static void test_vm_map(void)
{
paddr_t paddr;
vaddr_t vaddr;
vm_as_t* as;
as = vm_get_kernel_as();
vaddr = (vaddr_t)0xb0000000;
paddr = phys_alloc();
vm_map(as, vaddr + PHYS_FRAME_SIZE * 0, paddr);
*(volatile uint32_t*)(vaddr + PHYS_FRAME_SIZE * 0) = 0x2a;
serial_printl("[?] --> %x\n",
*(volatile uint32_t*)(vaddr + PHYS_FRAME_SIZE * 0));
vm_map(as, vaddr + PHYS_FRAME_SIZE * 1, paddr);
*(volatile uint32_t*)(vaddr + PHYS_FRAME_SIZE * 1) = 0x2a;
serial_printl("[?] --> %x\n",
*(volatile uint32_t*)(vaddr + PHYS_FRAME_SIZE * 1));
vm_map(as, vaddr + PHYS_FRAME_SIZE * 2, paddr);
*(volatile uint32_t*)(vaddr + PHYS_FRAME_SIZE * 2) = 0x2a;
serial_printl("[?] --> %x\n",
*(volatile uint32_t*)(vaddr + PHYS_FRAME_SIZE * 2));
phys_free(paddr);
}
int main(int argc, char** argv)
{
//定义变量并初始化
kern_return_t kr;
vm_task_t target_task=mach_task_self();
vm_address_t *address=NULL;
vm_size_t size=vm_page_size;
vm_address_t mask;//mask不会设置；
boolean_t anywhere=TURE;
memory_object_t memory_object=MEMORY_OBJECT_NULL;
vm_offset_t offset=0;
boolean_t copy=FALSE;
vm_prot_t cur_protection=(VM_PORT_READ|VM_PORT_WRITE);//当前保护属性，读写
vm_prot_t max_protection=(VM_PORT_READ|VM_PORT_WRITE);//最大保护属性，读写
vm_inherit_t inheritance=VM_INHERIT_SHARE;//共享，设置共享
kr=0;
//建立一个新的内存对象，使用vm_map()；
kr=vm_map(target_task,*address,size,mask,anywhere,
memory_object,offset,copy,cur_protection,
max_protection, inheritance);
if(kr)
{
mach_error("the value of error is",kr);
printf("vm_map() is exiting:%d\n",kr);
return kr;
}
printf("vm_map is ok\n");
}
/* Implement io_map_cntl as described in <hurd/io.defs>. */
kern_return_t
diskfs_S_io_map_cntl (struct protid *cred,
memory_object_t *ctlobj,
mach_msg_type_name_t *ctlobj_type)
{
if (!cred)
return EOPNOTSUPP;
assert (__vm_page_size >= sizeof (struct shared_io));
mutex_lock (&cred->po->np->lock);
if (!cred->mapped)
{
default_pager_object_create (diskfs_default_pager, &cred->shared_object,
__vm_page_size);
vm_map (mach_task_self (), (vm_address_t *)&cred->mapped, vm_page_size,
0, 1, cred->shared_object, 0, 0,
VM_PROT_READ|VM_PROT_WRITE, VM_PROT_READ|VM_PROT_WRITE, 0);
cred->mapped->shared_page_magic = SHARED_PAGE_MAGIC;
cred->mapped->conch_status = USER_HAS_NOT_CONCH;
spin_lock_init (&cred->mapped->lock);
*ctlobj = cred->shared_object;
*ctlobj_type = MACH_MSG_TYPE_COPY_SEND;
mutex_unlock (&cred->po->np->lock);
return 0;
}
else
{
mutex_unlock (&cred->po->np->lock);
return EBUSY;
}
}
PageAllocationAligned PageAllocationAligned::allocate(size_t size, size_t alignment, OSAllocator::Usage usage, bool writable)
{
ASSERT(isPageAligned(size));
ASSERT(isPageAligned(alignment));
ASSERT(isPowerOfTwo(alignment));
ASSERT(size >= alignment);
size_t alignmentMask = alignment - 1;
#if OS(DARWIN)
int flags = VM_FLAGS_ANYWHERE;
if (usage != OSAllocator::UnknownUsage)
flags |= usage;
int protection = PROT_READ;
if (writable)
protection |= PROT_WRITE;
vm_address_t address = 0;
vm_map(current_task(), &address, size, alignmentMask, flags, MEMORY_OBJECT_NULL, 0, FALSE, protection, PROT_READ | PROT_WRITE, VM_INHERIT_DEFAULT);
return PageAllocationAligned(reinterpret_cast<void*>(address), size);
#else
size_t alignmentDelta = alignment - pageSize();
// Resererve with suffcient additional VM to correctly align.
size_t reservationSize = size + alignmentDelta;
void* reservationBase = OSAllocator::reserveUncommitted(reservationSize, usage, writable, false);
// Select an aligned region within the reservation and commit.
void* alignedBase = reinterpret_cast<uintptr_t>(reservationBase) & alignmentMask
? reinterpret_cast<void*>((reinterpret_cast<uintptr_t>(reservationBase) & ~alignmentMask) + alignment)
: reservationBase;
OSAllocator::commit(alignedBase, size, writable, false);
return PageAllocationAligned(alignedBase, size, reservationBase, reservationSize);
#endif
}
void pm_free_page(addr_t addr)
{
if(!paging_enabled)
panic(PANIC_MEM | PANIC_NOSYNC, "Called free page without paging environment");
if(addr < pm_location || (((addr > highest_page) || addr < lowest_page)
&& memory_has_been_mapped)) {
panic(PANIC_MEM | PANIC_NOSYNC, "tried to free invalic physical address");
return;
}
if(current_task) current_task->freed++;
mutex_on(&pm_mutex);
if(pm_stack_max <= pm_stack)
{
vm_map(pm_stack_max, addr, PAGE_PRESENT | PAGE_WRITE, MAP_CRIT);
memset((void *)pm_stack_max, 0, PAGE_SIZE);
pm_stack_max += PAGE_SIZE;
} else {
*(unsigned int *)(pm_stack) = addr;
pm_stack += sizeof(unsigned int);
--pm_used_pages;
}
if(current_task && current_task->num_pages)
current_task->num_pages--;
mutex_off(&pm_mutex);
}
static void test_vm_unmap(void)
{
paddr_t paddr;
vaddr_t vaddr;
vm_as_t* as;
as = vm_get_kernel_as();
serial_printl("[?] test unmap\n");
vaddr = (vaddr_t)0xb0000000;
paddr = phys_alloc();
vm_map(as, vaddr, paddr);
/* no page fault
*/
*(uint32_t*)vaddr = 0x2a;
vm_unmap(as, vaddr);
phys_free(paddr);
/* page fault
*/
*(uint32_t*)vaddr = 0x2a;
}
void
vfork_end(struct proc *p)
{
void *stack;
task_t self = task_self();
DPRINTF(("vfork_end: org=%x saved=%x\n", p->p_stackbase,
p->p_stacksaved));
/*
* Restore parent's stack
*/
if (vm_map(p->p_task, p->p_stackbase, USTACK_SIZE, &stack) != 0)
return;
memcpy(stack, p->p_stacksaved, USTACK_SIZE);
vm_free(self, p->p_stacksaved);
vm_free(self, stack);
/*
* Resume parent
*/
p->p_vforked = 0;
task_resume(p->p_task);
}
void
attack(void)
{
object_t *objp = (object_t *)random();
object_t obj = (object_t)random();
char *name = (char *)random();
void *msg = (void *)random();
size_t size = (size_t)random();
task_t self = task_self();
void *addr = (void *)random();
int attr = random() & 7;
thread_t t = (thread_t)random();
thread_t *tp = (thread_t *)random();
object_create(NULL, NULL);
object_create(NULL, objp);
object_create(name, NULL);
object_create(name, objp);
object_destroy(0);
object_destroy(obj);
object_lookup(NULL, objp);
object_lookup(name, NULL);
object_lookup(name, objp);
msg_send(0, msg, size);
msg_send(obj, NULL, size);
msg_send(obj, msg, 0);
msg_send(0, msg, 0);
msg_send(0, NULL, size);
msg_send(obj, msg, size);
msg_receive(0, msg, size);
msg_receive(obj, NULL, size);
msg_receive(obj, msg, 0);
msg_receive(0, msg, 0);
msg_receive(0, NULL, size);
msg_receive(obj, msg, size);
msg_reply(0, msg, size);
msg_reply(obj, NULL, size);
msg_reply(obj, msg, 0);
msg_reply(0, msg, 0);
msg_reply(0, NULL, size);
msg_reply(obj, msg, size);
vm_allocate(self, addr, size, 1);
vm_allocate(self, &addr, size, 1);
vm_free(self, addr);
vm_attribute(self, addr, attr);
vm_map(self, addr, size, &addr);
thread_create(self, tp);
thread_suspend(t);
thread_terminate(t);
}
int
xf86ReadBIOS(unsigned long Base,unsigned long Offset,unsigned char *Buf,int Len)
{
mach_port_t device,iopl_dev;
memory_object_t iopl_mem;
vm_address_t addr = (vm_address_t)0; /* serach starting address */
kern_return_t err;
err = get_privileged_ports (NULL, &device);
if( err )
{
errno = err;
FatalError("xf86ReadBIOS() can't get_privileged_ports. (%s)\n",strerror(errno));
}
err = device_open(device,D_READ|D_WRITE,"iopl",&iopl_dev);
mach_port_deallocate (mach_task_self (), device);
if( err )
{
errno = err;
FatalError("xf86ReadBIOS() can't device_open. (%s)\n",strerror(errno));
}
err = device_map(iopl_dev,VM_PROT_READ|VM_PROT_WRITE, Base , BIOS_SIZE ,&iopl_mem,0);
if( err )
{
errno = err;
FatalError("xf86ReadBIOS() can't device_map. (%s)\n",strerror(errno));
}
err = vm_map(mach_task_self(),
&addr,
BIOS_SIZE,
0,
TRUE,
iopl_mem,
Base,
FALSE,
VM_PROT_READ|VM_PROT_WRITE,
VM_PROT_READ|VM_PROT_WRITE,
VM_INHERIT_SHARE);
mach_port_deallocate(mach_task_self(),iopl_mem);
if( err )
{
errno = err;
FatalError("xf86ReadBIOS() can't vm_map. (%s)\n",strerror(errno));
}
memcpy(Buf,(void*)((int)addr + Offset), Len);
err = vm_deallocate(mach_task_self(), addr, BIOS_SIZE);
if( err )
{
errno = err;
FatalError("xf86ReadBIOS() can't vm_deallocate. (%s)\n",strerror(errno));
}
return Len;
}
/*
* Contiguous space allocator for items of less than a page size.
