blob: 9099d2c1c18f46e6950ea9793b3c99e3b3392d1c [file] [log] [blame]
/* See COPYRIGHT for copyright information. */
//#define DEBUG
#include <ros/common.h>
#include <ros/limits.h>
#include <arch/types.h>
#include <arch/arch.h>
#include <arch/mmu.h>
#include <arch/console.h>
#include <time.h>
#include <error.h>
#include <elf.h>
#include <string.h>
#include <assert.h>
#include <process.h>
#include <schedule.h>
#include <pmap.h>
#include <umem.h>
#include <mm.h>
#include <trap.h>
#include <syscall.h>
#include <kmalloc.h>
#include <profiler.h>
#include <stdio.h>
#include <hashtable.h>
#include <bitmask.h>
#include <smp.h>
#include <arsc_server.h>
#include <event.h>
#include <kprof.h>
#include <termios.h>
#include <manager.h>
#include <ros/procinfo.h>
#include <rcu.h>
static int execargs_stringer(struct proc *p, char *d, size_t slen,
char *path, size_t path_l,
char *argenv, size_t argenv_l);
/* Global, used by the kernel monitor for syscall debugging. */
bool systrace_loud = FALSE;
/* Helper, given the trace record, pretty-print the trace's contents into the
* trace's pretty buf. 'entry' says whether we're an entry record or not
* (exit). Returns the number of bytes put into the pretty_buf. */
static size_t systrace_fill_pretty_buf(struct systrace_record *trace,
bool entry)
{
size_t len = 0;
struct timespec ts_start = tsc2timespec(trace->start_timestamp);
struct timespec ts_end = tsc2timespec(trace->end_timestamp);
/* Slightly different formats between entry and exit. Entry has retval
* set to ---, and begins with E. Exit begins with X. */
if (entry) {
len = snprintf(trace->pretty_buf, SYSTR_PRETTY_BUF_SZ - len,
"E [%7d.%09d]-[%7d.%09d] Syscall %3d (%12s):(0x%llx, "
"0x%llx, 0x%llx, 0x%llx, 0x%llx, 0x%llx) ret: --- "
"proc: %d core: %2d vcore: %2d errno: --- data: ",
ts_start.tv_sec,
ts_start.tv_nsec,
ts_end.tv_sec,
ts_end.tv_nsec,
trace->syscallno,
syscall_table[trace->syscallno].name,
trace->arg0,
trace->arg1,
trace->arg2,
trace->arg3,
trace->arg4,
trace->arg5,
trace->pid,
trace->coreid,
trace->vcoreid);
} else {
len = snprintf(trace->pretty_buf, SYSTR_PRETTY_BUF_SZ - len,
"X [%7d.%09d]-[%7d.%09d] Syscall %3d (%12s):(0x%llx, "
"0x%llx, 0x%llx, 0x%llx, 0x%llx, 0x%llx) ret: 0x%llx "
"proc: %d core: %2d vcore: -- errno: %3d data: ",
ts_start.tv_sec,
ts_start.tv_nsec,
ts_end.tv_sec,
ts_end.tv_nsec,
trace->syscallno,
syscall_table[trace->syscallno].name,
trace->arg0,
trace->arg1,
trace->arg2,
trace->arg3,
trace->arg4,
trace->arg5,
trace->retval,
trace->pid,
trace->coreid,
trace->errno);
}
len += printdump(trace->pretty_buf + len, trace->datalen,
SYSTR_PRETTY_BUF_SZ - len - 1,
trace->data);
len += snprintf(trace->pretty_buf + len, SYSTR_PRETTY_BUF_SZ - len,
"\n");
return len;
}
/* If some syscalls block, then they can really hurt the user and the
* kernel. For instance, if you blocked another call because the trace queue is
* full, the 2LS will want to yield the vcore, but then *that* call would block
* too. Since that caller was in vcore context, the core will just spin
* forever.
*
* Even worse, some syscalls operate on the calling core or current context,
* thus accessing pcpui. If we block, then that old context is gone. Worse, we
* could migrate and then be operating on a different core. Imagine
* SYS_halt_core. Doh! */
static bool sysc_can_block(unsigned int sysc_num)
{
switch (sysc_num) {
case SYS_proc_yield:
case SYS_fork:
case SYS_exec:
case SYS_pop_ctx:
case SYS_getvcoreid:
case SYS_halt_core:
case SYS_vc_entry:
case SYS_change_vcore:
case SYS_change_to_m:
return FALSE;
}
return TRUE;
}
/* Helper: spits out our trace to the various sinks. */
static void systrace_output(struct systrace_record *trace,
struct strace *strace, bool entry)
{
ERRSTACK(1);
size_t pretty_len;
/* qio ops can throw, especially the blocking qwrite. I had it block on
* the outbound path of sys_proc_destroy(). The rendez immediately
* throws. */
if (waserror()) {
poperror();
return;
}
pretty_len = systrace_fill_pretty_buf(trace, entry);
if (strace) {
/* At this point, we're going to emit the exit trace. It's just
* a question of whether or not we block while doing it. */
if (strace->drop_overflow || !sysc_can_block(trace->syscallno))
qiwrite(strace->q, trace->pretty_buf, pretty_len);
else
qwrite(strace->q, trace->pretty_buf, pretty_len);
}
if (systrace_loud)
printk("%s", trace->pretty_buf);
poperror();
}
static bool should_strace(struct proc *p, struct syscall *sysc)
{
unsigned int sysc_num;
if (systrace_loud)
return TRUE;
if (!p->strace || !p->strace->tracing)
return FALSE;
/* TOCTTOU concerns - sysc is __user. */
sysc_num = ACCESS_ONCE(sysc->num);
if (qfull(p->strace->q)) {
if (p->strace->drop_overflow || !sysc_can_block(sysc_num)) {
atomic_inc(&p->strace->nr_drops);
return FALSE;
}
}
if (sysc_num > MAX_SYSCALL_NR)
return FALSE;
return test_bit(sysc_num, p->strace->trace_set);
}
/* Helper, copies len bytes from u_data to the trace->data, if there's room. */
static void copy_tracedata_from_user(struct systrace_record *trace,
long u_data, size_t len)
{
size_t copy_amt;
copy_amt = MIN(sizeof(trace->data) - trace->datalen, len);
copy_from_user(trace->data + trace->datalen, (void*)u_data, copy_amt);
trace->datalen += copy_amt;
}
/* Helper, snprintfs to the trace, if there's room. */
static void snprintf_to_trace(struct systrace_record *trace, const char *fmt,
...)
{
va_list ap;
int rc;
va_start(ap, fmt);
rc = vsnprintf((char*)trace->data + trace->datalen,
sizeof(trace->data) - trace->datalen, fmt, ap);
va_end(ap);
if (!snprintf_error(rc, sizeof(trace->data) - trace->datalen))
trace->datalen += rc;
}
static bool trace_data_full(struct systrace_record *trace)
{
return trace->datalen == sizeof(trace->data);
}
static bool systrace_has_error(struct systrace_record *trace)
{
return syscall_retval_is_error(trace->syscallno, trace->retval);
}
/* Starts a trace for p running sysc, attaching it to kthread. Pairs with
* systrace_finish_trace(). */
static void systrace_start_trace(struct kthread *kthread, struct syscall *sysc)
{
struct proc *p = current;
struct systrace_record *trace;
kthread->strace = 0;
if (!should_strace(p, sysc))
return;
/* TODO: consider a block_alloc and qpass, though note that we actually
* write the same trace in twice (entry and exit). */
trace = kpages_alloc(SYSTR_BUF_SZ, MEM_ATOMIC);
if (p->strace) {
if (!trace) {
atomic_inc(&p->strace->nr_drops);
return;
}
/* Avoiding the atomic op. We sacrifice accuracy for less
* overhead. */
p->strace->appx_nr_sysc++;
} else {
if (!trace)
return;
}
/* if you ever need to debug just one strace function, this is
* handy way to do it: just bail out if it's not the one you
* want.
* if (sysc->num != SYS_exec)
* return; */
trace->start_timestamp = read_tsc();
trace->end_timestamp = 0;
trace->syscallno = sysc->num;
trace->arg0 = sysc->arg0;
trace->arg1 = sysc->arg1;
trace->arg2 = sysc->arg2;
trace->arg3 = sysc->arg3;
trace->arg4 = sysc->arg4;
trace->arg5 = sysc->arg5;
trace->retval = 0;
trace->pid = p->pid;
trace->coreid = core_id();
trace->vcoreid = proc_get_vcoreid(p);
trace->pretty_buf = (char*)trace + sizeof(struct systrace_record);
trace->datalen = 0;
trace->data[0] = 0;
switch (sysc->num) {
case SYS_write:
case SYS_openat:
case SYS_chdir:
case SYS_nmount:
copy_tracedata_from_user(trace, sysc->arg1, sysc->arg2);
break;
case SYS_stat:
case SYS_lstat:
case SYS_access:
case SYS_unlink:
case SYS_mkdir:
case SYS_rmdir:
case SYS_wstat:
copy_tracedata_from_user(trace, sysc->arg0, sysc->arg1);
break;
case SYS_link:
case SYS_symlink:
case SYS_rename:
case SYS_nbind:
copy_tracedata_from_user(trace, sysc->arg0, sysc->arg1);
snprintf_to_trace(trace, " -> ");
copy_tracedata_from_user(trace, sysc->arg2, sysc->arg3);
break;
case SYS_nunmount:
copy_tracedata_from_user(trace, sysc->arg2, sysc->arg3);
break;
case SYS_exec:
trace->datalen = execargs_stringer(current,
(char *)trace->data,
sizeof(trace->data),
(char *)sysc->arg0,
sysc->arg1,
(char *)sysc->arg2,
sysc->arg3);
break;
case SYS_proc_create:
trace->datalen = execargs_stringer(current,
(char *)trace->data,
sizeof(trace->data),
(char *)sysc->arg0,
sysc->arg1,
(char *)sysc->arg2,
sysc->arg3);
break;
case SYS_tap_fds:
for (size_t i = 0; i < (size_t)sysc->arg1; i++) {
struct fd_tap_req *tap_reqs = (struct
fd_tap_req*)sysc->arg0;
int fd, cmd, filter;
tap_reqs += i;
copy_from_user(&fd, &tap_reqs->fd, sizeof(fd));
copy_from_user(&cmd, &tap_reqs->cmd, sizeof(cmd));
copy_from_user(&filter, &tap_reqs->filter,
sizeof(filter));
snprintf_to_trace(trace, "%d (%d 0x%x), ", fd, cmd,
filter);
if (trace_data_full(trace))
break;
}
break;
}
systrace_output(trace, p->strace, TRUE);
kthread->strace = trace;
}
/* Finishes the trace on kthread for p, with retval being the return from the
* syscall we're tracing. Pairs with systrace_start_trace(). */
static void systrace_finish_trace(struct kthread *kthread, long retval)
{
struct proc *p = current;
struct systrace_record *trace;
if (!kthread->strace)
return;
trace = kthread->strace;
trace->end_timestamp = read_tsc();
trace->retval = retval;
trace->coreid = core_id();
/* Can't trust the vcoreid of an exit record. This'll be ignored later.
