Added memory management using paging

This was quite a roller-coaster and most things are slower now, but it works and is way more secure. I still need to implement things like shared memory for the WM/GUI system but other than that everything is supported.

6 files changed, 663 insertions, 67 deletions

diff --git a/kernel/features/fs.c b/kernel/features/fs.c
index 687d7ad..c8ad317 100644
--- a/kernel/features/fs.c
+++ b/kernel/features/fs.c
@@ -62,13 +62,15 @@ static struct mount_info *vfs_recursive_find(char *path)
 
 static struct mount_info *vfs_find_mount_info(const char *path)
 {
-	assert(path[0] == '/');
+	if (path[0] != '/')
+		return NULL;
 	return vfs_recursive_find(strdup(path));
 }
 
 struct device *vfs_find_dev(const char *path)
 {
-	assert(path[0] == '/');
+	if (path[0] != '/')
+		return NULL;
 	struct mount_info *m = vfs_find_mount_info(path);
 	if (m->dev->vfs->type == VFS_DEVFS) // TODO: ?
 		return device_get_by_name(path + strlen(m->path) + 1);
@@ -120,7 +122,7 @@ s32 vfs_mount(struct device *dev, const char *path)
 
 s32 vfs_read(const char *path, void *buf, u32 offset, u32 count)
 {
-	/* printf("%s READ: %s\n", proc_current()->name, path); */
+	/* printf("%s READ: %s\n", proc_current() ? proc_current()->name : "Unknown", path); */
 	if (!count)
 		return 0;
 
@@ -131,7 +133,8 @@ s32 vfs_read(const char *path, void *buf, u32 offset, u32 count)
 		path++;
 
 	struct mount_info *m = vfs_find_mount_info(path);
-	assert(m && m->dev && m->dev->vfs && m->dev->vfs->read && m->dev->vfs->perm);
+	if (!(m && m->dev && m->dev->vfs && m->dev->vfs->read && m->dev->vfs->perm))
+		return -1;
 
 	u32 len = strlen(m->path);
 	if (len > 1)
@@ -145,7 +148,7 @@ s32 vfs_read(const char *path, void *buf, u32 offset, u32 count)
 
 s32 vfs_write(const char *path, void *buf, u32 offset, u32 count)
 {
-	/* printf("%s WRITE: %s\n", proc_current()->name, path); */
+	/* printf("%s WRITE: %s\n", proc_current() ? proc_current()->name : "Unknown", path); */
 	if (!count)
 		return 0;
 
@@ -156,7 +159,8 @@ s32 vfs_write(const char *path, void *buf, u32 offset, u32 count)
 		path++;
 
 	struct mount_info *m = vfs_find_mount_info(path);
-	assert(m && m->dev && m->dev->vfs && m->dev->vfs->write && m->dev->vfs->perm);
+	if (!(m && m->dev && m->dev->vfs && m->dev->vfs->write && m->dev->vfs->perm))
+		return -1;
 
 	u32 len = strlen(m->path);
 	if (len > 1)
@@ -168,6 +172,25 @@ s32 vfs_write(const char *path, void *buf, u32 offset, u32 count)
 	return m->dev->vfs->write(path, buf, offset, count, m->dev);
 }
 
+s32 vfs_ioctl(const char *path, u32 request, void *arg1, void *arg2, void *arg3)
+{
+	while (*path == ' ')
+		path++;
+
+	struct mount_info *m = vfs_find_mount_info(path);
+	if (!(m && m->dev && m->dev->vfs && m->dev->vfs->ioctl && m->dev->vfs->perm))
+		return -1;
+
+	u32 len = strlen(m->path);
+	if (len > 1)
+		path += len;
+
+	if (!m->dev->vfs->perm(path, VFS_WRITE, m->dev) && !proc_super())
+		return -1;
+
+	return m->dev->vfs->ioctl(path, request, arg1, arg2, arg3, m->dev);
+}
+
 s32 vfs_stat(const char *path, struct stat *buf)
 {
 	while (*path == ' ')
@@ -177,12 +200,16 @@ s32 vfs_stat(const char *path, struct stat *buf)
 		return -1;
 
 	struct mount_info *m = vfs_find_mount_info(path);
-	assert(m && m->dev && m->dev->vfs && m->dev->vfs->stat);
+	if (!(m && m->dev && m->dev->vfs && m->dev->vfs->stat && m->dev->vfs->perm))
+		return -1;
 
 	u32 len = strlen(m->path);
 	if (len > 1)
 		path += len;
 
+	if (!m->dev->vfs->perm(path, VFS_WRITE, m->dev) && !proc_super())
+		return -1;
+
 	return m->dev->vfs->stat(path, buf, m->dev);
 }
 