*/
static union header *
kget_space(vm_offset_t allocsize)
{
vm_size_t space_to_add = 0;
vm_offset_t new_space = 0;
union header *addr;
while (kalloc_next_space + allocsize > kalloc_end_of_space) {
/*
* Add at least one page to allocation area.
*/
space_to_add = round_page(allocsize);
if (new_space == 0) {
/*
* Allocate memory.
* Try to make it contiguous with the last
* allocation area.
*/
new_space = kalloc_end_of_space;
if (vm_map(mach_task_self(),
&new_space, space_to_add, (vm_offset_t) 0, TRUE,
MEMORY_OBJECT_NULL, (vm_offset_t) 0, FALSE,
VM_PROT_DEFAULT, VM_PROT_ALL, VM_INHERIT_DEFAULT)
!= KERN_SUCCESS)
return 0;
continue;
}
/*
* Memory was allocated in a previous iteration.
* Check whether the new region is contiguous with the
* old one.
*/
if (new_space != kalloc_end_of_space) {
/*
* Throw away the remainder of the old space,
* and start a new one.
*/
kalloc_wasted_space +=
kalloc_end_of_space - kalloc_next_space;
kalloc_next_space = new_space;
}
kalloc_end_of_space = new_space + space_to_add;
new_space = 0;
}
addr = (union header *)kalloc_next_space;
kalloc_next_space += allocsize;
if (new_space != 0)
(void) vm_deallocate(mach_task_self(), new_space, space_to_add);
return addr;
}
void _start() {
vm_map(0x10000000, (0xAFFFF - 0xA0000), 0xA0000);
busy_wait(0x5FFFFFF);
uint8_t color = 0;
do {
fill_screen(color);
busy_wait(0xFFFFF);
color = (color++) % 255;
} while(1);
}
/**************************************************************************
* Video Memory Mapping section
***************************************************************************/
static pointer
mapVidMem(int ScreenNum, unsigned long Base, unsigned long Size, int Flags)
{
mach_port_t device,iopl_dev;
memory_object_t iopl_mem;
kern_return_t err;
vm_address_t addr=(vm_address_t)0;
err = get_privileged_ports (NULL, &device);
if( err )
{
errno = err;
FatalError("xf86MapVidMem() can't get_privileged_ports. (%s)\n",strerror(errno));
}
err = device_open(device,D_READ|D_WRITE,"iopl",&iopl_dev);
mach_port_deallocate (mach_task_self(), device);
if( err )
{
errno = err;
FatalError("xf86MapVidMem() can't device_open. (%s)\n",strerror(errno));
}
err = device_map(iopl_dev,VM_PROT_READ|VM_PROT_WRITE, Base , Size ,&iopl_mem,0);
if( err )
{
errno = err;
FatalError("xf86MapVidMem() can't device_map. (%s)\n",strerror(errno));
}
err = vm_map(mach_task_self(),
&addr,
Size,
0, /* mask */
TRUE, /* anywhere */
iopl_mem,
(vm_offset_t)Base,
FALSE, /* copy on write */
VM_PROT_READ|VM_PROT_WRITE,
VM_PROT_READ|VM_PROT_WRITE,
VM_INHERIT_SHARE);
mach_port_deallocate(mach_task_self(),iopl_mem);
if( err )
{
errno = err;
FatalError("xf86MapVidMem() can't vm_map.(iopl_mem) (%s)\n",strerror(errno));
}
mach_port_deallocate(mach_task_self(),iopl_dev);
if( err )
{
errno = err;
FatalError("xf86MapVidMem() can't mach_port_deallocate.(iopl_dev) (%s)\n",strerror(errno));
}
return (pointer)addr;
}
void setup_thread(thread_params_t *params)
{
context_t user_context;
uint32_t phys_page;
int i;
interrupt_status_t intr_status;
thread_table_t *thread= thread_get_current_thread_entry();
/* Copy thread parameters. */
int arg = params->arg;
void (*func)(int) = params->func;
process_id_t pid = thread->process_id = params->pid;
thread->pagetable = params->pagetable;
params->done = 1; /* OK, we don't need params any more. */
intr_status = _interrupt_disable();
spinlock_acquire(&process_table_slock);
/* Set up userspace environment. */
memoryset(&user_context, 0, sizeof(user_context));
user_context.cpu_regs[MIPS_REGISTER_A0] = arg;
user_context.pc = (uint32_t)func;
/* Allocate thread stack */
if (process_table[pid].bot_free_stack != 0) {
/* Reuse old thread stack. */
user_context.cpu_regs[MIPS_REGISTER_SP] =
process_table[pid].bot_free_stack
+ CONFIG_USERLAND_STACK_SIZE*PAGE_SIZE
- 4; /* Space for the thread argument */
process_table[pid].bot_free_stack =
*(uint32_t*)process_table[pid].bot_free_stack;
} else {
/* Allocate physical pages (frames) for the stack. */
for (i = 0; i < CONFIG_USERLAND_STACK_SIZE; i++) {
phys_page = pagepool_get_phys_page();
KERNEL_ASSERT(phys_page != 0);
vm_map(thread->pagetable, phys_page,
process_table[pid].stack_end - (i+1)*PAGE_SIZE, 1);
}
user_context.cpu_regs[MIPS_REGISTER_SP] =
process_table[pid].stack_end-4; /* Space for the thread argument */
process_table[pid].stack_end -= PAGE_SIZE*CONFIG_USERLAND_STACK_SIZE;
}
tlb_fill(thread->pagetable);
spinlock_release(&process_table_slock);
_interrupt_set_state(intr_status);
thread_goto_userland(&user_context);
}
static void test_page_fault(void)
{
volatile uint32_t* vaddr_0 = (uint32_t*)0xb0000000;
volatile uint32_t* vaddr_1 = (uint32_t*)0xb0001000;
volatile uint32_t* vaddr_2 = (uint32_t*)0xb0002000;
paddr_t paddr = phys_alloc();
vm_as_t* as;
as = vm_get_kernel_as();
vm_map(as, (vaddr_t)vaddr_0, paddr);
vm_map(as, (vaddr_t)vaddr_1, paddr);
vm_map(as, (vaddr_t)vaddr_2, paddr);
*vaddr_0 = 0x2a;
if ((*vaddr_0 != 0x2a) ||
(*vaddr_0 != *vaddr_1) ||
(*vaddr_0 != *vaddr_2))
BUG();
}
int main(int argc, char *argv[])
{
kern_return_t kr;
mach_port_t netmemory_server;
mach_port_t netmemory_object;
mach_port_t memory_object;
netname_name_t netmemory_object_name;
vm_size_t size=vm_page_size;
vm_address_t address;
boolean_t anywhere=TRUE;
//
kr=netname_look_up(name_server_port,
"",
"netmemoryserver",
&netmemory_server);
if (kr)
return kr;
//
kr=netname_look_up(name_server_port,"",
netmemory_object_name,
&netmemory_object);
if (kr)
{
printf("netmemory is not find\n");
return kr;
}
//
kr=netmemory_cache(netmemory_server,
netmemory_object,
&memory_object);
if (kr)
return kr;
//
kr=vm_map(mach_task_self(),&address,size,0,anywhere,
memory_object,0,FALSE,VM_PROT_DEFAULT,
VM_PROT_DEFAULT,VM_INHERIT_SHARE);
if (kr)
return kr;
//
printf("address is %d\n",address);
//address=100;
return 0;
}
/* Create a new pager in USER_PAGER with NPAGES pages, and return a
mapping to the memory in *USER. */
error_t
user_pager_create (struct user_pager *user_pager, unsigned int npages,
struct cons_display **user)
{
error_t err;
struct user_pager_info *upi;
upi = calloc (1, sizeof (struct user_pager_info)
+ sizeof (vm_address_t) * npages);
if (!upi)
return errno;
upi->memobj_npages = npages;
/* XXX Are the values 1 and MEMORY_OBJECT_COPY_DELAY correct? */
user_pager->pager = pager_create (upi, pager_bucket,
1, MEMORY_OBJECT_COPY_DELAY, 0);
if (!user_pager->pager)
{
free (upi);
return errno;
}
user_pager->memobj = pager_get_port (user_pager->pager);
ports_port_deref (user_pager->pager);
mach_port_insert_right (mach_task_self (), user_pager->memobj,
user_pager->memobj, MACH_MSG_TYPE_MAKE_SEND);
*user = 0;
err = vm_map (mach_task_self (),
(vm_address_t *) user,
(vm_size_t) npages * vm_page_size,
(vm_address_t) 0,
1 /* ! (flags & MAP_FIXED) */,
