static void sigchld_handler(int signal, siginfo_t *siginfo, void *data)
{
char *r;
if (futex_get(&task_entries->start) == CR_STATE_RESTORE_SIGCHLD) {
pr_debug("%ld: Collect a zombie with (pid %d, %d)\n",
sys_getpid(), siginfo->si_pid, siginfo->si_pid);
futex_dec_and_wake(&task_entries->nr_in_progress);
futex_dec_and_wake(&zombies_inprogress);
task_entries->nr_threads--;
task_entries->nr_tasks--;
mutex_unlock(&task_entries->zombie_lock);
return;
}
if (siginfo->si_code & CLD_EXITED)
r = " exited, status=";
else if (siginfo->si_code & CLD_KILLED)
r = " killed by signal ";
else
r = "disappeared with ";
pr_info("Task %d %s %d\n", siginfo->si_pid, r, siginfo->si_status);
futex_abort_and_wake(&task_entries->nr_in_progress);
/* sa_restorer may be unmaped, so we can't go back to userspace*/
sys_kill(sys_getpid(), SIGSTOP);
sys_exit_group(1);
}
__attribute__((noreturn)) static void fail(const char *filename,
const char *message,
const char *item1, int value1,
const char *item2, int value2) {
char valbuf1[32];
char valbuf2[32];
struct kernel_iovec iov[] = {
STRING_IOV("bootstrap_helper: ", 1),
{ (void *) filename, my_strlen(filename) },
STRING_IOV(": ", 1),
{ (void *) message, my_strlen(message) },
{ (void *) item1, item1 == NULL ? 0 : my_strlen(item1) },
STRING_IOV("=", item1 != NULL),
{ NULL, 0 }, /* iov[6] */
STRING_IOV(", ", item1 != NULL && item2 != NULL),
{ (void *) item2, item2 == NULL ? 0 : my_strlen(item2) },
STRING_IOV("=", item2 != NULL),
{ NULL, 0 }, /* iov[10] */
{ "\n", 1 },
};
const int niov = sizeof(iov) / sizeof(iov[0]);
if (item1 != NULL)
iov_int_string(value1, &iov[6], valbuf1, sizeof(valbuf1));
if (item2 != NULL)
iov_int_string(value2, &iov[10], valbuf2, sizeof(valbuf2));
sys_writev(2, iov, niov);
sys_exit_group(2);
while (1) *(volatile int *) 0 = 0; /* Crash. */
}
/*
* Threads restoration via sigreturn. Note it's locked
* routine and calls for unlock at the end.
*/
long __export_restore_thread(struct thread_restore_args *args)
{
struct rt_sigframe *rt_sigframe;
k_rtsigset_t to_block;
unsigned long new_sp;
int my_pid = sys_gettid();
int ret;
if (my_pid != args->pid) {
pr_err("Thread pid mismatch %d/%d\n", my_pid, args->pid);
goto core_restore_end;
}
/* All signals must be handled by thread leader */
ksigfillset(&to_block);
ret = sys_sigprocmask(SIG_SETMASK, &to_block, NULL, sizeof(k_rtsigset_t));
if (ret) {
pr_err("Unable to block signals %d", ret);
goto core_restore_end;
}
rt_sigframe = (void *)args->mem_zone.rt_sigframe;
if (restore_thread_common(rt_sigframe, args))
goto core_restore_end;
ret = restore_creds(&args->ta->creds);
if (ret)
goto core_restore_end;
ret = restore_dumpable_flag(&args->ta->mm);
if (ret)
goto core_restore_end;
pr_info("%ld: Restored\n", sys_gettid());
restore_finish_stage(CR_STATE_RESTORE);
if (restore_signals(args->siginfo, args->siginfo_nr, false))
goto core_restore_end;
restore_finish_stage(CR_STATE_RESTORE_SIGCHLD);
restore_pdeath_sig(args);
restore_finish_stage(CR_STATE_RESTORE_CREDS);
futex_dec_and_wake(&thread_inprogress);
new_sp = (long)rt_sigframe + SIGFRAME_OFFSET;
rst_sigreturn(new_sp);
core_restore_end:
pr_err("Restorer abnormal termination for %ld\n", sys_getpid());
futex_abort_and_wake(&task_entries->nr_in_progress);
sys_exit_group(1);
return -1;
}
/*
* Threads restoration via sigreturn. Note it's locked
* routine and calls for unlock at the end.
