +342,7 @@ kern_status_t sys_vm_controller_detach_object(
 	if (status != KERN_OK) {
 		object_unref(ctrl_obj);
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
@@ -223,6 +353,7 @@ kern_status_t sys_vm_controller_detach_object(
 	if (!ctrl || !vmo) {
 		object_unref(ctrl_obj);
 		object_unref(vmo_obj);
+		put_current_task(self);
 		return KERN_INVALID_ARGUMENT;
 	}
 
@@ -234,6 +365,7 @@ kern_status_t sys_vm_controller_detach_object(
 
 	object_unref(ctrl_obj);
 	object_unref(vmo_obj);
+	put_current_task(self);
 
 	return status;
 }
@@ -246,7 +378,7 @@ kern_status_t sys_vm_controller_supply_pages(
 	off_t src_offset,
 	size_t count)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	kern_status_t status = KERN_OK;
 	unsigned long flags;
@@ -257,6 +389,7 @@ kern_status_t sys_vm_controller_supply_pages(
 	status = task_resolve_handle(self, ctrl_handle, &ctrl_obj, &ctrl_flags);
 	if (status != KERN_OK) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
@@ -264,6 +397,7 @@ kern_status_t sys_vm_controller_supply_pages(
 	if (status != KERN_OK) {
 		object_unref(ctrl_obj);
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
@@ -272,6 +406,7 @@ kern_status_t sys_vm_controller_supply_pages(
 		object_unref(ctrl_obj);
 		object_unref(dst_obj);
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
@@ -284,11 +419,14 @@ kern_status_t sys_vm_controller_supply_pages(
 		object_unref(ctrl_obj);
 		object_unref(dst_obj);
 		object_unref(src_obj);
+		put_current_task(self);
 		return KERN_INVALID_ARGUMENT;
 	}
 
 	vm_controller_lock_irqsave(ctrl, &flags);
 	vm_object_lock_pair(src, dst);
+	equeue_key_t requester_key = dst->vo_key;
+
 	status = vm_controller_supply_pages(
 		ctrl,
 		dst,
@@ -297,11 +435,18 @@ kern_status_t sys_vm_controller_supply_pages(
 		src_offset,
 		count);
 	vm_object_unlock_pair(src, dst);
+	vm_controller_fulfill_requests(
+		ctrl,
+		requester_key,
+		dst_offset,
+		count,
+		status);
 	vm_controller_unlock_irqrestore(ctrl, flags);
 
 	object_unref(ctrl_obj);
 	object_unref(dst_obj);
 	object_unref(src_obj);
+	put_current_task(self);
 
 	return status;
 }

syscall/vm-object.c

@@ -11,13 +11,15 @@ kern_status_t sys_vm_object_create(
 	vm_prot_t prot,
 	kern_handle_t *out_handle)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	if ((name || name_len) && !validate_access_r(self, name, name_len)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
 	if (!validate_access_w(self, out_handle, sizeof *out_handle)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -30,6 +32,7 @@ kern_status_t sys_vm_object_create(
 	kern_status_t status
 		= task_open_handle(self, &obj->vo_base, 0, out_handle);
 	object_unref(&obj->vo_base);
+	put_current_task(self);
 
 	return status;
 }
@@ -41,13 +44,15 @@ kern_status_t sys_vm_object_read(
 	size_t count,
 	size_t *nr_read)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	if (!validate_access_w(self, dst, count)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
 	if (nr_read && !validate_access_w(self, nr_read, sizeof *nr_read)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -55,15 +60,19 @@ kern_status_t sys_vm_object_read(
 	handle_flags_t flags = 0;
 	kern_status_t status = task_resolve_handle(self, object, &obj, &flags);
 	if (status != KERN_OK) {
+		put_current_task(self);
 		return status;
 	}
 
 	struct vm_object *vmo = vm_object_cast(obj);
 	if (!vmo) {
+		put_current_task(self);
 		return KERN_INVALID_ARGUMENT;
 	}
 
-	return vm_object_read(vmo, dst, offset, count, nr_read);
+	status = vm_object_read(vmo, dst, offset, count, nr_read);
+	put_current_task(self);
+	return status;
 }
 
 kern_status_t sys_vm_object_write(
@@ -73,14 +82,16 @@ kern_status_t sys_vm_object_write(
 	size_t count,
 	size_t *nr_written)
 {
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	if (!validate_access_r(self, src, count)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
 	if (nr_written
 	    && !validate_access_w(self, nr_written, sizeof *nr_written)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -88,15 +99,19 @@ kern_status_t sys_vm_object_write(
 	handle_flags_t flags = 0;
 	kern_status_t status = task_resolve_handle(self, object, &obj, &flags);
 	if (status != KERN_OK) {
+		put_current_task(self);
 		return status;
 	}
 
 	struct vm_object *vmo = vm_object_cast(obj);
 	if (!vmo) {
+		put_current_task(self);
 		return KERN_INVALID_ARGUMENT;
 	}
 
