int
vslock(
user_addr_t addr,
user_size_t len)
{
kern_return_t kret;
vm_map_t map;
map = current_map();
kret = vm_map_wire(map,
vm_map_trunc_page(addr,
vm_map_page_mask(map)),
vm_map_round_page(addr+len,
vm_map_page_mask(map)),
VM_PROT_READ | VM_PROT_WRITE,
FALSE);
switch (kret) {
case KERN_SUCCESS:
return (0);
case KERN_INVALID_ADDRESS:
case KERN_NO_SPACE:
return (ENOMEM);
case KERN_PROTECTION_FAILURE:
return (EACCES);
default:
return (EINVAL);
}
}
int
minherit(__unused proc_t p, struct minherit_args *uap, __unused register_t *retval)
{
mach_vm_offset_t addr;
mach_vm_size_t size;
register vm_inherit_t inherit;
vm_map_t user_map;
kern_return_t result;
AUDIT_ARG(addr, uap->addr);
AUDIT_ARG(len, uap->len);
AUDIT_ARG(value, uap->inherit);
addr = (mach_vm_offset_t)uap->addr;
size = (mach_vm_size_t)uap->len;
inherit = uap->inherit;
user_map = current_map();
result = mach_vm_inherit(user_map, addr, size,
inherit);
switch (result) {
case KERN_SUCCESS:
return (0);
case KERN_PROTECTION_FAILURE:
return (EACCES);
}
return (EINVAL);
}
int
munmap(__unused proc_t p, struct munmap_args *uap, __unused register_t *retval)
{
mach_vm_offset_t user_addr;
mach_vm_size_t user_size;
kern_return_t result;
user_addr = (mach_vm_offset_t) uap->addr;
user_size = (mach_vm_size_t) uap->len;
AUDIT_ARG(addr, user_addr);
AUDIT_ARG(len, user_size);
if (user_addr & PAGE_MASK_64) {
/* UNIX SPEC: user address is not page-aligned, return EINVAL */
return EINVAL;
}
if (user_addr + user_size < user_addr)
return(EINVAL);
if (user_size == 0) {
/* UNIX SPEC: size is 0, return EINVAL */
return EINVAL;
}
result = mach_vm_deallocate(current_map(), user_addr, user_size);
if (result != KERN_SUCCESS) {
return(EINVAL);
}
return(0);
}
int
useracc(
user_addr_t addr,
user_size_t len,
int prot)
{
return (vm_map_check_protection(
current_map(),
vm_map_trunc_page(addr), vm_map_round_page(addr+len),
prot == B_READ ? VM_PROT_READ : VM_PROT_WRITE));
}
bool ShortLoopOptimizer::process(BlockBegin* loop_header) {
TRACE_VALUE_NUMBERING(tty->print_cr("** loop header block"));
_too_complicated_loop = false;
_loop_blocks.clear();
_loop_blocks.append(loop_header);
for (int i = 0; i < _loop_blocks.length(); i++) {
BlockBegin* block = _loop_blocks.at(i);
TRACE_VALUE_NUMBERING(tty->print_cr("processing loop block B%d", block->block_id()));
if (block->is_set(BlockBegin::exception_entry_flag)) {
// this would be too complicated
return false;
}
// add predecessors to worklist
for (int j = block->number_of_preds() - 1; j >= 0; j--) {
BlockBegin* pred = block->pred_at(j);
if (pred->is_set(BlockBegin::osr_entry_flag)) {
return false;
}
ValueMap* pred_map = value_map_of(pred);
if (pred_map != NULL) {
current_map()->kill_map(pred_map);
} else if (!_loop_blocks.contains(pred)) {
if (_loop_blocks.length() >= ValueMapMaxLoopSize) {
return false;
}
_loop_blocks.append(pred);
}
}
// use the instruction visitor for killing values
for (Value instr = block->next(); instr != NULL; instr = instr->next()) {
instr->visit(this);
if (_too_complicated_loop) {
return false;
}
}
}
bool optimistic = this->_gvn->compilation()->is_optimistic();
if (UseLoopInvariantCodeMotion && optimistic) {
LoopInvariantCodeMotion code_motion(this, _gvn, loop_header, &_loop_blocks);
}
TRACE_VALUE_NUMBERING(tty->print_cr("** loop successfully optimized"));
return true;
}
void main()
{
unsigned int init_length, x, y, seed;
unsigned long generation = 0;
char gen_text[80];
long bios_time, start_bios_time;
cellmap current_map(cellmap_height, cellmap_width);
cellmap next_map(cellmap_height, cellmap_width);
// Get the seed; seed randomly if 0 entered
cout << "Seed (0 for random seed): ";
cin >> seed;
if (seed == 0) seed = (unsigned) time(NULL);
// Randomly initialize the initial cell map
cout << "Initializing...";
srand(seed);
init_length = (cellmap_height * cellmap_width) / 2;
do {
x = random(cellmap_width);
y = random(cellmap_height);
next_map.set_cell(x, y);
} while (--init_length);
current_map.copy_cells(next_map); // put init map in current_map
enter_display_mode();
// Keep recalculating and redisplaying generations until a key
// is pressed
show_text(0, MSG_LINE, "Generation: ");
start_bios_time = _bios_timeofday(_TIME_GETCLOCK, &bios_time);
do {
generation++;
sprintf(gen_text, "%10lu", generation);
show_text(1, GENERATION_LINE, gen_text);
// Recalculate and draw the next generation
current_map.next_generation(next_map);
// Make current_map current again
current_map.copy_cells(next_map);
#if LIMIT_18_HZ
// Limit to a maximum of 18.2 frames per second, for visibility
do {
_bios_timeofday(_TIME_GETCLOCK, &bios_time);
} while (start_bios_time == bios_time);
start_bios_time = bios_time;
#endif
} while (!kbhit());
getch(); // clear keypress
exit_display_mode();
cout << "Total generations: " << generation << "\nSeed: " <<
seed << "\n";
}
int
vsunlock(
user_addr_t addr,
user_size_t len,
__unused int dirtied)
{
#if FIXME /* [ */
pmap_t pmap;
vm_page_t pg;
vm_map_offset_t vaddr;
ppnum_t paddr;
#endif /* FIXME ] */
kern_return_t kret;
vm_map_t map;
map = current_map();
#if FIXME /* [ */
if (dirtied) {
pmap = get_task_pmap(current_task());
for (vaddr = vm_map_trunc_page(addr, PAGE_MASK);
vaddr < vm_map_round_page(addr+len, PAGE_MASK);
vaddr += PAGE_SIZE) {
paddr = pmap_extract(pmap, vaddr);
pg = PHYS_TO_VM_PAGE(paddr);
vm_page_set_modified(pg);
}
}
#endif /* FIXME ] */
#ifdef lint
dirtied++;
#endif /* lint */
kret = vm_map_unwire(map,
vm_map_trunc_page(addr,
vm_map_page_mask(map)),
vm_map_round_page(addr+len,
vm_map_page_mask(map)),
FALSE);
switch (kret) {
case KERN_SUCCESS:
return (0);
case KERN_INVALID_ADDRESS:
case KERN_NO_SPACE:
return (ENOMEM);
case KERN_PROTECTION_FAILURE:
return (EACCES);
default:
return (EINVAL);
}
}
int
madvise(__unused proc_t p, struct madvise_args *uap, __unused register_t *retval)
{
vm_map_t user_map;
mach_vm_offset_t start;
mach_vm_size_t size;
vm_behavior_t new_behavior;
kern_return_t result;
/*
* Since this routine is only advisory, we default to conservative
* behavior.
*/
switch (uap->behav) {
case MADV_RANDOM:
new_behavior = VM_BEHAVIOR_RANDOM;
break;
case MADV_SEQUENTIAL:
new_behavior = VM_BEHAVIOR_SEQUENTIAL;
break;
case MADV_NORMAL:
new_behavior = VM_BEHAVIOR_DEFAULT;
break;
case MADV_WILLNEED:
new_behavior = VM_BEHAVIOR_WILLNEED;
break;
case MADV_DONTNEED:
new_behavior = VM_BEHAVIOR_DONTNEED;
break;
default:
return(EINVAL);
}
start = (mach_vm_offset_t) uap->addr;
size = (mach_vm_size_t) uap->len;
user_map = current_map();
result = mach_vm_behavior_set(user_map, start, size, new_behavior);
switch (result) {
case KERN_SUCCESS:
return (0);
case KERN_INVALID_ADDRESS:
return (ENOMEM);
}
return (EINVAL);
}
int
useracc(
user_addr_t addr,
user_size_t len,
int prot)
{
vm_map_t map;
map = current_map();
return (vm_map_check_protection(
map,
vm_map_trunc_page(addr,
vm_map_page_mask(map)),
vm_map_round_page(addr+len,
vm_map_page_mask(map)),
prot == B_READ ? VM_PROT_READ : VM_PROT_WRITE));
}
int copyoutmap(
vm_map_t map,
char *fromaddr,
char *toaddr,
int length)
{
if (vm_map_pmap(map) == kernel_pmap) {
/* assume a correct copy */
memcpy(toaddr, fromaddr, length);
return 0;
}
if (current_map() == map)
return copyout(fromaddr, toaddr, length);
return 1;
}
int
munlock(__unused proc_t p, struct munlock_args *uap, __unused register_t *retval)
{
mach_vm_offset_t addr;
mach_vm_size_t size;
vm_map_t user_map;
kern_return_t result;
AUDIT_ARG(addr, uap->addr);
AUDIT_ARG(addr, uap->len);
addr = (mach_vm_offset_t) uap->addr;
size = (mach_vm_size_t)uap->len;
user_map = current_map();
/* JMM - need to remove all wirings by spec - this just removes one */
result = mach_vm_wire(host_priv_self(), user_map, addr, size, VM_PROT_NONE);
return (result == KERN_SUCCESS ? 0 : ENOMEM);
}
int
mlock(__unused proc_t p, struct mlock_args *uap, __unused int32_t *retvalval)
{
vm_map_t user_map;
vm_map_offset_t addr;
vm_map_size_t size, pageoff;
kern_return_t result;
AUDIT_ARG(addr, uap->addr);
AUDIT_ARG(len, uap->len);
addr = (vm_map_offset_t) uap->addr;
size = (vm_map_size_t)uap->len;
/* disable wrap around */
if (addr + size < addr)
return (EINVAL);
if (size == 0)
return (0);
user_map = current_map();
pageoff = (addr & vm_map_page_mask(user_map));
addr -= pageoff;
size = vm_map_round_page(size+pageoff, vm_map_page_mask(user_map));
/* have to call vm_map_wire directly to pass "I don't know" protections */
result = vm_map_wire(user_map, addr, addr+size, VM_PROT_NONE | VM_PROT_MEMORY_TAG_MAKE(VM_KERN_MEMORY_MLOCK), TRUE);
if (result == KERN_RESOURCE_SHORTAGE)
return EAGAIN;
else if (result == KERN_PROTECTION_FAILURE)
return EACCES;
else if (result != KERN_SUCCESS)
return ENOMEM;
return 0; /* KERN_SUCCESS */
}
mach_msg_return_t
mach_msg_receive(
mach_msg_header_t *msg,
mach_msg_option_t option,
mach_msg_size_t rcv_size,
mach_port_t rcv_name,
mach_msg_timeout_t time_out,
mach_port_t notify)
{
ipc_thread_t self = current_thread();
ipc_space_t space = current_space();
vm_map_t map = current_map();
ipc_object_t object;
ipc_mqueue_t mqueue;
ipc_kmsg_t kmsg;
mach_port_seqno_t seqno;
mach_msg_return_t mr;
mr = ipc_mqueue_copyin(space, rcv_name, &mqueue, &object);
if (mr != MACH_MSG_SUCCESS)
return mr;
/* hold ref for object; mqueue is locked */
/*
* ipc_mqueue_receive may not return, because if we block
* then our kernel stack may be discarded. So we save
* state here for mach_msg_receive_continue to pick up.
