arch/s390/kernel/vdso.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * vdso setup for s390
   4  *
   5  *  Copyright IBM Corp. 2008
   6  *  Author(s): Martin Schwidefsky (schwidefsky@de.ibm.com)
   7  */
   8
   9 #include <linux/init.h>
  10 #include <linux/errno.h>
  11 #include <linux/sched.h>
  12 #include <linux/kernel.h>
  13 #include <linux/mm.h>
  14 #include <linux/smp.h>
  15 #include <linux/stddef.h>
  16 #include <linux/unistd.h>
  17 #include <linux/slab.h>
  18 #include <linux/user.h>
  19 #include <linux/elf.h>
  20 #include <linux/security.h>
  21 #include <linux/memblock.h>
  22 #include <linux/compat.h>
  23 #include <asm/asm-offsets.h>
  24 #include <asm/processor.h>
  25 #include <asm/mmu.h>
  26 #include <asm/mmu_context.h>
  27 #include <asm/sections.h>
  28 #include <asm/vdso.h>
  29 #include <asm/facility.h>
  30
  31 extern char vdso64_start, vdso64_end;
  32 static void *vdso64_kbase = &vdso64_start;
  33 static unsigned int vdso64_pages;
  34 static struct page **vdso64_pagelist;
  35
  36 /*
  37  * Should the kernel map a VDSO page into processes and pass its
  38  * address down to glibc upon exec()?
  39  */
  40 unsigned int __read_mostly vdso_enabled = 1;
  41
  42 static vm_fault_t vdso_fault(const struct vm_special_mapping *sm,
  43                       struct vm_area_struct *vma, struct vm_fault *vmf)
  44 {
  45         struct page **vdso_pagelist;
  46         unsigned long vdso_pages;
  47
  48         vdso_pagelist = vdso64_pagelist;
  49         vdso_pages = vdso64_pages;
  50
  51         if (vmf->pgoff >= vdso_pages)
  52                 return VM_FAULT_SIGBUS;
  53
  54         vmf->page = vdso_pagelist[vmf->pgoff];
  55         get_page(vmf->page);
  56         return 0;
  57 }
  58
  59 static int vdso_mremap(const struct vm_special_mapping *sm,
  60                        struct vm_area_struct *vma)
  61 {
  62         unsigned long vdso_pages;
  63
  64         vdso_pages = vdso64_pages;
  65
  66         if ((vdso_pages << PAGE_SHIFT) != vma->vm_end - vma->vm_start)
  67                 return -EINVAL;
  68
  69         if (WARN_ON_ONCE(current->mm != vma->vm_mm))
  70                 return -EFAULT;
  71
  72         current->mm->context.vdso_base = vma->vm_start;
  73         return 0;
  74 }
  75
  76 static const struct vm_special_mapping vdso_mapping = {
  77         .name = "[vdso]",
  78         .fault = vdso_fault,
  79         .mremap = vdso_mremap,
  80 };
  81
  82 static int __init vdso_setup(char *str)
  83 {
  84         bool enabled;
  85
  86         if (!kstrtobool(str, &enabled))
  87                 vdso_enabled = enabled;
  88         return 1;
  89 }
  90 __setup("vdso=", vdso_setup);
  91
  92 /*
  93  * The vdso data page
  94  */
  95 static union {
  96         struct vdso_data        data;
  97         u8                      page[PAGE_SIZE];
  98 } vdso_data_store __page_aligned_data;
  99 struct vdso_data *vdso_data = &vdso_data_store.data;
 100
 101 /*
 102  * Setup vdso data page.
 103  */
 104 static void __init vdso_init_data(struct vdso_data *vd)
 105 {
 106         vd->ectg_available = test_facility(31);
 107 }
 108
 109 /*
 110  * Allocate/free per cpu vdso data.
 111  */
 112 #define SEGMENT_ORDER   2
 113
 114 /*
 115  * The initial vdso_data structure for the boot CPU. Eventually
 116  * it is replaced with a properly allocated structure in vdso_init.
 117  * This is necessary because a valid S390_lowcore.vdso_per_cpu_data
 118  * pointer is required to be able to return from an interrupt or
 119  * program check. See the exit paths in entry.S.
