arch/x86/kernel/sys_x86_64.c

   1 #include <linux/errno.h>
   2 #include <linux/sched.h>
   3 #include <linux/syscalls.h>
   4 #include <linux/mm.h>
   5 #include <linux/fs.h>
   6 #include <linux/smp.h>
   7 #include <linux/sem.h>
   8 #include <linux/msg.h>
   9 #include <linux/shm.h>
  10 #include <linux/stat.h>
  11 #include <linux/mman.h>
  12 #include <linux/file.h>
  13 #include <linux/utsname.h>
  14 #include <linux/personality.h>
  15 #include <linux/random.h>
  16 #include <linux/uaccess.h>
  17
  18 #include <asm/ia32.h>
  19 #include <asm/syscalls.h>
  20
  21 /*
  22  * Align a virtual address to avoid aliasing in the I$ on AMD F15h.
  23  *
  24  * @flags denotes the allocation direction - bottomup or topdown -
  25  * or vDSO; see call sites below.
  26  */
  27 unsigned long align_addr(unsigned long addr, struct file *filp,
  28                          enum align_flags flags)
  29 {
  30         unsigned long tmp_addr;
  31
  32         /* handle 32- and 64-bit case with a single conditional */
  33         if (va_align.flags < 0 || !(va_align.flags & (2 - mmap_is_ia32())))
  34                 return addr;
  35
  36         if (!(current->flags & PF_RANDOMIZE))
  37                 return addr;
  38
  39         if (!((flags & ALIGN_VDSO) || filp))
  40                 return addr;
  41
  42         tmp_addr = addr;
  43
  44         /*
  45          * We need an address which is <= than the original
  46          * one only when in topdown direction.
  47          */
  48         if (!(flags & ALIGN_TOPDOWN))
  49                 tmp_addr += va_align.mask;
  50
  51         tmp_addr &= ~va_align.mask;
  52
  53         return tmp_addr;
  54 }
  55
  56 static int __init control_va_addr_alignment(char *str)
  57 {
  58         /* guard against enabling this on other CPU families */
  59         if (va_align.flags < 0)
  60                 return 1;
  61
  62         if (*str == 0)
  63                 return 1;
  64
  65         if (*str == '=')
  66                 str++;
  67
  68         if (!strcmp(str, "32"))
  69                 va_align.flags = ALIGN_VA_32;
  70         else if (!strcmp(str, "64"))
  71                 va_align.flags = ALIGN_VA_64;
  72         else if (!strcmp(str, "off"))
  73                 va_align.flags = 0;
  74         else if (!strcmp(str, "on"))
  75                 va_align.flags = ALIGN_VA_32 | ALIGN_VA_64;
  76         else
  77                 return 0;
  78
  79         return 1;
  80 }
  81 __setup("align_va_addr", control_va_addr_alignment);
  82
  83 SYSCALL_DEFINE6(mmap, unsigned long, addr, unsigned long, len,
  84                 unsigned long, prot, unsigned long, flags,
  85                 unsigned long, fd, unsigned long, off)
  86 {
  87         long error;
  88         error = -EINVAL;
  89         if (off & ~PAGE_MASK)
  90                 goto out;
  91
  92         error = sys_mmap_pgoff(addr, len, prot, flags, fd, off >> PAGE_SHIFT);
  93 out:
  94         return error;
  95 }
  96
  97 static void find_start_end(unsigned long flags, unsigned long *begin,
  98                            unsigned long *end)
  99 {
 100         if (!test_thread_flag(TIF_IA32) && (flags & MAP_32BIT)) {
 101                 unsigned long new_begin;
 102                 /* This is usually used needed to map code in small
 103                    model, so it needs to be in the first 31bit. Limit
 104                    it to that.  This means we need to move the
 105                    unmapped base down for this case. This can give
 106                    conflicts with the heap, but we assume that glibc
 107                    malloc knows how to fall back to mmap. Give it 1GB
 108                    of playground for now. -AK */
 109                 *begin = 0x40000000;
 110                 *end = 0x80000000;
 111                 if (current->flags & PF_RANDOMIZE) {
 112                         new_begin = randomize_range(*begin, *begin + 0x02000000, 0);
 113                         if (new_begin)
 114                                 *begin = new_begin;
 115                 }
 116         } else {
 117                 *begin = TASK_UNMAPPED_BASE;
 118                 *end = TASK_SIZE;
 119         }
 120 }
 121
 122 unsigned long
 123 arch_get_unmapped_area(struct file *filp, unsigned long addr,
 124                 unsigned long len, unsigned long pgoff, unsigned long flags)
 125 {
 126         struct mm_struct *mm = current->mm;
 127         struct vm_area_struct *vma;
 128         unsigned long start_addr;
 129         unsigned long begin, end;
 130
 131         if (flags & MAP_FIXED)
 132                 return addr;
 133
 134         find_start_end(flags, &begin, &end);
 135
 136         if (len > end)
 137                 return -ENOMEM;
 138
 139         if (addr) {
 140                 addr = PAGE_ALIGN(addr);
 141                 vma = find_vma(mm, addr);
 142                 if (end - len >= addr &&
 143                     (!vma || addr + len <= vma->vm_start))
 144                         return addr;
 145         }
 146         if (((flags & MAP_32BIT) || test_thread_flag(TIF_IA32))
 147             && len <= mm->cached_hole_size) {
 148                 mm->cached_hole_size = 0;
 149                 mm->free_area_cache = begin;
 150         }
 151         addr = mm->free_area_cache;
 152         if (addr < begin)
 153                 addr = begin;
 154         start_addr = addr;
 155
 156 full_search:
 157
 158         addr = align_addr(addr, filp, 0);
 159
 160         for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
 161                 /* At this point:  (!vma || addr < vma->vm_end). */
 162                 if (end - len < addr) {
 163                         /*
 164                          * Start a new search - just in case we missed
 165                          * some holes.
