toaruos/kernel/sys/syscall.c

1116 lines
28 KiB
C

/* vim: tabstop=4 shiftwidth=4 noexpandtab
* This file is part of ToaruOS and is released under the terms
* of the NCSA / University of Illinois License - see LICENSE.md
* Copyright (C) 2011-2018 K. Lange
* Copyright (C) 2012 Markus Schober
*
* Syscall Tables
*
*/
#include <kernel/system.h>
#include <kernel/process.h>
#include <kernel/logging.h>
#include <kernel/fs.h>
#include <kernel/pipe.h>
#include <kernel/version.h>
#include <kernel/shm.h>
#include <kernel/printf.h>
#include <kernel/module.h>
#include <sys/utsname.h>
#include <syscall_nums.h>
static char hostname[256];
static size_t hostname_len = 0;
#define FD_INRANGE(FD) \
((FD) < (int)current_process->fds->length && (FD) >= 0)
#define FD_ENTRY(FD) \
(current_process->fds->entries[(FD)])
#define FD_CHECK(FD) \
(FD_INRANGE(FD) && FD_ENTRY(FD))
#define FD_OFFSET(FD) \
(current_process->fds->offsets[(FD)])
#define FD_MODE(FD) \
(current_process->fds->modes[(FD)])
#define PTR_INRANGE(PTR) \
((uintptr_t)(PTR) > current_process->image.entry)
#define PTR_VALIDATE(PTR) \
ptr_validate((void *)(PTR), __func__)
static void ptr_validate(void * ptr, const char * syscall) {
if (ptr && !PTR_INRANGE(ptr)) {
debug_print(ERROR, "SEGFAULT: invalid pointer passed to %s. (0x%x < 0x%x)",
syscall, (uintptr_t)ptr, current_process->image.entry);
HALT_AND_CATCH_FIRE("Segmentation fault", NULL);
}
}
void validate(void * ptr) {
ptr_validate(ptr, "syscall");
}
/*
* Exit the current task.
* DOES NOT RETURN!
*/
static int __attribute__((noreturn)) sys_exit(int retval) {
/* Deschedule the current task */
task_exit((retval & 0xFF) << 8);
for (;;) ;
}
static int sys_read(int fd, char * ptr, int len) {
if (FD_CHECK(fd)) {
PTR_VALIDATE(ptr);
fs_node_t * node = FD_ENTRY(fd);
if (!(FD_MODE(fd) & 01)) {
debug_print(WARNING, "access denied (read, fd=%d, mode=%d, %s, %s)", fd, FD_MODE(fd), node->name, current_process->name);
return -EACCES;
}
uint32_t out = read_fs(node, FD_OFFSET(fd), len, (uint8_t *)ptr);
FD_OFFSET(fd) += out;
return (int)out;
}
return -EBADF;
}
static int sys_ioctl(int fd, int request, void * argp) {
if (FD_CHECK(fd)) {
PTR_VALIDATE(argp);
return ioctl_fs(FD_ENTRY(fd), request, argp);
}
return -EBADF;
}
static int sys_readdir(int fd, int index, struct dirent * entry) {
if (FD_CHECK(fd)) {
PTR_VALIDATE(entry);
struct dirent * kentry = readdir_fs(FD_ENTRY(fd), (uint32_t)index);
if (kentry) {
memcpy(entry, kentry, sizeof *entry);
free(kentry);
return 1;
} else {
return 0;
}
}
return -EBADF;
}
static int sys_write(int fd, char * ptr, int len) {
if (FD_CHECK(fd)) {
PTR_VALIDATE(ptr);
fs_node_t * node = FD_ENTRY(fd);
if (!(FD_MODE(fd) & 02)) {
debug_print(WARNING, "access denied (write, fd=%d)", fd);
return -EACCES;
}
uint32_t out = write_fs(node, FD_OFFSET(fd), len, (uint8_t *)ptr);
FD_OFFSET(fd) += out;
return out;
}
return -EBADF;
}
static int sys_waitpid(int pid, int * status, int options) {
if (status && !PTR_INRANGE(status)) {
return -EINVAL;
}
return waitpid(pid, status, options);
}
static int sys_open(const char * file, int flags, int mode) {
PTR_VALIDATE(file);
debug_print(NOTICE, "open(%s) flags=0x%x; mode=0x%x", file, flags, mode);
fs_node_t * node = kopen((char *)file, flags);
int access_bits = 0;
if (!(flags & O_WRONLY) || (flags & O_RDWR)) {
if (node && !has_permission(node, 04)) {
debug_print(WARNING, "access denied (read, sys_open, file=%s)", file);