user_pager->memobj, 0 /* (vm_offset_t) offset */,
0 /* ! (flags & MAP_SHARED) */,
VM_PROT_READ | VM_PROT_WRITE,
VM_PROT_READ | VM_PROT_WRITE,
VM_INHERIT_NONE);
if (err)
{
/* UPI will be cleaned up by libpager. */
mach_port_deallocate (mach_task_self (), user_pager->memobj);
return err;
}
return 0;
}
void* syscall_memlimit(void* new_end){
uint32_t phys_page;
interrupt_status_t intr_status;
intr_status = _interrupt_disable();
spinlock_acquire(&process_table_slock);
process_control_block_t *curr_proc = process_get_current_process_entry();
// checks if the new_end is NULL
if (new_end == NULL){
spinlock_release(&process_table_slock);
_interrupt_set_state(intr_status);
return (void*)curr_proc->heap_end;
}
// checks if we are trying to shrink the memory, if so we error
// and return NULL
if ((uint32_t)new_end < curr_proc->heap_end){
spinlock_release(&process_table_slock);
_interrupt_set_state(intr_status);
return NULL;
}
// loop where we alloc the physical and the virtual memoery, we also check
// if we have enough physical memory.
for (uint32_t i = (curr_proc->heap_end / PAGE_SIZE +1);
i <= ((uint32_t)new_end / PAGE_SIZE);i++){
// we allocate some physical memory.
phys_page = physmem_allocblock();
//check if we got enough physical memory, if we do not we error
if (phys_page == 0){
spinlock_release(&process_table_slock);
_interrupt_set_state(intr_status);
return NULL;
}
// maps the virtual memory
vm_map(thread_get_current_thread_entry()->pagetable,
phys_page, i * PAGE_SIZE, 1);
}
// if nothing fails we at last set heap_end to new_end
curr_proc->heap_end = (uint32_t)new_end;
spinlock_release(&process_table_slock);
_interrupt_set_state(intr_status);
return new_end;
}
bool AUMCircularBufferInit(AUMCircularBuffer *buffer, int length) {
buffer->length = round_page(length); // We need whole page sizes
// Temporarily allocate twice the length, so we have the contiguous address space to
// support a second instance of the buffer directly after
vm_address_t bufferAddress;
if ( !checkResult(vm_allocate(mach_task_self(), &bufferAddress, buffer->length * 2, TRUE /* (don't use the current bufferAddress value) */),
"Buffer allocation") ) return false;
// Now replace the second half of the allocation with a virtual copy of the first half. Deallocate the second half...
if ( !checkResult(vm_deallocate(mach_task_self(), bufferAddress + buffer->length, buffer->length),
"Buffer deallocation") ) return false;
// Then create a memory entry that refers to the buffer
vm_size_t entry_length = buffer->length;
mach_port_t memoryEntry;
if ( !checkResult(mach_make_memory_entry(mach_task_self(), &entry_length, bufferAddress, VM_PROT_READ|VM_PROT_WRITE, &memoryEntry, 0),
"Create memory entry") ) {
vm_deallocate(mach_task_self(), bufferAddress, buffer->length);
return false;
}
// And map the memory entry to the address space immediately after the buffer
vm_address_t virtualAddress = bufferAddress + buffer->length;
if ( !checkResult(vm_map(mach_task_self(), &virtualAddress, buffer->length, 0, FALSE, memoryEntry, 0, FALSE, VM_PROT_READ | VM_PROT_WRITE, VM_PROT_READ | VM_PROT_WRITE, VM_INHERIT_DEFAULT),
"Map buffer memory") ) {
vm_deallocate(mach_task_self(), bufferAddress, buffer->length);
return false;
}
if ( virtualAddress != bufferAddress+buffer->length ) {
printf("Couldn't map buffer memory to end of buffer\n");
vm_deallocate(mach_task_self(), virtualAddress, buffer->length);
vm_deallocate(mach_task_self(), bufferAddress, buffer->length);
return false;
}
buffer->buffer = (void*)bufferAddress;
buffer->fillCount = 0;
buffer->head = buffer->tail = 0;
return true;
}
void* allocateJSBlock(unsigned size)
{
// See: JavaScriptCore/runtime/Collector.cpp
OLYMPIA_ASSERT(size == BLOCK_SIZE);
#if defined(OLYMPIA_LINUX) //|| defined(OLYMPIA_MAC)
#if ENABLE(JSC_MULTIPLE_THREADS)
#error Need to initialize pagesize safely.
#endif
static size_t pagesize = pageSize();
size_t extra = 0;
if (BLOCK_SIZE > pagesize)
extra = BLOCK_SIZE - pagesize;
void* mmapResult = mmap(NULL, BLOCK_SIZE + extra, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON, -1, 0);
uintptr_t address = reinterpret_cast<uintptr_t>(mmapResult);
size_t adjust = 0;
if ((address & BLOCK_OFFSET_MASK) != 0)
adjust = BLOCK_SIZE - (address & BLOCK_OFFSET_MASK);
if (adjust > 0)
munmap(reinterpret_cast<char*>(address), adjust);
if (adjust < extra)
munmap(reinterpret_cast<char*>(address + adjust + BLOCK_SIZE), extra - adjust);
address += adjust;
return reinterpret_cast<void*>(address);
#elif defined(OLYMPIA_WINDOWS)
#if COMPILER(MINGW) && !COMPILER(MINGW64)
void* address = __mingw_aligned_malloc(BLOCK_SIZE, BLOCK_SIZE);
#else
void* address = _aligned_malloc(BLOCK_SIZE, BLOCK_SIZE);
#endif
memset(address, 0, BLOCK_SIZE);
return address;
#elif defined(OLYMPIA_MAC)
vm_address_t address = 0;
vm_map(current_task(), &address, BLOCK_SIZE, BLOCK_OFFSET_MASK, VM_FLAGS_ANYWHERE | VM_TAG_FOR_COLLECTOR_MEMORY, MEMORY_OBJECT_NULL, 0, FALSE, VM_PROT_DEFAULT, VM_PROT_DEFAULT, VM_INHERIT_DEFAULT);
return reinterpret_cast<void*>(address);
#endif
}
void
diskfs_start_disk_pager (struct user_pager_info *upi,
struct port_bucket *pager_bucket, int may_cache,
size_t size, void **image)
{
error_t err;
mach_port_t disk_pager_port;
/* Make a thread to service paging requests. */
cthread_detach (cthread_fork ((cthread_fn_t) service_paging_requests,
(any_t)pager_bucket));
/* Create the pager. */
diskfs_disk_pager = pager_create (upi, pager_bucket,
may_cache, MEMORY_OBJECT_COPY_NONE);
assert (diskfs_disk_pager);
/* Get a port to the disk pager. */
disk_pager_port = pager_get_port (diskfs_disk_pager);
mach_port_insert_right (mach_task_self (), disk_pager_port, disk_pager_port,
MACH_MSG_TYPE_MAKE_SEND);
/* Now map the disk image. */
err = vm_map (mach_task_self (), (vm_address_t *)image, size,
0, 1, disk_pager_port, 0, 0,
VM_PROT_READ | (diskfs_readonly ? 0 : VM_PROT_WRITE),
VM_PROT_READ | VM_PROT_WRITE,
VM_INHERIT_NONE);
if (err)
error (2, err, "cannot vm_map whole disk");
/* Set up the signal preemptor to catch faults on the disk image. */
preemptor.first = (vm_address_t) *image;
preemptor.last = ((vm_address_t) *image + size);
hurd_preempt_signals (&preemptor);
/* We have the mapping; we no longer need the send right. */
mach_port_deallocate (mach_task_self (), disk_pager_port);
}
static void
vproc_shmem_init(void)
{
vm_address_t vm_addr = 0;
mach_port_t shmem_port;
kern_return_t kr;
kr = vproc_mig_setup_shmem(bootstrap_port, &shmem_port);
//assert(kr == 0);
if (kr) {
/* rdar://problem/6416724
* If we fail to set up a shared memory page, just allocate a local chunk
* of memory. This way, processes can still introspect their own transaction
* counts if they're being run under a debugger. Moral of the story: Debug
* from the environment you intend to run in.
