elf: working loader / syscall: introduce the syscall api.

kernel/src/exec/elf.c

@@ -1,8 +1,13 @@
+#include "exec/exec.h"
+#include "mm/memop.h"
+#include "mm/pmm.h"
+#include "mm/vmm.h"
+#include <mm/liballoc/liballoc.h>
 #include <exec/elf.h>
 #include <stdint.h>
 #include <sys/log.h>
 
-void elf_load(char *data) {
+program_t *elf_load(char *data) {
     Elf64_Ehdr *ehdr = (Elf64_Ehdr*)data;
     if (   ehdr->e_ident[EI_MAG0] != ELFMAG0
         || ehdr->e_ident[EI_MAG1] != ELFMAG1
@@ -10,7 +15,7 @@ void elf_load(char *data) {
         || ehdr->e_ident[EI_MAG3] != ELFMAG3)
     {
         log("elf - loading failed: magic is incorrect\n");
-        return;
+        return NULL;
     }
 
     log("elf - e_ident[EI_DATA]: %d\n", ehdr->e_ident[EI_DATA]);
@@ -19,25 +24,100 @@ void elf_load(char *data) {
     log("elf - e_machine: %d\n", ehdr->e_machine);
     log("elf - e_entry: %p\n", ehdr->e_entry);
     log("elf - e_type: %p\n", ehdr->e_type);
+    log("elf - e_phnum: %p\n", ehdr->e_phnum);
 
     if (   ehdr->e_ident[EI_CLASS] != ELFCLASS64
         || ehdr->e_machine != EM_X86_64)
     {
         log("elf - loading failed: is the file built for amd64?\n");
-        return;
+        return NULL;
     }
 
     if (ehdr->e_type != ET_EXEC)
     {
         log("elf - loading failed: ELF type isn't ET_EXEC\n");
-        return;
+        return NULL;
     }
 
-    Elf64_Phdr *phdr = (Elf64_Phdr*)data + ehdr->e_phoff;
+    // There's the interesing part
-    for (uint16_t i = 0; i < ehdr->e_phnum; i++) {
+    elf_program_t *ret = malloc(sizeof(elf_program_t)); // Allocate memory for the program.
+
+    pagemap_t *pm = vmm_alloc_pm(); // Allocate a pagemap, so that we can map the program inside.
+    ret->program.pm = pm;
+    ret->program.entry = ehdr->e_entry;
+    ret->ehdr = ehdr;
+
+    Elf64_Phdr *phdr = (Elf64_Phdr *)((uint8_t *)data + ehdr->e_phoff);
+
+    for (uint16_t i = 0; i <= ehdr->e_phnum; i++)
+    {
+        log("elf - ELF segment type: %d\n", phdr[i].p_type);
         if (phdr[i].p_type != PT_LOAD)
             continue;
 
