/**
* Gets the virtual memory map the specified object is mapped into.
*
* @returns VM map handle on success, NULL if no map.
* @param pMem The memory object.
*/
DECLINLINE(vm_map_t) rtR0MemObjDarwinGetMap(PRTR0MEMOBJINTERNAL pMem)
{
switch (pMem->enmType)
{
case RTR0MEMOBJTYPE_PAGE:
case RTR0MEMOBJTYPE_LOW:
case RTR0MEMOBJTYPE_CONT:
return kernel_map;
case RTR0MEMOBJTYPE_PHYS:
case RTR0MEMOBJTYPE_PHYS_NC:
return NULL; /* pretend these have no mapping atm. */
case RTR0MEMOBJTYPE_LOCK:
return pMem->u.Lock.R0Process == NIL_RTR0PROCESS
? kernel_map
: get_task_map((task_t)pMem->u.Lock.R0Process);
case RTR0MEMOBJTYPE_RES_VIRT:
return pMem->u.ResVirt.R0Process == NIL_RTR0PROCESS
? kernel_map
: get_task_map((task_t)pMem->u.ResVirt.R0Process);
case RTR0MEMOBJTYPE_MAPPING:
return pMem->u.Mapping.R0Process == NIL_RTR0PROCESS
? kernel_map
: get_task_map((task_t)pMem->u.Mapping.R0Process);
default:
return NULL;
}
}
/*
* fork_create_child
*
* Description: Common operations associated with the creation of a child
* process
*
* Parameters: parent_task parent task
* child_proc child process
* inherit_memory TRUE, if the parents address space is
* to be inherited by the child
* is64bit TRUE, if the child being created will
* be associated with a 64 bit process
* rather than a 32 bit process
*
* Note: This code is called in the fork() case, from the execve() call
* graph, if implementing an execve() following a vfork(), from
* the posix_spawn() call graph (which implicitly includes a
* vfork() equivalent call, and in the system bootstrap case.
*
* It creates a new task and thread (and as a side effect of the
* thread creation, a uthread), which is then associated with the
* process 'child'. If the parent process address space is to
* be inherited, then a flag indicates that the newly created
* task should inherit this from the child task.
*
* As a special concession to bootstrapping the initial process
* in the system, it's possible for 'parent_task' to be TASK_NULL;
* in this case, 'inherit_memory' MUST be FALSE.
*/
thread_t
fork_create_child(task_t parent_task, proc_t child_proc, int inherit_memory, int is64bit)
{
thread_t child_thread = NULL;
task_t child_task;
kern_return_t result;
/* Create a new task for the child process */
result = task_create_internal(parent_task,
inherit_memory,
is64bit,
&child_task);
if (result != KERN_SUCCESS) {
printf("execve: task_create_internal failed. Code: %d\n", result);
goto bad;
}
/* Set the child process task to the new task */
child_proc->task = child_task;
/* Set child task process to child proc */
set_bsdtask_info(child_task, child_proc);
/* Propagate CPU limit timer from parent */
if (timerisset(&child_proc->p_rlim_cpu))
task_vtimer_set(child_task, TASK_VTIMER_RLIM);
/* Set/clear 64 bit vm_map flag */
if (is64bit)
vm_map_set_64bit(get_task_map(child_task));
else
vm_map_set_32bit(get_task_map(child_task));
#if CONFIG_MACF
/* Update task for MAC framework */
/* valid to use p_ucred as child is still not running ... */
mac_task_label_update_cred(child_proc->p_ucred, child_task);
#endif
/*
* Set child process BSD visible scheduler priority if nice value
* inherited from parent
*/
if (child_proc->p_nice != 0)
resetpriority(child_proc);
/* Create a new thread for the child process */
result = thread_create(child_task, &child_thread);
if (result != KERN_SUCCESS) {
printf("execve: thread_create failed. Code: %d\n", result);
task_deallocate(child_task);
child_task = NULL;
}
bad:
thread_yield_internal(1);
return(child_thread);
}
int
cs_allow_invalid(struct proc *p)
{
#if MACH_ASSERT
lck_mtx_assert(&p->p_mlock, LCK_MTX_ASSERT_NOTOWNED);
#endif
#if CONFIG_MACF && CONFIG_ENFORCE_SIGNED_CODE
/* There needs to be a MAC policy to implement this hook, or else the
* kill bits will be cleared here every time. If we have
* CONFIG_ENFORCE_SIGNED_CODE, we can assume there is a policy
* implementing the hook.