*/
trace->vcoreid = -1;
trace->errno = get_errno();
trace->datalen = 0;
/* Only try to do the trace data if we didn't do it on entry */
if (systrace_has_error(trace)) {
snprintf_to_trace(trace, "errstr: %s", current_errstr());
} else {
switch (trace->syscallno) {
case SYS_read:
if (retval <= 0)
break;
copy_tracedata_from_user(trace, trace->arg1, retval);
break;
case SYS_getcwd:
if (retval < 0)
break;
copy_tracedata_from_user(trace, trace->arg0, retval);
break;
case SYS_readlink:
if (retval <= 0)
break;
copy_tracedata_from_user(trace, trace->arg0,
trace->arg1);
snprintf_to_trace(trace, " -> ");
copy_tracedata_from_user(trace, trace->arg2, retval);
break;
}
}
systrace_output(trace, p->strace, FALSE);
kpages_free(kthread->strace, SYSTR_BUF_SZ);
kthread->strace = 0;
}
#ifdef CONFIG_SYSCALL_STRING_SAVING
static void alloc_sysc_str(struct kthread *kth)
{
kth->name = kmalloc(SYSCALL_STRLEN, MEM_ATOMIC);
if (!kth->name)
return;
kth->name[0] = 0;
}
static void free_sysc_str(struct kthread *kth)
{
char *str = kth->name;
kth->name = 0;
kfree(str);
}
#define sysc_save_str(...) \
{ \
struct per_cpu_info *pcpui = this_pcpui_ptr(); \
\
if (pcpui->cur_kthread->name) \
snprintf(pcpui->cur_kthread->name, SYSCALL_STRLEN, \
__VA_ARGS__); \
}
#else
static void alloc_sysc_str(struct kthread *kth)
{
}
static void free_sysc_str(struct kthread *kth)
{
}
#define sysc_save_str(...)
#endif /* CONFIG_SYSCALL_STRING_SAVING */
/* Helper to finish a syscall, signalling if appropriate */
static void finish_sysc(struct syscall *sysc, struct proc *p, long retval)
{
sysc->retval = retval;
/* Atomically turn on the LOCK and SC_DONE flag. The lock tells
* userspace we're messing with the flags and to not proceed. We use it
* instead of CASing with userspace. We need the atomics since we're
* racing with userspace for the event_queue registration. The 'lock'
* tells userspace to not muck with the flags while we're signalling. */
atomic_or(&sysc->flags, SC_K_LOCK | SC_DONE);
__signal_syscall(sysc, p);
atomic_and(&sysc->flags, ~SC_K_LOCK);
}
/* Helper that "finishes" the current async syscall. This should be used with
* care when we are not using the normal syscall completion path.
*
* Do *NOT* complete the same syscall twice. This is catastrophic for _Ms, and
* a bad idea for _S.
*
* It is possible for another user thread to see the syscall being done early -
* they just need to be careful with the weird proc management calls (as in,
* don't trust an async fork).
*
* *sysc is in user memory, and should be pinned (TODO: UMEM). There may be
* issues with unpinning this if we never return. */
static void finish_current_sysc(long retval)
{
/* Need to re-load pcpui, in case we migrated */
struct per_cpu_info *pcpui = this_pcpui_ptr();
struct syscall *sysc = pcpui->cur_kthread->sysc;
assert(sysc);
/* Some 9ns paths set errstr, but not errno. glibc will ignore errstr.
* this is somewhat hacky, since errno might get set unnecessarily */
if ((current_errstr()[0] != 0) && !get_errno())
set_errno(EUNSPECIFIED);
sysc->err = pcpui->cur_kthread->errno;
strncpy(sysc->errstr, pcpui->cur_kthread->errstr, MAX_ERRSTR_LEN);
free_sysc_str(pcpui->cur_kthread);
systrace_finish_trace(pcpui->cur_kthread, retval);
pcpui = this_pcpui_ptr(); /* reload again */
finish_sysc(pcpui->cur_kthread->sysc, pcpui->cur_proc, retval);
pcpui->cur_kthread->sysc = NULL;
}
/* Callable by any function while executing a syscall (or otherwise, actually).
*/
void set_errno(int errno)
{
struct per_cpu_info *pcpui = this_pcpui_ptr();
if (pcpui->cur_kthread)
pcpui->cur_kthread->errno = errno;
}
/* Callable by any function while executing a syscall (or otherwise, actually).
*/
int get_errno(void)
{
struct per_cpu_info *pcpui = this_pcpui_ptr();
if (pcpui->cur_kthread)
return pcpui->cur_kthread->errno;
/* if there's no errno to get, that's not an error I guess. */
return 0;
}
void unset_errno(void)
{
struct per_cpu_info *pcpui = this_pcpui_ptr();
if (!pcpui->cur_kthread)
return;
pcpui->cur_kthread->errno = 0;
pcpui->cur_kthread->errstr[0] = '\0';
}
void vset_errstr(const char *fmt, va_list ap)
{
struct per_cpu_info *pcpui = this_pcpui_ptr();
if (!pcpui->cur_kthread)
return;
vsnprintf(pcpui->cur_kthread->errstr, MAX_ERRSTR_LEN, fmt, ap);
/* TODO: likely not needed */
pcpui->cur_kthread->errstr[MAX_ERRSTR_LEN - 1] = '\0';
}
void set_errstr(const char *fmt, ...)
{
va_list ap;
assert(fmt);
va_start(ap, fmt);
vset_errstr(fmt, ap);
va_end(ap);
}
char *current_errstr(void)
{
struct per_cpu_info *pcpui = this_pcpui_ptr();
if (!pcpui->cur_kthread)
return "no errstr";
return pcpui->cur_kthread->errstr;
}
void set_error(int error, const char *fmt, ...)
{
va_list ap;
set_errno(error);
assert(fmt);
va_start(ap, fmt);
vset_errstr(fmt, ap);
va_end(ap);
}
struct errbuf *get_cur_errbuf(void)
{
return this_pcpui_var(cur_kthread)->errbuf;
}
void set_cur_errbuf(struct errbuf *ebuf)
{
this_pcpui_var(cur_kthread)->errbuf = ebuf;
}
char *get_cur_genbuf(void)
{
struct per_cpu_info *pcpui = this_pcpui_ptr();
assert(pcpui->cur_kthread);
return pcpui->cur_kthread->generic_buf;
}
/* Helper, looks up proc* for pid and ensures p controls that proc. 0 o/w */
static struct proc *get_controllable_proc(struct proc *p, pid_t pid)
{
struct proc *target = pid2proc(pid);
if (!target) {
set_error(ESRCH, "no proc for pid %d", pid);
return 0;
}
if (!proc_controls(p, target)) {
set_error(EPERM, "can't control pid %d", pid);
proc_decref(target);
return 0;
}
return target;
}
static int unpack_argenv(struct argenv *argenv, size_t argenv_l,
int *argc_p, char ***argv_p,
int *envc_p, char ***envp_p)
{
int argc = argenv->argc;
int envc = argenv->envc;
char **argv = (char**)argenv->buf;
char **envp = argv + argc;
char *argbuf = (char*)(envp + envc);
uintptr_t argbuf_offset = (uintptr_t)(argbuf - (char*)(argenv));
if (((char*)argv - (char*)argenv) > argenv_l)
return -1;
if (((char*)argv + (argc * sizeof(char**)) - (char*)argenv) > argenv_l)
return -1;
if (((char*)envp - (char*)argenv) > argenv_l)
return -1;
if (((char*)envp + (envc * sizeof(char**)) - (char*)argenv) > argenv_l)
return -1;
if (((char*)argbuf - (char*)argenv) > argenv_l)
return -1;
for (int i = 0; i < argc; i++) {
if ((uintptr_t)(argv[i] + argbuf_offset) > argenv_l)
return -1;
argv[i] += (uintptr_t)argbuf;
}
for (int i = 0; i < envc; i++) {
if ((uintptr_t)(envp[i] + argbuf_offset) > argenv_l)
return -1;
envp[i] += (uintptr_t)argbuf;
}
*argc_p = argc;
*argv_p = argv;
*envc_p = envc;
*envp_p = envp;
return 0;
}
/************** Utility Syscalls **************/
static int sys_null(void)
{
return 0;
}
/* Diagnostic function: blocks the kthread/syscall, to help userspace test its
* async I/O handling. */
static int sys_block(struct proc *p, unsigned long usec)
{
sysc_save_str("block for %lu usec", usec);
kthread_usleep(usec);
return 0;
}
/* Pause execution for a number of nanoseconds.
