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/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 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
  48 {
  49         if (likely(vma->vm_mm == current->mm))
  50                 return current->nsproxy->time_ns->vvar_page;
  51         /*
  52          * VM_PFNMAP | VM_IO protect .fault() handler from being called
  53          * through interfaces like /proc/$pid/mem or
  54          * process_vm_{readv,writev}() as long as there's no .access()
  55          * in special_mapping_vmops().
  56          * For more details check_vma_flags() and __access_remote_vm()
  57          */
  58         WARN(1, "vvar_page accessed remotely");
  59         return NULL;
  60 }
  61
  62 /*
  63  * The VVAR page layout depends on whether a task belongs to the root or
  64  * non-root time namespace. Whenever a task changes its namespace, the VVAR
  65  * page tables are cleared and then they will be re-faulted with a
  66  * corresponding layout.
  67  * See also the comment near timens_setup_vdso_data() for details.
  68  */
  69 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
  70 {
  71         struct mm_struct *mm = task->mm;
  72         struct vm_area_struct *vma;
  73
  74         mmap_read_lock(mm);
  75         for (vma = mm->mmap; vma; vma = vma->vm_next) {
  76                 unsigned long size = vma->vm_end - vma->vm_start;
  77
  78                 if (!vma_is_special_mapping(vma, &vvar_mapping))
  79                         continue;
  80                 zap_page_range(vma, vma->vm_start, size);
  81                 break;
  82         }
  83         mmap_read_unlock(mm);
  84         return 0;
  85 }
  86 #else
  87 static inline struct page *find_timens_vvar_page(struct vm_area_struct *vma)
  88 {
  89         return NULL;
  90 }
  91 #endif
  92
  93 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
  94                              struct vm_area_struct *vma, struct vm_fault *vmf)
  95 {
  96         struct page *timens_page = find_timens_vvar_page(vma);
  97         unsigned long addr, pfn;
  98         vm_fault_t err;
  99
 100         switch (vmf->pgoff) {
 101         case VVAR_DATA_PAGE_OFFSET:
 102                 pfn = virt_to_pfn(vdso_data);
 103                 if (timens_page) {
 104                         /*
 105                          * Fault in VVAR page too, since it will be accessed
 106                          * to get clock data anyway.
 107                          */
 108                         addr = vmf->address + VVAR_TIMENS_PAGE_OFFSET * PAGE_SIZE;
 109                         err = vmf_insert_pfn(vma, addr, pfn);
 110                         if (unlikely(err & VM_FAULT_ERROR))
 111                                 return err;
 112                         pfn = page_to_pfn(timens_page);
 113                 }
 114                 break;
 115 #ifdef CONFIG_TIME_NS
 116         case VVAR_TIMENS_PAGE_OFFSET:
 117                 /*
 118                  * If a task belongs to a time namespace then a namespace
 119                  * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
 120                  * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
 121                  * offset.
 122                  * See also the comment near timens_setup_vdso_data().
 123                  */
 124                 if (!timens_page)
 125                         return VM_FAULT_SIGBUS;
 126                 pfn = virt_to_pfn(vdso_data);
 127                 break;
 128 #endif /* CONFIG_TIME_NS */
 129         default:
 130                 return VM_FAULT_SIGBUS;
 131         }
 132         return vmf_insert_pfn(vma, vmf->address, pfn);
 133 }
 134
 135 static int vdso_mremap(const struct vm_special_mapping *sm,
 136                        struct vm_area_struct *vma)
 137 {
 138         current->mm->context.vdso_base = vma->vm_start;
 139         return 0;
 140 }
 141
 142 static struct vm_special_mapping vvar_mapping = {
 143         .name = "[vvar]",
 144         .fault = vvar_fault,
 145 };
 146
 147 static struct vm_special_mapping vdso64_mapping = {
 148         .name = "[vdso]",
 149         .mremap = vdso_mremap,
 150 };
 151
 152 static struct vm_special_mapping vdso32_mapping = {
 153         .name = "[vdso]",
 154         .mremap = vdso_mremap,
 155 };
 156
 157 int vdso_getcpu_init(void)
 158 {
 159         set_tod_programmable_field(smp_processor_id());
 160         return 0;
 161 }
 162 early_initcall(vdso_getcpu_init); /* Must be called before SMP init */
