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 unsigned long align_vdso_addr(unsigned long addr)
  56 {
  57         unsigned long align_mask = get_align_mask();
  58         addr = (addr + align_mask) & ~align_mask;
  59         return addr | get_align_bits();
  60 }
  61
  62 static int __init control_va_addr_alignment(char *str)
  63 {
  64         /* guard against enabling this on other CPU families */
  65         if (va_align.flags < 0)
  66                 return 1;
  67
  68         if (*str == 0)
  69                 return 1;
  70
  71         if (*str == '=')
  72                 str++;
  73
  74         if (!strcmp(str, "32"))
  75                 va_align.flags = ALIGN_VA_32;
  76         else if (!strcmp(str, "64"))
  77                 va_align.flags = ALIGN_VA_64;
  78         else if (!strcmp(str, "off"))
  79                 va_align.flags = 0;
  80         else if (!strcmp(str, "on"))
  81                 va_align.flags = ALIGN_VA_32 | ALIGN_VA_64;
  82         else
  83                 return 0;
  84
  85         return 1;
  86 }
  87 __setup("align_va_addr", control_va_addr_alignment);
  88
  89 SYSCALL_DEFINE6(mmap, unsigned long, addr, unsigned long, len,
  90                 unsigned long, prot, unsigned long, flags,
  91                 unsigned long, fd, unsigned long, off)
  92 {
  93         long error;
  94         error = -EINVAL;
  95         if (off & ~PAGE_MASK)
  96                 goto out;
  97
  98         error = ksys_mmap_pgoff(addr, len, prot, flags, fd, off >> PAGE_SHIFT);
  99 out:
 100         return error;
 101 }
 102
 103 static void find_start_end(unsigned long addr, unsigned long flags,
 104                 unsigned long *begin, unsigned long *end)
 105 {
 106         if (!in_32bit_syscall() && (flags & MAP_32BIT)) {
 107                 /* This is usually used needed to map code in small
 108                    model, so it needs to be in the first 31bit. Limit
 109                    it to that.  This means we need to move the
 110                    unmapped base down for this case. This can give
 111                    conflicts with the heap, but we assume that glibc
 112                    malloc knows how to fall back to mmap. Give it 1GB
 113                    of playground for now. -AK */
 114                 *begin = 0x40000000;
 115                 *end = 0x80000000;
 116                 if (current->flags & PF_RANDOMIZE) {
 117                         *begin = randomize_page(*begin, 0x02000000);
 118                 }
 119                 return;
 120         }
 121
 122         *begin  = get_mmap_base(1);
 123         if (in_32bit_syscall())
 124                 *end = task_size_32bit();
 125         else
 126                 *end = task_size_64bit(addr > DEFAULT_MAP_WINDOW);
 127 }
 128
 129 unsigned long
 130 arch_get_unmapped_area(struct file *filp, unsigned long addr,
 131                 unsigned long len, unsigned long pgoff, unsigned long flags)
 132 {
 133         struct mm_struct *mm = current->mm;
 134         struct vm_area_struct *vma;
 135         struct vm_unmapped_area_info info;
 136         unsigned long begin, end;
 137
 138         if (flags & MAP_FIXED)
 139                 return addr;
 140
 141         find_start_end(addr, flags, &begin, &end);
 142
 143         if (len > end)
 144                 return -ENOMEM;
 145
 146         if (addr) {
 147                 addr = PAGE_ALIGN(addr);
 148                 vma = find_vma(mm, addr);
 149                 if (end - len >= addr &&
 150                     (!vma || addr + len <= vm_start_gap(vma)))
 151                         return addr;
 152         }
 153
 154         info.flags = 0;
 155         info.length = len;
 156         info.low_limit = begin;
 157         info.high_limit = end;
 158         info.align_mask = 0;
 159         info.align_offset = pgoff << PAGE_SHIFT;
 160         if (filp) {
 161                 info.align_mask = get_align_mask();
 162                 info.align_offset += get_align_bits();
 163         }
 164         return vm_unmapped_area(&info);
 165 }
 166
 167 unsigned long
 168 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
 169                           const unsigned long len, const unsigned long pgoff,
 170                           const unsigned long flags)
 171 {
 172         struct vm_area_struct *vma;
 173         struct mm_struct *mm = current->mm;
 174         unsigned long addr = addr0;
 175         struct vm_unmapped_area_info info;
 176
 177         /* requested length too big for entire address space */
 178         if (len > TASK_SIZE)
 179                 return -ENOMEM;
 180
 181         /* No address checking. See comment at mmap_address_hint_valid() */
 182         if (flags & MAP_FIXED)
 183                 return addr;
 184
 185         /* for MAP_32BIT mappings we force the legacy mmap base */
 186         if (!in_32bit_syscall() && (flags & MAP_32BIT))
 187                 goto bottomup;
 188
 189         /* requesting a specific address */
 190         if (addr) {
 191                 addr &= PAGE_MASK;
 192                 if (!mmap_address_hint_valid(addr, len))
 193                         goto get_unmapped_area;
 194
 195                 vma = find_vma(mm, addr);
 196                 if (!vma || addr + len <= vm_start_gap(vma))
 197                         return addr;
 198         }
 199 get_unmapped_area:
 200
 201         info.flags = VM_UNMAPPED_AREA_TOPDOWN;
 202         info.length = len;
 203         info.low_limit = PAGE_SIZE;
 204         info.high_limit = get_mmap_base(0);
 205
 206         /*
 207          * If hint address is above DEFAULT_MAP_WINDOW, look for unmapped area
 208          * in the full address space.
 209          *
 210          * !in_32bit_syscall() check to avoid high addresses for x32
 211          * (and make it no op on native i386).
 212          */
 213         if (addr > DEFAULT_MAP_WINDOW && !in_32bit_syscall())
 214                 info.high_limit += TASK_SIZE_MAX - DEFAULT_MAP_WINDOW;
 215
 216         info.align_mask = 0;
 217         info.align_offset = pgoff << PAGE_SHIFT;
 218         if (filp) {
 219                 info.align_mask = get_align_mask();
 220                 info.align_offset += get_align_bits();
 221         }
 222         addr = vm_unmapped_area(&info);
 223         if (!(addr & ~PAGE_MASK))
 224                 return addr;
 225         VM_BUG_ON(addr != -ENOMEM);
 226
 227 bottomup:
 228         /*
 229          * A failed mmap() very likely causes application failure,
 230          * so fall back to the bottom-up function here. This scenario
 231          * can happen with large stack limits and large mmap()
 232          * allocations.
 233          */
 234         return arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
 235 }