Merge branch 'address-spaces'

apps/src/sym.c

@@ -1,8 +1,11 @@
 #include <stdio.h>
 #include <string.h>
+#include <unistd.h>
 
 int main()
 {
+    sleep(6);
+
     FILE* syms = fopen("/sys/moon.sym", "r");
     if (!syms)
     {

kernel/include/memory/AddressSpace.h

@@ -0,0 +1,27 @@
+#pragma once
+#include "memory/Paging.h"
+
+struct AddressSpace
+{
+    static AddressSpace create();
+
+    void destroy();
+
+    void detach();
+
+    AddressSpace clone();
+
+    PageTable* get_pml4()
+    {
+        return m_pml4;
+    }
+
+    bool is_cloned()
+    {
+        return m_cloned;
+    }
+
+  private:
+    PageTable* m_pml4;
+    bool m_cloned;
+};

kernel/include/memory/MemoryManager.h

@@ -29,4 +29,6 @@ namespace MemoryManager
 
     void release_page(void* page);
     void release_pages(void* pages, uint64_t count);
+
+    void protect(void* page, uint64_t count, int flags);
 }

kernel/include/memory/VMM.h

@@ -1,4 +1,5 @@
 #pragma once
+#include "memory/AddressSpace.h"
 #include "memory/Paging.h"
 
 enum Flags
@@ -11,6 +12,17 @@ namespace VMM
 {
     void init(); // Fetch page table from cr3
 
+    void switch_to_user_address_space(AddressSpace& space);
+    void switch_to_previous_user_address_space();
+    void switch_back_to_kernel_address_space();
+
+    void enter_syscall_context();
+    void exit_syscall_context();
+
+    void apply_address_space();
+
+    bool is_using_kernel_address_space();
+
     void map(uint64_t vaddr, uint64_t paddr, int flags);
     void remap(uint64_t vaddr, int flags);
     void unmap(uint64_t vaddr);
@@ -27,4 +39,7 @@ namespace VMM
 
     void decompose_vaddr(uint64_t vaddr, uint64_t& page_index, uint64_t& pt_index, uint64_t& pd_index,
                          uint64_t& pdp_index);
+    uint64_t recompose_vaddr(uint64_t page_index, uint64_t pt_index, uint64_t pd_index, uint64_t pdp_index);
+
+    void install_kernel_page_directory_into_address_space(AddressSpace& space);
 };

kernel/include/sys/Syscall.h

@@ -20,6 +20,8 @@
 namespace Syscall
 {
     void entry(Context* context);
+
+    char* strdup_from_user(const char* user_string);
 }
 
 void sys_exit(Context* context, int status);

kernel/include/thread/Scheduler.h

@@ -10,7 +10,8 @@ namespace Scheduler
     void exit(int status);
     void sleep(unsigned long ms);
     void add_kernel_task(void (*task)(void));
-    void add_user_task(void* task);
+
+    Task* create_user_task();
 
     void load_user_task(const char* filename);
 

kernel/include/thread/Task.h

@@ -1,6 +1,7 @@
 #pragma once
 #include "fs/FileDescriptor.h"
 #include "interrupts/Context.h"
+#include "memory/AddressSpace.h"
 #include "sys/elf/Image.h"
 
 #define TASK_MAX_FDS 8
@@ -43,6 +44,8 @@ struct Task
     ELFImage* image = nullptr;
 
     Descriptor files[TASK_MAX_FDS];
+
+    AddressSpace address_space;
 };
 
 void set_context_from_task(Task& task, Context* ctx);

kernel/src/interrupts/Entry.cpp

@@ -51,6 +51,8 @@ extern "C" void common_handler(Context* context)
             StackTracer tracer(context->rbp);
             tracer.trace_with_ip(context->rip);
 
+            hang(); // FIXME: Remove this when multiple address spaces are working.
+
             Scheduler::task_misbehave(context, -3);
         }
     }

kernel/src/main.cpp

@@ -16,6 +16,7 @@
 #include "io/PIC.h"
 #include "io/Serial.h"
 #include "log/Log.h"
+#include "memory/AddressSpace.h"
 #include "memory/Memory.h"
 #include "memory/MemoryManager.h"
 #include "memory/MemoryMap.h"
@@ -56,12 +57,6 @@ extern "C" void _start()
 
     kinfoln("Loaded IDT");
 
