arch/arm64/kernel/vdso.c

   1 /*
   2  * VDSO implementations.
   3  *
   4  * Copyright (C) 2012 ARM Limited
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  17  *
  18  * Author: Will Deacon <will.deacon@arm.com>
  19  */
  20
  21 #include <linux/cache.h>
  22 #include <linux/clocksource.h>
  23 #include <linux/elf.h>
  24 #include <linux/err.h>
  25 #include <linux/errno.h>
  26 #include <linux/gfp.h>
  27 #include <linux/kernel.h>
  28 #include <linux/mm.h>
  29 #include <linux/sched.h>
  30 #include <linux/signal.h>
  31 #include <linux/slab.h>
  32 #include <linux/timekeeper_internal.h>
  33 #include <linux/vmalloc.h>
  34
  35 #include <asm/cacheflush.h>
  36 #include <asm/signal32.h>
  37 #include <asm/vdso.h>
  38 #include <asm/vdso_datapage.h>
  39
  40 extern char vdso_start[], vdso_end[];
  41 static unsigned long vdso_pages __ro_after_init;
  42
  43 /*
  44  * The vDSO data page.
  45  */
  46 static union {
  47         struct vdso_data        data;
  48         u8                      page[PAGE_SIZE];
  49 } vdso_data_store __page_aligned_data;
  50 struct vdso_data *vdso_data = &vdso_data_store.data;
  51
  52 #ifdef CONFIG_COMPAT
  53 /*
  54  * Create and map the vectors page for AArch32 tasks.
  55  */
  56 #define C_VECTORS       0
  57 #define C_SIGPAGE       1
  58 #define C_PAGES         (C_SIGPAGE + 1)
  59 static struct page *aarch32_vdso_pages[C_PAGES] __ro_after_init;
  60 static const struct vm_special_mapping aarch32_vdso_spec[C_PAGES] = {
  61         {
  62                 .name   = "[vectors]", /* ABI */
  63                 .pages  = &aarch32_vdso_pages[C_VECTORS],
  64         },
  65         {
  66                 .name   = "[sigpage]", /* ABI */
  67                 .pages  = &aarch32_vdso_pages[C_SIGPAGE],
  68         },
  69 };
  70
  71 static int aarch32_alloc_kuser_vdso_page(void)
  72 {
  73         extern char __kuser_helper_start[], __kuser_helper_end[];
  74         int kuser_sz = __kuser_helper_end - __kuser_helper_start;
  75         unsigned long vdso_page;
  76
  77         if (!IS_ENABLED(CONFIG_KUSER_HELPERS))
  78                 return 0;
  79
  80         vdso_page = get_zeroed_page(GFP_ATOMIC);
  81         if (!vdso_page)
  82                 return -ENOMEM;
  83
  84         memcpy((void *)(vdso_page + 0x1000 - kuser_sz), __kuser_helper_start,
  85                kuser_sz);
  86         aarch32_vdso_pages[C_VECTORS] = virt_to_page(vdso_page);
  87         flush_dcache_page(aarch32_vdso_pages[C_VECTORS]);
  88         return 0;
  89 }
  90
  91 static int __init aarch32_alloc_vdso_pages(void)
  92 {
  93         extern char __aarch32_sigret_code_start[], __aarch32_sigret_code_end[];
  94         int sigret_sz = __aarch32_sigret_code_end - __aarch32_sigret_code_start;
  95         unsigned long sigpage;
  96         int ret;
  97
  98         sigpage = get_zeroed_page(GFP_ATOMIC);
  99         if (!sigpage)
 100                 return -ENOMEM;
 101
 102         memcpy((void *)sigpage, __aarch32_sigret_code_start, sigret_sz);
 103         aarch32_vdso_pages[C_SIGPAGE] = virt_to_page(sigpage);
 104         flush_dcache_page(aarch32_vdso_pages[C_SIGPAGE]);
 105
 106         ret = aarch32_alloc_kuser_vdso_page();
 107         if (ret)
 108                 free_page(sigpage);
 109
 110         return ret;
 111 }
 112 arch_initcall(aarch32_alloc_vdso_pages);
 113
 114 static int aarch32_kuser_helpers_setup(struct mm_struct *mm)
 115 {
 116         void *ret;
 117
 118         if (!IS_ENABLED(CONFIG_KUSER_HELPERS))
 119                 return 0;
 120
 121         /*
 122          * Avoid VM_MAYWRITE for compatibility with arch/arm/, where it's
 123          * not safe to CoW the page containing the CPU exception vectors.