 163
 164 static int map_vdso(unsigned long addr, unsigned long vdso_mapping_len)
 165 {
 166         unsigned long vvar_start, vdso_text_start, vdso_text_len;
 167         struct vm_special_mapping *vdso_mapping;
 168         struct mm_struct *mm = current->mm;
 169         struct vm_area_struct *vma;
 170         int rc;
 171
 172         BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES);
 173         if (mmap_write_lock_killable(mm))
 174                 return -EINTR;
 175
 176         if (is_compat_task()) {
 177                 vdso_text_len = vdso32_end - vdso32_start;
 178                 vdso_mapping = &vdso32_mapping;
 179         } else {
 180                 vdso_text_len = vdso64_end - vdso64_start;
 181                 vdso_mapping = &vdso64_mapping;
 182         }
 183         vvar_start = get_unmapped_area(NULL, addr, vdso_mapping_len, 0, 0);
 184         rc = vvar_start;
 185         if (IS_ERR_VALUE(vvar_start))
 186                 goto out;
 187         vma = _install_special_mapping(mm, vvar_start, VVAR_NR_PAGES*PAGE_SIZE,
 188                                        VM_READ|VM_MAYREAD|VM_IO|VM_DONTDUMP|
 189                                        VM_PFNMAP,
 190                                        &vvar_mapping);
 191         rc = PTR_ERR(vma);
 192         if (IS_ERR(vma))
 193                 goto out;
 194         vdso_text_start = vvar_start + VVAR_NR_PAGES * PAGE_SIZE;
 195         /* VM_MAYWRITE for COW so gdb can set breakpoints */
 196         vma = _install_special_mapping(mm, vdso_text_start, vdso_text_len,
 197                                        VM_READ|VM_EXEC|
 198                                        VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
 199                                        vdso_mapping);
 200         if (IS_ERR(vma)) {
 201                 do_munmap(mm, vvar_start, PAGE_SIZE, NULL);
 202                 rc = PTR_ERR(vma);
 203         } else {
 204                 current->mm->context.vdso_base = vdso_text_start;
 205                 rc = 0;
 206         }
 207 out:
 208         mmap_write_unlock(mm);
 209         return rc;
 210 }
 211
 212 static unsigned long vdso_addr(unsigned long start, unsigned long len)
 213 {
 214         unsigned long addr, end, offset;
 215
 216         /*
 217          * Round up the start address. It can start out unaligned as a result
 218          * of stack start randomization.
 219          */
 220         start = PAGE_ALIGN(start);
 221
 222         /* Round the lowest possible end address up to a PMD boundary. */
 223         end = (start + len + PMD_SIZE - 1) & PMD_MASK;
 224         if (end >= VDSO_BASE)
 225                 end = VDSO_BASE;
 226         end -= len;
 227
 228         if (end > start) {
 229                 offset = get_random_int() % (((end - start) >> PAGE_SHIFT) + 1);
 230                 addr = start + (offset << PAGE_SHIFT);
 231         } else {
 232                 addr = start;
 233         }
 234         return addr;
 235 }
 236
 237 unsigned long vdso_size(void)
 238 {
 239         unsigned long size = VVAR_NR_PAGES * PAGE_SIZE;
 240
 241         if (is_compat_task())
 242                 size += vdso32_end - vdso32_start;
 243         else
 244                 size += vdso64_end - vdso64_start;
 245         return PAGE_ALIGN(size);
 246 }
 247
 248 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 249 {
 250         unsigned long addr = VDSO_BASE;
 251         unsigned long size = vdso_size();
 252
 253         if (current->flags & PF_RANDOMIZE)
 254                 addr = vdso_addr(current->mm->start_stack + PAGE_SIZE, size);
 255         return map_vdso(addr, size);
 256 }
 257
 258 static struct page ** __init vdso_setup_pages(void *start, void *end)
 259 {
 260         int pages = (end - start) >> PAGE_SHIFT;
 261         struct page **pagelist;
 262         int i;
 263
 264         pagelist = kcalloc(pages + 1, sizeof(struct page *), GFP_KERNEL);
 265         if (!pagelist)
 266                 panic("%s: Cannot allocate page list for VDSO", __func__);
 267         for (i = 0; i < pages; i++)
 268                 pagelist[i] = virt_to_page(start + i * PAGE_SIZE);
 269         return pagelist;
 270 }
 271
 272 static int __init vdso_init(void)
 273 {
 274         vdso64_mapping.pages = vdso_setup_pages(vdso64_start, vdso64_end);
 275         if (IS_ENABLED(CONFIG_COMPAT))
 276                 vdso32_mapping.pages = vdso_setup_pages(vdso32_start, vdso32_end);
 277         return 0;
 278 }
 279 arch_initcall(vdso_init);