*/
long __export_restore_thread(struct thread_restore_args *args)
{
struct rt_sigframe *rt_sigframe;
unsigned long new_sp;
int my_pid = sys_gettid();
int ret;
if (my_pid != args->pid) {
pr_err("Thread pid mismatch %d/%d\n", my_pid, args->pid);
goto core_restore_end;
}
rt_sigframe = (void *)args->mem_zone.rt_sigframe + 8;
if (restore_thread_common(rt_sigframe, args))
goto core_restore_end;
mutex_unlock(&args->ta->rst_lock);
ret = restore_creds(&args->ta->creds);
if (ret)
goto core_restore_end;
pr_info("%ld: Restored\n", sys_gettid());
restore_finish_stage(CR_STATE_RESTORE);
if (restore_signals(args->siginfo, args->siginfo_nr, false))
goto core_restore_end;
restore_finish_stage(CR_STATE_RESTORE_SIGCHLD);
restore_finish_stage(CR_STATE_RESTORE_CREDS);
futex_dec_and_wake(&thread_inprogress);
new_sp = (long)rt_sigframe + SIGFRAME_OFFSET;
ARCH_RT_SIGRETURN(new_sp);
core_restore_end:
pr_err("Restorer abnormal termination for %ld\n", sys_getpid());
futex_abort_and_wake(&task_entries->nr_in_progress);
sys_exit_group(1);
return -1;
}
/*
* The main routine to restore task via sigreturn.
* This one is very special, we never return there
* but use sigreturn facility to restore core registers
* and jump execution to some predefined ip read from
* core file.
*/
long __export_restore_task(struct task_restore_args *args)
{
long ret = -1;
int i;
VmaEntry *vma_entry;
unsigned long va;
struct rt_sigframe *rt_sigframe;
unsigned long new_sp;
k_rtsigset_t to_block;
pid_t my_pid = sys_getpid();
rt_sigaction_t act;
bootstrap_start = args->bootstrap_start;
bootstrap_len = args->bootstrap_len;
#ifdef CONFIG_VDSO
vdso_rt_size = args->vdso_rt_size;
#endif
task_entries = args->task_entries;
helpers = args->helpers;
n_helpers = args->n_helpers;
*args->breakpoint = rst_sigreturn;
ksigfillset(&act.rt_sa_mask);
act.rt_sa_handler = sigchld_handler;
act.rt_sa_flags = SA_SIGINFO | SA_RESTORER | SA_RESTART;
act.rt_sa_restorer = cr_restore_rt;
sys_sigaction(SIGCHLD, &act, NULL, sizeof(k_rtsigset_t));
log_set_fd(args->logfd);
log_set_loglevel(args->loglevel);
cap_last_cap = args->cap_last_cap;
pr_info("Switched to the restorer %d\n", my_pid);
#ifdef CONFIG_VDSO
if (vdso_do_park(&args->vdso_sym_rt, args->vdso_rt_parked_at, vdso_rt_size))
goto core_restore_end;
#endif
if (unmap_old_vmas((void *)args->premmapped_addr, args->premmapped_len,
bootstrap_start, bootstrap_len))
goto core_restore_end;
/* Shift private vma-s to the left */
for (i = 0; i < args->nr_vmas; i++) {
vma_entry = args->tgt_vmas + i;
if (!vma_entry_is(vma_entry, VMA_AREA_REGULAR))
continue;
if (!vma_priv(vma_entry))
continue;
if (vma_entry->end >= TASK_SIZE)
continue;
if (vma_entry->start > vma_entry->shmid)
break;
if (vma_remap(vma_premmaped_start(vma_entry),
vma_entry->start, vma_entry_len(vma_entry)))
goto core_restore_end;
}
/* Shift private vma-s to the right */
for (i = args->nr_vmas - 1; i >= 0; i--) {
vma_entry = args->tgt_vmas + i;
if (!vma_entry_is(vma_entry, VMA_AREA_REGULAR))
continue;
if (!vma_priv(vma_entry))
continue;
if (vma_entry->start > TASK_SIZE)
continue;
if (vma_entry->start < vma_entry->shmid)
break;
if (vma_remap(vma_premmaped_start(vma_entry),
vma_entry->start, vma_entry_len(vma_entry)))
goto core_restore_end;
}
/*
* OK, lets try to map new one.
