arch/x86/kernel/sys_x86_64.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/compat.h>
   3 #include <linux/errno.h>
   4 #include <linux/sched.h>
   5 #include <linux/sched/mm.h>
   6 #include <linux/syscalls.h>
   7 #include <linux/mm.h>
   8 #include <linux/fs.h>
   9 #include <linux/smp.h>
  10 #include <linux/sem.h>
  11 #include <linux/msg.h>
  12 #include <linux/shm.h>
  13 #include <linux/stat.h>
  14 #include <linux/mman.h>
  15 #include <linux/file.h>
  16 #include <linux/utsname.h>
  17 #include <linux/personality.h>
  18 #include <linux/random.h>
  19 #include <linux/uaccess.h>
  20 #include <linux/elf.h>
  21
  22 #include <asm/elf.h>
  23 #include <asm/ia32.h>
  24
  25 /*
  26  * Align a virtual address to avoid aliasing in the I$ on AMD F15h.
  27  */
  28 static unsigned long get_align_mask(void)
  29 {
  30         /* handle 32- and 64-bit case with a single conditional */
  31         if (va_align.flags < 0 || !(va_align.flags & (2 - mmap_is_ia32())))
  32                 return 0;
  33
  34         if (!(current->flags & PF_RANDOMIZE))
  35                 return 0;
  36
  37         return va_align.mask;
  38 }
  39
  40 /*
  41  * To avoid aliasing in the I$ on AMD F15h, the bits defined by the
  42  * va_align.bits, [12:upper_bit), are set to a random value instead of
  43  * zeroing them. This random value is computed once per boot. This form
  44  * of ASLR is known as "per-boot ASLR".
  45  *
  46  * To achieve this, the random value is added to the info.align_offset
  47  * value before calling vm_unmapped_area() or ORed directly to the
  48  * address.
  49  */
  50 static unsigned long get_align_bits(void)
  51 {
  52         return va_align.bits & get_align_mask();
  53 }
  54
  55 static int __init control_va_addr_alignment(char *str)
  56 {
  57         /* guard against enabling this on other CPU families */
  58         if (va_align.flags < 0)
  59                 return 1;
  60
  61         if (*str == 0)
  62                 return 1;
  63
  64         if (!strcmp(str, "32"))
  65                 va_align.flags = ALIGN_VA_32;
  66         else if (!strcmp(str, "64"))
  67                 va_align.flags = ALIGN_VA_64;
  68         else if (!strcmp(str, "off"))
  69                 va_align.flags = 0;
  70         else if (!strcmp(str, "on"))
  71                 va_align.flags = ALIGN_VA_32 | ALIGN_VA_64;
  72         else
  73                 pr_warn("invalid option value: 'align_va_addr=%s'\n", str);
  74
  75         return 1;
  76 }
  77 __setup("align_va_addr=", control_va_addr_alignment);
  78
  79 SYSCALL_DEFINE6(mmap, unsigned long, addr, unsigned long, len,
  80                 unsigned long, prot, unsigned long, flags,
  81                 unsigned long, fd, unsigned long, off)
  82 {
  83         if (off & ~PAGE_MASK)
  84                 return -EINVAL;
  85
  86         return ksys_mmap_pgoff(addr, len, prot, flags, fd, off >> PAGE_SHIFT);
  87 }
  88
  89 static void find_start_end(unsigned long addr, unsigned long flags,
  90                 unsigned long *begin, unsigned long *end)
  91 {
  92         if (!in_32bit_syscall() && (flags & MAP_32BIT)) {
  93                 /* This is usually used needed to map code in small
  94                    model, so it needs to be in the first 31bit. Limit
  95                    it to that.  This means we need to move the
  96                    unmapped base down for this case. This can give
  97                    conflicts with the heap, but we assume that glibc
  98                    malloc knows how to fall back to mmap. Give it 1GB
  99                    of playground for now. -AK */
 100                 *begin = 0x40000000;
 101                 *end = 0x80000000;
 102                 if (current->flags & PF_RANDOMIZE) {
 103                         *begin = randomize_page(*begin, 0x02000000);
 104                 }
 105                 return;
 106         }
 107
 108         *begin  = get_mmap_base(1);
 109         if (in_32bit_syscall())
 110                 *end = task_size_32bit();
 111         else
 112                 *end = task_size_64bit(addr > DEFAULT_MAP_WINDOW);
 113 }
 114
 115 static inline unsigned long stack_guard_placement(vm_flags_t vm_flags)
 116 {
 117         if (vm_flags & VM_SHADOW_STACK)
