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 VMA_ITERATOR(vmi, mm, 0); 73 struct vm_area_struct *vma; 74 75 mmap_read_lock(mm); 76 for_each_vma(vmi, vma) { 77 unsigned long size = vma->vm_end - vma->vm_start; 78 79 if (!vma_is_special_mapping(vma, &vvar_mapping)) 80 continue; 81 zap_page_range(vma, vma->vm_start, size); 82 break; 83 } 84 mmap_read_unlock(mm); 85 return 0; 86 } 87 #else 88 static inline struct page *find_timens_vvar_page(struct vm_area_struct *vma) 89 { 90 return NULL; 91 } 92 #endif 93 94 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm, 95 struct vm_area_struct *vma, struct vm_fault *vmf) 96 { 97 struct page *timens_page = find_timens_vvar_page(vma); 98 unsigned long addr, pfn; 99 vm_fault_t err; 100 101 switch (vmf->pgoff) { 102 case VVAR_DATA_PAGE_OFFSET: 103 pfn = virt_to_pfn(vdso_data); 104 if (timens_page) { 105 /* 106 * Fault in VVAR page too, since it will be accessed 107 * to get clock data anyway. 108 */ 109 addr = vmf->address + VVAR_TIMENS_PAGE_OFFSET * PAGE_SIZE; 110 err = vmf_insert_pfn(vma, addr, pfn); 111 if (unlikely(err & VM_FAULT_ERROR)) 112 return err; 113 pfn = page_to_pfn(timens_page); 114 } 115 break; 116 #ifdef CONFIG_TIME_NS 117 case VVAR_TIMENS_PAGE_OFFSET: 118 /* 119 * If a task belongs to a time namespace then a namespace 120 * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and 121 * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET 122 * offset. 123 * See also the comment near timens_setup_vdso_data(). 124 */ 125 if (!timens_page) 126 return VM_FAULT_SIGBUS; 127 pfn = virt_to_pfn(vdso_data); 128 break; 129 #endif /* CONFIG_TIME_NS */ 130 default: 131 return VM_FAULT_SIGBUS; 132 } 133 return vmf_insert_pfn(vma, vmf->address, pfn); 134 } 135 136 static int vdso_mremap(const struct vm_special_mapping *sm, 137 struct vm_area_struct *vma) 138 { 139 current->mm->context.vdso_base = vma->vm_start; 140 return 0; 141 } 142 143 static struct vm_special_mapping vvar_mapping = { 144 .name = "[vvar]", 145 .fault = vvar_fault, 146 }; 147 148 static struct vm_special_mapping vdso64_mapping = { 149 .name = "[vdso]", 150 .mremap = vdso_mremap, 151 }; 152 153 static struct vm_special_mapping vdso32_mapping = { 154 .name = "[vdso]", 155 .mremap = vdso_mremap, 156 }; 157 158 int vdso_getcpu_init(void) 159 { 160 set_tod_programmable_field(smp_processor_id()); 161 return 0; 162 } 163 early_initcall(vdso_getcpu_init); /* Must be called before SMP init */ 164 165 static int map_vdso(unsigned long addr, unsigned long vdso_mapping_len) 166 { 167 unsigned long vvar_start, vdso_text_start, vdso_text_len; 168 struct vm_special_mapping *vdso_mapping; 169 struct mm_struct *mm = current->mm; 170 struct vm_area_struct *vma; 171 int rc; 172 173 BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES); 174 if (mmap_write_lock_killable(mm)) 175 return -EINTR; 176 177 if (is_compat_task()) { 178 vdso_text_len = vdso32_end - vdso32_start; 179 vdso_mapping = &vdso32_mapping; 180 } else { 181 vdso_text_len = vdso64_end - vdso64_start; 182 vdso_mapping = &vdso64_mapping; 183 } 184 vvar_start = get_unmapped_area(NULL, addr, vdso_mapping_len, 0, 0); 185 rc = vvar_start; 186 if (IS_ERR_VALUE(vvar_start)) 187 goto out; 188 vma = _install_special_mapping(mm, vvar_start, VVAR_NR_PAGES*PAGE_SIZE, 189 VM_READ|VM_MAYREAD|VM_IO|VM_DONTDUMP| 190 VM_PFNMAP, 191 &vvar_mapping); 192 rc = PTR_ERR(vma); 193 if (IS_ERR(vma)) 194 goto out; 195 vdso_text_start = vvar_start + VVAR_NR_PAGES * PAGE_SIZE; 196 /* VM_MAYWRITE for COW so gdb can set breakpoints */ 197 vma = _install_special_mapping(mm, vdso_text_start, vdso_text_len, 198 VM_READ|VM_EXEC| 199 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC, 200 vdso_mapping); 201 if (IS_ERR(vma)) { 202 do_munmap(mm, vvar_start, PAGE_SIZE, NULL); 203 rc = PTR_ERR(vma); 204 } else { 205 current->mm->context.vdso_base = vdso_text_start; 206 rc = 0; 207 } 208 out: 209 mmap_write_unlock(mm); 210 return rc; 211 } 212 213 static unsigned long vdso_addr(unsigned long start, unsigned long len) 214 { 215 unsigned long addr, end, offset; 216 217 /* 218 * Round up the start address. It can start out unaligned as a result 219 * of stack start randomization. 220 */ 221 start = PAGE_ALIGN(start); 222 223 /* Round the lowest possible end address up to a PMD boundary. */ 224 end = (start + len + PMD_SIZE - 1) & PMD_MASK; 225 if (end >= VDSO_BASE) 226 end = VDSO_BASE; 227 end -= len; 228 229 if (end > start) { 230 offset = prandom_u32_max(((end - start) >> PAGE_SHIFT) + 1); 231 addr = start + (offset << PAGE_SHIFT); 232 } else { 233 addr = start; 234 } 235 return addr; 236 } 237 238 unsigned long vdso_size(void) 239 { 240 unsigned long size = VVAR_NR_PAGES * PAGE_SIZE; 241 242 if (is_compat_task()) 243 size += vdso32_end - vdso32_start; 244 else 245 size += vdso64_end - vdso64_start; 246 return PAGE_ALIGN(size); 247 } 248 249 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp) 250 { 251 unsigned long addr = VDSO_BASE; 252 unsigned long size = vdso_size(); 253 254 if (current->flags & PF_RANDOMIZE) 255 addr = vdso_addr(current->mm->start_stack + PAGE_SIZE, size); 256 return map_vdso(addr, size); 257 } 258 259 static struct page ** __init vdso_setup_pages(void *start, void *end) 260 { 261 int pages = (end - start) >> PAGE_SHIFT; 262 struct page **pagelist; 263 int i; 264 265 pagelist = kcalloc(pages + 1, sizeof(struct page *), GFP_KERNEL); 266 if (!pagelist) 267 panic("%s: Cannot allocate page list for VDSO", __func__); 268 for (i = 0; i < pages; i++) 269 pagelist[i] = virt_to_page(start + i * PAGE_SIZE); 270 return pagelist; 271 } 272 273 static int __init vdso_init(void) 274 { 275 vdso64_mapping.pages = vdso_setup_pages(vdso64_start, vdso64_end); 276 if (IS_ENABLED(CONFIG_COMPAT)) 277 vdso32_mapping.pages = vdso_setup_pages(vdso32_start, vdso32_end); 278 return 0; 279 } 280 arch_initcall(vdso_init); 281