*/
for (i = 0; i < args->nr_vmas; i++) {
vma_entry = args->tgt_vmas + i;
if (!vma_entry_is(vma_entry, VMA_AREA_REGULAR))
continue;
if (vma_priv(vma_entry))
continue;
va = restore_mapping(vma_entry);
if (va != vma_entry->start) {
pr_err("Can't restore %"PRIx64" mapping with %lx\n", vma_entry->start, va);
goto core_restore_end;
}
}
#ifdef CONFIG_VDSO
/*
* Proxify vDSO.
*/
for (i = 0; i < args->nr_vmas; i++) {
if (vma_entry_is(&args->tgt_vmas[i], VMA_AREA_VDSO) ||
vma_entry_is(&args->tgt_vmas[i], VMA_AREA_VVAR)) {
if (vdso_proxify("dumpee", &args->vdso_sym_rt,
args->vdso_rt_parked_at,
i, args->tgt_vmas, args->nr_vmas))
goto core_restore_end;
break;
}
}
#endif
/*
* Walk though all VMAs again to drop PROT_WRITE
* if it was not there.
*/
for (i = 0; i < args->nr_vmas; i++) {
vma_entry = args->tgt_vmas + i;
if (!(vma_entry_is(vma_entry, VMA_AREA_REGULAR)))
continue;
if (vma_entry_is(vma_entry, VMA_ANON_SHARED)) {
struct shmem_info *entry;
entry = find_shmem(args->shmems, args->nr_shmems,
vma_entry->shmid);
if (entry && entry->pid == my_pid &&
entry->start == vma_entry->start)
futex_set_and_wake(&entry->lock, 1);
}
if (vma_entry->prot & PROT_WRITE)
continue;
sys_mprotect(decode_pointer(vma_entry->start),
vma_entry_len(vma_entry),
vma_entry->prot);
}
/*
* Finally restore madivse() bits
*/
for (i = 0; i < args->nr_vmas; i++) {
unsigned long m;
vma_entry = args->tgt_vmas + i;
if (!vma_entry->has_madv || !vma_entry->madv)
continue;
for (m = 0; m < sizeof(vma_entry->madv) * 8; m++) {
if (vma_entry->madv & (1ul << m)) {
ret = sys_madvise(vma_entry->start,
vma_entry_len(vma_entry),
m);
if (ret) {
pr_err("madvise(%"PRIx64", %"PRIu64", %ld) "
"failed with %ld\n",
vma_entry->start,
vma_entry_len(vma_entry),
m, ret);
goto core_restore_end;
}
}
}
}
ret = 0;
/*
* Tune up the task fields.
*/
ret |= sys_prctl_safe(PR_SET_NAME, (long)args->comm, 0, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_START_CODE, (long)args->mm.mm_start_code, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_END_CODE, (long)args->mm.mm_end_code, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_START_DATA, (long)args->mm.mm_start_data, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_END_DATA, (long)args->mm.mm_end_data, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_START_STACK, (long)args->mm.mm_start_stack, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_START_BRK, (long)args->mm.mm_start_brk, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_BRK, (long)args->mm.mm_brk, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_ARG_START, (long)args->mm.mm_arg_start, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_ARG_END, (long)args->mm.mm_arg_end, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_ENV_START, (long)args->mm.mm_env_start, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_ENV_END, (long)args->mm.mm_env_end, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_AUXV, (long)args->mm_saved_auxv, args->mm_saved_auxv_size);
if (ret)
goto core_restore_end;
/*
* Because of requirements applied from kernel side
* we need to restore /proc/pid/exe symlink late,
* after old existing VMAs are superseded with
* new ones from image file.
*/
ret = restore_self_exe_late(args);
if (ret)
goto core_restore_end;
/*
* We need to prepare a valid sigframe here, so
* after sigreturn the kernel will pick up the
* registers from the frame, set them up and
* finally pass execution to the new IP.
*/
rt_sigframe = (void *)args->t->mem_zone.rt_sigframe;
if (restore_thread_common(rt_sigframe, args->t))
goto core_restore_end;
/*
* Threads restoration. This requires some more comments. This
* restorer routine and thread restorer routine has the following
* memory map, prepared by a caller code.