close_fs(node);
return -EACCES;
} else {
access_bits |= 01;
}
}
if ((flags & O_RDWR) || (flags & O_APPEND) || (flags & O_WRONLY) || (flags & O_TRUNC)) {
if (node && !has_permission(node, 02)) {
debug_print(WARNING, "access denied (write, sys_open, file=%s)", file);
close_fs(node);
return -EACCES;
}
if (node) {
truncate_fs(node);
}
access_bits |= 02;
}
if (!node && (flags & O_CREAT)) {
/* TODO check directory permissions */
debug_print(NOTICE, "- file does not exist and create was requested.");
/* Um, make one */
int result = create_file_fs((char *)file, mode);
if (!result) {
node = kopen((char *)file, flags);
} else {
return result;
}
}
if (!node) {
debug_print(NOTICE, "File does not exist; someone should be setting errno?");
return -ENOENT;
}
int fd = process_append_fd((process_t *)current_process, node);
FD_MODE(fd) = access_bits;
if (flags & O_APPEND) {
FD_OFFSET(fd) = node->length;
} else {
FD_OFFSET(fd) = 0;
}
debug_print(INFO, "[open] pid=%d %s -> %d", getpid(), file, fd);
return fd;
}
static int sys_access(const char * file, int flags) {
PTR_VALIDATE(file);
debug_print(INFO, "access(%s, 0x%x) from pid=%d", file, flags, getpid());
fs_node_t * node = kopen((char *)file, 0);
if (!node) return -ENOENT;
close_fs(node);
return 0;
}
static int sys_close(int fd) {
if (FD_CHECK(fd)) {
close_fs(FD_ENTRY(fd));
FD_ENTRY(fd) = NULL;
return 0;
}
return -EBADF;
}
static int sys_sbrk(int size) {
process_t * proc = (process_t *)current_process;
if (proc->group != 0) {
proc = process_from_pid(proc->group);
}
spin_lock(proc->image.lock);
uintptr_t ret = proc->image.heap;
uintptr_t i_ret = ret;
ret = (ret + 0xfff) & ~0xfff; /* Rounds ret to 0x1000 in O(1) */
proc->image.heap += (ret - i_ret) + size;
while (proc->image.heap > proc->image.heap_actual) {
proc->image.heap_actual += 0x1000;
assert(proc->image.heap_actual % 0x1000 == 0);
alloc_frame(get_page(proc->image.heap_actual, 1, current_directory), 0, 1);
invalidate_tables_at(proc->image.heap_actual);
}
spin_unlock(proc->image.lock);
return ret;
}
static int sys_getpid(void) {
/* The user actually wants the pid of the originating thread (which can be us). */
if (current_process->group) {
return current_process->group;
} else {
/* We are the leader */
return current_process->id;
}
}
/* Actual getpid() */
static int sys_gettid(void) {
return getpid();
}
static int sys_execve(const char * filename, char *const argv[], char *const envp[]) {
PTR_VALIDATE(argv);
PTR_VALIDATE(filename);
PTR_VALIDATE(envp);
debug_print(WARNING, "%d = exec(%s", current_process->id, filename);
for (char * const * arg = argv; *arg; ++arg) {
debug_print(WARNING, " %s", *arg);
}
debug_print(WARNING, " )");
int argc = 0;
int envc = 0;
while (argv[argc]) {
PTR_VALIDATE(argv[argc]);
++argc;
}
if (envp) {
while (envp[envc]) {
PTR_VALIDATE(envp[envc]);
++envc;
}
}
debug_print(INFO, "Allocating space for arguments...");
char ** argv_ = malloc(sizeof(char *) * (argc + 1));
for (int j = 0; j < argc; ++j) {
argv_[j] = malloc((strlen(argv[j]) + 1) * sizeof(char));
memcpy(argv_[j], argv[j], strlen(argv[j]) + 1);
}
argv_[argc] = 0;
char ** envp_;
if (envp && envc) {
envp_ = malloc(sizeof(char *) * (envc + 1));
for (int j = 0; j < envc; ++j) {
envp_[j] = malloc((strlen(envp[j]) + 1) * sizeof(char));
memcpy(envp_[j], envp[j], strlen(envp[j]) + 1);
}
envp_[envc] = 0;
} else {