-        log("elf - Loadable program header: vaddr: %p", phdr[i].p_vaddr);
+        uint64_t vaddr_start = ALIGN_DOWN(phdr[i].p_vaddr, PMM_PAGE_SIZE);
+        uint64_t vaddr_end = ALIGN_UP(phdr[i].p_vaddr + phdr[i].p_memsz, PMM_PAGE_SIZE);
+        uint64_t offset = phdr[i].p_offset;
+
+        uint64_t flags = VMM_PRESENT;
+        if (phdr[i].p_flags & PF_W)
+            flags |= VMM_WRITABLE;
+        if (!(phdr[i].p_flags & PF_X))
+            flags |= VMM_NX;
+
+        flags |= VMM_USER; // User mode access
+
+        log("elf - loading ELF program header %u: vaddr 0x%llx - 0x%llx, offset 0x%llx, filesz 0x%llx, size 0x%llx, flags 0x%llx\n",
+              i, vaddr_start, vaddr_end, offset, phdr[i].p_filesz, phdr[i].p_memsz, flags);
+
+        uint64_t page_offset = phdr[i].p_vaddr & (PMM_PAGE_SIZE - 1);
+
+        for (uint64_t vaddr = vaddr_start; vaddr < vaddr_end; vaddr += PMM_PAGE_SIZE)
+        {
+            uint64_t phys = (uint64_t)pmm_request_page();
+            if (!phys)
+            {
+                log("elf - pmm page alloc failed. out of memory?\n");
+                return 0;
+            }
+
+            vmm_map(pm, vaddr, phys, flags);
+            memset((void *)HIGHER_HALF(phys), 0, PMM_PAGE_SIZE);
+
+            uint64_t file_page_offset = offset + (vaddr - vaddr_start);
+            uint64_t file_data_end = offset + phdr[i].p_filesz;
+
+            if (file_page_offset < file_data_end)
+            {
+                uint64_t bytes_from_start = vaddr - vaddr_start;
+                uint64_t page_data_offset = 0;
+
+                if (bytes_from_start == 0 && page_offset > 0)
+                {
+                    page_data_offset = page_offset;
+                }
+
+                uint64_t copy_offset = file_page_offset;
+                uint64_t copy_size = PMM_PAGE_SIZE - page_data_offset;
+
+                if (copy_offset + copy_size > file_data_end)
+                {
+                    copy_size = file_data_end - copy_offset;
+                }
+
+                if (copy_size > 0)
+                {
+                    void *dest = (void *)(HIGHER_HALF(phys) + page_data_offset);
+                    void *src = (uint8_t *)data + copy_offset;
+                    memcpy(dest, src, copy_size);
+
+                    //log("elf - copied 0x%llx bytes from ELF file offset 0x%llx to vaddr 0x%llx (phys 0x%llx)\n",
+                    //      copy_size, copy_offset, vaddr + page_data_offset, phys + page_data_offset);
+                }
+            }
+        }
     }
+
+    log("elf - loaded ELF program in memory.\n");
+    return (program_t*)ret;
 }

kernel/src/exec/elf.h

@@ -1,5 +1,6 @@
 #pragma once
 
+#include "exec/exec.h"
 #include <stdint.h>
 
 // ELF magic.
@@ -115,4 +116,10 @@ typedef struct {
     uint64_t   p_align;
 } Elf64_Phdr;
 
-void elf_load(char *data);
+typedef struct {
+    program_t program;
+
+    Elf64_Ehdr *ehdr;
+} elf_program_t;
+
+program_t *elf_load(char *data);

kernel/src/exec/exec.h

@@ -0,0 +1,25 @@
+#pragma once
+
+/*
+ * A program (or a script, if a shebang is found inside), literally.
+ */
+#include "mm/vmm.h"
+
+#define EXEC_TYPE_ELF 0
+
+typedef struct {
+    // The path to the program/script that will be executed.
+    char *path;
+
+    // The pagemap where the program's code is loaded.
+    pagemap_t *pm;
+
+    // The path to the first instruction. This will be passed to the new process's rip register.
+    uint64_t entry;
+
+    // The program type. Used to get additional, unneeded information out of a program
+    int type;
+
+    // That is what Soaplin needs to know. Executable file loaders are encouraged to extend this structure
+    // to include info such as the EHDR for the ELF loader...
+} program_t;

kernel/src/fs/vfs.h

@@ -1 +1,27 @@
 #pragma once
+
+// btw fuck sun
+
+#include <stdint.h>
+
+#define VNODE_TYPE_FILE 0
+#define VNODE_TYPE_DIR  1
+#define VNODE_TYPE_DEV  2
+
+typedef struct __vnode {
+    char name[128];
+    int  type;
+
+    struct __vnode *children;
+    struct __vnode *next;
+    struct __vnode *parent;
+
+    void(*write)(void *buffer, uint64_t off, uint64_t size);
+    void(*read)(void *buffer, uint64_t off, uint64_t size);
+} vnode;
+
+typedef struct __vfs_mount {
+    struct vfs_node *mount_point;       // The directory in the main VFS where this is mounted
+    struct vfs_node *mounted_root;      // The root node of the mounted filesystem
+    struct vfs_mount *next;             // Pointer to next mount point
+} vfs_mount;

