// Based on ARM Reference Manual
// Chapter B4 Virtual Memory System Architecture
// B4.7 Hardware page table translation
status_t v2p_aarch32 (vmi_instance_t vmi,
addr_t dtb,
addr_t vaddr,
page_info_t *info)
{
status_t status = VMI_FAILURE;
dbprint(VMI_DEBUG_PTLOOKUP, "--ARM AArch32 PTLookup: vaddr = 0x%.16"PRIx64", dtb = 0x%.16"PRIx64"\n", vaddr, dtb);
get_first_level_descriptor(vmi, dtb, vaddr, info);
dbprint(VMI_DEBUG_PTLOOKUP, "--ARM AArch32 PTLookup: fld_location = 0x%"PRIx32"\n", info->arm_aarch32.fld_location);
dbprint(VMI_DEBUG_PTLOOKUP, "--ARM AArch32 PTLookup: fld_value = 0x%"PRIx32"\n", info->arm_aarch32.fld_value);
switch (info->arm_aarch32.fld_value & VMI_BIT_MASK(0,1)) {
case 0b01: {
dbprint(VMI_DEBUG_PTLOOKUP, "--ARM AArch32 PTLookup: the entry gives the physical address of a coarse second-level table\n");
get_coarse_second_level_descriptor(vmi, vaddr, info);
dbprint(VMI_DEBUG_PTLOOKUP, "--ARM AArch32 PTLookup: l2d = 0x%"PRIx32"\n", info->arm_aarch32.sld_value);
switch (info->arm_aarch32.sld_value & VMI_BIT_MASK(0,1)) {
case 0b01:
// large page
info->size = VMI_PS_64KB;
info->paddr = (info->arm_aarch32.sld_value & VMI_BIT_MASK(16,31)) | (vaddr & VMI_BIT_MASK(0,15));
status = VMI_SUCCESS;
break;
case 0b10:
case 0b11:
// small page
info->size = VMI_PS_4KB;
info->paddr = (info->arm_aarch32.sld_value & VMI_BIT_MASK(12,31)) | (vaddr & VMI_BIT_MASK(0,11));
status = VMI_SUCCESS;
default:
break;
}
break;
}
case 0b10: {
if (!VMI_GET_BIT(info->arm_aarch32.fld_value, 18)) {
dbprint(VMI_DEBUG_PTLOOKUP, "--ARM AArch32 PTLookup: the entry is a section descriptor for its associated modified virtual addresses\n");
info->size = VMI_PS_1MB;
info->paddr = (info->arm_aarch32.fld_value & VMI_BIT_MASK(20,31)) | (vaddr & VMI_BIT_MASK(0,19));
} else {
dbprint(VMI_DEBUG_PTLOOKUP, "--ARM AArch32 PTLookup: the entry is a supersection descriptor for its associated modified virtual addresses\n");
info->size = VMI_PS_16MB;
info->paddr = (info->arm_aarch32.fld_value & VMI_BIT_MASK(24,31)) | (vaddr & VMI_BIT_MASK(0,23));
}
status = VMI_SUCCESS;
break;
}
case 0b11: {
dbprint(VMI_DEBUG_PTLOOKUP, "--ARM AArch32 PTLookup: the entry gives the physical address of a fine second-level table\n");
get_fine_second_level_descriptor(vmi, vaddr, info);
dbprint(VMI_DEBUG_PTLOOKUP, "--ARM AArch32 PTLookup: sld = 0x%"PRIx32"\n", info->arm_aarch32.sld_value);
switch (info->arm_aarch32.sld_value & VMI_BIT_MASK(0,1)) {
case 0b01:
// large page
info->size = VMI_PS_64KB;
info->paddr = (info->arm_aarch32.sld_value & VMI_BIT_MASK(16,31)) | (vaddr & VMI_BIT_MASK(0,15));
status = VMI_SUCCESS;
break;
case 0b10:
// small page
info->size = VMI_PS_4KB;
info->paddr = (info->arm_aarch32.sld_value & VMI_BIT_MASK(12,31)) | (vaddr & VMI_BIT_MASK(0,11));
status = VMI_SUCCESS;
break;
case 0b11:
// tiny page
info->size = VMI_PS_1KB;
info->paddr = (info->arm_aarch32.sld_value & VMI_BIT_MASK(10,31)) | (vaddr & VMI_BIT_MASK(0,9));
status = VMI_SUCCESS;
break;
default:
break;
}
break;
}
default:
break;
}
dbprint(VMI_DEBUG_PTLOOKUP, "--ARM PTLookup: PA = 0x%"PRIx64"\n", info->paddr);
return status;
}
/*
* check that this vm uses a paging method that we support
* and set pm/cr3/pae/pse/lme flags optionally on the given pointers
*/
status_t probe_memory_layout_x86(vmi_instance_t vmi) {
// To get the paging layout, the following bits are needed:
// 1. CR0.PG
// 2. CR4.PAE
// 3. Either (a) IA32_EFER.LME, or (b) the guest's address width (32 or
// 64). Not all backends allow us to read an MSR; in particular, Xen's PV
// backend doessn't.
