X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=blobdiff_plain;f=palacios%2Fsrc%2Fpalacios%2Fvmm_shadow_paging.c;h=97dac472c6c6dd5244719736f59f18a030924edd;hb=master;hp=9e70bb2ac040a3dbe79456b79fb98998e7e40a6b;hpb=a5c5675571882a9b8a7594ef07fe303b195ef9ae;p=palacios-OLD.git diff --git a/palacios/src/palacios/vmm_shadow_paging.c b/palacios/src/palacios/vmm_shadow_paging.c index 9e70bb2..97dac47 100644 --- a/palacios/src/palacios/vmm_shadow_paging.c +++ b/palacios/src/palacios/vmm_shadow_paging.c @@ -1,3 +1,23 @@ +/* + * This file is part of the Palacios Virtual Machine Monitor developed + * by the V3VEE Project with funding from the United States National + * Science Foundation and the Department of Energy. + * + * The V3VEE Project is a joint project between Northwestern University + * and the University of New Mexico. You can find out more at + * http://www.v3vee.org + * + * Copyright (c) 2008, Jack Lange + * Copyright (c) 2008, The V3VEE Project + * All rights reserved. + * + * Author: Jack Lange + * + * This is free software. You are permitted to use, + * redistribute, and modify it as specified in the file "V3VEE_LICENSE". + */ + + #include @@ -18,6 +38,36 @@ + +DEFINE_HASHTABLE_INSERT(add_cr3_to_cache, addr_t, struct hashtable *); +DEFINE_HASHTABLE_SEARCH(find_cr3_in_cache, addr_t, struct hashtable *); +DEFINE_HASHTABLE_REMOVE(del_cr3_from_cache, addr_t, struct hashtable *, 0); + + +DEFINE_HASHTABLE_INSERT(add_pte_map, addr_t, addr_t); +DEFINE_HASHTABLE_SEARCH(find_pte_map, addr_t, addr_t); +DEFINE_HASHTABLE_REMOVE(del_pte_map, addr_t, addr_t, 0); + + + + +static uint_t pte_hash_fn(addr_t key) { + return hash_long(key, 32); +} + +static int pte_equals(addr_t key1, addr_t key2) { + return (key1 == key2); +} + +static uint_t cr3_hash_fn(addr_t key) { + return hash_long(key, 32); +} + +static int cr3_equals(addr_t key1, addr_t key2) { + return (key1 == key2); +} + + static int handle_shadow_pte32_fault(struct guest_info* info, addr_t fault_addr, pf_error_t error_code, @@ -26,18 +76,152 @@ static int handle_shadow_pte32_fault(struct guest_info* info, static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr, pf_error_t error_code); -int init_shadow_page_state(struct guest_info * info) { +int v3_init_shadow_page_state(struct guest_info * info) { struct shadow_page_state * state = &(info->shdw_pg_state); - state->guest_mode = PDE32; - state->shadow_mode = PDE32; state->guest_cr3 = 0; state->shadow_cr3 = 0; + + state->cr3_cache = create_hashtable(0, &cr3_hash_fn, &cr3_equals); + + state->cached_cr3 = 0; + state->cached_ptes = NULL; + return 0; } -int handle_shadow_pagefault(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) { + + + +/* + For now we'll do something a little more lightweight +int cache_page_tables32(struct guest_info * info, addr_t pde) { + struct shadow_page_state * state = &(info->shdw_pg_state); + addr_t pde_host_addr; + pde32_t * tmp_pde; + struct hashtable * pte_cache = NULL; + int i = 0; + + + pte_cache = (struct hashtable *)find_cr3_in_cache(state->cr3_cache, pde); + if (pte_cache != NULL) { + PrintError("CR3 already present in cache\n"); + state->current_ptes = pte_cache; + return 1; + } else { + PrintError("Creating new CR3 cache entry\n"); + pte_cache = create_hashtable(0, &pte_hash_fn, &pte_equals); + state->current_ptes = pte_cache; + add_cr3_to_cache(state->cr3_cache, pde, pte_cache); + } + + if (guest_pa_to_host_va(info, pde, &pde_host_addr) == -1) { + PrintError("Could not lookup host address of guest PDE\n"); + return -1; + } + + tmp_pde = (pde32_t *)pde_host_addr; + + add_pte_map(pte_cache, pde, pde_host_addr); + + + for (i = 0; i < MAX_PDE32_ENTRIES; i++) { + if ((tmp_pde[i].present) && (tmp_pde[i].large_page == 0)) { + addr_t pte_host_addr; + + if (guest_pa_to_host_va(info, (addr_t)(PDE32_T_ADDR(tmp_pde[i])), &pte_host_addr) == -1) { + PrintError("Could not lookup host address of guest PDE\n"); + return -1; + } + + add_pte_map(pte_cache, (addr_t)(PDE32_T_ADDR(tmp_pde[i])), pte_host_addr); + } + } + + + return 0; +} +*/ + +int v3_cache_page_tables32(struct guest_info * info, addr_t pde) { + struct shadow_page_state * state = &(info->shdw_pg_state); + addr_t pde_host_addr; + pde32_t * tmp_pde; + struct hashtable * pte_cache = NULL; + int i = 0; + + if (pde == state->cached_cr3) { + return 1; + } + + if (state->cached_ptes != NULL) { + hashtable_destroy(state->cached_ptes, 0, 0); + state->cached_ptes = NULL; + } + + state->cached_cr3 = pde; + + pte_cache = create_hashtable(0, &pte_hash_fn, &pte_equals); + state->cached_ptes = pte_cache; + + if (guest_pa_to_host_va(info, pde, &pde_host_addr) == -1) { + PrintError("Could not lookup host address of guest PDE\n"); + return -1; + } + + tmp_pde = (pde32_t *)pde_host_addr; + + add_pte_map(pte_cache, pde, pde_host_addr); + + + for (i = 0; i < MAX_PDE32_ENTRIES; i++) { + if ((tmp_pde[i].present) && (tmp_pde[i].large_page == 0)) { + addr_t pte_host_addr; + + if (guest_pa_to_host_va(info, (addr_t)(PDE32_T_ADDR(tmp_pde[i])), &pte_host_addr) == -1) { + PrintError("Could not lookup host address of guest PDE\n"); + return -1; + } + + add_pte_map(pte_cache, (addr_t)(PDE32_T_ADDR(tmp_pde[i])), pte_host_addr); + } + } + + return 0; + +} + + + +int v3_replace_shdw_page32(struct guest_info * info, addr_t location, pte32_t * new_page, pte32_t * old_page) { + pde32_t * shadow_pd = (pde32_t *)CR3_TO_PDE32(info->shdw_pg_state.shadow_cr3); + pde32_t * shadow_pde = (pde32_t *)&(shadow_pd[PDE32_INDEX(location)]); + + if (shadow_pde->large_page == 0) { + pte32_t * shadow_pt = (pte32_t *)(addr_t)PDE32_T_ADDR((*shadow_pde)); + pte32_t * shadow_pte = (pte32_t *)&(shadow_pt[PTE32_INDEX(location)]); + + //if (shadow_pte->present == 1) { + *(uint_t *)old_page = *(uint_t *)shadow_pte; + //} + + *(uint_t *)shadow_pte = *(uint_t *)new_page; + + } else { + // currently unhandled + return -1; + } + + return 0; +} + + + + + + +int v3_handle_shadow_pagefault(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) { if (info->mem_mode == PHYSICAL_MEM) { // If paging is not turned on we need to handle the special cases @@ -60,10 +244,10 @@ int handle_shadow_pagefault(struct guest_info * info, addr_t fault_addr, pf_erro } } -addr_t create_new_shadow_pt32() { +addr_t v3_create_new_shadow_pt32() { void * host_pde = 0; - host_pde = V3_AllocPages(1); + host_pde = V3_VAddr(V3_AllocPages(1)); memset(host_pde, 0, PAGE_SIZE); return (addr_t)host_pde; @@ -136,12 +320,13 @@ static int handle_large_pagefault32(struct guest_info * info, if (host_page_type == HOST_REGION_INVALID) { // Inject a machine check in the guest - PrintDebug("Invalid Guest Address in page table (0x%x)\n", guest_fault_pa); + PrintDebug("Invalid Guest Address in page table (0x%p)\n", (void *)guest_fault_pa); v3_raise_exception(info, MC_EXCEPTION); return 0; } if (host_page_type == HOST_REGION_PHYSICAL_MEMORY) { + struct shadow_page_state * state = &(info->shdw_pg_state); addr_t shadow_pa = get_shadow_addr(info, guest_fault_pa); shadow_pte->page_base_addr = PT32_BASE_ADDR(shadow_pa); @@ -154,7 +339,16 @@ static int handle_large_pagefault32(struct guest_info * info, * Allow everything */ shadow_pte->user_page = 1; - shadow_pte->writable = 1; + + if (find_pte_map(state->cached_ptes, PT32_PAGE_ADDR(guest_fault_pa)) != NULL) { + // Check if the entry is a page table... + PrintDebug("Marking page as Guest Page Table (large page)\n"); + shadow_pte->vmm_info = PT32_GUEST_PT; + shadow_pte->writable = 0; + } else { + shadow_pte->writable = 1; + } + //set according to VMM policy shadow_pte->write_through = 0; @@ -164,11 +358,19 @@ static int handle_large_pagefault32(struct guest_info * info, } else { // Handle hooked pages as well as other special pages - if (handle_special_page_fault(info, fault_addr, PT32_PAGE_ADDR(guest_fault_pa), error_code) == -1) { - PrintError("Special Page Fault handler returned error for address: %x\n", fault_addr); + if (handle_special_page_fault(info, fault_addr, guest_fault_pa, error_code) == -1) { + PrintError("Special Page Fault handler returned error for address: %p\n", (void *)fault_addr); return -1; } } + } else if ((shadow_pte_access == PT_WRITE_ERROR) && + (shadow_pte->vmm_info == PT32_GUEST_PT)) { + + struct shadow_page_state * state = &(info->shdw_pg_state); + PrintDebug("Write operation on Guest PAge Table Page (large page)\n"); + state->cached_cr3 = 0; + shadow_pte->writable = 1; + } else { PrintError("Error in large page fault handler...\n"); PrintError("This case should have been handled at the top level handler\n"); @@ -183,16 +385,16 @@ static int handle_large_pagefault32(struct guest_info * info, static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) { pde32_t * guest_pd = NULL; pde32_t * shadow_pd = (pde32_t *)CR3_TO_PDE32(info->shdw_pg_state.shadow_cr3); - addr_t guest_cr3 = CR3_TO_PDE32(info->shdw_pg_state.guest_cr3); + addr_t guest_cr3 = (addr_t) V3_PAddr( CR3_TO_PDE32(info->shdw_pg_state.guest_cr3) ); pt_access_status_t guest_pde_access; pt_access_status_t shadow_pde_access; pde32_t * guest_pde = NULL; pde32_t * shadow_pde = (pde32_t *)&(shadow_pd[PDE32_INDEX(fault_addr)]); - PrintDebug("Shadow page fault handler\n"); + PrintDebug("Shadow page fault handler: %p\n", (void*) fault_addr ); if (guest_pa_to_host_va(info, guest_cr3, (addr_t*)&guest_pd) == -1) { - PrintError("Invalid Guest PDE Address: 0x%x\n", guest_cr3); + PrintError("Invalid Guest PDE Address: 0x%p\n", (void *)guest_cr3); return -1; } @@ -208,7 +410,7 @@ static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr /* Was the page fault caused by the Guest's page tables? */ if (is_guest_pf(guest_pde_access, shadow_pde_access) == 1) { PrintDebug("Injecting PDE pf to guest: (guest access error=%d) (pf error code=%d)\n", - guest_pde_access, error_code); + *(uint_t *)&guest_pde_access, *(uint_t *)&error_code); inject_guest_pf(info, fault_addr, error_code); return 0; } @@ -216,7 +418,7 @@ static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr if (shadow_pde_access == PT_ENTRY_NOT_PRESENT) { - pte32_t * shadow_pt = (pte32_t *)create_new_shadow_pt32(); + pte32_t * shadow_pt = (pte32_t *)v3_create_new_shadow_pt32(); shadow_pde->present = 1; shadow_pde->user_page = guest_pde->user_page; @@ -232,11 +434,12 @@ static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr guest_pde->accessed = 1; - shadow_pde->pt_base_addr = PD32_BASE_ADDR(shadow_pt); + shadow_pde->pt_base_addr = PD32_BASE_ADDR((addr_t)V3_PAddr(shadow_pt)); if (guest_pde->large_page == 0) { shadow_pde->writable = guest_pde->writable; } else { + // ?? What if guest pde is dirty a this point? ((pde32_4MB_t *)guest_pde)->dirty = 0; shadow_pde->writable = 0; } @@ -246,7 +449,7 @@ static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr // // PTE fault // - pte32_t * shadow_pt = (pte32_t *)PDE32_T_ADDR((*shadow_pde)); + pte32_t * shadow_pt = (pte32_t *)V3_VAddr( (void*)(addr_t) PDE32_T_ADDR(*shadow_pde) ); if (guest_pde->large_page == 0) { pte32_t * guest_pt = NULL; @@ -276,6 +479,7 @@ static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr // Page Directory Entry marked read-only // Its a large page and we need to update the dirty bit in the guest // + PrintDebug("Large