*/
void *_vm_addr = calloc(1, sizeof(struct vproc_shmem_s));
if( !_vm_addr ) {
return;
}
vm_addr = (vm_address_t)_vm_addr;
} else {
kr = vm_map(mach_task_self(), &vm_addr, getpagesize(), 0, true, shmem_port, 0, false,
VM_PROT_READ|VM_PROT_WRITE, VM_PROT_READ|VM_PROT_WRITE, VM_INHERIT_NONE);
//assert(kr == 0);
if (kr) return;
kr = mach_port_deallocate(mach_task_self(), shmem_port);
//assert(kr == 0);
}
vproc_shmem = (struct vproc_shmem_s *)vm_addr;
}
static int
vfork_start(struct proc *p)
{
void *stack;
task_t self = task_self();
/*
* Save parent's stack
*/
if (vm_map(p->p_task, p->p_stackbase, USTACK_SIZE, &stack) != 0)
return ENOMEM;
if (vm_allocate(self, &p->p_stacksaved, USTACK_SIZE, 1) != 0)
return ENOMEM;
memcpy(p->p_stacksaved, stack, USTACK_SIZE);
vm_free(self, stack);
p->p_vforked = 1;
DPRINTF(("vfork_start: saved=%x org=%x\n", p->p_stacksaved,
p->p_stackbase));
return 0;
}
/* Implement the diskfs_lookup callback from the diskfs library. See
<hurd/diskfs.h> for the interface specification. */
error_t
diskfs_lookup_hard (struct node *dp, const char *name, enum lookup_type type,
struct node **npp, struct dirstat *ds, struct protid *cred)
{
error_t err;
ino_t inum;
int namelen;
int spec_dotdot;
struct node *np = 0;
int retry_dotdot = 0;
vm_prot_t prot =
(type == LOOKUP) ? VM_PROT_READ : (VM_PROT_READ | VM_PROT_WRITE);
memory_object_t memobj;
vm_address_t buf = 0;
vm_size_t buflen = 0;
int blockaddr;
int idx, lastidx;
int looped;
if ((type == REMOVE) || (type == RENAME))
assert (npp);
if (npp)
*npp = 0;
spec_dotdot = type & SPEC_DOTDOT;
type &= ~SPEC_DOTDOT;
namelen = strlen (name);
if (namelen > FAT_NAME_MAX)
return ENAMETOOLONG;
try_again:
if (ds)
{
ds->type = LOOKUP;
ds->mapbuf = 0;
ds->mapextent = 0;
}
if (buf)
{
munmap ((caddr_t) buf, buflen);
buf = 0;
}
if (ds && (type == CREATE || type == RENAME))
ds->stat = LOOKING;
/* Map in the directory contents. */
memobj = diskfs_get_filemap (dp, prot);
if (memobj == MACH_PORT_NULL)
return errno;
buf = 0;
/* We allow extra space in case we have to do an EXTEND. */
buflen = round_page (dp->dn_stat.st_size + DIRBLKSIZ);
err = vm_map (mach_task_self (),
&buf, buflen, 0, 1, memobj, 0, 0, prot, prot, 0);
mach_port_deallocate (mach_task_self (), memobj);
inum = 0;
diskfs_set_node_atime (dp);
/* Start the lookup at DP->dn->dir_idx. */
idx = dp->dn->dir_idx;
if (idx << LOG2_DIRBLKSIZ > dp->dn_stat.st_size)
idx = 0; /* just in case */
blockaddr = buf + (idx << LOG2_DIRBLKSIZ);
looped = (idx == 0);
lastidx = idx;
if (lastidx == 0)
lastidx = dp->dn_stat.st_size >> LOG2_DIRBLKSIZ;
while (!looped || idx < lastidx)
{
err = dirscanblock (blockaddr, dp, idx, name, namelen, type, ds, &inum);
if (!err)
{
dp->dn->dir_idx = idx;
break;
}
if (err != ENOENT)
{
munmap ((caddr_t) buf, buflen);
return err;
}
blockaddr += DIRBLKSIZ;
idx++;
if (blockaddr - buf >= dp->dn_stat.st_size && !looped)
{
/* We've gotten to the end; start back at the beginning. */
looped = 1;
blockaddr = buf;
idx = 0;
}
}
diskfs_set_node_atime (dp);
if (diskfs_synchronous)
diskfs_node_update (dp, 1);
/* If err is set here, it's ENOENT, and we don't want to
think about that as an error yet. */
err = 0;
if (inum && npp)
{
if (namelen != 2 || name[0] != '.' || name[1] != '.')
{
if (inum == dp->cache_id)
{
np = dp;
diskfs_nref (np);
}
else
{
err = diskfs_cached_lookup_in_dirbuf (inum, &np, buf);
if (err)
goto out;
}
}
/* We are looking up "..". */
/* Check to see if this is the root of the filesystem. */
else if (dp == diskfs_root_node)
{
err = EAGAIN;
goto out;
}
/* We can't just do diskfs_cached_lookup, because we would then
deadlock. So we do this. Ick. */
else if (retry_dotdot)
{
/* Check to see that we got the same answer as last time. */
if (inum != retry_dotdot)
{
/* Drop what we *thought* was .. (but isn't any more) and
try *again*. */
diskfs_nput (np);
mutex_unlock (&dp->lock);
err = diskfs_cached_lookup_in_dirbuf (inum, &np, buf);
mutex_lock (&dp->lock);
if (err)
goto out;
retry_dotdot = inum;
goto try_again;
}
/* Otherwise, we got it fine and np is already set properly. */
}
else if (!spec_dotdot)
{
/* Lock them in the proper order, and then
repeat the directory scan to see if this is still
right. */
mutex_unlock (&dp->lock);
err = diskfs_cached_lookup_in_dirbuf (inum, &np, buf);
mutex_lock (&dp->lock);
if (err)
goto out;
retry_dotdot = inum;
goto try_again;
}
/* Here below are the spec dotdot cases. */
else if (type == RENAME || type == REMOVE)
np = ifind (inum);
else if (type == LOOKUP)
{
diskfs_nput (dp);
err = diskfs_cached_lookup_in_dirbuf (inum, &np, buf);
if (err)
goto out;
}
else
assert (0);
}
if ((type == CREATE || type == RENAME) && !inum && ds && ds->stat == LOOKING)
{
/* We didn't find any room, so mark ds to extend the dir. */
ds->type = CREATE;
ds->stat = EXTEND;
ds->idx = dp->dn_stat.st_size >> LOG2_DIRBLKSIZ;
}
static void*
commpage_allocate(
vm_map_t submap, // commpage32_map or commpage_map64
size_t area_used, // _COMM_PAGE32_AREA_USED or _COMM_PAGE64_AREA_USED
vm_prot_t uperm)
{
vm_offset_t kernel_addr = 0; // address of commpage in kernel map
vm_offset_t zero = 0;
vm_size_t size = area_used; // size actually populated
vm_map_entry_t entry;
ipc_port_t handle;
kern_return_t kr;
if (submap == NULL)
panic("commpage submap is null");
if ((kr = vm_map(kernel_map,
&kernel_addr,
area_used,
0,
VM_FLAGS_ANYWHERE | VM_MAKE_TAG(VM_KERN_MEMORY_OSFMK),
NULL,
0,
FALSE,
VM_PROT_ALL,
VM_PROT_ALL,
VM_INHERIT_NONE)))
panic("cannot allocate commpage %d", kr);
if ((kr = vm_map_wire(kernel_map,
kernel_addr,
kernel_addr+area_used,
VM_PROT_DEFAULT|VM_PROT_MEMORY_TAG_MAKE(VM_KERN_MEMORY_OSFMK),
FALSE)))
panic("cannot wire commpage: %d", kr);
/*
* Now that the object is created and wired into the kernel map, mark it so that no delay
* copy-on-write will ever be performed on it as a result of mapping it into user-space.
