arch/s390/kernel/vdso.c

   1 /*
   2  * vdso setup for s390
   3  *
   4  *  Copyright IBM Corp. 2008
   5  *  Author(s): Martin Schwidefsky (schwidefsky@de.ibm.com)
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License (version 2 only)
   9  * as published by the Free Software Foundation.
  10  */
  11
  12 #include <linux/init.h>
  13 #include <linux/errno.h>
  14 #include <linux/sched.h>
  15 #include <linux/kernel.h>
  16 #include <linux/mm.h>
  17 #include <linux/smp.h>
  18 #include <linux/stddef.h>
  19 #include <linux/unistd.h>
  20 #include <linux/slab.h>
  21 #include <linux/user.h>
  22 #include <linux/elf.h>
  23 #include <linux/security.h>
  24 #include <linux/bootmem.h>
  25 #include <linux/compat.h>
  26 #include <asm/asm-offsets.h>
  27 #include <asm/pgtable.h>
  28 #include <asm/processor.h>
  29 #include <asm/mmu.h>
  30 #include <asm/mmu_context.h>
  31 #include <asm/sections.h>
  32 #include <asm/vdso.h>
  33 #include <asm/facility.h>
  34
  35 #ifdef CONFIG_COMPAT
  36 extern char vdso32_start, vdso32_end;
  37 static void *vdso32_kbase = &vdso32_start;
  38 static unsigned int vdso32_pages;
  39 static struct page **vdso32_pagelist;
  40 #endif
  41
  42 extern char vdso64_start, vdso64_end;
  43 static void *vdso64_kbase = &vdso64_start;
  44 static unsigned int vdso64_pages;
  45 static struct page **vdso64_pagelist;
  46
  47 /*
  48  * Should the kernel map a VDSO page into processes and pass its
  49  * address down to glibc upon exec()?
  50  */
  51 unsigned int __read_mostly vdso_enabled = 1;
  52
  53 static int vdso_fault(const struct vm_special_mapping *sm,
  54                       struct vm_area_struct *vma, struct vm_fault *vmf)
  55 {
  56         struct page **vdso_pagelist;
  57         unsigned long vdso_pages;
  58
  59         vdso_pagelist = vdso64_pagelist;
  60         vdso_pages = vdso64_pages;
  61 #ifdef CONFIG_COMPAT
  62         if (is_compat_task()) {
  63                 vdso_pagelist = vdso32_pagelist;
  64                 vdso_pages = vdso32_pages;
  65         }
  66 #endif
  67
  68         if (vmf->pgoff >= vdso_pages)
  69                 return VM_FAULT_SIGBUS;
  70
  71         vmf->page = vdso_pagelist[vmf->pgoff];
  72         get_page(vmf->page);
  73         return 0;
  74 }
  75
  76 static int vdso_mremap(const struct vm_special_mapping *sm,
  77                        struct vm_area_struct *vma)
  78 {
  79         unsigned long vdso_pages;
  80
  81         vdso_pages = vdso64_pages;
  82 #ifdef CONFIG_COMPAT
  83         if (is_compat_task())
  84                 vdso_pages = vdso32_pages;
  85 #endif
  86
  87         if ((vdso_pages << PAGE_SHIFT) != vma->vm_end - vma->vm_start)
  88                 return -EINVAL;
  89
  90         if (WARN_ON_ONCE(current->mm != vma->vm_mm))
  91                 return -EFAULT;
  92
  93         current->mm->context.vdso_base = vma->vm_start;
  94         return 0;
  95 }
  96
  97 static const struct vm_special_mapping vdso_mapping = {
  98         .name = "[vdso]",
  99         .fault = vdso_fault,
 100         .mremap = vdso_mremap,
 101 };
 102
 103 static int __init vdso_setup(char *s)
 104 {
 105         unsigned long val;
 106         int rc;
 107
 108         rc = 0;
 109         if (strncmp(s, "on", 3) == 0)
 110                 vdso_enabled = 1;
 111         else if (strncmp(s, "off", 4) == 0)
 112                 vdso_enabled = 0;
 113         else {
 114                 rc = kstrtoul(s, 0, &val);
 115                 vdso_enabled = rc ? 0 : !!val;
 116         }
 117         return !rc;
 118 }
 119 __setup("vdso=", vdso_setup);
 120
 121 /*
 122  * The vdso data page
 123  */
 124 static union {
 125         struct vdso_data        data;
 126         u8                      page[PAGE_SIZE];
 127 } vdso_data_store __page_aligned_data;
 128 struct vdso_data *vdso_data = &vdso_data_store.data;
 129
 130 /*
 131  * Setup vdso data page.
