arch/x86/mm/pti.c

   1 /*
   2  * Copyright(c) 2017 Intel Corporation. All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of version 2 of the GNU General Public License as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * This code is based in part on work published here:
  14  *
  15  *      https://github.com/IAIK/KAISER
  16  *
  17  * The original work was written by and and signed off by for the Linux
  18  * kernel by:
  19  *
  20  *   Signed-off-by: Richard Fellner <richard.fellner@student.tugraz.at>
  21  *   Signed-off-by: Moritz Lipp <moritz.lipp@iaik.tugraz.at>
  22  *   Signed-off-by: Daniel Gruss <daniel.gruss@iaik.tugraz.at>
  23  *   Signed-off-by: Michael Schwarz <michael.schwarz@iaik.tugraz.at>
  24  *
  25  * Major changes to the original code by: Dave Hansen <dave.hansen@intel.com>
  26  * Mostly rewritten by Thomas Gleixner <tglx@linutronix.de> and
  27  *                     Andy Lutomirsky <luto@amacapital.net>
  28  */
  29 #include <linux/kernel.h>
  30 #include <linux/errno.h>
  31 #include <linux/string.h>
  32 #include <linux/types.h>
  33 #include <linux/bug.h>
  34 #include <linux/init.h>
  35 #include <linux/spinlock.h>
  36 #include <linux/mm.h>
  37 #include <linux/uaccess.h>
  38
  39 #include <asm/cpufeature.h>
  40 #include <asm/hypervisor.h>
  41 #include <asm/vsyscall.h>
  42 #include <asm/cmdline.h>
  43 #include <asm/pti.h>
  44 #include <asm/pgtable.h>
  45 #include <asm/pgalloc.h>
  46 #include <asm/tlbflush.h>
  47 #include <asm/desc.h>
  48
  49 #undef pr_fmt
  50 #define pr_fmt(fmt)     "Kernel/User page tables isolation: " fmt
  51
  52 /* Backporting helper */
  53 #ifndef __GFP_NOTRACK
  54 #define __GFP_NOTRACK   0
  55 #endif
  56
  57 static void __init pti_print_if_insecure(const char *reason)
  58 {
  59         if (boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
  60                 pr_info("%s\n", reason);
  61 }
  62
  63 static void __init pti_print_if_secure(const char *reason)
  64 {
  65         if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
  66                 pr_info("%s\n", reason);
  67 }
  68
  69 void __init pti_check_boottime_disable(void)
  70 {
  71         char arg[5];
  72         int ret;
  73
  74         if (hypervisor_is_type(X86_HYPER_XEN_PV)) {
  75                 pti_print_if_insecure("disabled on XEN PV.");
  76                 return;
  77         }
  78
  79         ret = cmdline_find_option(boot_command_line, "pti", arg, sizeof(arg));
  80         if (ret > 0)  {
  81                 if (ret == 3 && !strncmp(arg, "off", 3)) {
  82                         pti_print_if_insecure("disabled on command line.");
  83                         return;
  84                 }
  85                 if (ret == 2 && !strncmp(arg, "on", 2)) {
  86                         pti_print_if_secure("force enabled on command line.");
  87                         goto enable;
  88                 }
  89                 if (ret == 4 && !strncmp(arg, "auto", 4))
  90                         goto autosel;
  91         }
  92
  93         if (cmdline_find_option_bool(boot_command_line, "nopti")) {
  94                 pti_print_if_insecure("disabled on command line.");
  95                 return;
  96         }
  97
  98 autosel:
  99         if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
 100                 return;
 101 enable:
 102         setup_force_cpu_cap(X86_FEATURE_PTI);
 103 }
 104
 105 pgd_t __pti_set_user_pgd(pgd_t *pgdp, pgd_t pgd)
 106 {
 107         /*
 108          * Changes to the high (kernel) portion of the kernelmode page
 109          * tables are not automatically propagated to the usermode tables.
 110          *
 111          * Users should keep in mind that, unlike the kernelmode tables,
 112          * there is no vmalloc_fault equivalent for the usermode tables.
 113          * Top-level entries added to init_mm's usermode pgd after boot
 114          * will not be automatically propagated to other mms.
 115          */
 116         if (!pgdp_maps_userspace(pgdp))
 117                 return pgd;
 118
 119         /*
 120          * The user page tables get the full PGD, accessible from
 121          * userspace:
 122          */
 123         kernel_to_user_pgdp(pgdp)->pgd = pgd.pgd;
 124
 125         /*
 126          * If this is normal user memory, make it NX in the kernel
 127          * pagetables so that, if we somehow screw up and return to
 128          * usermode with the kernel CR3 loaded, we'll get a page fault
 129          * instead of allowing user code to execute with the wrong CR3.
 130          *
 131          * As exceptions, we don't set NX if:
 132          *  - _PAGE_USER is not set.  This could be an executable
 133          *     EFI runtime mapping or something similar, and the kernel
 134          *     may execute from it
 135          *  - we don't have NX support
 136          *  - we're clearing the PGD (i.e. the new pgd is not present).
