patches/kernel/0196-x86-mm-pti-Add-functions-to-clone-kernel-PMDs.patch

   1 From bcb5ffbfba8c6c557ad536eb9084040b8e52923e Mon Sep 17 00:00:00 2001
   2 From: Andy Lutomirski <luto@kernel.org>
   3 Date: Mon, 4 Dec 2017 15:07:42 +0100
   4 Subject: [PATCH 196/241] x86/mm/pti: Add functions to clone kernel PMDs
   5 MIME-Version: 1.0
   6 Content-Type: text/plain; charset=UTF-8
   7 Content-Transfer-Encoding: 8bit
   8
   9 CVE-2017-5754
  10
  11 Provide infrastructure to:
  12
  13  - find a kernel PMD for a mapping which must be visible to user space for
  14    the entry/exit code to work.
  15
  16  - walk an address range and share the kernel PMD with it.
  17
  18 This reuses a small part of the original KAISER patches to populate the
  19 user space page table.
  20
  21 [ tglx: Made it universally usable so it can be used for any kind of shared
  22         mapping. Add a mechanism to clear specific bits in the user space
  23         visible PMD entry. Folded Andys simplifactions ]
  24
  25 Originally-by: Dave Hansen <dave.hansen@linux.intel.com>
  26 Signed-off-by: Andy Lutomirski <luto@kernel.org>
  27 Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
  28 Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
  29 Cc: Borislav Petkov <bp@alien8.de>
  30 Cc: Borislav Petkov <bpetkov@suse.de>
  31 Cc: Brian Gerst <brgerst@gmail.com>
  32 Cc: Dave Hansen <dave.hansen@linux.intel.com>
  33 Cc: David Laight <David.Laight@aculab.com>
  34 Cc: Denys Vlasenko <dvlasenk@redhat.com>
  35 Cc: Eduardo Valentin <eduval@amazon.com>
  36 Cc: Greg KH <gregkh@linuxfoundation.org>
  37 Cc: H. Peter Anvin <hpa@zytor.com>
  38 Cc: Josh Poimboeuf <jpoimboe@redhat.com>
  39 Cc: Juergen Gross <jgross@suse.com>
  40 Cc: Linus Torvalds <torvalds@linux-foundation.org>
  41 Cc: Peter Zijlstra <peterz@infradead.org>
  42 Cc: Will Deacon <will.deacon@arm.com>
  43 Cc: aliguori@amazon.com
  44 Cc: daniel.gruss@iaik.tugraz.at
  45 Cc: hughd@google.com
  46 Cc: keescook@google.com
  47 Signed-off-by: Ingo Molnar <mingo@kernel.org>
  48 (cherry picked from commit 03f4424f348e8be95eb1bbeba09461cd7b867828)
  49 Signed-off-by: Andy Whitcroft <apw@canonical.com>
  50 Signed-off-by: Kleber Sacilotto de Souza <kleber.souza@canonical.com>
  51 (cherry picked from commit 262ab7e8665e88581d20ccaefa107340457224bb)
  52 Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
  53 ---
  54  arch/x86/mm/pti.c | 127 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
  55  1 file changed, 127 insertions(+)
  56
  57 diff --git a/arch/x86/mm/pti.c b/arch/x86/mm/pti.c
  58 index 69a983365392..d58bcee470fc 100644
  59 --- a/arch/x86/mm/pti.c
  60 +++ b/arch/x86/mm/pti.c
  61 @@ -48,6 +48,11 @@
  62  #undef pr_fmt
  63  #define pr_fmt(fmt)     "Kernel/User page tables isolation: " fmt
  64
  65 +/* Backporting helper */
  66 +#ifndef __GFP_NOTRACK
  67 +#define __GFP_NOTRACK  0
  68 +#endif
  69 +
  70  static void __init pti_print_if_insecure(const char *reason)
  71  {
  72         if (boot_cpu_has_bug(X86_BUG_CPU_INSECURE))
  73 @@ -137,6 +142,128 @@ pgd_t __pti_set_user_pgd(pgd_t *pgdp, pgd_t pgd)
  74         return pgd;
  75  }
  76
  77 +/*
  78 + * Walk the user copy of the page tables (optionally) trying to allocate
  79 + * page table pages on the way down.
  80 + *
  81 + * Returns a pointer to a P4D on success, or NULL on failure.