status_t ret = VMI_FAILURE;
page_mode_t pm = VMI_PM_UNKNOWN;
uint8_t dom_addr_width = 0; // domain address width (bytes)
/* pull info from registers, if we can */
reg_t cr0, cr3, cr4, efer;
int pae = 0, pse = 0, lme = 0;
uint8_t msr_efer_lme = 0; // LME bit in MSR_EFER
/* get the control register values */
if (driver_get_vcpureg(vmi, &cr0, CR0, 0) == VMI_FAILURE) {
errprint("**failed to get CR0\n");
goto _exit;
}
/* PG Flag --> CR0, bit 31 == 1 --> paging enabled */
if (!VMI_GET_BIT(cr0, 31)) {
dbprint(VMI_DEBUG_CORE, "Paging disabled for this VM, only physical addresses supported.\n");
vmi->page_mode = VMI_PM_UNKNOWN;
vmi->pae = 0;
vmi->pse = 0;
vmi->lme = 0;
ret = VMI_SUCCESS;
goto _exit;
}
//
// Paging enabled (PG==1)
//
if (driver_get_vcpureg(vmi, &cr4, CR4, 0) == VMI_FAILURE) {
errprint("**failed to get CR4\n");
goto _exit;
}
/* PSE Flag --> CR4, bit 5 */
pae = VMI_GET_BIT(cr4, 5);
dbprint(VMI_DEBUG_CORE, "**set pae = %d\n", pae);
/* PSE Flag --> CR4, bit 4 */
pse = VMI_GET_BIT(cr4, 4);
dbprint(VMI_DEBUG_CORE, "**set pse = %d\n", pse);
ret = driver_get_vcpureg(vmi, &efer, MSR_EFER, 0);
if (VMI_SUCCESS == ret) {
lme = VMI_GET_BIT(efer, 8);
dbprint(VMI_DEBUG_CORE, "**set lme = %d\n", lme);
} else {
dbprint(VMI_DEBUG_CORE, "**failed to get MSR_EFER, trying method #2\n");
// does this trick work in all cases?
ret = driver_get_address_width(vmi, &dom_addr_width);
if (VMI_FAILURE == ret) {
errprint
("Failed to get domain address width. Giving up.\n");
goto _exit;
}
lme = (8 == dom_addr_width);
dbprint
(VMI_DEBUG_CORE, "**found guest address width is %d bytes; assuming IA32_EFER.LME = %d\n",
dom_addr_width, lme);
} // if
// Get current cr3 for sanity checking
if (driver_get_vcpureg(vmi, &cr3, CR3, 0) == VMI_FAILURE) {
errprint("**failed to get CR3\n");
goto _exit;
}
// now determine addressing mode
if (0 == pae) {
dbprint(VMI_DEBUG_CORE, "**32-bit paging\n");
pm = VMI_PM_LEGACY;
cr3 &= 0xFFFFF000ull;
}
// PAE == 1; determine IA-32e or PAE
else if (lme) { // PAE == 1, LME == 1
dbprint(VMI_DEBUG_CORE, "**IA-32e paging\n");
pm = VMI_PM_IA32E;
cr3 &= 0xFFFFFFFFFFFFF000ull;
} else { // PAE == 1, LME == 0
dbprint(VMI_DEBUG_CORE, "**PAE paging\n");
pm = VMI_PM_PAE;
cr3 &= 0xFFFFFFE0;
} // if-else
dbprint(VMI_DEBUG_CORE, "**sanity checking cr3 = 0x%.16"PRIx64"\n", cr3);
/* testing to see CR3 value */
if (!driver_is_pv(vmi) && cr3 >= vmi->max_physical_address) { // sanity check on CR3
dbprint(VMI_DEBUG_CORE, "** Note cr3 value [0x%"PRIx64"] exceeds memsize [0x%"PRIx64"]\n",
cr3, vmi->size);
}
vmi->page_mode = pm;
vmi->pae = pae;
vmi->pse = pse;
vmi->lme = lme;
_exit:
return ret;
}
static status_t init_task_kaslr_test(vmi_instance_t vmi, addr_t page_vaddr) {
status_t ret = VMI_FAILURE;
uint32_t pid;
addr_t init_task = page_vaddr + (vmi->init_task & VMI_BIT_MASK(0,11));
linux_instance_t linux_instance = vmi->os_data;
access_context_t ctx = {
.translate_mechanism = VMI_TM_PROCESS_DTB,
.dtb = vmi->kpgd
};
ctx.addr = init_task + linux_instance->pid_offset;
if ( VMI_FAILURE == vmi_read_32(vmi, &ctx, &pid) )
return ret;
if ( pid )
return ret;
ctx.addr = init_task + linux_instance->name_offset;
char* init_task_name = vmi_read_str(vmi, &ctx);
if ( init_task_name && !strncmp("swapper", init_task_name, 7) )
ret = VMI_SUCCESS;
free(init_task_name);
return ret;
}
status_t init_kaslr(vmi_instance_t vmi) {
/*
* Let's check if we can translate init_task first as is.
