SYSCALL_DEFINE3(sparc_sigaction, int, sig,
struct old_sigaction __user *,act,
struct old_sigaction __user *,oact)
{
WARN_ON_ONCE(sig >= 0);
return sys_sigaction(-sig, act, oact);
}
void
test_rw(int signum, unsigned char *act, unsigned char *oldact, size_t sigsetsize)
{
int prot_rw = PROT_READ | PROT_WRITE;
int ret_sys, ret_sim;
memrand(test_rw_sys, sizeof(test_rw_sys));
memcpy(test_rw_sim, test_rw_sys, sizeof(test_rw_sys));
ret_sys = sys_sigaction(signum, act, oldact, sigsetsize);
ret_sim = sim_sigaction(signum,
act == NULL ? NULL : test_rw_sim + (act - test_rw_sys),
oldact == NULL ? NULL : test_rw_sim + (oldact - test_rw_sys),
sigsetsize, prot_rw, prot_rw);
assert(ret_sys == ret_sim);
assert(memcmp(test_rw_sys, test_rw_sim, sizeof(test_rw_sys)) == 0);
}
static int dump_sigact(struct parasite_dump_sa_args *da)
{
int sig, ret = 0;
for (sig = 1; sig < SIGMAX; sig++) {
if (sig == SIGKILL || sig == SIGSTOP)
continue;
ret = sys_sigaction(sig, NULL, &da->sas[sig], sizeof(rt_sigset_t));
if (ret < 0) {
sys_write_msg("sys_sigaction failed\n");
break;
}
}
return ret;
}
void
test_edge(int prot1, int prot2, unsigned char *act, unsigned char *oldact, int ret)
{
int ret_sys, ret_sim;
assert(act == NULL || oldact == NULL);
mprotect_nofail(test_edge_base, test_edge_size, prot1);
mprotect_nofail(test_edge_base + test_edge_size, test_edge_size, prot2);
ret_sys = sys_sigaction(SIG1, act, oldact, SIGSETSIZE);
/* Check that the real syscall gave the expected return value. */
assert(ret_sys == ret);
/* If the real syscall succeeded, run the simulated syscall and compare. */
if (ret == 0) {
unsigned char tmp[SIGACTSZ];
int prot = PROT_READ | PROT_WRITE;
ret_sim = sim_sigaction(SIG1, act, tmp, SIGSETSIZE, prot, prot);
assert(ret_sim == ret);
assert(oldact == NULL || memcmp(oldact, tmp, SIGACTSZ) == 0);
}
}
/* The protection is specified by t1 and t2, referring to test_prots.
* The value 0 means a null pointer so there is no memory to protect.
*/
void
test_prot(int signum, int t1, int t2, size_t sigsetsize)
{
unsigned char sim_new[SIGACTSZ + 2 * MARGIN];
unsigned char sim_old[SIGACTSZ + 2 * MARGIN];
const size_t sim_array_size = sizeof(sim_new);
int prot1 = test_prots[t1];
int prot2 = test_prots[t2];
int ret_sys, ret_sim;
memrand(sim_new, sim_array_size);
memcpy(test_prot_mem1, sim_new, sim_array_size);
memrand(sim_old, sim_array_size);
memcpy(test_prot_mem2, sim_old, sim_array_size);
if (t1 != 0)
mprotect_nofail(test_prot_mem1, sim_array_size, prot1);
if (t2 != 0)
mprotect_nofail(test_prot_mem2, sim_array_size, prot2);
ret_sys = sys_sigaction(signum,
t1 == 0 ? NULL : test_prot_mem1 + MARGIN,
t2 == 0 ? NULL : test_prot_mem2 + MARGIN,
sigsetsize);
if (t1 != 0)
mprotect_nofail(test_prot_mem1, sim_array_size, PROT_READ | PROT_WRITE);
if (t2 != 0)
mprotect_nofail(test_prot_mem2, sim_array_size, PROT_READ | PROT_WRITE);
ret_sim = sim_sigaction(signum,
t1 == 0 ? NULL : sim_new + MARGIN,
t2 == 0 ? NULL : sim_old + MARGIN,
sigsetsize, prot1, prot2);
assert(ret_sys == ret_sim ||
/* 32-bit on a 64-bit kernel returns -ENXIO for invalid oact (i#1984) */
(ret_sys == -ENXIO && ret_sim == -EFAULT));
assert(memcmp(test_prot_mem1, sim_new, sim_array_size) == 0 &&
memcmp(test_prot_mem2, sim_old, sim_array_size) == 0);
}
static void ListerThread(struct ListerParams *args) {
int found_parent = 0;
pid_t clone_pid = sys_gettid(), ppid = sys_getppid();
char proc_self_task[80], marker_name[48], *marker_path;
const char *proc_paths[3];
const char *const *proc_path = proc_paths;
int proc = -1, marker = -1, num_threads = 0;
int max_threads = 0, sig;
struct kernel_stat marker_sb, proc_sb;
stack_t altstack;
/* Create "marker" that we can use to detect threads sharing the same
* address space and the same file handles. By setting the FD_CLOEXEC flag
* we minimize the risk of misidentifying child processes as threads;
* and since there is still a race condition, we will filter those out
* later, anyway.