@@ -192,7 +219,8 @@ s32 vfs_wait(const char *path, u32 func_ptr)
 		path++;
 
 	struct mount_info *m = vfs_find_mount_info(path);
-	assert(m && m->dev && m->dev->vfs);
+	if (!(m && m->dev && m->dev->vfs))
+		return -1;
 
 	// Default wait
 	if (!m->dev->vfs->wait) {
@@ -280,10 +308,19 @@ static s32 devfs_read(const char *path, void *buf, u32 offset, u32 count, struct
 {
 	struct device *target = device_get_by_name(path + 1);
 	if (!target || !target->read)
-		return 0;
+		return -1;
 	return target->read(buf, offset, count, dev);
 }
 
+static s32 devfs_ioctl(const char *path, u32 request, void *arg1, void *arg2, void *arg3,
+		       struct device *dev)
+{
+	struct device *target = device_get_by_name(path + 1);
+	if (!target || !target->ioctl)
+		return -1;
+	return target->ioctl(request, arg1, arg2, arg3, dev);
+}
+
 static u8 devfs_perm(const char *path, enum vfs_perm perm, struct device *dev)
 {
 	(void)path;
@@ -309,6 +346,7 @@ void device_install(void)
 	struct vfs *vfs = zalloc(sizeof(*vfs));
 	vfs->type = VFS_DEVFS;
 	vfs->read = devfs_read;
+	vfs->ioctl = devfs_ioctl;
 	vfs->perm = devfs_perm;
 	vfs->ready = devfs_ready;
 	struct device *dev = zalloc(sizeof(*dev));
diff --git a/kernel/features/load.c b/kernel/features/load.c
index 8a4aae3..9e6db79 100644
--- a/kernel/features/load.c
+++ b/kernel/features/load.c
@@ -3,28 +3,36 @@
 #include <fs.h>
 #include <load.h>
 #include <mem.h>
+#include <mm.h>
 #include <str.h>
 
-void proc_load(struct proc *proc, void *data)
-{
-	u32 stack = (u32)malloc(0x2000) + 0x1000;
-
-	proc->regs.ebp = (u32)stack;
-	proc->regs.useresp = (u32)stack;
-	proc->regs.eip = (u32)data;
-	proc->entry = (u32)data;
-}
+#define PROC_STACK_SIZE 0x4000
 
 int bin_load(const char *path, struct proc *proc)
 {
 	struct stat s = { 0 };
 	vfs_stat(path, &s);
-	char *data = malloc(s.size);
-	if (!vfs_read(path, data, 0, s.size))
-		return 1;
 
 	strcpy(proc->name, path);
-	proc_load(proc, data);
 