 137          */
 138         if ((pgd.pgd & (_PAGE_USER|_PAGE_PRESENT)) == (_PAGE_USER|_PAGE_PRESENT) &&
 139             (__supported_pte_mask & _PAGE_NX))
 140                 pgd.pgd |= _PAGE_NX;
 141
 142         /* return the copy of the PGD we want the kernel to use: */
 143         return pgd;
 144 }
 145
 146 /*
 147  * Walk the user copy of the page tables (optionally) trying to allocate
 148  * page table pages on the way down.
 149  *
 150  * Returns a pointer to a P4D on success, or NULL on failure.
 151  */
 152 static p4d_t *pti_user_pagetable_walk_p4d(unsigned long address)
 153 {
 154         pgd_t *pgd = kernel_to_user_pgdp(pgd_offset_k(address));
 155         gfp_t gfp = (GFP_KERNEL | __GFP_NOTRACK | __GFP_ZERO);
 156
 157         if (address < PAGE_OFFSET) {
 158                 WARN_ONCE(1, "attempt to walk user address\n");
 159                 return NULL;
 160         }
 161
 162         if (pgd_none(*pgd)) {
 163                 unsigned long new_p4d_page = __get_free_page(gfp);
 164                 if (!new_p4d_page)
 165                         return NULL;
 166
 167                 if (pgd_none(*pgd)) {
 168                         set_pgd(pgd, __pgd(_KERNPG_TABLE | __pa(new_p4d_page)));
 169                         new_p4d_page = 0;
 170                 }
 171                 if (new_p4d_page)
 172                         free_page(new_p4d_page);
 173         }
 174         BUILD_BUG_ON(pgd_large(*pgd) != 0);
 175
 176         return p4d_offset(pgd, address);
 177 }
 178
 179 /*
 180  * Walk the user copy of the page tables (optionally) trying to allocate
 181  * page table pages on the way down.
 182  *
 183  * Returns a pointer to a PMD on success, or NULL on failure.
 184  */
 185 static pmd_t *pti_user_pagetable_walk_pmd(unsigned long address)
 186 {
 187         gfp_t gfp = (GFP_KERNEL | __GFP_NOTRACK | __GFP_ZERO);
 188         p4d_t *p4d = pti_user_pagetable_walk_p4d(address);
 189         pud_t *pud;
 190
 191         BUILD_BUG_ON(p4d_large(*p4d) != 0);
 192         if (p4d_none(*p4d)) {
 193                 unsigned long new_pud_page = __get_free_page(gfp);
 194                 if (!new_pud_page)
 195                         return NULL;
 196
 197                 if (p4d_none(*p4d)) {
 198                         set_p4d(p4d, __p4d(_KERNPG_TABLE | __pa(new_pud_page)));
 199                         new_pud_page = 0;
 200                 }
 201                 if (new_pud_page)
 202                         free_page(new_pud_page);
 203         }
 204
 205         pud = pud_offset(p4d, address);
 206         /* The user page tables do not use large mappings: */
 207         if (pud_large(*pud)) {
 208                 WARN_ON(1);
 209                 return NULL;
 210         }
 211         if (pud_none(*pud)) {
 212                 unsigned long new_pmd_page = __get_free_page(gfp);
 213                 if (!new_pmd_page)
 214                         return NULL;
 215
 216                 if (pud_none(*pud)) {
 217                         set_pud(pud, __pud(_KERNPG_TABLE | __pa(new_pmd_page)));
 218                         new_pmd_page = 0;
 219                 }
 220                 if (new_pmd_page)
 221                         free_page(new_pmd_page);
 222         }
 223
 224         return pmd_offset(pud, address);
 225 }
 226
 227 #ifdef CONFIG_X86_VSYSCALL_EMULATION
 228 /*
 229  * Walk the shadow copy of the page tables (optionally) trying to allocate
 230  * page table pages on the way down.  Does not support large pages.
 231  *
 232  * Note: this is only used when mapping *new* kernel data into the
 233  * user/shadow page tables.  It is never used for userspace data.
 234  *
 235  * Returns a pointer to a PTE on success, or NULL on failure.