-	return vm_object_write(vmo, src, offset, count, nr_written);
+	status = vm_object_write(vmo, src, offset, count, nr_written);
+	put_current_task(self);
+	return status;
 }
 
 kern_status_t sys_vm_object_copy(
@@ -114,10 +129,11 @@ kern_status_t sys_vm_object_copy(
 	       src_offset,
 	       count,
 	       nr_copied);
-	struct task *self = current_task();
+	struct task *self = get_current_task();
 
 	if (nr_copied
 	    && !validate_access_w(self, nr_copied, sizeof *nr_copied)) {
+		put_current_task(self);
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -131,16 +147,19 @@ kern_status_t sys_vm_object_copy(
 	status = task_resolve_handle(self, dst, &dst_obj, &dst_flags);
 	if (status != KERN_OK) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
 	status = task_resolve_handle(self, src, &src_obj, &src_flags);
 	if (status != KERN_OK) {
 		task_unlock_irqrestore(self, flags);
+		put_current_task(self);
 		return status;
 	}
 
 	task_unlock_irqrestore(self, flags);
+	put_current_task(self);
 
 	struct vm_object *dst_vmo = vm_object_cast(dst_obj);
 	struct vm_object *src_vmo = vm_object_cast(src_obj);

vm/address-space.c

@@ -6,6 +6,7 @@
 #include <kernel/printk.h>
 #include <kernel/sched.h>
 #include <kernel/task.h>
+#include <kernel/thread.h>
 #include <kernel/util.h>
 #include <kernel/vm-controller.h>
 #include <kernel/vm-object.h>
@@ -705,6 +706,7 @@ kern_status_t address_space_map(
 	struct vm_object *object,
 	off_t object_offset,
 	size_t length,
+	vm_flags_t flags,
 	vm_prot_t prot,
 	virt_addr_t *out)
 {
@@ -762,6 +764,7 @@ kern_status_t address_space_map(
 	area->vma_space = root;
 	area->vma_object = object;
 	area->vma_prot = prot;
+	area->vma_flags = flags;
 	area->vma_object_offset = object_offset;
 	area->vma_base = map_address;
 	area->vma_limit = map_address + length - 1;
@@ -1206,6 +1209,267 @@ kern_status_t address_space_release(
 	return status;
 }
 