+	struct page_dir *prev;
+	memory_backup_dir(&prev);
+	memory_switch_dir(proc->page_dir);
+
+	u32 size = PAGE_ALIGN_UP(s.size);
+	u32 data = (u32)memory_alloc(proc->page_dir, size, MEMORY_USER | MEMORY_CLEAR);
+
+	if (!vfs_read(proc->name, (void *)data, 0, s.size)) {
+		memory_switch_dir(prev);
+		return 1;
+	}
+
+	u32 stack = (u32)memory_alloc(proc->page_dir, PROC_STACK_SIZE, MEMORY_USER | MEMORY_CLEAR);
+	proc->regs.ebp = stack;
+	proc->regs.useresp = stack;
+	proc->regs.eip = data;
+	proc->entry = data;
+
+	memory_switch_dir(prev);
 	return 0;
 }
diff --git a/kernel/features/mm.c b/kernel/features/mm.c
new file mode 100644
index 0000000..9eca438
--- /dev/null
+++ b/kernel/features/mm.c
@@ -0,0 +1,498 @@
+// Hugely inspired by the implementation in skiftOS: MIT License, Copyright (c) 2020 N. Van Bossuyt
+// MIT License, Copyright (c) 2021 Marvin Borner
+
+#include <assert.h>
+#include <cpu.h>
+#include <def.h>
+#include <mem.h>
+#include <mm.h>
+#include <print.h>
+
+static struct page_dir kernel_dir ALIGNED(PAGE_SIZE) = { 0 };
+static struct page_table kernel_tables[256] ALIGNED(PAGE_SIZE) = { 0 };
+
+/**
+ * Lowlevel paging
+ */
+
+/*static void paging_disable(void)
+{
+	cr0_set(cr0_get() | 0x7fffffff);
+}*/
+
+static void paging_enable(void)
+{
+	cr0_set(cr0_get() | 0x80000000);
+}
+
+static void paging_switch_dir(u32 dir)
+{
+	cr3_set(dir);
+}
+
+extern void paging_invalidate_tlb(void);
+
+void page_fault_handler(struct regs *r)
+{
+	// Check error code
+	const char *type = (r->err_code & 1) ? "present" : "non-present";
+	const char *operation = (r->err_code & 2) ? "write" : "read";
+	const char *super = (r->err_code & 4) ? "User" : "Super";
+
+	// Check cr2 address
+	u32 vaddr;
+	__asm__ volatile("movl %%cr2, %%eax" : "=a"(vaddr));
+	struct proc *proc = proc_current();
+	struct page_dir *dir = NULL;
+	if (proc && proc->page_dir) {
+		dir = proc->page_dir;
+		printf("Stack is at %x, entry at %x\n", virtual_to_physical(dir, proc->regs.ebp),
+		       virtual_to_physical(dir, proc->entry));
+	} else {
+		dir = &kernel_dir;
+	}
+	u32 paddr = virtual_to_physical(dir, vaddr);
+
+	// Print!
+	printf("%s process tried to %s a %s page at [vaddr=%x; paddr=%x]\n", super, operation, type,
+	       vaddr, paddr);
+
+	isr_panic(r);
+}
+
+/**
+ * Physical
+ */
+
+static u32 memory_used = 0;
+static u32 memory_total = 0;
+static u32 best_bet = 0;
+static u8 memory[PAGE_COUNT * PAGE_COUNT / 8] = { 0 };
+
+static u8 physical_page_is_used(u32 addr)
+{
+	u32 page = addr / PAGE_SIZE;
+	return memory[page / 8] & (1 << (page % 8));
+}
+
+static void physical_page_set_used(u32 address)
+{
+	u32 page = address / PAGE_SIZE;
+
+	if (page == best_bet)
+		best_bet++;
+
+	memory[page / 8] |= 1 << (page % 8);
+}
+
+static void physical_page_set_free(u32 address)
+{
+	u32 page = address / PAGE_SIZE;
+
+	if (page < best_bet)
+		best_bet = page;
+
+	memory[page / 8] &= ~(1 << (page % 8));
+}
+
+static void physical_set_used(struct memory_range range)
+{
+	assert(PAGE_ALIGNED(range.base) && PAGE_ALIGNED(range.size));
+
+	for (u32 i = 0; i < range.size / PAGE_SIZE; i++) {
+		u32 addr = range.base + i * PAGE_SIZE;
+		if (!physical_page_is_used(addr)) {
+			memory_used += PAGE_SIZE;
+			physical_page_set_used(addr);
+		}
+	}
+}
+
+static void physical_set_free(struct memory_range range)
+{
+	assert(PAGE_ALIGNED(range.base) && PAGE_ALIGNED(range.size));
+
+	for (u32 i = 0; i < range.size / PAGE_SIZE; i++) {
+		u32 addr = range.base + i * PAGE_SIZE;
+
+		if (physical_page_is_used(addr)) {
+			memory_used -= PAGE_SIZE;
+			physical_page_set_free(addr);
+		}
+	}
+}
+
+static u8 physical_is_used(struct memory_range range)
+{
+	assert(PAGE_ALIGNED(range.base) && PAGE_ALIGNED(range.size));
+
+	for (u32 i = 0; i < range.size / PAGE_SIZE; i++) {
+		u32 addr = range.base + i * PAGE_SIZE;
+
+		if (physical_page_is_used(addr))
+			return 1;
+	}
+
+	return 0;
+}
+
+struct memory_range physical_alloc(u32 size)
+{
+	assert(PAGE_ALIGNED(size));
+
+	for (u32 i = best_bet; i < ((memory_total - size) / PAGE_SIZE); i++) {
+		struct memory_range range = memory_range(i * PAGE_SIZE, size);
+
+		if (!physical_is_used(range)) {
+			physical_set_used(range);
+			return range;
+		}
+	}
+
+	panic("Out of physical memory!\n");
+	return memory_range(0, 0);
+}
+
+void physical_free(struct memory_range range)
+{
+	assert(PAGE_ALIGNED(range.base) && PAGE_ALIGNED(range.size));
+	physical_set_free(range);
+}
+
+/**
+ * Virtual
+ */
+
+#define PDI(vaddr) (((vaddr) >> 22) & 0x03ff)