* The current implementation rounds up to the nearest microsecond. If the
* syscall is aborted, we return the remaining time the call would have ran
* in the 'rem' parameter. */
static int sys_nanosleep(struct proc *p,
const struct timespec *req,
struct timespec *rem)
{
ERRSTACK(1);
uint64_t usec;
struct timespec kreq, krem = {0, 0};
uint64_t tsc = read_tsc();
/* Check the input arguments. */
if (memcpy_from_user(p, &kreq, req, sizeof(struct timespec))) {
set_errno(EFAULT);
return -1;
}
if (rem && memcpy_to_user(p, rem, &krem, sizeof(struct timespec))) {
set_errno(EFAULT);
return -1;
}
if (kreq.tv_sec < 0) {
set_errno(EINVAL);
return -1;
}
if ((kreq.tv_nsec < 0) || (kreq.tv_nsec > 999999999)) {
set_errno(EINVAL);
return -1;
}
/* Convert timespec to usec. Ignore overflow on the tv_sec field. */
usec = kreq.tv_sec * 1000000;
usec += DIV_ROUND_UP(kreq.tv_nsec, 1000);
/* Attempt to sleep. If we get aborted, copy the remaining time into
* 'rem' and return. We assume the tsc is sufficient to tell how much
* time is remaining (i.e. it only overflows on the order of hundreds of
* years, which should be sufficiently long enough to ensure we don't
* overflow). */
if (waserror()) {
krem = tsc2timespec(read_tsc() - tsc);
if (rem &&
memcpy_to_user(p, rem, &krem, sizeof(struct timespec)))
set_errno(EFAULT);
poperror();
return -1;
}
sysc_save_str("nanosleep for %lu usec", usec);
kthread_usleep(usec);
poperror();
return 0;
}
static int sys_cache_invalidate(void)
{
#ifdef CONFIG_X86
wbinvd();
#endif
return 0;
}
/* sys_reboot(): called directly from dispatch table. */
/* Returns the id of the physical core this syscall is executed on. */
static uint32_t sys_getpcoreid(void)
{
return core_id();
}
// TODO: Temporary hack until thread-local storage is implemented on i386 and
// this is removed from the user interface
static size_t sys_getvcoreid(struct proc *p)
{
return proc_get_vcoreid(p);
}
/************** Process management syscalls **************/
/* Helper for proc_create and fork */
static void inherit_strace(struct proc *parent, struct proc *child)
{
if (parent->strace && parent->strace->inherit) {
/* Refcnt on both, put in the child's ->strace. */
kref_get(&parent->strace->users, 1);
kref_get(&parent->strace->procs, 1);
child->strace = parent->strace;
}
}
/* Creates a process from the file 'path'. The process is not runnable by
* default, so it needs it's status to be changed so that the next call to
* schedule() will try to run it. */
static int sys_proc_create(struct proc *p, char *path, size_t path_l,
char *argenv, size_t argenv_l, int flags)
{
int pid = 0;
char *t_path;
struct file_or_chan *program;
struct proc *new_p;
int argc, envc;
char **argv, **envp;
struct argenv *kargenv;
t_path = copy_in_path(p, path, path_l);
if (!t_path)
return -1;
program = foc_open(t_path, O_EXEC | O_READ, 0);
if (!program)
goto error_with_path;
if (!is_valid_elf(program)) {
set_errno(ENOEXEC);
goto error_with_file;
}
/* Check the size of the argenv array, error out if too large. */
if ((argenv_l < sizeof(struct argenv)) || (argenv_l > ARG_MAX)) {
set_error(EINVAL, "The argenv array has an invalid size: %lu\n",
argenv_l);
goto error_with_file;
}
/* Copy the argenv array into a kernel buffer. Delay processing of the
* array to load_elf(). */
kargenv = user_memdup_errno(p, argenv, argenv_l);
if (!kargenv) {
set_error(EINVAL, "Failed to copy in the args");
goto error_with_file;
}
/* Unpack the argenv array into more usable variables. Integrity
* checking done along side this as well. */
if (unpack_argenv(kargenv, argenv_l, &argc, &argv, &envc, &envp)) {
set_error(EINVAL, "Failed to unpack the args");
goto error_with_kargenv;
}
/* TODO: need to split the proc creation, since you must load after
* setting args/env, since auxp gets set up there. */
//new_p = proc_create(program, 0, 0);
if (proc_alloc(&new_p, current, flags)) {
set_error(ENOMEM, "Failed to alloc new proc");
goto error_with_kargenv;
}
inherit_strace(p, new_p);
/* close the CLOEXEC ones, even though this isn't really an exec */
close_fdt(&new_p->open_files, TRUE);
/* Load the elf. */
if (load_elf(new_p, program, argc, argv, envc, envp)) {
set_error(EINVAL, "Failed to load elf");
goto error_with_proc;
}
/* progname is argv0, which accounts for symlinks */
proc_set_progname(new_p, argc ? argv[0] : NULL);
proc_replace_binary_path(new_p, t_path);
foc_decref(program);
user_memdup_free(p, kargenv);
__proc_ready(new_p);
pid = new_p->pid;
profiler_notify_new_process(new_p);
/* give up the reference created in proc_create() */
proc_decref(new_p);
return pid;
error_with_proc:
/* proc_destroy will decref once, which is for the ref created in
* proc_create(). We don't decref again (the usual "+1 for existing"),
* since the scheduler, which usually handles that, hasn't heard about
* the process (via __proc_ready()). */
proc_destroy(new_p);
error_with_kargenv:
user_memdup_free(p, kargenv);
error_with_file:
foc_decref(program);
error_with_path:
free_path(p, t_path);
return -1;
}
/* Makes process PID runnable. Consider moving the functionality to process.c
*/
static error_t sys_proc_run(struct proc *p, unsigned pid)
{
error_t retval = 0;
struct proc *target = get_controllable_proc(p, pid);
if (!target)
return -1;
if (target->state != PROC_CREATED) {
set_errno(EINVAL);
proc_decref(target);
return -1;
}
/* Note a proc can spam this for someone it controls. Seems safe - if
* it isn't we can change it. */
proc_wakeup(target);
proc_decref(target);
return 0;
}
/* Destroy proc pid. If this is called by the dying process, it will never
* return. o/w it will return 0 on success, or an error. Errors include:
* - ESRCH: if there is no such process with pid
* - EPERM: if caller does not control pid */
static error_t sys_proc_destroy(struct proc *p, pid_t pid, int exitcode)
{
error_t r;
struct proc *p_to_die = get_controllable_proc(p, pid);
if (!p_to_die)
return -1;
if (p_to_die == p) {
p->exitcode = exitcode;
printd("[PID %d] proc exiting gracefully (code %d)\n",
p->pid,exitcode);
} else {
p_to_die->exitcode = exitcode;
printd("[%d] destroying proc %d\n", p->pid, p_to_die->pid);
}
proc_destroy(p_to_die);
proc_decref(p_to_die);
return 0;
}
static int sys_proc_yield(struct proc *p, bool being_nice)
{
/* proc_yield() often doesn't return - we need to finish the syscall
* early. If it doesn't return, it expects to eat our reference (for
* now). */
finish_current_sysc(0);
proc_incref(p, 1);
proc_yield(p, being_nice);
proc_decref(p);
/* Shouldn't return, to prevent the chance of mucking with cur_sysc. */
smp_idle();
assert(0);
}
static int sys_change_vcore(struct proc *p, uint32_t vcoreid,
bool enable_my_notif)
{
/* Note retvals can be negative, but we don't mess with errno in case
* callers use this in low-level code and want to extract the 'errno'.
*/
return proc_change_to_vcore(p, vcoreid, enable_my_notif);
}
static ssize_t sys_fork(env_t* e)
{
uintptr_t temp;
int ret;
// TODO: right now we only support fork for single-core processes
if (e->state != PROC_RUNNING_S) {
set_errno(EINVAL);
return -1;
}
env_t* env;
ret = proc_alloc(&env, current, PROC_DUP_FGRP);
assert(!ret);
assert(env != NULL);
proc_set_progname(env, e->progname);
/* Can't really fork if we don't have a current_ctx to fork */
if (!current_ctx) {
proc_destroy(env);
proc_decref(env);
set_errno(EINVAL);
return -1;
}
assert(current == this_pcpui_var(owning_proc));
copy_current_ctx_to(&env->scp_ctx);
/* Make the new process have the same VMRs as the older. This will copy
* the contents of non MAP_SHARED pages to the new VMRs. */
if (duplicate_vmrs(e, env)) {
proc_destroy(env);
proc_decref(env);
set_errno(ENOMEM);
return -1;
}
/* Switch to the new proc's address space and finish the syscall. We'll
* never naturally finish this syscall for the new proc, since its
* memory is cloned before we return for the original process. If we
* ever do CoW for forked memory, this will be the first place that gets
* CoW'd. */
temp = switch_to(env);
finish_sysc(current_kthread->sysc, env, 0);
switch_back(env, temp);
/* Copy some state from the original proc into the new proc. */
env->env_flags = e->env_flags;
inherit_strace(e, env);
/* In general, a forked process should be a fresh process, and we copy
* over whatever stuff is needed between procinfo/procdata. */
*env->procdata = *e->procdata;
env->procinfo->program_end = e->procinfo->program_end;
/* FYI: once we call ready, the proc is open for concurrent usage */
__proc_ready(env);
proc_wakeup(env);
// don't decref the new process.
// that will happen when the parent waits for it.
// TODO: if the parent doesn't wait, we need to change the child's
// parent when the parent dies, or at least decref it
printd("[PID %d] fork PID %d\n", e->pid, env->pid);
ret = env->pid;
profiler_notify_new_process(env);
proc_decref(env); /* give up the reference created in proc_alloc() */
return ret;
}
/* string for sys_exec arguments. Assumes that d is pointing to zero'd
* storage or storage that does not require null termination or
* provides the null. */
static int execargs_stringer(struct proc *p, char *d, size_t slen,
char *path, size_t path_l,
char *argenv, size_t argenv_l)
{
int argc, envc, i;
char **argv, **envp;
struct argenv *kargenv;
int amt;
char *s = d;
char *e = d + slen;
if (path_l > slen)
path_l = slen;
if (memcpy_from_user(p, d, path, path_l)) {
s = seprintf(s, e, "Invalid exec path");
return s - d;
}
s += path_l;
/* yes, this code is cloned from below. I wrote a helper but
* Barret and I concluded after talking about it that the
* helper was not really helper-ful, as it has almost 10
* arguments. Please, don't suggest a cpp macro. Thank you. */
/* Check the size of the argenv array, error out if too large. */
if ((argenv_l < sizeof(struct argenv)) || (argenv_l > ARG_MAX)) {
s = seprintf(s, e,
"The argenv array has an invalid size: %lu\n",
argenv_l);
return s - d;
}
/* Copy the argenv array into a kernel buffer. */
kargenv = user_memdup_errno(p, argenv, argenv_l);
if (!kargenv) {
s = seprintf(s, e,
"Failed to copy in the args and environment");
return s - d;
}
/* Unpack the argenv array into more usable variables. Integrity
* checking done along side this as well. */
if (unpack_argenv(kargenv, argenv_l, &argc, &argv, &envc, &envp)) {
s = seprintf(s, e, "Failed to unpack the args");
user_memdup_free(p, kargenv);
return s - d;
}
s = seprintf(s, e, "[%d]{", argc);
for (i = 0; i < argc; i++)
s = seprintf(s, e, "%s, ", argv[i]);
s = seprintf(s, e, "}");
user_memdup_free(p, kargenv);
return s - d;
}
/* Load the binary "path" into the current process, and start executing it.
* argv and envp are magically bundled in procinfo for now. Keep in sync with
* glibc's sysdeps/ros/execve.c. Once past a certain point, this function won't
* return. It assumes (and checks) that it is current. Don't give it an extra
* refcnt'd *p (syscall won't do that).
* Note: if someone batched syscalls with this call, they could clobber their
* old memory (and will likely PF and die). Don't do it... */
static int sys_exec(struct proc *p, char *path, size_t path_l,
char *argenv, size_t argenv_l)
{
int ret = -1;
char *t_path = NULL;
struct file_or_chan *program;
int argc, envc;
char **argv, **envp;
struct argenv *kargenv;
/* We probably want it to never be allowed to exec if it ever was _M */
if (p->state != PROC_RUNNING_S) {
set_error(EINVAL, "Can't exec an MCP");
return -1;
}
/* Check the size of the argenv array, error out if too large. */
if ((argenv_l < sizeof(struct argenv)) || (argenv_l > ARG_MAX)) {
set_error(EINVAL, "The argenv array has an invalid size: %lu\n",
argenv_l);
return -1;
}
if (p != this_pcpui_var(owning_proc)) {
warn("Proc %d tried to exec and wasn't owning_proc", p->pid);
set_error(EAGAIN, "exec may have blocked during execution");
return -1;
}
assert(current_ctx);
/* Before this, we shouldn't have blocked (maybe with strace, though we
* explicitly don't block exec for strace). The owning proc, cur_proc,
* and cur_ctx checks should catch that. After this, we might still
* block, such as on accessing the filesystem.
*
* After this point, we're treated like a yield - we're waiting until
* something wakes us. The kthread might block, error and fail, or
* succeed. We shouldn't return to userspace before one of those. The
* only way out of this function is via smp_idle, not returning the way
* we came.
*
* Under normal situations, the only thing that will wake us is this
* kthread completing. I think you can trigger wakeups with events and
* async syscalls started before the exec. I'm not sure if that could
* trigger more bugs or if that would hurt the kernel. If so, we could
* add an EXEC_LIMBO state.
