static unsigned long get_mmap_base(elf_bin_t *elf)
{
unsigned long addr, pg_start, pg_end, mmap_min = ULONG_MAX, mmap_max = 0;
int i;
for (i=0; i<elf->hdr.e_phnum; i++)
if ( elf->phdr[i].p_type == PT_LOAD)
{
pg_start = PAGE_BASE(elf->phdr[i].p_vaddr);
pg_end = PAGE_NEXT(elf->phdr[i].p_vaddr + elf->phdr[i].p_memsz);
if (pg_start < mmap_min)
mmap_min = pg_start;
if (pg_end > mmap_max)
mmap_max = pg_end;
}
if (mmap_min > mmap_max)
mmap_min = mmap_max;
addr = do_mmap2(mmap_min, mmap_max-mmap_min, PROT_NONE,
MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
if (addr & PG_MASK) /* not on page boundary -> error code */
return addr;
sys_munmap(addr, mmap_max-mmap_min);
return addr-mmap_min;
}
static int
qcntl_ioctl (struct inode *inode, struct file *filp, unsigned int cmd, unsigned long arg, int minor)
{
struct shmiqreq req;
struct vm_area_struct *vma;
int v;
switch (cmd) {
/*
* The address space is already mapped as a /dev/zero
* mapping. FIXME: check that /dev/zero is what the user
* had mapped before :-)
*/
case QIOCATTACH: {
unsigned long vaddr;
int s;
v = verify_area (VERIFY_READ, (void *) arg,
sizeof (struct shmiqreq));
if (v)
return v;
if (copy_from_user(&req, (void *) arg, sizeof (req)))
return -EFAULT;
/*
* Do not allow to attach to another region if it has
* already been attached
*/
if (shmiqs [minor].mapped) {
printk("SHMIQ:The thingie is already mapped\n");
return -EINVAL;
}
vaddr = (unsigned long) req.user_vaddr;
vma = find_vma (current->mm, vaddr);
if (!vma) {
printk ("SHMIQ: could not find %lx the vma\n",
vaddr);
return -EINVAL;
}
s = req.arg * sizeof (struct shmqevent) +
sizeof (struct sharedMemoryInputQueue);
v = sys_munmap (vaddr, s);
down_write(¤t->mm->mmap_sem);
do_munmap(current->mm, vaddr, s);
do_mmap(filp, vaddr, s, PROT_READ | PROT_WRITE,
MAP_PRIVATE|MAP_FIXED, 0);
up_write(¤t->mm->mmap_sem);
shmiqs[minor].events = req.arg;
shmiqs[minor].mapped = 1;
return 0;
}
}
return -EINVAL;
}
static void rst_tcp_socks_all(struct rst_tcp_sock *arr, int size)
{
int i;
if (size == 0)
return;
for (i =0; arr[i].sk >= 0; i++)
rst_tcp_repair_off(arr + i);
sys_munmap(arr, size);
}
int process2()
{
int* momo2 = (int*)sys_mmap(sizeof(int) * 10);
int a2 = vmem_translate((uint32_t)momo2, current_process);
sys_yield();
sys_munmap(momo2, sizeof(int) * 10);
a2 = vmem_translate((uint32_t)momo2, current_process);
momo2++;
return a2;
}
static int fini(void)
{
int ret;
ret = fini_thread();
sys_munmap(tid_state, TID_STATE_SIZE(nr_tid_state));
log_set_fd(-1);
sys_close(tsock);
return ret;
}
/*
* This function unmaps all VMAs, which don't belong to
* the restored process or the restorer.