+#define PTI(vaddr) (((vaddr) >> 12) & 0x03ff)
+
+u8 virtual_present(struct page_dir *dir, u32 vaddr)
+{
+	u32 pdi = PDI(vaddr);
+	union page_dir_entry *dir_entry = &dir->entries[pdi];
+	if (!dir_entry->bits.present)
+		return 0;
+
+	struct page_table *table = (struct page_table *)(dir_entry->bits.address * PAGE_SIZE);
+
+	u32 pti = PTI(vaddr);
+	union page_table_entry *table_entry = &table->entries[pti];
+
+	return table_entry->bits.present;
+}
+
+u32 virtual_to_physical(struct page_dir *dir, u32 vaddr)
+{
+	u32 pdi = PDI(vaddr);
+	union page_dir_entry *dir_entry = &dir->entries[pdi];
+	if (!dir_entry->bits.present)
+		return 0;
+
+	struct page_table *table = (struct page_table *)(dir_entry->bits.address * PAGE_SIZE);
+
+	u32 pti = PTI(vaddr);
+	union page_table_entry *table_entry = &table->entries[pti];
+	if (!table_entry->bits.present)
+		return 0;
+
+	return (table_entry->bits.address * PAGE_SIZE) + (vaddr & (PAGE_SIZE - 1));
+}
+
+void virtual_map(struct page_dir *dir, struct memory_range prange, u32 vaddr, u32 flags)
+{
+	for (u32 i = 0; i < prange.size / PAGE_SIZE; i++) {
+		u32 offset = i * PAGE_SIZE;
+
+		u32 pdi = PDI(vaddr + offset);
+		union page_dir_entry *dir_entry = &dir->entries[pdi];
+		struct page_table *table =
+			(struct page_table *)(dir_entry->bits.address * PAGE_SIZE);
+
+		if (!dir_entry->bits.present) {
+			table = memory_alloc_identity(dir, MEMORY_CLEAR);
+			dir_entry->bits.present = 1;
+			dir_entry->bits.writable = 1;
+			dir_entry->bits.user = 1;
+			dir_entry->bits.address = (u32)(table) >> 12;
+		}
+
+		u32 pti = PTI(vaddr + offset);
+		union page_table_entry *table_entry = &table->entries[pti];
+		table_entry->bits.present = 1;
+		table_entry->bits.writable = 1;
+		table_entry->bits.user = flags & MEMORY_USER;
+		table_entry->bits.address = (prange.base + offset) >> 12;
+	}
+
+	paging_invalidate_tlb();
+}
+
+struct memory_range virtual_alloc(struct page_dir *dir, struct memory_range prange, u32 flags)
+{
+	u8 user = flags & MEMORY_USER;
+	u32 vaddr = 0;
+	u32 size = 0;
+
+	for (u32 i = (user ? 256 : 1) * PAGE_COUNT; i < (user ? PAGE_COUNT : 256) * PAGE_COUNT;
+	     i++) {
+		u32 addr = i * PAGE_SIZE;
+		if (!virtual_present(dir, addr)) {
+			if (size == 0)
+				vaddr = addr;
+
+			size += PAGE_SIZE;
+
+			if (size == prange.size) {
+				virtual_map(dir, prange, vaddr, flags);
+				return memory_range(vaddr, size);
+			}
+		} else {
+			size = 0;
+		}
+	}
+
+	panic("Out of virtual memory!\n");
+	return memory_range(0, 0);
+}
+
+void virtual_free(struct page_dir *dir, struct memory_range vrange)
+{
+	for (u32 i = 0; i < vrange.size / PAGE_SIZE; i++) {
+		u32 offset = i * PAGE_SIZE;
+
+		u32 pdi = PDI(vrange.base + offset);
+		union page_dir_entry *dir_entry = &dir->entries[pdi];
+		if (!dir_entry->bits.present)
+			continue;
+
+		struct page_table *table =
+			(struct page_table *)(dir_entry->bits.address * PAGE_SIZE);
+
+		u32 pti = PTI(vrange.base + offset);
+		union page_table_entry *table_entry = &table->entries[pti];
+
+		if (table_entry->bits.present)
+			table_entry->uint = 0;
+	}
+
+	paging_invalidate_tlb();
+}
+
+struct page_dir *virtual_create_dir(void)
+{
+	struct page_dir *dir = memory_alloc(&kernel_dir, sizeof(*dir), MEMORY_CLEAR);
+
+	memset(dir, 0, sizeof(*dir));
+
+	for (u32 i = 0; i < 256; i++) {
+		union page_dir_entry *dir_entry = &dir->entries[i];
+
+		dir_entry->bits.present = 1;
+		dir_entry->bits.writable = 1;
+		dir_entry->bits.user = 0;
+		dir_entry->bits.address = (u32)&kernel_tables[i] / PAGE_SIZE;
+	}
+
+	return dir;
+}
+
+void virtual_destroy_dir(struct page_dir *dir)
+{
+	assert(dir != &kernel_dir);
+
+	for (u32 i = 256; i < PAGE_COUNT; i++) {
+		union page_dir_entry *dir_entry = &dir->entries[i];
+		if (dir_entry->bits.present) {
+			struct page_table *table =
+				(struct page_table *)(dir_entry->bits.address * PAGE_SIZE);
+			for (u32 j = 0; j < PAGE_COUNT; j++) {
+				union page_table_entry *table_entry = &table->entries[j];
+				if (table_entry->bits.present) {
+					u32 paddr = table_entry->bits.address * PAGE_SIZE;
+					physical_free(memory_range(paddr, PAGE_SIZE));
+				}
+			}
+
+			memory_free(&kernel_dir, memory_range((u32)table, sizeof(*table)));
+		}
+	}
+
+	memory_free(&kernel_dir, memory_range((u32)dir, sizeof(*dir)));
+}
+
+struct page_dir *virtual_kernel_dir(void)
+{
+	return &kernel_dir;
+}
+
+/**
+ * Memory wrappers
+ */
+
+void *memory_alloc(struct page_dir *dir, u32 size, u32 flags)
+{
+	assert(PAGE_ALIGNED(size));