*
* | <-- low addresses high addresses --> |
* +-------------------------------------------------------+-----------------------+
* | this proc body | own stack | rt_sigframe space | thread restore zone |
* +-------------------------------------------------------+-----------------------+
*
* where each thread restore zone is the following
*
* | <-- low addresses high addresses --> |
* +--------------------------------------------------------------------------+
* | thread restore proc | thread1 stack | thread1 rt_sigframe |
* +--------------------------------------------------------------------------+
*/
if (args->nr_threads > 1) {
struct thread_restore_args *thread_args = args->thread_args;
long clone_flags = CLONE_VM | CLONE_FILES | CLONE_SIGHAND |
CLONE_THREAD | CLONE_SYSVSEM;
long last_pid_len;
long parent_tid;
int i, fd;
fd = args->fd_last_pid;
ret = sys_flock(fd, LOCK_EX);
if (ret) {
pr_err("Can't lock last_pid %d\n", fd);
goto core_restore_end;
}
for (i = 0; i < args->nr_threads; i++) {
char last_pid_buf[16], *s;
/* skip self */
if (thread_args[i].pid == args->t->pid)
continue;
new_sp = restorer_stack(thread_args + i);
last_pid_len = vprint_num(last_pid_buf, sizeof(last_pid_buf), thread_args[i].pid - 1, &s);
sys_lseek(fd, 0, SEEK_SET);
ret = sys_write(fd, s, last_pid_len);
if (ret < 0) {
pr_err("Can't set last_pid %ld/%s\n", ret, last_pid_buf);
goto core_restore_end;
}
/*
* To achieve functionality like libc's clone()
* we need a pure assembly here, because clone()'ed
* thread will run with own stack and we must not
* have any additional instructions... oh, dear...
*/
RUN_CLONE_RESTORE_FN(ret, clone_flags, new_sp, parent_tid, thread_args, args->clone_restore_fn);
}
ret = sys_flock(fd, LOCK_UN);
if (ret) {
pr_err("Can't unlock last_pid %ld\n", ret);
goto core_restore_end;
}
}
sys_close(args->fd_last_pid);
restore_rlims(args);
ret = create_posix_timers(args);
if (ret < 0) {
pr_err("Can't restore posix timers %ld\n", ret);
goto core_restore_end;
}
ret = timerfd_arm(args);
if (ret < 0) {
pr_err("Can't restore timerfd %ld\n", ret);
goto core_restore_end;
}
pr_info("%ld: Restored\n", sys_getpid());
futex_set(&zombies_inprogress, args->nr_zombies);
restore_finish_stage(CR_STATE_RESTORE);
futex_wait_while_gt(&zombies_inprogress, 0);
if (wait_helpers(args) < 0)
goto core_restore_end;
ksigfillset(&to_block);
ret = sys_sigprocmask(SIG_SETMASK, &to_block, NULL, sizeof(k_rtsigset_t));
if (ret) {
pr_err("Unable to block signals %ld", ret);
goto core_restore_end;
}
sys_sigaction(SIGCHLD, &args->sigchld_act, NULL, sizeof(k_rtsigset_t));
ret = restore_signals(args->siginfo, args->siginfo_nr, true);
if (ret)
goto core_restore_end;
ret = restore_signals(args->t->siginfo, args->t->siginfo_nr, false);
if (ret)
goto core_restore_end;
restore_finish_stage(CR_STATE_RESTORE_SIGCHLD);
rst_tcp_socks_all(args);
/*
* Writing to last-pid is CAP_SYS_ADMIN protected,
* turning off TCP repair is CAP_SYS_NED_ADMIN protected,
* thus restore* creds _after_ all of the above.
*/
ret = restore_creds(&args->creds);
ret = ret || restore_dumpable_flag(&args->mm);
ret = ret || restore_pdeath_sig(args->t);
futex_set_and_wake(&thread_inprogress, args->nr_threads);
restore_finish_stage(CR_STATE_RESTORE_CREDS);
if (ret)
BUG();
/* Wait until children stop to use args->task_entries */
futex_wait_while_gt(&thread_inprogress, 1);
log_set_fd(-1);
/*
* The code that prepared the itimers makes shure the
* code below doesn't fail due to bad timing values.
*/
#define itimer_armed(args, i) \
(args->itimers[i].it_interval.tv_sec || \
args->itimers[i].it_interval.tv_usec)
if (itimer_armed(args, 0))
sys_setitimer(ITIMER_REAL, &args->itimers[0], NULL);
if (itimer_armed(args, 1))
sys_setitimer(ITIMER_VIRTUAL, &args->itimers[1], NULL);
if (itimer_armed(args, 2))
sys_setitimer(ITIMER_PROF, &args->itimers[2], NULL);
restore_posix_timers(args);
sys_munmap(args->rst_mem, args->rst_mem_size);
/*
* Sigframe stack.