*
* Note that we will 'hard block' if we block at all. We can't return
* to userspace and then asynchronously finish the exec later. */
spin_lock(&p->proc_lock);
/* We only need the context for the error case. We have to save it now,
* since once we leave this core, such as when the kthread blocks, the
* old SCP's context will be gone. */
__proc_save_context_s(p);
/* We are no longer owning, but we are still current, like any
* kthread-that-blocked-on-behalf of a process. I think one invariant
* for SCPs is: "RUNNING_S <==> is the owning proc". */
clear_owning_proc(core_id());
__proc_set_state(p, PROC_WAITING);
spin_unlock(&p->proc_lock);
/* Copy the argenv array into a kernel buffer. */
kargenv = user_memdup_errno(p, argenv, argenv_l);
if (!kargenv) {
set_error(EINVAL, "Failed to copy in the args and environment");
goto out_error;
}
/* Unpack the argenv array into more usable variables. Integrity
* checking done along side this as well. */
if (unpack_argenv(kargenv, argenv_l, &argc, &argv, &envc, &envp)) {
set_error(EINVAL, "Failed to unpack the args");
goto out_error_kargenv;
}
t_path = copy_in_path(p, path, path_l);
if (!t_path) {
user_memdup_free(p, kargenv);
goto out_error_kargenv;
}
program = foc_open(t_path, O_EXEC | O_READ, 0);
if (!program)
goto out_error_tpath;
if (!is_valid_elf(program)) {
set_error(ENOEXEC, "Program was not a valid ELF");
goto out_error_program;
}
/* This is the point of no return for the process. Any errors here lead
* to destruction. */
/* progname is argv0, which accounts for symlinks */
proc_replace_binary_path(p, t_path);
/* p now owns the t_path, and it'll get freed when we destroy p. */
t_path = NULL;
proc_set_progname(p, argc ? argv[0] : NULL);
proc_init_procdata(p);
p->procinfo->program_end = 0;
/* When we destroy our memory regions, accessing cur_sysc would PF */
current_kthread->sysc = 0;
unmap_and_destroy_vmrs(p);
/* close the CLOEXEC ones */
close_fdt(&p->open_files, TRUE);
env_user_mem_free(p, 0, UMAPTOP);
if (load_elf(p, program, argc, argv, envc, envp)) {
set_error(EINVAL, "Failed to load elf");
/* At this point, we destroyed memory and can't return to the
* app. We can't use the error cases, since they assume we'll
* return. */
foc_decref(program);
user_memdup_free(p, kargenv);
/* We finish the trace and not the sysc, since the sysc is gone.
*/
systrace_finish_trace(current_kthread, -1);
/* Note this is an inedible reference, but proc_destroy now
* returns */
proc_destroy(p);
/* We don't want to do anything else - we just need to not
* accidentally return to the user (hence the all_out) */
goto all_out;
}
printd("[PID %d] exec %s\n", p->pid, foc_to_name(program));
foc_decref(program);
user_memdup_free(p, kargenv);
systrace_finish_trace(current_kthread, 0);
proc_wakeup(p);
goto all_out;
out_error_program:
foc_decref(program);
out_error_tpath:
/* Note the t_path is passed to proc_replace_binary_path in the non
* out_error cases. */
free_path(p, t_path);
out_error_kargenv:
user_memdup_free(p, kargenv);
out_error:
finish_current_sysc(-1);
proc_wakeup(p);
all_out:
/* This free and setting sysc = NULL may happen twice (early errors do
* it), but they are idempotent. */
free_sysc_str(current_kthread);
current_kthread->sysc = NULL;
/* we can't return, since we'd write retvals to the old location of the
* syscall struct (which has been freed and is in the old userspace) (or
* has already been written to).*/
disable_irq(); /* abandon_core/clear_own wants irqs disabled */
abandon_core();
smp_idle(); /* will reenable interrupts */
}
/* Helper, will attempt a particular wait on a proc. Returns the pid of the
* process if we waited on it successfully, and the status will be passed back
* in ret_status (kernel memory). Returns 0 if the wait failed and we should
* try again. Returns -1 if we should abort. Only handles DYING. Callers
* need to lock to protect the children tailq and reaping bits. Callers must
* decref the child on success. */
static pid_t __try_wait(struct proc *parent, struct proc *child,
int *ret_status, int options)
{
if (proc_is_dying(child)) {
/* Disown returns -1 if it's already been disowned or we should
* o/w abort. This can happen if we have concurrent waiters,
* both with pointers to the child (only one should reap). Note
* that if we don't do this, we could go to sleep and never
* receive a cv_signal. */
if (__proc_disown_child(parent, child))
return -1;
/* despite disowning, the child won't be freed til we drop this
* ref held by this function, so it is safe to access the
* memory.
*
* Note the exit code one byte in the 0xff00 spot. Check out
* glibc's posix/sys/wait.h and bits/waitstatus.h for more info.
* If we ever deal with signalling and stopping, we'll need to
* do some more work here.*/
*ret_status = (child->exitcode & 0xff) << 8;
return child->pid;
}
return 0;
}
/* Helper, like __try_wait, but attempts a wait on any of the children,
* returning the specific PID we waited on, 0 to try again (a waitable exists),
* and -1 to abort (no children/waitables exist). Callers need to lock to
* protect the children tailq and reaping bits. Callers must decref the child,
* if successful. */
static pid_t __try_wait_any(struct proc *parent, int *ret_status, int options,
struct proc **child)
{
struct proc *i, *temp;
pid_t retval;
if (TAILQ_EMPTY(&parent->children))
return -1;
/* Could have concurrent waiters mucking with the tailq, caller must
* lock */
TAILQ_FOREACH_SAFE(i, &parent->children, sibling_link, temp) {
retval = __try_wait(parent, i, ret_status, options);
/* This catches a thread causing a wait to fail but not taking
* the child off the list before unlocking. Should never
* happen. */
assert(retval != -1);
/* Succeeded, return the pid of the child we waited on */
if (retval) {
*child = i;
return retval;
}
}
assert(retval == 0);
return 0;
}
/* Waits on a particular child, returns the pid of the child waited on, and
* puts the ret status in *ret_status. Returns the pid if we succeeded, 0 if
* the child was not waitable and WNOHANG, and -1 on error. */
static pid_t wait_one(struct proc *parent, struct proc *child, int *ret_status,
int options)
{
pid_t retval;
cv_lock(&parent->child_wait);
/* retval == 0 means we should block */
retval = __try_wait(parent, child, ret_status, options);
if ((retval == 0) && (options & WNOHANG))
goto out_unlock;
while (!retval) {
cpu_relax();
cv_wait(&parent->child_wait);
/* If we're dying, then we don't need to worry about waiting.
* We don't do this yet, but we'll need this outlet when we deal
* with orphaned children and having init inherit them. */
if (proc_is_dying(parent))
goto out_unlock;
/* Any child can wake us up, but we check for the particular
* child we care about */
retval = __try_wait(parent, child, ret_status, options);
}
if (retval == -1) {
/* Child was already waited on by a concurrent syscall. */
set_errno(ECHILD);
}
/* Fallthrough */
out_unlock:
cv_unlock(&parent->child_wait);
if (retval > 0)
proc_decref(child);
return retval;
}
/* Waits on any child, returns the pid of the child waited on, and puts the ret
* status in *ret_status. Is basically a waitpid(-1, ... ); See wait_one for
* more details. Returns -1 if there are no children to wait on, and returns 0
* if there are children and we need to block but WNOHANG was set. */
static pid_t wait_any(struct proc *parent, int *ret_status, int options)
{
pid_t retval;
struct proc *child;
cv_lock(&parent->child_wait);
retval = __try_wait_any(parent, ret_status, options, &child);
if ((retval == 0) && (options & WNOHANG))
goto out_unlock;
while (!retval) {
cpu_relax();
cv_wait(&parent->child_wait);
if (proc_is_dying(parent))
goto out_unlock;
/* Any child can wake us up from the CV. This is a linear
* __try_wait scan. If we have a lot of children, we could
* optimize this. */
retval = __try_wait_any(parent, ret_status, options, &child);
}
if (retval == -1)
assert(TAILQ_EMPTY(&parent->children));
/* Fallthrough */
out_unlock:
cv_unlock(&parent->child_wait);
if (retval > 0)
proc_decref(child);
return retval;
}
/* Note: we only allow waiting on children (no such thing as threads, for
* instance). Right now we only allow waiting on termination (not signals),
* and we don't have a way for parents to disown their children (such as
* ignoring SIGCHLD, see man 2 waitpid's Notes).
*
* We don't bother with stop/start signals here, though we can probably build
* it in the helper above.
*
* Returns the pid of who we waited on, or -1 on error, or 0 if we couldn't
* wait (WNOHANG). */
static pid_t sys_waitpid(struct proc *parent, pid_t pid, int *status,
int options)
{
struct proc *child;
pid_t retval = 0;
int ret_status = 0;
sysc_save_str("waitpid on %d", pid);
/* -1 is the signal for 'any child' */
if (pid == -1) {
retval = wait_any(parent, &ret_status, options);
goto out;
}
child = pid2proc(pid);
if (!child) {
set_errno(ECHILD); /* ECHILD also used for no proc */
retval = -1;
goto out;
}
if (!(parent->pid == child->ppid)) {
set_errno(ECHILD);
retval = -1;
goto out_decref;
}
retval = wait_one(parent, child, &ret_status, options);
/* fall-through */
out_decref:
proc_decref(child);
out:
/* ignoring / don't care about memcpy's retval here. */
if (status)
memcpy_to_user(parent, status, &ret_status, sizeof(ret_status));
printd("[PID %d] waited for PID %d, got retval %d (status 0x%x)\n",
parent->pid, pid, retval, ret_status);
return retval;
}
/************** Memory Management Syscalls **************/
static void *sys_mmap(struct proc *p, uintptr_t addr, size_t len, int prot,
int flags, int fd, off_t offset)
{
return mmap(p, addr, len, prot, flags, fd, offset);
}
static intreg_t sys_mprotect(struct proc *p, void *addr, size_t len, int prot)
{
return mprotect(p, (uintptr_t)addr, len, prot);
}
static intreg_t sys_munmap(struct proc *p, void *addr, size_t len)
{
return munmap(p, (uintptr_t)addr, len);
}
static ssize_t sys_shared_page_alloc(env_t* p1,
void **_addr, pid_t p2_id,
int p1_flags, int p2_flags
)
{
printk("[kernel] shared page alloc is deprecated/unimplemented.\n");
return -1;
}
static int sys_shared_page_free(env_t* p1, void *addr, pid_t p2)
{
return -1;
}
/* Helper, to do the actual provisioning of a resource to a proc */
static int prov_resource(struct proc *target, unsigned int res_type,
long res_val)
{
switch (res_type) {
case (RES_CORES):
/* in the off chance we have a kernel scheduler that can't
* provision, we'll need to change this. */
return provision_core(target, res_val);
default:
printk("[kernel] got provisioning for unknown resource %d\n",
res_type);
set_errno(ENOENT); /* or EINVAL? */
return -1;
}
}
/* Rough syscall to provision res_val of type res_type to target_pid */
static int sys_provision(struct proc *p, int target_pid,
unsigned int res_type, long res_val)
{
struct proc *target = pid2proc(target_pid);
int retval;
if (!target) {
if (target_pid == 0)
return prov_resource(0, res_type, res_val);
/* debugging interface */
if (target_pid == -1)
print_coreprov_map();
set_errno(ESRCH);
return -1;
}
retval = prov_resource(target, res_type, res_val);
proc_decref(target);
return retval;
}
/* Untested. Will notify the target on the given vcore, if the caller controls
* the target. Will honor the target's wanted/vcoreid. u_ne can be NULL. */
static int sys_notify(struct proc *p, int target_pid, unsigned int ev_type,
struct event_msg *u_msg)
{
struct event_msg local_msg = {0};
struct proc *target = get_controllable_proc(p, target_pid);
if (!target)
return -1;
/* if the user provided an ev_msg, copy it in and use that */
if (u_msg) {
if (memcpy_from_user(p, &local_msg, u_msg,
sizeof(struct event_msg))) {
proc_decref(target);
set_errno(EINVAL);
return -1;
}
} else {
local_msg.ev_type = ev_type;
}
send_kernel_event(target, &local_msg, 0);
proc_decref(target);
return 0;
}
/* Will notify the calling process on the given vcore, independently of WANTED
* or advertised vcoreid. If you change the parameters, change pop_user_ctx().