*/
self->ith_msg = msg;
self->ith_option = option;
self->ith_rcv_size = rcv_size;
self->ith_timeout = time_out;
self->ith_notify = notify;
self->ith_object = object;
self->ith_mqueue = mqueue;
if (option & MACH_RCV_LARGE) {
mr = ipc_mqueue_receive(mqueue, option & MACH_RCV_TIMEOUT,
rcv_size, time_out,
FALSE, mach_msg_receive_continue,
&kmsg, &seqno);
/* mqueue is unlocked */
ipc_object_release(object);
if (mr != MACH_MSG_SUCCESS) {
if (mr == MACH_RCV_TOO_LARGE) {
mach_msg_size_t real_size =
(mach_msg_size_t) (vm_offset_t) kmsg;
assert(real_size > rcv_size);
(void) copyout(&real_size,
&msg->msgh_size,
sizeof(mach_msg_size_t));
}
return mr;
}
kmsg->ikm_header.msgh_seqno = seqno;
assert(kmsg->ikm_header.msgh_size <= rcv_size);
} else {
mr = ipc_mqueue_receive(mqueue, option & MACH_RCV_TIMEOUT,
MACH_MSG_SIZE_MAX, time_out,
FALSE, mach_msg_receive_continue,
&kmsg, &seqno);
/* mqueue is unlocked */
ipc_object_release(object);
if (mr != MACH_MSG_SUCCESS)
return mr;
kmsg->ikm_header.msgh_seqno = seqno;
if (kmsg->ikm_header.msgh_size > rcv_size) {
ipc_kmsg_copyout_dest(kmsg, space);
(void) ipc_kmsg_put(msg, kmsg, sizeof *msg);
return MACH_RCV_TOO_LARGE;
}
}
if (option & MACH_RCV_NOTIFY) {
if (notify == MACH_PORT_NULL)
mr = MACH_RCV_INVALID_NOTIFY;
else
mr = ipc_kmsg_copyout(kmsg, space, map, notify);
} else
mr = ipc_kmsg_copyout(kmsg, space, map, MACH_PORT_NULL);
if (mr != MACH_MSG_SUCCESS) {
if ((mr &~ MACH_MSG_MASK) == MACH_RCV_BODY_ERROR) {
(void) ipc_kmsg_put(msg, kmsg,
kmsg->ikm_header.msgh_size);
} else {
ipc_kmsg_copyout_dest(kmsg, space);
(void) ipc_kmsg_put(msg, kmsg, sizeof *msg);
}
return mr;
}
return ipc_kmsg_put(msg, kmsg, kmsg->ikm_header.msgh_size);
}
void kill_field(ciField* field) {
current_map()->kill_field(field);
}
void kill_array(ValueType* type) {
current_map()->kill_array(type);
}
uap->behav == MADV_FREE_REUSABLE)) {
printf("** FOURK_COMPAT: %d[%s] "
"failing madvise(0x%llx,0x%llx,%s)\n",
p->p_pid, p->p_comm, start, size,
((uap->behav == MADV_FREE_REUSABLE)
? "MADV_FREE_REUSABLE"
: "MADV_FREE"));
DTRACE_VM3(fourk_compat_madvise,
uint64_t, start,
uint64_t, size,
int, uap->behav);
return EINVAL;
}
#endif /* __arm64__ */
user_map = current_map();
result = mach_vm_behavior_set(user_map, start, size, new_behavior);
switch (result) {
case KERN_SUCCESS:
return 0;
case KERN_INVALID_ADDRESS:
return EINVAL;
case KERN_NO_SPACE:
return ENOMEM;
}
return EINVAL;
}
int
// implementation for abstract methods of ValueNumberingVisitor
void kill_memory() {
current_map()->kill_memory();
}
void kill_field(ciField* field, bool all_offsets) {
current_map()->kill_field(field, all_offsets);
assert(field->type()->basic_type() >= 0 && field->type()->basic_type() <= T_ARRAY, "Invalid type");
_has_field_store[field->type()->basic_type()] = true;
}
int
mprotect(__unused proc_t p, struct mprotect_args *uap, __unused int32_t *retval)
{
vm_prot_t prot;
mach_vm_offset_t user_addr;
mach_vm_size_t user_size;
kern_return_t result;
vm_map_t user_map;
#if CONFIG_MACF
int error;
#endif
AUDIT_ARG(addr, uap->addr);
AUDIT_ARG(len, uap->len);
AUDIT_ARG(value32, uap->prot);
user_map = current_map();
user_addr = (mach_vm_offset_t) uap->addr;
user_size = (mach_vm_size_t) uap->len;
prot = (vm_prot_t)(uap->prot & (VM_PROT_ALL | VM_PROT_TRUSTED | VM_PROT_STRIP_READ));
if (user_addr & vm_map_page_mask(user_map)) {
/* UNIX SPEC: user address is not page-aligned, return EINVAL */
return EINVAL;
}
#ifdef notyet
/* Hmm .. */
#if defined(VM_PROT_READ_IS_EXEC)
if (prot & VM_PROT_READ)
prot |= VM_PROT_EXECUTE;
#endif
#endif /* notyet */
#if 3936456
if (prot & (VM_PROT_EXECUTE | VM_PROT_WRITE))
prot |= VM_PROT_READ;
#endif /* 3936456 */
#if defined(__arm64__)
if (prot & VM_PROT_STRIP_READ)
prot &= ~(VM_PROT_READ | VM_PROT_STRIP_READ);
#endif
#if CONFIG_MACF
/*
* The MAC check for mprotect is of limited use for 2 reasons:
* Without mmap revocation, the caller could have asked for the max
* protections initially instead of a reduced set, so a mprotect
* check would offer no new security.
* It is not possible to extract the vnode from the pager object(s)
* of the target memory range.
* However, the MAC check may be used to prevent a process from,
* e.g., making the stack executable.
*/
error = mac_proc_check_mprotect(p, user_addr,
user_size, prot);
if (error)
return (error);
#endif
if(prot & VM_PROT_TRUSTED) {
#if CONFIG_DYNAMIC_CODE_SIGNING
/* CODE SIGNING ENFORCEMENT - JIT support */
/* The special protection value VM_PROT_TRUSTED requests that we treat
* this page as if it had a valid code signature.
* If this is enabled, there MUST be a MAC policy implementing the
* mac_proc_check_mprotect() hook above. Otherwise, Codesigning will be
* compromised because the check would always succeed and thusly any
* process could sign dynamically. */
result = vm_map_sign(
user_map,
vm_map_trunc_page(user_addr,
vm_map_page_mask(user_map)),
vm_map_round_page(user_addr+user_size,
vm_map_page_mask(user_map)));
switch (result) {
case KERN_SUCCESS:
break;
case KERN_INVALID_ADDRESS:
/* UNIX SPEC: for an invalid address range, return ENOMEM */
return ENOMEM;
default:
return EINVAL;
}
#else
return ENOTSUP;
#endif
}
prot &= ~VM_PROT_TRUSTED;
result = mach_vm_protect(user_map, user_addr, user_size,
FALSE, prot);
switch (result) {
case KERN_SUCCESS:
return (0);
case KERN_PROTECTION_FAILURE:
return (EACCES);
case KERN_INVALID_ADDRESS:
/* UNIX SPEC: for an invalid address range, return ENOMEM */
return ENOMEM;
}
return (EINVAL);
}
mach_msg_return_t
mach_msg_overwrite(
mach_msg_header_t *msg,
mach_msg_option_t option,
mach_msg_size_t send_size,
mach_msg_size_t rcv_size,
mach_port_name_t rcv_name,
__unused mach_msg_timeout_t msg_timeout,
__unused mach_port_name_t notify,
__unused mach_msg_header_t *rcv_msg,
__unused mach_msg_size_t rcv_msg_size)
{
ipc_space_t space = current_space();
vm_map_t map = current_map();
ipc_kmsg_t kmsg;
mach_port_seqno_t seqno;
mach_msg_return_t mr;
mach_msg_trailer_size_t trailer_size;
if (option & MACH_SEND_MSG) {
mach_msg_size_t msg_and_trailer_size;
mach_msg_max_trailer_t *max_trailer;
if ((send_size < sizeof(mach_msg_header_t)) || (send_size & 3))
return MACH_SEND_MSG_TOO_SMALL;
if (send_size > MACH_MSG_SIZE_MAX - MAX_TRAILER_SIZE)
return MACH_SEND_TOO_LARGE;
msg_and_trailer_size = send_size + MAX_TRAILER_SIZE;
kmsg = ipc_kmsg_alloc(msg_and_trailer_size);
if (kmsg == IKM_NULL)
return MACH_SEND_NO_BUFFER;
(void) memcpy((void *) kmsg->ikm_header, (const void *) msg, send_size);
kmsg->ikm_header->msgh_size = send_size;
/*
* Reserve for the trailer the largest space (MAX_TRAILER_SIZE)
* However, the internal size field of the trailer (msgh_trailer_size)
* is initialized to the minimum (sizeof(mach_msg_trailer_t)), to optimize
* the cases where no implicit data is requested.