+static struct vm_area *area_duplicate(struct vm_area *area)
+{
+	struct vm_area *out = vm_cache_alloc(&vm_area_cache, VM_NORMAL);
+	if (!out) {
+		return NULL;
+	}
+
+	out->vma_prot = area->vma_prot;
+	out->vma_flags = area->vma_flags;
+	out->vma_object_offset = area->vma_object_offset;
+	out->vma_base = area->vma_base;
+	out->vma_limit = area->vma_limit;
+
+	return out;
+}
+
+static kern_status_t update_area_pte_cow(
+	struct address_space *src,
+	struct address_space *dest,
+	struct vm_area *area)
+{
+	if (!area->vma_object) {
+		return KERN_OK;
+	}
+
+	for (virt_addr_t i = area->vma_base; i < area->vma_limit;
+	     i += VM_PAGE_SIZE) {
+		off_t pg_offset = i - area->vma_base + area->vma_object_offset;
+		struct vm_page *pg = vm_object_get_page(
+			area->vma_object,
+			pg_offset,
+			0,
+			NULL);
+		vm_prot_t temp_prot = area->vma_prot;
+		temp_prot &= ~VM_PROT_WRITE;
+
+		if (pg) {
+			pmap_add(
+				src->s_pmap,
+				i,
+				vm_page_get_pfn(pg),
+				temp_prot,
+				PMAP_NORMAL);
+			pmap_add(
+				dest->s_pmap,
+				i,
+				vm_page_get_pfn(pg),
+				temp_prot,
+				PMAP_NORMAL);
+			tracek("PTE %zx -> %zx [%x]",
+			       i,
+			       vm_page_get_paddr(pg),
+			       temp_prot);
+		}
+	}
+
+	return KERN_OK;
+}
+
+static kern_status_t prepare_duplicate_areas(
+	struct address_space *src,
+	struct address_space *dest)
+{
+	struct btree_node *cur_node = btree_first(&src->s_mappings);
+	while (cur_node) {
+		struct vm_area *tmp_area
+			= BTREE_CONTAINER(struct vm_area, vma_node, cur_node);
+		if (!tmp_area->vma_object) {
+			cur_node = btree_next(cur_node);
+			continue;
+		}
+
+		struct vm_object *src_vmo = tmp_area->vma_object;
+		vm_object_lock(src_vmo);
+
+		struct vm_object *dest_vmo_link = NULL;
+		struct vm_object *dest_vmo_cow = NULL;
+		struct queue_entry *cur_entry
+			= queue_first(&src_vmo->vo_mappings);
+
+		while (cur_entry) {
+			struct vm_area *src_area = QUEUE_CONTAINER(
+				struct vm_area,
+				vma_object_entry,
+				cur_entry);
+			if (src_area->vma_space != src) {
+				cur_entry = queue_next(cur_entry);
+				continue;
+			}
+
+			struct vm_area *dest_area = get_entry(
+				&dest->s_mappings,
+				src_area->vma_base,
+				GET_ENTRY_EXACT);
+			if (!dest_area) {
+				/* this shouldn't happen. the duplicate vm_areas
+				 * were already created by
+				 * address_space_duplicate */
+				panic("create_duplicate_vmo: corresponding "
+				      "vm_area is missing");
+			}
+
+			if (dest_area->vma_object) {
+				cur_entry = queue_next(cur_entry);
+				continue;
+			}
+
+			struct vm_object *dest_vmo = NULL;
+			if (src_area->vma_flags & VM_SHARED) {
+				dest_vmo = dest_vmo_link;
+			} else {
+				dest_vmo = dest_vmo_cow;
+			}
+
+			if (!dest_vmo) {
+				tracek("[%zx-%zx %x] creating COW duplicate of "
+				       "vmo %p",
+				       src_area->vma_base,
+				       src_area->vma_limit,
+				       src_area->vma_prot,
+				       src_vmo);
+
+				if (src_area->vma_flags & VM_SHARED) {
+					dest_vmo_link = src_vmo;
+					object_ref(&dest_vmo_link->vo_base);
+
+					dest_vmo = dest_vmo_link;
+				} else {
+					dest_vmo_cow = vm_object_duplicate_cow(
+						src_vmo);
+					dest_vmo = dest_vmo_cow;
+				}
+
+				tracek("[%zx-%zx %x] created COW duplicate of "
+				       "vmo %p -> %p",
+				       src_area->vma_base,
+				       src_area->vma_limit,
+				       src_area->vma_prot,
+				       src_vmo,
+				       dest_vmo);
+			}
+
+			object_ref(&dest_vmo->vo_base);
+			dest_area->vma_object = dest_vmo;
+			update_area_pte_cow(src, dest, src_area);
+			cur_entry = queue_next(cur_entry);
+		}
+
+		if (dest_vmo_link) {
+			object_unref(&dest_vmo_link->vo_base);
+		}
+
+		if (dest_vmo_cow) {
+			object_unref(&dest_vmo_cow->vo_base);
+		}
+
+		vm_object_unlock(src_vmo);
+
+		cur_node = btree_next(cur_node);
+	}
+
+	return KERN_OK;
+}
+
+kern_status_t address_space_duplicate(
+	struct address_space *dest,
+	struct address_space *src)
+{
+	// address_space_dump(src);
+	/* clang-format off
+	 * strategy for COW address space duplication:
+	 * 1. duplicate each vm_area in the address space
+	 *    a. all details except for the vm_object pointer are copied.
+	 *    b. create a duplicate vm_object, where all the details are the
+	 *       same, but don't copy the pages or vm_page pointers.
+	 *    c. if the vm_object is attached to a vm_controller, don't inform
+	 *       the controller yet.
+	 *    d. for both the original and duplicate vm_area, duplicate the PTE
+	 *       entries, changing all of them to read-only. increment the
+	 *       p_cow_ref counters for all committed vm_pages.
+	 *    e. use the vm_object's vm_area list, and the vm_area's vma_space
+	 *       pointer, to ensure that only one duplicate is created for each
+	 *       unique vm-object referenced by an address-space.
+	 * 2. when a page fault occurs:
+	 *    a. find the relevant vm_area as normal.
+	 *    b. if the faulted page is present and the vm_area's prot flags
+	 *       should allow the access, a COW is required.
+	 *    c. if the relevant page is already present in the vm_area's
+	 *       vm_object, this is the original vm_area. otherwise, this is the
+	 *       clone vm_area.
+	 *    d. if this is the source vm_area:
+	 *       i. decrement p_cow_ref in the page. if it is 0, skip to step v.
+	 *       ii. remove the relevant page from the vm_area
+	 *       iii. allocate a new page and copy the data.
+	 *       iv. add the new page to the vm_object at the same offset.
+	 *       v. change the PTE entry to the proper protection flags.
+	 *       vi. resume the faulting task.
+	 *    e. otherwise, if this is the clone vm_area:
+	 *       i. if the vm-object has a controller, send
+	 *          PAGE_REQUEST_DUPLICATE to it. the controller needs to
+	 *          prepare itself to receive page requests from this vm-object,
+	 *          which includes priving it an equeue_key_t.
+	 *       i. use the physical address stored in the PTE to find the
+	 *          relevant vm_page.
+	 *       ii. decrement p_cow_ref in the page.
+	 *       iii. if p_cow_ref is > 0, allocate a new page and copy the data.
+	 *            otherwise, use the existing page as-is.
+	 *       iv. add the page from step iii to the vm_object at the correct
+	 *           offset.
+	 *       v. change the PTE entry to the proper protection flags.
+	 *       vi. resume the faulting task.
+	 * 3. when destroying a vm_area:
+	 *    a. for pages already present in a vm-object, handle as normal.
+	 *    b. for pages not present in a vm-object, but for which a valid PTE
+	 *       exists, use the PTE physical address to find the vm_page.
+	 *    c. decrement p_cow_ref in this page.
+	 *    d. if p_cow_ref == 0, de-allocate the page.
+	 * clang-format on
+	 */
+	struct btree_node *cur = btree_first(&src->s_mappings);
+	while (cur) {
+		struct vm_area *src_area
+			= BTREE_CONTAINER(struct vm_area, vma_node, cur);
+		struct vm_area *dest_area = area_duplicate(src_area);
+		tracek("duplicated vm_area [%zx-%zx] %p -> %p",
+		       src_area->vma_base,
+		       src_area->vma_limit,
+		       src_area,
+		       dest_area);
+		/* TODO handle OOM */
+		put_entry(&dest->s_mappings, dest_area);
+
+		cur = btree_next(cur);
+	}
+
+	cur = btree_first(&src->s_reserved);
+	while (cur) {
+		struct vm_area *src_area
+			= BTREE_CONTAINER(struct vm_area, vma_node, cur);
+		struct vm_area *dest_area = area_duplicate(src_area);
+		tracek("duplicated vm_area [r] [%zx-%zx] %p -> %p",
+		       src_area->vma_base,
+		       src_area->vma_limit,
+		       src_area,
+		       dest_area);
+		/* TODO handle OOM */
+		put_entry(&dest->s_reserved, dest_area);
+
+		cur = btree_next(cur);
+	}
+
+	tracek("preparing duplicate areas");
+	kern_status_t status = prepare_duplicate_areas(src, dest);
+	tracek("prepared duplicate areas");
+	if (status != KERN_OK) {
+		return status;
+	}
+
+	return KERN_OK;
+}
+
 bool address_space_validate_access(
 	struct address_space *region,
 	virt_addr_t ptr,
@@ -1261,13 +1525,26 @@ static kern_status_t request_missing_page(
 	/* here:
 	 * `region` is locked.
 	 * `object` is unlocked.
-	 * `irq_flags` must be restored when `region` is unlocked.
-	 * the relevant page in `object` may or may not be committed.
-	 * if it isn't, it needs to be requested.
+	 * `irq_flags` must be restored when `region` is
+	 * unlocked.
+	 * the relevant page in `object` may or may
+	 * not be committed. if it isn't, it needs to be
+	 * requested.
 	 */
 	vm_object_lock(object);
 	address_space_unlock(region);
 