+
+	if (!size)
+		goto err;
+
+	struct memory_range prange = physical_alloc(size);
+	if (prange.size == 0)
+		goto err;
+
+	u32 vaddr = virtual_alloc(dir, prange, flags).base;
+	if (!vaddr) {
+		physical_free(prange);
+		goto err;
+	}
+
+	if (flags & MEMORY_CLEAR)
+		memset((void *)vaddr, 0, size);
+
+	return (void *)vaddr;
+
+err:
+	print("Memory allocation error!\n");
+	return 0;
+}
+
+void *memory_alloc_identity(struct page_dir *dir, u32 flags)
+{
+	for (u32 i = 1; i < 256 * PAGE_COUNT; i++) {
+		struct memory_range range = memory_range(i * PAGE_SIZE, PAGE_SIZE);
+
+		if (!virtual_present(dir, range.base) && !physical_is_used(range)) {
+			physical_set_used(range);
+			virtual_map(dir, range, range.base, flags);
+			if (flags & MEMORY_CLEAR)
+				memset((void *)range.base, 0, PAGE_SIZE);
+			return (void *)range.base;
+		}
+	}
+
+	return 0;
+}
+
+void memory_free(struct page_dir *dir, struct memory_range vrange)
+{
+	assert(PAGE_ALIGNED(vrange.base) && PAGE_ALIGNED(vrange.size));
+
+	for (u32 i = 0; i < vrange.size / PAGE_SIZE; i++) {
+		u32 vaddr = vrange.base + i * PAGE_SIZE;
+		if (virtual_present(dir, vaddr)) {
+			struct memory_range page_prange =
+				memory_range(virtual_to_physical(dir, vaddr), PAGE_SIZE);
+			struct memory_range page_vrange = memory_range(vaddr, PAGE_SIZE);
+			physical_free(page_prange);
+			virtual_free(dir, page_vrange);
+		}
+	}
+}
+
+void memory_map_identity(struct page_dir *dir, struct memory_range prange, u32 flags)
+{
+	assert(PAGE_ALIGNED(prange.base) && PAGE_ALIGNED(prange.size));
+
+	physical_set_used(prange);
+	virtual_map(dir, prange, prange.base, flags);
+	if (flags & MEMORY_CLEAR)
+		memset((void *)prange.base, 0, prange.size);
+}
+
+void memory_switch_dir(struct page_dir *dir)
+{
+	paging_switch_dir(virtual_to_physical(&kernel_dir, (u32)dir));
+}
+
+void memory_backup_dir(struct page_dir **backup)
+{
+	struct proc *proc = proc_current();
+	struct page_dir *dir = proc ? proc->page_dir : virtual_kernel_dir();
+	*backup = dir;
+}
+
+struct memory_range memory_range_from(u32 base, u32 size)
+{
+	u32 align = PAGE_SIZE - base % PAGE_SIZE;
+
+	if (base % PAGE_SIZE == 0) {
+		align = 0;
+	}
+
+	base += align;
+	size -= align;
+
+	size -= size % PAGE_SIZE;
+
+	return memory_range(base, size);
+}
+
+struct memory_range memory_range_around(u32 base, u32 size)
+{
+	u32 align = base % PAGE_SIZE;
+
+	base -= align;
+	size += align;
+
+	size += PAGE_SIZE - size % PAGE_SIZE;
+
+	return memory_range(base, size);
+}
+
+extern u32 kernel_start;
+extern u32 kernel_end;
+static struct memory_range kernel_memory_range(void)
+{
+	return memory_range_around((u32)&kernel_start, (u32)&kernel_end - (u32)&kernel_start);
+}
+
+void memory_install(struct mem_info *mem_info, struct vid_info *vid_info)
+{
+	for (struct mmap_boot *p = mem_info->start; (u32)(p - mem_info->start) < mem_info->size;
+	     p++) {
+		if (p->hbase || !p->acpi || !p->type)
+			continue;
+
+		u32 size = p->lsize;
+		if (p->hsize)
+			size = U32_MAX - p->lbase;
+
+		/* printf("Memory region: %x-%x\n", p->lbase, p->lbase + size); */
+		if (p->type == MEMORY_AVAILABLE) {
+			physical_set_free(memory_range_around(p->lbase, size / PAGE_SIZE));
+			memory_total += size;
+		} else if (p->type == MEMORY_DEFECT) {
+			printf("Defect memory at 0x%x-0x%x!\n", p->lbase, p->lbase + size);
+		}
+	}
+
+	for (u32 i = 0; i < 256; i++) {
+		union page_dir_entry *dir_entry = &kernel_dir.entries[i];
+		dir_entry->bits.present = 1;
+		dir_entry->bits.writable = 1;
+		dir_entry->bits.user = 0;
+		dir_entry->bits.address = (u32)&kernel_tables[i] / PAGE_SIZE;
+	}
+
+	memory_used = 0;
+	printf("Detected memory: %dKiB (%dMiB)\n", memory_total >> 10, memory_total >> 20);
+
+	// Map kernel
+	memory_map_identity(&kernel_dir, kernel_memory_range(), MEMORY_NONE);
+
+	// Map kernel stack
+	memory_map_identity(&kernel_dir, memory_range_around(STACK_START - STACK_SIZE, STACK_SIZE),
+			    MEMORY_NONE);
+
+	// Map VBE data
+	memory_map_identity(&kernel_dir, memory_range_around((u32)vid_info->vbe, 0x1000),
+			    MEMORY_NONE);
+
+	// Unmap NULL byte/page
+	struct memory_range zero = memory_range(0, PAGE_SIZE);
+	virtual_free(&kernel_dir, zero);
+	physical_set_used(zero);
+
+	memory_switch_dir(&kernel_dir);
+	paging_enable();
+}
diff --git a/kernel/features/proc.asm b/kernel/features/proc.asm
index 1a2ba65..dfc3448 100644
--- a/kernel/features/proc.asm
+++ b/kernel/features/proc.asm
@@ -28,3 +28,9 @@ proc_jump_userspace:
 	push dword [_eip]
 