 166                          */
 167                         if (start_addr != begin) {
 168                                 start_addr = addr = begin;
 169                                 mm->cached_hole_size = 0;
 170                                 goto full_search;
 171                         }
 172                         return -ENOMEM;
 173                 }
 174                 if (!vma || addr + len <= vma->vm_start) {
 175                         /*
 176                          * Remember the place where we stopped the search:
 177                          */
 178                         mm->free_area_cache = addr + len;
 179                         return addr;
 180                 }
 181                 if (addr + mm->cached_hole_size < vma->vm_start)
 182                         mm->cached_hole_size = vma->vm_start - addr;
 183
 184                 addr = vma->vm_end;
 185                 addr = align_addr(addr, filp, 0);
 186         }
 187 }
 188
 189
 190 unsigned long
 191 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
 192                           const unsigned long len, const unsigned long pgoff,
 193                           const unsigned long flags)
 194 {
 195         struct vm_area_struct *vma;
 196         struct mm_struct *mm = current->mm;
 197         unsigned long addr = addr0;
 198
 199         /* requested length too big for entire address space */
 200         if (len > TASK_SIZE)
 201                 return -ENOMEM;
 202
 203         if (flags & MAP_FIXED)
 204                 return addr;
 205
 206         /* for MAP_32BIT mappings we force the legact mmap base */
 207         if (!test_thread_flag(TIF_IA32) && (flags & MAP_32BIT))
 208                 goto bottomup;
 209
 210         /* requesting a specific address */
 211         if (addr) {
 212                 addr = PAGE_ALIGN(addr);
 213                 vma = find_vma(mm, addr);
 214                 if (TASK_SIZE - len >= addr &&
 215                                 (!vma || addr + len <= vma->vm_start))
 216                         return addr;
 217         }
 218
 219         /* check if free_area_cache is useful for us */
 220         if (len <= mm->cached_hole_size) {
 221                 mm->cached_hole_size = 0;
 222                 mm->free_area_cache = mm->mmap_base;
 223         }
 224
 225         /* either no address requested or can't fit in requested address hole */
 226         addr = mm->free_area_cache;
 227
 228         /* make sure it can fit in the remaining address space */
 229         if (addr > len) {
 230                 unsigned long tmp_addr = align_addr(addr - len, filp,
 231                                                     ALIGN_TOPDOWN);
 232
 233                 vma = find_vma(mm, tmp_addr);
 234                 if (!vma || tmp_addr + len <= vma->vm_start)
 235                         /* remember the address as a hint for next time */
 236                         return mm->free_area_cache = tmp_addr;
 237         }
 238
 239         if (mm->mmap_base < len)
 240                 goto bottomup;
 241
 242         addr = mm->mmap_base-len;
 243
 244         do {
 245                 addr = align_addr(addr, filp, ALIGN_TOPDOWN);
 246
 247                 /*
 248                  * Lookup failure means no vma is above this address,
 249                  * else if new region fits below vma->vm_start,
 250                  * return with success:
 251                  */
 252                 vma = find_vma(mm, addr);
 253                 if (!vma || addr+len <= vma->vm_start)
 254                         /* remember the address as a hint for next time */
 255                         return mm->free_area_cache = addr;
 256
 257                 /* remember the largest hole we saw so far */
 258                 if (addr + mm->cached_hole_size < vma->vm_start)
 259                         mm->cached_hole_size = vma->vm_start - addr;
 260
 261                 /* try just below the current vma->vm_start */
 262                 addr = vma->vm_start-len;
 263         } while (len < vma->vm_start);
 264
 265 bottomup:
 266         /*
 267          * A failed mmap() very likely causes application failure,
 268          * so fall back to the bottom-up function here. This scenario
 269          * can happen with large stack limits and large mmap()
 270          * allocations.
 271          */
 272         mm->cached_hole_size = ~0UL;
 273         mm->free_area_cache = TASK_UNMAPPED_BASE;
 274         addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
 275         /*
 276          * Restore the topdown base:
 277          */
 278         mm->free_area_cache = mm->mmap_base;
 279         mm->cached_hole_size = ~0UL;
 280
 281         return addr;
 282 }