envp_ = malloc(sizeof(char *));
envp_[0] = NULL;
}
debug_print(INFO,"Releasing all shmem regions...");
shm_release_all((process_t *)current_process);
current_process->cmdline = argv_;
debug_print(INFO,"Executing...");
/* Discard envp */
return exec((char *)filename, argc, (char **)argv_, (char **)envp_);
}
static int sys_seek(int fd, int offset, int whence) {
if (FD_CHECK(fd)) {
if (fd < 3) {
return 0;
}
switch (whence) {
case 0:
FD_OFFSET(fd) = offset;
break;
case 1:
FD_OFFSET(fd) += offset;
break;
case 2:
FD_OFFSET(fd) = FD_ENTRY(fd)->length + offset;
break;
}
return FD_OFFSET(fd);
}
return -EBADF;
}
static int stat_node(fs_node_t * fn, uintptr_t st) {
struct stat * f = (struct stat *)st;
PTR_VALIDATE(f);
if (!fn) {
memset(f, 0x00, sizeof(struct stat));
debug_print(INFO, "stat: This file doesn't exist");
return -1;
}
f->st_dev = (uint16_t)(((uint32_t)fn->device & 0xFFFF0) >> 8);
f->st_ino = fn->inode;
uint32_t flags = 0;
if (fn->flags & FS_FILE) { flags |= _IFREG; }
if (fn->flags & FS_DIRECTORY) { flags |= _IFDIR; }
if (fn->flags & FS_CHARDEVICE) { flags |= _IFCHR; }
if (fn->flags & FS_BLOCKDEVICE) { flags |= _IFBLK; }
if (fn->flags & FS_PIPE) { flags |= _IFIFO; }
if (fn->flags & FS_SYMLINK) { flags |= _IFLNK; }
f->st_mode = fn->mask | flags;
f->st_nlink = fn->nlink;
f->st_uid = fn->uid;
f->st_gid = fn->gid;
f->st_rdev = 0;
f->st_size = fn->length;
f->st_atime = fn->atime;
f->st_mtime = fn->mtime;
f->st_ctime = fn->ctime;
f->st_blksize = 512; /* whatever */
if (fn->get_size) {
f->st_size = fn->get_size(fn);
}
return 0;
}
static int sys_statf(char * file, uintptr_t st) {
int result;
PTR_VALIDATE(file);
PTR_VALIDATE(st);
fs_node_t * fn = kopen(file, 0);
result = stat_node(fn, st);
if (fn) {
close_fs(fn);
}
return result;
}
static int sys_chmod(char * file, int mode) {
int result;
PTR_VALIDATE(file);
fs_node_t * fn = kopen(file, 0);
if (fn) {
/* Can group members change bits? I think it's only owners. */
if (current_process->user != 0 && current_process->user != fn->uid) {
close_fs(fn);
return -EACCES;
}
result = chmod_fs(fn, mode);
close_fs(fn);
return result;
} else {
return -ENOENT;
}
}
static int sys_chown(char * file, int uid, int gid) {
int result;
PTR_VALIDATE(file);
fs_node_t * fn = kopen(file, 0);
if (fn) {
/* TODO: Owners can change groups... */
if (current_process->user != 0) {
close_fs(fn);
return -EACCES;
}
result = chown_fs(fn, uid, gid);
close_fs(fn);
return result;
} else {
return -ENOENT;
}
}
static int sys_stat(int fd, uintptr_t st) {
PTR_VALIDATE(st);
if (FD_CHECK(fd)) {
return stat_node(FD_ENTRY(fd), st);
}
return -EBADF;
}
static int sys_mkpipe(void) {
fs_node_t * node = make_pipe(4096 * 2);
open_fs(node, 0);
int fd = process_append_fd((process_t *)current_process, node);
FD_MODE(fd) = 03; /* read write */
return fd;
}
static int sys_dup2(int old, int new) {
return process_move_fd((process_t *)current_process, old, new);
}
static int sys_getuid(void) {
return current_process->real_user;
}
static int sys_geteuid(void) {
return current_process->user;
}
static int sys_setuid(user_t new_uid) {
if (current_process->user == USER_ROOT_UID) {
current_process->user = new_uid;
current_process->real_user = new_uid;
return 0;
}
return -EPERM;
}
static int sys_uname(struct utsname * name) {
PTR_VALIDATE(name);
char version_number[256];
sprintf(version_number, __kernel_version_format,