+	kern_status_t status = KERN_OK;
+
+	if (object->vo_flags & VMO_LAZY_ATTACH) {
+		status = vm_object_attach_cow(object, irq_flags);
+	}
+
+	if (status != KERN_OK) {
+		vm_object_unlock_irqrestore(object, *irq_flags);
+		return status;
+	}
+
 	struct vm_page *pg = vm_object_get_page(
 		object,
 		object_offset,
@@ -1281,7 +1558,7 @@ static kern_status_t request_missing_page(
 
 	/* now: `region` is unlocked, and `object` is locked */
 
-	kern_status_t status = pmap_add(
+	status = pmap_add(
 		region->s_pmap,
 		addr,
 		vm_page_get_pfn(pg),
@@ -1292,7 +1569,126 @@ static kern_status_t request_missing_page(
 	return status;
 }
 
-/* this function must be called with `region` locked */
+/* handle a write to a page that is present and /should be/ writeable, but is
+ * currently read-only due to COW. */
+static kern_status_t handle_cow_access(
+	struct address_space *region,
+	virt_addr_t addr,
+	enum pmap_fault_flags flags,
+	unsigned long *irq_flags)
+{
+	struct vm_area *area
+		= get_entry(&region->s_mappings, addr, GET_ENTRY_EXACT);
+	if (!area || !area->vma_object) {
+		/* no mapping exists (this shouldn't happen) */
+		address_space_unlock_irqrestore(region, *irq_flags);
+		return KERN_NO_ENTRY;
+	}
+
+	if ((area->vma_prot & (VM_PROT_WRITE | VM_PROT_USER))
+	    != (VM_PROT_WRITE | VM_PROT_USER)) {
+		/* access denied */
+		address_space_unlock_irqrestore(region, *irq_flags);
+		return KERN_ACCESS_DENIED;
+	}
+
+	tracek("cow access %zx", addr);
+
+	off_t object_offset = addr - area->vma_base + area->vma_object_offset;
+	vm_object_lock(area->vma_object);
+	struct vm_page *pg
+		= vm_object_get_page(area->vma_object, object_offset, 0, NULL);
+
+	bool add_page = false;
+	if (!pg) {
+		/* the page hasn't been added to the vmo yet. */
+		pfn_t pfn;
+		vm_prot_t prot;
+		kern_status_t status
+			= pmap_get(region->s_pmap, addr, &pfn, &prot);
+		if (status != KERN_OK) {
+			vm_object_unlock(area->vma_object);
+			address_space_unlock_irqrestore(region, *irq_flags);
+			return status;
+		}
+
+		add_page = true;
+		pg = vm_page_get(pfn * VM_PAGE_SIZE);
+		tracek("recovered page %zx (%p, %u) from the aether",
+		       vm_page_get_paddr(pg),
+		       pg,
+		       pg->p_cow_ref);
+
+		if (!pg) {
+			/* still can't find the page */
+			panic("cannot resolve cow-reference to page %zx",
+			      pfn * VM_PAGE_SIZE);
+		}
+
+		if (!pg->p_cow_ref) {
+			/* still can't find the page */
+			panic("cow-reference page %zx cow-ref=0",
+			      pfn * VM_PAGE_SIZE);
+		}
+	}
+
+	atomic_t cow_ref = atomic_sub_fetch(&pg->p_cow_ref, 1);
+	tracek("decrement cow-ref page %zx -> now %u",
+	       vm_page_get_paddr(pg),
+	       pg->p_cow_ref);
+	if (cow_ref > 0) {
+		/* another task is (or will be) referencing this page. create
+		 * a copy */
+		struct vm_page *dup = vm_page_alloc(VM_PAGE_4K, VM_NORMAL);
+		if (!dup) {
+			vm_object_unlock(area->vma_object);
+			address_space_unlock_irqrestore(region, *irq_flags);
+			return KERN_NO_MEMORY;
+		}
+
+		void *src = vm_page_get_vaddr(pg);
+		void *dest = vm_page_get_vaddr(dup);
+		memcpy(dest, src, vm_page_get_size_bytes(dup));
+
+		if (!add_page) {
+			btree_delete(&area->vma_object->vo_pages, &pg->p_bnode);
+		}
+
+		vm_object_put_page(area->vma_object, object_offset, dup);
+		tracek("splitting cow page %zx -> %zx",
+		       vm_page_get_paddr(pg),
+		       vm_page_get_paddr(dup));
+		tracek("put page %zx into area [%zx-%zx] at %zx",
+		       vm_page_get_paddr(dup),
+		       area->vma_base,
+		       area->vma_limit,
+		       object_offset);
+		pg = dup;
+	} else if (add_page) {
+		/* we are the last task referencing this page. add it to our
+		 * vmo and continue */
+		tracek("claimed cow page %zx", vm_page_get_paddr(pg));
+		vm_object_put_page(area->vma_object, object_offset, pg);
+	} else {
+		/* we are the last task referencing this page, and it is already
+		 * in our vmo. */
+		tracek("reclaimed cow page %zx", vm_page_get_paddr(pg));
+	}
+
+	kern_status_t status = pmap_add(
+		region->s_pmap,
+		addr,
+		vm_page_get_pfn(pg),
+		area->vma_prot,
+		PMAP_NORMAL);
+
+	vm_object_unlock(area->vma_object);
+	address_space_unlock_irqrestore(region, *irq_flags);
+
+	return status;
+}
+
+/* this function must be called with `region` unlocked */
 kern_status_t address_space_demand_map(
 	struct address_space *region,
 	virt_addr_t addr,
@@ -1306,6 +1702,13 @@ kern_status_t address_space_demand_map(
 	unsigned long irq_flags;
 	address_space_lock_irqsave(region, &irq_flags);
 