*/
static int sys_self_notify(struct proc *p, uint32_t vcoreid,
unsigned int ev_type, struct event_msg *u_msg,
bool priv)
{
struct event_msg local_msg = {0};
/* if the user provided an ev_msg, copy it in and use that */
if (u_msg) {
if (memcpy_from_user(p, &local_msg, u_msg,
sizeof(struct event_msg))) {
set_errno(EINVAL);
return -1;
}
} else {
local_msg.ev_type = ev_type;
}
if (local_msg.ev_type >= MAX_NR_EVENT) {
printk("[kernel] received self-notify for vcoreid %d, "
"ev_type %d, u_msg %p, u_msg->type %d\n", vcoreid,
ev_type, u_msg, u_msg ? u_msg->ev_type : 0);
return -1;
}
/* this will post a message and IPI, regardless of
* wants/needs/debutantes.*/
post_vcore_event(p, &local_msg, vcoreid,
priv ? EVENT_VCORE_PRIVATE : 0);
proc_notify(p, vcoreid);
return 0;
}
static int sys_send_event(struct proc *p, struct event_queue *ev_q,
struct event_msg *u_msg, uint32_t vcoreid)
{
struct event_msg local_msg = {0};
if (memcpy_from_user_errno(p, &local_msg, u_msg,
sizeof(struct event_msg))) {
return -1;
}
if (!is_user_rwaddr(ev_q, sizeof(struct event_queue))) {
set_error(EINVAL, "bad event_queue %p", ev_q);
return -1;
}
send_event(p, ev_q, &local_msg, vcoreid);
return 0;
}
/* Puts the calling core into vcore context, if it wasn't already, via a
* self-IPI / active notification. Barring any weird unmappings, we just send
* ourselves a __notify. */
static int sys_vc_entry(struct proc *p)
{
send_kernel_message(core_id(), __notify, (long)p, 0, 0, KMSG_ROUTINE);
return 0;
}
/* This will halt the core, waking on an IRQ. These could be kernel IRQs for
* things like timers or devices, or they could be IPIs for RKMs (__notify for
* an evq with IPIs for a syscall completion, etc). With arch support, this
* will also wake on a write to notif_pending.
*
* We don't need to finish the syscall early (worried about the syscall struct,
* on the vcore's stack). The syscall will finish before any __preempt RKM
* executes, so the vcore will not restart somewhere else before the syscall
* completes (unlike with yield, where the syscall itself adjusts the vcore
* structures).
*
* In the future, RKM code might avoid sending IPIs if the core is already in
* the kernel. That code will need to check the CPU's state in some manner, and
* send if the core is halted/idle. Or perhaps use mwait, if there's arch
* support.
*
* The core must wake up for RKMs, including RKMs that arrive while the kernel
* is trying to halt.
*
* If our hardware supports something like monitor/mwait, we'll abort if
* notif_pending was or gets set. Note that whoever writes notif_pending may
* send an IPI regardless of whether or not we have mwait. That's up to the
* ev_q settings (so basically userspace). If userspace doesn't want an IPI, a
* notif will wake it up, but it won't break it out of a uthread loop. */
static int sys_halt_core(struct proc *p, unsigned long usec)
{
struct per_cpu_info *pcpui = this_pcpui_ptr();
struct preempt_data *vcpd;
/* The user can only halt CG cores! (ones it owns) */
if (management_core())
return -1;
rcu_report_qs();
disable_irq();
/* both for accounting and possible RKM optimizations */
__set_cpu_state(pcpui, CPU_STATE_IDLE);
wrmb();
if (has_routine_kmsg()) {
__set_cpu_state(pcpui, CPU_STATE_KERNEL);
enable_irq();
return 0;
}
vcpd = &p->procdata->vcore_preempt_data[pcpui->owning_vcoreid];
/* We pretend to not be in vcore context so other cores will send us
* IPIs (__notify). If we do get a __notify, we'll have set
* notif_disabled back on before we handle the message, since it's a
* routine KMSG. Note that other vcores will think we are not in vcore
* context. This is no different to when we pop contexts: 'briefly'
* leave VC ctx, check notif_pending, and (possibly) abort and set
* notif_disabled. */
vcpd->notif_disabled = false;
cpu_halt_notif_pending(vcpd);
__set_cpu_state(pcpui, CPU_STATE_KERNEL);
vcpd->notif_disabled = true;
enable_irq();
return 0;
}
/* Changes a process into _M mode, or -EINVAL if it already is an mcp.
* __proc_change_to_m() returns and we'll eventually finish the sysc later. The
* original context may restart on a remote core before we return and finish,
* but that's fine thanks to the async kernel interface. */
static int sys_change_to_m(struct proc *p)
{
int retval = proc_change_to_m(p);
/* convert the kernel error code into (-1, errno) */
if (retval) {
set_errno(-retval);
retval = -1;
}
return retval;
}
/* Assists the user/2LS by atomically running *ctx and leaving vcore context.
* Normally, the user can do this themselves, but x86 VM contexts need kernel
* support. The caller ought to be in vcore context, and if a notif is pending,
* then the calling vcore will restart in a fresh VC ctx (as if it was notified
* or did a sys_vc_entry).
*
* Note that this will set the TLS too, which is part of the context. Parlib's
* pop_user_ctx currently does *not* do this, since the TLS is managed
* separately. If you want to use this syscall for testing, you'll need to 0
* out fsbase and conditionally write_msr in proc_pop_ctx(). */
static int sys_pop_ctx(struct proc *p, struct user_context *ctx)
{
int pcoreid = core_id();
struct per_cpu_info *pcpui = &per_cpu_info[pcoreid];
int vcoreid = pcpui->owning_vcoreid;
struct preempt_data *vcpd = &p->procdata->vcore_preempt_data[vcoreid];
/* With change_to, there's a bunch of concerns about changing the vcore
* map, since the kernel may have already locked and sent preempts,
* deaths, etc.
*
* In this case, we don't care as much. Other than notif_pending and
* notif_disabled, it's more like we're just changing a few registers in
* cur_ctx. We can safely order-after any kernel messages or other
* changes, as if the user had done all of the changes we'll make and
* then did a no-op syscall.
*
* Since we are mucking with current_ctx, it is important that we don't
* block before or during this syscall. */
arch_finalize_ctx(pcpui->cur_ctx);
if (copy_from_user(pcpui->cur_ctx, ctx, sizeof(struct user_context))) {
/* The 2LS isn't really in a position to handle errors. At the
* very least, we can print something and give them a fresh vc
* ctx. */
printk("[kernel] unable to copy user_ctx, 2LS bug\n");
memset(pcpui->cur_ctx, 0, sizeof(struct user_context));
proc_init_ctx(pcpui->cur_ctx, vcoreid, vcpd->vcore_entry,
vcpd->vcore_stack, vcpd->vcore_tls_desc);
return -1;
}
proc_secure_ctx(pcpui->cur_ctx);
/* The caller leaves vcore context no matter what. We'll put them back
* in if they missed a message. */
vcpd->notif_disabled = FALSE;
wrmb(); /* order disabled write before pending read */
if (vcpd->notif_pending)
send_kernel_message(pcoreid, __notify, (long)p, 0, 0,
KMSG_ROUTINE);
return 0;
}
static int sys_vmm_add_gpcs(struct proc *p, unsigned int nr_more_gpcs,
struct vmm_gpcore_init *gpcis)
{
ERRSTACK(1);
struct vmm *vmm = &p->vmm;
/* We do a copy_from_user in __vmm_add_gpcs, but it ought to be clear
* from the syscall.c code if we did our error checking. */
if (!is_user_rwaddr(gpcis, sizeof(struct vmm_gpcore_init) *
nr_more_gpcs)) {
set_error(EINVAL, "bad user addr %p + %p", gpcis,
sizeof(struct vmm_gpcore_init) * nr_more_gpcs);
return -1;
}
qlock(&vmm->qlock);
if (waserror()) {
qunlock(&vmm->qlock);
poperror();
return -1;
}
__vmm_struct_init(p);
__vmm_add_gpcs(p, nr_more_gpcs, gpcis);
qunlock(&vmm->qlock);
poperror();
return nr_more_gpcs;
}
static int sys_vmm_poke_guest(struct proc *p, int guest_pcoreid)
{
return vmm_poke_guest(p, guest_pcoreid);
}
static int sys_vmm_ctl(struct proc *p, int cmd, unsigned long arg1,
unsigned long arg2, unsigned long arg3,
unsigned long arg4)
{
ERRSTACK(1);
int ret;
struct vmm *vmm = &p->vmm;
/* Protects against concurrent setters and for gets that are not atomic
* reads (say, multiple exec ctls). */
qlock(&vmm->qlock);
if (waserror()) {
qunlock(&vmm->qlock);
poperror();
return -1;
}
__vmm_struct_init(p);
switch (cmd) {
case VMM_CTL_GET_EXITS:
if (vmm->amd)
error(ENOTSUP, "AMD VMMs unsupported");
ret = vmx_ctl_get_exits(&vmm->vmx);
break;
case VMM_CTL_SET_EXITS:
if (arg1 & ~VMM_CTL_ALL_EXITS)
error(EINVAL, "Bad vmm_ctl_exits %x (%x)", arg1,
VMM_CTL_ALL_EXITS);
if (vmm->amd)
error(ENOTSUP, "AMD VMMs unsupported");
ret = vmx_ctl_set_exits(&vmm->vmx, arg1);
break;
case VMM_CTL_GET_FLAGS:
ret = vmm->flags;
break;
case VMM_CTL_SET_FLAGS:
if (arg1 & ~VMM_CTL_ALL_FLAGS)
error(EINVAL,
"Bad vmm_ctl flags. Got 0x%lx, allowed 0x%lx\n",
arg1, VMM_CTL_ALL_FLAGS);
vmm->flags = arg1;
ret = 0;
break;
default:
error(EINVAL, "Bad vmm_ctl cmd %d", cmd);
}
qunlock(&vmm->qlock);
poperror();
return ret;
}
/* Pokes the ksched for the given resource for target_pid. If the target pid
* == 0, we just poke for the calling process. The common case is poking for
* self, so we avoid the lookup.