__kernel_version_major,
__kernel_version_minor,
__kernel_version_lower,
__kernel_version_suffix);
char version_string[256];
sprintf(version_string, "%s %s %s",
__kernel_version_codename,
__kernel_build_date,
__kernel_build_time);
strcpy(name->sysname, __kernel_name);
strcpy(name->nodename, hostname);
strcpy(name->release, version_number);
strcpy(name->version, version_string);
strcpy(name->machine, __kernel_arch);
strcpy(name->domainname, "");
return 0;
}
static int sys_signal(uint32_t signum, uintptr_t handler) {
if (signum > NUMSIGNALS) {
return -EINVAL;
}
uintptr_t old = current_process->signals.functions[signum];
current_process->signals.functions[signum] = handler;
return (int)old;
}
/*
static void inspect_memory (uintptr_t vaddr) {
// Please use this scary hack of a function as infrequently as possible.
shmem_debug_frame(vaddr);
}
*/
extern void idt_load(uint32_t *);
static int sys_reboot(void) {
debug_print(NOTICE, "[kernel] Reboot requested from process %d by user #%d", current_process->id, current_process->user);
if (current_process->user != USER_ROOT_UID) {
return -EPERM;
} else {
debug_print(ERROR, "[kernel] Good bye!");
/* Goodbye, cruel world */
IRQ_OFF;
uintptr_t phys;
uint32_t * virt = (void*)kvmalloc_p(0x1000, &phys);
virt[0] = 0;
virt[1] = 0;
virt[2] = 0;
idt_load(virt);
uint8_t out = 0x02;
while ((out & 0x02) != 0) {
out = inportb(0x64);
}
outportb(0x64, 0xFE); /* Reset */
STOP;
}
return 0;
}
static int sys_chdir(char * newdir) {
PTR_VALIDATE(newdir);
char * path = canonicalize_path(current_process->wd_name, newdir);
fs_node_t * chd = kopen(path, 0);
if (chd) {
if ((chd->flags & FS_DIRECTORY) == 0) {
close_fs(chd);
return -ENOTDIR;
}
if (!has_permission(chd, 01)) {
close_fs(chd);
return -EACCES;
}
close_fs(chd);
free(current_process->wd_name);
current_process->wd_name = malloc(strlen(path) + 1);
memcpy(current_process->wd_name, path, strlen(path) + 1);
return 0;
} else {
return -ENOENT;
}
}
static int sys_getcwd(char * buf, size_t size) {
if (buf) {
PTR_VALIDATE(buf);
size_t len = strlen(current_process->wd_name) + 1;
return (int)memcpy(buf, current_process->wd_name, MIN(size, len));
}
return 0;
}
static int sys_sethostname(char * new_hostname) {
if (current_process->user == USER_ROOT_UID) {
PTR_VALIDATE(new_hostname);
size_t len = strlen(new_hostname) + 1;
if (len > 256) {
return -ENAMETOOLONG;
}
hostname_len = len;
memcpy(hostname, new_hostname, hostname_len);
return 0;
} else {
return -EPERM;
}
}
static int sys_gethostname(char * buffer) {
PTR_VALIDATE(buffer);
memcpy(buffer, hostname, hostname_len);
return hostname_len;
}
extern int mkdir_fs(char *name, uint16_t permission);
static int sys_mkdir(char * path, uint32_t mode) {
return mkdir_fs(path, mode);
}
/*
* Yield the rest of the quantum;
* useful for busy waiting and other such things
*/
static int sys_yield(void) {
switch_task(1);
return 1;
}
/*
* System Function
*/
static int sys_sysfunc(int fn, char ** args) {
/* System Functions are special debugging system calls */
if (current_process->user == USER_ROOT_UID) {
switch (fn) {
case 3:
debug_print(ERROR, "sync is currently unimplemented");
//ext2_disk_sync();
return 0;
case 4:
/* Request kernel output to file descriptor in arg0*/
debug_print(NOTICE, "Setting output to file object in process %d's fd=%d!", getpid(), (int)args);
debug_file = current_process->fds->entries[(int)args];
return 0;
case 5:
{
char *arg;