 120  */
 121 struct vdso_data boot_vdso_data __initdata;
 122
 123 void __init vdso_alloc_boot_cpu(struct lowcore *lowcore)
 124 {
 125         lowcore->vdso_per_cpu_data = (unsigned long) &boot_vdso_data;
 126 }
 127
 128 int vdso_alloc_per_cpu(struct lowcore *lowcore)
 129 {
 130         unsigned long segment_table, page_table, page_frame;
 131         struct vdso_per_cpu_data *vd;
 132
 133         segment_table = __get_free_pages(GFP_KERNEL, SEGMENT_ORDER);
 134         page_table = get_zeroed_page(GFP_KERNEL);
 135         page_frame = get_zeroed_page(GFP_KERNEL);
 136         if (!segment_table || !page_table || !page_frame)
 137                 goto out;
 138         arch_set_page_dat(virt_to_page(segment_table), SEGMENT_ORDER);
 139         arch_set_page_dat(virt_to_page(page_table), 0);
 140
 141         /* Initialize per-cpu vdso data page */
 142         vd = (struct vdso_per_cpu_data *) page_frame;
 143         vd->cpu_nr = lowcore->cpu_nr;
 144         vd->node_id = cpu_to_node(vd->cpu_nr);
 145
 146         /* Set up page table for the vdso address space */
 147         memset64((u64 *)segment_table, _SEGMENT_ENTRY_EMPTY, _CRST_ENTRIES);
 148         memset64((u64 *)page_table, _PAGE_INVALID, PTRS_PER_PTE);
 149
 150         *(unsigned long *) segment_table = _SEGMENT_ENTRY + page_table;
 151         *(unsigned long *) page_table = _PAGE_PROTECT + page_frame;
 152
 153         lowcore->vdso_asce = segment_table +
 154                 _ASCE_TABLE_LENGTH + _ASCE_USER_BITS + _ASCE_TYPE_SEGMENT;
 155         lowcore->vdso_per_cpu_data = page_frame;
 156
 157         return 0;
 158
 159 out:
 160         free_page(page_frame);
 161         free_page(page_table);
 162         free_pages(segment_table, SEGMENT_ORDER);
 163         return -ENOMEM;
 164 }
 165
 166 void vdso_free_per_cpu(struct lowcore *lowcore)
 167 {
 168         unsigned long segment_table, page_table, page_frame;
 169
 170         segment_table = lowcore->vdso_asce & PAGE_MASK;
 171         page_table = *(unsigned long *) segment_table;
 172         page_frame = *(unsigned long *) page_table;
 173
 174         free_page(page_frame);
 175         free_page(page_table);
 176         free_pages(segment_table, SEGMENT_ORDER);
 177 }
 178
 179 /*
 180  * This is called from binfmt_elf, we create the special vma for the
 181  * vDSO and insert it into the mm struct tree
 182  */
 183 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 184 {
 185         struct mm_struct *mm = current->mm;
 186         struct vm_area_struct *vma;
 187         unsigned long vdso_pages;
 188         unsigned long vdso_base;
 189         int rc;
 190
 191         if (!vdso_enabled)
 192                 return 0;
 193
 194         if (is_compat_task())
 195                 return 0;
 196
 197         vdso_pages = vdso64_pages;
 198         /*
 199          * vDSO has a problem and was disabled, just don't "enable" it for
 200          * the process
 201          */
 202         if (vdso_pages == 0)
 203                 return 0;
 204
 205         /*
 206          * pick a base address for the vDSO in process space. We try to put
 207          * it at vdso_base which is the "natural" base for it, but we might
 208          * fail and end up putting it elsewhere.
 209          */
 210         if (mmap_write_lock_killable(mm))
 211                 return -EINTR;
 212         vdso_base = get_unmapped_area(NULL, 0, vdso_pages << PAGE_SHIFT, 0, 0);
 213         if (IS_ERR_VALUE(vdso_base)) {
 214                 rc = vdso_base;
 215                 goto out_up;
 216         }
 217
 218         /*
 219          * our vma flags don't have VM_WRITE so by default, the process
 220          * isn't allowed to write those pages.
 221          * gdb can break that with ptrace interface, and thus trigger COW
 222          * on those pages but it's then your responsibility to never do that
 223          * on the "data" page of the vDSO or you'll stop getting kernel
 224          * updates and your nice userland gettimeofday will be totally dead.
 225          * It's fine to use that for setting breakpoints in the vDSO code
 226          * pages though.
 227          */
 228         vma = _install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
 229                                        VM_READ|VM_EXEC|
 230                                        VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
 231                                        &vdso_mapping);
 232         if (IS_ERR(vma)) {
 233                 rc = PTR_ERR(vma);
 234                 goto out_up;
 235         }
 236
 237         current->mm->context.vdso_base = vdso_base;
 238         rc = 0;
 239
 240 out_up:
 241         mmap_write_unlock(mm);
 242         return rc;
 243 }
 244
 245 static int __init vdso_init(void)
 246 {
 247         int i;
 248
 249         vdso_init_data(vdso_data);
 250
 251         /* Calculate the size of the 64 bit vDSO */
 252         vdso64_pages = ((&vdso64_end - &vdso64_start
 253                          + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
 254
 255         /* Make sure pages are in the correct state */
 256         vdso64_pagelist = kcalloc(vdso64_pages + 1, sizeof(struct page *),
 257                                   GFP_KERNEL);
 258         BUG_ON(vdso64_pagelist == NULL);
 259         for (i = 0; i < vdso64_pages - 1; i++) {
 260                 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
 261                 get_page(pg);
 262                 vdso64_pagelist[i] = pg;
 263         }
 264         vdso64_pagelist[vdso64_pages - 1] = virt_to_page(vdso_data);
 265         vdso64_pagelist[vdso64_pages] = NULL;
 266         if (vdso_alloc_per_cpu(&S390_lowcore))
 267                 BUG();
 268
 269         get_page(virt_to_page(vdso_data));
 270
 271         return 0;
 272 }
 273 early_initcall(vdso_init);