page write error... Setting dirty bit and returning\n"); ((pde32_4MB_t *)guest_pde)->dirty = 1; shadow_pde->writable = guest_pde->writable; @@ -308,7 +512,7 @@ static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr return 0; } - PrintDebug("Returning end of PDE function (rip=%x)\n", info->rip); + PrintDebug("Returning end of PDE function (rip=%p)\n", (void *)(addr_t)(info->rip)); return 0; } @@ -361,7 +565,7 @@ static int handle_shadow_pte32_fault(struct guest_info * info, if (shadow_pte_access == PT_ENTRY_NOT_PRESENT) { - addr_t guest_pa = PTE32_T_ADDR((*guest_pte)); + addr_t guest_pa = PTE32_T_ADDR((*guest_pte)) + PT32_PAGE_OFFSET(fault_addr); // Page Table Entry Not Present @@ -369,7 +573,7 @@ static int handle_shadow_pte32_fault(struct guest_info * info, if (host_page_type == HOST_REGION_INVALID) { // Inject a machine check in the guest - PrintDebug("Invalid Guest Address in page table (0x%x)\n", guest_pa); + PrintDebug("Invalid Guest Address in page table (0x%p)\n", (void *)guest_pa); v3_raise_exception(info, MC_EXCEPTION); return 0; } @@ -377,6 +581,7 @@ static int handle_shadow_pte32_fault(struct guest_info * info, // else... if (host_page_type == HOST_REGION_PHYSICAL_MEMORY) { + struct shadow_page_state * state = &(info->shdw_pg_state); addr_t shadow_pa = get_shadow_addr(info, guest_pa); shadow_pte->page_base_addr = PT32_BASE_ADDR(shadow_pa); @@ -392,18 +597,35 @@ static int handle_shadow_pte32_fault(struct guest_info * info, guest_pte->accessed = 1; + if (find_pte_map(state->cached_ptes, PT32_PAGE_ADDR(guest_pa)) != NULL) { + // Check if the entry is a page table... + PrintDebug("Marking page as Guest Page Table %d\n", shadow_pte->writable); + shadow_pte->vmm_info = PT32_GUEST_PT; + } + if (guest_pte->dirty == 1) { shadow_pte->writable = guest_pte->writable; } else if ((guest_pte->dirty == 0) && (error_code.write == 1)) { shadow_pte->writable = guest_pte->writable; guest_pte->dirty = 1; - } else if ((guest_pte->dirty = 0) && (error_code.write == 0)) { + + if (shadow_pte->vmm_info == PT32_GUEST_PT) { + // Well that was quick... + struct shadow_page_state * state = &(info->shdw_pg_state); + PrintDebug("Immediate Write operation on Guest PAge Table Page\n"); + state->cached_cr3 = 0; + } + + } else if ((guest_pte->dirty == 0) && (error_code.write == 0)) { // was = shadow_pte->writable = 0; } + + + } else { // Page fault handled by hook functions if (handle_special_page_fault(info, fault_addr, guest_pa, error_code) == -1) { - PrintError("Special Page fault handler returned error for address: %x\n", fault_addr); + PrintError("Special Page fault handler returned error for address: %p\n", (void *)fault_addr); return -1; } } @@ -414,6 +636,13 @@ static int handle_shadow_pte32_fault(struct guest_info * info, PrintDebug("Shadow PTE Write Error\n"); guest_pte->dirty = 1; shadow_pte->writable = guest_pte->writable; + + if (shadow_pte->vmm_info == PT32_GUEST_PT) { + struct shadow_page_state * state = &(info->shdw_pg_state); + PrintDebug("Write operation on Guest PAge Table Page\n"); + state->cached_cr3 = 0; + } + return 0; } else { @@ -434,56 +663,64 @@ static int handle_shadow_pte32_fault(struct guest_info * info, /* Currently Does not work with Segmentation!!! */ -int handle_shadow_invlpg(struct guest_info * info) { - if (info->mem_mode != VIRTUAL_MEM) { - // Paging must be turned on... - // should handle with some sort of fault I think - PrintError("ERROR: INVLPG called in non paged mode\n"); - return -1; - } +int v3_handle_shadow_invlpg(struct guest_info * info) +{ + if (info->mem_mode != VIRTUAL_MEM) { + // Paging must be turned on... + // should handle with some sort of fault I think + PrintError("ERROR: INVLPG called in non paged