PTR_VALIDATE(args);
for (arg = args[0]; arg; arg++)
PTR_VALIDATE(arg);
debug_print(NOTICE, "Replacing process %d's file descriptors with pointers to %s", getpid(), (char *)args);
fs_node_t * repdev = kopen((char *)args, 0);
while (current_process->fds->length < 3) {
process_append_fd((process_t *)current_process, repdev);
}
FD_ENTRY(0) = repdev;
FD_ENTRY(1) = repdev;
FD_ENTRY(2) = repdev;
}
return 0;
case 6:
debug_print(WARNING, "writing contents of file %s to sdb", args[0]);
{
PTR_VALIDATE(args);
PTR_VALIDATE(args[0]);
fs_node_t * file = kopen((char *)args[0], 0);
if (!file) {
return -EINVAL;
}
size_t length = file->length;
uint8_t * buffer = malloc(length);
read_fs(file, 0, length, (uint8_t *)buffer);
close_fs(file);
debug_print(WARNING, "Finished reading file, going to write it now.");
fs_node_t * f = kopen("/dev/sdb", 0);
if (!f) {
return 1;
}
write_fs(f, 0, length, buffer);
free(buffer);
return 0;
}
case 7:
debug_print(NOTICE, "Spawning debug hook as child of process %d.", getpid());
if (debug_hook) {
fs_node_t * tty = FD_ENTRY(0);
return create_kernel_tasklet(debug_hook, "[kttydebug]", tty);
} else {
return -EINVAL;
}
case 8:
debug_print(NOTICE, "Loading module %s.", args[0]);
{
/* Check file existence */
fs_node_t * file = kopen(args[0], 0);
if (!file) {
return 1;
}
close_fs(file);
module_data_t * mod_info = module_load(args[0]);
if (!mod_info) {
return 2;
}
return 0;
}
}
}
switch (fn) {
/* The following functions are here to support the loader and are probably bad. */
case 9:
{
process_t * proc = (process_t *)current_process;
if (proc->group != 0) {
proc = process_from_pid(proc->group);
}
spin_lock(proc->image.lock);
/* Set new heap start */
proc->image.heap = (uintptr_t)args[0];
proc->image.heap_actual = proc->image.heap & 0xFFFFF000;
assert(proc->image.heap_actual % 0x1000 == 0);
alloc_frame(get_page(proc->image.heap_actual, 1, current_directory), 0, 1);
invalidate_tables_at(proc->image.heap_actual);
while (proc->image.heap > proc->image.heap_actual) {
proc->image.heap_actual += 0x1000;
alloc_frame(get_page(proc->image.heap_actual, 1, current_directory), 0, 1);
invalidate_tables_at(proc->image.heap_actual);
}
spin_unlock(proc->image.lock);
return 0;
}
case 10:
{
/* Load pages to fit region. */
uintptr_t address = (uintptr_t)args[0];
/* TODO: These virtual address bounds should be in a header somewhere */
if (address < 0x20000000) return -EINVAL;
/* TODO: Upper bounds */
size_t size = (size_t)args[1];
/* TODO: Other arguments for read/write? */
if (address & 0xFFF) {
size += address & 0xFFF;
address &= 0xFFFFF000;
}
process_t * proc = (process_t *)current_process;
if (proc->group != 0) {
proc = process_from_pid(proc->group);
}
spin_lock(proc->image.lock);
for (size_t x = 0; x < size; x += 0x1000) {
alloc_frame(get_page(address + x, 1, current_directory), 0, 1);
invalidate_tables_at(address + x);
}
spin_unlock(proc->image.lock);
return 0;
}
case 11:
{
/* Set command line (meant for threads to set descriptions) */
int count = 0;
char **arg = args;
PTR_VALIDATE(args);
while (*arg) {
PTR_VALIDATE(*args);
count++;
arg++;
}
/*
* XXX We have a pretty obvious leak in storing command lines, since
* we never free them! Unfortunately, at the moment, they are
* shared between different processes, so until that gets fixed
* we're going to be just as bad as the rest of the codebase and
* just not free the previous value.