*/
max_trailer = (mach_msg_max_trailer_t *) ((vm_offset_t)kmsg->ikm_header + send_size);
max_trailer->msgh_sender = current_thread()->task->sec_token;
max_trailer->msgh_audit = current_thread()->task->audit_token;
max_trailer->msgh_trailer_type = MACH_MSG_TRAILER_FORMAT_0;
max_trailer->msgh_trailer_size = MACH_MSG_TRAILER_MINIMUM_SIZE;
mr = ipc_kmsg_copyin(kmsg, space, map, &option);
if (mr != MACH_MSG_SUCCESS) {
ipc_kmsg_free(kmsg);
return mr;
}
do {
mr = ipc_kmsg_send(kmsg, MACH_MSG_OPTION_NONE, MACH_MSG_TIMEOUT_NONE);
} while (mr == MACH_SEND_INTERRUPTED);
assert(mr == MACH_MSG_SUCCESS);
}
if (option & MACH_RCV_MSG) {
thread_t self = current_thread();
do {
ipc_object_t object;
ipc_mqueue_t mqueue;
mr = ipc_mqueue_copyin(space, rcv_name,
&mqueue, &object);
if (mr != MACH_MSG_SUCCESS)
return mr;
/* hold ref for object */
self->ith_continuation = (void (*)(mach_msg_return_t))0;
ipc_mqueue_receive(mqueue,
MACH_MSG_OPTION_NONE,
MACH_MSG_SIZE_MAX,
MACH_MSG_TIMEOUT_NONE,
THREAD_ABORTSAFE);
mr = self->ith_state;
kmsg = self->ith_kmsg;
seqno = self->ith_seqno;
io_release(object);
} while (mr == MACH_RCV_INTERRUPTED);
if (mr != MACH_MSG_SUCCESS)
return mr;
trailer_size = ipc_kmsg_add_trailer(kmsg, space, option, current_thread(), seqno, TRUE,
kmsg->ikm_header->msgh_remote_port->ip_context);
if (rcv_size < (kmsg->ikm_header->msgh_size + trailer_size)) {
ipc_kmsg_copyout_dest(kmsg, space);
(void) memcpy((void *) msg, (const void *) kmsg->ikm_header, sizeof *msg);
ipc_kmsg_free(kmsg);
return MACH_RCV_TOO_LARGE;
}
mr = ipc_kmsg_copyout(kmsg, space, map, MACH_MSG_BODY_NULL, option);
if (mr != MACH_MSG_SUCCESS) {
if ((mr &~ MACH_MSG_MASK) == MACH_RCV_BODY_ERROR) {
ipc_kmsg_put_to_kernel(msg, kmsg,
kmsg->ikm_header->msgh_size + trailer_size);
} else {
ipc_kmsg_copyout_dest(kmsg, space);
(void) memcpy((void *) msg, (const void *) kmsg->ikm_header, sizeof *msg);
ipc_kmsg_free(kmsg);
}
return mr;
}
(void) memcpy((void *) msg, (const void *) kmsg->ikm_header,
kmsg->ikm_header->msgh_size + trailer_size);
ipc_kmsg_free(kmsg);
}
return MACH_MSG_SUCCESS;
}
static vm_map_t
_current_map(void)
{
return current_map();
}
void kill_array(ValueType* type) {
current_map()->kill_array(type);
BasicType basic_type = as_BasicType(type); assert(basic_type >= 0 && basic_type <= T_ARRAY, "Invalid type");
_has_indexed_store[basic_type] = true;
}
int
mincore(__unused proc_t p, struct mincore_args *uap, __unused register_t *retval)
{
mach_vm_offset_t addr, first_addr, end;
vm_map_t map;
user_addr_t vec;
int error;
int vecindex, lastvecindex;
int mincoreinfo=0;
int pqueryinfo;
kern_return_t ret;
int numref;
char c;
map = current_map();
/*
* Make sure that the addresses presented are valid for user
* mode.
*/
first_addr = addr = mach_vm_trunc_page(uap->addr);
end = addr + mach_vm_round_page(uap->len);
if (end < addr)
return (EINVAL);
/*
* Address of byte vector
*/
vec = uap->vec;
map = current_map();
/*
* Do this on a map entry basis so that if the pages are not
* in the current processes address space, we can easily look
* up the pages elsewhere.
*/
lastvecindex = -1;
for( ; addr < end; addr += PAGE_SIZE ) {
pqueryinfo = 0;
ret = mach_vm_page_query(map, addr, &pqueryinfo, &numref);
if (ret != KERN_SUCCESS)
pqueryinfo = 0;
mincoreinfo = 0;
if (pqueryinfo & VM_PAGE_QUERY_PAGE_PRESENT)
mincoreinfo |= MINCORE_INCORE;
if (pqueryinfo & VM_PAGE_QUERY_PAGE_REF)
mincoreinfo |= MINCORE_REFERENCED;
if (pqueryinfo & VM_PAGE_QUERY_PAGE_DIRTY)
mincoreinfo |= MINCORE_MODIFIED;
/*
* calculate index into user supplied byte vector
*/
vecindex = (addr - first_addr)>> PAGE_SHIFT;
/*
* If we have skipped map entries, we need to make sure that
* the byte vector is zeroed for those skipped entries.
*/
while((lastvecindex + 1) < vecindex) {
c = 0;
error = copyout(&c, vec + lastvecindex, 1);
if (error) {
return (EFAULT);
}
++lastvecindex;
}
/*
* Pass the page information to the user
*/
c = (char)mincoreinfo;
error = copyout(&c, vec + vecindex, 1);
if (error) {
return (EFAULT);
}
lastvecindex = vecindex;
}
/*
* Zero the last entries in the byte vector.
*/
vecindex = (end - first_addr) >> PAGE_SHIFT;
while((lastvecindex + 1) < vecindex) {
c = 0;
error = copyout(&c, vec + lastvecindex, 1);
if (error) {
return (EFAULT);
}
++lastvecindex;
}
return (0);
}
*/
mach_msg_return_t
mach_msg_overwrite(
mach_msg_header_t *msg,
mach_msg_option_t option,
mach_msg_size_t send_size,
mach_msg_size_t rcv_size,
mach_port_name_t rcv_name,
__unused mach_msg_timeout_t msg_timeout,
mach_msg_priority_t override,
__unused mach_msg_header_t *rcv_msg,
__unused mach_msg_size_t rcv_msg_size)
{
ipc_space_t space = current_space();
vm_map_t map = current_map();
ipc_kmsg_t kmsg;
mach_port_seqno_t seqno;
mach_msg_return_t mr;
mach_msg_trailer_size_t trailer_size;
if (option & MACH_SEND_MSG) {
mach_msg_size_t msg_and_trailer_size;
mach_msg_max_trailer_t *max_trailer;
if ((send_size & 3) ||
send_size < sizeof(mach_msg_header_t) ||
(send_size < sizeof(mach_msg_body_t) && (msg->msgh_bits & MACH_MSGH_BITS_COMPLEX)))
return MACH_SEND_MSG_TOO_SMALL;
if (send_size > MACH_MSG_SIZE_MAX - MAX_TRAILER_SIZE)
kern_return_t
map_fd_funneled(
int fd,
vm_object_offset_t offset,
vm_offset_t *va,
boolean_t findspace,
vm_size_t size)
{
kern_return_t result;
struct fileproc *fp;
struct vnode *vp;
void * pager;
vm_offset_t map_addr=0;
vm_size_t map_size;
int err=0;
vm_map_t my_map;
proc_t p = current_proc();
struct vnode_attr vattr;
/*
* Find the inode; verify that it's a regular file.
*/
err = fp_lookup(p, fd, &fp, 0);
if (err)
return(err);
if (fp->f_fglob->fg_type != DTYPE_VNODE){
err = KERN_INVALID_ARGUMENT;
goto bad;
}
if (!(fp->f_fglob->fg_flag & FREAD)) {
err = KERN_PROTECTION_FAILURE;
goto bad;
}
vp = (struct vnode *)fp->f_fglob->fg_data;
err = vnode_getwithref(vp);
if(err != 0)
goto bad;
if (vp->v_type != VREG) {
(void)vnode_put(vp);
err = KERN_INVALID_ARGUMENT;
goto bad;
}
AUDIT_ARG(vnpath, vp, ARG_VNODE1);
/*
* POSIX: mmap needs to update access time for mapped files
*/
if ((vnode_vfsvisflags(vp) & MNT_NOATIME) == 0) {
VATTR_INIT(&vattr);
nanotime(&vattr.va_access_time);
VATTR_SET_ACTIVE(&vattr, va_access_time);
vnode_setattr(vp, &vattr, vfs_context_current());
}
if (offset & PAGE_MASK_64) {
printf("map_fd: file offset not page aligned(%d : %s)\n",p->p_pid, p->p_comm);
(void)vnode_put(vp);
err = KERN_INVALID_ARGUMENT;
goto bad;
}
map_size = round_page(size);
/*
* Allow user to map in a zero length file.
*/
if (size == 0) {
(void)vnode_put(vp);
err = KERN_SUCCESS;
goto bad;
}
/*
* Map in the file.
*/
pager = (void *)ubc_getpager(vp);
if (pager == NULL) {
(void)vnode_put(vp);
err = KERN_FAILURE;
goto bad;
}
my_map = current_map();
result = vm_map_64(
my_map,
&map_addr, map_size, (vm_offset_t)0,
VM_FLAGS_ANYWHERE, pager, offset, TRUE,
VM_PROT_DEFAULT, VM_PROT_ALL,
VM_INHERIT_DEFAULT);
if (result != KERN_SUCCESS) {
(void)vnode_put(vp);
err = result;
goto bad;
}
if (!findspace) {
vm_offset_t dst_addr;
vm_map_copy_t tmp;
if (copyin(CAST_USER_ADDR_T(va), &dst_addr, sizeof (dst_addr)) ||
trunc_page_32(dst_addr) != dst_addr) {
(void) vm_map_remove(
my_map,
map_addr, map_addr + map_size,
VM_MAP_NO_FLAGS);
(void)vnode_put(vp);
err = KERN_INVALID_ADDRESS;
goto bad;
}
result = vm_map_copyin(my_map, (vm_map_address_t)map_addr,
(vm_map_size_t)map_size, TRUE, &tmp);
if (result != KERN_SUCCESS) {
(void) vm_map_remove(my_map, vm_map_trunc_page(map_addr),
vm_map_round_page(map_addr + map_size),
VM_MAP_NO_FLAGS);
(void)vnode_put(vp);
err = result;
goto bad;
}
result = vm_map_copy_overwrite(my_map,
(vm_map_address_t)dst_addr, tmp, FALSE);
if (result != KERN_SUCCESS) {
vm_map_copy_discard(tmp);
(void)vnode_put(vp);
err = result;
goto bad;
}
} else {
if (copyout(&map_addr, CAST_USER_ADDR_T(va), sizeof (map_addr))) {
(void) vm_map_remove(my_map, vm_map_trunc_page(map_addr),
vm_map_round_page(map_addr + map_size),
VM_MAP_NO_FLAGS);
(void)vnode_put(vp);
err = KERN_INVALID_ADDRESS;
goto bad;
}
}
ubc_setthreadcred(vp, current_proc(), current_thread());
(void)ubc_map(vp, (PROT_READ | PROT_EXEC));
(void)vnode_put(vp);
err = 0;
bad:
fp_drop(p, fd, fp, 0);
return (err);
}
void
mach_msg_receive_continue(void)
{
ipc_thread_t self = current_thread();
ipc_space_t space = current_space();
vm_map_t map = current_map();
mach_msg_header_t *msg = self->ith_msg;
mach_msg_option_t option = self->ith_option;
mach_msg_size_t rcv_size = self->ith_rcv_size;
mach_msg_timeout_t time_out = self->ith_timeout;
mach_port_t notify = self->ith_notify;
ipc_object_t object = self->ith_object;
ipc_mqueue_t mqueue = self->ith_mqueue;
ipc_kmsg_t kmsg;
mach_port_seqno_t seqno;
mach_msg_return_t mr;
if (option & MACH_RCV_LARGE) {
mr = ipc_mqueue_receive(mqueue, option & MACH_RCV_TIMEOUT,
rcv_size, time_out,
TRUE, mach_msg_receive_continue,
&kmsg, &seqno);
/* mqueue is unlocked */
ipc_object_release(object);
if (mr != MACH_MSG_SUCCESS) {
if (mr == MACH_RCV_TOO_LARGE) {
mach_msg_size_t real_size =
(mach_msg_size_t) (vm_offset_t) kmsg;
assert(real_size > rcv_size);
(void) copyout(&real_size,
&msg->msgh_size,
sizeof(mach_msg_size_t));