*/
if ((marker = sys_socket(PF_LOCAL, SOCK_DGRAM, 0)) < 0 ||
sys_fcntl(marker, F_SETFD, FD_CLOEXEC) < 0) {
failure:
args->result = -1;
args->err = errno;
if (marker >= 0)
NO_INTR(sys_close(marker));
sig_marker = marker = -1;
if (proc >= 0)
NO_INTR(sys_close(proc));
sig_proc = proc = -1;
sys__exit(1);
}
/* Compute search paths for finding thread directories in /proc */
local_itoa(strrchr(strcpy(proc_self_task, "/proc/"), '\000'), ppid);
strcpy(marker_name, proc_self_task);
marker_path = marker_name + strlen(marker_name);
strcat(proc_self_task, "/task/");
proc_paths[0] = proc_self_task; /* /proc/$$/task/ */
proc_paths[1] = "/proc/"; /* /proc/ */
proc_paths[2] = NULL;
/* Compute path for marker socket in /proc */
local_itoa(strcpy(marker_path, "/fd/") + 4, marker);
if (sys_stat(marker_name, &marker_sb) < 0) {
goto failure;
}
/* Catch signals on an alternate pre-allocated stack. This way, we can
* safely execute the signal handler even if we ran out of memory.
*/
memset(&altstack, 0, sizeof(altstack));
altstack.ss_sp = args->altstack_mem;
altstack.ss_flags = 0;
altstack.ss_size = ALT_STACKSIZE;
sys_sigaltstack(&altstack, (const stack_t *)NULL);
/* Some kernels forget to wake up traced processes, when the
* tracer dies. So, intercept synchronous signals and make sure
* that we wake up our tracees before dying. It is the caller's
* responsibility to ensure that asynchronous signals do not
* interfere with this function.
*/
sig_marker = marker;
sig_proc = -1;
for (sig = 0; sig < sizeof(sync_signals)/sizeof(*sync_signals); sig++) {
struct kernel_sigaction sa;
memset(&sa, 0, sizeof(sa));
sa.sa_sigaction_ = SignalHandler;
sys_sigfillset(&sa.sa_mask);
sa.sa_flags = SA_ONSTACK|SA_SIGINFO|SA_RESETHAND;
sys_sigaction(sync_signals[sig], &sa, (struct kernel_sigaction *)NULL);
}
/* Read process directories in /proc/... */
for (;;) {
/* Some kernels know about threads, and hide them in "/proc"
* (although they are still there, if you know the process
* id). Threads are moved into a separate "task" directory. We
* check there first, and then fall back on the older naming
* convention if necessary.
*/
if ((sig_proc = proc = c_open(*proc_path, O_RDONLY|O_DIRECTORY, 0)) < 0) {
if (*++proc_path != NULL)
continue;
goto failure;
}
if (sys_fstat(proc, &proc_sb) < 0)
goto failure;
/* Since we are suspending threads, we cannot call any libc
* functions that might acquire locks. Most notably, we cannot
* call malloc(). So, we have to allocate memory on the stack,
* instead. Since we do not know how much memory we need, we
* make a best guess. And if we guessed incorrectly we retry on
* a second iteration (by jumping to "detach_threads").
*
* Unless the number of threads is increasing very rapidly, we
* should never need to do so, though, as our guestimate is very
* conservative.
*/
if (max_threads < proc_sb.st_nlink + 100)
max_threads = proc_sb.st_nlink + 100;
/* scope */ {
pid_t pids[max_threads];
int added_entries = 0;
sig_num_threads = num_threads;
sig_pids = pids;
for (;;) {
struct kernel_dirent *entry;
char buf[4096];
ssize_t nbytes = sys_getdents(proc, (struct kernel_dirent *)buf,
sizeof(buf));
if (nbytes < 0)
goto failure;
else if (nbytes == 0) {
if (added_entries) {
/* Need to keep iterating over "/proc" in multiple
* passes until we no longer find any more threads. This
* algorithm eventually completes, when all threads have
* been suspended.