+	const enum pmap_fault_flags cow_flags
+		= PMAP_FAULT_WRITE | PMAP_FAULT_PRESENT;
+
+	if ((flags & cow_flags) == cow_flags) {
+		return handle_cow_access(region, addr, flags, &irq_flags);
+	}
+
 	struct vm_area *area
 		= get_entry(&region->s_mappings, addr, GET_ENTRY_EXACT);
 	if (!area || !area->vma_object) {
@@ -1348,6 +1751,8 @@ kern_status_t address_space_demand_map(
 #endif
 
 	if (!pg) {
+		vm_object_unlock(area->vma_object);
+		address_space_unlock_irqrestore(region, irq_flags);
 		return KERN_FATAL_ERROR;
 	}
 

vm/vm-controller.c

@@ -9,13 +9,12 @@
 #define VM_CONTROLLER_CAST(p)                                                  \
 	OBJECT_C_CAST(struct vm_controller, vc_base, &vm_controller_type, p)
 
-BTREE_DEFINE_SIMPLE_INSERT(struct vm_object, vo_ctrl_node, vo_key, put_object)
 BTREE_DEFINE_SIMPLE_GET(
-	struct vm_object,
-	equeue_key_t,
-	vo_ctrl_node,
-	vo_key,
-	get_object)
+	struct vm_request,
+	uint64_t,
+	req_node,
+	req_id,
+	get_request)
 
 static struct object_type vm_controller_type = {
 	.ob_name = "vm-controller",
@@ -23,14 +22,14 @@ static struct object_type vm_controller_type = {
 	.ob_header_offset = offsetof(struct vm_controller, vc_base),
 };
 
-static struct vm_cache page_request_cache = {
-	.c_name = "page-request",
-	.c_obj_size = sizeof(struct page_request),
+static struct vm_cache vm_request_cache = {
+	.c_name = "vm-request",
+	.c_obj_size = sizeof(struct vm_request),
 };
 
 kern_status_t vm_controller_type_init(void)
 {
-	vm_cache_init(&page_request_cache);
+	vm_cache_init(&vm_request_cache);
 	return object_type_register(&vm_controller_type);
 }
 
@@ -51,19 +50,19 @@ struct vm_controller *vm_controller_create(void)
 	return ctrl;
 }
 
-static struct page_request *get_next_request(struct vm_controller *ctrl)
+static struct vm_request *get_next_request(struct vm_controller *ctrl)
 {
 	struct btree_node *cur = btree_first(&ctrl->vc_requests);
 	while (cur) {
-		struct page_request *req
-			= BTREE_CONTAINER(struct page_request, req_node, cur);
+		struct vm_request *req
+			= BTREE_CONTAINER(struct vm_request, req_node, cur);
 		spin_lock(&req->req_lock);
 		switch (req->req_status) {
-		case PAGE_REQUEST_PENDING:
-			req->req_status = PAGE_REQUEST_IN_PROGRESS;
+		case VM_REQUEST_PENDING:
+			req->req_status = VM_REQUEST_IN_PROGRESS;
 			ctrl->vc_requests_waiting--;
 			return req;
-		case PAGE_REQUEST_ASYNC:
+		case VM_REQUEST_ASYNC:
 			btree_delete(&ctrl->vc_requests, &req->req_node);
 			ctrl->vc_requests_waiting--;
 			return req;
@@ -78,7 +77,7 @@ static struct page_request *get_next_request(struct vm_controller *ctrl)
 	return NULL;
 }
 