-    PIC::remap();
-    PIC::enable_master(0b11111100); // enable keyboard and PIT
-    PIC::enable_slave(0b11111111);
-
-    kinfoln("Prepared PIC");
-
     PIT::initialize(1000); // 1000 times per second
 
     kinfoln("Prepared PIT");
@@ -79,6 +74,7 @@ extern "C" void _start()
     });
 
     Scheduler::load_user_task("/bin/init");
+    Scheduler::load_user_task("/bin/sym");
 
     kinfoln("Prepared scheduler tasks");
 
@@ -87,7 +83,9 @@ extern "C" void _start()
 
     Init::finish_kernel_boot();
 
-    Interrupts::enable(); // Task switching commences here
+    PIC::remap();
+    PIC::enable_master(0b11111100); // enable keyboard and PIT
+    PIC::enable_slave(0b11111111);
 
     kinfoln("Interrupts enabled");
 

kernel/src/memory/AddressSpace.cpp

@@ -0,0 +1,93 @@
+#define MODULE "vmm"
+
+#include "memory/AddressSpace.h"
+#include "log/Log.h"
+#include "memory/PMM.h"
+#include "memory/VMM.h"
+
+AddressSpace AddressSpace::create()
+{
+    AddressSpace result;
+    result.m_pml4 = (PageTable*)PMM::request_page();
+    VMM::install_kernel_page_directory_into_address_space(result);
+    return result;
+}
+
+void AddressSpace::destroy()
+{
+    if (m_cloned)
+    {
+        kdbgln("Will not destroy a cloned address space, I don't own it");
+        return;
+    }
+    uint64_t pages_freed = 0;
+    for (int i = 0; i < 512; i++)
+    {
+        PageDirectoryEntry& pdp_pde = m_pml4->entries[i];
+        if (!pdp_pde.present) continue;
+        if (pdp_pde.larger_pages)
+        {
+            pages_freed++;
+            PMM::free_page((void*)pdp_pde.get_address());
+            continue;
+        }
+        PageTable* pdp = (PageTable*)pdp_pde.get_address();
+        for (int j = 0; j < 511; j++) // skip the last page directory, it's the kernel one
+        {
+            PageDirectoryEntry& pd_pde = pdp->entries[j];
+            if (!pd_pde.present) continue;
+            if (pd_pde.larger_pages)
+            {
+                pages_freed++;
+                PMM::free_page((void*)pd_pde.get_address());
+                continue;
+            }
+            PageTable* pd = (PageTable*)pd_pde.get_address();
+            for (int k = 0; k < 512; k++)
+            {
+                PageDirectoryEntry& pt_pde = pd->entries[k];
+                if (!pt_pde.present) continue;
+                if (pt_pde.larger_pages)
+                {
+                    pages_freed++;
+                    PMM::free_page((void*)pt_pde.get_address());
+                    continue;
+                }
+                PageTable* pt = (PageTable*)pt_pde.get_address();
+                for (int l = 0; l < 512; l++)
+                {
+                    PageDirectoryEntry& pde = pt->entries[l];
+                    if (!pde.present) continue;
+                    pages_freed++;
+                    PMM::free_page((void*)pde.get_address());
+                }
+                pages_freed++;
+                PMM::free_page(pt);
+            }
+            pages_freed++;
+            PMM::free_page(pd);
+        }
+        pages_freed++;
+        PMM::free_page(pdp);
+    }
+    pages_freed++;
+    PMM::free_page(m_pml4);
+
+    kdbgln("Reclaimed %ld pages from address space!", pages_freed);
+}
+
+void AddressSpace::detach()
+{
+    if (!m_cloned) return;
+    m_pml4 = (PageTable*)PMM::request_page();
+    VMM::install_kernel_page_directory_into_address_space(*this);
+    m_cloned = false;
+}
+
+AddressSpace AddressSpace::clone()
+{
+    AddressSpace result;
+    result.m_pml4 = m_pml4;
+    result.m_cloned = true;
+    return result;
+}

kernel/src/memory/MemoryManager.cpp

@@ -190,3 +190,8 @@ void MemoryManager::release_pages(void* pages, uint64_t count)
     }
     KernelHeap::free_virtual_pages((uint64_t)pages, count);
 }
+
+void MemoryManager::protect(void* page, uint64_t count, int flags)
+{
+    for (uint64_t i = 0; i < count; i++) { VMM::remap((uint64_t)page + (i * PAGE_SIZE), flags); }
+}

kernel/src/memory/VMM.cpp

@@ -9,29 +9,88 @@
 #include "utils/Addresses.h"
 #include "utils/Registers.h"
 