 	iret
+
+global paging_invalidate_tlb
+paging_invalidate_tlb:
+	mov eax, cr3
+	mov cr3, eax
+	ret
diff --git a/kernel/features/proc.c b/kernel/features/proc.c
index cdbe8b1..b93f7c8 100644
--- a/kernel/features/proc.c
+++ b/kernel/features/proc.c
@@ -6,6 +6,7 @@
 #include <fs.h>
 #include <load.h>
 #include <mem.h>
+#include <mm.h>
 #include <print.h>
 #include <proc.h>
 #include <stack.h>
@@ -19,7 +20,7 @@ struct list *proc_list = NULL;
 struct node *idle_proc = NULL;
 struct node *current = NULL;
 
-// TODO: Use less memcpy and only copy relevant registers
+// TODO: Use less memcpy and only copy relevant registers (rewrite for efficiency argh)
 // TODO: 20 priority queues (https://www.kernel.org/doc/html/latest/scheduler/sched-nice-design.html)
 void scheduler(struct regs *regs)
 {
@@ -56,6 +57,7 @@ void scheduler(struct regs *regs)
 		}
 	}
 
+	memory_switch_dir(((struct proc *)current->data)->page_dir);
 	memcpy(regs, &((struct proc *)current->data)->regs, sizeof(struct regs));
 
 	if (regs->cs != GDT_USER_CODE_OFFSET) {
@@ -74,12 +76,6 @@ void scheduler(struct regs *regs)
 	/* printf("{%d}", ((struct proc *)current->data)->pid); */
 }
 
-static void kernel_idle(void)
-{
-	while (1)
-		;
-}
-
 void proc_print(void)
 {
 	struct node *node = proc_list->head;
@@ -103,7 +99,7 @@ u8 proc_super(void)
 {
 	struct proc *proc = proc_current();
 	if (proc)
-		return proc->super;
+		return proc->priv == PROC_PRIV_ROOT || proc->priv == PROC_PRIV_KERNEL;
 	else if (current_pid == 0)
 		return 1; // Kernel has super permissions
 	else
@@ -161,6 +157,9 @@ void proc_yield(struct regs *r)
 
 void proc_enable_waiting(u32 id, enum proc_wait_type type)
 {
+	struct page_dir *dir_bak;
+	memory_backup_dir(&dir_bak);
+
 	struct proc *proc_bak = proc_current();
 	if (!proc_bak)
 		return;
@@ -183,8 +182,11 @@ void proc_enable_waiting(u32 id, enum proc_wait_type type)
 				struct regs *r = &p->regs;
 				u32 (*func)(u32, u32, u32, u32) =
 					(u32(*)(u32, u32, u32, u32))w->ids[i].func_ptr;
-				if (w->ids[i].func_ptr)
+				if (w->ids[i].func_ptr) {
+					memory_switch_dir(p->page_dir);
 					r->eax = func(r->ebx, r->ecx, r->edx, r->esi);
+					memory_switch_dir(dir_bak);
+				}
 				memset(&w->ids[i], 0, sizeof(w->ids[i]));
 				p->wait.id_cnt--;
 				p->state = PROC_RUNNING;
@@ -237,20 +239,37 @@ end:
 	p->state = PROC_SLEEPING;
 }
 