* If such a delayed copy ever occurred, we could remove the kernel's wired mapping - and
* that would be a real disaster.
*
* JMM - What we really need is a way to create it like this in the first place.
*/
if (!(kr = vm_map_lookup_entry( kernel_map, vm_map_trunc_page(kernel_addr, VM_MAP_PAGE_MASK(kernel_map)), &entry) || entry->is_sub_map))
panic("cannot find commpage entry %d", kr);
VME_OBJECT(entry)->copy_strategy = MEMORY_OBJECT_COPY_NONE;
if ((kr = mach_make_memory_entry( kernel_map, // target map
&size, // size
kernel_addr, // offset (address in kernel map)
uperm, // protections as specified
&handle, // this is the object handle we get
NULL ))) // parent_entry (what is this?)
panic("cannot make entry for commpage %d", kr);
if ((kr = vm_map_64( submap, // target map (shared submap)
&zero, // address (map into 1st page in submap)
area_used, // size
0, // mask
VM_FLAGS_FIXED, // flags (it must be 1st page in submap)
handle, // port is the memory entry we just made
0, // offset (map 1st page in memory entry)
FALSE, // copy
uperm, // cur_protection (R-only in user map)
uperm, // max_protection
VM_INHERIT_SHARE ))) // inheritance
panic("cannot map commpage %d", kr);
ipc_port_release(handle);
/* Make the kernel mapping non-executable. This cannot be done
* at the time of map entry creation as mach_make_memory_entry
* cannot handle disjoint permissions at this time.
*/
kr = vm_protect(kernel_map, kernel_addr, area_used, FALSE, VM_PROT_READ | VM_PROT_WRITE);
assert (kr == KERN_SUCCESS);
return (void*)(intptr_t)kernel_addr; // return address in kernel map
}
/**
* Starts one userland process. The thread calling this function will
* be used to run the process and will therefore never return from
* this function. This function asserts that no errors occur in
* process startup (the executable file exists and is a valid ecoff
* file, enough memory is available, file operations succeed...).
* Therefore this function is not suitable to allow startup of
* arbitrary processes.
*
* @executable The name of the executable to be run in the userland
* process
*/
void process_start(uint32_t pid)
{
thread_table_t *my_entry;
pagetable_t *pagetable;
uint32_t phys_page;
context_t user_context;
uint32_t stack_bottom;
elf_info_t elf;
openfile_t file;
const char* executable;
int i;
interrupt_status_t intr_status;
my_entry = thread_get_current_thread_entry();
my_entry->process_id = pid;
executable = process_table[pid].executable;
/* If the pagetable of this thread is not NULL, we are trying to
run a userland process for a second time in the same thread.
This is not possible. */
KERNEL_ASSERT(my_entry->pagetable == NULL);
pagetable = vm_create_pagetable(thread_get_current_thread());
KERNEL_ASSERT(pagetable != NULL);
intr_status = _interrupt_disable();
my_entry->pagetable = pagetable;
_interrupt_set_state(intr_status);
file = vfs_open((char *)executable);
/* Make sure the file existed and was a valid ELF file */
KERNEL_ASSERT(file >= 0);
KERNEL_ASSERT(elf_parse_header(&elf, file));
/* Trivial and naive sanity check for entry point: */
KERNEL_ASSERT(elf.entry_point >= PAGE_SIZE);
/* Calculate the number of pages needed by the whole process
(including userland stack). Since we don't have proper tlb
handling code, all these pages must fit into TLB. */
KERNEL_ASSERT(elf.ro_pages + elf.rw_pages + CONFIG_USERLAND_STACK_SIZE
<= _tlb_get_maxindex() + 1);
/* Allocate and map stack */
for(i = 0; i < CONFIG_USERLAND_STACK_SIZE; i++) {
phys_page = pagepool_get_phys_page();
KERNEL_ASSERT(phys_page != 0);
vm_map(my_entry->pagetable, phys_page,
(USERLAND_STACK_TOP & PAGE_SIZE_MASK) - i*PAGE_SIZE, 1);
}
/* Allocate and map pages for the segments. We assume that
segments begin at page boundary. (The linker script in tests
directory creates this kind of segments) */
for(i = 0; i < (int)elf.ro_pages; i++) {
phys_page = pagepool_get_phys_page();
KERNEL_ASSERT(phys_page != 0);
vm_map(my_entry->pagetable, phys_page,
elf.ro_vaddr + i*PAGE_SIZE, 1);
}
for(i = 0; i < (int)elf.rw_pages; i++) {
phys_page = pagepool_get_phys_page();
KERNEL_ASSERT(phys_page != 0);
vm_map(my_entry->pagetable, phys_page,
elf.rw_vaddr + i*PAGE_SIZE, 1);
}
/* Put the mapped pages into TLB. Here we again assume that the
pages fit into the TLB. After writing proper TLB exception
handling this call should be skipped. */
//intr_status = _interrupt_disable();
//tlb_fill(my_entry->pagetable);
//_interrupt_set_state(intr_status);
/* Now we may use the virtual addresses of the segments. */
/* Zero the pages. */
memoryset((void *)elf.ro_vaddr, 0, elf.ro_pages*PAGE_SIZE);
memoryset((void *)elf.rw_vaddr, 0, elf.rw_pages*PAGE_SIZE);
stack_bottom = (USERLAND_STACK_TOP & PAGE_SIZE_MASK) -
(CONFIG_USERLAND_STACK_SIZE-1)*PAGE_SIZE;
memoryset((void *)stack_bottom, 0, CONFIG_USERLAND_STACK_SIZE*PAGE_SIZE);
/* Copy segments */
if (elf.ro_size > 0) {
/* Make sure that the segment is in proper place. */
KERNEL_ASSERT(elf.ro_vaddr >= PAGE_SIZE);
KERNEL_ASSERT(vfs_seek(file, elf.ro_location) == VFS_OK);
KERNEL_ASSERT(vfs_read(file, (void *)elf.ro_vaddr, elf.ro_size)
== (int)elf.ro_size);
}
if (elf.rw_size > 0) {
/* Make sure that the segment is in proper place. */
KERNEL_ASSERT(elf.rw_vaddr >= PAGE_SIZE);
KERNEL_ASSERT(vfs_seek(file, elf.rw_location) == VFS_OK);
KERNEL_ASSERT(vfs_read(file, (void *)elf.rw_vaddr, elf.rw_size)
== (int)elf.rw_size);
}
/* Set the dirty bit to zero (read-only) on read-only pages. */
for(i = 0; i < (int)elf.ro_pages; i++) {
vm_set_dirty(my_entry->pagetable, elf.ro_vaddr + i*PAGE_SIZE, 0);
}
/* Insert page mappings again to TLB to take read-only bits into use */
//intr_status = _interrupt_disable();
//tlb_fill(my_entry->pagetable);
//_interrupt_set_state(intr_status);
/* Initialize the user context. (Status register is handled by
thread_goto_userland) */
memoryset(&user_context, 0, sizeof(user_context));
user_context.cpu_regs[MIPS_REGISTER_SP] = USERLAND_STACK_TOP;
user_context.pc = elf.entry_point;
vfs_close(file);
thread_goto_userland(&user_context);
KERNEL_PANIC("thread_goto_userland failed.");
}
static
load_return_t
load_segment(
struct segment_command *scp,
void * pager,
off_t pager_offset,
off_t macho_size,
__unused off_t end_of_file,
vm_map_t map,
load_result_t *result
)
{
kern_return_t ret;
vm_offset_t map_addr, map_offset;
vm_size_t map_size, seg_size, delta_size;
vm_prot_t initprot;
vm_prot_t maxprot;
/*
* Make sure what we get from the file is really ours (as specified
* by macho_size).
*/
if (scp->fileoff + scp->filesize > macho_size)
return (LOAD_BADMACHO);
/*
* Make sure the segment is page-aligned in the file.
*/
if ((scp->fileoff & PAGE_MASK) != 0)
return LOAD_BADMACHO;
seg_size = round_page(scp->vmsize);
if (seg_size == 0)
return(KERN_SUCCESS);
/*
* Round sizes to page size.
*/
map_size = round_page(scp->filesize);
map_addr = trunc_page(scp->vmaddr);
#if 0 /* XXX (4596982) this interferes with Rosetta */
if (map_addr == 0 &&
map_size == 0 &&
seg_size != 0 &&
(scp->initprot & VM_PROT_ALL) == VM_PROT_NONE &&
(scp->maxprot & VM_PROT_ALL) == VM_PROT_NONE) {
/*
* This is a "page zero" segment: it starts at address 0,
* is not mapped from the binary file and is not accessible.
* User-space should never be able to access that memory, so
* make it completely off limits by raising the VM map's
* minimum offset.