-// FIXME: There is a lot of duplicate code in this file. This should probably be refactored.
+static PageTable* kernel_pml4;
+static PageTable* current_pml4;
+static AddressSpace* user_address_space;
 
-static PageTable* PML4;
+void VMM::switch_back_to_kernel_address_space()
+{
+    if (current_pml4 != kernel_pml4) { current_pml4 = kernel_pml4; }
+}
+
+void VMM::switch_to_user_address_space(AddressSpace& space)
+{
+    user_address_space = &space;
+    current_pml4 = user_address_space->get_pml4();
+}
+
+void VMM::switch_to_previous_user_address_space()
+{
+    current_pml4 = user_address_space->get_pml4();
+}
+
+void VMM::enter_syscall_context()
+{
+    if (current_pml4 != kernel_pml4)
+    {
+        current_pml4 = kernel_pml4;
+        apply_address_space();
+        switch_to_previous_user_address_space();
+    }
+}
+
+void VMM::exit_syscall_context()
+{
+    if (current_pml4 != user_address_space->get_pml4()) { switch_to_previous_user_address_space(); }
+    apply_address_space();
+}
+
+void VMM::apply_address_space()
+{
+    write_cr3(current_pml4);
+}
+
+bool VMM::is_using_kernel_address_space()
+{
+    return current_pml4 == kernel_pml4;
+}
 
 void VMM::init()
 {
-    PML4 = (PageTable*)read_cr3();
+    kernel_pml4 = (PageTable*)read_cr3();
+    current_pml4 = kernel_pml4;
 }
 
 void VMM::unmap(uint64_t vaddr)
 {
     vaddr = round_down_to_nearest_page(vaddr);
 
-    PageDirectoryEntry* pde = find_pde(PML4, vaddr);
+    PageDirectoryEntry* pde = find_pde(current_pml4, vaddr);
     if (!pde) return; // Already unmapped
 
     memset(pde, 0, sizeof(PageDirectoryEntry));
     flush_tlb(vaddr);
 }
 
+void VMM::remap(uint64_t vaddr, int flags)
+{
+    vaddr = round_down_to_nearest_page(vaddr);
+
+    PageDirectoryEntry* pde = find_pde(current_pml4, vaddr);
+    if (!pde) return; // Not mapped
+
+    if (flags & User) propagate_user(current_pml4, vaddr);
+    else
+        pde->user = false;
+    if (flags & ReadWrite) propagate_read_write(current_pml4, vaddr);
+    else
+        pde->read_write = false;
+    flush_tlb(vaddr);
+}
+
 uint64_t VMM::get_physical(uint64_t vaddr)
 {
-    PageDirectoryEntry* pde = find_pde(PML4, round_down_to_nearest_page(vaddr));
+    PageDirectoryEntry* pde = find_pde(current_pml4, round_down_to_nearest_page(vaddr));
     if (!pde) return UINT64_MAX; // Not mapped
 
     return pde->get_address() | (vaddr % PAGE_SIZE);
@@ -39,7 +98,7 @@ uint64_t VMM::get_physical(uint64_t vaddr)
 
 uint64_t VMM::get_flags(uint64_t vaddr)
 {
-    PageDirectoryEntry* pde = find_pde(PML4, round_down_to_nearest_page(vaddr));
+    PageDirectoryEntry* pde = find_pde(current_pml4, round_down_to_nearest_page(vaddr));
     if (!pde) return 0; // Not mapped
 
     uint64_t flags = 0;
@@ -51,35 +110,38 @@ uint64_t VMM::get_flags(uint64_t vaddr)
 void VMM::map(uint64_t vaddr, uint64_t paddr, int flags)
 {
     vaddr = round_down_to_nearest_page(vaddr);
-    PageDirectoryEntry* pde = find_pde(PML4, vaddr);
+    PageDirectoryEntry* pde = find_pde(current_pml4, vaddr);
     bool will_flush_tlb = true;
     if (!pde)
     {
-        pde = create_pde_if_not_exists(PML4, vaddr);
+        pde = create_pde_if_not_exists(current_pml4, vaddr);
         will_flush_tlb = false;
     }
     else if (pde->larger_pages)
     {
         unmap(vaddr);
-        pde = create_pde_if_not_exists(PML4, vaddr);
+        pde = create_pde_if_not_exists(current_pml4, vaddr);
-        will_flush_tlb = false;
     }
 
