Home | History | Annotate | Line # | Download | only in include
asan.h revision 1.9
      1 /*	$NetBSD: asan.h,v 1.9 2020/09/10 14:10:46 maxv Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 2018-2020 Maxime Villard, m00nbsd.net
      5  * All rights reserved.
      6  *
      7  * This code is part of the KASAN subsystem of the NetBSD kernel.
      8  *
      9  * Redistribution and use in source and binary forms, with or without
     10  * modification, are permitted provided that the following conditions
     11  * are met:
     12  * 1. Redistributions of source code must retain the above copyright
     13  *    notice, this list of conditions and the following disclaimer.
     14  * 2. Redistributions in binary form must reproduce the above copyright
     15  *    notice, this list of conditions and the following disclaimer in the
     16  *    documentation and/or other materials provided with the distribution.
     17  *
     18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
     19  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     20  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     21  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
     22  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
     23  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
     24  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
     25  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
     26  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     28  * SUCH DAMAGE.
     29  */
     30 
     31 #include <sys/ksyms.h>
     32 
     33 #include <uvm/uvm.h>
     34 
     35 #include <amd64/pmap.h>
     36 #include <amd64/vmparam.h>
     37 
     38 #ifdef __HAVE_PCPU_AREA
     39 #error "PCPU area not allowed with KASAN"
     40 #endif
     41 #ifdef __HAVE_DIRECT_MAP
     42 #error "DMAP not allowed with KASAN"
     43 #endif
     44 
     45 #define __MD_VIRTUAL_SHIFT	47	/* 48bit address space, cut half */
     46 #define __MD_KERNMEM_BASE	0xFFFF800000000000 /* kern mem base address */
     47 
     48 #define __MD_SHADOW_SIZE	(1ULL << (__MD_VIRTUAL_SHIFT - KASAN_SHADOW_SCALE_SHIFT))
     49 #define KASAN_MD_SHADOW_START	(VA_SIGN_NEG((L4_SLOT_KASAN * NBPD_L4)))
     50 #define KASAN_MD_SHADOW_END	(KASAN_MD_SHADOW_START + __MD_SHADOW_SIZE)
     51 
     52 /* -------------------------------------------------------------------------- */
     53 
     54 /*
     55  * Early mapping, used to map just the stack at boot time. We rely on the fact
     56  * that VA = PA + KERNBASE.
     57  */
     58 
     59 static bool __md_early __read_mostly = true;
     60 static uint8_t __md_earlypages[8 * PAGE_SIZE] __aligned(PAGE_SIZE);
     61 static size_t __md_earlytaken = 0;
     62 
     63 static paddr_t
     64 __md_early_palloc(void)
     65 {
     66 	paddr_t ret;
     67 
     68 	KASSERT(__md_earlytaken < 8);
     69 
     70 	ret = (paddr_t)(&__md_earlypages[0] + __md_earlytaken * PAGE_SIZE);
     71 	__md_earlytaken++;
     72 
     73 	ret -= KERNBASE;
     74 
     75 	return ret;
     76 }
     77 
     78 static void
     79 __md_early_shadow_map_page(vaddr_t va)
     80 {
     81 	extern struct bootspace bootspace;
     82 	const pt_entry_t pteflags = PTE_W | pmap_pg_nx | PTE_P;
     83 	pt_entry_t *pdir = (pt_entry_t *)bootspace.pdir;