*
* The restorer memory is two regions -- area with restorer, its stack
* and arguments and the one with private vmas of the tasks we restore
* (a.k.a. premmaped area):
*
* 0 TASK_SIZE
* +----+====+----+====+---+
*
* Thus to unmap old memory we have to do 3 unmaps:
* [ 0 -- 1st area start ]
* [ 1st end -- 2nd start ]
* [ 2nd start -- TASK_SIZE ]
*/
static int unmap_old_vmas(void *premmapped_addr, unsigned long premmapped_len,
void *bootstrap_start, unsigned long bootstrap_len)
{
unsigned long s1, s2;
void *p1, *p2;
int ret;
if ((void *) premmapped_addr < bootstrap_start) {
p1 = premmapped_addr;
s1 = premmapped_len;
p2 = bootstrap_start;
s2 = bootstrap_len;
} else {
p2 = premmapped_addr;
s2 = premmapped_len;
p1 = bootstrap_start;
s1 = bootstrap_len;
}
ret = sys_munmap(NULL, p1 - NULL);
if (ret) {
pr_err("Unable to unmap (%p-%p): %d\n", NULL, p1, ret);
return -1;
}
ret = sys_munmap(p1 + s1, p2 - (p1 + s1));
if (ret) {
pr_err("Unable to unmap (%p-%p): %d\n", p1 + s1, p2, ret);
return -1;
}
ret = sys_munmap(p2 + s2, (void *) TASK_SIZE - (p2 + s2));
if (ret) {
pr_err("Unable to unmap (%p-%p): %d\n",
p2 + s2, (void *)TASK_SIZE, ret);
return -1;
}
return 0;
}
int process1()
{
int* momo1 = (int*)sys_mmap(sizeof(int) * 10);
int a1 = vmem_translate((uint32_t)momo1, current_process);
sys_yield();
sys_munmap(momo1, sizeof(int) * 10);
a1 = vmem_translate((uint32_t)momo1, current_process);
momo1[40000] = 50; // error : data_handler
momo1++;
return a1;
}
void qbytes_free_munmap(qbytes_t* b) {
err_t err;
/* I don't believe this is required, but
* I've heard here and there it might be for NFS...
*
rc = msync(b->data, b->len, MS_ASYNC);
if( rc ) fprintf(stderr, "Warning - in qbytes_free_munmap, msync failed with %i, errno %i\n", rc, errno);
*/
err = sys_munmap(b->data, b->len);
assert( !err );
_qbytes_free_qbytes(b);
}
/*
* Like cr_sendfile_buffered(), but for HUGETLBFS destination file.
* Uses temporary mmap()s of a chunk of len HPAGE_SIZE at a time.
*
* Note: Caller is responsible for checking count==0 or {dst,src}_ppos==NULL.
*/
static loff_t
cr_sendfile_hugedst(cr_errbuf_t *eb, struct file *dst_filp, struct file *src_filp, loff_t *src_ppos, loff_t count)
{
loff_t bytes_left = count;
loff_t retval;
struct mm_struct *mm = current->mm;
unsigned long map_addr = 0;
unsigned long map_pgoff = 0;
unsigned long map_flags = MAP_SHARED;
CRI_ASSERT((count & (HPAGE_SIZE-1)) == 0);
CRI_ASSERT(dst_filp->f_pos == 0);
CRI_ASSERT(src_ppos = &src_filp->f_pos);
for (bytes_left = count; bytes_left; bytes_left -= HPAGE_SIZE) {
unsigned long tmp;
down_write(&mm->mmap_sem);
tmp = do_mmap_pgoff(dst_filp, map_addr, HPAGE_SIZE, PROT_READ|PROT_WRITE, map_flags, map_pgoff);
up_write(&mm->mmap_sem);
if (IS_ERR((void*)tmp)) {
CR_ERR_EB(eb, "do_mmap(HUGE dst file) returned %ld", (long)tmp);
retval = tmp;
goto out_err;
}
map_addr = tmp;
map_pgoff += (HPAGE_SIZE >> PAGE_SHIFT);
map_flags |= MAP_FIXED;
retval = cr_uread(eb, src_filp, (void *)map_addr, HPAGE_SIZE);
if (retval < 0) goto out_unmap;
}
retval = count;
dst_filp->f_pos = count;
out_unmap:
if (map_addr) {
(void)sys_munmap(map_addr, HPAGE_SIZE); // XXX: check for error (unless on error path already)?
}
out_err:
return retval;
}
uint32_t elf_load_fd(struct sys_state *sys, int fd, int flags, struct elf_load_info *info)
{
uint32_t entry;
stat_s stat;
void *base;
int err;
err = sys_fstat(fd, &stat);
if (err) {
debug("Failed to stat ELF (%d)\n", err);
sys_exit(1);
}
base = sys_mmap(NULL, stat.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
if (IS_ERROR(base)) {
debug("Failed to mmap ELF file\n");
sys_exit(1);
}
entry = elf_load(sys, base, flags, info);
sys_munmap(base, stat.st_size);
return entry;
}
/*
* 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;
}
static void
syscall_handler (struct intr_frame *f)
{
int syscall_number;
ASSERT( sizeof(syscall_number) == 4 ); // assuming x86
// The system call number is in the 32-bit word at the caller's stack pointer.
memread_user(f->esp, &syscall_number, sizeof(syscall_number));
_DEBUG_PRINTF ("[DEBUG] system call, number = %d!\n", syscall_number);
// Store the esp, which is needed in the page fault handler.