     pde->set_address(round_down_to_nearest_page(paddr));
-    if (flags & User) propagate_user(PML4, vaddr);
+    if (flags & User) propagate_user(current_pml4, vaddr);
-    if (flags & ReadWrite) propagate_read_write(PML4, vaddr);
+    else
+        pde->user = false;
+    if (flags & ReadWrite) propagate_read_write(current_pml4, vaddr);
+    else
+        pde->read_write = false;
     if (will_flush_tlb) flush_tlb(vaddr);
 }
 
 PageDirectoryEntry* VMM::find_pde(PageTable* root, uint64_t vaddr)
 {
-    uint64_t page_index, pt_index, pd_index, pdp_index;
+    uint64_t page_index;
     PageDirectoryEntry* pde;
     PageTable* pt = root;
 
-    decompose_vaddr(vaddr, page_index, pt_index, pd_index, pdp_index);
+    uint64_t indexes[3];
 
-    uint64_t indexes[3] = {pdp_index, pd_index, pt_index};
+    decompose_vaddr(vaddr, page_index, indexes[2], indexes[1], indexes[0]);
 
     for (int i = 0; i < 3;
          i++) // Walk through the page map level 4, page directory pointer, and page directory to find the page table.
@@ -98,11 +160,13 @@ PageDirectoryEntry* VMM::find_pde(PageTable* root, uint64_t vaddr)
 
 PageDirectoryEntry* VMM::create_pde_if_not_exists(PageTable* root, uint64_t vaddr)
 {
-    uint64_t page_index, pt_index, pd_index, pdp_index;
+    uint64_t page_index;
     PageDirectoryEntry* pde;
     PageTable* pt = root;
 
-    decompose_vaddr(vaddr, page_index, pt_index, pd_index, pdp_index);
+    uint64_t indexes[3];
+
+    decompose_vaddr(vaddr, page_index, indexes[2], indexes[1], indexes[0]);
 
     auto pde_create_if_not_present = [&]() {
         pt = (PageTable*)PMM::request_page();
@@ -112,8 +176,6 @@ PageDirectoryEntry* VMM::create_pde_if_not_exists(PageTable* root, uint64_t vadd
         pde->present = true;
     };
 
-    uint64_t indexes[3] = {pdp_index, pd_index, pt_index};
-
     for (int i = 0; i < 3; i++)
     {
         pde = &pt->entries[indexes[i]];
@@ -130,13 +192,13 @@ PageDirectoryEntry* VMM::create_pde_if_not_exists(PageTable* root, uint64_t vadd
 
 void VMM::propagate_read_write(PageTable* root, uint64_t vaddr)
 {
-    uint64_t page_index, pt_index, pd_index, pdp_index;
+    uint64_t page_index;
     PageDirectoryEntry* pde;
     PageTable* pt = root;
 
-    decompose_vaddr(vaddr, page_index, pt_index, pd_index, pdp_index);
+    uint64_t indexes[3];
 
-    uint64_t indexes[3] = {pdp_index, pd_index, pt_index};
+    decompose_vaddr(vaddr, page_index, indexes[2], indexes[1], indexes[0]);
 
     for (int i = 0; i < 3; i++)
     {
@@ -158,13 +220,13 @@ void VMM::propagate_read_write(PageTable* root, uint64_t vaddr)
 
 void VMM::propagate_user(PageTable* root, uint64_t vaddr)
 {
-    uint64_t page_index, pt_index, pd_index, pdp_index;
+    uint64_t page_index;
     PageDirectoryEntry* pde;
     PageTable* pt = root;
 
-    decompose_vaddr(vaddr, page_index, pt_index, pd_index, pdp_index);
+    uint64_t indexes[3];
 