*
* Not sure if you could harm someone via asking the kernel to look at them, so
* we'll do a 'controls' check for now. In the future, we might have something
* in the ksched that limits or penalizes excessive pokes. */
static int sys_poke_ksched(struct proc *p, int target_pid,
unsigned int res_type)
{
struct proc *target;
int retval = 0;
if (!target_pid) {
poke_ksched(p, res_type);
return 0;
}
target = pid2proc(target_pid);
if (!target) {
set_errno(ESRCH);
return -1;
}
if (!proc_controls(p, target)) {
set_errno(EPERM);
retval = -1;
goto out;
}
poke_ksched(target, res_type);
out:
proc_decref(target);
return retval;
}
static int sys_abort_sysc(struct proc *p, struct syscall *sysc)
{
return abort_sysc(p, (uintptr_t)sysc);
}
static int sys_abort_sysc_fd(struct proc *p, int fd)
{
/* Consider checking for a bad fd. Doesn't matter now, since we only
* look for actual syscalls blocked that had used fd. */
return abort_all_sysc_fd(p, fd);
}
static unsigned long sys_populate_va(struct proc *p, uintptr_t va,
unsigned long nr_pgs)
{
return populate_va(p, ROUNDDOWN(va, PGSIZE), nr_pgs);
}
static intreg_t sys_read(struct proc *p, int fd, void *buf, size_t len)
{
if (!is_user_rwaddr(buf, len)) {
set_error(EINVAL, "bad user addr %p + %p", buf, len);
return -1;
}
sysc_save_str("read on fd %d", fd);
return sysread(fd, buf, len);
}
static intreg_t sys_write(struct proc *p, int fd, const void *buf, size_t len)
{
/* We'll let this one include read-only areas, unlike most other
* syscalls that take bufs created and written by the user. */
if (!is_user_raddr(buf, len)) {
set_error(EINVAL, "bad user addr %p + %p", buf, len);
return -1;
}
sysc_save_str("write on fd %d", fd);
return syswrite(fd, (void*)buf, len);
}
/* Checks args/reads in the path, opens the file (relative to fromfd if the path
* is not absolute), and inserts it into the process's open file list. */
static intreg_t sys_openat(struct proc *p, int fromfd, const char *path,
size_t path_l, int oflag, int mode)
{
int fd;
char *t_path;
printd("File %s Open attempt oflag %x mode %x\n", path, oflag, mode);
if ((oflag & O_PATH) && (oflag & O_ACCMODE)) {
set_error(EINVAL, "Cannot open O_PATH with any I/O perms (O%o)",
oflag);
return -1;
}
t_path = copy_in_path(p, path, path_l);
if (!t_path)
return -1;
sysc_save_str("open %s at fd %d", t_path, fromfd);
fd = sysopenat(fromfd, t_path, oflag);
/* successful lookup with CREATE and EXCL is an error */
if (fd != -1) {
if ((oflag & O_CREATE) && (oflag & O_EXCL)) {
set_errno(EEXIST);
sysclose(fd);
free_path(p, t_path);
return -1;
}
} else {
if (oflag & O_CREATE) {
mode &= ~p->umask;
mode &= S_PMASK;
static_assert(!(DMMODE_BITS & S_PMASK));
fd = syscreate(t_path, oflag, mode);
}
}
free_path(p, t_path);
printd("File %s Open, fd=%d\n", path, fd);
return fd;
}
static intreg_t sys_close(struct proc *p, int fd)
{
return sysclose(fd);
}
static intreg_t sys_fstat(struct proc *p, int fd, struct kstat *u_stat)
{
struct kstat *kbuf;
kbuf = kmalloc(sizeof(struct kstat), 0);
if (!kbuf) {
set_errno(ENOMEM);
return -1;
}
if (sysfstatakaros(fd, (struct kstat *)kbuf) < 0) {
kfree(kbuf);
return -1;
}
/* TODO: UMEM: pin the memory, copy directly, and skip the kernel buffer
*/
if (memcpy_to_user_errno(p, u_stat, kbuf, sizeof(struct kstat))) {
kfree(kbuf);
return -1;
}
kfree(kbuf);
return 0;
}
/* sys_stat() and sys_lstat() do nearly the same thing, differing in how they
* treat a symlink for the final item, which (probably) will be controlled by
* the lookup flags */
static intreg_t stat_helper(struct proc *p, const char *path, size_t path_l,
struct kstat *u_stat, int flags)
{
struct kstat *kbuf;
char *t_path = copy_in_path(p, path, path_l);
int retval = 0;
if (!t_path)
return -1;
kbuf = kmalloc(sizeof(struct kstat), 0);
if (!kbuf) {
set_errno(ENOMEM);
retval = -1;
goto out_with_path;
}
retval = sysstatakaros(t_path, (struct kstat *)kbuf, flags);
if (retval < 0)
goto out_with_kbuf;
/* TODO: UMEM: pin the memory, copy directly, and skip the kernel buffer
*/
if (memcpy_to_user_errno(p, u_stat, kbuf, sizeof(struct kstat)))
retval = -1;
/* Fall-through */
out_with_kbuf:
kfree(kbuf);
out_with_path:
free_path(p, t_path);
return retval;
}
/* Follow a final symlink */
static intreg_t sys_stat(struct proc *p, const char *path, size_t path_l,
struct kstat *u_stat)
{
return stat_helper(p, path, path_l, u_stat, 0);
}
/* Don't follow a final symlink */
static intreg_t sys_lstat(struct proc *p, const char *path, size_t path_l,
struct kstat *u_stat)
{
return stat_helper(p, path, path_l, u_stat, O_NOFOLLOW);
}
intreg_t sys_fcntl(struct proc *p, int fd, int cmd, unsigned long arg1,
unsigned long arg2, unsigned long arg3, unsigned long arg4)
{
switch (cmd) {
case (F_DUPFD):
/* TODO: glibc uses regular DUPFD for dup2, which is racy. */
return sysdup(fd, arg1, FALSE);
case (F_GETFD):
return fd_get_fd_flags(&p->open_files, fd);
case (F_SETFD):
if (arg1 & ~FD_VALID_FLAGS) {
set_error(EINVAL, "Bad FD flags %p, valid are %p", arg1,
FD_VALID_FLAGS);
return -1;
}
return fd_set_fd_flags(&p->open_files, fd, arg1);
case (F_SYNC):
return fd_chan_ctl(fd, CCTL_SYNC, 0, 0, 0, 0);
case (F_GETFL):
return fd_getfl(fd);
case (F_SETFL):
return fd_chan_ctl(fd, CCTL_SET_FL, arg1, 0, 0, 0);
default:
/* chanctl and fcntl share flags */
if (cmd >= F_CHANCTL_BASE)
return fd_chan_ctl(fd, cmd, arg1, arg2, arg3, arg4);
set_error(EINVAL, "Unsupported fcntl cmd %d", cmd);
return -1;
}
}
static intreg_t sys_access(struct proc *p, const char *path, size_t path_l,
int mode)
{
int retval;
struct dir *dir;
char *t_path = copy_in_path(p, path, path_l);
if (!t_path)
return -1;
dir = sysdirstat(t_path);
if (!dir)
goto out;
if ((mode == F_OK) ||
caller_has_dir_perms(dir, access_bits_to_omode(mode)))
retval = 0;
kfree(dir);
out:
free_path(p, t_path);
return retval;
}
intreg_t sys_umask(struct proc *p, int mask)
{
int old_mask = p->umask;
p->umask = mask & S_PMASK;
return old_mask;
}
/* 64 bit seek, with the off64_t passed in via two (potentially 32 bit) off_ts.
* We're supporting both 32 and 64 bit kernels/userspaces, but both use the
* llseek syscall with 64 bit parameters. */
static intreg_t sys_llseek(struct proc *p, int fd, off_t offset_hi,
off_t offset_lo, off64_t *result, int whence)
{
off64_t retoff = 0;
off64_t tempoff = 0;
int ret = 0;
tempoff = offset_hi;
tempoff <<= 32;
tempoff |= offset_lo;
retoff = sysseek(fd, tempoff, whence);
ret = (retoff < 0);
if (ret)
return -1;
if (memcpy_to_user_errno(p, result, &retoff, sizeof(off64_t)))
return -1;
return 0;
}
intreg_t sys_link(struct proc *p, char *old_path, size_t old_l,
char *new_path, size_t new_l)
{
int ret;
char *t_oldpath = copy_in_path(p, old_path, old_l);
if (t_oldpath == NULL)
return -1;
char *t_newpath = copy_in_path(p, new_path, new_l);
if (t_newpath == NULL) {
free_path(p, t_oldpath);
return -1;
}
set_error(ENOSYS, "no link");
ret = -1;
free_path(p, t_oldpath);
free_path(p, t_newpath);
return ret;
}
intreg_t sys_unlink(struct proc *p, const char *path, size_t path_l)
{
int retval;
char *t_path = copy_in_path(p, path, path_l);
if (!t_path)
return -1;
retval = sysremove(t_path);
free_path(p, t_path);
return retval;
}
intreg_t sys_symlink(struct proc *p, char *old_path, size_t old_l,
char *new_path, size_t new_l)
{
int ret;
char *t_oldpath = copy_in_path(p, old_path, old_l);
if (t_oldpath == NULL)
return -1;
char *t_newpath = copy_in_path(p, new_path, new_l);
if (t_newpath == NULL) {
free_path(p, t_oldpath);
return -1;
}
ret = syssymlink(t_newpath, t_oldpath);
free_path(p, t_oldpath);
free_path(p, t_newpath);
return ret;
}
intreg_t sys_readlink(struct proc *p, char *path, size_t path_l,
char *u_buf, size_t buf_l)
{
char *symname = NULL;
ssize_t copy_amt;
int ret = -1;
char *t_path = copy_in_path(p, path, path_l);
struct dir *dir = NULL;
if (t_path == NULL)
return -1;
dir = sysdirlstat(t_path);
if (!(dir->mode & DMSYMLINK))
set_error(EINVAL, "not a symlink: %s", t_path);
else
symname = dir->ext;
free_path(p, t_path);
if (symname){
copy_amt = strnlen(symname, buf_l - 1) + 1;
if (!memcpy_to_user_errno(p, u_buf, symname, copy_amt))
ret = copy_amt - 1;
}
kfree(dir);
return ret;
}
static intreg_t sys_chdir(struct proc *p, pid_t pid, const char *path,
size_t path_l)
{
int retval;
char *t_path;
struct proc *target = get_controllable_proc(p, pid);
if (!target)
return -1;
if ((target != p) && (target->state != PROC_CREATED)) {
proc_decref(target);
set_error(EINVAL, "pid %d has already started", pid);
return -1;
}
t_path = copy_in_path(p, path, path_l);
if (!t_path) {
proc_decref(target);
return -1;
}
retval = syschdir(target, t_path);
free_path(p, t_path);
proc_decref(target);
return retval;
}
static intreg_t sys_fchdir(struct proc *p, pid_t pid, int fd)
{
int retval;
struct proc *target = get_controllable_proc(p, pid);
if (!target)
return -1;
if ((target != p) && (target->state != PROC_CREATED)) {
proc_decref(target);
set_error(EINVAL, "pid %d has already started", pid);
return -1;
}
retval = sysfchdir(target, fd);
proc_decref(target);
return retval;
}
/* Note cwd_l is not a strlen, it's an absolute size.
