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 ([email protected])
   7  */
   8
   9 #include <linux/binfmts.h>
  10 #include <linux/compat.h>
  11 #include <linux/elf.h>
  12 #include <linux/errno.h>
  13 #include <linux/init.h>
  14 #include <linux/kernel.h>
  15 #include <linux/mm.h>
  16 #include <linux/slab.h>
  17 #include <linux/smp.h>
  18 #include <linux/time_namespace.h>
  19 #include <linux/random.h>
  20 #include <vdso/datapage.h>
  21 #include <asm/vdso.h>
  22
  23 extern char vdso64_start[], vdso64_end[];
  24 extern char vdso32_start[], vdso32_end[];
  25
  26 static struct vm_special_mapping vvar_mapping;
  27
  28 static union {
  29         struct vdso_data        data[CS_BASES];
  30         u8                      page[PAGE_SIZE];
  31 } vdso_data_store __page_aligned_data;
  32
  33 struct vdso_data *vdso_data = vdso_data_store.data;
  34
  35 enum vvar_pages {
  36         VVAR_DATA_PAGE_OFFSET,
  37         VVAR_TIMENS_PAGE_OFFSET,
  38         VVAR_NR_PAGES,
  39 };
  40
  41 #ifdef CONFIG_TIME_NS
  42 struct vdso_data *arch_get_vdso_data(void *vvar_page)
  43 {
  44         return (struct vdso_data *)(vvar_page);
  45 }
  46
  47 /*
  48  * The VVAR page layout depends on whether a task belongs to the root or
  49  * non-root time namespace. Whenever a task changes its namespace, the VVAR
  50  * page tables are cleared and then they will be re-faulted with a
  51  * corresponding layout.
  52  * See also the comment near timens_setup_vdso_data() for details.
  53  */
  54 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
  55 {
  56         struct mm_struct *mm = task->mm;
  57         VMA_ITERATOR(vmi, mm, 0);
  58         struct vm_area_struct *vma;
  59
  60         mmap_read_lock(mm);
  61         for_each_vma(vmi, vma) {
  62                 unsigned long size = vma->vm_end - vma->vm_start;
  63
  64                 if (!vma_is_special_mapping(vma, &vvar_mapping))
  65                         continue;
  66                 zap_page_range(vma, vma->vm_start, size);
  67                 break;
  68         }
  69         mmap_read_unlock(mm);
  70         return 0;
  71 }
  72 #endif
  73
  74 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
  75                              struct vm_area_struct *vma, struct vm_fault *vmf)
  76 {
  77         struct page *timens_page = find_timens_vvar_page(vma);
  78         unsigned long addr, pfn;
  79         vm_fault_t err;
  80
  81         switch (vmf->pgoff) {
  82         case VVAR_DATA_PAGE_OFFSET:
  83                 pfn = virt_to_pfn(vdso_data);
  84                 if (timens_page) {
  85                         /*
  86                          * Fault in VVAR page too, since it will be accessed
  87                          * to get clock data anyway.
  88                          */
  89                         addr = vmf->address + VVAR_TIMENS_PAGE_OFFSET * PAGE_SIZE;
  90                         err = vmf_insert_pfn(vma, addr, pfn);
  91                         if (unlikely(err & VM_FAULT_ERROR))
  92                                 return err;
  93                         pfn = page_to_pfn(timens_page);
  94                 }
  95                 break;
  96 #ifdef CONFIG_TIME_NS
  97         case VVAR_TIMENS_PAGE_OFFSET:
  98                 /*
  99                  * If a task belongs to a time namespace then a namespace
 100                  * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
 101                  * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
 102                  * offset.
 103                  * See also the comment near timens_setup_vdso_data().