-    uint64_t indexes[3] = {pdp_index, pd_index, pt_index};
+    decompose_vaddr(vaddr, page_index, indexes[2], indexes[1], indexes[0]);
 
     for (int i = 0; i < 3; i++)
     {
@@ -201,3 +263,29 @@ void VMM::decompose_vaddr(uint64_t vaddr, uint64_t& page_index, uint64_t& pt_ind
     vaddr >>= 9;
     pdp_index = vaddr & 0x1ff;
 }
+
+uint64_t VMM::recompose_vaddr(uint64_t page_index, uint64_t pt_index, uint64_t pd_index, uint64_t pdp_index)
+{
+    return pdp_index << 39 | pd_index << 30 | pt_index << 21 | page_index << 12;
+}
+
+void VMM::install_kernel_page_directory_into_address_space(AddressSpace& space)
+{
+    PageTable* space_pml4 = space.get_pml4();
+    PageTable* kernel_last_pdp = (PageTable*)kernel_pml4->entries[511].get_address();
+    PageTable* kernel_last_pd = (PageTable*)kernel_last_pdp->entries[511].get_address();
+
+    PageTable* space_last_pdp = (PageTable*)PMM::request_page();
+
+    PageDirectoryEntry& space_last_pdp_pde = space_pml4->entries[511];
+
+    space_last_pdp_pde.present = true;
+    space_last_pdp_pde.read_write = true;
+    space_last_pdp_pde.set_address((uint64_t)space_last_pdp);
+
+    PageDirectoryEntry& space_last_pd_pde = space_last_pdp->entries[511];
+
+    space_last_pd_pde.present = true;
+    space_last_pd_pde.read_write = true;
+    space_last_pd_pde.set_address((uint64_t)kernel_last_pd);
+}

kernel/src/misc/reboot.cpp

@@ -8,6 +8,7 @@
 #include "interrupts/Interrupts.h"
 #include "io/IO.h"
 #include "log/Log.h"
+#include "memory/VMM.h"
 #include "misc/hang.h"
 #include "std/string.h"
 
@@ -85,6 +86,11 @@ static void try_idt_triple_fault()
 [[noreturn]] void reboot()
 {
     Interrupts::disable();
+    if (!VMM::is_using_kernel_address_space())
+    {
+        VMM::switch_back_to_kernel_address_space();
+        VMM::apply_address_space();
+    }
     kinfoln("Attempting reboot using ACPI");
     try_acpi_reboot();
     kinfoln("Attempting reboot using keyboard RESET pulsing");

kernel/src/sys/Syscall.cpp

@@ -1,11 +1,14 @@
 #include "sys/Syscall.h"
 #include "errno.h"
 #include "io/Serial.h"
+#include "memory/VMM.h"
+#include "std/string.h"
 #include "thread/Scheduler.h"
 
 void Syscall::entry(Context* context)
 {
     asm volatile("cli");
+    VMM::enter_syscall_context();
     switch (context->rax)
     {
     case SYS_exit: sys_exit(context, (int)context->rdi); break;
@@ -24,4 +27,12 @@ void Syscall::entry(Context* context)
     case SYS_exec: sys_exec(context, (const char*)context->rdi); break;
     default: context->rax = -ENOSYS; break;
     }
+    VMM::exit_syscall_context();
+}
+
+char* Syscall::strdup_from_user(const char* user_string)
+{
+    uint64_t phys = VMM::get_physical((uint64_t)user_string);
+    if (phys == (uint64_t)-1) { return nullptr; }
+    return strdup((const char*)phys);
 }