-struct proc *proc_make(void)
+struct proc *proc_make(enum proc_priv priv)
 {
 	struct proc *proc = zalloc(sizeof(*proc));
 	proc->pid = current_pid++;
-	proc->super = 0;
+	proc->priv = priv;
 	proc->messages = stack_new();
 	proc->state = PROC_RUNNING;
 
+	if (priv == PROC_PRIV_KERNEL)
+		proc->page_dir = virtual_kernel_dir();
+	else
+		proc->page_dir = virtual_create_dir();
+
 	if (current)
 		list_add(proc_list, proc);
 
 	return proc;
 }
 
+void proc_stack_push(struct proc *proc, u32 data)
+{
+	struct page_dir *prev;
+	memory_backup_dir(&prev);
+	memory_switch_dir(proc->page_dir);
+
+	proc->regs.useresp -= sizeof(data);
+	*(u32 *)proc->regs.useresp = data;
+
+	memory_switch_dir(prev);
+}
+
 // TODO: Procfs needs a simpler interface structure (memcmp and everything sucks)
 
 static const char *procfs_parse_path(const char **path, u32 *pid)
@@ -286,6 +305,11 @@ static enum stream_defaults procfs_stream(const char *path)
 	}
 }
 
+struct procfs_message {
+	u8 *data;
+	u32 size;
+};
+
 static s32 procfs_write(const char *path, void *buf, u32 offset, u32 count, struct device *dev)
 {
 	u32 pid = 0;
@@ -299,7 +323,10 @@ static s32 procfs_write(const char *path, void *buf, u32 offset, u32 count, stru
 		if (!memcmp(path, "msg", 4)) {
 			void *msg_data = malloc(count);
 			memcpy(msg_data, buf, count);
-			stack_push_bot(p->messages, msg_data); // TODO: Use offset
+			struct procfs_message *msg = malloc(sizeof(*msg));
+			msg->data = msg_data;
+			msg->size = count;
+			stack_push_bot(p->messages, msg); // TODO: Use offset
 			proc_enable_waiting(pid, PROC_WAIT_MSG);
 			return count;
 		} else if (!memcmp(path, "io/", 3)) {
@@ -321,7 +348,7 @@ static s32 procfs_write(const char *path, void *buf, u32 offset, u32 count, stru
 		}
 	}
 
-	printf("%s - off: %d, cnt: %d, buf: %x, dev %x\n", path, offset, count, buf, dev);
+	printf("ERR: %s - off: %d, cnt: %d, buf: %x, dev %x\n", path, offset, count, buf, dev);
 	return -1;
 }
 
@@ -352,12 +379,13 @@ static s32 procfs_read(const char *path, void *buf, u32 offset, u32 count, struc
 			if (stack_empty(p->messages)) {
 				return -1; // This shouldn't happen
 			} else {
-				u8 *msg = stack_pop(p->messages);
+				struct procfs_message *msg = stack_pop(p->messages);
 				if (!msg)
 					return -1;
-				memcpy(buf, msg + offset, count);
+				memcpy(buf, msg->data + offset, MIN(count, msg->size));
+				free(msg->data);
 				free(msg);
-				return count;
+				return MIN(count, msg->size);
 			}
 		} else if (!memcmp(path, "io/", 3)) {
 			path += 3;
@@ -465,32 +493,26 @@ void proc_init(void)
 	vfs_mount(dev, "/proc/");
 
 	// Idle proc
-	struct proc *kernel_proc = proc_make();
-	void (*func)(void) = kernel_idle;
-	proc_load(kernel_proc, *(void **)&func);
-	strcpy(kernel_proc->name, "idle");
+	struct proc *kernel_proc = proc_make(PROC_PRIV_NONE);
+	bin_load("/bin/idle", kernel_proc);
 	kernel_proc->state = PROC_SLEEPING;
 	idle_proc = list_add(proc_list, kernel_proc);
 
-	struct node *new = list_add(proc_list, proc_make());
+	// Init proc (root)
+	struct node *new = list_add(proc_list, proc_make(PROC_PRIV_ROOT));
 	bin_load("/bin/init", new->data);
 	current = new;
+	proc_stack_push(new->data, 0);
 