}
thread_syscall_return(mr);
/*NOTREACHED*/
}
kmsg->ikm_header.msgh_seqno = seqno;
assert(kmsg->ikm_header.msgh_size <= rcv_size);
} else {
mr = ipc_mqueue_receive(mqueue, option & MACH_RCV_TIMEOUT,
MACH_MSG_SIZE_MAX, time_out,
TRUE, mach_msg_receive_continue,
&kmsg, &seqno);
/* mqueue is unlocked */
ipc_object_release(object);
if (mr != MACH_MSG_SUCCESS) {
thread_syscall_return(mr);
/*NOTREACHED*/
}
kmsg->ikm_header.msgh_seqno = seqno;
if (kmsg->ikm_header.msgh_size > rcv_size) {
ipc_kmsg_copyout_dest(kmsg, space);
(void) ipc_kmsg_put(msg, kmsg, sizeof *msg);
thread_syscall_return(MACH_RCV_TOO_LARGE);
/*NOTREACHED*/
}
}
if (option & MACH_RCV_NOTIFY) {
if (notify == MACH_PORT_NULL)
mr = MACH_RCV_INVALID_NOTIFY;
else
mr = ipc_kmsg_copyout(kmsg, space, map, notify);
} else
mr = ipc_kmsg_copyout(kmsg, space, map, MACH_PORT_NULL);
if (mr != MACH_MSG_SUCCESS) {
if ((mr &~ MACH_MSG_MASK) == MACH_RCV_BODY_ERROR) {
(void) ipc_kmsg_put(msg, kmsg,
kmsg->ikm_header.msgh_size);
} else {
ipc_kmsg_copyout_dest(kmsg, space);
(void) ipc_kmsg_put(msg, kmsg, sizeof *msg);
}
thread_syscall_return(mr);
/*NOTREACHED*/
}
mr = ipc_kmsg_put(msg, kmsg, kmsg->ikm_header.msgh_size);
thread_syscall_return(mr);
/*NOTREACHED*/
}
int
msync_nocancel(__unused proc_t p, struct msync_nocancel_args *uap, __unused register_t *retval)
{
mach_vm_offset_t addr;
mach_vm_size_t size;
int flags;
vm_map_t user_map;
int rv;
vm_sync_t sync_flags=0;
addr = (mach_vm_offset_t) uap->addr;
size = (mach_vm_size_t)uap->len;
KERNEL_DEBUG_CONSTANT((BSDDBG_CODE(DBG_BSD_SC_EXTENDED_INFO, SYS_msync) | DBG_FUNC_NONE), (uint32_t)(addr >> 32), (uint32_t)(size >> 32), 0, 0, 0);
if (addr & PAGE_MASK_64) {
/* UNIX SPEC: user address is not page-aligned, return EINVAL */
return EINVAL;
}
if (size == 0) {
/*
* We cannot support this properly without maintaining
* list all mmaps done. Cannot use vm_map_entry as they could be
* split or coalesced by indepenedant actions. So instead of
* inaccurate results, lets just return error as invalid size
* specified
*/
return (EINVAL); /* XXX breaks posix apps */
}
flags = uap->flags;
/* disallow contradictory flags */
if ((flags & (MS_SYNC|MS_ASYNC)) == (MS_SYNC|MS_ASYNC))
return (EINVAL);
if (flags & MS_KILLPAGES)
sync_flags |= VM_SYNC_KILLPAGES;
if (flags & MS_DEACTIVATE)
sync_flags |= VM_SYNC_DEACTIVATE;
if (flags & MS_INVALIDATE)
sync_flags |= VM_SYNC_INVALIDATE;
if ( !(flags & (MS_KILLPAGES | MS_DEACTIVATE))) {
if (flags & MS_ASYNC)
sync_flags |= VM_SYNC_ASYNCHRONOUS;
else
sync_flags |= VM_SYNC_SYNCHRONOUS;
}
sync_flags |= VM_SYNC_CONTIGUOUS; /* complain if holes */
user_map = current_map();
rv = mach_vm_msync(user_map, addr, size, sync_flags);
switch (rv) {
case KERN_SUCCESS:
break;
case KERN_INVALID_ADDRESS: /* hole in region being sync'ed */
return (ENOMEM);
case KERN_FAILURE:
return (EIO);
default:
return (EINVAL);
}
return (0);
}
/*
* XXX Internally, we use VM_PROT_* somewhat interchangeably, but the correct
* XXX usage is PROT_* from an interface perspective. Thus the values of
* XXX VM_PROT_* and PROT_* need to correspond.
*/
int
mmap(proc_t p, struct mmap_args *uap, user_addr_t *retval)
{
/*
* Map in special device (must be SHARED) or file
*/
struct fileproc *fp;
register struct vnode *vp;
int flags;
int prot, file_prot;
int err=0;
vm_map_t user_map;
kern_return_t result;
mach_vm_offset_t user_addr;
mach_vm_size_t user_size;
vm_object_offset_t pageoff;
vm_object_offset_t file_pos;
int alloc_flags=0;
boolean_t docow;
vm_prot_t maxprot;
void *handle;
vm_pager_t pager;
int mapanon=0;
int fpref=0;
int error =0;
int fd = uap->fd;
user_addr = (mach_vm_offset_t)uap->addr;
user_size = (mach_vm_size_t) uap->len;
AUDIT_ARG(addr, user_addr);
AUDIT_ARG(len, user_size);
AUDIT_ARG(fd, uap->fd);
prot = (uap->prot & VM_PROT_ALL);
#if 3777787
/*
* Since the hardware currently does not support writing without
* read-before-write, or execution-without-read, if the request is
* for write or execute access, we must imply read access as well;
* otherwise programs expecting this to work will fail to operate.
*/
if (prot & (VM_PROT_EXECUTE | VM_PROT_WRITE))
prot |= VM_PROT_READ;
#endif /* radar 3777787 */
flags = uap->flags;
vp = NULLVP;
/*
* The vm code does not have prototypes & compiler doesn't do the'
* the right thing when you cast 64bit value and pass it in function
* call. So here it is.
*/
file_pos = (vm_object_offset_t)uap->pos;
/* make sure mapping fits into numeric range etc */
if (file_pos + user_size > (vm_object_offset_t)-PAGE_SIZE_64)
return (EINVAL);
/*
* Align the file position to a page boundary,
* and save its page offset component.
*/
pageoff = (file_pos & PAGE_MASK);
file_pos -= (vm_object_offset_t)pageoff;
/* Adjust size for rounding (on both ends). */
user_size += pageoff; /* low end... */
user_size = mach_vm_round_page(user_size); /* hi end */
/*
* Check for illegal addresses. Watch out for address wrap... Note
* that VM_*_ADDRESS are not constants due to casts (argh).
*/
if (flags & MAP_FIXED) {
/*
* The specified address must have the same remainder
* as the file offset taken modulo PAGE_SIZE, so it
* should be aligned after adjustment by pageoff.
*/
user_addr -= pageoff;
if (user_addr & PAGE_MASK)
return (EINVAL);
}
#ifdef notyet
/* DO not have apis to get this info, need to wait till then*/
/*
* XXX for non-fixed mappings where no hint is provided or
* the hint would fall in the potential heap space,
* place it after the end of the largest possible heap.
*
* There should really be a pmap call to determine a reasonable
* location.
*/
else if (addr < mach_vm_round_page(p->p_vmspace->vm_daddr + MAXDSIZ))
addr = mach_vm_round_page(p->p_vmspace->vm_daddr + MAXDSIZ);
#endif
alloc_flags = 0;
if (flags & MAP_ANON) {
/*
* Mapping blank space is trivial. Use positive fds as the alias
* value for memory tracking.
*/
if (fd != -1) {
/*
* Use "fd" to pass (some) Mach VM allocation flags,
* (see the VM_FLAGS_* definitions).
*/
alloc_flags = fd & (VM_FLAGS_ALIAS_MASK |
VM_FLAGS_PURGABLE);
if (alloc_flags != fd) {
/* reject if there are any extra flags */
return EINVAL;
}
}
handle = NULL;
maxprot = VM_PROT_ALL;
file_pos = 0;
mapanon = 1;
} else {
struct vnode_attr va;
vfs_context_t ctx = vfs_context_current();
/*
* Mapping file, get fp for validation. Obtain vnode and make
* sure it is of appropriate type.
*/
err = fp_lookup(p, fd, &fp, 0);
if (err)
return(err);
fpref = 1;
if(fp->f_fglob->fg_type == DTYPE_PSXSHM) {
uap->addr = (user_addr_t)user_addr;
uap->len = (user_size_t)user_size;
uap->prot = prot;
uap->flags = flags;
uap->pos = file_pos;
error = pshm_mmap(p, uap, retval, fp, (off_t)pageoff);
goto bad;
}
if (fp->f_fglob->fg_type != DTYPE_VNODE) {
error = EINVAL;
goto bad;
}
vp = (struct vnode *)fp->f_fglob->fg_data;
error = vnode_getwithref(vp);
if(error != 0)
goto bad;
if (vp->v_type != VREG && vp->v_type != VCHR) {
(void)vnode_put(vp);
error = EINVAL;
goto bad;
}
AUDIT_ARG(vnpath, vp, ARG_VNODE1);
/*
* POSIX: mmap needs to update access time for mapped files
*/
if ((vnode_vfsvisflags(vp) & MNT_NOATIME) == 0) {
VATTR_INIT(&va);
nanotime(&va.va_access_time);
VATTR_SET_ACTIVE(&va, va_access_time);
vnode_setattr(vp, &va, ctx);
}
/*
* XXX hack to handle use of /dev/zero to map anon memory (ala
* SunOS).
*/
if (vp->v_type == VCHR || vp->v_type == VSTR) {
(void)vnode_put(vp);
error = ENODEV;
goto bad;
} else {
/*
* Ensure that file and memory protections are
* compatible. Note that we only worry about
* writability if mapping is shared; in this case,
* current and max prot are dictated by the open file.
* XXX use the vnode instead? Problem is: what
* credentials do we use for determination? What if
* proc does a setuid?
*/
maxprot = VM_PROT_EXECUTE; /* ??? */
if (fp->f_fglob->fg_flag & FREAD)
maxprot |= VM_PROT_READ;
else if (prot & PROT_READ) {
(void)vnode_put(vp);
error = EACCES;
goto bad;
}
/*
* If we are sharing potential changes (either via
* MAP_SHARED or via the implicit sharing of character
* device mappings), and we are trying to get write
* permission although we opened it without asking
* for it, bail out.
*/
if ((flags & MAP_SHARED) != 0) {
if ((fp->f_fglob->fg_flag & FWRITE) != 0) {
/*
* check for write access
*
* Note that we already made this check when granting FWRITE
* against the file, so it seems redundant here.
*/
error = vnode_authorize(vp, NULL, KAUTH_VNODE_CHECKIMMUTABLE, ctx);
/* if not granted for any reason, but we wanted it, bad */
if ((prot & PROT_WRITE) && (error != 0)) {
vnode_put(vp);
goto bad;
}
/* if writable, remember */
if (error == 0)
maxprot |= VM_PROT_WRITE;
} else if ((prot & PROT_WRITE) != 0) {
(void)vnode_put(vp);
error = EACCES;
goto bad;
}
} else
maxprot |= VM_PROT_WRITE;
handle = (void *)vp;
#if CONFIG_MACF
error = mac_file_check_mmap(vfs_context_ucred(ctx),
fp->f_fglob, prot, flags, &maxprot);
if (error) {
(void)vnode_put(vp);
goto bad;
}
#endif /* MAC */
}
}
if (user_size == 0) {
if (!mapanon)
(void)vnode_put(vp);
error = 0;
goto bad;
}
/*
* We bend a little - round the start and end addresses
* to the nearest page boundary.