*/
if( 0 != mac_proc_check_run_cs_invalid(p)) {
if(cs_debug) printf("CODE SIGNING: cs_allow_invalid() "
"not allowed: pid %d\n",
p->p_pid);
return 0;
}
if(cs_debug) printf("CODE SIGNING: cs_allow_invalid() "
"allowed: pid %d\n",
p->p_pid);
proc_lock(p);
p->p_csflags &= ~(CS_KILL | CS_HARD);
proc_unlock(p);
vm_map_switch_protect(get_task_map(p->task), FALSE);
#endif
return (p->p_csflags & (CS_KILL | CS_HARD)) == 0;
}
__private_extern__ kern_return_t
chudxnu_task_write(
task_t task,
uint64_t useraddr,
void *kernaddr,
vm_size_t size)
{
kern_return_t ret = KERN_SUCCESS;
boolean_t old_level;
if(ml_at_interrupt_context()) {
return KERN_FAILURE; // can't poke into tasks on interrupt stack
}
/*
* pmap layer requires interrupts to be on
*/
old_level = ml_set_interrupts_enabled(TRUE);
if(current_task()==task) {
if(copyout(kernaddr, useraddr, size)) {
ret = KERN_FAILURE;
}
} else {
vm_map_t map = get_task_map(task);
ret = vm_map_write_user(map, kernaddr, useraddr, size);
}
ml_set_interrupts_enabled(old_level);
return ret;
}
/* hw.pagesize and hw.tbfrequency are expected as 64 bit values */
static int
sysctl_pagesize
(__unused struct sysctl_oid *oidp, __unused void *arg1, __unused int arg2, struct sysctl_req *req)
{
vm_map_t map = get_task_map(current_task());
long long l = vm_map_page_size(map);
return sysctl_io_number(req, l, sizeof(l), NULL, NULL);
}
/*
* cloneproc
*
* Description: Create a new process from a specified process.
*
* Parameters: parent_task The parent task to be cloned, or
* TASK_NULL is task characteristics
* are not to be inherited
* be cloned, or TASK_NULL if the new
* task is not to inherit the VM
* characteristics of the parent
* parent_proc The parent process to be cloned
* inherit_memory True if the child is to inherit
* memory from the parent; if this is
* non-NULL, then the parent_task must
* also be non-NULL
*
* Returns: !NULL pointer to new child thread
* NULL Failure (unspecified)
*
* Note: On return newly created child process has signal lock held
* to block delivery of signal to it if called with lock set.
* fork() code needs to explicity remove this lock before
* signals can be delivered
*
* In the case of bootstrap, this function can be called from
* bsd_utaskbootstrap() in order to bootstrap the first process;
* the net effect is to provide a uthread structure for the
* kernel process associated with the kernel task.
*
* XXX: Tristating using the value parent_task as the major key
* and inherit_memory as the minor key is something we should
* refactor later; we owe the current semantics, ultimately,
* to the semantics of task_create_internal. For now, we will
* live with this being somewhat awkward.
*/
thread_t
cloneproc(task_t parent_task, proc_t parent_proc, int inherit_memory)
{
task_t child_task;
proc_t child_proc;
thread_t child_thread = NULL;
if ((child_proc = forkproc(parent_proc)) == NULL) {
/* Failed to allocate new process */
goto bad;
}
child_thread = fork_create_child(parent_task, child_proc, inherit_memory, (parent_task == TASK_NULL) ? FALSE : (parent_proc->p_flag & P_LP64));
if (child_thread == NULL) {
/*
* Failed to create thread; now we must deconstruct the new
* process previously obtained from forkproc().
*/
forkproc_free(child_proc);
goto bad;
}
child_task = get_threadtask(child_thread);
if (parent_proc->p_flag & P_LP64) {
task_set_64bit(child_task, TRUE);
OSBitOrAtomic(P_LP64, (UInt32 *)&child_proc->p_flag);
#ifdef __ppc__
/*
* PPC51: ppc64 is limited to 51-bit addresses.
* Memory above that limit is handled specially at
* the pmap level.
*/
pmap_map_sharedpage(child_task, get_map_pmap(get_task_map(child_task)));
#endif /* __ppc__ */
} else {
task_set_64bit(child_task, FALSE);
OSBitAndAtomic(~((uint32_t)P_LP64), (UInt32 *)&child_proc->p_flag);
}
/* make child visible */
pinsertchild(parent_proc, child_proc);
/*
* Make child runnable, set start time.