kernel/src/main.c

@@ -1,4 +1,5 @@
 #include "exec/elf.h"
+#include "exec/exec.h"
 #include "mm/pmm.h"
 #include "mm/vma.h"
 #include "mm/vmm.h"
@@ -6,6 +7,7 @@
 #include "sched/sched.h"
 #include "sys/arch/x86_64/pit.h"
 #include "sys/arch/x86_64/sse.h"
+#include "sys/syscall.h"
 #include <sys/log.h>
 #include <stdint.h>
 #include <stddef.h>
@@ -116,7 +118,6 @@ void kmain(void) {
     gdt_init(&kstack[8192]);
     idt_init();
     fpu_activate();
-
     sse_init();
 
     pmm_init();
@@ -129,11 +130,7 @@ void kmain(void) {
             asm("hlt");
     }
 
-    char *a = malloc(1);
+    syscall_init();
-    *a = 32;
-    log("Allocated 1 byte at 0x%.16llx\n", (uint64_t)a);
-    free(a);
-
     pit_init(1000);
     sched_init();
     //user_init();
@@ -141,7 +138,8 @@ void kmain(void) {
     struct limine_file *f = module_request.response->modules[0];
     log("kmain - %s\n", f->path);
 
-    elf_load((char*)f->address);
+    program_t *prog = elf_load((char*)f->address);
+    sched_create("Init", prog->entry, prog->pm, SCHED_USER_PROCESS);
 
     log("kernel - Soaplin initialized sucessfully.\n");
     while (1)

kernel/src/sched/sched.c

@@ -11,6 +11,18 @@ sched_process *curr_proc;
 int current_pid = 0;
 int standby = 0;
 
+void map_range_to_pagemap(pagemap_t *dest_pagemap, pagemap_t *src_pagemap, uint64_t start, uint64_t size, uint64_t flags)
+{
+    for (uint64_t offset = 0; offset < size; offset += PMM_PAGE_SIZE)
+    {
+        uint64_t phys = virt_to_phys(src_pagemap, start + offset);
+        if (phys)
+        {
+            vmm_map(dest_pagemap, start + offset, phys, flags);
+        }
+    }
+}
+
 void sched_init() {
     // TODO: It may be good to implement heap memory to save space.
     
@@ -71,7 +83,7 @@ sched_process *sched_create(char *name, uint64_t entry_point, pagemap_t* pm, uin
         proc->regs.ss = 0x30;
     }
     else if (flags == SCHED_USER_PROCESS) {
-        proc->regs.cs = 0x43; // Run in kernel mode
+        proc->regs.cs = 0x43; // Run in user mode
         proc->regs.ss = 0x3B;
     }
     proc->regs.rflags = 0x202; // Enable interrupts
@@ -83,6 +95,8 @@ sched_process *sched_create(char *name, uint64_t entry_point, pagemap_t* pm, uin
 
     current_pid++;
 
+    map_range_to_pagemap(pm, vmm_kernel_pm, 0x1000, 0x10000, VMM_PRESENT | VMM_WRITABLE | VMM_USER);
+
     if (standby) {
         // Disable standby mode as there's actually something to
         // run, now.

kernel/src/sys/arch/x86_64/interrupts.c

@@ -5,6 +5,7 @@
 #include "sys/arch/x86_64/pic.h"
 #include "sys/arch/x86_64/rtc.h"
 #include "sys/log.h"
+#include "sys/syscall.h"
 //#include "sys/sched.h"
 #include <stdint.h>
 #include <sys/arch/x86_64/idt.h>
@@ -71,10 +72,7 @@ void exception_handler(registers_t *regs) {
     }
     else if (regs->int_no == 0x80)
     {
-        log("syscall - Hello World! Current process: %s\n", curr_proc->name);
+        syscall_handle(regs);
-        if (curr_proc->flags == SCHED_USER_PROCESS)
-            log("syscall - Btw we made it to userspace, baby!\n", curr_proc->name);
-    
     }
     //logln(info, "arch/ints", "Received interrupt %d\n", regs->int_no);
     pic_ack(regs->int_no - 32);