*/
current_process->cmdline = malloc(sizeof(char*)*(count+1));
int i = 0;
while (i < count) {
current_process->cmdline[i] = strdup(args[i]);
i++;
}
current_process->cmdline[i] = NULL;
return 0;
}
case 12:
/*
* Print a debug message to the kernel console
* XXX: This probably should be a thing normal users can do.
*/
PTR_VALIDATE(args);
debug_print(WARNING, "0x%x 0x%x 0x%x 0x%x", args[0], args[1], args[2], args[3]);
_debug_print(args[0], (uintptr_t)args[1], (uint32_t)args[2], args[3] ? args[3] : "(null)");
return 0;
case 13:
/*
* Set VGA text-mode cursor location
* (Not actually used to place a cursor, we use this to move the cursor off screen)
*/
PTR_VALIDATE(args);
outportb(0x3D4, 14);
outportb(0x3D5, (unsigned int)args[0]);
outportb(0x3D4, 15);
outportb(0x3D5, (unsigned int)args[1]);
return 0;
default:
debug_print(ERROR, "Bad system function %d", fn);
break;
}
return -EINVAL; /* Bad system function or access failure */
}
static int sys_sleepabs(unsigned long seconds, unsigned long subseconds) {
/* Mark us as asleep until <some time period> */
sleep_until((process_t *)current_process, seconds, subseconds);
/* Switch without adding us to the queue */
switch_task(0);
if (seconds > timer_ticks || (seconds == timer_ticks && subseconds >= timer_subticks)) {
return 0;
} else {
return 1;
}
}
static int sys_sleep(unsigned long seconds, unsigned long subseconds) {
unsigned long s, ss;
relative_time(seconds, subseconds * 10, &s, &ss);
return sys_sleepabs(s, ss);
}
static int sys_umask(int mode) {
current_process->mask = mode & 0777;
return 0;
}
static int sys_unlink(char * file) {
PTR_VALIDATE(file);
return unlink_fs(file);
}
static int sys_fork(void) {
return (int)fork();
}
static int sys_clone(uintptr_t new_stack, uintptr_t thread_func, uintptr_t arg) {
if (!new_stack || !PTR_INRANGE(new_stack)) return -EINVAL;
if (!thread_func || !PTR_INRANGE(thread_func)) return -EINVAL;
return (int)clone(new_stack, thread_func, arg);
}
static int sys_shm_obtain(char * path, size_t * size) {
PTR_VALIDATE(path);
PTR_VALIDATE(size);
return (int)shm_obtain(path, size);
}
static int sys_shm_release(char * path) {
PTR_VALIDATE(path);
return shm_release(path);
}
static int sys_kill(pid_t process, uint32_t signal) {
if (process < -1) {
return group_send_signal(-process, signal, 0);
} else {
return send_signal(process, signal, 0);
}
}
static int sys_gettimeofday(struct timeval * tv, void * tz) {
PTR_VALIDATE(tv);
PTR_VALIDATE(tz);
return gettimeofday(tv, tz);
}
static int sys_openpty(int * master, int * slave, char * name, void * _ign0, void * size) {
/* We require a place to put these when we are done. */
if (!master || !slave) return -EINVAL;
if (master && !PTR_INRANGE(master)) return -EINVAL;
if (slave && !PTR_INRANGE(slave)) return -EINVAL;
if (size && !PTR_INRANGE(size)) return -EINVAL;
/* Create a new pseudo terminal */
fs_node_t * fs_master;
fs_node_t * fs_slave;
pty_create(size, &fs_master, &fs_slave);
/* Append the master and slave to the calling process */
*master = process_append_fd((process_t *)current_process, fs_master);