 124          */
 125         ret = _install_special_mapping(mm, AARCH32_VECTORS_BASE, PAGE_SIZE,
 126                                        VM_READ | VM_EXEC |
 127                                        VM_MAYREAD | VM_MAYEXEC,
 128                                        &aarch32_vdso_spec[C_VECTORS]);
 129
 130         return PTR_ERR_OR_ZERO(ret);
 131 }
 132
 133 static int aarch32_sigreturn_setup(struct mm_struct *mm)
 134 {
 135         unsigned long addr;
 136         void *ret;
 137
 138         addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
 139         if (IS_ERR_VALUE(addr)) {
 140                 ret = ERR_PTR(addr);
 141                 goto out;
 142         }
 143
 144         /*
 145          * VM_MAYWRITE is required to allow gdb to Copy-on-Write and
 146          * set breakpoints.
 147          */
 148         ret = _install_special_mapping(mm, addr, PAGE_SIZE,
 149                                        VM_READ | VM_EXEC | VM_MAYREAD |
 150                                        VM_MAYWRITE | VM_MAYEXEC,
 151                                        &aarch32_vdso_spec[C_SIGPAGE]);
 152         if (IS_ERR(ret))
 153                 goto out;
 154
 155         mm->context.vdso = (void *)addr;
 156
 157 out:
 158         return PTR_ERR_OR_ZERO(ret);
 159 }
 160
 161 int aarch32_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 162 {
 163         struct mm_struct *mm = current->mm;
 164         int ret;
 165
 166         if (down_write_killable(&mm->mmap_sem))
 167                 return -EINTR;
 168
 169         ret = aarch32_kuser_helpers_setup(mm);
 170         if (ret)
 171                 goto out;
 172
 173         ret = aarch32_sigreturn_setup(mm);
 174
 175 out:
 176         up_write(&mm->mmap_sem);
 177         return ret;
 178 }
 179 #endif /* CONFIG_COMPAT */
 180
 181 static int vdso_mremap(const struct vm_special_mapping *sm,
 182                 struct vm_area_struct *new_vma)
 183 {
 184         unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
 185         unsigned long vdso_size = vdso_end - vdso_start;
 186
 187         if (vdso_size != new_size)
 188                 return -EINVAL;
 189
 190         current->mm->context.vdso = (void *)new_vma->vm_start;
 191
 192         return 0;
 193 }
 194
 195 static struct vm_special_mapping vdso_spec[2] __ro_after_init = {
 196         {
 197                 .name   = "[vvar]",
 198         },
 199         {
 200                 .name   = "[vdso]",
 201                 .mremap = vdso_mremap,
 202         },
 203 };
 204
 205 static int __init vdso_init(void)
 206 {
 207         int i;
 208         struct page **vdso_pagelist;
 209         unsigned long pfn;
 210
 211         if (memcmp(vdso_start, "\177ELF", 4)) {
 212                 pr_err("vDSO is not a valid ELF object!\n");
 213                 return -EINVAL;
 214         }
 215
 216         vdso_pages = (vdso_end - vdso_start) >> PAGE_SHIFT;
 217
 218         /* Allocate the vDSO pagelist, plus a page for the data. */
 219         vdso_pagelist = kcalloc(vdso_pages + 1, sizeof(struct page *),
 220                                 GFP_KERNEL);
 221         if (vdso_pagelist == NULL)
 222                 return -ENOMEM;
 223
 224         /* Grab the vDSO data page. */
 225         vdso_pagelist[0] = phys_to_page(__pa_symbol(vdso_data));
 226
 227
 228         /* Grab the vDSO code pages. */
 229         pfn = sym_to_pfn(vdso_start);
 230
 231         for (i = 0; i < vdso_pages; i++)
 232                 vdso_pagelist[i + 1] = pfn_to_page(pfn + i);
 233
 234         vdso_spec[0].pages = &vdso_pagelist[0];
 235         vdso_spec[1].pages = &vdso_pagelist[1];
 236