     84 	paddr_t pa;
     85 
     86 	if (!pmap_valid_entry(pdir[pl4_pi(va)])) {
     87 		pa = __md_early_palloc();
     88 		pdir[pl4_pi(va)] = pa | pteflags;
     89 	}
     90 	pdir = (pt_entry_t *)((pdir[pl4_pi(va)] & PTE_FRAME) + KERNBASE);
     91 
     92 	if (!pmap_valid_entry(pdir[pl3_pi(va)])) {
     93 		pa = __md_early_palloc();
     94 		pdir[pl3_pi(va)] = pa | pteflags;
     95 	}
     96 	pdir = (pt_entry_t *)((pdir[pl3_pi(va)] & PTE_FRAME) + KERNBASE);
     97 
     98 	if (!pmap_valid_entry(pdir[pl2_pi(va)])) {
     99 		pa = __md_early_palloc();
    100 		pdir[pl2_pi(va)] = pa | pteflags;
    101 	}
    102 	pdir = (pt_entry_t *)((pdir[pl2_pi(va)] & PTE_FRAME) + KERNBASE);
    103 
    104 	if (!pmap_valid_entry(pdir[pl1_pi(va)])) {
    105 		pa = __md_early_palloc();
    106 		pdir[pl1_pi(va)] = pa | pteflags | pmap_pg_g;
    107 	}
    108 }
    109 
    110 /* -------------------------------------------------------------------------- */
    111 
    112 static inline int8_t *
    113 kasan_md_addr_to_shad(const void *addr)
    114 {
    115 	vaddr_t va = (vaddr_t)addr;
    116 	return (int8_t *)(KASAN_MD_SHADOW_START +
    117 	    ((va - __MD_KERNMEM_BASE) >> KASAN_SHADOW_SCALE_SHIFT));
    118 }
    119 
    120 static inline bool
    121 kasan_md_unsupported(vaddr_t addr)
    122 {
    123 	return (addr >= (vaddr_t)PTE_BASE &&
    124 	    addr < ((vaddr_t)PTE_BASE + NBPD_L4));
    125 }
    126 
    127 static paddr_t
    128 __md_palloc(void)
    129 {
    130 	/* The page is zeroed. */
    131 	return pmap_get_physpage();
    132 }
    133 
    134 static inline paddr_t
    135 __md_palloc_large(void)
    136 {
    137 	struct pglist pglist;
    138 	int ret;
    139 
    140 	if (!uvm.page_init_done)
    141 		return 0;
    142 
    143 	ret = uvm_pglistalloc(NBPD_L2, 0, ~0UL, NBPD_L2, 0,
    144 	    &pglist, 1, 0);
    145 	if (ret != 0)
    146 		return 0;
    147 
    148 	/* The page may not be zeroed. */
    149 	return VM_PAGE_TO_PHYS(TAILQ_FIRST(&pglist));
    150 }
    151 
    152 static void
    153 kasan_md_shadow_map_page(vaddr_t va)
    154 {
    155 	const pt_entry_t pteflags = PTE_W | pmap_pg_nx | PTE_P;
    156 	paddr_t pa;
    157 
    158 	if (__predict_false(__md_early)) {
    159 		__md_early_shadow_map_page(va);
    160 		return;
    161 	}
    162 
    163 	if (!pmap_valid_entry(L4_BASE[pl4_i(va)])) {
    164 		pa = __md_palloc();
    165 		L4_BASE[pl4_i(va)] = pa | pteflags;
    166 	}
    167 	if (!pmap_valid_entry(L3_BASE[pl3_i(va)])) {
    168 		pa = __md_palloc();
    169 		L3_BASE[pl3_i(va)] = pa | pteflags;
    170 	}
    171 	if (!pmap_valid_entry(L2_BASE[pl2_i(va)])) {
    172 		if ((pa = __md_palloc_large()) != 0) {
    173 			L2_BASE[pl2_i(va)] = pa | pteflags | PTE_PS |
    174 			    pmap_pg_g;
    175 			__insn_barrier();
    176 			__builtin_memset((void *)va, 0, NBPD_L2);
    177 			return;
    178 		}
    179 		pa = __md_palloc();
    180 		L2_BASE[pl2_i(va)] = pa | pteflags;
    181 	} else if (L2_BASE[pl2_i(va)] & PTE_PS) {
    182 		return;
    183 	}
    184 	if (!pmap_valid_entry(L1_BASE[pl1_i(va)])) {
    185 		pa = __md_palloc();
    186 		L1_BASE[pl1_i(va)] = pa | pteflags | pmap_pg_g;
    187 	}
    188 }
    189 
    190 /*
    191  * Map only the current stack. We will map the rest in kasan_init.