*slave = process_append_fd((process_t *)current_process, fs_slave);
FD_MODE(*master) = 03;
FD_MODE(*slave) = 03;
open_fs(fs_master, 0);
open_fs(fs_slave, 0);
/* Return success */
return 0;
}
static int sys_pipe(int pipes[2]) {
if (pipes && !PTR_INRANGE(pipes)) {
return -EFAULT;
}
fs_node_t * outpipes[2];
make_unix_pipe(outpipes);
open_fs(outpipes[0], 0);
open_fs(outpipes[1], 0);
pipes[0] = process_append_fd((process_t *)current_process, outpipes[0]);
pipes[1] = process_append_fd((process_t *)current_process, outpipes[1]);
FD_MODE(pipes[0]) = 03;
FD_MODE(pipes[1]) = 03;
return 0;
}
static int sys_mount(char * arg, char * mountpoint, char * type, unsigned long flags, void * data) {
/* TODO: Make use of flags and data from mount command. */
(void)flags;
(void)data;
if (current_process->user != USER_ROOT_UID) {
return -EPERM;
}
if (PTR_INRANGE(arg) && PTR_INRANGE(mountpoint) && PTR_INRANGE(type)) {
return vfs_mount_type(type, arg, mountpoint);
}
return -EFAULT;
}
static int sys_symlink(char * target, char * name) {
PTR_VALIDATE(target);
PTR_VALIDATE(name);
return symlink_fs(target, name);
}
static int sys_readlink(const char * file, char * ptr, int len) {
PTR_VALIDATE(file);
fs_node_t * node = kopen((char *) file, O_PATH | O_NOFOLLOW);
if (!node) {
return -ENOENT;
}
int rv = readlink_fs(node, ptr, len);
close_fs(node);
return rv;
}
static int sys_lstat(char * file, uintptr_t st) {
int result;
PTR_VALIDATE(file);
PTR_VALIDATE(st);
fs_node_t * fn = kopen(file, O_PATH | O_NOFOLLOW);
result = stat_node(fn, st);
if (fn) {
close_fs(fn);
}
return result;
}
static int sys_fswait(int c, int fds[]) {
PTR_VALIDATE(fds);
for (int i = 0; i < c; ++i) {
if (!FD_CHECK(fds[i])) return -EBADF;
}
fs_node_t ** nodes = malloc(sizeof(fs_node_t *)*(c+1));
for (int i = 0; i < c; ++i) {
nodes[i] = FD_ENTRY(fds[i]);
}
nodes[c] = NULL;
int result = process_wait_nodes((process_t *)current_process, nodes, -1);
free(nodes);
return result;
}
static int sys_fswait_timeout(int c, int fds[], int timeout) {
PTR_VALIDATE(fds);
for (int i = 0; i < c; ++i) {
if (!FD_CHECK(fds[i])) return -EBADF;
}
fs_node_t ** nodes = malloc(sizeof(fs_node_t *)*(c+1));
for (int i = 0; i < c; ++i) {
nodes[i] = FD_ENTRY(fds[i]);
}
nodes[c] = NULL;
int result = process_wait_nodes((process_t *)current_process, nodes, timeout);
free(nodes);
return result;
}
static int sys_setsid(void) {
if (current_process->job == current_process->group) {
return -EPERM;
}
current_process->session = current_process->group;
current_process->job = current_process->group;
return current_process->session;
}
static int sys_setpgid(pid_t pid, pid_t pgid) {
if (pgid < 0) {
return -EINVAL;
}
process_t * proc;
if (pid == 0) {
proc = (process_t*)current_process;
} else {
proc = process_from_pid(pid);
}
if (!proc) {
debug_print(WARNING, "not found");
return -ESRCH;
}
if (proc->session != current_process->session) {
debug_print(WARNING, "child is in different sesion");
return -EPERM;
}
if (proc->session == proc->group) {
debug_print(WARNING, "process is session leader");
return -EPERM;
}
if (pgid == 0) {