*/
ret = vm_map_raise_min_offset(map, (vm_map_offset_t) seg_size);
if (ret != KERN_SUCCESS) {
return LOAD_FAILURE;
}
return LOAD_SUCCESS;
}
#endif
map_offset = pager_offset + scp->fileoff;
if (map_size > 0) {
initprot = (scp->initprot) & VM_PROT_ALL;
maxprot = (scp->maxprot) & VM_PROT_ALL;
/*
* Map a copy of the file into the address space.
*/
ret = vm_map(map,
&map_addr, map_size, (vm_offset_t)0,
VM_FLAGS_FIXED, pager, map_offset, TRUE,
initprot, maxprot,
VM_INHERIT_DEFAULT);
if (ret != KERN_SUCCESS)
return(LOAD_NOSPACE);
/*
* If the file didn't end on a page boundary,
* we need to zero the leftover.
*/
delta_size = map_size - scp->filesize;
#if FIXME
if (delta_size > 0) {
vm_offset_t tmp;
ret = vm_allocate(kernel_map, &tmp, delta_size, VM_FLAGS_ANYWHERE);
if (ret != KERN_SUCCESS)
return(LOAD_RESOURCE);
if (copyout(tmp, map_addr + scp->filesize,
delta_size)) {
(void) vm_deallocate(
kernel_map, tmp, delta_size);
return(LOAD_FAILURE);
}
(void) vm_deallocate(kernel_map, tmp, delta_size);
}
#endif /* FIXME */
}
/*
* If the virtual size of the segment is greater
* than the size from the file, we need to allocate
* zero fill memory for the rest.
*/
delta_size = seg_size - map_size;
if (delta_size > 0) {
vm_offset_t tmp = map_addr + map_size;
ret = vm_map(map, &tmp, delta_size, 0, VM_FLAGS_FIXED,
NULL, 0, FALSE,
scp->initprot, scp->maxprot,
VM_INHERIT_DEFAULT);
if (ret != KERN_SUCCESS)
return(LOAD_NOSPACE);
}
if ( (scp->fileoff == 0) && (scp->filesize != 0) )
result->mach_header = map_addr;
if (scp->flags & SG_PROTECTED_VERSION_1) {
ret = unprotect_segment_64((uint64_t) scp->fileoff,
(uint64_t) scp->filesize,
map,
(vm_map_offset_t) map_addr,
(vm_map_size_t) map_size);
} else {
ret = LOAD_SUCCESS;
}
return ret;
}
void
init_mapped_time(void)
{
kern_return_t kr;
mach_port_t pager;
int new_res;
tvalspec_t rtc_time;
kr = host_get_clock_service(host_port,
REALTIME_CLOCK,
&rt_clock);
if (kr != KERN_SUCCESS) {
MACH3_DEBUG(1, kr,
("init_mapped_time: "
"host_get_clock_service(REALTIME_CLOCK)"));
panic("unable to get real time clock");
}
kr = host_get_clock_control(privileged_host_port,
REALTIME_CLOCK,
&rt_clock_ctrl);
if (kr != KERN_SUCCESS) {
MACH3_DEBUG(1, kr,
("init_mapped_time: "
"host_get_clock_control(REALTIME_CLOCK)"));
} else {
/* ask for 500 microsecond resolution */
new_res = 500000;
#if 0
kr = clock_set_attributes(rt_clock_ctrl,
CLOCK_ALARM_CURRES,
(clock_attr_t) &new_res,
1);
if (kr != KERN_SUCCESS) {
MACH3_DEBUG(2, kr,
("init_mapped_time: "
"clock_set_attributes(%d nsec)",
new_res));
}
#endif
}
kr = clock_map_time(rt_clock, &pager);
if (kr != KERN_SUCCESS) {
MACH3_DEBUG(1, kr, ("init_mapped_time: clock_map_time"));
panic("unable to map real time clock");
}
kr = vm_map(mach_task_self(),
(vm_address_t *)&serv_mtime,
sizeof(mapped_tvalspec_t),
0,
TRUE,
pager,
0,
0,
VM_PROT_READ,
VM_PROT_READ,
VM_INHERIT_NONE);
if (kr != D_SUCCESS) {
MACH3_DEBUG(1, kr, ("init_mapped_time: vm_map"));
panic("unable to vm_map real time clock");
}
kr = mach_port_deallocate(mach_task_self(), pager);
if (kr != KERN_SUCCESS) {
MACH3_DEBUG(1, kr, ("init_mapped_time: mach_port_deallocate"));
panic("unable to deallocate pager");
}
/* calculate origin of rtclock (ie. time of boot) so that we
* can use rtclock to generate the current time
*/
kr = host_get_clock_service(host_port,
BATTERY_CLOCK,
&bb_clock);
if (kr != KERN_SUCCESS) {
MACH3_DEBUG(1, kr,
("init_mapped_time: "
"host_get_clock_service(BATTERY_CLOCK)"));
panic("unable to get battery backed clock");
}
kr = host_get_clock_control(privileged_host_port,
BATTERY_CLOCK,
&bb_clock_ctrl);
if (kr != KERN_SUCCESS) {
MACH3_DEBUG(1, kr,
("init_mapped_time: "
"host_get_clock_control(BATTERY_CLOCK)"));
return;
}
kr = clock_get_time(bb_clock, &base_time);
if (kr != KERN_SUCCESS) {
MACH3_DEBUG(1, kr, ("init_mapped_time: clock_get_time"));
}
MTS_TO_TS(serv_mtime, &rtc_time);
SUB_TVALSPEC(&base_time, &rtc_time);
}
void
osfmach3_insert_vm_struct(
struct mm_struct *mm,
struct vm_area_struct *vmp)
{
memory_object_t mem_obj;
vm_offset_t mem_obj_offset;
kern_return_t kr;
unsigned short vm_flags;
boolean_t is_shared;
vm_prot_t cur_prot, max_prot;
vm_address_t user_addr, wanted_addr;
vm_size_t size;
unsigned int id;
struct shmid_ds *shp;
struct osfmach3_mach_task_struct *mach_task;
extern struct shmid_ds *shm_segs[SHMMNI];
if (vmp->vm_flags & VM_REMAPPING) {
/* don't mess with Mach VM: it's only Linux remappings */
return;
}
#ifdef VMA_DEBUG
if (vma_debug) {
printk("VMA:osfmach3_insert_vm_struct: mm=0x%p, vmp=0x%p\n",
mm, vmp);
}
#endif /* VMA_DEBUG */
mach_task = mm->mm_mach_task;
if (vmp->vm_inode == NULL) {
if (vmp->vm_pte != 0) {
/* shared memory */
id = SWP_OFFSET(vmp->vm_pte) & SHM_ID_MASK;
shp = shm_segs[id];
if (shp != IPC_UNUSED) {
mem_obj = (mach_port_t) shp->shm_pages;
mem_obj_offset = 0;
} else {
mem_obj = MEMORY_OBJECT_NULL;
mem_obj_offset = 0;
}
} else {
mem_obj = MEMORY_OBJECT_NULL;
mem_obj_offset = 0;
}
} else if (S_ISREG(vmp->vm_inode->i_mode)) {
mem_obj = inode_pager_setup(vmp->vm_inode);
if (mem_obj == MEMORY_OBJECT_NULL) {
panic("osfmach3_insert_vm_struct: can't setup pager");
}
mem_obj_offset = (vm_offset_t) vmp->vm_offset;
} else if (vmp->vm_inode->i_mem_object != NULL) {
/* special file, but with a pager already setup */
mem_obj = vmp->vm_inode->i_mem_object->imo_mem_obj;
mem_obj_offset = (vm_offset_t) vmp->vm_offset;
} else {
panic("osfmach3_insert_vm_struct: non-regular file");
}
vm_flags = vmp->vm_flags;
cur_prot = VM_PROT_NONE;
if (vm_flags & VM_READ)
cur_prot |= VM_PROT_READ;
if (vm_flags & VM_WRITE)
cur_prot |= VM_PROT_WRITE;
if (vm_flags & VM_EXEC)
cur_prot |= VM_PROT_EXECUTE;
max_prot = VM_PROT_ALL;
is_shared = (vmp->vm_flags & VM_SHARED) != 0;
user_addr = vmp->vm_start;
wanted_addr = user_addr;
size = vmp->vm_end - vmp->vm_start;
#ifdef VMA_DEBUG
if (vma_debug) {
printk("VMA: vm_map(task=0x%x, user_addr=0x%x, size=0x%x, "
"mem_obj=0x%x, offset=0x%x, %sCOPY, cur_prot=0x%x, "
"max_prot=0x%x, %s)\n",
mach_task->mach_task_port,
user_addr,
size,
mem_obj,
mem_obj_offset,
is_shared ? "!" : "",
cur_prot,
max_prot,
is_shared ? "INHERIT_SHARE" : "INHERIT_COPY");
}
#endif /* VMA_DEBUG */
server_thread_blocking(FALSE);
kr = vm_map(mach_task->mach_task_port,
&user_addr,
size,
0, /* no mask */
FALSE, /* not anywhere */
mem_obj,
mem_obj_offset,
!is_shared,
cur_prot,
max_prot,
is_shared ? VM_INHERIT_SHARE : VM_INHERIT_COPY);
server_thread_unblocking(FALSE);
if (kr != KERN_SUCCESS) {
printk("Failure: vm_map(task=0x%x, user_addr=0x%x, size=0x%x, "
"mem_obj=0x%x, offset=0x%x, %sCOPY, cur_prot=0x%x, "
"max_prot=0x%x, %s)\n",
mach_task->mach_task_port,
user_addr,
size,
mem_obj,
mem_obj_offset,
is_shared ? "!" : "",
cur_prot,
max_prot,
is_shared ? "INHERIT_SHARE" : "INHERIT_COPY");
MACH3_DEBUG(1, kr, ("osfmach3_insert_vm_struct: vm_map"));
printk("osfmach3_insert_vm_struct: can't map\n");
}
if (user_addr != wanted_addr) {
printk("vm_map: mapped at 0x%x instead of 0x%x\n",
user_addr, wanted_addr);
printk("osfmach3_insert_vm_struct: mapping at wrong address\n");
}
if (vmp->vm_flags & VM_LOCKED) {
extern mach_port_t privileged_host_port;
server_thread_blocking(FALSE);
kr = vm_wire(privileged_host_port,
mach_task->mach_task_port,
user_addr,
size,
cur_prot);
server_thread_unblocking(FALSE);
if (kr != KERN_SUCCESS) {
MACH3_DEBUG(2, kr,
("osfmach3_insert_vm_struct: "
"vm_wire(task=0x%x, addr=0x%x, size=0x%x, "
"prot=0x%x)",
mach_task->mach_task_port,
user_addr,
size,
cur_prot));
printk("osfmach3_insert_vm_struct: vm_wire failed\n");
}
}
#if 0
if (vmp->vm_inode != NULL) {
/*
* If mem_obj was already cached in the kernel, we got an
* extra reference on its i_mem_object structure (inode_pager).