 236  */
 237 static __init pte_t *pti_user_pagetable_walk_pte(unsigned long address)
 238 {
 239         gfp_t gfp = (GFP_KERNEL | __GFP_NOTRACK | __GFP_ZERO);
 240         pmd_t *pmd = pti_user_pagetable_walk_pmd(address);
 241         pte_t *pte;
 242
 243         /* We can't do anything sensible if we hit a large mapping. */
 244         if (pmd_large(*pmd)) {
 245                 WARN_ON(1);
 246                 return NULL;
 247         }
 248
 249         if (pmd_none(*pmd)) {
 250                 unsigned long new_pte_page = __get_free_page(gfp);
 251                 if (!new_pte_page)
 252                         return NULL;
 253
 254                 if (pmd_none(*pmd)) {
 255                         set_pmd(pmd, __pmd(_KERNPG_TABLE | __pa(new_pte_page)));
 256                         new_pte_page = 0;
 257                 }
 258                 if (new_pte_page)
 259                         free_page(new_pte_page);
 260         }
 261
 262         pte = pte_offset_kernel(pmd, address);
 263         if (pte_flags(*pte) & _PAGE_USER) {
 264                 WARN_ONCE(1, "attempt to walk to user pte\n");
 265                 return NULL;
 266         }
 267         return pte;
 268 }
 269
 270 static void __init pti_setup_vsyscall(void)
 271 {
 272         pte_t *pte, *target_pte;
 273         unsigned int level;
 274
 275         pte = lookup_address(VSYSCALL_ADDR, &level);
 276         if (!pte || WARN_ON(level != PG_LEVEL_4K) || pte_none(*pte))
 277                 return;
 278
 279         target_pte = pti_user_pagetable_walk_pte(VSYSCALL_ADDR);
 280         if (WARN_ON(!target_pte))
 281                 return;
 282
 283         *target_pte = *pte;
 284         set_vsyscall_pgtable_user_bits(kernel_to_user_pgdp(swapper_pg_dir));
 285 }
 286 #else
 287 static void __init pti_setup_vsyscall(void) { }
 288 #endif
 289
 290 static void __init
 291 pti_clone_pmds(unsigned long start, unsigned long end, pmdval_t clear)
 292 {
 293         unsigned long addr;
 294
 295         /*
 296          * Clone the populated PMDs which cover start to end. These PMD areas
 297          * can have holes.
 298          */
 299         for (addr = start; addr < end; addr += PMD_SIZE) {
 300                 pmd_t *pmd, *target_pmd;
 301                 pgd_t *pgd;
 302                 p4d_t *p4d;
 303                 pud_t *pud;
 304
 305                 pgd = pgd_offset_k(addr);
 306                 if (WARN_ON(pgd_none(*pgd)))
 307                         return;
 308                 p4d = p4d_offset(pgd, addr);
 309                 if (WARN_ON(p4d_none(*p4d)))
 310                         return;
 311                 pud = pud_offset(p4d, addr);
 312                 if (pud_none(*pud))
 313                         continue;
 314                 pmd = pmd_offset(pud, addr);
 315                 if (pmd_none(*pmd))
 316                         continue;
 317
 318                 target_pmd = pti_user_pagetable_walk_pmd(addr);
 319                 if (WARN_ON(!target_pmd))
 320                         return;
 321
 322                 /*
 323                  * Copy the PMD.  That is, the kernelmode and usermode
 324                  * tables will share the last-level page tables of this
 325                  * address range
 326                  */
 327                 *target_pmd = pmd_clear_flags(*pmd, clear);
 328         }
 329 }
 330
 331 /*
 332  * Clone a single p4d (i.e. a top-level entry on 4-level systems and a
 333  * next-level entry on 5-level systems.
 334  */
 335 static void __init pti_clone_p4d(unsigned long addr)
 336 {
 337         p4d_t *kernel_p4d, *user_p4d;
 338         pgd_t *kernel_pgd;
 339
 340         user_p4d = pti_user_pagetable_walk_p4d(addr);
 341         kernel_pgd = pgd_offset_k(addr);
 342         kernel_p4d = p4d_offset(kernel_pgd, addr);
 343         *user_p4d = *kernel_p4d;
 344 }
 345
 346 /*
 347  * Clone the CPU_ENTRY_AREA into the user space visible page table.
 348  */
 349 static void __init pti_clone_user_shared(void)
 350 {
 351         pti_clone_p4d(CPU_ENTRY_AREA_BASE);
 352 }
 353
 354 /*
 355  * Clone the ESPFIX P4D into the user space visinble page table
 356  */
 357 static void __init pti_setup_espfix64(void)
 358 {
 359 #ifdef CONFIG_X86_ESPFIX64
 360         pti_clone_p4d(ESPFIX_BASE_ADDR);
 361 #endif
 362 }
 363
 364 /*
 365  * Clone the populated PMDs of the entry and irqentry text and force it RO.
 366  */
 367 static void __init pti_clone_entry_text(void)
 368 {
 369         pti_clone_pmds((unsigned long) __entry_text_start,
 370                         (unsigned long) __irqentry_text_end,
 371                        _PAGE_RW | _PAGE_GLOBAL);
 372 }
 373
 374 /*
 375  * Initialize kernel page table isolation
 376  */
 377 void __init pti_init(void)
 378 {
 379         if (!static_cpu_has(X86_FEATURE_PTI))
 380                 return;
 381
 382         pr_info("enabled\n");
 383
 384         pti_clone_user_shared();
 385         pti_clone_entry_text();
 386         pti_setup_espfix64();
 387         pti_setup_vsyscall();
 388 }