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 __init 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                 set_pgd(pgd, __pgd(_KERNPG_TABLE | __pa(new_p4d_page)));
 168         }
 169         BUILD_BUG_ON(pgd_large(*pgd) != 0);
 170
 171         return p4d_offset(pgd, address);
 172 }
 173
 174 /*
 175  * Walk the user copy of the page tables (optionally) trying to allocate
 176  * page table pages on the way down.
 177  *
 178  * Returns a pointer to a PMD on success, or NULL on failure.
 179  */
 180 static __init pmd_t *pti_user_pagetable_walk_pmd(unsigned long address)
 181 {
 182         gfp_t gfp = (GFP_KERNEL | __GFP_NOTRACK | __GFP_ZERO);
 183         p4d_t *p4d = pti_user_pagetable_walk_p4d(address);
 184         pud_t *pud;
 185
 186         BUILD_BUG_ON(p4d_large(*p4d) != 0);
 187         if (p4d_none(*p4d)) {
 188                 unsigned long new_pud_page = __get_free_page(gfp);
 189                 if (!new_pud_page)
 190                         return NULL;
 191
 192                 set_p4d(p4d, __p4d(_KERNPG_TABLE | __pa(new_pud_page)));
 193         }
 194
 195         pud = pud_offset(p4d, address);
 196         /* The user page tables do not use large mappings: */
 197         if (pud_large(*pud)) {
 198                 WARN_ON(1);
 199                 return NULL;
 200         }
 201         if (pud_none(*pud)) {
 202                 unsigned long new_pmd_page = __get_free_page(gfp);
 203                 if (!new_pmd_page)
 204                         return NULL;
 205
 206                 set_pud(pud, __pud(_KERNPG_TABLE | __pa(new_pmd_page)));
 207         }
 208
 209         return pmd_offset(pud, address);
 210 }
 211
 212 #ifdef CONFIG_X86_VSYSCALL_EMULATION
 213 /*
 214  * Walk the shadow copy of the page tables (optionally) trying to allocate
 215  * page table pages on the way down.  Does not support large pages.
 216  *
 217  * Note: this is only used when mapping *new* kernel data into the
 218  * user/shadow page tables.  It is never used for userspace data.
 219  *
 220  * Returns a pointer to a PTE on success, or NULL on failure.
 221  */
 222 static __init pte_t *pti_user_pagetable_walk_pte(unsigned long address)
 223 {
 224         gfp_t gfp = (GFP_KERNEL | __GFP_NOTRACK | __GFP_ZERO);
 225         pmd_t *pmd = pti_user_pagetable_walk_pmd(address);
 226         pte_t *pte;
 227
 228         /* We can't do anything sensible if we hit a large mapping. */
 229         if (pmd_large(*pmd)) {
 230                 WARN_ON(1);
 231                 return NULL;
 232         }
 233
 234         if (pmd_none(*pmd)) {
 235                 unsigned long new_pte_page = __get_free_page(gfp);
 236                 if (!new_pte_page)
 237                         return NULL;
 238
 239                 set_pmd(pmd, __pmd(_KERNPG_TABLE | __pa(new_pte_page)));
 240         }
 241
 242         pte = pte_offset_kernel(pmd, address);
 243         if (pte_flags(*pte) & _PAGE_USER) {
 244                 WARN_ONCE(1, "attempt to walk to user pte\n");
 245                 return NULL;
 246         }
 247         return pte;
 248 }
 249
 250 static void __init pti_setup_vsyscall(void)
 251 {
 252         pte_t *pte, *target_pte;
 253         unsigned int level;
 254
 255         pte = lookup_address(VSYSCALL_ADDR, &level);
 256         if (!pte || WARN_ON(level != PG_LEVEL_4K) || pte_none(*pte))
 257                 return;
 258
 259         target_pte = pti_user_pagetable_walk_pte(VSYSCALL_ADDR);
 260         if (WARN_ON(!target_pte))
 261                 return;
 262
 263         *target_pte = *pte;
 264         set_vsyscall_pgtable_user_bits(kernel_to_user_pgdp(swapper_pg_dir));
 265 }
 266 #else
 267 static void __init pti_setup_vsyscall(void) { }
 268 #endif
 269
 270 static void __init
 271 pti_clone_pmds(unsigned long start, unsigned long end, pmdval_t clear)
 272 {
 273         unsigned long addr;
 274
 275         /*
 276          * Clone the populated PMDs which cover start to end. These PMD areas
 277          * can have holes.