kernel/src/sys/syscall.c

@@ -0,0 +1,48 @@
+#include "sched/sched.h"
+#include <stdint.h>
+#include <stddef.h>
+#include <sys/log.h>
+#include <sys/syscall.h>
+
+static syscall syscall_table[1024];
+
+static int syscall_initialized = 0;
+
+// Stub function for undefined syscalls.
+static uint64_t __syscall_undefined() { return 0; }
+
+void syscall_handle(registers_t *regs) {
+    if (regs->rax > 1024) {
+        log("syscall - syscall_handle was called with rax better than what Soaplin supports (1024). did you forget to set rax?\n");
+        return;
+    }
+
+    if (curr_proc == NULL || curr_proc->regs.cs != 0x43) {
+        log("syscall - syscall_handle was called by the kernel. is this wanted?\n");
+        return;
+    }
+
+    if (syscall_table[regs->rax] == (syscall)__syscall_undefined) {
+        log("syscall - syscall_handle was called with an undefined system call. (%d)\n", regs->rax);
+        return;
+    }
+
+    regs->rax = syscall_table[regs->rax](regs->rsp, regs->rbp, regs->r12, regs->r13, regs->r14, regs->r15);
+    return;
+}
+
+void syscall_register(int id, syscall handler) {
+    if (syscall_table[id] != (syscall)__syscall_undefined)
+    {
+        log("syscall - warning: syscall_register has been called to try replacing an existing syscall.\n");
+        return;
+    }
+
+    syscall_table[id] = handler;
+    log("syscall - System call %d has been set to %p", handler);
+}
+
+void syscall_init() {
+    for (int i = 0; i < 1024; i++)
+        syscall_table[i] = (syscall)__syscall_undefined;
+}

kernel/src/sys/syscall.h

@@ -0,0 +1,18 @@
+#pragma once
+
+#include "sys/arch/x86_64/idt.h"
+#include <stdint.h>
+
+/// A function that defines a system call.
+/// NOTE: Arguments are defined as uint64_t, but you can simply put your own type, and then cast your function to syscall.
+typedef uint64_t(*syscall)(uint64_t rsp, uint64_t rbp, uint64_t r12, uint64_t r13, uint64_t r14, uint64_t r15);
+
+/// Registers a system call.
+/// NOTE: an existing system call cannot be replaced by another.
+void syscall_register(int id, syscall handler);
+
+/// Called by the interupt handler, or the "syscall" instruction handler
+void syscall_handle(registers_t*);
+
+/// Initialize the system calls.
+void syscall_init();

kernel/src/sys/uname.h

@@ -0,0 +1,13 @@
+#pragma once
+
+typedef struct __uname {
+    char sysname[128];    /* Operating system name */
+    char nodename[128];   /* Name within communications network
+                          to which the node is attached, if any */
+    char release[128];    /* Operating system release */
+    char version[128];    /* Operating system version */
+    char machine[128];    /* Hardware type identifier */
+#ifdef _GNU_SOURCE
+    char domainname[]; /* NIS or YP domain name */
+#endif
+} uname_t;

testing/link.ld

@@ -1,20 +1,23 @@
+OUTPUT_FORMAT("elf64-x86-64")
+ENTRY(_start)
+
 SECTIONS {
-  . = 0x1000;
+    . = 0x100000;
 
-  .text : {
+    .text : {
-    *(.text*)
+        *(.text)
-  }
+    } 
     
-  .rodata : {
+    .rodata : {
-    *(.rodata*)
+        *(.rodata)
-  }
+    }
 
-  .data : {
+    .data : {
-    *(.data*)
+        *(.data)
-  }
+    }
 
-  .bss : {
+    .bss : {
-    *(.bss*)
+        *(COMMON)
-    *(COMMON)
+        *(.bss)
-  }
+    }
 }