*/
user_size = mach_vm_round_page(user_size);
if (file_pos & PAGE_MASK_64) {
if (!mapanon)
(void)vnode_put(vp);
error = EINVAL;
goto bad;
}
user_map = current_map();
if ((flags & MAP_FIXED) == 0) {
alloc_flags |= VM_FLAGS_ANYWHERE;
user_addr = mach_vm_round_page(user_addr);
} else {
if (user_addr != mach_vm_trunc_page(user_addr)) {
if (!mapanon)
(void)vnode_put(vp);
error = EINVAL;
goto bad;
}
/*
* mmap(MAP_FIXED) will replace any existing mappings in the
* specified range, if the new mapping is successful.
* If we just deallocate the specified address range here,
* another thread might jump in and allocate memory in that
* range before we get a chance to establish the new mapping,
* and we won't have a chance to restore the old mappings.
* So we use VM_FLAGS_OVERWRITE to let Mach VM know that it
* has to deallocate the existing mappings and establish the
* new ones atomically.
*/
alloc_flags |= VM_FLAGS_FIXED | VM_FLAGS_OVERWRITE;
}
if (flags & MAP_NOCACHE)
alloc_flags |= VM_FLAGS_NO_CACHE;
/*
* Lookup/allocate object.
*/
if (handle == NULL) {
pager = NULL;
#ifdef notyet
/* Hmm .. */
#if defined(VM_PROT_READ_IS_EXEC)
if (prot & VM_PROT_READ)
prot |= VM_PROT_EXECUTE;
if (maxprot & VM_PROT_READ)
maxprot |= VM_PROT_EXECUTE;
#endif
#endif
#if 3777787
if (prot & (VM_PROT_EXECUTE | VM_PROT_WRITE))
prot |= VM_PROT_READ;
if (maxprot & (VM_PROT_EXECUTE | VM_PROT_WRITE))
maxprot |= VM_PROT_READ;
#endif /* radar 3777787 */
result = vm_map_enter_mem_object(user_map,
&user_addr, user_size,
0, alloc_flags,
IPC_PORT_NULL, 0, FALSE,
prot, maxprot,
(flags & MAP_SHARED) ?
VM_INHERIT_SHARE :
VM_INHERIT_DEFAULT);
if (result != KERN_SUCCESS)
goto out;
} else {
pager = (vm_pager_t)ubc_getpager(vp);
if (pager == NULL) {
(void)vnode_put(vp);
error = ENOMEM;
goto bad;
}
/*
* Set credentials:
* FIXME: if we're writing the file we need a way to
* ensure that someone doesn't replace our R/W creds
* with ones that only work for read.
*/
ubc_setthreadcred(vp, p, current_thread());
docow = FALSE;
if ((flags & (MAP_ANON|MAP_SHARED)) == 0) {
docow = TRUE;
}
#ifdef notyet
/* Hmm .. */
#if defined(VM_PROT_READ_IS_EXEC)
if (prot & VM_PROT_READ)
prot |= VM_PROT_EXECUTE;
if (maxprot & VM_PROT_READ)
maxprot |= VM_PROT_EXECUTE;
#endif
#endif /* notyet */
#if 3777787
if (prot & (VM_PROT_EXECUTE | VM_PROT_WRITE))
prot |= VM_PROT_READ;
if (maxprot & (VM_PROT_EXECUTE | VM_PROT_WRITE))
maxprot |= VM_PROT_READ;
#endif /* radar 3777787 */
result = vm_map_enter_mem_object(user_map,
&user_addr, user_size,
0, alloc_flags,
(ipc_port_t)pager, file_pos,
docow, prot, maxprot,
(flags & MAP_SHARED) ?
VM_INHERIT_SHARE :
VM_INHERIT_DEFAULT);
if (result != KERN_SUCCESS) {
(void)vnode_put(vp);
goto out;
}
file_prot = prot & (PROT_READ | PROT_WRITE | PROT_EXEC);
if (docow) {
/* private mapping: won't write to the file */
file_prot &= ~PROT_WRITE;
}
(void) ubc_map(vp, file_prot);
}
if (!mapanon)
(void)vnode_put(vp);
out:
switch (result) {
case KERN_SUCCESS:
*retval = user_addr + pageoff;
error = 0;
break;
case KERN_INVALID_ADDRESS:
case KERN_NO_SPACE:
error = ENOMEM;
break;
case KERN_PROTECTION_FAILURE:
error = EACCES;
break;
default:
error = EINVAL;
break;
}
bad:
if (fpref)
fp_drop(p, fd, fp, 0);
KERNEL_DEBUG_CONSTANT((BSDDBG_CODE(DBG_BSD_SC_EXTENDED_INFO, SYS_mmap) | DBG_FUNC_NONE), fd, (uint32_t)(*retval), (uint32_t)user_size, error, 0);
KERNEL_DEBUG_CONSTANT((BSDDBG_CODE(DBG_BSD_SC_EXTENDED_INFO2, SYS_mmap) | DBG_FUNC_NONE), (uint32_t)(*retval >> 32), (uint32_t)(user_size >> 32),
(uint32_t)(file_pos >> 32), (uint32_t)file_pos, 0);
return(error);
}
mach_msg_return_t
mach_msg_trap(
mach_msg_header_t *msg,
mach_msg_option_t option,
mach_msg_size_t send_size,
mach_msg_size_t rcv_size,
mach_port_t rcv_name,
mach_msg_timeout_t time_out,
mach_port_t notify)
{
mach_msg_return_t mr;
/* first check for common cases */
if (option == (MACH_SEND_MSG|MACH_RCV_MSG)) {
ipc_thread_t self = current_thread();
ipc_space_t space = self->task->itk_space;
ipc_kmsg_t kmsg;
ipc_port_t dest_port;
ipc_object_t rcv_object;
ipc_mqueue_t rcv_mqueue;
mach_msg_size_t reply_size;
/*
* This case is divided into ten sections, each
* with a label. There are five optimized
* sections and six unoptimized sections, which
* do the same thing but handle all possible
* cases and are slower.
*
* The five sections for an RPC are
* 1) Get request message into a buffer.
* (fast_get or slow_get)
* 2) Copyin request message and rcv_name.
* (fast_copyin or slow_copyin)
* 3) Enqueue request and dequeue reply.
* (fast_send_receive or
* slow_send and slow_receive)
* 4) Copyout reply message.
* (fast_copyout or slow_copyout)
* 5) Put reply message to user's buffer.
* (fast_put or slow_put)
*
* Keep the locking hierarchy firmly in mind.
* (First spaces, then ports, then port sets,
* then message queues.) Only a non-blocking
* attempt can be made to acquire locks out of
* order, or acquire two locks on the same level.
* Acquiring two locks on the same level will
* fail if the objects are really the same,
* unless simple locking is disabled. This is OK,
* because then the extra unlock does nothing.
*
* There are two major reasons these RPCs can't use
* ipc_thread_switch, and use slow_send/slow_receive:
* 1) Kernel RPCs.
* 2) Servers fall behind clients, so
* client doesn't find a blocked server thread and
* server finds waiting messages and can't block.
*/
/*
fast_get:
*/
/*
* optimized ipc_kmsg_get
*
* No locks, references, or messages held.
* We must clear ikm_cache before copyinmsg.
*/
if ((send_size > IKM_SAVED_MSG_SIZE) ||
(send_size < sizeof(mach_msg_header_t)) ||
(send_size & 3) ||
((kmsg = ikm_cache()) == IKM_NULL))
goto slow_get;
ikm_cache() = IKM_NULL;
ikm_check_initialized(kmsg, IKM_SAVED_KMSG_SIZE);
if (copyinmsg(msg, &kmsg->ikm_header,
send_size)) {
ikm_free(kmsg);
goto slow_get;
}
kmsg->ikm_header.msgh_size = send_size;
fast_copyin:
/*
* optimized ipc_kmsg_copyin/ipc_mqueue_copyin
*
* We have the request message data in kmsg.
* Must still do copyin, send, receive, etc.
*
* If the message isn't simple, we can't combine
* ipc_kmsg_copyin_header and ipc_mqueue_copyin,
* because copyin of the message body might
* affect rcv_name.
*/
switch (kmsg->ikm_header.msgh_bits) {
case MACH_MSGH_BITS(MACH_MSG_TYPE_COPY_SEND,
MACH_MSG_TYPE_MAKE_SEND_ONCE): {
ipc_entry_t table;
ipc_entry_num_t size;
ipc_port_t reply_port;
/* sending a request message */
{
mach_port_index_t index;
mach_port_gen_t gen;
{
mach_port_t reply_name =
kmsg->ikm_header.msgh_local_port;
if (reply_name != rcv_name)
goto slow_copyin;
/* optimized ipc_entry_lookup of reply_name */
index = MACH_PORT_INDEX(reply_name);
gen = MACH_PORT_GEN(reply_name);
}
is_read_lock(space);
assert(space->is_active);
size = space->is_table_size;
table = space->is_table;
if (index >= size)
goto abort_request_copyin;
{
ipc_entry_t entry;
ipc_entry_bits_t bits;
entry = &table[index];
bits = entry->ie_bits;
/* check generation number and type bit */
if ((bits & (IE_BITS_GEN_MASK|
MACH_PORT_TYPE_RECEIVE)) !=
(gen | MACH_PORT_TYPE_RECEIVE))
goto abort_request_copyin;
reply_port = (ipc_port_t) entry->ie_object;
assert(reply_port != IP_NULL);
}
}
/* optimized ipc_entry_lookup of dest_name */
{
mach_port_index_t index;
mach_port_gen_t gen;
{
mach_port_t dest_name =
kmsg->ikm_header.msgh_remote_port;
index = MACH_PORT_INDEX(dest_name);
gen = MACH_PORT_GEN(dest_name);
}
if (index >= size)
goto abort_request_copyin;
{
ipc_entry_t entry;
ipc_entry_bits_t bits;
entry = &table[index];
bits = entry->ie_bits;
/* check generation number and type bit */
if ((bits & (IE_BITS_GEN_MASK|MACH_PORT_TYPE_SEND)) !=
(gen | MACH_PORT_TYPE_SEND))
goto abort_request_copyin;
assert(IE_BITS_UREFS(bits) > 0);
dest_port = (ipc_port_t) entry->ie_object;
assert(dest_port != IP_NULL);
}
}
/*
* To do an atomic copyin, need simultaneous
* locks on both ports and the space. If
* dest_port == reply_port, and simple locking is
* enabled, then we will abort. Otherwise it's
* OK to unlock twice.
*/
ip_lock(dest_port);
if (!ip_active(dest_port) ||
!ip_lock_try(reply_port)) {
ip_unlock(dest_port);
goto abort_request_copyin;
}
is_read_unlock(space);
assert(dest_port->ip_srights > 0);
dest_port->ip_srights++;
ip_reference(dest_port);
assert(ip_active(reply_port));
assert(reply_port->ip_receiver_name ==
kmsg->ikm_header.msgh_local_port);
assert(reply_port->ip_receiver == space);
reply_port->ip_sorights++;
ip_reference(reply_port);
kmsg->ikm_header.msgh_bits =
MACH_MSGH_BITS(MACH_MSG_TYPE_PORT_SEND,
MACH_MSG_TYPE_PORT_SEND_ONCE);
kmsg->ikm_header.msgh_remote_port =
(mach_port_t) dest_port;
kmsg->ikm_header.msgh_local_port =
(mach_port_t) reply_port;
/* make sure we can queue to the destination */
if (dest_port->ip_receiver == ipc_space_kernel) {
/*
* The kernel server has a reference to
* the reply port, which it hands back
* to us in the reply message. We do
* not need to keep another reference to
* it.