 278          */
 279         for (addr = start; addr < end; addr += PMD_SIZE) {
 280                 pmd_t *pmd, *target_pmd;
 281                 pgd_t *pgd;
 282                 p4d_t *p4d;
 283                 pud_t *pud;
 284
 285                 pgd = pgd_offset_k(addr);
 286                 if (WARN_ON(pgd_none(*pgd)))
 287                         return;
 288                 p4d = p4d_offset(pgd, addr);
 289                 if (WARN_ON(p4d_none(*p4d)))
 290                         return;
 291                 pud = pud_offset(p4d, addr);
 292                 if (pud_none(*pud))
 293                         continue;
 294                 pmd = pmd_offset(pud, addr);
 295                 if (pmd_none(*pmd))
 296                         continue;
 297
 298                 target_pmd = pti_user_pagetable_walk_pmd(addr);
 299                 if (WARN_ON(!target_pmd))
 300                         return;
 301
 302                 /*
 303                  * Copy the PMD.  That is, the kernelmode and usermode
 304                  * tables will share the last-level page tables of this
 305                  * address range
 306                  */
 307                 *target_pmd = pmd_clear_flags(*pmd, clear);
 308         }
 309 }
 310
 311 /*
 312  * Clone a single p4d (i.e. a top-level entry on 4-level systems and a
 313  * next-level entry on 5-level systems.
 314  */
 315 static void __init pti_clone_p4d(unsigned long addr)
 316 {
 317         p4d_t *kernel_p4d, *user_p4d;
 318         pgd_t *kernel_pgd;
 319
 320         user_p4d = pti_user_pagetable_walk_p4d(addr);
 321         kernel_pgd = pgd_offset_k(addr);
 322         kernel_p4d = p4d_offset(kernel_pgd, addr);
 323         *user_p4d = *kernel_p4d;
 324 }
 325
 326 /*
 327  * Clone the CPU_ENTRY_AREA into the user space visible page table.
 328  */
 329 static void __init pti_clone_user_shared(void)
 330 {
 331         pti_clone_p4d(CPU_ENTRY_AREA_BASE);
 332 }
 333
 334 /*
 335  * Clone the ESPFIX P4D into the user space visinble page table
 336  */
 337 static void __init pti_setup_espfix64(void)
 338 {
 339 #ifdef CONFIG_X86_ESPFIX64
 340         pti_clone_p4d(ESPFIX_BASE_ADDR);
 341 #endif
 342 }
 343
 344 /*
 345  * Clone the populated PMDs of the entry and irqentry text and force it RO.
 346  */
 347 static void __init pti_clone_entry_text(void)
 348 {
 349         pti_clone_pmds((unsigned long) __entry_text_start,
 350                         (unsigned long) __irqentry_text_end,
 351                        _PAGE_RW | _PAGE_GLOBAL);
 352 }
 353
 354 /*
 355  * Initialize kernel page table isolation
 356  */
 357 void __init pti_init(void)
 358 {
 359         if (!static_cpu_has(X86_FEATURE_PTI))
 360                 return;
 361
 362         pr_info("enabled\n");
 363
 364         pti_clone_user_shared();
 365         pti_clone_entry_text();
 366         pti_setup_espfix64();
 367         pti_setup_vsyscall();
 368 }