// refer to exception.c:page_fault() (see manual 4.3.3)
thread_current()->current_esp = f->esp;
// Dispatch w.r.t system call number
// SYS_*** constants are defined in syscall-nr.h
switch (syscall_number) {
case SYS_HALT: // 0
{
sys_halt();
NOT_REACHED();
break;
}
case SYS_EXIT: // 1
{
int exitcode;
memread_user(f->esp + 4, &exitcode, sizeof(exitcode));
sys_exit(exitcode);
NOT_REACHED();
break;
}
case SYS_EXEC: // 2
{
void* cmdline;
memread_user(f->esp + 4, &cmdline, sizeof(cmdline));
int return_code = sys_exec((const char*) cmdline);
f->eax = (uint32_t) return_code;
break;
}
case SYS_WAIT: // 3
{
pid_t pid;
memread_user(f->esp + 4, &pid, sizeof(pid_t));
int ret = sys_wait(pid);
f->eax = (uint32_t) ret;
break;
}
case SYS_CREATE: // 4
{
const char* filename;
unsigned initial_size;
bool return_code;
memread_user(f->esp + 4, &filename, sizeof(filename));
memread_user(f->esp + 8, &initial_size, sizeof(initial_size));
return_code = sys_create(filename, initial_size);
f->eax = return_code;
break;
}
case SYS_REMOVE: // 5
{
const char* filename;
bool return_code;
memread_user(f->esp + 4, &filename, sizeof(filename));
return_code = sys_remove(filename);
f->eax = return_code;
break;
}
case SYS_OPEN: // 6
{
const char* filename;
int return_code;
memread_user(f->esp + 4, &filename, sizeof(filename));
return_code = sys_open(filename);
f->eax = return_code;
break;
}
case SYS_FILESIZE: // 7
{
int fd, return_code;
memread_user(f->esp + 4, &fd, sizeof(fd));
return_code = sys_filesize(fd);
f->eax = return_code;
break;
}
case SYS_READ: // 8
{
int fd, return_code;
void *buffer;
unsigned size;
memread_user(f->esp + 4, &fd, sizeof(fd));
memread_user(f->esp + 8, &buffer, sizeof(buffer));
memread_user(f->esp + 12, &size, sizeof(size));
return_code = sys_read(fd, buffer, size);
f->eax = (uint32_t) return_code;
break;
}
case SYS_WRITE: // 9
{
int fd, return_code;
const void *buffer;
unsigned size;
memread_user(f->esp + 4, &fd, sizeof(fd));
memread_user(f->esp + 8, &buffer, sizeof(buffer));
memread_user(f->esp + 12, &size, sizeof(size));
return_code = sys_write(fd, buffer, size);
f->eax = (uint32_t) return_code;
break;
}
case SYS_SEEK: // 10
{
int fd;
unsigned position;
memread_user(f->esp + 4, &fd, sizeof(fd));
memread_user(f->esp + 8, &position, sizeof(position));
sys_seek(fd, position);
break;
}
case SYS_TELL: // 11
{
int fd;
unsigned return_code;
memread_user(f->esp + 4, &fd, sizeof(fd));
return_code = sys_tell(fd);
f->eax = (uint32_t) return_code;
break;
}
case SYS_CLOSE: // 12
{
int fd;
memread_user(f->esp + 4, &fd, sizeof(fd));
sys_close(fd);
break;
}
#ifdef VM
case SYS_MMAP: // 13
{
int fd;
void *addr;
memread_user(f->esp + 4, &fd, sizeof(fd));
memread_user(f->esp + 8, &addr, sizeof(addr));
mmapid_t ret = sys_mmap (fd, addr);
f->eax = ret;
break;
}
case SYS_MUNMAP: // 14
{
mmapid_t mid;
memread_user(f->esp + 4, &mid, sizeof(mid));
sys_munmap(mid);
break;
}
#endif
#ifdef FILESYS
case SYS_CHDIR: // 15
{
const char* filename;
int return_code;
memread_user(f->esp + 4, &filename, sizeof(filename));
return_code = sys_chdir(filename);
f->eax = return_code;
break;
}
case SYS_MKDIR: // 16
{
const char* filename;
int return_code;
memread_user(f->esp + 4, &filename, sizeof(filename));
return_code = sys_mkdir(filename);
f->eax = return_code;
break;
}
case SYS_READDIR: // 17
{
int fd;
char *name;
int return_code;
memread_user(f->esp + 4, &fd, sizeof(fd));
memread_user(f->esp + 8, &name, sizeof(name));
return_code = sys_readdir(fd, name);
f->eax = return_code;
break;
}
case SYS_ISDIR: // 18
{
int fd;
int return_code;
memread_user(f->esp + 4, &fd, sizeof(fd));
return_code = sys_isdir(fd);
f->eax = return_code;
break;
}
case SYS_INUMBER: // 19
{
int fd;
int return_code;
memread_user(f->esp + 4, &fd, sizeof(fd));
return_code = sys_inumber(fd);
f->eax = return_code;
break;
}
#endif
/* unhandled case */
default:
printf("[ERROR] system call %d is unimplemented!\n", syscall_number);
// ensure that waiting (parent) process should wake up and terminate.