kernel/src/sys/elf/ELFLoader.cpp

@@ -8,6 +8,7 @@
 #include "log/Log.h"
 #include "memory/Memory.h"
 #include "memory/MemoryManager.h"
+#include "memory/VMM.h"
 #include "misc/utils.h"
 #include "std/stdlib.h"
 #include "std/string.h"
@@ -67,7 +68,9 @@ ELFImage* ELFLoader::load_elf_from_vfs(VFS::Node* node)
 {
     Elf64_Ehdr elf_ehdr;
     ASSERT(VFS::read(node, 0, sizeof(elf_ehdr), (char*)&elf_ehdr) >= 0);
-    ASSERT(strncmp((const char*)elf_ehdr.e_ident, ELFMAG, SELFMAG) == 0);
+    ASSERT(strncmp((const char*)elf_ehdr.e_ident, ELFMAG, SELFMAG) ==
+           0); // If you haven't checked the ELF executable with check_elf_image() first, then an assertion fail is your
+               // fault =D
     ASSERT(elf_ehdr.e_ident[EI_CLASS] == ELFCLASS64);
     ASSERT(elf_ehdr.e_ident[EI_DATA] == ELFDATA2LSB);
     ASSERT(elf_ehdr.e_type == ET_EXEC);
@@ -86,13 +89,24 @@ ELFImage* ELFLoader::load_elf_from_vfs(VFS::Node* node)
             kdbgln("Loading loadable segment at address %lx, file size %ld, mem size %ld, permissions %s", phdr.p_vaddr,
                    phdr.p_filesz, phdr.p_memsz, format_permissions(phdr.p_flags));
             ASSERT(phdr.p_vaddr);
+
             uint64_t pages = Utilities::get_blocks_from_size(PAGE_SIZE, (phdr.p_vaddr % PAGE_SIZE) + phdr.p_memsz);
-            void* buffer =
+            void* buffer = (void*)((uint64_t)MemoryManager::get_pages_at(round_down_to_nearest_page(phdr.p_vaddr),
-                (void*)((uint64_t)MemoryManager::get_pages_at(round_down_to_nearest_page(phdr.p_vaddr), pages,
+                                                                         pages, MAP_READ_WRITE) +
-                                                              phdr.p_flags & 2 ? MAP_READ_WRITE | MAP_USER : MAP_USER) +
+                                   (phdr.p_vaddr % PAGE_SIZE));
-                        (phdr.p_vaddr % PAGE_SIZE));
+
+            if (VMM::is_using_kernel_address_space()) { VMM::switch_to_previous_user_address_space(); }
+            VMM::apply_address_space();
+
             VFS::read(node, phdr.p_offset, phdr.p_filesz, (char*)buffer);
             memset((void*)((uint64_t)buffer + phdr.p_filesz), 0, phdr.p_memsz - phdr.p_filesz);
+
+            VMM::switch_back_to_kernel_address_space();
+            VMM::apply_address_space();
+            VMM::switch_to_previous_user_address_space();
+
+            MemoryManager::protect(buffer, pages, phdr.p_flags & 2 ? MAP_READ_WRITE | MAP_USER : MAP_USER);
+
             image = (ELFImage*)krealloc(image, (sizeof(ELFImage) - sizeof(ELFSection)) +
                                                    (image->section_count + 1) * sizeof(ELFSection));
             ELFSection& section = image->sections[image->section_count];

kernel/src/sys/exec.cpp

@@ -5,30 +5,38 @@
 #include "interrupts/Interrupts.h"
 #include "memory/MemoryManager.h"
 #include "memory/PMM.h"
+#include "memory/VMM.h"
 #include "std/stdlib.h"
 #include "std/string.h"
+#include "sys/Syscall.h"
 #include "sys/elf/ELFLoader.h"
 #include "thread/Scheduler.h"
 
 void sys_exec(Context* context, const char* pathname)
 {
-    if (!pathname)
+    /*context->rax = -ENOSYS; // FIXME: Make exec() work under separate address spaces.
+    return;*/
+
+    char* kpathname = Syscall::strdup_from_user(pathname);
+    if (!kpathname)
     {
         context->rax = -EFAULT;
         return;
     }
 
-    kinfoln("exec(): executing %s", pathname);
+    kinfoln("exec(): executing %s", kpathname);
 
-    VFS::Node* program = VFS::resolve_path(pathname);
+    VFS::Node* program = VFS::resolve_path(kpathname);
     if (!program)
     {
+        kfree(kpathname);
         context->rax = -ENOENT;
         return;
     }
 
     if (program->type == VFS_DIRECTORY)
     {
+        kfree(kpathname);
         context->rax = -EISDIR;
         return;
     }
@@ -36,6 +44,7 @@ void sys_exec(Context* context, const char* pathname)
     long memusage;
     if ((memusage = ELFLoader::check_elf_image(program)) < 0)
     {
+        kfree(kpathname);
         context->rax = -ENOEXEC;
         return;
     }
@@ -43,12 +52,16 @@ void sys_exec(Context* context, const char* pathname)
     uint64_t allocated_stack = (uint64_t)MemoryManager::get_pages(TASK_PAGES_IN_STACK, MAP_READ_WRITE | MAP_USER);
     if (!allocated_stack)
     {
+        kfree(kpathname);
         context->rax = -ENOMEM;
         return;
     }
 
+    uint64_t allocated_stack_phys = VMM::get_physical(allocated_stack);
+
     if ((uint64_t)memusage > PMM::get_free())
     {
+        kfree(kpathname);
         MemoryManager::release_pages((void*)allocated_stack, TASK_PAGES_IN_STACK);
         context->rax = -ENOMEM;
         return;
@@ -62,19 +75,22 @@ void sys_exec(Context* context, const char* pathname)
 
     // At this point, pretty much nothing can fail.
 