 	_eip = ((struct proc *)new->data)->regs.eip;
 	_esp = ((struct proc *)new->data)->regs.useresp;
-	((struct proc *)new->data)->super = 1;
-
-	u32 argc = 2;
-	char **argv = malloc(sizeof(*argv) * (argc + 1));
-	argv[0] = strdup("init");
-	argv[1] = (char *)boot_passed->vbe;
-	argv[2] = NULL;
-
-	((u32 *)_esp)[0] = argc; // First argument (argc)
-	((u32 *)_esp)[1] = (u32)argv; // Second argument (argv)
 
 	printf("Jumping to userspace!\n");
+	memory_switch_dir(((struct proc *)new->data)->page_dir);
+
+	// You're waiting for a train. A train that will take you far away...
 	proc_jump_userspace();
+
 	while (1) {
 	};
 }
diff --git a/kernel/features/syscall.c b/kernel/features/syscall.c
index 61c7479..68bcbaa 100644
--- a/kernel/features/syscall.c
+++ b/kernel/features/syscall.c
@@ -5,6 +5,7 @@
 #include <interrupts.h>
 #include <load.h>
 #include <mem.h>
+#include <mm.h>
 #include <net.h>
 #include <print.h>
 #include <proc.h>
@@ -25,12 +26,13 @@ static void syscall_handler(struct regs *r)
 		loop();
 		break;
 	}
-	case SYS_MALLOC: {
-		r->eax = (u32)malloc(r->ebx);
+	case SYS_ALLOC: {
+		r->eax = (u32)memory_alloc(proc_current()->page_dir, r->ebx,
+					   MEMORY_CLEAR | MEMORY_USER);
 		break;
 	}
 	case SYS_FREE: {
-		free((void *)r->ebx);
+		memory_free(proc_current()->page_dir, memory_range(r->ebx, r->ecx));
 		break;
 	}
 	case SYS_STAT: {
@@ -52,6 +54,11 @@ static void syscall_handler(struct regs *r)
 		r->eax = (u32)vfs_write((char *)r->ebx, (void *)r->ecx, r->edx, r->esi);
 		break;
 	}
+	case SYS_IOCTL: {
+		r->eax = (u32)vfs_ioctl((char *)r->ebx, r->ecx, (void *)r->edx, (void *)r->esi,
+					(void *)r->edi);
+		break;
+	}
 	case SYS_POLL: {
 		s32 ret = vfs_poll((const char **)r->ebx);
 		if (ret == PROC_MAX_WAIT_IDS + 1)
@@ -62,17 +69,10 @@ static void syscall_handler(struct regs *r)
 	}
 	case SYS_EXEC: {
 		char *path = (char *)r->ebx;
-		struct proc *proc = proc_make();
+		struct proc *proc = proc_make(PROC_PRIV_NONE);
 		r->eax = (u32)bin_load(path, proc);
-		u32 argc = 3; // TODO: Add argc evaluator
-		char **argv = malloc(sizeof(*argv) * (argc + 1));
-		argv[0] = (char *)r->ecx;
-		argv[1] = (char *)r->edx;
-		argv[2] = (char *)r->esi;
-		argv[3] = (char *)r->edi;
-		argv[4] = NULL;
-		((u32 *)proc->regs.useresp)[0] = argc;
-		((u32 *)proc->regs.useresp)[1] = (u32)argv;
+		// TODO: Reimplement argc,argv
+		proc_stack_push(proc, 0);
 		if (r->eax)
 			proc_exit(proc, (int)r->eax);
 		proc_yield(r);
@@ -82,6 +82,30 @@ static void syscall_handler(struct regs *r)
 		proc_exit(proc_current(), (int)r->ebx);
 		break;
 	}
+	case SYS_BOOT: { // TODO: Move
+		if (r->ebx != SYS_BOOT_MAGIC || !proc_super()) {
+			r->eax = -1;
+			break;
+		}
+		switch (r->ecx) {
+		case SYS_BOOT_REBOOT:
+			print("Rebooting...\n");
+			outb(0x64, 0xfe);
+			__asm__ volatile("ud2");
+			break;
+		case SYS_BOOT_SHUTDOWN:
+			print("Shutting down...\n");
+			outw(0xB004, 0x2000);
+			outw(0x604, 0x2000);
+			outw(0x4004, 0x3400);
+			outb(0x64, 0xfe);
+			__asm__ volatile("ud2");
+			break;
+		default:
+			r->eax = -1;
+		}
+		break;
+	}
 	case SYS_YIELD: {
 		proc_yield(r);
 		break;
@@ -123,7 +147,7 @@ static void syscall_handler(struct regs *r)
 		break;
 	}
 	default: {
-		print("Unknown syscall!\n");
+		printf("Unknown syscall %d!\n", num);
 		break;
 	}
 	}