sys_exit(-1);
break;
}
}
/*
* 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;
}
static void brk_fini(void)
{
sys_munmap(brk_start, MAX_HEAP_SIZE);
}
int munmap(unsigned long addr, size_t len)
{
return sys_munmap(addr, len);
}
int munmap(uintptr_t addr, size_t len)
{
return sys_munmap(addr, len);
}
static int syscall_dispatch(uint32_t sysnum, uint32_t args, regs_t *regs)
{
switch (sysnum) {
case SYS_waitpid:
return sys_waitpid((waitpid_args_t *)args);
case SYS_exit:
do_exit((int)args);
panic("exit failed!\n");
return 0;
case SYS_thr_exit:
kthread_exit((void *)args);
panic("thr_exit failed!\n");
return 0;
case SYS_thr_yield:
sched_make_runnable(curthr);
sched_switch();
return 0;
case SYS_fork:
return sys_fork(regs);
case SYS_getpid:
return curproc->p_pid;
case SYS_sync:
sys_sync();
return 0;
#ifdef __MOUNTING__
case SYS_mount:
return sys_mount((mount_args_t *) args);
case SYS_umount:
return sys_umount((argstr_t *) args);
#endif
case SYS_mmap:
return (int) sys_mmap((mmap_args_t *) args);
case SYS_munmap:
return sys_munmap((munmap_args_t *) args);
case SYS_open:
return sys_open((open_args_t *) args);
case SYS_close:
return sys_close((int)args);
case SYS_read:
return sys_read((read_args_t *)args);
case SYS_write:
return sys_write((write_args_t *)args);
case SYS_dup:
return sys_dup((int)args);
case SYS_dup2:
return sys_dup2((dup2_args_t *)args);
case SYS_mkdir:
return sys_mkdir((mkdir_args_t *)args);
case SYS_rmdir:
return sys_rmdir((argstr_t *)args);
case SYS_unlink:
return sys_unlink((argstr_t *)args);
case SYS_link:
return sys_link((link_args_t *)args);
case SYS_rename:
return sys_rename((rename_args_t *)args);
case SYS_chdir:
return sys_chdir((argstr_t *)args);
case SYS_getdents:
return sys_getdents((getdents_args_t *)args);
case SYS_brk:
return (int) sys_brk((void *)args);
case SYS_lseek:
return sys_lseek((lseek_args_t *)args);
case SYS_halt:
sys_halt();
return -1;
case SYS_set_errno:
curthr->kt_errno = (int)args;
return 0;
case SYS_errno:
return curthr->kt_errno;
case SYS_execve:
return sys_execve((execve_args_t *)args, regs);
case SYS_stat:
return sys_stat((stat_args_t *)args);
case SYS_uname:
return sys_uname((struct utsname *)args);
case SYS_debug:
return sys_debug((argstr_t *)args);
case SYS_kshell:
return sys_kshell((int)args);
default:
dbg(DBG_ERROR, "ERROR: unknown system call: %d (args: %#08x)\n", sysnum, args);
curthr->kt_errno = ENOSYS;
return -1;
}
}
void __export_unmap(void)
{
sys_munmap(bootstrap_start, bootstrap_len);
}
void __export_unmap(void)
{
sys_munmap(bootstrap_start, bootstrap_len - vdso_rt_size);
}
static void
syscall_handler (struct intr_frame *f)
{
/* Check to see that we can read supplied user memory pointer, using the
check_mem_ptr helper() function, in get_word_from_stack(). If the
check fails, the process is terminated. */
int syscall_number = (int)get_word_on_stack(f, 0);
switch(syscall_number) {
case SYS_HALT:
{
sys_halt();
break;
}
case SYS_EXIT:
{
int status = (int)get_word_on_stack(f, 1);
sys_exit(status);
/* Returns exit status to the kernel. */
f->eax = status;
break;
}
case SYS_EXEC:
{