*/
new_sp = (long)rt_sigframe + SIGFRAME_OFFSET;
/*
* Prepare the stack and call for sigreturn,
* pure assembly since we don't need any additional
* code insns from gcc.
*/
rst_sigreturn(new_sp);
core_restore_end:
futex_abort_and_wake(&task_entries->nr_in_progress);
pr_err("Restorer fail %ld\n", sys_getpid());
sys_exit_group(1);
return -1;
}
long syscall_emu(long call, long arg1, long arg2, long arg3,
long arg4, long arg5, long arg6)
{
long ret;
switch (call)
{
case __NR_brk:
case __NR_mmap2:
case __NR_mmap:
case __NR_mremap:
case __NR_mprotect:
case __NR_madvise:
case __NR_sigaltstack:
case __NR_signal:
case __NR_sigaction:
case __NR_sigreturn:
case __NR_rt_sigaction:
case __NR_rt_sigreturn:
case __NR_fork:
case __NR_vfork:
case __NR_clone:
case __NR_exit:
case __NR_execve:
case __NR_exit_group:
break;
case __NR_read:
case __NR_readv:
case __NR_open:
case __NR_creat:
case __NR_dup:
case __NR_dup2:
case __NR_openat:
case __NR_pipe:
case __NR_socketcall:
ret = syscall_intr(call,arg1,arg2,arg3,arg4,arg5,arg6);
if ( taint_flag == TAINT_ON )
do_taint(ret,call,arg1,arg2,arg3,arg4,arg5,arg6);
return ret;
case __NR_ipc:
if ( arg1 == SHMAT )
break;
/* fall through */
default:
return syscall_intr(call,arg1,arg2,arg3,arg4,arg5,arg6);
}
ret = call;
if (!try_block_signals())
return ret; /* we have a signal in progress, revert to pre-syscall state */
switch (call)
{
/* these calls are all non-blocking right?
* blocked signals during blocking calls is a bad thing
*/
case __NR_brk:
ret = user_brk(arg1);
break;
case __NR_mmap2:
ret = user_mmap2(arg1,arg2,arg3,arg4,arg5,arg6);
break;
case __NR_mmap:
ret = user_old_mmap((struct kernel_mmap_args *)arg1);
break;
case __NR_mremap:
ret = user_mremap(arg1,arg2,arg3,arg4,arg5);
break;
case __NR_mprotect:
ret = user_mprotect(arg1,arg2,arg3);
break;
case __NR_madvise:
ret = user_madvise(arg1,arg2,arg3);
break;
case __NR_ipc:
if (arg1 == SHMAT)
ret = user_shmat(arg2,(char *)arg5,arg3,(unsigned long *)arg4);
else
die("should not have caught IPC call: %d", arg1);
break;
case __NR_sigaltstack:
ret = user_sigaltstack((stack_t *)arg1, (stack_t *)arg2);
break;
case __NR_signal:
{
ret = (long)user_signal(arg1, (kernel_sighandler_t)arg2);
break;
}
case __NR_sigaction:
ret = user_sigaction(arg1, (struct kernel_old_sigaction *)arg2,
(struct kernel_old_sigaction *)arg3);
break;
case __NR_rt_sigaction:
ret = user_rt_sigaction(arg1, (struct kernel_sigaction *)arg2,
(struct kernel_sigaction *)arg3, arg4);
break;
case __NR_sigreturn:
user_sigreturn();
break;
case __NR_rt_sigreturn:
user_rt_sigreturn();
break;
case __NR_vfork:
ret = user_clone(SIGCHLD, 0, NULL, NULL, NULL);
// ret = user_clone(CLONE_VFORK | CLONE_VM | SIGCHLD, 0, NULL, NULL, NULL);
break;
case __NR_fork:
ret = user_clone(SIGCHLD, 0, NULL, NULL, NULL);
break;
case __NR_clone:
ret = user_clone(arg1, arg2, (void *)arg3, (void *)arg4, (void*)arg5);
break;
case __NR_exit:
user_exit(arg1);
break;
case __NR_execve:
ret = user_execve((char *)arg1, (char **)arg2, (char **)arg3);
break;
case __NR_exit_group:
if (dump_on_exit)
{
long regs[] = { call, arg2, arg3, arg1, get_thread_ctx()->user_esp, arg6, arg4, arg5 };
do_taint_dump(regs);
}
sys_exit_group(arg1);
default:
die("unimplemented syscall");
break;
}
unblock_signals();
return ret;
}
/*
* The main routine to restore task via sigreturn.