-static kern_status_t try_enqueue(struct btree *tree, struct page_request *req)
+static kern_status_t try_enqueue(struct btree *tree, struct vm_request *req)
 {
 	if (!tree->b_root) {
 		tree->b_root = &req->req_node;
@@ -88,8 +87,8 @@ static kern_status_t try_enqueue(struct btree *tree, struct page_request *req)
 
 	struct btree_node *cur = tree->b_root;
 	while (1) {
-		struct page_request *cur_node
-			= BTREE_CONTAINER(struct page_request, req_node, cur);
+		struct vm_request *cur_node
+			= BTREE_CONTAINER(struct vm_request, req_node, cur);
 		struct btree_node *next = NULL;
 
 		if (req->req_id > cur_node->req_id) {
@@ -119,7 +118,7 @@ static kern_status_t try_enqueue(struct btree *tree, struct page_request *req)
 
 static kern_status_t send_request_async(
 	struct vm_controller *ctrl,
-	struct page_request *req)
+	struct vm_request *req)
 {
 	fill_random(&req->req_id, sizeof req->req_id);
 	while (!try_enqueue(&ctrl->vc_requests, req)) {
@@ -136,9 +135,9 @@ static kern_status_t send_request_async(
 
 kern_status_t vm_controller_recv(
 	struct vm_controller *ctrl,
-	equeue_packet_page_request_t *out)
+	equeue_packet_vm_request_t *out)
 {
-	struct page_request *req = NULL;
+	struct vm_request *req = NULL;
 
 	req = get_next_request(ctrl);
 	if (!req) {
@@ -151,15 +150,30 @@ kern_status_t vm_controller_recv(
 			VM_CONTROLLER_SIGNAL_REQUEST_RECEIVED);
 	}
 
-	out->req_vmo = req->req_object;
+	vm_object_lock(req->req_object);
+	out->req_id = req->req_id;
+	out->req_vmo = req->req_object->vo_key;
 	out->req_type = req->req_type;
+
+	switch (req->req_type) {
+	case VM_REQUEST_READ:
+	case VM_REQUEST_DIRTY:
 		out->req_offset = req->req_offset;
 		out->req_length = req->req_length;
+		break;
+	case VM_REQUEST_ATTACH:
+		out->req_src_vmo = req->req_object->vo_key;
+		break;
+	default:
+		break;
+	}
 
+	vm_object_unlock(req->req_object);
 	spin_unlock(&req->req_lock);
 
-	if (req->req_status == PAGE_REQUEST_ASYNC) {
-		vm_cache_free(&page_request_cache, req);
+	if (req->req_status == VM_REQUEST_ASYNC) {
+		put_current_thread(req->req_sender);
+		vm_cache_free(&vm_request_cache, req);
 	}
 
 	return KERN_OK;
@@ -207,6 +221,52 @@ kern_status_t vm_controller_create_object(
 	return KERN_OK;
 }
 
+kern_status_t vm_controller_prepare_attach(
+	struct vm_controller *ctrl,
+	uint64_t req_id,
+	struct vm_object **out_vmo)
+{
+	struct vm_request *req = get_request(&ctrl->vc_requests, req_id);
+	if (!req) {
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	spin_lock(&req->req_lock);
+	req->req_status = VM_REQUEST_IN_PROGRESS;
+	*out_vmo = req->req_object;
+	spin_unlock(&req->req_lock);
+
+	return KERN_OK;
+}
+
+kern_status_t vm_controller_finish_attach(
+	struct vm_controller *ctrl,
+	uint64_t req_id,
+	equeue_key_t new_key)
+{
+	struct vm_request *req = get_request(&ctrl->vc_requests, req_id);
+	if (!req) {
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	spin_lock(&req->req_lock);
+	struct vm_object *vmo = req->req_object;
+	spin_unlock(&req->req_lock);
+
+	vm_object_lock(vmo);
+	vmo->vo_key = new_key;
+	vmo->vo_flags &= ~VMO_LAZY_ATTACH;
+	vm_object_unlock(vmo);
+
+	spin_lock(&req->req_lock);
+	req->req_status = VM_REQUEST_COMPLETE;
+	req->req_result = KERN_OK;
+	thread_awaken(req->req_sender);
+	spin_unlock(&req->req_lock);
+
+	return KERN_OK;
+}
+
 kern_status_t vm_controller_detach_object(
 	struct vm_controller *ctrl,
 	struct vm_object *vmo)
@@ -215,16 +275,23 @@ kern_status_t vm_controller_detach_object(
 		return KERN_INVALID_ARGUMENT;
 	}
 