const char *cmd_line = (const char *)get_word_on_stack(f, 1);
pid_t pid = sys_exec(cmd_line);
/* Returns new processes pid. */
f->eax = pid;
break;
}
case SYS_WAIT:
{
pid_t pid = (pid_t)get_word_on_stack(f, 1);
/* Returns child's exit status (pid argument is pid of this child). */
f->eax = sys_wait(pid);
break;
}
case SYS_CREATE:
{
const char *filename = (const char *)get_word_on_stack(f, 1);
unsigned initial_size = (unsigned)get_word_on_stack(f, 2);
/* Returns true to the kernel if creation is successful. */
f->eax = (int)sys_create(filename, initial_size);
break;
}
case SYS_REMOVE:
{
const char *filename = (const char *)get_word_on_stack(f, 1);
/* Returns true if successful, and false otherwise. */
f->eax = sys_remove(filename);
break;
}
case SYS_OPEN:
{
const char *filename = (const char *)get_word_on_stack(f, 1);
/* Returns file descriptor of opened file, or -1 if it could not
be opened. */
f->eax = sys_open(filename);
break;
}
case SYS_FILESIZE:
{
int fd = (int)get_word_on_stack(f, 1);
/* Returns size of file in bytes. */
f->eax = sys_filesize(fd);
break;
}
case SYS_READ:
{
int fd = (int)get_word_on_stack(f, 1);
void *buffer = (void *)get_word_on_stack(f, 2);
unsigned size = (unsigned)get_word_on_stack(f, 3);
/* Returns number of bytes actually read, or -1 if it could not
be read. */
f->eax = sys_read(fd, buffer, size, f);
break;
}
case SYS_WRITE:
{
int fd = (int)get_word_on_stack(f, 1);
void *buffer = (void *)get_word_on_stack(f, 2);
unsigned size = (unsigned)get_word_on_stack(f, 3);
/* Returns number of bytes written. */
f->eax = sys_write(fd, buffer, size);
break;
}
case SYS_SEEK:
{
int fd = (int)get_word_on_stack(f, 1);
unsigned position = (int)get_word_on_stack(f, 2);
sys_seek(fd, position);
break;
}
case SYS_TELL:
{
int fd = (int)get_word_on_stack(f, 1);
/* Returns the position of the next byte to be read or written in open
file 'fd' (in bytes, from start of file). */
f->eax = sys_tell(fd);
break;
}
case SYS_CLOSE:
{
int fd = (int)get_word_on_stack(f, 1);
sys_close(fd);
break;
}
case SYS_MMAP:
{
int fd = (int)get_word_on_stack(f, 1);
void *addr = (void *)get_word_on_stack(f, 2);
f->eax = sys_mmap(fd, addr);
break;
}
case SYS_MUNMAP:
{
mapid_t mapping = (mapid_t)get_word_on_stack(f, 1);
sys_munmap(mapping);
break;
}
default:
{
NOT_REACHED();
}
}
}
int vdso_proxify(char *who, struct vdso_symtable *sym_rt,
unsigned long vdso_rt_parked_at, size_t index,
VmaEntry *vmas, size_t nr_vmas)
{
VmaEntry *vma_vdso = NULL, *vma_vvar = NULL;
struct vdso_symtable s = VDSO_SYMTABLE_INIT;
bool remap_rt = false;
/*
* Figure out which kind of vdso tuple we get.
*/
if (vma_entry_is(&vmas[index], VMA_AREA_VDSO))
vma_vdso = &vmas[index];
else if (vma_entry_is(&vmas[index], VMA_AREA_VVAR))
vma_vvar = &vmas[index];
if (index < (nr_vmas - 1)) {
if (vma_entry_is(&vmas[index + 1], VMA_AREA_VDSO))
vma_vdso = &vmas[index + 1];
else if (vma_entry_is(&vmas[index + 1], VMA_AREA_VVAR))
vma_vvar = &vmas[index + 1];
}
if (!vma_vdso) {
pr_err("Can't find vDSO area in image\n");
return -1;
}
/*
* vDSO mark overwrites Elf program header of proxy vDSO thus
* it must never ever be greater in size.