* This one is very special, we never return there
* but use sigreturn facility to restore core registers
* and jump execution to some predefined ip read from
* core file.
*/
long __export_restore_task(struct task_restore_core_args *args)
{
long ret = -1;
VmaEntry *vma_entry;
u64 va;
unsigned long premmapped_end = args->premmapped_addr + args->premmapped_len;
struct rt_sigframe *rt_sigframe;
unsigned long new_sp;
pid_t my_pid = sys_getpid();
rt_sigaction_t act;
task_entries = args->task_entries;
ksigfillset(&act.rt_sa_mask);
act.rt_sa_handler = sigchld_handler;
act.rt_sa_flags = SA_SIGINFO | SA_RESTORER | SA_RESTART;
act.rt_sa_restorer = cr_restore_rt;
sys_sigaction(SIGCHLD, &act, NULL, sizeof(k_rtsigset_t));
log_set_fd(args->logfd);
log_set_loglevel(args->loglevel);
cap_last_cap = args->cap_last_cap;
pr_info("Switched to the restorer %d\n", my_pid);
for (vma_entry = args->self_vmas; vma_entry->start != 0; vma_entry++) {
unsigned long addr = vma_entry->start;
unsigned long len;
if (!vma_entry_is(vma_entry, VMA_AREA_REGULAR))
continue;
pr_debug("Examine %"PRIx64"-%"PRIx64"\n", vma_entry->start, vma_entry->end);
if (addr < args->premmapped_addr) {
if (vma_entry->end >= args->premmapped_addr)
len = args->premmapped_addr - addr;
else
len = vma_entry->end - vma_entry->start;
if (sys_munmap((void *) addr, len)) {
pr_err("munmap fail for %lx - %lx\n", addr, addr + len);
goto core_restore_end;
}
}
if (vma_entry->end >= TASK_SIZE)
continue;
if (vma_entry->end > premmapped_end) {
if (vma_entry->start < premmapped_end)
addr = premmapped_end;
len = vma_entry->end - addr;
if (sys_munmap((void *) addr, len)) {
pr_err("munmap fail for %lx - %lx\n", addr, addr + len);
goto core_restore_end;
}
}
}
sys_munmap(args->self_vmas,
((void *)(vma_entry + 1) - ((void *)args->self_vmas)));
/* Shift private vma-s to the left */
for (vma_entry = args->tgt_vmas; vma_entry->start != 0; vma_entry++) {
if (!vma_entry_is(vma_entry, VMA_AREA_REGULAR))
continue;
if (!vma_priv(vma_entry))
continue;
if (vma_entry->end >= TASK_SIZE)
continue;
if (vma_entry->start > vma_entry->shmid)
break;
if (vma_remap(vma_premmaped_start(vma_entry),
vma_entry->start, vma_entry_len(vma_entry)))
goto core_restore_end;
}
/* Shift private vma-s to the right */
for (vma_entry = args->tgt_vmas + args->nr_vmas -1;
vma_entry >= args->tgt_vmas; vma_entry--) {
if (!vma_entry_is(vma_entry, VMA_AREA_REGULAR))
continue;
if (!vma_priv(vma_entry))
continue;
if (vma_entry->start > TASK_SIZE)
continue;
if (vma_entry->start < vma_entry->shmid)
break;
if (vma_remap(vma_premmaped_start(vma_entry),
vma_entry->start, vma_entry_len(vma_entry)))
goto core_restore_end;
}
/*
* OK, lets try to map new one.
*/
for (vma_entry = args->tgt_vmas; vma_entry->start != 0; vma_entry++) {
if (!vma_entry_is(vma_entry, VMA_AREA_REGULAR))
continue;
if (vma_priv(vma_entry))
continue;
va = restore_mapping(vma_entry);
if (va != vma_entry->start) {
pr_err("Can't restore %"PRIx64" mapping with %"PRIx64"\n", vma_entry->start, va);
goto core_restore_end;
}
}
/*
* Walk though all VMAs again to drop PROT_WRITE
* if it was not there.
