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 #include <asm/syscalls.h>
  25
  26 /*
  27  * Align a virtual address to avoid aliasing in the I$ on AMD F15h.
  28  */
  29 static unsigned long get_align_mask(void)
  30 {
  31         /* handle 32- and 64-bit case with a single conditional */
  32         if (va_align.flags < 0 || !(va_align.flags & (2 - mmap_is_ia32())))
  33                 return 0;
  34
  35         if (!(current->flags & PF_RANDOMIZE))
  36                 return 0;
  37
  38         return va_align.mask;
  39 }
  40
  41 /*
  42  * To avoid aliasing in the I$ on AMD F15h, the bits defined by the
  43  * va_align.bits, [12:upper_bit), are set to a random value instead of
  44  * zeroing them. This random value is computed once per boot. This form
  45  * of ASLR is known as "per-boot ASLR".
  46  *
  47  * To achieve this, the random value is added to the info.align_offset
  48  * value before calling vm_unmapped_area() or ORed directly to the
  49  * address.
  50  */
  51 static unsigned long get_align_bits(void)
  52 {
  53         return va_align.bits & get_align_mask();
  54 }
  55
  56 unsigned long align_vdso_addr(unsigned long addr)
  57 {
  58         unsigned long align_mask = get_align_mask();
  59         addr = (addr + align_mask) & ~align_mask;
  60         return addr | get_align_bits();
  61 }
  62
  63 static int __init control_va_addr_alignment(char *str)
  64 {
  65         /* guard against enabling this on other CPU families */
  66         if (va_align.flags < 0)
  67                 return 1;
  68
  69         if (*str == 0)
  70                 return 1;
  71
  72         if (*str == '=')
  73                 str++;
  74
  75         if (!strcmp(str, "32"))
  76                 va_align.flags = ALIGN_VA_32;
  77         else if (!strcmp(str, "64"))
  78                 va_align.flags = ALIGN_VA_64;
  79         else if (!strcmp(str, "off"))
  80                 va_align.flags = 0;
  81         else if (!strcmp(str, "on"))
  82                 va_align.flags = ALIGN_VA_32 | ALIGN_VA_64;
  83         else
  84                 return 0;
  85
  86         return 1;
  87 }
  88 __setup("align_va_addr", control_va_addr_alignment);
  89
  90 SYSCALL_DEFINE6(mmap, unsigned long, addr, unsigned long, len,
  91                 unsigned long, prot, unsigned long, flags,
  92                 unsigned long, fd, unsigned long, off)
  93 {
  94         long error;
  95         error = -EINVAL;
  96         if (off & ~PAGE_MASK)
  97                 goto out;
  98
  99         error = ksys_mmap_pgoff(addr, len, prot, flags, fd, off >> PAGE_SHIFT);
 100 out:
 101         return error;
 102 }
 103
 104 static void find_start_end(unsigned long addr, unsigned long flags,
 105                 unsigned long *begin, unsigned long *end)
 106 {
 107         if (!in_32bit_syscall() && (flags & MAP_32BIT)) {
 108                 /* This is usually used needed to map code in small
 109                    model, so it needs to be in the first 31bit. Limit
 110                    it to that.  This means we need to move the
 111                    unmapped base down for this case. This can give
 112                    conflicts with the heap, but we assume that glibc
 113                    malloc knows how to fall back to mmap. Give it 1GB
 114                    of playground for now. -AK */
 115                 *begin = 0x40000000;
 116                 *end = 0x80000000;
 117                 if (current->flags & PF_RANDOMIZE) {
 118                         *begin = randomize_page(*begin, 0x02000000);
 119                 }
 120                 return;
 121         }
 122
 123         *begin  = get_mmap_base(1);
 124         if (in_32bit_syscall())
 125                 *end = task_size_32bit();
 126         else
 127                 *end = task_size_64bit(addr > DEFAULT_MAP_WINDOW);
 128 }
 129
 130 unsigned long
 131 arch_get_unmapped_area(struct file *filp, unsigned long addr,