*/
ip_unlock(reply_port);
assert(ip_active(dest_port));
ip_unlock(dest_port);
goto kernel_send;
}
if (dest_port->ip_msgcount >= dest_port->ip_qlimit)
goto abort_request_send_receive;
/* optimized ipc_mqueue_copyin */
if (reply_port->ip_pset != IPS_NULL)
goto abort_request_send_receive;
rcv_object = (ipc_object_t) reply_port;
io_reference(rcv_object);
rcv_mqueue = &reply_port->ip_messages;
imq_lock(rcv_mqueue);
io_unlock(rcv_object);
goto fast_send_receive;
abort_request_copyin:
is_read_unlock(space);
goto slow_copyin;
abort_request_send_receive:
ip_unlock(dest_port);
ip_unlock(reply_port);
goto slow_send;
}
case MACH_MSGH_BITS(MACH_MSG_TYPE_MOVE_SEND_ONCE, 0): {
ipc_entry_num_t size;
ipc_entry_t table;
/* sending a reply message */
{
mach_port_t reply_name =
kmsg->ikm_header.msgh_local_port;
if (reply_name != MACH_PORT_NULL)
goto slow_copyin;
}
is_write_lock(space);
assert(space->is_active);
/* optimized ipc_entry_lookup */
size = space->is_table_size;
table = space->is_table;
{
ipc_entry_t entry;
mach_port_gen_t gen;
mach_port_index_t index;
{
mach_port_t dest_name =
kmsg->ikm_header.msgh_remote_port;
index = MACH_PORT_INDEX(dest_name);
gen = MACH_PORT_GEN(dest_name);
}
if (index >= size)
goto abort_reply_dest_copyin;
entry = &table[index];
/* check generation, collision bit, and type bit */
if ((entry->ie_bits & (IE_BITS_GEN_MASK|
IE_BITS_COLLISION|
MACH_PORT_TYPE_SEND_ONCE)) !=
(gen | MACH_PORT_TYPE_SEND_ONCE))
goto abort_reply_dest_copyin;
/* optimized ipc_right_copyin */
assert(IE_BITS_TYPE(entry->ie_bits) ==
MACH_PORT_TYPE_SEND_ONCE);
assert(IE_BITS_UREFS(entry->ie_bits) == 1);
assert((entry->ie_bits & IE_BITS_MAREQUEST) == 0);
if (entry->ie_request != 0)
goto abort_reply_dest_copyin;
dest_port = (ipc_port_t) entry->ie_object;
assert(dest_port != IP_NULL);
ip_lock(dest_port);
if (!ip_active(dest_port)) {
ip_unlock(dest_port);
goto abort_reply_dest_copyin;
}
assert(dest_port->ip_sorights > 0);
/* optimized ipc_entry_dealloc */
entry->ie_next = table->ie_next;
table->ie_next = index;
entry->ie_bits = gen;
entry->ie_object = IO_NULL;
}
kmsg->ikm_header.msgh_bits =
MACH_MSGH_BITS(MACH_MSG_TYPE_PORT_SEND_ONCE,
0);
kmsg->ikm_header.msgh_remote_port =
(mach_port_t) dest_port;
/* make sure we can queue to the destination */
assert(dest_port->ip_receiver != ipc_space_kernel);
/* optimized ipc_entry_lookup/ipc_mqueue_copyin */
{
ipc_entry_t entry;
ipc_entry_bits_t bits;
{
mach_port_index_t index;
mach_port_gen_t gen;
index = MACH_PORT_INDEX(rcv_name);
gen = MACH_PORT_GEN(rcv_name);
if (index >= size)
goto abort_reply_rcv_copyin;
entry = &table[index];
bits = entry->ie_bits;
/* check generation number */
if ((bits & IE_BITS_GEN_MASK) != gen)
goto abort_reply_rcv_copyin;
}
/* check type bits; looking for receive or set */
if (bits & MACH_PORT_TYPE_PORT_SET) {
ipc_pset_t rcv_pset;
rcv_pset = (ipc_pset_t) entry->ie_object;
assert(rcv_pset != IPS_NULL);
ips_lock(rcv_pset);
assert(ips_active(rcv_pset));
rcv_object = (ipc_object_t) rcv_pset;
rcv_mqueue = &rcv_pset->ips_messages;
} else if (bits & MACH_PORT_TYPE_RECEIVE) {
ipc_port_t rcv_port;
rcv_port = (ipc_port_t) entry->ie_object;
assert(rcv_port != IP_NULL);
if (!ip_lock_try(rcv_port))
goto abort_reply_rcv_copyin;
assert(ip_active(rcv_port));
if (rcv_port->ip_pset != IPS_NULL) {
ip_unlock(rcv_port);
goto abort_reply_rcv_copyin;
}
rcv_object = (ipc_object_t) rcv_port;
rcv_mqueue = &rcv_port->ip_messages;
} else
goto abort_reply_rcv_copyin;
}
is_write_unlock(space);
io_reference(rcv_object);
imq_lock(rcv_mqueue);
io_unlock(rcv_object);
goto fast_send_receive;
abort_reply_dest_copyin:
is_write_unlock(space);
goto slow_copyin;
abort_reply_rcv_copyin:
ip_unlock(dest_port);
is_write_unlock(space);
goto slow_send;
}
default:
goto slow_copyin;
}
/*NOTREACHED*/
fast_send_receive:
/*
* optimized ipc_mqueue_send/ipc_mqueue_receive
*
* Finished get/copyin of kmsg and copyin of rcv_name.
* space is unlocked, dest_port is locked,
* we can queue kmsg to dest_port,
* rcv_mqueue is locked, rcv_object holds a ref,
* if rcv_object is a port it isn't in a port set
*
* Note that if simple locking is turned off,
* then we could have dest_mqueue == rcv_mqueue
* and not abort when we try to lock dest_mqueue.
*/
assert(ip_active(dest_port));
assert(dest_port->ip_receiver != ipc_space_kernel);
assert((dest_port->ip_msgcount < dest_port->ip_qlimit) ||
(MACH_MSGH_BITS_REMOTE(kmsg->ikm_header.msgh_bits) ==
MACH_MSG_TYPE_PORT_SEND_ONCE));
assert((kmsg->ikm_header.msgh_bits &
MACH_MSGH_BITS_CIRCULAR) == 0);
{
ipc_mqueue_t dest_mqueue;
ipc_thread_t receiver;
{
ipc_pset_t dest_pset;
dest_pset = dest_port->ip_pset;
if (dest_pset == IPS_NULL)
dest_mqueue = &dest_port->ip_messages;
else
dest_mqueue = &dest_pset->ips_messages;
}
if (!imq_lock_try(dest_mqueue)) {
abort_send_receive:
ip_unlock(dest_port);
imq_unlock(rcv_mqueue);
ipc_object_release(rcv_object);
goto slow_send;
}
receiver = ipc_thread_queue_first(&dest_mqueue->imq_threads);
if ((receiver == ITH_NULL) ||
(ipc_kmsg_queue_first(&rcv_mqueue->imq_messages)
!= IKM_NULL)) {
imq_unlock(dest_mqueue);
goto abort_send_receive;
}
/*
* There is a receiver thread waiting, and
* there is no reply message for us to pick up.
* We have hope of hand-off, so save state.
*/
self->ith_msg = msg;
self->ith_rcv_size = rcv_size;
self->ith_object = rcv_object;
self->ith_mqueue = rcv_mqueue;
if ((receiver->swap_func == (void (*)()) mach_msg_continue) &&
thread_handoff(self, mach_msg_continue, receiver)) {
assert(current_thread() == receiver);
/*
* We can use the optimized receive code,
* because the receiver is using no options.
*/
} else if ((receiver->swap_func ==
(void (*)()) exception_raise_continue) &&
thread_handoff(self, mach_msg_continue, receiver)) {
counter(c_mach_msg_trap_block_exc++);
assert(current_thread() == receiver);
/*
* We are a reply message coming back through
* the optimized exception-handling path.
* Finish with rcv_mqueue and dest_mqueue,
* and then jump to exception code with
* dest_port still locked. We don't bother
* with a sequence number in this case.
*/
ipc_thread_enqueue_macro(
&rcv_mqueue->imq_threads, self);
self->ith_state = MACH_RCV_IN_PROGRESS;
self->ith_msize = MACH_MSG_SIZE_MAX;
imq_unlock(rcv_mqueue);
ipc_thread_rmqueue_first_macro(
&dest_mqueue->imq_threads, receiver);
imq_unlock(dest_mqueue);
exception_raise_continue_fast(dest_port, kmsg);
/*NOTREACHED*/
return MACH_MSG_SUCCESS;
} else if ((send_size <= receiver->ith_msize) &&
thread_handoff(self, mach_msg_continue, receiver)) {
assert(current_thread() == receiver);
if ((receiver->swap_func ==
(void (*)()) mach_msg_receive_continue) &&
((receiver->ith_option & MACH_RCV_NOTIFY) == 0)) {
/*
* We can still use the optimized code.
*/
} else {
counter(c_mach_msg_trap_block_slow++);
/*
* We are running as the receiver,
* but we can't use the optimized code.
* Finish send/receive processing.
*/
dest_port->ip_msgcount++;
ip_unlock(dest_port);
ipc_thread_enqueue_macro(
&rcv_mqueue->imq_threads, self);
self->ith_state = MACH_RCV_IN_PROGRESS;
self->ith_msize = MACH_MSG_SIZE_MAX;
imq_unlock(rcv_mqueue);
ipc_thread_rmqueue_first_macro(
&dest_mqueue->imq_threads, receiver);
receiver->ith_state = MACH_MSG_SUCCESS;
receiver->ith_kmsg = kmsg;
receiver->ith_seqno = dest_port->ip_seqno++;
imq_unlock(dest_mqueue);
/*
* Call the receiver's continuation.
*/
receiver->wait_result = THREAD_AWAKENED;
(*receiver->swap_func)();
/*NOTREACHED*/
return MACH_MSG_SUCCESS;
}
} else {
/*
* The receiver can't accept the message,
* or we can't switch to the receiver.
*/
imq_unlock(dest_mqueue);
goto abort_send_receive;
}
counter(c_mach_msg_trap_block_fast++);
/*
* Safe to unlock dest_port now that we are
* committed to this path, because we hold
* dest_mqueue locked. We never bother changing
* dest_port->ip_msgcount.
*/
ip_unlock(dest_port);
/*
* We need to finish preparing self for its
* time asleep in rcv_mqueue.
*/
ipc_thread_enqueue_macro(&rcv_mqueue->imq_threads, self);
self->ith_state = MACH_RCV_IN_PROGRESS;
self->ith_msize = MACH_MSG_SIZE_MAX;
imq_unlock(rcv_mqueue);
/*
* Finish extracting receiver from dest_mqueue.
*/
ipc_thread_rmqueue_first_macro(
&dest_mqueue->imq_threads, receiver);
kmsg->ikm_header.msgh_seqno = dest_port->ip_seqno++;
imq_unlock(dest_mqueue);
/*
* We don't have to do any post-dequeue processing of
* the message. We never incremented ip_msgcount, we
* know it has no msg-accepted request, and blocked
* senders aren't a worry because we found the port
* with a receiver waiting.