*/
for (vma_entry = args->tgt_vmas; vma_entry->start != 0; vma_entry++) {
if (!(vma_entry_is(vma_entry, VMA_AREA_REGULAR)))
continue;
if (vma_entry_is(vma_entry, VMA_ANON_SHARED)) {
struct shmem_info *entry;
entry = find_shmem(args->shmems,
vma_entry->shmid);
if (entry && entry->pid == my_pid &&
entry->start == vma_entry->start)
futex_set_and_wake(&entry->lock, 1);
}
if (vma_entry->prot & PROT_WRITE)
continue;
sys_mprotect(decode_pointer(vma_entry->start),
vma_entry_len(vma_entry),
vma_entry->prot);
}
/*
* Finally restore madivse() bits
*/
for (vma_entry = args->tgt_vmas; vma_entry->start != 0; vma_entry++) {
unsigned long i;
if (!vma_entry->has_madv || !vma_entry->madv)
continue;
for (i = 0; i < sizeof(vma_entry->madv) * 8; i++) {
if (vma_entry->madv & (1ul << i)) {
ret = sys_madvise(vma_entry->start,
vma_entry_len(vma_entry),
i);
if (ret) {
pr_err("madvise(%"PRIx64", %"PRIu64", %ld) "
"failed with %ld\n",
vma_entry->start,
vma_entry_len(vma_entry),
i, ret);
goto core_restore_end;
}
}
}
}
sys_munmap(args->tgt_vmas,
((void *)(vma_entry + 1) - ((void *)args->tgt_vmas)));
ret = sys_munmap(args->shmems, SHMEMS_SIZE);
if (ret < 0) {
pr_err("Can't unmap shmem %ld\n", ret);
goto core_restore_end;
}
/*
* Tune up the task fields.
*/
ret |= sys_prctl_safe(PR_SET_NAME, (long)args->comm, 0, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_START_CODE, (long)args->mm.mm_start_code, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_END_CODE, (long)args->mm.mm_end_code, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_START_DATA, (long)args->mm.mm_start_data, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_END_DATA, (long)args->mm.mm_end_data, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_START_STACK, (long)args->mm.mm_start_stack, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_START_BRK, (long)args->mm.mm_start_brk, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_BRK, (long)args->mm.mm_brk, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_ARG_START, (long)args->mm.mm_arg_start, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_ARG_END, (long)args->mm.mm_arg_end, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_ENV_START, (long)args->mm.mm_env_start, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_ENV_END, (long)args->mm.mm_env_end, 0);
ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_AUXV, (long)args->mm_saved_auxv, args->mm_saved_auxv_size);
if (ret)
goto core_restore_end;
/*
* Because of requirements applied from kernel side
* we need to restore /proc/pid/exe symlink late,
* after old existing VMAs are superseded with
* new ones from image file.
*/
ret = restore_self_exe_late(args);
if (ret)
goto core_restore_end;
/*
* We need to prepare a valid sigframe here, so
* after sigreturn the kernel will pick up the
* registers from the frame, set them up and
* finally pass execution to the new IP.
*/
rt_sigframe = (void *)args->t->mem_zone.rt_sigframe + 8;
if (restore_thread_common(rt_sigframe, args->t))
goto core_restore_end;
/*
* Threads restoration. This requires some more comments. This
* restorer routine and thread restorer routine has the following
* memory map, prepared by a caller code.