*/
BUILD_BUG_ON(sizeof(struct vdso_mark) > sizeof(Elf64_Phdr));
/*
* Find symbols in vDSO zone read from image.
*/
if (vdso_fill_symtable((void *)vma_vdso->start, vma_entry_len(vma_vdso), &s))
return -1;
/*
* Proxification strategy
*
* - There might be two vDSO zones: vdso code and optionally vvar data
* - To be able to use in-place remapping we need
*
* a) Size and order of vDSO zones are to match
* b) Symbols offsets must match
* c) Have same number of vDSO zones
*/
if (vma_entry_len(vma_vdso) == vdso_vma_size(sym_rt)) {
size_t i;
for (i = 0; i < ARRAY_SIZE(s.symbols); i++) {
if (s.symbols[i].offset != sym_rt->symbols[i].offset)
break;
}
if (i == ARRAY_SIZE(s.symbols)) {
if (vma_vvar && sym_rt->vvar_start != VVAR_BAD_ADDR) {
remap_rt = (vvar_vma_size(sym_rt) == vma_entry_len(vma_vvar));
if (remap_rt) {
long delta_rt = sym_rt->vvar_start - sym_rt->vma_start;
long delta_this = vma_vvar->start - vma_vdso->start;
remap_rt = (delta_rt ^ delta_this) < 0 ? false : true;
}
} else
remap_rt = true;
}
}
pr_debug("image [vdso] %lx-%lx [vvar] %lx-%lx\n",
vma_vdso->start, vma_vdso->end,
vma_vvar ? vma_vvar->start : VVAR_BAD_ADDR,
vma_vvar ? vma_vvar->end : VVAR_BAD_ADDR);
/*
* Easy case -- the vdso from image has same offsets, order and size
* as runtime, so we simply remap runtime vdso to dumpee position
* without generating any proxy.
*
* Note we may remap VVAR vdso as well which might not yet been mapped
* by a caller code. So drop VMA_AREA_REGULAR from it and caller would
* not touch it anymore.
*/
if (remap_rt) {
int ret = 0;
pr_info("Runtime vdso/vvar matches dumpee, remap inplace\n");
if (sys_munmap((void *)vma_vdso->start, vma_entry_len(vma_vdso))) {
pr_err("Failed to unmap %s\n", who);
return -1;
}
if (vma_vvar) {
if (sys_munmap((void *)vma_vvar->start, vma_entry_len(vma_vvar))) {
pr_err("Failed to unmap %s\n", who);
return -1;
}
if (vma_vdso->start < vma_vvar->start) {
ret = vdso_remap(who, vdso_rt_parked_at, vma_vdso->start, vdso_vma_size(sym_rt));
vdso_rt_parked_at += vdso_vma_size(sym_rt);
ret |= vdso_remap(who, vdso_rt_parked_at, vma_vvar->start, vvar_vma_size(sym_rt));
} else {
ret = vdso_remap(who, vdso_rt_parked_at, vma_vvar->start, vvar_vma_size(sym_rt));
vdso_rt_parked_at += vvar_vma_size(sym_rt);
ret |= vdso_remap(who, vdso_rt_parked_at, vma_vdso->start, vdso_vma_size(sym_rt));
}
} else
ret = vdso_remap(who, vdso_rt_parked_at, vma_vdso->start, vdso_vma_size(sym_rt));
return ret;
}
/*
* Now complex case -- we need to proxify calls. We redirect
* calls from dumpee vdso to runtime vdso, making dumpee
* to operate as proxy vdso.
*/
pr_info("Runtime vdso mismatches dumpee, generate proxy\n");
/*
* Don't forget to shift if vvar is before vdso.
*/
if (sym_rt->vvar_start != VDSO_BAD_ADDR &&
sym_rt->vvar_start < sym_rt->vma_start)
vdso_rt_parked_at += vvar_vma_size(sym_rt);
if (vdso_redirect_calls(vdso_rt_parked_at,
vma_vdso->start,
sym_rt, &s)) {
pr_err("Failed to proxify dumpee contents\n");
return -1;
}
/*
* Put a special mark into runtime vdso, thus at next checkpoint
* routine we could detect this vdso and do not dump it, since
* it's auto-generated every new session if proxy required.