  82 + */
  83 +static p4d_t *pti_user_pagetable_walk_p4d(unsigned long address)
  84 +{
  85 +       pgd_t *pgd = kernel_to_user_pgdp(pgd_offset_k(address));
  86 +       gfp_t gfp = (GFP_KERNEL | __GFP_NOTRACK | __GFP_ZERO);
  87 +
  88 +       if (address < PAGE_OFFSET) {
  89 +               WARN_ONCE(1, "attempt to walk user address\n");
  90 +               return NULL;
  91 +       }
  92 +
  93 +       if (pgd_none(*pgd)) {
  94 +               unsigned long new_p4d_page = __get_free_page(gfp);
  95 +               if (!new_p4d_page)
  96 +                       return NULL;
  97 +
  98 +               if (pgd_none(*pgd)) {
  99 +                       set_pgd(pgd, __pgd(_KERNPG_TABLE | __pa(new_p4d_page)));
 100 +                       new_p4d_page = 0;
 101 +               }
 102 +               if (new_p4d_page)
 103 +                       free_page(new_p4d_page);
 104 +       }
 105 +       BUILD_BUG_ON(pgd_large(*pgd) != 0);
 106 +
 107 +       return p4d_offset(pgd, address);
 108 +}
 109 +
 110 +/*
 111 + * Walk the user copy of the page tables (optionally) trying to allocate
 112 + * page table pages on the way down.
 113 + *
 114 + * Returns a pointer to a PMD on success, or NULL on failure.
 115 + */
 116 +static pmd_t *pti_user_pagetable_walk_pmd(unsigned long address)
 117 +{
 118 +       gfp_t gfp = (GFP_KERNEL | __GFP_NOTRACK | __GFP_ZERO);
 119 +       p4d_t *p4d = pti_user_pagetable_walk_p4d(address);
 120 +       pud_t *pud;
 121 +
 122 +       BUILD_BUG_ON(p4d_large(*p4d) != 0);
 123 +       if (p4d_none(*p4d)) {
 124 +               unsigned long new_pud_page = __get_free_page(gfp);
 125 +               if (!new_pud_page)
 126 +                       return NULL;
 127 +
 128 +               if (p4d_none(*p4d)) {
 129 +                       set_p4d(p4d, __p4d(_KERNPG_TABLE | __pa(new_pud_page)));
 130 +                       new_pud_page = 0;
 131 +               }
 132 +               if (new_pud_page)
 133 +                       free_page(new_pud_page);
 134 +       }
 135 +
 136 +       pud = pud_offset(p4d, address);
 137 +       /* The user page tables do not use large mappings: */
 138 +       if (pud_large(*pud)) {
 139 +               WARN_ON(1);
 140 +               return NULL;
 141 +       }
 142 +       if (pud_none(*pud)) {
 143 +               unsigned long new_pmd_page = __get_free_page(gfp);
 144 +               if (!new_pmd_page)
 145 +                       return NULL;
 146 +
 147 +               if (pud_none(*pud)) {
 148 +                       set_pud(pud, __pud(_KERNPG_TABLE | __pa(new_pmd_page)));
 149 +                       new_pmd_page = 0;
 150 +               }
 151 +               if (new_pmd_page)
 152 +                       free_page(new_pmd_page);
 153 +       }
 154 +
 155 +       return pmd_offset(pud, address);
 156 +}
 157 +
 158 +static void __init
 159 +pti_clone_pmds(unsigned long start, unsigned long end, pmdval_t clear)
 160 +{
 161 +       unsigned long addr;
 162 +
 163 +       /*
 164 +        * Clone the populated PMDs which cover start to end. These PMD areas
 165 +        * can have holes.
 166 +        */
 167 +       for (addr = start; addr < end; addr += PMD_SIZE) {
 168 +               pmd_t *pmd, *target_pmd;
 169 +               pgd_t *pgd;
 170 +               p4d_t *p4d;
 171 +               pud_t *pud;
 172 +
 173 +               pgd = pgd_offset_k(addr);
 174 +               if (WARN_ON(pgd_none(*pgd)))
 175 +                       return;
 176 +               p4d = p4d_offset(pgd, addr);
 177 +               if (WARN_ON(p4d_none(*p4d)))
 178 +                       return;
 179 +               pud = pud_offset(p4d, addr);
 180 +               if (pud_none(*pud))
 181 +                       continue;
 182 +               pmd = pmd_offset(pud, addr);
 183 +               if (pmd_none(*pmd))
 184 +                       continue;
 185 +
 186 +               target_pmd = pti_user_pagetable_walk_pmd(addr);
 187 +               if (WARN_ON(!target_pmd))
 188 +                       return;
 189 +
 190 +               /*
 191 +                * Copy the PMD.  That is, the kernelmode and usermode
 192 +                * tables will share the last-level page tables of this
 193 +                * address range
 194 +                */
 195 +               *target_pmd = pmd_clear_flags(*pmd, clear);
 196 +       }
 197 +}
 198 +
 199  /*
 200   * Initialize kernel page table isolation
 201   */
 202 --
 203 2.14.2
 204