arch/x86/kernel/sys_x86_64.c

   1 #include <linux/errno.h>
   2 #include <linux/sched.h>
   3 #include <linux/sched/mm.h>
   4 #include <linux/syscalls.h>
   5 #include <linux/mm.h>
   6 #include <linux/fs.h>
   7 #include <linux/smp.h>
   8 #include <linux/sem.h>
   9 #include <linux/msg.h>
  10 #include <linux/shm.h>
  11 #include <linux/stat.h>
  12 #include <linux/mman.h>
  13 #include <linux/file.h>
  14 #include <linux/utsname.h>
  15 #include <linux/personality.h>
  16 #include <linux/random.h>
  17 #include <linux/uaccess.h>
  18 #include <linux/elf.h>
  19
  20 #include <asm/elf.h>
  21 #include <asm/compat.h>
  22 #include <asm/ia32.h>
  23 #include <asm/syscalls.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 = sys_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 flags, unsigned long *begin,
 104                            unsigned long *end)
 105 {
 106         if (!in_compat_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         *end    = in_compat_syscall() ? tasksize_32bit() : tasksize_64bit();
 124 }
 125
 126 unsigned long
 127 arch_get_unmapped_area(struct file *filp, unsigned long addr,
 128                 unsigned long len, unsigned long pgoff, unsigned long flags)
 129 {
 130         struct mm_struct *mm = current->mm;
 131         struct vm_area_struct *vma;
 132         struct vm_unmapped_area_info info;
 133         unsigned long begin, end;
 134
 135         if (flags & MAP_FIXED)
 136                 return addr;
 137
 138         find_start_end(flags, &begin, &end);
 139
 140         if (len > end)
 141                 return -ENOMEM;
 142
 143         if (addr) {
 144                 addr = PAGE_ALIGN(addr);
 145                 vma = find_vma(mm, addr);
 146                 if (end - len >= addr &&
 147                     (!vma || addr + len <= vm_start_gap(vma)))
 148                         return addr;
 149         }
 150
 151         info.flags = 0;
 152         info.length = len;
 153         info.low_limit = begin;
 154         info.high_limit = end;
 155         info.align_mask = 0;
 156         info.align_offset = pgoff << PAGE_SHIFT;
 157         if (filp) {
 158                 info.align_mask = get_align_mask();
 159                 info.align_offset += get_align_bits();
 160         }
 161         return vm_unmapped_area(&info);
 162 }
 163
 164 unsigned long
 165 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
 166                           const unsigned long len, const unsigned long pgoff,
 167                           const unsigned long flags)
 168 {
 169         struct vm_area_struct *vma;
 170         struct mm_struct *mm = current->mm;
 171         unsigned long addr = addr0;
 172         struct vm_unmapped_area_info info;
 173
 174         /* requested length too big for entire address space */
 175         if (len > TASK_SIZE)
 176                 return -ENOMEM;
 177
 178         if (flags & MAP_FIXED)
 179                 return addr;
 180
 181         /* for MAP_32BIT mappings we force the legacy mmap base */
 182         if (!in_compat_syscall() && (flags & MAP_32BIT))
 183                 goto bottomup;
 184
 185         /* requesting a specific address */
 186         if (addr) {
 187                 addr = PAGE_ALIGN(addr);
 188                 vma = find_vma(mm, addr);
 189                 if (TASK_SIZE - len >= addr &&
 190                                 (!vma || addr + len <= vm_start_gap(vma)))
 191                         return addr;
 192         }
 193
 194         info.flags = VM_UNMAPPED_AREA_TOPDOWN;
 195         info.length = len;
 196         info.low_limit = PAGE_SIZE;
 197         info.high_limit = get_mmap_base(0);
 198         info.align_mask = 0;
 199         info.align_offset = pgoff << PAGE_SHIFT;
 200         if (filp) {
 201                 info.align_mask = get_align_mask();
 202                 info.align_offset += get_align_bits();
 203         }
 204         addr = vm_unmapped_area(&info);
 205         if (!(addr & ~PAGE_MASK))
 206                 return addr;
 207         VM_BUG_ON(addr != -ENOMEM);
 208
 209 bottomup:
 210         /*
 211          * A failed mmap() very likely causes application failure,
 212          * so fall back to the bottom-up function here. This scenario
 213          * can happen with large stack limits and large mmap()
 214          * allocations.
 215          */
 216         return arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
 217 }