*/
child_proc->p_stat = SRUN;
bad:
return(child_thread);
}
void log_unnest_badness(vm_map_t m, vm_map_offset_t s, vm_map_offset_t e) {
struct timeval tv;
const char *pcommstr;
if (shared_region_unnest_logging == 0)
return;
if (shared_region_unnest_logging == 1) {
microtime(&tv);
if ((tv.tv_sec - last_unnest_log_time) < vm_shared_region_unnest_log_interval) {
if (shared_region_unnest_log_count++ > shared_region_unnest_log_count_threshold)
return;
}
else {
last_unnest_log_time = tv.tv_sec;
shared_region_unnest_log_count = 0;
}
}
pcommstr = current_proc()->p_comm;
printf("%s (map: %p) triggered DYLD shared region unnest for map: %p, region 0x%qx->0x%qx. While not abnormal for debuggers, this increases system memory footprint until the target exits.\n", current_proc()->p_comm, get_task_map(current_proc()->task), m, (uint64_t)s, (uint64_t)e);
}
/*
* Supporting some variables requires us to do "real" work. We
* gather some of that here.
*/
static int
sysctl_hw_generic(__unused struct sysctl_oid *oidp, __unused void *arg1,
int arg2, struct sysctl_req *req)
{
char dummy[65];
int epochTemp;
ml_cpu_info_t cpu_info;
int val, doquad;
long long qval;
host_basic_info_data_t hinfo;
kern_return_t kret;
mach_msg_type_number_t count = HOST_BASIC_INFO_COUNT;
/*
* Test and mask off the 'return quad' flag.
* Note that only some things here support it.
*/
doquad = arg2 & CTLHW_RETQUAD;
arg2 &= ~CTLHW_RETQUAD;
ml_cpu_get_info(&cpu_info);
#define BSD_HOST 1
kret = host_info((host_t)BSD_HOST, HOST_BASIC_INFO, (host_info_t)&hinfo, &count);
/*
* Handle various OIDs.
*
* OIDs that can return int or quad set val and qval and then break.
* Errors and int-only values return inline.
*/
switch (arg2) {
case HW_NCPU:
if (kret == KERN_SUCCESS) {
return(SYSCTL_RETURN(req, hinfo.max_cpus));
} else {
return(EINVAL);
}
case HW_AVAILCPU:
if (kret == KERN_SUCCESS) {
return(SYSCTL_RETURN(req, hinfo.avail_cpus));
} else {
return(EINVAL);
}
case HW_LOCAL_PHYSICALCPU:
if (kret == KERN_SUCCESS) {
return(SYSCTL_RETURN(req, hinfo.physical_cpu));
} else {
return(EINVAL);
}
case HW_LOCAL_PHYSICALCPUMAX:
if (kret == KERN_SUCCESS) {
return(SYSCTL_RETURN(req, hinfo.physical_cpu_max));
} else {
return(EINVAL);
}
case HW_LOCAL_LOGICALCPU:
if (kret == KERN_SUCCESS) {
return(SYSCTL_RETURN(req, hinfo.logical_cpu));
} else {
return(EINVAL);
}
case HW_LOCAL_LOGICALCPUMAX:
if (kret == KERN_SUCCESS) {
return(SYSCTL_RETURN(req, hinfo.logical_cpu_max));
} else {
return(EINVAL);
}
case HW_PAGESIZE:
{
vm_map_t map = get_task_map(current_task());
val = vm_map_page_size(map);
qval = (long long)val;
break;
}
case HW_CACHELINE:
val = cpu_info.cache_line_size;
qval = (long long)val;
break;
case HW_L1ICACHESIZE:
val = cpu_info.l1_icache_size;
qval = (long long)val;
break;
case HW_L1DCACHESIZE:
val = cpu_info.l1_dcache_size;
qval = (long long)val;
break;
case HW_L2CACHESIZE:
if (cpu_info.l2_cache_size == 0xFFFFFFFF)
return(EINVAL);
val = cpu_info.l2_cache_size;
qval = (long long)val;
break;
case HW_L3CACHESIZE:
if (cpu_info.l3_cache_size == 0xFFFFFFFF)
return(EINVAL);
val = cpu_info.l3_cache_size;
qval = (long long)val;
break;
/*
* Deprecated variables. We still support these for
* backwards compatibility purposes only.