* Same as with readlink, we give them a null-terminated string, and we return
* strlen, which doesn't include the \0. If we can't give them the \0, we'll
* error out. Our readlink also does that, which is not POSIX-like. */
intreg_t sys_getcwd(struct proc *p, char *u_cwd, size_t cwd_l)
{
ssize_t retval = -1;
size_t copy_amt;
char *k_cwd;
k_cwd = sysgetcwd();
if (!k_cwd) {
set_error(EINVAL, "unable to getcwd");
return -1;
}
copy_amt = strlen(k_cwd) + 1;
if (copy_amt > cwd_l) {
set_error(ERANGE, "getcwd buf too small, needed %d", copy_amt);
goto out;
}
if (!memcpy_to_user_errno(p, u_cwd, k_cwd, copy_amt))
retval = copy_amt - 1;
out:
kfree(k_cwd);
return retval;
}
intreg_t sys_mkdir(struct proc *p, const char *path, size_t path_l, int mode)
{
int retval;
char *t_path = copy_in_path(p, path, path_l);
if (!t_path)
return -1;
mode &= ~p->umask;
mode &= S_PMASK;
static_assert(!(DMMODE_BITS & S_PMASK));
retval = syscreate(t_path, O_READ, DMDIR | mode);
if (retval >= 0) {
sysclose(retval);
retval = 0;
}
free_path(p, t_path);
return retval;
}
intreg_t sys_rmdir(struct proc *p, const char *path, size_t path_l)
{
int retval;
char *t_path = copy_in_path(p, path, path_l);
if (!t_path)
return -1;
retval = sysremove(t_path);
free_path(p, t_path);
return retval;
}
intreg_t sys_tcgetattr(struct proc *p, int fd, void *termios_p)
{
int retval = 0;
/* TODO: actually support this call on tty FDs. Right now, we just fake
* what my linux box reports for a bash pty. */
struct termios *kbuf = kmalloc(sizeof(struct termios), 0);
kbuf->c_iflag = 0x2d02;
kbuf->c_oflag = 0x0005;
kbuf->c_cflag = 0x04bf;
kbuf->c_lflag = 0x8a3b;
kbuf->c_line = 0x0;
kbuf->c_ispeed = 0xf;
kbuf->c_ospeed = 0xf;
kbuf->c_cc[0] = 0x03;
kbuf->c_cc[1] = 0x1c;
kbuf->c_cc[2] = 0x7f;
kbuf->c_cc[3] = 0x15;
kbuf->c_cc[4] = 0x04;
kbuf->c_cc[5] = 0x00;
kbuf->c_cc[6] = 0x01;
kbuf->c_cc[7] = 0xff;
kbuf->c_cc[8] = 0x11;
kbuf->c_cc[9] = 0x13;
kbuf->c_cc[10] = 0x1a;
kbuf->c_cc[11] = 0xff;
kbuf->c_cc[12] = 0x12;
kbuf->c_cc[13] = 0x0f;
kbuf->c_cc[14] = 0x17;
kbuf->c_cc[15] = 0x16;
kbuf->c_cc[16] = 0xff;
kbuf->c_cc[17] = 0x00;
kbuf->c_cc[18] = 0x00;
kbuf->c_cc[19] = 0x00;
kbuf->c_cc[20] = 0x00;
kbuf->c_cc[21] = 0x00;
kbuf->c_cc[22] = 0x00;
kbuf->c_cc[23] = 0x00;
kbuf->c_cc[24] = 0x00;
kbuf->c_cc[25] = 0x00;
kbuf->c_cc[26] = 0x00;
kbuf->c_cc[27] = 0x00;
kbuf->c_cc[28] = 0x00;
kbuf->c_cc[29] = 0x00;
kbuf->c_cc[30] = 0x00;
kbuf->c_cc[31] = 0x00;
if (memcpy_to_user_errno(p, termios_p, kbuf, sizeof(struct termios)))
retval = -1;
kfree(kbuf);
return retval;
}
intreg_t sys_tcsetattr(struct proc *p, int fd, int optional_actions,
const void *termios_p)
{
/* TODO: do this properly too. For now, we just say 'it worked' */
return 0;
}
/* TODO: we don't have any notion of UIDs or GIDs yet, but don't let that stop a
* process from thinking it can do these. The other alternative is to have
* glibc return 0 right away, though someone might want to do something with
* these calls. Someday. */
intreg_t sys_setuid(struct proc *p, uid_t uid)
{
return 0;
}
intreg_t sys_setgid(struct proc *p, gid_t gid)
{
return 0;
}
/* long bind(char* src_path, char* onto_path, int flag);
*
* The naming for the args in bind is messy historically. We do:
* bind src_path onto_path
* plan9 says bind NEW OLD, where new is *src*, and old is *onto*.
* Linux says mount --bind OLD NEW, where OLD is *src* and NEW is *onto*. */
intreg_t sys_nbind(struct proc *p,
char *src_path, size_t src_l,
char *onto_path, size_t onto_l,
unsigned int flag)
{
int ret;
char *t_srcpath = copy_in_path(p, src_path, src_l);
if (t_srcpath == NULL) {
printd("srcpath dup failed ptr %p size %d\n", src_path, src_l);
return -1;
}
char *t_ontopath = copy_in_path(p, onto_path, onto_l);
if (t_ontopath == NULL) {
free_path(p, t_srcpath);
printd("ontopath dup failed ptr %p size %d\n", onto_path,
onto_l);
return -1;
}
printd("sys_nbind: %s -> %s flag %d\n", t_srcpath, t_ontopath, flag);
ret = sysbind(t_srcpath, t_ontopath, flag);
free_path(p, t_srcpath);
free_path(p, t_ontopath);
return ret;
}
/* int mount(int fd, int afd, char* onto_path, int flag, char* aname); */
intreg_t sys_nmount(struct proc *p,
int fd,
char *onto_path, size_t onto_l,
unsigned int flag
/* we ignore these */
/* no easy way to pass this many args anyway. *
int afd,
char *auth, size_t auth_l*/)
{
int ret;
int afd;
afd = -1;
char *t_ontopath = copy_in_path(p, onto_path, onto_l);
if (t_ontopath == NULL)
return -1;
ret = sysmount(fd, afd, t_ontopath, flag, /* spec or auth */"/");
free_path(p, t_ontopath);
return ret;
}
/* Unmount undoes the operation of a bind or mount. Check out
* http://plan9.bell-labs.com/magic/man2html/1/bind . Though our mount takes an
* FD, not servename (aka src_path), so it's not quite the same.
*
* To translate between Plan 9 and Akaros, old -> onto_path. new -> src_path.
*
* For unmount, src_path / new is optional. If set, we only unmount the
* bindmount that came from src_path. */
intreg_t sys_nunmount(struct proc *p, char *src_path, int src_l,
char *onto_path, int onto_l)
{
int ret;
char *t_ontopath, *t_srcpath;
t_ontopath = copy_in_path(p, onto_path, onto_l);
if (t_ontopath == NULL)
return -1;
if (src_path) {
t_srcpath = copy_in_path(p, src_path, src_l);
if (t_srcpath == NULL) {
free_path(p, t_ontopath);
return -1;
}
} else {
t_srcpath = 0;
}
ret = sysunmount(t_srcpath, t_ontopath);
free_path(p, t_ontopath);
free_path(p, t_srcpath); /* you can free a null path */
return ret;
}
intreg_t sys_fd2path(struct proc *p, int fd, void *u_buf, size_t len)
{
int ret = 0;
struct chan *ch;
ERRSTACK(1);
/* UMEM: Check the range, can PF later and kill if the page isn't
* present */
if (!is_user_rwaddr(u_buf, len)) {
set_error(EINVAL, "bad user addr %p + %p", u_buf, len);
return -1;
}
/* fdtochan throws */
if (waserror()) {
poperror();
return -1;
}
ch = fdtochan(&current->open_files, fd, -1, FALSE, TRUE);
if (snprintf(u_buf, len, "%s", channame(ch)) >= len) {
set_error(ERANGE, "fd2path buf too small, needed %d", ret);
ret = -1;
}
cclose(ch);
poperror();
return ret;
}
intreg_t sys_wstat(struct proc *p, char *path, size_t path_l,
uint8_t *stat_m, size_t stat_sz, int flags)
{
int retval = 0;
char *t_path;
if (!is_user_rwaddr(stat_m, stat_sz)) {
set_error(EINVAL, "bad user addr %p + %p", stat_m, stat_sz);
return -1;
}
t_path = copy_in_path(p, path, path_l);
if (!t_path)
return -1;
retval = syswstat(t_path, stat_m, stat_sz);
free_path(p, t_path);
return retval;
}
intreg_t sys_fwstat(struct proc *p, int fd, uint8_t *stat_m, size_t stat_sz,
int flags)
{
if (!is_user_rwaddr(stat_m, stat_sz)) {
set_error(EINVAL, "bad user addr %p + %p", stat_m, stat_sz);
return -1;
}
return sysfwstat(fd, stat_m, stat_sz);
}
intreg_t sys_rename(struct proc *p, char *old_path, size_t old_path_l,
char *new_path, size_t new_path_l)
{
char *from_path = copy_in_path(p, old_path, old_path_l);
char *to_path = copy_in_path(p, new_path, new_path_l);
int ret;
if ((!from_path) || (!to_path))
return -1;
ret = sysrename(from_path, to_path);
free_path(p, from_path);
free_path(p, to_path);
return ret;
}
/* Careful: if an FD is busy, we don't close the old object, it just fails */
static intreg_t sys_dup_fds_to(struct proc *p, unsigned int pid,
struct childfdmap *map, unsigned int nentries)
{
ssize_t ret = 0;
struct proc *child;
int slot;
if (!is_user_rwaddr(map, sizeof(struct childfdmap) * nentries)) {
set_error(EINVAL, "bad user addr %p + %p", map,
sizeof(struct childfdmap) * nentries);
return -1;
}
child = get_controllable_proc(p, pid);
if (!child)
return -1;
for (int i = 0; i < nentries; i++) {
map[i].ok = -1;
if (!sys_dup_to(p, map[i].parentfd, child, map[i].childfd)) {
map[i].ok = 0;
ret++;
continue;
}
/* probably a bug, could send EBADF, maybe via 'ok' */
printk("[kernel] dup_fds_to: couldn't find %d\n", map[i].parentfd);
}
proc_decref(child);
return ret;
}
/* 0 on success, anything else is an error, with errno/errstr set */
static int handle_tap_req(struct proc *p, struct fd_tap_req *req)
{
switch (req->cmd) {
case (FDTAP_CMD_ADD):
return add_fd_tap(p, req);
case (FDTAP_CMD_REM):
return remove_fd_tap(p, req->fd);
default:
set_error(ENOSYS, "FD Tap Command %d not supported", req->cmd);
return -1;
}
}
/* Processes up to nr_reqs tap requests. If a request errors out, we stop
* immediately. Returns the number processed. If done != nr_reqs, check errno
* and errstr for the last failure, which is for tap_reqs[done]. */
static intreg_t sys_tap_fds(struct proc *p, struct fd_tap_req *tap_reqs,
size_t nr_reqs)
{
struct fd_tap_req *req_i = tap_reqs;
int done;
if (!is_user_rwaddr(tap_reqs, sizeof(struct fd_tap_req) * nr_reqs)) {
set_error(EINVAL, "bad user addr %p + %p", tap_reqs,
sizeof(struct fd_tap_req) * nr_reqs);
return 0;
}
for (done = 0; done < nr_reqs; done++, req_i++) {
if (handle_tap_req(p, req_i))
break;
}
return done;
}
/************** Syscall Invokation **************/
const struct sys_table_entry syscall_table[] = {