-	struct page_request *req
-		= vm_cache_alloc(&page_request_cache, VM_NORMAL);
-	req->req_type = PAGE_REQUEST_DETACH;
-	req->req_status = PAGE_REQUEST_ASYNC;
-	req->req_object = vmo->vo_key;
-	req->req_sender = current_thread();
+	if (vmo->vo_flags & VMO_LAZY_ATTACH) {
+		/* this vmo isn't actually attached to this controller yet.
+		 * this can happen if a controller-attached vmo was duplicated
+		 * via copy-on-write, and the duplicate vmo has not yet been
+		 * accessed. */
+		vmo->vo_ctrl = NULL;
+		return KERN_OK;
+	}
+
+	struct vm_request *req = vm_cache_alloc(&vm_request_cache, VM_NORMAL);
+	req->req_type = VM_REQUEST_DETACH;
+	req->req_status = VM_REQUEST_ASYNC;
+	req->req_object = vmo;
+	req->req_sender = get_current_thread();
 	send_request_async(ctrl, req);
 
 	vmo->vo_ctrl = NULL;
-	vmo->vo_key = 0;
 	object_unref(&ctrl->vc_base);
 
 	return KERN_OK;
@@ -232,16 +299,16 @@ kern_status_t vm_controller_detach_object(
 
 static void wait_for_reply(
 	struct vm_controller *ctrl,
-	struct page_request *req,
+	struct vm_request *req,
 	unsigned long *lock_flags)
 {
 	struct wait_item waiter;
-	struct thread *self = current_thread();
+	struct thread *self = get_current_thread();
 
 	wait_item_init(&waiter, self);
 	for (;;) {
 		self->tr_state = THREAD_SLEEPING;
-		if (req->req_status == PAGE_REQUEST_COMPLETE) {
+		if (req->req_status == VM_REQUEST_COMPLETE) {
 			break;
 		}
 
@@ -251,9 +318,10 @@ static void wait_for_reply(
 	}
 
 	self->tr_state = THREAD_READY;
+	put_current_thread(self);
 }
 
-static void fulfill_requests(
+void vm_controller_fulfill_requests(
 	struct vm_controller *ctrl,
 	equeue_key_t object,
 	off_t offset,
@@ -263,8 +331,8 @@ static void fulfill_requests(
 	off_t limit = offset + length - 1;
 	struct btree_node *cur = btree_first(&ctrl->vc_requests);
 	while (cur) {
-		struct page_request *req
-			= BTREE_CONTAINER(struct page_request, req_node, cur);
+		struct vm_request *req
+			= BTREE_CONTAINER(struct vm_request, req_node, cur);
 		spin_lock(&req->req_lock);
 		bool match = false;
 		off_t req_base = req->req_offset;
@@ -276,12 +344,14 @@ static void fulfill_requests(
 			match = true;
 		}
 
-		if (req->req_object != object) {
+		vm_object_lock(req->req_object);
+		if (req->req_object->vo_key != object) {
 			match = false;
 		}
+		vm_object_unlock(req->req_object);
 
 		if (match) {
-			req->req_status = PAGE_REQUEST_COMPLETE;
+			req->req_status = VM_REQUEST_COMPLETE;
 			req->req_result = result;
 			thread_awaken(req->req_sender);
 		}
@@ -314,14 +384,13 @@ kern_status_t vm_controller_supply_pages(
 		src_offset,
 		count,
 		NULL);
-	fulfill_requests(ctrl, dst->vo_key, dst_offset, count, status);
 
 	return status;
 }
 
 kern_status_t vm_controller_send_request(
 	struct vm_controller *ctrl,
-	struct page_request *req,
+	struct vm_request *req,
 	unsigned long *irq_flags)
 {
 	fill_random(&req->req_id, sizeof req->req_id);

vm/vm-object.c

@@ -1,4 +1,5 @@
 #include <kernel/address-space.h>
+#include <kernel/panic.h>
 #include <kernel/printk.h>
 #include <kernel/sched.h>
 #include <kernel/util.h>
@@ -31,9 +32,10 @@ static kern_status_t vm_object_cleanup(struct object *obj)
 
 	if (vmo->vo_ctrl) {
 		unsigned long flags;
-		vm_controller_lock_irqsave(vmo->vo_ctrl, &flags);
+		struct vm_controller *ctrl = vmo->vo_ctrl;
+		vm_controller_lock_irqsave(ctrl, &flags);
 		vm_controller_detach_object(vmo->vo_ctrl, vmo);
-		vm_controller_unlock_irqrestore(vmo->vo_ctrl, flags);
+		vm_controller_unlock_irqrestore(ctrl, flags);
 	}
 
 	return KERN_OK;
@@ -276,6 +278,75 @@ extern struct vm_object *vm_object_create_in_place(
 	return vmo;
 }
 