    192  */
    193 static void
    194 kasan_md_early_init(void *stack)
    195 {
    196 	kasan_shadow_map(stack, USPACE);
    197 	__md_early = false;
    198 }
    199 
    200 /*
    201  * Create the shadow mapping. We don't create the 'User' area, because we
    202  * exclude it from the monitoring. The 'Main' area is created dynamically
    203  * in pmap_growkernel.
    204  */
    205 static void
    206 kasan_md_init(void)
    207 {
    208 	extern struct bootspace bootspace;
    209 	size_t i;
    210 
    211 	CTASSERT((__MD_SHADOW_SIZE / NBPD_L4) == NL4_SLOT_KASAN);
    212 
    213 	/* Kernel. */
    214 	for (i = 0; i < BTSPACE_NSEGS; i++) {
    215 		if (bootspace.segs[i].type == BTSEG_NONE) {
    216 			continue;
    217 		}
    218 		kasan_shadow_map((void *)bootspace.segs[i].va,
    219 		    bootspace.segs[i].sz);
    220 	}
    221 
    222 	/* Boot region. */
    223 	kasan_shadow_map((void *)bootspace.boot.va, bootspace.boot.sz);
    224 
    225 	/* Module map. */
    226 	kasan_shadow_map((void *)bootspace.smodule,
    227 	    (size_t)(bootspace.emodule - bootspace.smodule));
    228 
    229 	/* The bootstrap spare va. */
    230 	kasan_shadow_map((void *)bootspace.spareva, PAGE_SIZE);
    231 }
    232 
    233 static inline bool
    234 __md_unwind_end(const char *name)
    235 {
    236 	if (!strcmp(name, "syscall") ||
    237 	    !strcmp(name, "alltraps") ||
    238 	    !strcmp(name, "handle_syscall") ||
    239 	    !strncmp(name, "Xtrap", 5) ||
    240 	    !strncmp(name, "Xintr", 5) ||
    241 	    !strncmp(name, "Xhandle", 7) ||
    242 	    !strncmp(name, "Xresume", 7) ||
    243 	    !strncmp(name, "Xstray", 6) ||
    244 	    !strncmp(name, "Xhold", 5) ||
    245 	    !strncmp(name, "Xrecurse", 8) ||
    246 	    !strcmp(name, "Xdoreti") ||
    247 	    !strncmp(name, "Xsoft", 5)) {
    248 		return true;
    249 	}
    250 
    251 	return false;
    252 }
    253 
    254 static void
    255 kasan_md_unwind(void)
    256 {
    257 	uint64_t *rbp, rip;
    258 	const char *mod;
    259 	const char *sym;
    260 	size_t nsym;
    261 	int error;
    262 
    263 	rbp = (uint64_t *)__builtin_frame_address(0);
    264 	nsym = 0;
    265 
    266 	while (1) {
    267 		/* 8(%rbp) contains the saved %rip. */
    268 		rip = *(rbp + 1);
    269 
    270 		if (rip < KERNBASE) {
    271 			break;
    272 		}
    273 		error = ksyms_getname(&mod, &sym, (vaddr_t)rip, KSYMS_PROC);
    274 		if (error) {
    275 			break;
    276 		}
    277 		printf("#%zu %p in %s <%s>\n", nsym, (void *)rip, sym, mod);
    278 		if (__md_unwind_end(sym)) {
    279 			break;
    280 		}
    281 
    282 		rbp = (uint64_t *)*(rbp);
    283 		if (rbp == 0) {
    284 			break;
    285 		}
    286 		nsym++;
    287 
    288 		if (nsym >= 15) {
    289 			break;
    290 		}
    291 	}
    292 }
    293