*/
uint32_t test;
access_context_t ctx = {
.translate_mechanism = VMI_TM_PROCESS_DTB,
.dtb = vmi->kpgd,
.addr = vmi->init_task
};
if ( VMI_SUCCESS == vmi_read_32(vmi, &ctx, &test) )
return VMI_SUCCESS;
status_t ret = VMI_FAILURE;
linux_instance_t linux_instance = vmi->os_data;
GSList *loop, *pages = vmi_get_va_pages(vmi, vmi->kpgd);
loop = pages;
while (loop) {
page_info_t *info = loop->data;
if ( !linux_instance->kaslr_offset ) {
switch(vmi->page_mode) {
case VMI_PM_AARCH64:
case VMI_PM_IA32E:
if ( VMI_GET_BIT(info->vaddr, 47) )
ret = init_task_kaslr_test(vmi, info->vaddr);
break;
default:
ret = init_task_kaslr_test(vmi, info->vaddr);
break;
};
if ( VMI_SUCCESS == ret ) {
linux_instance->kaslr_offset = info->vaddr - (vmi->init_task & ~VMI_BIT_MASK(0,11));
vmi->init_task += linux_instance->kaslr_offset;
dbprint(VMI_DEBUG_MISC, "**calculated KASLR offset: 0x%"PRIx64"\n", linux_instance->kaslr_offset);
}
}
g_free(info);
loop = loop->next;
}
g_slist_free(pages);
return ret;
}
status_t linux_init(vmi_instance_t vmi) {
status_t rc;
os_interface_t os_interface = NULL;
if (vmi->config == NULL) {
errprint("No config table found\n");
return VMI_FAILURE;
}
if (vmi->os_data != NULL) {
errprint("os data already initialized, reinitializing\n");
free(vmi->os_data);
}
vmi->os_data = safe_malloc(sizeof(struct linux_instance));
bzero(vmi->os_data, sizeof(struct linux_instance));
linux_instance_t linux_instance = vmi->os_data;
g_hash_table_foreach(vmi->config, (GHFunc)linux_read_config_ghashtable_entries, vmi);
if(linux_instance->rekall_profile)
rc = init_from_rekall_profile(vmi);
else
rc = linux_symbol_to_address(vmi, "init_task", NULL, &vmi->init_task);
if (VMI_FAILURE == rc) {
errprint("Could not get init_task from Rekall profile or System.map\n");
goto _exit;
}
vmi->init_task = canonical_addr(vmi->init_task);
#if defined(ARM32) || defined(ARM64)
rc = driver_get_vcpureg(vmi, &vmi->kpgd, TTBR1, 0);
#elif defined(I386) || defined(X86_64)
rc = driver_get_vcpureg(vmi, &vmi->kpgd, CR3, 0);
#endif
/*
* The driver failed to get us a pagetable.
* As a fall-back, try to init using heuristics.
* This path is taken in FILE mode as well.
*/
if (VMI_FAILURE == rc)
if (VMI_FAILURE == linux_filemode_init(vmi))
goto _exit;
if ( VMI_FAILURE == init_kaslr(vmi) ) {
dbprint(VMI_DEBUG_MISC, "**failed to determine KASLR offset\n");
goto _exit;
}
dbprint(VMI_DEBUG_MISC, "**set vmi->kpgd (0x%.16"PRIx64").\n", vmi->kpgd);
os_interface = safe_malloc(sizeof(struct os_interface));
bzero(os_interface, sizeof(struct os_interface));
os_interface->os_get_offset = linux_get_offset;
os_interface->os_pid_to_pgd = linux_pid_to_pgd;
os_interface->os_pgd_to_pid = linux_pgd_to_pid;
os_interface->os_ksym2v = linux_symbol_to_address;
os_interface->os_usym2rva = NULL;
os_interface->os_v2sym = linux_system_map_address_to_symbol;
os_interface->os_read_unicode_struct = NULL;
os_interface->os_teardown = linux_teardown;
vmi->os_interface = os_interface;
return VMI_SUCCESS;
_exit:
free(vmi->os_data);
vmi->os_data = NULL;
return VMI_FAILURE;
}