+struct vm_object *vm_object_duplicate_cow(struct vm_object *vmo)
+{
+	struct object *obj = object_create(&vm_object_type);
+	if (!obj) {
+		return NULL;
+	}
+
+	struct vm_object *out = VM_OBJECT_CAST(obj);
+
+	memcpy(out->vo_name, vmo->vo_name, sizeof out->vo_name);
+	out->vo_flags = vmo->vo_flags | VMO_LAZY_ATTACH;
+	out->vo_ctrl = vmo->vo_ctrl;
+	out->vo_key = vmo->vo_key;
+	out->vo_prot = vmo->vo_prot;
+	out->vo_size = vmo->vo_size;
+	memcpy(out->vo_name, vmo->vo_name, sizeof vmo->vo_name);
+
+	struct btree_node *cur = btree_first(&vmo->vo_pages);
+	while (cur) {
+		struct vm_page *pg
+			= BTREE_CONTAINER(struct vm_page, p_bnode, cur);
+		int x = atomic_fetch_add(&pg->p_cow_ref, 1);
+		if (x == 0) {
+			/* this is the first cow-reference for this page, so
+			 * the address-space it belongs to will need one too */
+			atomic_fetch_add(&pg->p_cow_ref, 1);
+		}
+
+		tracek("convert page %zx to cow %u",
+		       vm_page_get_paddr(pg),
+		       pg->p_cow_ref);
+
+		cur = btree_next(cur);
+	}
+
+	return out;
+}
+
+kern_status_t vm_object_attach_cow(
+	struct vm_object *vmo,
+	unsigned long *irq_flags)
+{
+	struct vm_controller *ctrl = vmo->vo_ctrl;
+	struct vm_request req = {0};
+	req.req_status = VM_REQUEST_PENDING;
+	req.req_type = VM_REQUEST_ATTACH;
+	req.req_length = vm_page_order_to_bytes(VM_PAGE_4K);
+	req.req_sender = get_current_thread();
+
+	object_ref(&vmo->vo_base);
+	req.req_object = vmo;
+
+	vm_object_unlock_irqrestore(vmo, *irq_flags);
+	vm_controller_lock_irqsave(ctrl, irq_flags);
+
+	spin_lock(&req.req_lock);
+
+	kern_status_t status
+		= vm_controller_send_request(ctrl, &req, irq_flags);
+
+	put_current_thread(req.req_sender);
+	spin_unlock(&req.req_lock);
+	vm_controller_unlock_irqrestore(ctrl, *irq_flags);
+	object_unref(&vmo->vo_base);
+	vm_object_lock_irqsave(vmo, irq_flags);
+
+	return status;
+}
+
 static struct vm_page *alloc_page(struct vm_object *vo, off_t offset)
 {
 	struct vm_page *page = NULL;
@@ -341,6 +412,54 @@ static struct vm_page *alloc_page(struct vm_object *vo, off_t offset)
 	return NULL;
 }
 
+kern_status_t vm_object_put_page(
+	struct vm_object *vo,
+	off_t offset,
+	struct vm_page *pg)
+{
+	struct btree_node *cur = vo->vo_pages.b_root;
+	if (!vo->vo_pages.b_root) {
+		vo->vo_pages.b_root = &pg->p_bnode;
+		return KERN_OK;
+	}
+
+	while (cur) {
+		struct vm_page *page
+			= BTREE_CONTAINER(struct vm_page, p_bnode, cur);
+		struct btree_node *next = NULL;
+
+		off_t base = page->p_vmo_offset;
+		off_t limit = base + vm_page_get_size_bytes(page);
+		if (offset < base) {
+			next = btree_left(cur);
+		} else if (offset >= limit) {
+			next = btree_right(cur);
+		} else {
+			panic("vm_object_put_page: page already exists");
+			return KERN_NAME_EXISTS;
+		}
+
+		if (next) {
+			cur = next;
+			continue;
+		}
+
+		pg->p_vmo_offset = offset;
+
+		if (offset < base) {
+			btree_put_left(cur, &pg->p_bnode);
+		} else {
+			btree_put_right(cur, &pg->p_bnode);
+		}
+
+		btree_insert_fixup(&vo->vo_pages, &pg->p_bnode);
+
+		return KERN_OK;
+	}
+
+	return KERN_FATAL_ERROR;
+}
+
 static struct vm_page *get_page(struct vm_object *vo, off_t offset)
 {
 	struct btree_node *cur = vo->vo_pages.b_root;
@@ -371,15 +490,15 @@ static kern_status_t request_page(
 	unsigned long *irq_flags)
 {
 	struct vm_controller *ctrl = vo->vo_ctrl;
-	struct page_request req = {0};
-	req.req_status = PAGE_REQUEST_PENDING;
-	req.req_type = PAGE_REQUEST_READ;
+	struct vm_request req = {0};
+	req.req_status = VM_REQUEST_PENDING;
+	req.req_type = VM_REQUEST_READ;
 	req.req_offset = offset;
 	req.req_length = vm_page_order_to_bytes(VM_PAGE_4K);
-	req.req_sender = current_thread();
+	req.req_sender = get_current_thread();
 
 	object_ref(&vo->vo_base);
-	req.req_object = vo->vo_key;
+	req.req_object = vo;
 
 	vm_object_unlock_irqrestore(vo, *irq_flags);
 	vm_controller_lock_irqsave(ctrl, irq_flags);
@@ -388,6 +507,7 @@ static kern_status_t request_page(
 	kern_status_t status
 		= vm_controller_send_request(ctrl, &req, irq_flags);
 
+	put_current_thread(req.req_sender);
 	spin_unlock(&req.req_lock);
 	vm_controller_unlock_irqrestore(ctrl, *irq_flags);
 	object_unref(&vo->vo_base);
@@ -412,7 +532,7 @@ struct vm_page *vm_object_get_page(
 		return pg;
 	}
 
-	if (!vo->vo_ctrl) {
+	if (!vo->vo_ctrl || !(flags & VMO_REQUEST_MISSING_PAGE)) {
 		return NULL;
 	}