 132                 unsigned long len, unsigned long pgoff, unsigned long flags)
 133 {
 134         struct mm_struct *mm = current->mm;
 135         struct vm_area_struct *vma;
 136         struct vm_unmapped_area_info info;
 137         unsigned long begin, end;
 138
 139         if (flags & MAP_FIXED)
 140                 return addr;
 141
 142         find_start_end(addr, flags, &begin, &end);
 143
 144         if (len > end)
 145                 return -ENOMEM;
 146
 147         if (addr) {
 148                 addr = PAGE_ALIGN(addr);
 149                 vma = find_vma(mm, addr);
 150                 if (end - len >= addr &&
 151                     (!vma || addr + len <= vm_start_gap(vma)))
 152                         return addr;
 153         }
 154
 155         info.flags = 0;
 156         info.length = len;
 157         info.low_limit = begin;
 158         info.high_limit = end;
 159         info.align_mask = 0;
 160         info.align_offset = pgoff << PAGE_SHIFT;
 161         if (filp) {
 162                 info.align_mask = get_align_mask();
 163                 info.align_offset += get_align_bits();
 164         }
 165         return vm_unmapped_area(&info);
 166 }
 167
 168 unsigned long
 169 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
 170                           const unsigned long len, const unsigned long pgoff,
 171                           const unsigned long flags)
 172 {
 173         struct vm_area_struct *vma;
 174         struct mm_struct *mm = current->mm;
 175         unsigned long addr = addr0;
 176         struct vm_unmapped_area_info info;
 177
 178         /* requested length too big for entire address space */
 179         if (len > TASK_SIZE)
 180                 return -ENOMEM;
 181
 182         /* No address checking. See comment at mmap_address_hint_valid() */
 183         if (flags & MAP_FIXED)
 184                 return addr;
 185
 186         /* for MAP_32BIT mappings we force the legacy mmap base */
 187         if (!in_32bit_syscall() && (flags & MAP_32BIT))
 188                 goto bottomup;
 189
 190         /* requesting a specific address */
 191         if (addr) {
 192                 addr &= PAGE_MASK;
 193                 if (!mmap_address_hint_valid(addr, len))
 194                         goto get_unmapped_area;
 195
 196                 vma = find_vma(mm, addr);
 197                 if (!vma || addr + len <= vm_start_gap(vma))
 198                         return addr;
 199         }
 200 get_unmapped_area:
 201
 202         info.flags = VM_UNMAPPED_AREA_TOPDOWN;
 203         info.length = len;
 204         info.low_limit = PAGE_SIZE;
 205         info.high_limit = get_mmap_base(0);
 206
 207         /*
 208          * If hint address is above DEFAULT_MAP_WINDOW, look for unmapped area
 209          * in the full address space.
 210          *
 211          * !in_32bit_syscall() check to avoid high addresses for x32
 212          * (and make it no op on native i386).
 213          */
 214         if (addr > DEFAULT_MAP_WINDOW && !in_32bit_syscall())
 215                 info.high_limit += TASK_SIZE_MAX - DEFAULT_MAP_WINDOW;
 216
 217         info.align_mask = 0;
 218         info.align_offset = pgoff << PAGE_SHIFT;
 219         if (filp) {
 220                 info.align_mask = get_align_mask();
 221                 info.align_offset += get_align_bits();
 222         }
 223         addr = vm_unmapped_area(&info);
 224         if (!(addr & ~PAGE_MASK))
 225                 return addr;
 226         VM_BUG_ON(addr != -ENOMEM);
 227
 228 bottomup:
 229         /*
 230          * A failed mmap() very likely causes application failure,
 231          * so fall back to the bottom-up function here. This scenario
 232          * can happen with large stack limits and large mmap()
 233          * allocations.
 234          */
 235         return arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
 236 }