*/
added_entries = 0;
sys_lseek(proc, 0, SEEK_SET);
continue;
}
break;
}
for (entry = (struct kernel_dirent *)buf;
entry < (struct kernel_dirent *)&buf[nbytes];
entry = (struct kernel_dirent *)((char *)entry+entry->d_reclen)) {
if (entry->d_ino != 0) {
const char *ptr = entry->d_name;
pid_t pid;
/* Some kernels hide threads by preceding the pid with a '.' */
if (*ptr == '.')
ptr++;
/* If the directory is not numeric, it cannot be a
* process/thread
*/
if (*ptr < '0' || *ptr > '9')
continue;
pid = local_atoi(ptr);
/* Attach (and suspend) all threads */
if (pid && pid != clone_pid) {
struct kernel_stat tmp_sb;
char fname[entry->d_reclen + 48];
strcat(strcat(strcpy(fname, "/proc/"),
entry->d_name), marker_path);
/* Check if the marker is identical to the one we created */
if (sys_stat(fname, &tmp_sb) >= 0 &&
marker_sb.st_ino == tmp_sb.st_ino) {
long i, j;
/* Found one of our threads, make sure it is no duplicate */
for (i = 0; i < num_threads; i++) {
/* Linear search is slow, but should not matter much for
* the typically small number of threads.
*/
if (pids[i] == pid) {
/* Found a duplicate; most likely on second pass */
goto next_entry;
}
}
/* Check whether data structure needs growing */
if (num_threads >= max_threads) {
/* Back to square one, this time with more memory */
NO_INTR(sys_close(proc));
goto detach_threads;
}
/* Attaching to thread suspends it */
pids[num_threads++] = pid;
sig_num_threads = num_threads;
if (sys_ptrace(PTRACE_ATTACH, pid, (void *)0,
(void *)0) < 0) {
/* If operation failed, ignore thread. Maybe it
* just died? There might also be a race
* condition with a concurrent core dumper or
* with a debugger. In that case, we will just
* make a best effort, rather than failing
* entirely.
*/
num_threads--;
sig_num_threads = num_threads;
goto next_entry;
}
while (sys_waitpid(pid, (int *)0, __WALL) < 0) {
if (errno != EINTR) {
sys_ptrace_detach(pid);
num_threads--;
sig_num_threads = num_threads;
goto next_entry;
}
}
if (sys_ptrace(PTRACE_PEEKDATA, pid, &i, &j) || i++ != j ||
sys_ptrace(PTRACE_PEEKDATA, pid, &i, &j) || i != j) {
/* Address spaces are distinct, even though both
* processes show the "marker". This is probably
* a forked child process rather than a thread.
*/
sys_ptrace_detach(pid);
num_threads--;
sig_num_threads = num_threads;
} else {
found_parent |= pid == ppid;
added_entries++;
}
}
}
}
next_entry:;
}
}
NO_INTR(sys_close(proc));
sig_proc = proc = -1;
/* If we failed to find any threads, try looking somewhere else in
* /proc. Maybe, threads are reported differently on this system.
*/
if (num_threads > 1 || !*++proc_path) {
NO_INTR(sys_close(marker));
sig_marker = marker = -1;
/* If we never found the parent process, something is very wrong.
* Most likely, we are running in debugger. Any attempt to operate
* on the threads would be very incomplete. Let's just report an
* error to the caller.
*/
if (!found_parent) {
ResumeAllProcessThreads(num_threads, pids);
sys__exit(3);
}
/* Now we are ready to call the callback,
* which takes care of resuming the threads for us.
*/
args->result = args->callback(args->parameter, num_threads,
pids, args->ap);
args->err = errno;
/* Callback should have resumed threads, but better safe than sorry */
if (ResumeAllProcessThreads(num_threads, pids)) {
/* Callback forgot to resume at least one thread, report error */
args->err = EINVAL;
args->result = -1;
}
sys__exit(0);
}
detach_threads:
/* Resume all threads prior to retrying the operation */
ResumeAllProcessThreads(num_threads, pids);
sig_pids = NULL;
num_threads = 0;
sig_num_threads = num_threads;
max_threads += 100;
}
}
}
/*
* 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;
}
int sigaction(int signum, const struct sigaction *act, struct sigaction *oldact)
{
return sys_sigaction(signum, act, oldact);
}
/*
* 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;
}