 104                  */
 105                 if (!timens_page)
 106                         return VM_FAULT_SIGBUS;
 107                 pfn = virt_to_pfn(vdso_data);
 108                 break;
 109 #endif /* CONFIG_TIME_NS */
 110         default:
 111                 return VM_FAULT_SIGBUS;
 112         }
 113         return vmf_insert_pfn(vma, vmf->address, pfn);
 114 }
 115
 116 static int vdso_mremap(const struct vm_special_mapping *sm,
 117                        struct vm_area_struct *vma)
 118 {
 119         current->mm->context.vdso_base = vma->vm_start;
 120         return 0;
 121 }
 122
 123 static struct vm_special_mapping vvar_mapping = {
 124         .name = "[vvar]",
 125         .fault = vvar_fault,
 126 };
 127
 128 static struct vm_special_mapping vdso64_mapping = {
 129         .name = "[vdso]",
 130         .mremap = vdso_mremap,
 131 };
 132
 133 static struct vm_special_mapping vdso32_mapping = {
 134         .name = "[vdso]",
 135         .mremap = vdso_mremap,
 136 };
 137
 138 int vdso_getcpu_init(void)
 139 {
 140         set_tod_programmable_field(smp_processor_id());
 141         return 0;
 142 }
 143 early_initcall(vdso_getcpu_init); /* Must be called before SMP init */
 144
 145 static int map_vdso(unsigned long addr, unsigned long vdso_mapping_len)
 146 {
 147         unsigned long vvar_start, vdso_text_start, vdso_text_len;
 148         struct vm_special_mapping *vdso_mapping;
 149         struct mm_struct *mm = current->mm;
 150         struct vm_area_struct *vma;
 151         int rc;
 152
 153         BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES);
 154         if (mmap_write_lock_killable(mm))
 155                 return -EINTR;
 156
 157         if (is_compat_task()) {
 158                 vdso_text_len = vdso32_end - vdso32_start;
 159                 vdso_mapping = &vdso32_mapping;
 160         } else {
 161                 vdso_text_len = vdso64_end - vdso64_start;
 162                 vdso_mapping = &vdso64_mapping;
 163         }
 164         vvar_start = get_unmapped_area(NULL, addr, vdso_mapping_len, 0, 0);
 165         rc = vvar_start;
 166         if (IS_ERR_VALUE(vvar_start))
 167                 goto out;
 168         vma = _install_special_mapping(mm, vvar_start, VVAR_NR_PAGES*PAGE_SIZE,
 169                                        VM_READ|VM_MAYREAD|VM_IO|VM_DONTDUMP|
 170                                        VM_PFNMAP,
 171                                        &vvar_mapping);
 172         rc = PTR_ERR(vma);
 173         if (IS_ERR(vma))
 174                 goto out;
 175         vdso_text_start = vvar_start + VVAR_NR_PAGES * PAGE_SIZE;
 176         /* VM_MAYWRITE for COW so gdb can set breakpoints */
 177         vma = _install_special_mapping(mm, vdso_text_start, vdso_text_len,
 178                                        VM_READ|VM_EXEC|
 179                                        VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
 180                                        vdso_mapping);
 181         if (IS_ERR(vma)) {
 182                 do_munmap(mm, vvar_start, PAGE_SIZE, NULL);
 183                 rc = PTR_ERR(vma);
 184         } else {
 185                 current->mm->context.vdso_base = vdso_text_start;
 186                 rc = 0;
 187         }
 188 out:
 189         mmap_write_unlock(mm);
 190         return rc;
 191 }
 192
 193 static unsigned long vdso_addr(unsigned long start, unsigned long len)
 194 {
 195         unsigned long addr, end, offset;
 196
 197         /*
 198          * Round up the start address. It can start out unaligned as a result
 199          * of stack start randomization.
 200          */
 201         start = PAGE_ALIGN(start);
 202
 203         /* Round the lowest possible end address up to a PMD boundary. */
 204         end = (start + len + PMD_SIZE - 1) & PMD_MASK;
 205         if (end >= VDSO_BASE)
 206                 end = VDSO_BASE;
 207         end -= len;
 208
 209         if (end > start) {
 210                 offset = get_random_u32_below(((end - start) >> PAGE_SHIFT) + 1);
 211                 addr = start + (offset << PAGE_SHIFT);
 212         } else {
 213                 addr = start;
 214         }
 215         return addr;
 216 }
 217
 218 unsigned long vdso_size(void)
 219 {
 220         unsigned long size = VVAR_NR_PAGES * PAGE_SIZE;
 221
 222         if (is_compat_task())
 223                 size += vdso32_end - vdso32_start;
 224         else
 225                 size += vdso64_end - vdso64_start;
 226         return PAGE_ALIGN(size);
 227 }
 228
 229 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 230 {
 231         unsigned long addr = VDSO_BASE;
 232         unsigned long size = vdso_size();
 233
 234         if (current->flags & PF_RANDOMIZE)
 235                 addr = vdso_addr(current->mm->start_stack + PAGE_SIZE, size);
 236         return map_vdso(addr, size);
 237 }
 238
 239 static struct page ** __init vdso_setup_pages(void *start, void *end)
 240 {
 241         int pages = (end - start) >> PAGE_SHIFT;
 242         struct page **pagelist;
 243         int i;
 244
 245         pagelist = kcalloc(pages + 1, sizeof(struct page *), GFP_KERNEL);
 246         if (!pagelist)
 247                 panic("%s: Cannot allocate page list for VDSO", __func__);
 248         for (i = 0; i < pages; i++)
 249                 pagelist[i] = virt_to_page(start + i * PAGE_SIZE);
 250         return pagelist;
 251 }
 252
 253 static int __init vdso_init(void)
 254 {
 255         vdso64_mapping.pages = vdso_setup_pages(vdso64_start, vdso64_end);
 256         if (IS_ENABLED(CONFIG_COMPAT))
 257                 vdso32_mapping.pages = vdso_setup_pages(vdso32_start, vdso32_end);
 258         return 0;
 259 }
 260 arch_initcall(vdso_init);