proc->job = proc->group;
} else {
process_t * pgroup = process_from_pid(pgid);
if (!pgroup) {
debug_print(WARNING, "bad session id");
return -EPERM;
}
if (pgroup->session != proc->session) {
debug_print(WARNING, "tried to move to different session");
return -EPERM;
}
proc->job = pgid;
}
return 0;
}
static int sys_getpgid(pid_t pid) {
process_t * proc;
if (pid == 0) {
proc = (process_t*)current_process;
} else {
proc = process_from_pid(pid);
}
if (!proc) {
return -ESRCH;
}
return proc->job;
}
/*
* System Call Internals
*/
static int (*syscalls[])() = {
/* System Call Table */
[SYS_EXT] = sys_exit,
[SYS_GETEUID] = sys_geteuid,
[SYS_OPEN] = sys_open,
[SYS_READ] = sys_read,
[SYS_WRITE] = sys_write,
[SYS_CLOSE] = sys_close,
[SYS_GETTIMEOFDAY] = sys_gettimeofday,
[SYS_EXECVE] = sys_execve,
[SYS_FORK] = sys_fork,
[SYS_GETPID] = sys_getpid,
[SYS_SBRK] = sys_sbrk,
[SYS_UNAME] = sys_uname,
[SYS_OPENPTY] = sys_openpty,
[SYS_SEEK] = sys_seek,
[SYS_STAT] = sys_stat,
[SYS_MKPIPE] = sys_mkpipe,
[SYS_DUP2] = sys_dup2,
[SYS_GETUID] = sys_getuid,
[SYS_SETUID] = sys_setuid,
[SYS_REBOOT] = sys_reboot,
[SYS_READDIR] = sys_readdir,
[SYS_CHDIR] = sys_chdir,
[SYS_GETCWD] = sys_getcwd,
[SYS_CLONE] = sys_clone,
[SYS_SETHOSTNAME] = sys_sethostname,
[SYS_GETHOSTNAME] = sys_gethostname,
[SYS_MKDIR] = sys_mkdir,
[SYS_SHM_OBTAIN] = sys_shm_obtain,
[SYS_SHM_RELEASE] = sys_shm_release,
[SYS_KILL] = sys_kill,
[SYS_SIGNAL] = sys_signal,
[SYS_GETTID] = sys_gettid,
[SYS_YIELD] = sys_yield,
[SYS_SYSFUNC] = sys_sysfunc,
[SYS_SLEEPABS] = sys_sleepabs,
[SYS_SLEEP] = sys_sleep,
[SYS_IOCTL] = sys_ioctl,
[SYS_ACCESS] = sys_access,
[SYS_STATF] = sys_statf,
[SYS_CHMOD] = sys_chmod,
[SYS_UMASK] = sys_umask,
[SYS_UNLINK] = sys_unlink,
[SYS_WAITPID] = sys_waitpid,
[SYS_PIPE] = sys_pipe,
[SYS_MOUNT] = sys_mount,
[SYS_SYMLINK] = sys_symlink,
[SYS_READLINK] = sys_readlink,
[SYS_LSTAT] = sys_lstat,
[SYS_FSWAIT] = sys_fswait,
[SYS_FSWAIT2] = sys_fswait_timeout,
[SYS_CHOWN] = sys_chown,
[SYS_SETSID] = sys_setsid,
[SYS_SETPGID] = sys_setpgid,
[SYS_GETPGID] = sys_getpgid,
};
uint32_t num_syscalls = sizeof(syscalls) / sizeof(*syscalls);
typedef uint32_t (*scall_func)(unsigned int, ...);
pid_t trace_pid = 0;
void syscall_handler(struct regs * r) {
if (r->eax >= num_syscalls) {
return;
}
uintptr_t location = (uintptr_t)syscalls[r->eax];
if (!location) {
return;
}
/* Update the syscall registers for this process */
current_process->syscall_registers = r;
if (trace_pid && current_process->id == trace_pid) {
debug_print(WARNING, "[syscall trace] %d (0x%x) 0x%x 0x%x 0x%x 0x%x 0x%x", r->eax, location, r->ebx, r->ecx, r->edx, r->esi, r->edi);
}
/* Call the syscall function */
scall_func func = (scall_func)location;
uint32_t ret = func(r->ebx, r->ecx, r->edx, r->esi, r->edi);
if ((current_process->syscall_registers == r) ||
(location != (uintptr_t)&fork && location != (uintptr_t)&clone)) {
r->eax = ret;
}
}
void syscalls_install(void) {
debug_print(NOTICE, "Initializing syscall table with %d functions", num_syscalls);
isrs_install_handler(0x7F, &syscall_handler);
}