*/
case HW_MACHINE:
bzero(dummy, sizeof(dummy));
if(!PEGetMachineName(dummy,64))
return(EINVAL);
dummy[64] = 0;
return(SYSCTL_OUT(req, dummy, strlen(dummy) + 1));
case HW_MODEL:
bzero(dummy, sizeof(dummy));
if(!PEGetModelName(dummy,64))
return(EINVAL);
dummy[64] = 0;
return(SYSCTL_OUT(req, dummy, strlen(dummy) + 1));
case HW_USERMEM:
{
int usermem = mem_size - vm_page_wire_count * page_size;
return(SYSCTL_RETURN(req, usermem));
}
case HW_EPOCH:
epochTemp = PEGetPlatformEpoch();
if (epochTemp == -1)
return(EINVAL);
return(SYSCTL_RETURN(req, epochTemp));
case HW_VECTORUNIT: {
int vector = cpu_info.vector_unit == 0? 0 : 1;
return(SYSCTL_RETURN(req, vector));
}
case HW_L2SETTINGS:
if (cpu_info.l2_cache_size == 0xFFFFFFFF)
return(EINVAL);
return(SYSCTL_RETURN(req, cpu_info.l2_settings));
case HW_L3SETTINGS:
if (cpu_info.l3_cache_size == 0xFFFFFFFF)
return(EINVAL);
return(SYSCTL_RETURN(req, cpu_info.l3_settings));
default:
return(ENOTSUP);
}
/*
* Callers may come to us with either int or quad buffers.
*/
if (doquad) {
return(SYSCTL_RETURN(req, qval));
}
return(SYSCTL_RETURN(req, val));
}
bool IOBufferMemoryDescriptor::initWithOptions(
IOOptionBits options,
vm_size_t capacity,
vm_offset_t alignment,
task_t inTask)
{
vm_map_t map = 0;
IOOptionBits iomdOptions = kIOMemoryAsReference | kIOMemoryTypeVirtual;
if (!capacity)
return false;
_options = options;
_capacity = capacity;
_physAddrs = 0;
_physSegCount = 0;
_buffer = 0;
// Grab the direction and the Auto Prepare bits from the Buffer MD options
iomdOptions |= options & (kIOMemoryDirectionMask | kIOMemoryAutoPrepare);
if ((options & kIOMemorySharingTypeMask) && (alignment < page_size))
alignment = page_size;
if ((inTask != kernel_task) && !(options & kIOMemoryPageable))
return false;
_alignment = alignment;
if (options & kIOMemoryPageable)
{
iomdOptions |= kIOMemoryBufferPageable;
if (inTask == kernel_task)
{
/* Allocate some kernel address space. */
_buffer = IOMallocPageable(capacity, alignment);
if (_buffer)
map = IOPageableMapForAddress((vm_address_t) _buffer);
}
else
{
kern_return_t kr;
if( !reserved) {
reserved = IONew( ExpansionData, 1 );
if( !reserved)
return( false );
}
map = get_task_map(inTask);
vm_map_reference(map);
reserved->map = map;
kr = vm_allocate( map, (vm_address_t *) &_buffer, round_page_32(capacity),
VM_FLAGS_ANYWHERE | VM_MAKE_TAG(VM_MEMORY_IOKIT) );
if( KERN_SUCCESS != kr)
return( false );
// we have to make sure that these pages don't get copied on fork.
kr = vm_inherit( map, (vm_address_t) _buffer, round_page_32(capacity), VM_INHERIT_NONE);
if( KERN_SUCCESS != kr)
return( false );
}
}
else
{
// @@@ gvdl: Need to remove this
// Buffer should never auto prepare they should be prepared explicitly
// But it never was enforced so what are you going to do?