*/
self = receiver;
space = self->task->itk_space;
msg = self->ith_msg;
rcv_size = self->ith_rcv_size;
rcv_object = self->ith_object;
/* inline ipc_object_release */
io_lock(rcv_object);
io_release(rcv_object);
io_check_unlock(rcv_object);
}
fast_copyout:
/*
* Nothing locked and no references held, except
* we have kmsg with msgh_seqno filled in. Must
* still check against rcv_size and do
* ipc_kmsg_copyout/ipc_kmsg_put.
*/
assert((ipc_port_t) kmsg->ikm_header.msgh_remote_port
== dest_port);
reply_size = kmsg->ikm_header.msgh_size;
if (rcv_size < reply_size)
goto slow_copyout;
/* optimized ipc_kmsg_copyout/ipc_kmsg_copyout_header */
switch (kmsg->ikm_header.msgh_bits) {
case MACH_MSGH_BITS(MACH_MSG_TYPE_PORT_SEND,
MACH_MSG_TYPE_PORT_SEND_ONCE): {
ipc_port_t reply_port =
(ipc_port_t) kmsg->ikm_header.msgh_local_port;
mach_port_t dest_name, reply_name;
/* receiving a request message */
if (!IP_VALID(reply_port))
goto slow_copyout;
is_write_lock(space);
assert(space->is_active);
/*
* To do an atomic copyout, need simultaneous
* locks on both ports and the space. If
* dest_port == reply_port, and simple locking is
* enabled, then we will abort. Otherwise it's
* OK to unlock twice.
*/
ip_lock(dest_port);
if (!ip_active(dest_port) ||
!ip_lock_try(reply_port))
goto abort_request_copyout;
if (!ip_active(reply_port)) {
ip_unlock(reply_port);
goto abort_request_copyout;
}
assert(reply_port->ip_sorights > 0);
ip_unlock(reply_port);
{
ipc_entry_t table;
ipc_entry_t entry;
mach_port_index_t index;
/* optimized ipc_entry_get */
table = space->is_table;
index = table->ie_next;
if (index == 0)
goto abort_request_copyout;
entry = &table[index];
table->ie_next = entry->ie_next;
entry->ie_request = 0;
{
mach_port_gen_t gen;
assert((entry->ie_bits &~ IE_BITS_GEN_MASK) == 0);
gen = entry->ie_bits + IE_BITS_GEN_ONE;
reply_name = MACH_PORT_MAKE(index, gen);
/* optimized ipc_right_copyout */
entry->ie_bits = gen | (MACH_PORT_TYPE_SEND_ONCE | 1);
}
assert(MACH_PORT_VALID(reply_name));
entry->ie_object = (ipc_object_t) reply_port;
is_write_unlock(space);
}
/* optimized ipc_object_copyout_dest */
assert(dest_port->ip_srights > 0);
ip_release(dest_port);
if (dest_port->ip_receiver == space)
dest_name = dest_port->ip_receiver_name;
else
dest_name = MACH_PORT_NULL;
if ((--dest_port->ip_srights == 0) &&
(dest_port->ip_nsrequest != IP_NULL)) {
ipc_port_t nsrequest;
mach_port_mscount_t mscount;
/* a rather rare case */
nsrequest = dest_port->ip_nsrequest;
mscount = dest_port->ip_mscount;
dest_port->ip_nsrequest = IP_NULL;
ip_unlock(dest_port);
ipc_notify_no_senders(nsrequest, mscount);
} else
ip_unlock(dest_port);
if (! ipc_port_flag_protected_payload(dest_port)) {
kmsg->ikm_header.msgh_bits = MACH_MSGH_BITS(
MACH_MSG_TYPE_PORT_SEND_ONCE,
MACH_MSG_TYPE_PORT_SEND);
kmsg->ikm_header.msgh_local_port = dest_name;
} else {
kmsg->ikm_header.msgh_bits = MACH_MSGH_BITS(
MACH_MSG_TYPE_PORT_SEND_ONCE,
MACH_MSG_TYPE_PROTECTED_PAYLOAD);
kmsg->ikm_header.msgh_protected_payload =
dest_port->ip_protected_payload;
}
kmsg->ikm_header.msgh_remote_port = reply_name;
goto fast_put;
abort_request_copyout:
ip_unlock(dest_port);
is_write_unlock(space);
goto slow_copyout;
}
case MACH_MSGH_BITS(MACH_MSG_TYPE_PORT_SEND_ONCE, 0): {
mach_port_t dest_name;
/* receiving a reply message */
ip_lock(dest_port);
if (!ip_active(dest_port))
goto slow_copyout;
/* optimized ipc_object_copyout_dest */
assert(dest_port->ip_sorights > 0);
if (dest_port->ip_receiver == space) {
ip_release(dest_port);
dest_port->ip_sorights--;
dest_name = dest_port->ip_receiver_name;
ip_unlock(dest_port);
} else {
ip_unlock(dest_port);
ipc_notify_send_once(dest_port);
dest_name = MACH_PORT_NULL;
}
if (! ipc_port_flag_protected_payload(dest_port)) {
kmsg->ikm_header.msgh_bits = MACH_MSGH_BITS(
0,
MACH_MSG_TYPE_PORT_SEND_ONCE);
kmsg->ikm_header.msgh_local_port = dest_name;
} else {
kmsg->ikm_header.msgh_bits = MACH_MSGH_BITS(
0,
MACH_MSG_TYPE_PROTECTED_PAYLOAD);
kmsg->ikm_header.msgh_protected_payload =
dest_port->ip_protected_payload;
}
kmsg->ikm_header.msgh_remote_port = MACH_PORT_NULL;
goto fast_put;
}
case MACH_MSGH_BITS_COMPLEX|
MACH_MSGH_BITS(MACH_MSG_TYPE_PORT_SEND_ONCE, 0): {
mach_port_t dest_name;
/* receiving a complex reply message */
ip_lock(dest_port);
if (!ip_active(dest_port))
goto slow_copyout;
/* optimized ipc_object_copyout_dest */
assert(dest_port->ip_sorights > 0);
if (dest_port->ip_receiver == space) {
ip_release(dest_port);
dest_port->ip_sorights--;
dest_name = dest_port->ip_receiver_name;
ip_unlock(dest_port);
} else {
ip_unlock(dest_port);
ipc_notify_send_once(dest_port);
dest_name = MACH_PORT_NULL;
}
if (! ipc_port_flag_protected_payload(dest_port)) {
kmsg->ikm_header.msgh_bits =
MACH_MSGH_BITS_COMPLEX
| MACH_MSGH_BITS(
0,
MACH_MSG_TYPE_PORT_SEND_ONCE);
kmsg->ikm_header.msgh_local_port = dest_name;
} else {
kmsg->ikm_header.msgh_bits =
MACH_MSGH_BITS_COMPLEX
| MACH_MSGH_BITS(
0,
MACH_MSG_TYPE_PROTECTED_PAYLOAD);
kmsg->ikm_header.msgh_protected_payload =
dest_port->ip_protected_payload;
}
kmsg->ikm_header.msgh_remote_port = MACH_PORT_NULL;
mr = ipc_kmsg_copyout_body(
(vm_offset_t) (&kmsg->ikm_header + 1),
(vm_offset_t) &kmsg->ikm_header
+ kmsg->ikm_header.msgh_size,
space,
current_map());
if (mr != MACH_MSG_SUCCESS) {
(void) ipc_kmsg_put(msg, kmsg,
kmsg->ikm_header.msgh_size);
return mr | MACH_RCV_BODY_ERROR;
}
goto fast_put;
}
default:
goto slow_copyout;
}
/*NOTREACHED*/
fast_put:
/*
* We have the reply message data in kmsg,
* and the reply message size in reply_size.
* Just need to copy it out to the user and free kmsg.
* We must check ikm_cache after copyoutmsg.
*/
ikm_check_initialized(kmsg, kmsg->ikm_size);
if ((kmsg->ikm_size != IKM_SAVED_KMSG_SIZE) ||
copyoutmsg(&kmsg->ikm_header, msg,
reply_size) ||
(ikm_cache() != IKM_NULL))
goto slow_put;
ikm_cache() = kmsg;
thread_syscall_return(MACH_MSG_SUCCESS);
/*NOTREACHED*/
return MACH_MSG_SUCCESS; /* help for the compiler */
/*
* The slow path has a few non-register temporary
* variables used only for call-by-reference.
*/
{
ipc_kmsg_t temp_kmsg;
mach_port_seqno_t temp_seqno;
ipc_object_t temp_rcv_object;
ipc_mqueue_t temp_rcv_mqueue;
slow_get:
/*
* No locks, references, or messages held.
* Still have to get the request, send it,
* receive reply, etc.
*/
mr = ipc_kmsg_get(msg, send_size, &temp_kmsg);
if (mr != MACH_MSG_SUCCESS) {
thread_syscall_return(mr);
/*NOTREACHED*/
}
kmsg = temp_kmsg;
/* try to get back on optimized path */
goto fast_copyin;
slow_copyin:
/*
* We have the message data in kmsg, but
* we still need to copyin, send it,
* receive a reply, and do copyout.
*/
mr = ipc_kmsg_copyin(kmsg, space, current_map(),
MACH_PORT_NULL);
if (mr != MACH_MSG_SUCCESS) {
ikm_free(kmsg);
thread_syscall_return(mr);
/*NOTREACHED*/
}
/* try to get back on optimized path */
if (kmsg->ikm_header.msgh_bits & MACH_MSGH_BITS_CIRCULAR)
goto slow_send;
dest_port = (ipc_port_t) kmsg->ikm_header.msgh_remote_port;
assert(IP_VALID(dest_port));
ip_lock(dest_port);
if (dest_port->ip_receiver == ipc_space_kernel) {
assert(ip_active(dest_port));
ip_unlock(dest_port);
goto kernel_send;
}
if (ip_active(dest_port) &&
((dest_port->ip_msgcount < dest_port->ip_qlimit) ||
(MACH_MSGH_BITS_REMOTE(kmsg->ikm_header.msgh_bits) ==
MACH_MSG_TYPE_PORT_SEND_ONCE)))
{
/*
* Try an optimized ipc_mqueue_copyin.
* It will work if this is a request message.
*/
ipc_port_t reply_port;
reply_port = (ipc_port_t)
kmsg->ikm_header.msgh_local_port;
if (IP_VALID(reply_port)) {
if (ip_lock_try(reply_port)) {
if (ip_active(reply_port) &&
reply_port->ip_receiver == space &&
reply_port->ip_receiver_name == rcv_name &&
reply_port->ip_pset == IPS_NULL)
{
/* Grab a reference to the reply port. */
rcv_object = (ipc_object_t) reply_port;
io_reference(rcv_object);
rcv_mqueue = &reply_port->ip_messages;
imq_lock(rcv_mqueue);
io_unlock(rcv_object);
goto fast_send_receive;
}
ip_unlock(reply_port);
}
}
}
ip_unlock(dest_port);
goto slow_send;
kernel_send:
/*
* Special case: send message to kernel services.
* The request message has been copied into the
* kmsg. Nothing is locked.
*/
{
ipc_port_t reply_port;
/*
* Perform the kernel function.
*/
kmsg = ipc_kobject_server(kmsg);
if (kmsg == IKM_NULL) {
/*
* No reply. Take the
* slow receive path.
*/
goto slow_get_rcv_port;
}
/*
* Check that:
* the reply port is alive
* we hold the receive right
* the name has not changed.