[SYS_null] = {(syscall_t)sys_null, "null"},
[SYS_block] = {(syscall_t)sys_block, "block"},
[SYS_cache_invalidate] = {(syscall_t)sys_cache_invalidate, "wbinv"},
[SYS_reboot] = {(syscall_t)reboot, "reboot!"},
[SYS_getpcoreid] = {(syscall_t)sys_getpcoreid, "getpcoreid"},
[SYS_getvcoreid] = {(syscall_t)sys_getvcoreid, "getvcoreid"},
[SYS_proc_create] = {(syscall_t)sys_proc_create, "proc_create"},
[SYS_proc_run] = {(syscall_t)sys_proc_run, "proc_run"},
[SYS_proc_destroy] = {(syscall_t)sys_proc_destroy, "proc_destroy"},
[SYS_proc_yield] = {(syscall_t)sys_proc_yield, "proc_yield"},
[SYS_change_vcore] = {(syscall_t)sys_change_vcore, "change_vcore"},
[SYS_fork] = {(syscall_t)sys_fork, "fork"},
[SYS_exec] = {(syscall_t)sys_exec, "exec"},
[SYS_waitpid] = {(syscall_t)sys_waitpid, "waitpid"},
[SYS_mmap] = {(syscall_t)sys_mmap, "mmap"},
[SYS_munmap] = {(syscall_t)sys_munmap, "munmap"},
[SYS_mprotect] = {(syscall_t)sys_mprotect, "mprotect"},
[SYS_shared_page_alloc] = {(syscall_t)sys_shared_page_alloc, "pa"},
[SYS_shared_page_free] = {(syscall_t)sys_shared_page_free, "pf"},
[SYS_provision] = {(syscall_t)sys_provision, "provision"},
[SYS_notify] = {(syscall_t)sys_notify, "notify"},
[SYS_self_notify] = {(syscall_t)sys_self_notify, "self_notify"},
[SYS_send_event] = {(syscall_t)sys_send_event, "send_event"},
[SYS_vc_entry] = {(syscall_t)sys_vc_entry, "vc_entry"},
[SYS_halt_core] = {(syscall_t)sys_halt_core, "halt_core"},
#ifdef CONFIG_ARSC_SERVER
[SYS_init_arsc] = {(syscall_t)sys_init_arsc, "init_arsc"},
#endif
[SYS_change_to_m] = {(syscall_t)sys_change_to_m, "change_to_m"},
[SYS_vmm_add_gpcs] = {(syscall_t)sys_vmm_add_gpcs, "vmm_add_gpcs"},
[SYS_vmm_poke_guest] = {(syscall_t)sys_vmm_poke_guest, "vmm_poke_guest"},
[SYS_vmm_ctl] = {(syscall_t)sys_vmm_ctl, "vmm_ctl"},
[SYS_poke_ksched] = {(syscall_t)sys_poke_ksched, "poke_ksched"},
[SYS_abort_sysc] = {(syscall_t)sys_abort_sysc, "abort_sysc"},
[SYS_abort_sysc_fd] = {(syscall_t)sys_abort_sysc_fd, "abort_sysc_fd"},
[SYS_populate_va] = {(syscall_t)sys_populate_va, "populate_va"},
[SYS_nanosleep] = {(syscall_t)sys_nanosleep, "nanosleep"},
[SYS_pop_ctx] = {(syscall_t)sys_pop_ctx, "pop_ctx"},
[SYS_read] = {(syscall_t)sys_read, "read"},
[SYS_write] = {(syscall_t)sys_write, "write"},
[SYS_openat] = {(syscall_t)sys_openat, "openat"},
[SYS_close] = {(syscall_t)sys_close, "close"},
[SYS_fstat] = {(syscall_t)sys_fstat, "fstat"},
[SYS_stat] = {(syscall_t)sys_stat, "stat"},
[SYS_lstat] = {(syscall_t)sys_lstat, "lstat"},
[SYS_fcntl] = {(syscall_t)sys_fcntl, "fcntl"},
[SYS_access] = {(syscall_t)sys_access, "access"},
[SYS_umask] = {(syscall_t)sys_umask, "umask"},
[SYS_llseek] = {(syscall_t)sys_llseek, "llseek"},
[SYS_link] = {(syscall_t)sys_link, "link"},
[SYS_unlink] = {(syscall_t)sys_unlink, "unlink"},
[SYS_symlink] = {(syscall_t)sys_symlink, "symlink"},
[SYS_readlink] = {(syscall_t)sys_readlink, "readlink"},
[SYS_chdir] = {(syscall_t)sys_chdir, "chdir"},
[SYS_fchdir] = {(syscall_t)sys_fchdir, "fchdir"},
[SYS_getcwd] = {(syscall_t)sys_getcwd, "getcwd"},
[SYS_mkdir] = {(syscall_t)sys_mkdir, "mkdir"},
[SYS_rmdir] = {(syscall_t)sys_rmdir, "rmdir"},
[SYS_tcgetattr] = {(syscall_t)sys_tcgetattr, "tcgetattr"},
[SYS_tcsetattr] = {(syscall_t)sys_tcsetattr, "tcsetattr"},
[SYS_setuid] = {(syscall_t)sys_setuid, "setuid"},
[SYS_setgid] = {(syscall_t)sys_setgid, "setgid"},
/* special! */
[SYS_nbind] ={(syscall_t)sys_nbind, "nbind"},
[SYS_nmount] ={(syscall_t)sys_nmount, "nmount"},
[SYS_nunmount] ={(syscall_t)sys_nunmount, "nunmount"},
[SYS_fd2path] ={(syscall_t)sys_fd2path, "fd2path"},
[SYS_wstat] ={(syscall_t)sys_wstat, "wstat"},
[SYS_fwstat] ={(syscall_t)sys_fwstat, "fwstat"},
[SYS_rename] ={(syscall_t)sys_rename, "rename"},
[SYS_dup_fds_to] = {(syscall_t)sys_dup_fds_to, "dup_fds_to"},
[SYS_tap_fds] = {(syscall_t)sys_tap_fds, "tap_fds"},
};
const int max_syscall = sizeof(syscall_table)/sizeof(syscall_table[0]);
/* Executes the given syscall.
*
* Note tf is passed in, which points to the tf of the context on the kernel
* stack. If any syscall needs to block, it needs to save this info, as well as
* any silly state.
*
* This syscall function is used by both local syscall and arsc, and should
* remain oblivious of the caller. */
intreg_t syscall(struct proc *p, uintreg_t sc_num, uintreg_t a0, uintreg_t a1,
uintreg_t a2, uintreg_t a3, uintreg_t a4, uintreg_t a5)
{
intreg_t ret = -1;
ERRSTACK(1);
if (sc_num > max_syscall || syscall_table[sc_num].call == NULL) {
printk("[kernel] Invalid syscall %d for proc %d\n", sc_num,
p->pid);
printk("\tArgs: %p, %p, %p, %p, %p, %p\n", a0, a1, a2, a3, a4,
a5);
print_user_ctx(this_pcpui_var(cur_ctx));
return -1;
}
/* N.B. This is going away. */
if (waserror()){
printk("Plan 9 system call returned via waserror()\n");
printk("String: '%s'\n", current_errstr());
/* if we got here, then the errbuf was right.
* no need to check!
*/
return -1;
}
//printd("before syscall errstack %p\n", errstack);
//printd("before syscall errstack base %p\n", get_cur_errbuf());
ret = syscall_table[sc_num].call(p, a0, a1, a2, a3, a4, a5);
//printd("after syscall errstack base %p\n", get_cur_errbuf());
if (get_cur_errbuf() != &errstack[0]) {
/* Can't trust coreid and vcoreid anymore, need to check the
* trace */
printk("[%16llu] Syscall %3d (%12s):(%p, %p, %p, %p, "
"%p, %p) proc: %d\n", read_tsc(),
sc_num, syscall_table[sc_num].name, a0, a1, a2, a3,
a4, a5, p->pid);
if (sc_num != SYS_fork)
panic("errstack mismatch");
}
return ret;
}
/* Execute the syscall on the local core */
void run_local_syscall(struct syscall *sysc)
{
struct per_cpu_info *pcpui = this_pcpui_ptr();
struct proc *p = pcpui->cur_proc;
long retval;
/* In lieu of pinning, we just check the sysc and will PF on the user
* addr later (if the addr was unmapped). Which is the plan for all
* UMEM. */
if (!is_user_rwaddr(sysc, sizeof(struct syscall))) {
printk("[kernel] bad user addr %p (+%p) in %s (user bug)\n",
sysc, sizeof(struct syscall), __FUNCTION__);
return;
}
pcpui->cur_kthread->sysc = sysc;/* let the core know which sysc it is */
unset_errno();
systrace_start_trace(pcpui->cur_kthread, sysc);
pcpui = this_pcpui_ptr(); /* reload again */
alloc_sysc_str(pcpui->cur_kthread);
/* syscall() does not return for exec and yield, so put any cleanup in
* there too. */
retval = syscall(pcpui->cur_proc, sysc->num, sysc->arg0, sysc->arg1,
sysc->arg2, sysc->arg3, sysc->arg4, sysc->arg5);
finish_current_sysc(retval);
}
/* A process can trap and call this function, which will set up the core to
* handle all the syscalls. a.k.a. "sys_debutante(needs, wants)". If there is
* at least one, it will run it directly. */
void prep_syscalls(struct proc *p, struct syscall *sysc, unsigned int nr_syscs)
{
/* Careful with pcpui here, we could have migrated */
if (!nr_syscs) {
printk("[kernel] No nr_sysc, probably a bug, user!\n");
return;
}
/* For all after the first call, send ourselves a KMSG (TODO). */
if (nr_syscs != 1)
warn("Only one supported (Debutante calls: %d)\n", nr_syscs);
/* Call the first one directly. (we already checked to make sure there
* is 1) */
run_local_syscall(sysc);
}
/* Call this when something happens on the syscall where userspace might want to
* get signaled. Passing p, since the caller should know who the syscall
* belongs to (probably is current).
*
* You need to have SC_K_LOCK set when you call this. */
void __signal_syscall(struct syscall *sysc, struct proc *p)
{
struct event_queue *ev_q;
struct event_msg local_msg;
/* User sets the ev_q then atomically sets the flag (races with SC_DONE)
*/
if (atomic_read(&sysc->flags) & SC_UEVENT) {
rmb(); /* read the ev_q after reading the flag */
ev_q = sysc->ev_q;
if (ev_q) {
memset(&local_msg, 0, sizeof(struct event_msg));
local_msg.ev_type = EV_SYSCALL;
local_msg.ev_arg3 = sysc;
if (!is_user_rwaddr(ev_q, sizeof(struct event_queue))) {
printk("[kernel] syscall had bad ev_q %p\n",
ev_q);
return;
}
send_event(p, ev_q, &local_msg, 0);
}
}
}
bool syscall_uses_fd(struct syscall *sysc, int fd)
{
switch (sysc->num) {
case (SYS_read):
case (SYS_write):
case (SYS_close):
case (SYS_fstat):
case (SYS_fcntl):
case (SYS_llseek):
case (SYS_nmount):
case (SYS_fd2path):
if (sysc->arg0 == fd)
return TRUE;
return FALSE;
case (SYS_mmap):
/* mmap always has to be special. =) */
if (sysc->arg4 == fd)
return TRUE;
return FALSE;
default:
return FALSE;
}
}
void print_sysc(struct proc *p, struct syscall *sysc)
{
uintptr_t old_p = switch_to(p);
printk("SYS_%d, flags %p, a0 %p, a1 %p, a2 %p, a3 %p, a4 %p, a5 %p\n",
sysc->num, atomic_read(&sysc->flags),
sysc->arg0, sysc->arg1, sysc->arg2, sysc->arg3, sysc->arg4,
sysc->arg5);
switch_back(p, old_p);
}
/* Called when we try to return from a panic. */
void kth_panic_sysc(struct kthread *kth)
{
kth->sysc = NULL;
/* We actually could block here, but that might be OK, since we cleared
* cur_kthread->sysc. As OK as anything is after a panic... */
systrace_finish_trace(kth, -12345);
}