mode\n"); + return -1; + } - if (info->cpu_mode == PROTECTED) { - char instr[15]; - int ret; - int index = 0; + if (info->cpu_mode != PROTECTED) + return 0; - ret = read_guest_va_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); - if (ret != 15) { - PrintError("Could not read instruction 0x%x (ret=%d)\n", info->rip, ret); - return -1; - } + uchar_t instr[15]; + int index = 0; + + int ret = read_guest_va_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr); + if (ret != 15) { + PrintError("Could not read instruction 0x%p (ret=%d)\n", (void *)(addr_t)(info->rip), ret); + return -1; + } - /* Can INVLPG work with Segments?? */ - while (is_prefix_byte(instr[index])) { - index++; - } + /* Can INVLPG work with Segments?? */ + while (is_prefix_byte(instr[index])) { + index++; + } - if ((instr[index] == (uchar_t)0x0f) && - (instr[index + 1] == (uchar_t)0x01)) { - - addr_t first_operand; - addr_t second_operand; - operand_type_t addr_type; - addr_t guest_cr3 = CR3_TO_PDE32(info->shdw_pg_state.guest_cr3); + if( instr[index + 0] != (uchar_t) 0x0f + || instr[index + 1] != (uchar_t) 0x01 + ) { + PrintError("invalid Instruction Opcode\n"); + PrintTraceMemDump(instr, 15); + return -1; + } - pde32_t * guest_pd = NULL; + addr_t first_operand; + addr_t second_operand; + addr_t guest_cr3 = (addr_t)V3_PAddr( (void*)(addr_t) CR3_TO_PDE32(info->shdw_pg_state.guest_cr3) ); - if (guest_pa_to_host_va(info, guest_cr3, (addr_t*)&guest_pd) == -1) { - PrintError("Invalid Guest PDE Address: 0x%x\n", guest_cr3); - return -1; - } + pde32_t * guest_pd = NULL; - + if (guest_pa_to_host_va(info, guest_cr3, (addr_t*)&guest_pd) == -1) + { + PrintError("Invalid Guest PDE Address: 0x%p\n", (void *)guest_cr3); + return -1; + } + index += 2; - index += 2; + v3_operand_type_t addr_type = decode_operands32(&(info->vm_regs), instr + index, &index, &first_operand, &second_operand, REG32); - addr_type = decode_operands32(&(info->vm_regs), instr + index, &index, &first_operand, &second_operand, REG32); + if (addr_type != MEM_OPERAND) { + PrintError("Invalid Operand type\n"); + return -1; + } - if (addr_type == MEM_OPERAND) { pde32_t * shadow_pd = (pde32_t *)CR3_TO_PDE32(info->shdw_pg_state.shadow_cr3); pde32_t * shadow_pde = (pde32_t *)&shadow_pd[PDE32_INDEX(first_operand)]; pde32_t * guest_pde; @@ -491,42 +728,29 @@ int handle_shadow_invlpg(struct guest_info * info) { //PrintDebug("PDE Index=%d\n", PDE32_INDEX(first_operand)); //PrintDebug("FirstOperand = %x\n", first_operand); - PrintDebug("Invalidating page for %x\n", first_operand); + PrintDebug("Invalidating page for %p\n", (void *)first_operand); guest_pde = (pde32_t *)&(guest_pd[PDE32_INDEX(first_operand)]); if (guest_pde->large_page == 1) { - shadow_pde->present = 0; - PrintDebug("Invalidating Large Page\n"); - } else { - - if (shadow_pde->present == 1) { - pte32_t * shadow_pt = (pte32_t *)PDE32_T_ADDR((*shadow_pde)); - pte32_t * shadow_pte = (pte32_t *)&shadow_pt[PTE32_INDEX(first_operand)]; + shadow_pde->present = 0; + PrintDebug("Invalidating Large Page\n"); + } else + if (shadow_pde->present == 1) { + pte32_t * shadow_pt = (pte32_t *)(addr_t)PDE32_T_ADDR((*shadow_pde)); + pte32_t * shadow_pte = (pte32_t *) V3_VAddr( (void*) &shadow_pt[PTE32_INDEX(first_operand)] ); #ifdef DEBUG_SHADOW_PAGING - PrintDebug("Setting not present\n"); - PrintPTE32(first_operand, shadow_pte); + PrintDebug("Setting not present\n"); + PrintPTE32(first_operand, shadow_pte ); #endif - shadow_pte->present = 0; - } + shadow_pte->present = 0; } info->rip += index; - } else { - PrintError("Invalid Operand type\n"); - return -1; - } - } else { - PrintError("invalid Instruction Opcode\n"); - PrintTraceMemDump(instr, 15); - return -1; - } - } - - return 0; + return 0; }