* the port is not in a set
* If any of these are not true,
* we cannot directly receive the reply
* message.
*/
reply_port = (ipc_port_t) kmsg->ikm_header.msgh_remote_port;
ip_lock(reply_port);
if ((!ip_active(reply_port)) ||
(reply_port->ip_receiver != space) ||
(reply_port->ip_receiver_name != rcv_name) ||
(reply_port->ip_pset != IPS_NULL))
{
ip_unlock(reply_port);
ipc_mqueue_send_always(kmsg);
goto slow_get_rcv_port;
}
rcv_mqueue = &reply_port->ip_messages;
imq_lock(rcv_mqueue);
/* keep port locked, and don`t change ref count yet */
/*
* If there are messages on the port
* or other threads waiting for a message,
* we cannot directly receive the reply.
*/
if ((ipc_thread_queue_first(&rcv_mqueue->imq_threads)
!= ITH_NULL) ||
(ipc_kmsg_queue_first(&rcv_mqueue->imq_messages)
!= IKM_NULL))
{
imq_unlock(rcv_mqueue);
ip_unlock(reply_port);
ipc_mqueue_send_always(kmsg);
goto slow_get_rcv_port;
}
/*
* We can directly receive this reply.
* Since the kernel reply never blocks,
* it holds no message_accepted request.
* Since there were no messages queued
* on the reply port, there should be
* no threads blocked waiting to send.
*/
assert(kmsg->ikm_marequest == IMAR_NULL);
assert(ipc_thread_queue_first(&reply_port->ip_blocked)
== ITH_NULL);
dest_port = reply_port;
kmsg->ikm_header.msgh_seqno = dest_port->ip_seqno++;
imq_unlock(rcv_mqueue);
/*
* inline ipc_object_release.
* Port is still locked.
* Reference count was not incremented.
*/
ip_check_unlock(reply_port);
/* copy out the kernel reply */
goto fast_copyout;
}
slow_send:
/*
* Nothing is locked. We have acquired kmsg, but
* we still need to send it and receive a reply.
*/
mr = ipc_mqueue_send(kmsg, MACH_MSG_OPTION_NONE,
MACH_MSG_TIMEOUT_NONE);
if (mr != MACH_MSG_SUCCESS) {
mr |= ipc_kmsg_copyout_pseudo(kmsg, space,
current_map());
assert(kmsg->ikm_marequest == IMAR_NULL);
(void) ipc_kmsg_put(msg, kmsg,
kmsg->ikm_header.msgh_size);
thread_syscall_return(mr);
/*NOTREACHED*/
}
slow_get_rcv_port:
/*
* We have sent the message. Copy in the receive port.
*/
mr = ipc_mqueue_copyin(space, rcv_name,
&temp_rcv_mqueue, &temp_rcv_object);
if (mr != MACH_MSG_SUCCESS) {
thread_syscall_return(mr);
/*NOTREACHED*/
}
rcv_mqueue = temp_rcv_mqueue;
rcv_object = temp_rcv_object;
/* hold ref for rcv_object; rcv_mqueue is locked */
/*
slow_receive:
*/
/*
* Now we have sent the request and copied in rcv_name,
* so rcv_mqueue is locked and hold ref for rcv_object.
* Just receive a reply and try to get back to fast path.
*
* ipc_mqueue_receive may not return, because if we block
* then our kernel stack may be discarded. So we save
* state here for mach_msg_continue to pick up.
*/
self->ith_msg = msg;
self->ith_rcv_size = rcv_size;
self->ith_object = rcv_object;
self->ith_mqueue = rcv_mqueue;
mr = ipc_mqueue_receive(rcv_mqueue,
MACH_MSG_OPTION_NONE,
MACH_MSG_SIZE_MAX,
MACH_MSG_TIMEOUT_NONE,
FALSE, mach_msg_continue,
&temp_kmsg, &temp_seqno);
/* rcv_mqueue is unlocked */
ipc_object_release(rcv_object);
if (mr != MACH_MSG_SUCCESS) {
thread_syscall_return(mr);
/*NOTREACHED*/
}
(kmsg = temp_kmsg)->ikm_header.msgh_seqno = temp_seqno;
dest_port = (ipc_port_t) kmsg->ikm_header.msgh_remote_port;
goto fast_copyout;
slow_copyout:
/*
* Nothing locked and no references held, except
* we have kmsg with msgh_seqno filled in. Must
* still check against rcv_size and do
* ipc_kmsg_copyout/ipc_kmsg_put.
*/
reply_size = kmsg->ikm_header.msgh_size;
if (rcv_size < reply_size) {
ipc_kmsg_copyout_dest(kmsg, space);
(void) ipc_kmsg_put(msg, kmsg, sizeof *msg);
thread_syscall_return(MACH_RCV_TOO_LARGE);
/*NOTREACHED*/
}
mr = ipc_kmsg_copyout(kmsg, space, current_map(),
MACH_PORT_NULL);
if (mr != MACH_MSG_SUCCESS) {
if ((mr &~ MACH_MSG_MASK) == MACH_RCV_BODY_ERROR) {
(void) ipc_kmsg_put(msg, kmsg,
kmsg->ikm_header.msgh_size);
} else {
ipc_kmsg_copyout_dest(kmsg, space);
(void) ipc_kmsg_put(msg, kmsg, sizeof *msg);
}
thread_syscall_return(mr);
/*NOTREACHED*/
}
/* try to get back on optimized path */
goto fast_put;
slow_put:
mr = ipc_kmsg_put(msg, kmsg, reply_size);
thread_syscall_return(mr);
/*NOTREACHED*/
}
} else if (option == MACH_SEND_MSG) {
ipc_space_t space = current_space();
vm_map_t map = current_map();
ipc_kmsg_t kmsg;
mr = ipc_kmsg_get(msg, send_size, &kmsg);
if (mr != MACH_MSG_SUCCESS)
return mr;
mr = ipc_kmsg_copyin(kmsg, space, map, MACH_PORT_NULL);
if (mr != MACH_MSG_SUCCESS) {
ikm_free(kmsg);
return mr;
}
mr = ipc_mqueue_send(kmsg, MACH_MSG_OPTION_NONE,
MACH_MSG_TIMEOUT_NONE);
if (mr != MACH_MSG_SUCCESS) {
mr |= ipc_kmsg_copyout_pseudo(kmsg, space, map);
assert(kmsg->ikm_marequest == IMAR_NULL);
(void) ipc_kmsg_put(msg, kmsg,
kmsg->ikm_header.msgh_size);
}
return mr;
} else if (option == MACH_RCV_MSG) {
ipc_thread_t self = current_thread();
ipc_space_t space = current_space();
vm_map_t map = current_map();
ipc_object_t object;
ipc_mqueue_t mqueue;
ipc_kmsg_t kmsg;
mach_port_seqno_t seqno;
mr = ipc_mqueue_copyin(space, rcv_name, &mqueue, &object);
if (mr != MACH_MSG_SUCCESS)
return mr;
/* hold ref for object; mqueue is locked */
/*
* ipc_mqueue_receive may not return, because if we block
* then our kernel stack may be discarded. So we save
* state here for mach_msg_continue to pick up.
*/
self->ith_msg = msg;
self->ith_rcv_size = rcv_size;
self->ith_object = object;
self->ith_mqueue = mqueue;
mr = ipc_mqueue_receive(mqueue,
MACH_MSG_OPTION_NONE,
MACH_MSG_SIZE_MAX,
MACH_MSG_TIMEOUT_NONE,
FALSE, mach_msg_continue,
&kmsg, &seqno);
/* mqueue is unlocked */
ipc_object_release(object);
if (mr != MACH_MSG_SUCCESS)
return mr;
kmsg->ikm_header.msgh_seqno = seqno;
if (rcv_size < kmsg->ikm_header.msgh_size) {
ipc_kmsg_copyout_dest(kmsg, space);
(void) ipc_kmsg_put(msg, kmsg, sizeof *msg);
return MACH_RCV_TOO_LARGE;
}
mr = ipc_kmsg_copyout(kmsg, space, map, MACH_PORT_NULL);
if (mr != MACH_MSG_SUCCESS) {
if ((mr &~ MACH_MSG_MASK) == MACH_RCV_BODY_ERROR) {
(void) ipc_kmsg_put(msg, kmsg,
kmsg->ikm_header.msgh_size);
} else {
ipc_kmsg_copyout_dest(kmsg, space);
(void) ipc_kmsg_put(msg, kmsg, sizeof *msg);
}
return mr;
}
return ipc_kmsg_put(msg, kmsg, kmsg->ikm_header.msgh_size);
} else if (option == MACH_MSG_OPTION_NONE) {
/*
* We can measure the "null mach_msg_trap"
* (syscall entry and thread_syscall_return exit)
* with this path.
*/
thread_syscall_return(MACH_MSG_SUCCESS);
/*NOTREACHED*/
}
if (option & MACH_SEND_MSG) {
mr = mach_msg_send(msg, option, send_size,
time_out, notify);
if (mr != MACH_MSG_SUCCESS)
return mr;
}
if (option & MACH_RCV_MSG) {
mr = mach_msg_receive(msg, option, rcv_size, rcv_name,
time_out, notify);
if (mr != MACH_MSG_SUCCESS)
return mr;
}
return MACH_MSG_SUCCESS;
}
int
mprotect(__unused proc_t p, struct mprotect_args *uap, __unused register_t *retval)
{
register vm_prot_t prot;
mach_vm_offset_t user_addr;
mach_vm_size_t user_size;
kern_return_t result;
vm_map_t user_map;
#if CONFIG_MACF
int error;
#endif
AUDIT_ARG(addr, uap->addr);
AUDIT_ARG(len, uap->len);
AUDIT_ARG(value, uap->prot);
user_addr = (mach_vm_offset_t) uap->addr;
user_size = (mach_vm_size_t) uap->len;
prot = (vm_prot_t)(uap->prot & VM_PROT_ALL);
if (user_addr & PAGE_MASK_64) {
/* UNIX SPEC: user address is not page-aligned, return EINVAL */
return EINVAL;
}
#ifdef notyet
/* Hmm .. */
#if defined(VM_PROT_READ_IS_EXEC)
if (prot & VM_PROT_READ)
prot |= VM_PROT_EXECUTE;
#endif
#endif /* notyet */
#if 3936456
if (prot & (VM_PROT_EXECUTE | VM_PROT_WRITE))
prot |= VM_PROT_READ;
#endif /* 3936456 */
user_map = current_map();
#if CONFIG_MACF
/*
* The MAC check for mprotect is of limited use for 2 reasons:
* Without mmap revocation, the caller could have asked for the max
* protections initially instead of a reduced set, so a mprotect
* check would offer no new security.
* It is not possible to extract the vnode from the pager object(s)
* of the target memory range.
* However, the MAC check may be used to prevent a process from,
* e.g., making the stack executable.
*/
error = mac_proc_check_mprotect(p, user_addr,
user_size, prot);
if (error)
return (error);
#endif
result = mach_vm_protect(user_map, user_addr, user_size,
FALSE, prot);
switch (result) {
case KERN_SUCCESS:
return (0);
case KERN_PROTECTION_FAILURE:
return (EACCES);
case KERN_INVALID_ADDRESS:
/* UNIX SPEC: for an invalid address range, return ENOMEM */
return ENOMEM;
}
return (EINVAL);
}