iomdOptions |= kIOMemoryAutoPrepare;
/* Allocate a wired-down buffer inside kernel space. */
if (options & kIOMemoryPhysicallyContiguous)
_buffer = IOMallocContiguous(capacity, alignment, 0);
else if (alignment > 1)
_buffer = IOMallocAligned(capacity, alignment);
else
_buffer = IOMalloc(capacity);
}
if (!_buffer)
return false;
_singleRange.v.address = (vm_address_t) _buffer;
_singleRange.v.length = capacity;
if (!super::initWithOptions(&_singleRange.v, 1, 0,
inTask, iomdOptions, /* System mapper */ 0))
return false;
if (options & kIOMemoryPageable) {
kern_return_t kr;
ipc_port_t sharedMem = (ipc_port_t) _memEntry;
vm_size_t size = round_page_32(_ranges.v[0].length);
// must create the entry before any pages are allocated
if( 0 == sharedMem) {
// set memory entry cache
vm_prot_t memEntryCacheMode = VM_PROT_READ | VM_PROT_WRITE;
switch (options & kIOMapCacheMask)
{
case kIOMapInhibitCache:
SET_MAP_MEM(MAP_MEM_IO, memEntryCacheMode);
break;
case kIOMapWriteThruCache:
SET_MAP_MEM(MAP_MEM_WTHRU, memEntryCacheMode);
break;
case kIOMapWriteCombineCache:
SET_MAP_MEM(MAP_MEM_WCOMB, memEntryCacheMode);
break;
case kIOMapCopybackCache:
SET_MAP_MEM(MAP_MEM_COPYBACK, memEntryCacheMode);
break;
case kIOMapDefaultCache:
default:
SET_MAP_MEM(MAP_MEM_NOOP, memEntryCacheMode);
break;
}
kr = mach_make_memory_entry( map,
&size, _ranges.v[0].address,
memEntryCacheMode, &sharedMem,
NULL );
if( (KERN_SUCCESS == kr) && (size != round_page_32(_ranges.v[0].length))) {
ipc_port_release_send( sharedMem );
kr = kIOReturnVMError;
}
if( KERN_SUCCESS != kr)
sharedMem = 0;
_memEntry = (void *) sharedMem;
}
}
setLength(capacity);
return true;
}
/*
* fork_create_child
*
* Description: Common operations associated with the creation of a child
* process
*
* Parameters: parent_task parent task
* parent_coalitions parent's set of coalitions
* child_proc child process
* inherit_memory TRUE, if the parents address space is
* to be inherited by the child
* is64bit TRUE, if the child being created will
* be associated with a 64 bit process
* rather than a 32 bit process
*
* Note: This code is called in the fork() case, from the execve() call
* graph, if implementing an execve() following a vfork(), from
* the posix_spawn() call graph (which implicitly includes a
* vfork() equivalent call, and in the system bootstrap case.
*
* It creates a new task and thread (and as a side effect of the
* thread creation, a uthread) in the parent coalition set, which is
* then associated with the process 'child'. If the parent
* process address space is to be inherited, then a flag
* indicates that the newly created task should inherit this from
* the child task.
*
* As a special concession to bootstrapping the initial process
* in the system, it's possible for 'parent_task' to be TASK_NULL;
* in this case, 'inherit_memory' MUST be FALSE.
*/
thread_t
fork_create_child(task_t parent_task, coalition_t *parent_coalitions, proc_t child_proc, int inherit_memory, int is64bit)
{
thread_t child_thread = NULL;
task_t child_task;
kern_return_t result;
/* Create a new task for the child process */
result = task_create_internal(parent_task,
parent_coalitions,
inherit_memory,
is64bit,
&child_task);
if (result != KERN_SUCCESS) {
printf("%s: task_create_internal failed. Code: %d\n",
__func__, result);
goto bad;
}
/* Set the child process task to the new task */
child_proc->task = child_task;
/* Set child task process to child proc */
set_bsdtask_info(child_task, child_proc);
/* Propagate CPU limit timer from parent */
if (timerisset(&child_proc->p_rlim_cpu))
task_vtimer_set(child_task, TASK_VTIMER_RLIM);
/* Set/clear 64 bit vm_map flag */
if (is64bit)
vm_map_set_64bit(get_task_map(child_task));
else
vm_map_set_32bit(get_task_map(child_task));
/*
* Set child process BSD visible scheduler priority if nice value
* inherited from parent
*/
if (child_proc->p_nice != 0)
resetpriority(child_proc);
/* Create a new thread for the child process */
result = thread_create_with_continuation(child_task, &child_thread, (thread_continue_t)proc_wait_to_return);
if (result != KERN_SUCCESS) {
printf("%s: thread_create failed. Code: %d\n",
__func__, result);
task_deallocate(child_task);
child_task = NULL;
}
/*
* Tag thread as being the first thread in its task.
*/
thread_set_tag(child_thread, THREAD_TAG_MAINTHREAD);
bad:
thread_yield_internal(1);
return(child_thread);
}