 132  */
 133 static void __init vdso_init_data(struct vdso_data *vd)
 134 {
 135         vd->ectg_available = test_facility(31);
 136 }
 137
 138 /*
 139  * Allocate/free per cpu vdso data.
 140  */
 141 #define SEGMENT_ORDER   2
 142
 143 int vdso_alloc_per_cpu(struct lowcore *lowcore)
 144 {
 145         unsigned long segment_table, page_table, page_frame;
 146         struct vdso_per_cpu_data *vd;
 147         u32 *psal, *aste;
 148         int i;
 149
 150         lowcore->vdso_per_cpu_data = __LC_PASTE;
 151
 152         if (!vdso_enabled)
 153                 return 0;
 154
 155         segment_table = __get_free_pages(GFP_KERNEL, SEGMENT_ORDER);
 156         page_table = get_zeroed_page(GFP_KERNEL | GFP_DMA);
 157         page_frame = get_zeroed_page(GFP_KERNEL);
 158         if (!segment_table || !page_table || !page_frame)
 159                 goto out;
 160         arch_set_page_dat(virt_to_page(segment_table), SEGMENT_ORDER);
 161         arch_set_page_dat(virt_to_page(page_table), 0);
 162
 163         /* Initialize per-cpu vdso data page */
 164         vd = (struct vdso_per_cpu_data *) page_frame;
 165         vd->cpu_nr = lowcore->cpu_nr;
 166         vd->node_id = cpu_to_node(vd->cpu_nr);
 167
 168         /* Set up access register mode page table */
 169         clear_table((unsigned long *) segment_table, _SEGMENT_ENTRY_EMPTY,
 170                     PAGE_SIZE << SEGMENT_ORDER);
 171         clear_table((unsigned long *) page_table, _PAGE_INVALID,
 172                     256*sizeof(unsigned long));
 173
 174         *(unsigned long *) segment_table = _SEGMENT_ENTRY + page_table;
 175         *(unsigned long *) page_table = _PAGE_PROTECT + page_frame;
 176
 177         psal = (u32 *) (page_table + 256*sizeof(unsigned long));
 178         aste = psal + 32;
 179
 180         for (i = 4; i < 32; i += 4)
 181                 psal[i] = 0x80000000;
 182
 183         lowcore->paste[4] = (u32)(addr_t) psal;
 184         psal[0] = 0x02000000;
 185         psal[2] = (u32)(addr_t) aste;
 186         *(unsigned long *) (aste + 2) = segment_table +
 187                 _ASCE_TABLE_LENGTH + _ASCE_USER_BITS + _ASCE_TYPE_SEGMENT;
 188         aste[4] = (u32)(addr_t) psal;
 189         lowcore->vdso_per_cpu_data = page_frame;
 190
 191         return 0;
 192
 193 out:
 194         free_page(page_frame);
 195         free_page(page_table);
 196         free_pages(segment_table, SEGMENT_ORDER);
 197         return -ENOMEM;
 198 }
 199
 200 void vdso_free_per_cpu(struct lowcore *lowcore)
 201 {
 202         unsigned long segment_table, page_table, page_frame;
 203         u32 *psal, *aste;
 204
 205         if (!vdso_enabled)
 206                 return;
 207
 208         psal = (u32 *)(addr_t) lowcore->paste[4];
 209         aste = (u32 *)(addr_t) psal[2];
 210         segment_table = *(unsigned long *)(aste + 2) & PAGE_MASK;
 211         page_table = *(unsigned long *) segment_table;
 212         page_frame = *(unsigned long *) page_table;
 213
 214         free_page(page_frame);
 215         free_page(page_table);
 216         free_pages(segment_table, SEGMENT_ORDER);
 217 }
 218
 219 static void vdso_init_cr5(void)
 220 {
 221         unsigned long cr5;
 222
 223         if (!vdso_enabled)
 224                 return;
 225         cr5 = offsetof(struct lowcore, paste);
 226         __ctl_load(cr5, 5, 5);
 227 }
 228
 229 /*
 230  * This is called from binfmt_elf, we create the special vma for the
 231  * vDSO and insert it into the mm struct tree
 232  */
 233 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 234 {
 235         struct mm_struct *mm = current->mm;
 236         struct vm_area_struct *vma;
 237         unsigned long vdso_pages;
 238         unsigned long vdso_base;
 239         int rc;
 240
 241         if (!vdso_enabled)
 242                 return 0;
 243         /*
 244          * Only map the vdso for dynamically linked elf binaries.