*
* | <-- low addresses high addresses --> |
* +-------------------------------------------------------+-----------------------+
* | this proc body | own stack | heap | rt_sigframe space | thread restore zone |
* +-------------------------------------------------------+-----------------------+
*
* where each thread restore zone is the following
*
* | <-- low addresses high addresses --> |
* +--------------------------------------------------------------------------+
* | thread restore proc | thread1 stack | thread1 heap | thread1 rt_sigframe |
* +--------------------------------------------------------------------------+
*/
if (args->nr_threads > 1) {
struct thread_restore_args *thread_args = args->thread_args;
long clone_flags = CLONE_VM | CLONE_FILES | CLONE_SIGHAND |
CLONE_THREAD | CLONE_SYSVSEM;
long last_pid_len;
long parent_tid;
int i, fd;
fd = sys_open(LAST_PID_PATH, O_RDWR, LAST_PID_PERM);
if (fd < 0) {
pr_err("Can't open last_pid %d\n", fd);
goto core_restore_end;
}
ret = sys_flock(fd, LOCK_EX);
if (ret) {
pr_err("Can't lock last_pid %d\n", fd);
goto core_restore_end;
}
for (i = 0; i < args->nr_threads; i++) {
char last_pid_buf[16], *s;
/* skip self */
if (thread_args[i].pid == args->t->pid)
continue;
mutex_lock(&args->rst_lock);
new_sp =
RESTORE_ALIGN_STACK((long)thread_args[i].mem_zone.stack,
sizeof(thread_args[i].mem_zone.stack));
last_pid_len = vprint_num(last_pid_buf, sizeof(last_pid_buf), thread_args[i].pid - 1, &s);
ret = sys_write(fd, s, last_pid_len);
if (ret < 0) {
pr_err("Can't set last_pid %ld/%s\n", ret, last_pid_buf);
goto core_restore_end;
}
/*
* To achieve functionality like libc's clone()
* we need a pure assembly here, because clone()'ed
* thread will run with own stack and we must not
* have any additional instructions... oh, dear...
*/
RUN_CLONE_RESTORE_FN(ret, clone_flags, new_sp, parent_tid, thread_args, args->clone_restore_fn);
}
ret = sys_flock(fd, LOCK_UN);
if (ret) {
pr_err("Can't unlock last_pid %ld\n", ret);
goto core_restore_end;
}
sys_close(fd);
}
restore_rlims(args);
pr_info("%ld: Restored\n", sys_getpid());
futex_set(&zombies_inprogress, args->nr_zombies);
restore_finish_stage(CR_STATE_RESTORE);
futex_wait_while_gt(&zombies_inprogress, 0);
sys_sigaction(SIGCHLD, &args->sigchld_act, NULL, sizeof(k_rtsigset_t));
ret = restore_signals(args->siginfo, args->siginfo_nr, true);
if (ret)
goto core_restore_end;
ret = restore_signals(args->t->siginfo, args->t->siginfo_nr, false);
if (ret)
goto core_restore_end;
restore_finish_stage(CR_STATE_RESTORE_SIGCHLD);
if (args->siginfo_size) {
ret = sys_munmap(args->siginfo, args->siginfo_size);
if (ret < 0) {
pr_err("Can't unmap signals %ld\n", ret);
goto core_restore_failed;
}
}
rst_tcp_socks_all(args->rst_tcp_socks, args->rst_tcp_socks_size);
/*
* Writing to last-pid is CAP_SYS_ADMIN protected,
* turning off TCP repair is CAP_SYS_NED_ADMIN protected,
* thus restore* creds _after_ all of the above.
*/
ret = restore_creds(&args->creds);
futex_set_and_wake(&thread_inprogress, args->nr_threads);
restore_finish_stage(CR_STATE_RESTORE_CREDS);
if (ret)
BUG();
/* Wait until children stop to use args->task_entries */
futex_wait_while_gt(&thread_inprogress, 1);
log_set_fd(-1);
/*
* The code that prepared the itimers makes shure the
* code below doesn't fail due to bad timing values.
*/
#define itimer_armed(args, i) \
(args->itimers[i].it_interval.tv_sec || \
args->itimers[i].it_interval.tv_usec)
if (itimer_armed(args, 0))
sys_setitimer(ITIMER_REAL, &args->itimers[0], NULL);
if (itimer_armed(args, 1))
sys_setitimer(ITIMER_VIRTUAL, &args->itimers[1], NULL);
if (itimer_armed(args, 2))
sys_setitimer(ITIMER_PROF, &args->itimers[2], NULL);
ret = sys_munmap(args->task_entries, TASK_ENTRIES_SIZE);
if (ret < 0) {
ret = ((long)__LINE__ << 16) | ((-ret) & 0xffff);
goto core_restore_failed;
}
/*
* Sigframe stack.
*/
new_sp = (long)rt_sigframe + SIGFRAME_OFFSET;
/*
* Prepare the stack and call for sigreturn,
* pure assembly since we don't need any additional
* code insns from gcc.
*/
ARCH_RT_SIGRETURN(new_sp);
core_restore_end:
futex_abort_and_wake(&task_entries->nr_in_progress);
pr_err("Restorer fail %ld\n", sys_getpid());
sys_exit_group(1);
return -1;
core_restore_failed:
ARCH_FAIL_CORE_RESTORE;
return ret;
}