* If it was the first time we mapped the inode, the memory
* object has just been initialized by the kernel and we
* got a reference in memory_object_init(). In both cases,
* we have to release a reference.
*/
ASSERT(mem_obj != MEMORY_OBJECT_NULL);
ASSERT(vmp->vm_inode->i_mem_object);
ASSERT(vmp->vm_inode->i_mem_object->imo_mem_obj_control);
inode_pager_release(vmp->vm_inode);
}
#endif
}
/**
* Map pages starting at @a task_addr from @a task into the current process. The mapping
* will be copy-on-write, and will be checked to ensure a minimum protection value of
* VM_PROT_READ.
*
* @param task The task from which the memory will be mapped.
* @param task_addr The task-relative address of the memory to be mapped. This is not required to fall on a page boundry.
* @param length The total size of the mapping to create.
* @param require_full If false, short mappings will be permitted in the case where a memory object of the requested length
* does not exist at the target address. It is the caller's responsibility to validate the resulting length of the
* mapping, eg, using plcrash_async_mobject_remap_address() and similar. If true, and the entire requested page range is
* not valid, the mapping request will fail.
* @param[out] result The in-process address at which the pages were mapped.
* @param[out] result_length The total size, in bytes, of the mapped pages.
*
* @return On success, returns PLCRASH_ESUCCESS. On failure, one of the plcrash_error_t error values will be returned, and no
* mapping will be performed.
*
* @note
* This code previously used vm_remap() to perform atomic remapping of process memory. However, this appeared
* to trigger a kernel bug (and resulting panic) on iOS 6.0 through 6.1.2, possibly fixed in 6.1.3. Note that
* no stable release of PLCrashReporter shipped with the vm_remap() code.
*
* Investigation of the failure seems to show an over-release of the target vm_map and backing vm_object, leading to
* NULL dereference, invalid memory references, and in some cases, deadlocks that result in watchdog timeouts.
*
* In one example case, the crash occurs in update_first_free_ll() as a NULL dereference of the vm_map_entry_t parameter.
* Analysis of the limited reports shows that this is called via vm_map_store_update_first_free(). No backtrace is
* available from the kernel panics, but analyzing the register state demonstrates:
* - A reference to vm_map_store_update_first_free() remains in the link register.
* - Of the following callers, one can be eliminated by register state:
* - vm_map_enter - not possible, r3 should be equal to r0
* - vm_map_clip_start - possible
* - vm_map_clip_unnest - possible
* - vm_map_clip_end - possible
*
* In the other panic seen in vm_object_reap_pages(), a value of 0x8008 is loaded and deferenced from the next pointer
* of an element within the vm_object's resident page queue (object->memq).
*
* Unfortunately, our ability to investigate has been extremely constrained by the following issues;
* - The panic is not easily or reliably reproducible
* - Apple's does not support iOS kernel debugging
* - There is no support for jailbreak kernel debugging against iOS 6.x devices at the time of writing.
*
* The work-around used here is to split the vm_remap() into distinct calls to mach_make_memory_entry_64() and
* vm_map(); this follows a largely distinct code path from vm_remap(). In testing by a large-scale user of PLCrashReporter,
* they were no longer able to reproduce the issue with this fix in place. Additionally, they've not been able to reproduce
* the issue on 6.1.3 devices, or had any reports of the issue occuring on 6.1.3 devices.
*
* The mach_make_memory_entry_64() API may not actually return an entry for the full requested length; this requires
* that we loop through the full range, requesting an entry for the remaining unallocated pages, and then mapping
* the pages in question. Since this requires multiple calls to vm_map(), we pre-allocate a contigious range of pages
* for the target mappings into which we'll insert (via overwrite) our own mappings.
*
* @note
* As a work-around for bugs in Apple's Mach-O/dyld implementation, we provide the @a require_full flag; if false,
* a successful mapping that is smaller than the requested range may be made, and will not return an error. This is necessary
* to allow our callers to work around bugs in update_dyld_shared_cache(1), which writes out a larger Mach-O VM segment
* size value than is actually available and mappable. See the plcrash_async_macho_map_segment() API documentation for
* more details. This bug has been reported to Apple as rdar://13707406.
*/
static plcrash_error_t plcrash_async_mobject_remap_pages_workaround (mach_port_t task,
pl_vm_address_t task_addr,
pl_vm_size_t length,
bool require_full,
pl_vm_address_t *result,
pl_vm_size_t *result_length)
{
kern_return_t kt;
/* Compute the total required page size. */
pl_vm_address_t base_addr = mach_vm_trunc_page(task_addr);
pl_vm_size_t total_size = mach_vm_round_page(length + (task_addr - base_addr));
/*
* If short mappings are permitted, determine the actual mappable size of the target range. Due
* to rdar://13707406 (update_dyld_shared_cache appears to write invalid LINKEDIT vmsize), an
* LC_SEGMENT-reported VM size may be far larger than the actual mapped pages. This would result
* in us making large (eg, 36MB) allocations in cases where the mappable range is actually much
* smaller, which can trigger out-of-memory conditions on smaller devices.
*/
if (!require_full) {
pl_vm_size_t verified_size = 0;
while (verified_size < total_size) {
memory_object_size_t entry_length = total_size - verified_size;
mach_port_t mem_handle;
/* Fetch an entry reference */
kt = mach_make_memory_entry_64(task, &entry_length, base_addr + verified_size, VM_PROT_READ, &mem_handle, MACH_PORT_NULL);
if (kt != KERN_SUCCESS) {
/* Once we hit an unmappable page, break */
break;
}
/* Drop the reference */
kt = mach_port_mod_refs(mach_task_self(), mem_handle, MACH_PORT_RIGHT_SEND, -1);
if (kt != KERN_SUCCESS) {
PLCF_DEBUG("mach_port_mod_refs(-1) failed: %d", kt);
}
/* Note the size */
verified_size += entry_length;
}
/* No valid page found at the task_addr */
if (verified_size == 0) {
PLCF_DEBUG("No mappable pages found at 0x%" PRIx64, (uint64_t) task_addr);
return PLCRASH_ENOMEM;
}
/* Reduce the total size to the verified size */
if (verified_size < total_size)
total_size = verified_size;
}
/*
* Set aside a memory range large enough for the total requested number of pages. Ideally the kernel
* will lazy-allocate the backing physical pages so that we don't waste actual memory on this
* pre-emptive page range reservation.