 118                 return PAGE_SIZE;
 119
 120         return 0;
 121 }
 122
 123 unsigned long
 124 arch_get_unmapped_area(struct file *filp, unsigned long addr, unsigned long len,
 125                        unsigned long pgoff, unsigned long flags, vm_flags_t vm_flags)
 126 {
 127         struct mm_struct *mm = current->mm;
 128         struct vm_area_struct *vma;
 129         struct vm_unmapped_area_info info = {};
 130         unsigned long begin, end;
 131
 132         if (flags & MAP_FIXED)
 133                 return addr;
 134
 135         find_start_end(addr, flags, &begin, &end);
 136
 137         if (len > end)
 138                 return -ENOMEM;
 139
 140         if (addr) {
 141                 addr = PAGE_ALIGN(addr);
 142                 vma = find_vma(mm, addr);
 143                 if (end - len >= addr &&
 144                     (!vma || addr + len <= vm_start_gap(vma)))
 145                         return addr;
 146         }
 147
 148         info.length = len;
 149         info.low_limit = begin;
 150         info.high_limit = end;
 151         info.align_offset = pgoff << PAGE_SHIFT;
 152         info.start_gap = stack_guard_placement(vm_flags);
 153         if (filp) {
 154                 info.align_mask = get_align_mask();
 155                 info.align_offset += get_align_bits();
 156         }
 157         return vm_unmapped_area(&info);
 158 }
 159
 160 unsigned long
 161 arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr0,
 162                           unsigned long len, unsigned long pgoff,
 163                           unsigned long flags, vm_flags_t vm_flags)
 164 {
 165         struct vm_area_struct *vma;
 166         struct mm_struct *mm = current->mm;
 167         unsigned long addr = addr0;
 168         struct vm_unmapped_area_info info = {};
 169
 170         /* requested length too big for entire address space */
 171         if (len > TASK_SIZE)
 172                 return -ENOMEM;
 173
 174         /* No address checking. See comment at mmap_address_hint_valid() */
 175         if (flags & MAP_FIXED)
 176                 return addr;
 177
 178         /* for MAP_32BIT mappings we force the legacy mmap base */
 179         if (!in_32bit_syscall() && (flags & MAP_32BIT))
 180                 goto bottomup;
 181
 182         /* requesting a specific address */
 183         if (addr) {
 184                 addr &= PAGE_MASK;
 185                 if (!mmap_address_hint_valid(addr, len))
 186                         goto get_unmapped_area;
 187
 188                 vma = find_vma(mm, addr);
 189                 if (!vma || addr + len <= vm_start_gap(vma))
 190                         return addr;
 191         }
 192 get_unmapped_area:
 193
 194         info.flags = VM_UNMAPPED_AREA_TOPDOWN;
 195         info.length = len;
 196         if (!in_32bit_syscall() && (flags & MAP_ABOVE4G))
 197                 info.low_limit = SZ_4G;
 198         else
 199                 info.low_limit = PAGE_SIZE;
 200
 201         info.high_limit = get_mmap_base(0);
 202         info.start_gap = stack_guard_placement(vm_flags);
 203
 204         /*
 205          * If hint address is above DEFAULT_MAP_WINDOW, look for unmapped area
 206          * in the full address space.
 207          *
 208          * !in_32bit_syscall() check to avoid high addresses for x32
 209          * (and make it no op on native i386).
 210          */
 211         if (addr > DEFAULT_MAP_WINDOW && !in_32bit_syscall())
 212                 info.high_limit += TASK_SIZE_MAX - DEFAULT_MAP_WINDOW;
 213
 214         info.align_offset = pgoff << PAGE_SHIFT;
 215         if (filp) {
 216                 info.align_mask = get_align_mask();
 217                 info.align_offset += get_align_bits();
 218         }
 219         addr = vm_unmapped_area(&info);
 220         if (!(addr & ~PAGE_MASK))
 221                 return addr;
 222         VM_BUG_ON(addr != -ENOMEM);
 223
 224 bottomup:
 225         /*
 226          * A failed mmap() very likely causes application failure,
 227          * so fall back to the bottom-up function here. This scenario
 228          * can happen with large stack limits and large mmap()
 229          * allocations.
 230          */
 231         return arch_get_unmapped_area(filp, addr0, len, pgoff, flags, 0);
 232 }