 245          */
 246         if (!uses_interp)
 247                 return 0;
 248
 249         vdso_pages = vdso64_pages;
 250 #ifdef CONFIG_COMPAT
 251         if (is_compat_task())
 252                 vdso_pages = vdso32_pages;
 253 #endif
 254         /*
 255          * vDSO has a problem and was disabled, just don't "enable" it for
 256          * the process
 257          */
 258         if (vdso_pages == 0)
 259                 return 0;
 260
 261         /*
 262          * pick a base address for the vDSO in process space. We try to put
 263          * it at vdso_base which is the "natural" base for it, but we might
 264          * fail and end up putting it elsewhere.
 265          */
 266         if (down_write_killable(&mm->mmap_sem))
 267                 return -EINTR;
 268         vdso_base = get_unmapped_area(NULL, 0, vdso_pages << PAGE_SHIFT, 0, 0);
 269         if (IS_ERR_VALUE(vdso_base)) {
 270                 rc = vdso_base;
 271                 goto out_up;
 272         }
 273
 274         /*
 275          * our vma flags don't have VM_WRITE so by default, the process
 276          * isn't allowed to write those pages.
 277          * gdb can break that with ptrace interface, and thus trigger COW
 278          * on those pages but it's then your responsibility to never do that
 279          * on the "data" page of the vDSO or you'll stop getting kernel
 280          * updates and your nice userland gettimeofday will be totally dead.
 281          * It's fine to use that for setting breakpoints in the vDSO code
 282          * pages though.
 283          */
 284         vma = _install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
 285                                        VM_READ|VM_EXEC|
 286                                        VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
 287                                        &vdso_mapping);
 288         if (IS_ERR(vma)) {
 289                 rc = PTR_ERR(vma);
 290                 goto out_up;
 291         }
 292
 293         current->mm->context.vdso_base = vdso_base;
 294         rc = 0;
 295
 296 out_up:
 297         up_write(&mm->mmap_sem);
 298         return rc;
 299 }
 300
 301 static int __init vdso_init(void)
 302 {
 303         int i;
 304
 305         if (!vdso_enabled)
 306                 return 0;
 307         vdso_init_data(vdso_data);
 308 #ifdef CONFIG_COMPAT
 309         /* Calculate the size of the 32 bit vDSO */
 310         vdso32_pages = ((&vdso32_end - &vdso32_start
 311                          + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
 312
 313         /* Make sure pages are in the correct state */
 314         vdso32_pagelist = kzalloc(sizeof(struct page *) * (vdso32_pages + 1),
 315                                   GFP_KERNEL);
 316         BUG_ON(vdso32_pagelist == NULL);
 317         for (i = 0; i < vdso32_pages - 1; i++) {
 318                 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
 319                 ClearPageReserved(pg);
 320                 get_page(pg);
 321                 vdso32_pagelist[i] = pg;
 322         }
 323         vdso32_pagelist[vdso32_pages - 1] = virt_to_page(vdso_data);
 324         vdso32_pagelist[vdso32_pages] = NULL;
 325 #endif
 326
 327         /* Calculate the size of the 64 bit vDSO */
 328         vdso64_pages = ((&vdso64_end - &vdso64_start
 329                          + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
 330
 331         /* Make sure pages are in the correct state */
 332         vdso64_pagelist = kzalloc(sizeof(struct page *) * (vdso64_pages + 1),
 333                                   GFP_KERNEL);
 334         BUG_ON(vdso64_pagelist == NULL);
 335         for (i = 0; i < vdso64_pages - 1; i++) {
 336                 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
 337                 ClearPageReserved(pg);
 338                 get_page(pg);
 339                 vdso64_pagelist[i] = pg;
 340         }
 341         vdso64_pagelist[vdso64_pages - 1] = virt_to_page(vdso_data);
 342         vdso64_pagelist[vdso64_pages] = NULL;
 343         if (vdso_alloc_per_cpu(&S390_lowcore))
 344                 BUG();
 345         vdso_init_cr5();
 346
 347         get_page(virt_to_page(vdso_data));
 348
 349         return 0;
 350 }
 351 early_initcall(vdso_init);