*/
pl_vm_address_t mapping_addr = 0x0;
pl_vm_size_t mapped_size = 0;
#ifdef PL_HAVE_MACH_VM
kt = mach_vm_allocate(mach_task_self(), &mapping_addr, total_size, VM_FLAGS_ANYWHERE);
#else
kt = vm_allocate(mach_task_self(), &mapping_addr, total_size, VM_FLAGS_ANYWHERE);
#endif
if (kt != KERN_SUCCESS) {
PLCF_DEBUG("Failed to allocate a target page range for the page remapping: %d", kt);
return PLCRASH_EINTERNAL;
}
/* Map the source pages into the allocated region, overwriting the existing page mappings */
while (mapped_size < total_size) {
/* Create a reference to the target pages. The returned entry may be smaller than the total length. */
memory_object_size_t entry_length = total_size - mapped_size;
mach_port_t mem_handle;
kt = mach_make_memory_entry_64(task, &entry_length, base_addr + mapped_size, VM_PROT_READ, &mem_handle, MACH_PORT_NULL);
if (kt != KERN_SUCCESS) {
/* No pages are found at the target. When validating the total length above, we already verified the
* availability of the requested pages; if they've now disappeared, we can treat it as an error,
* even if !require_full was specified */
PLCF_DEBUG("mach_make_memory_entry_64() failed: %d", kt);
/* Clean up the reserved pages */
kt = vm_deallocate(mach_task_self(), mapping_addr, total_size);
if (kt != KERN_SUCCESS) {
PLCF_DEBUG("vm_deallocate() failed: %d", kt);
}
/* Return error */
return PLCRASH_ENOMEM;
}
/* Map the pages into our local task, overwriting the allocation used to reserve the target space above. */
pl_vm_address_t target_address = mapping_addr + mapped_size;
#ifdef PL_HAVE_MACH_VM
kt = mach_vm_map(mach_task_self(), &target_address, entry_length, 0x0, VM_FLAGS_FIXED|VM_FLAGS_OVERWRITE, mem_handle, 0x0, TRUE, VM_PROT_READ, VM_PROT_READ, VM_INHERIT_COPY);
#else
kt = vm_map(mach_task_self(), &target_address, entry_length, 0x0, VM_FLAGS_FIXED|VM_FLAGS_OVERWRITE, mem_handle, 0x0, TRUE, VM_PROT_READ, VM_PROT_READ, VM_INHERIT_COPY);
#endif /* !PL_HAVE_MACH_VM */
if (kt != KERN_SUCCESS) {
PLCF_DEBUG("vm_map() failure: %d", kt);
/* Clean up the reserved pages */
kt = vm_deallocate(mach_task_self(), mapping_addr, total_size);
if (kt != KERN_SUCCESS) {
PLCF_DEBUG("vm_deallocate() failed: %d", kt);
}
/* Drop the memory handle */
kt = mach_port_mod_refs(mach_task_self(), mem_handle, MACH_PORT_RIGHT_SEND, -1);
if (kt != KERN_SUCCESS) {
PLCF_DEBUG("mach_port_mod_refs(-1) failed: %d", kt);
}
return PLCRASH_ENOMEM;
}
/* Drop the memory handle */
kt = mach_port_mod_refs(mach_task_self(), mem_handle, MACH_PORT_RIGHT_SEND, -1);
if (kt != KERN_SUCCESS) {
PLCF_DEBUG("mach_port_mod_refs(-1) failed: %d", kt);
}
/* Adjust the total mapping size */
mapped_size += entry_length;
}
*result = mapping_addr;
*result_length = mapped_size;
return PLCRASH_ESUCCESS;
}
/* Return non-zero on error. */
int setup_new_process(TID_t thread,
const char *executable, const char **argv_src,
virtaddr_t *entry_point, virtaddr_t *stack_top)
{
pagetable_t *pagetable;
elf_info_t elf;
openfile_t file;
uintptr_t phys_page;
int i, res;
thread_table_t *thread_entry = thread_get_thread_entry(thread);
int argc = 1;
virtaddr_t argv_begin;
virtaddr_t argv_dst;
int argv_elem_size;
virtaddr_t argv_elem_dst;
file = vfs_open((char *)executable);
/* Make sure the file existed and was a valid ELF file */
if (file < 0) {
return -1;
}
res = elf_parse_header(&elf, file);
if (res < 0) {
return -1;
}
/* Trivial and naive sanity check for entry point: */
if (elf.entry_point < PAGE_SIZE) {
return -1;
}
*entry_point = elf.entry_point;
pagetable = vm_create_pagetable(thread);
thread_entry->pagetable = pagetable;
/* Allocate and map stack */
for(i = 0; i < CONFIG_USERLAND_STACK_SIZE; i++) {
phys_page = physmem_allocblock();
KERNEL_ASSERT(phys_page != 0);
/* Zero the page */
memoryset((void*)ADDR_PHYS_TO_KERNEL(phys_page), 0, PAGE_SIZE);
vm_map(pagetable, phys_page,
(USERLAND_STACK_TOP & PAGE_SIZE_MASK) - i*PAGE_SIZE, 1);
}
/* Allocate and map pages for the segments. We assume that
segments begin at page boundary. (The linker script in tests
directory creates this kind of segments) */
for(i = 0; i < (int)elf.ro_pages; i++) {
int left_to_read = elf.ro_size - i*PAGE_SIZE;
phys_page = physmem_allocblock();
KERNEL_ASSERT(phys_page != 0);
/* Zero the page */
memoryset((void*)ADDR_PHYS_TO_KERNEL(phys_page), 0, PAGE_SIZE);
/* Fill the page from ro segment */
if (left_to_read > 0) {
KERNEL_ASSERT(vfs_seek(file, elf.ro_location + i*PAGE_SIZE) == VFS_OK);
KERNEL_ASSERT(vfs_read(file, (void*)ADDR_PHYS_TO_KERNEL(phys_page),
MIN(PAGE_SIZE, left_to_read))
== (int) MIN(PAGE_SIZE, left_to_read));
}
vm_map(pagetable, phys_page,
elf.ro_vaddr + i*PAGE_SIZE, 0);
}
for(i = 0; i < (int)elf.rw_pages; i++) {
int left_to_read = elf.rw_size - i*PAGE_SIZE;
phys_page = physmem_allocblock();
KERNEL_ASSERT(phys_page != 0);
/* Zero the page */
memoryset((void*)ADDR_PHYS_TO_KERNEL(phys_page), 0, PAGE_SIZE);
/* Fill the page from rw segment */
if (left_to_read > 0) {
KERNEL_ASSERT(vfs_seek(file, elf.rw_location + i*PAGE_SIZE) == VFS_OK);
KERNEL_ASSERT(vfs_read(file, (void*)ADDR_PHYS_TO_KERNEL(phys_page),
MIN(PAGE_SIZE, left_to_read))
== (int) MIN(PAGE_SIZE, left_to_read));
}
vm_map(pagetable, phys_page,
elf.rw_vaddr + i*PAGE_SIZE, 1);
}
/* Set up argc and argv on the stack. */
/* Start by preparing ancillary information for the new process argv. */
if (argv_src != NULL)
for (i = 0; argv_src[i] != NULL; i++) {
argc++;
}
argv_begin = USERLAND_STACK_TOP - (argc * sizeof(virtaddr_t));
argv_dst = argv_begin;
/* Prepare for copying executable. */
argv_elem_size = strlen(executable) + 1;
argv_elem_dst = argv_dst - wordpad(argv_elem_size);
/* Copy executable to argv[0] location. */
vm_memwrite(pagetable,
argv_elem_size,
argv_elem_dst,
executable);
/* Set argv[i] */
vm_memwrite(pagetable,
sizeof(virtaddr_t),
argv_dst,
&argv_elem_dst);
/* Move argv_dst to &argv[1]. */
argv_dst += sizeof(virtaddr_t);
if (argv_src != NULL) {
for (i = 0; argv_src[i] != NULL; i++) {
/* Compute the size of argv[i+1] */
argv_elem_size = strlen(argv_src[i]) + 1;
argv_elem_dst -= wordpad(argv_elem_size);
/* Write the 'i+1'th element of argv */
vm_memwrite(pagetable,
argv_elem_size,
argv_elem_dst,
argv_src[i]);
/* Write argv[i+1] */
vm_memwrite(pagetable,
sizeof(virtaddr_t),
argv_dst,
&argv_elem_dst);
/* Move argv_dst to next element of argv. */
argv_dst += sizeof(virtaddr_t);
}
}
/* Write argc to the stack. */
vm_memwrite(pagetable,
sizeof(int),
argv_elem_dst - sizeof(int),
&argc);
/* Write argv to the stack. */
vm_memwrite(pagetable,
sizeof(virtaddr_t),
argv_elem_dst - sizeof(int) - sizeof(virtaddr_t),
&argv_begin);
/* Stack pointer points at argv. */
*stack_top = argv_elem_dst - sizeof(int) - sizeof(virtaddr_t);
return 0;
}