*/
sys_mprotect((void *)vdso_rt_parked_at, vdso_vma_size(sym_rt), PROT_WRITE);
vdso_put_mark((void *)vdso_rt_parked_at, vma_vdso->start, vma_vvar ? vma_vvar->start : VVAR_BAD_ADDR);
sys_mprotect((void *)vdso_rt_parked_at, vdso_vma_size(sym_rt), VDSO_PROT);
return 0;
}
int
cloudabi_sys_mem_advise(struct thread *td,
struct cloudabi_sys_mem_advise_args *uap)
{
struct madvise_args madvise_args = {
.addr = uap->addr,
.len = uap->len
};
switch (uap->advice) {
case CLOUDABI_ADVICE_DONTNEED:
madvise_args.behav = MADV_DONTNEED;
break;
case CLOUDABI_ADVICE_NORMAL:
madvise_args.behav = MADV_NORMAL;
break;
case CLOUDABI_ADVICE_RANDOM:
madvise_args.behav = MADV_RANDOM;
break;
case CLOUDABI_ADVICE_SEQUENTIAL:
madvise_args.behav = MADV_SEQUENTIAL;
break;
case CLOUDABI_ADVICE_WILLNEED:
madvise_args.behav = MADV_WILLNEED;
break;
default:
return (EINVAL);
}
return (sys_madvise(td, &madvise_args));
}
int
cloudabi_sys_mem_lock(struct thread *td, struct cloudabi_sys_mem_lock_args *uap)
{
struct mlock_args mlock_args = {
.addr = uap->addr,
.len = uap->len
};
return (sys_mlock(td, &mlock_args));
}
int
cloudabi_sys_mem_map(struct thread *td, struct cloudabi_sys_mem_map_args *uap)
{
struct mmap_args mmap_args = {
.addr = uap->addr,
.len = uap->len,
.fd = uap->fd,
.pos = uap->off
};
int error;
/* Translate flags. */
if (uap->flags & CLOUDABI_MAP_ANON)
mmap_args.flags |= MAP_ANON;
if (uap->flags & CLOUDABI_MAP_FIXED)
mmap_args.flags |= MAP_FIXED;
if (uap->flags & CLOUDABI_MAP_PRIVATE)
mmap_args.flags |= MAP_PRIVATE;
if (uap->flags & CLOUDABI_MAP_SHARED)
mmap_args.flags |= MAP_SHARED;
/* Translate protection. */
error = convert_mprot(uap->prot, &mmap_args.prot);
if (error != 0)
return (error);
return (sys_mmap(td, &mmap_args));
}
int
cloudabi_sys_mem_protect(struct thread *td,
struct cloudabi_sys_mem_protect_args *uap)
{
struct mprotect_args mprotect_args = {
.addr = uap->addr,
.len = uap->len,
};
int error;
/* Translate protection. */
error = convert_mprot(uap->prot, &mprotect_args.prot);
if (error != 0)
return (error);
return (sys_mprotect(td, &mprotect_args));
}
int
cloudabi_sys_mem_sync(struct thread *td, struct cloudabi_sys_mem_sync_args *uap)
{
struct msync_args msync_args = {
.addr = uap->addr,
.len = uap->len,
};
/* Convert flags. */
switch (uap->flags & (CLOUDABI_MS_ASYNC | CLOUDABI_MS_SYNC)) {
case CLOUDABI_MS_ASYNC:
msync_args.flags |= MS_ASYNC;
break;
case CLOUDABI_MS_SYNC:
msync_args.flags |= MS_SYNC;
break;
default:
return (EINVAL);
}
if ((uap->flags & CLOUDABI_MS_INVALIDATE) != 0)
msync_args.flags |= MS_INVALIDATE;
return (sys_msync(td, &msync_args));
}
int
cloudabi_sys_mem_unlock(struct thread *td,
struct cloudabi_sys_mem_unlock_args *uap)
{
struct munlock_args munlock_args = {
.addr = uap->addr,
.len = uap->len
};
return (sys_munlock(td, &munlock_args));
}
int
cloudabi_sys_mem_unmap(struct thread *td,
struct cloudabi_sys_mem_unmap_args *uap)
{
struct munmap_args munmap_args = {
.addr = uap->addr,
.len = uap->len
};
return (sys_munmap(td, &munmap_args));
}