 237         return 0;
 238 }
 239 arch_initcall(vdso_init);
 240
 241 int arch_setup_additional_pages(struct linux_binprm *bprm,
 242                                 int uses_interp)
 243 {
 244         struct mm_struct *mm = current->mm;
 245         unsigned long vdso_base, vdso_text_len, vdso_mapping_len;
 246         void *ret;
 247
 248         vdso_text_len = vdso_pages << PAGE_SHIFT;
 249         /* Be sure to map the data page */
 250         vdso_mapping_len = vdso_text_len + PAGE_SIZE;
 251
 252         if (down_write_killable(&mm->mmap_sem))
 253                 return -EINTR;
 254         vdso_base = get_unmapped_area(NULL, 0, vdso_mapping_len, 0, 0);
 255         if (IS_ERR_VALUE(vdso_base)) {
 256                 ret = ERR_PTR(vdso_base);
 257                 goto up_fail;
 258         }
 259         ret = _install_special_mapping(mm, vdso_base, PAGE_SIZE,
 260                                        VM_READ|VM_MAYREAD,
 261                                        &vdso_spec[0]);
 262         if (IS_ERR(ret))
 263                 goto up_fail;
 264
 265         vdso_base += PAGE_SIZE;
 266         mm->context.vdso = (void *)vdso_base;
 267         ret = _install_special_mapping(mm, vdso_base, vdso_text_len,
 268                                        VM_READ|VM_EXEC|
 269                                        VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
 270                                        &vdso_spec[1]);
 271         if (IS_ERR(ret))
 272                 goto up_fail;
 273
 274
 275         up_write(&mm->mmap_sem);
 276         return 0;
 277
 278 up_fail:
 279         mm->context.vdso = NULL;
 280         up_write(&mm->mmap_sem);
 281         return PTR_ERR(ret);
 282 }
 283
 284 /*
 285  * Update the vDSO data page to keep in sync with kernel timekeeping.
 286  */
 287 void update_vsyscall(struct timekeeper *tk)
 288 {
 289         u32 use_syscall = !tk->tkr_mono.clock->archdata.vdso_direct;
 290
 291         ++vdso_data->tb_seq_count;
 292         smp_wmb();
 293
 294         vdso_data->use_syscall                  = use_syscall;
 295         vdso_data->xtime_coarse_sec             = tk->xtime_sec;
 296         vdso_data->xtime_coarse_nsec            = tk->tkr_mono.xtime_nsec >>
 297                                                         tk->tkr_mono.shift;
 298         vdso_data->wtm_clock_sec                = tk->wall_to_monotonic.tv_sec;
 299         vdso_data->wtm_clock_nsec               = tk->wall_to_monotonic.tv_nsec;
 300
 301         /* Read without the seqlock held by clock_getres() */
 302         WRITE_ONCE(vdso_data->hrtimer_res, hrtimer_resolution);
 303
 304         if (!use_syscall) {
 305                 /* tkr_mono.cycle_last == tkr_raw.cycle_last */
 306                 vdso_data->cs_cycle_last        = tk->tkr_mono.cycle_last;
 307                 vdso_data->raw_time_sec         = tk->raw_sec;
 308                 vdso_data->raw_time_nsec        = tk->tkr_raw.xtime_nsec;
 309                 vdso_data->xtime_clock_sec      = tk->xtime_sec;
 310                 vdso_data->xtime_clock_nsec     = tk->tkr_mono.xtime_nsec;
 311                 vdso_data->cs_mono_mult         = tk->tkr_mono.mult;
 312                 vdso_data->cs_raw_mult          = tk->tkr_raw.mult;
 313                 /* tkr_mono.shift == tkr_raw.shift */
 314                 vdso_data->cs_shift             = tk->tkr_mono.shift;
 315         }
 316
 317         smp_wmb();
 318         ++vdso_data->tb_seq_count;
 319 }
 320
 321 void update_vsyscall_tz(void)
 322 {
 323         vdso_data->tz_minuteswest       = sys_tz.tz_minuteswest;
 324         vdso_data->tz_dsttime           = sys_tz.tz_dsttime;
 325 }