-    ELFLoader::release_elf_image(Scheduler::current_task()->image);
-
     ELFImage* image = ELFLoader::load_elf_from_vfs(program);
     ASSERT(image); // If check_elf_image succeeded, load_elf_from_vfs MUST succeed, unless something has gone terribly
                    // wrong.
 
-    MemoryManager::release_pages((void*)task->allocated_stack, TASK_PAGES_IN_STACK);
-
     task->allocated_stack = allocated_stack;
 
+    for (uint64_t i = 0; i < TASK_PAGES_IN_STACK; i++)
+    {
+        VMM::map(allocated_stack + (i * PAGE_SIZE), allocated_stack_phys + (i * PAGE_SIZE), MAP_READ_WRITE | MAP_USER);
+    }
+
     Scheduler::reset_task(task, image);
 
     set_context_from_task(*task, context);
 
+    kfree(kpathname);
+
     return;
 }

kernel/src/sys/stdio.cpp

@@ -4,7 +4,10 @@
 #include "interrupts/Context.h"
 #include "io/Serial.h"
 #include "log/Log.h"
+#include "memory/VMM.h"
 #include "render/TextRenderer.h"
+#include "std/stdlib.h"
+#include "sys/Syscall.h"
 #include "thread/Scheduler.h"
 #include "thread/Task.h"
 
@@ -79,7 +82,7 @@ void sys_write(Context* context, int fd, size_t size, const char* addr)
         context->rax = -EBADF;
         return;
     }
-    ssize_t result = current_task->files[fd].write(size, addr);
+    ssize_t result = current_task->files[fd].write(size, (const char*)VMM::get_physical((uint64_t)addr));
     context->rax = (size_t)result;
     return;
 }
@@ -99,9 +102,17 @@ void sys_open(Context* context, const char* filename, int flags)
         return;
     }
 
-    VFS::Node* node = VFS::resolve_path(filename);
+    char* kernel_filename = Syscall::strdup_from_user(filename);
+    if (!kernel_filename)
+    {
+        context->rax = -EFAULT;
+        return;
+    }
+
+    VFS::Node* node = VFS::resolve_path(kernel_filename);
     if (!node)
     {
+        kfree(kernel_filename);
         context->rax = -ENOENT;
         return;
     }
@@ -110,16 +121,18 @@ void sys_open(Context* context, const char* filename, int flags)
     bool can_write = (flags & OPEN_WRITE) > 0;
     if (!can_read && !can_write)
     {
+        kfree(kernel_filename);
         context->rax = -EINVAL;
         return;
     }
 
-    kdbgln("open(): opening %s %s, allocated file descriptor %d", filename,
+    kdbgln("open(): opening %s %s, allocated file descriptor %d", kernel_filename,
            (can_read && can_write) ? "rw"
            : can_read              ? "r-"
                                    : "-w",
            fd);
 
+    kfree(kernel_filename);
     current_task->files[fd].open(node, can_read, can_write);
     context->rax = fd;
     return;
@@ -143,7 +156,7 @@ void sys_read(Context* context, int fd, size_t size, char* buffer)
         context->rax = -EBADF;
         return;
     }
-    ssize_t result = current_task->files[fd].read(size, buffer);
+    ssize_t result = current_task->files[fd].read(size, (char*)VMM::get_physical((uint64_t)buffer));
     context->rax = (size_t)result;
     return;
 }

kernel/src/thread/Scheduler.cpp

@@ -87,21 +87,16 @@ void Scheduler::add_kernel_task(void (*task)(void))
             new_task->id, new_task->regs.rsp, task_num);
 }
 
-void Scheduler::add_user_task(void* task)
+Task* Scheduler::create_user_task()
 {
     Task* new_task = new Task;
     ASSERT(new_task);
     memset(&new_task->regs, 0, sizeof(Context));
     new_task->user_task = true;
     new_task->id = free_tid++;
-    new_task->regs.rip = (uint64_t)task;
     new_task->allocated_stack = (uint64_t)MemoryManager::get_pages(
         TASK_PAGES_IN_STACK, MAP_READ_WRITE | MAP_USER); // 16 KB is enough for everyone, right?
     new_task->regs.rsp = get_top_of_stack(new_task->allocated_stack, TASK_PAGES_IN_STACK);
-    new_task->regs.cs = 0x18 | 0x03;
-    new_task->regs.ss = 0x20 | 0x03;
-    new_task->regs.ds = 0x20 | 0x03;
-    new_task->regs.rflags = (1 << 21) | (1 << 9); // enable interrupts
     new_task->task_sleep = 0;
     new_task->task_time = 0;
     new_task->cpu_time = 0;
@@ -110,10 +105,7 @@ void Scheduler::add_user_task(void* task)
     base_task->prev_task = new_task;
     new_task->next_task = base_task;
     end_task = new_task;
-    new_task->state = new_task->Running;
     task_num++;
-    kinfoln("Adding user task: starts at %lx, tid %ld, stack at %lx, total tasks: %ld", new_task->regs.rip,
-            new_task->id, new_task->regs.rsp, task_num);
 }
 
 void Scheduler::load_user_task(const char* filename)
@@ -130,6 +122,8 @@ void Scheduler::load_user_task(const char* filename)
     ASSERT(new_task);
     memset(&new_task->regs, 0, sizeof(Context));
     new_task->id = free_tid++;
+    new_task->address_space = AddressSpace::create();
+    VMM::switch_to_user_address_space(new_task->address_space);
     ELFImage* image = ELFLoader::load_elf_from_filesystem(
         filename); // FIXME: TOCTOU? Right now, impossible, since interrupts are disabled and SMP is not a thing. But in
                    // the future, it might be possible.
@@ -156,6 +150,7 @@ void Scheduler::load_user_task(const char* filename)
     task_num++;
     kinfoln("Adding user task: loaded at %lx, tid %ld, stack at %lx, total tasks: %ld", new_task->regs.rip,
             new_task->id, new_task->regs.rsp, task_num);
+    VMM::switch_back_to_kernel_address_space();
     Interrupts::pop();
 }
 
@@ -182,12 +177,27 @@ void Scheduler::reap_task(Task* task)
     task_num--;
     Task* exiting_task = task;
     ASSERT(task->id != 0); // WHY IN THE WORLD WOULD WE BE REAPING THE IDLE TASK?
+    if (exiting_task->is_user_task())
+    {
+        VMM::switch_back_to_kernel_address_space();
+        VMM::apply_address_space();
+        VMM::switch_to_user_address_space(exiting_task->address_space);
+    }
     kinfoln("reaping task %ld, exited with code %ld", exiting_task->id, exiting_task->exit_status);
     if (exiting_task->allocated_stack)
         MemoryManager::release_pages((void*)exiting_task->allocated_stack, TASK_PAGES_IN_STACK);
     if (exiting_task->image) // FIXME: Also free pages the task has mmap-ed but not munmap-ed.
     {
-        ELFLoader::release_elf_image(exiting_task->image);
+        // ELFLoader::release_elf_image(exiting_task->image);
+        kfree(exiting_task->image);
+    }
+    if (exiting_task->is_user_task())
+    {
+        VMM::switch_back_to_kernel_address_space();
+        VMM::apply_address_space();
+        Interrupts::push_and_enable();
+        exiting_task->address_space.destroy();
+        Interrupts::pop();
     }
     for (int i = 0; i < TASK_MAX_FDS; i++) { exiting_task->files[i].close(); }
     delete exiting_task;
@@ -316,7 +326,17 @@ void Scheduler::task_yield(Context* context)
                 {
                     task_save_floating(*original_task);
                 }
-                if (sched_current_task->is_user_task()) { task_restore_floating(*sched_current_task); }
+                if (sched_current_task->is_user_task())
+                {
+                    VMM::switch_to_user_address_space(sched_current_task->address_space);
+                    VMM::apply_address_space();
+                    task_restore_floating(*sched_current_task);
+                }
+                else if (!was_idle && original_task->is_user_task() && !sched_current_task->is_user_task())
+                {
+                    VMM::switch_back_to_kernel_address_space();
+                    VMM::apply_address_space();
+                }
             }
             sched_current_task->task_time = 20;
             set_context_from_task(*sched_current_task, context);