1 // SPDX-License-Identifier: GPL-2.0-or-later 2 3 /* 4 * Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp. 5 * <benh@kernel.crashing.org> 6 */ 7 8 #include <linux/errno.h> 9 #include <linux/sched.h> 10 #include <linux/kernel.h> 11 #include <linux/mm.h> 12 #include <linux/smp.h> 13 #include <linux/stddef.h> 14 #include <linux/unistd.h> 15 #include <linux/slab.h> 16 #include <linux/user.h> 17 #include <linux/elf.h> 18 #include <linux/security.h> 19 #include <linux/memblock.h> 20 #include <linux/syscalls.h> 21 #include <linux/time_namespace.h> 22 #include <vdso/datapage.h> 23 24 #include <asm/syscall.h> 25 #include <asm/processor.h> 26 #include <asm/mmu.h> 27 #include <asm/mmu_context.h> 28 #include <asm/machdep.h> 29 #include <asm/cputable.h> 30 #include <asm/sections.h> 31 #include <asm/firmware.h> 32 #include <asm/vdso.h> 33 #include <asm/vdso_datapage.h> 34 #include <asm/setup.h> 35 36 /* The alignment of the vDSO */ 37 #define VDSO_ALIGNMENT (1 << 16) 38 39 extern char vdso32_start, vdso32_end; 40 extern char vdso64_start, vdso64_end; 41 42 long sys_ni_syscall(void); 43 44 /* 45 * The vdso data page (aka. systemcfg for old ppc64 fans) is here. 46 * Once the early boot kernel code no longer needs to muck around 47 * with it, it will become dynamically allocated 48 */ 49 static union { 50 struct vdso_arch_data data; 51 u8 page[PAGE_SIZE]; 52 } vdso_data_store __page_aligned_data; 53 struct vdso_arch_data *vdso_data = &vdso_data_store.data; 54 55 enum vvar_pages { 56 VVAR_DATA_PAGE_OFFSET, 57 VVAR_TIMENS_PAGE_OFFSET, 58 VVAR_NR_PAGES, 59 }; 60 61 static int vdso_mremap(const struct vm_special_mapping *sm, struct vm_area_struct *new_vma, 62 unsigned long text_size) 63 { 64 unsigned long new_size = new_vma->vm_end - new_vma->vm_start; 65 66 if (new_size != text_size) 67 return -EINVAL; 68 69 current->mm->context.vdso = (void __user *)new_vma->vm_start; 70 71 return 0; 72 } 73 74 static int vdso32_mremap(const struct vm_special_mapping *sm, struct vm_area_struct *new_vma) 75 { 76 return vdso_mremap(sm, new_vma, &vdso32_end - &vdso32_start); 77 } 78 79 static int vdso64_mremap(const struct vm_special_mapping *sm, struct vm_area_struct *new_vma) 80 { 81 return vdso_mremap(sm, new_vma, &vdso64_end - &vdso64_start); 82 } 83 84 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm, 85 struct vm_area_struct *vma, struct vm_fault *vmf); 86 87 static struct vm_special_mapping vvar_spec __ro_after_init = { 88 .name = "[vvar]", 89 .fault = vvar_fault, 90 }; 91 92 static struct vm_special_mapping vdso32_spec __ro_after_init = { 93 .name = "[vdso]", 94 .mremap = vdso32_mremap, 95 }; 96 97 static struct vm_special_mapping vdso64_spec __ro_after_init = { 98 .name = "[vdso]", 99 .mremap = vdso64_mremap, 100 }; 101 102 #ifdef CONFIG_TIME_NS 103 struct vdso_data *arch_get_vdso_data(void *vvar_page) 104 { 105 return ((struct vdso_arch_data *)vvar_page)->data; 106 } 107 108 /* 109 * The vvar mapping contains data for a specific time namespace, so when a task 110 * changes namespace we must unmap its vvar data for the old namespace. 111 * Subsequent faults will map in data for the new namespace. 112 * 113 * For more details see timens_setup_vdso_data(). 114 */ 115 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns) 116 { 117 struct mm_struct *mm = task->mm; 118 VMA_ITERATOR(vmi, mm, 0); 119 struct vm_area_struct *vma; 120 121 mmap_read_lock(mm); 122 for_each_vma(vmi, vma) { 123 unsigned long size = vma->vm_end - vma->vm_start; 124 125 if (vma_is_special_mapping(vma, &vvar_spec)) 126 zap_page_range(vma, vma->vm_start, size); 127 } 128 mmap_read_unlock(mm); 129 130 return 0; 131 } 132 133 static struct page *find_timens_vvar_page(struct vm_area_struct *vma) 134 { 135 if (likely(vma->vm_mm == current->mm)) 136 return current->nsproxy->time_ns->vvar_page; 137 138 /* 139 * VM_PFNMAP | VM_IO protect .fault() handler from being called 140 * through interfaces like /proc/$pid/mem or 141 * process_vm_{readv,writev}() as long as there's no .access() 142 * in special_mapping_vmops. 143 * For more details check_vma_flags() and __access_remote_vm() 144 */ 145 WARN(1, "vvar_page accessed remotely"); 146 147 return NULL; 148 } 149 #else 150 static struct page *find_timens_vvar_page(struct vm_area_struct *vma) 151 { 152 return NULL; 153 } 154 #endif 155 156 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm, 157 struct vm_area_struct *vma, struct vm_fault *vmf) 158 { 159 struct page *timens_page = find_timens_vvar_page(vma); 160 unsigned long pfn; 161 162 switch (vmf->pgoff) { 163 case VVAR_DATA_PAGE_OFFSET: 164 if (timens_page) 165 pfn = page_to_pfn(timens_page); 166 else 167 pfn = virt_to_pfn(vdso_data); 168 break; 169 #ifdef CONFIG_TIME_NS 170 case VVAR_TIMENS_PAGE_OFFSET: 171 /* 172 * If a task belongs to a time namespace then a namespace 173 * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and 174 * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET 175 * offset. 176 * See also the comment near timens_setup_vdso_data(). 177 */ 178 if (!timens_page) 179 return VM_FAULT_SIGBUS; 180 pfn = virt_to_pfn(vdso_data); 181 break; 182 #endif /* CONFIG_TIME_NS */ 183 default: 184 return VM_FAULT_SIGBUS; 185 } 186 187 return vmf_insert_pfn(vma, vmf->address, pfn); 188 } 189 190 /* 191 * This is called from binfmt_elf, we create the special vma for the 192 * vDSO and insert it into the mm struct tree 193 */ 194 static int __arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp) 195 { 196 unsigned long vdso_size, vdso_base, mappings_size; 197 struct vm_special_mapping *vdso_spec; 198 unsigned long vvar_size = VVAR_NR_PAGES * PAGE_SIZE; 199 struct mm_struct *mm = current->mm; 200 struct vm_area_struct *vma; 201 202 if (is_32bit_task()) { 203 vdso_spec = &vdso32_spec; 204 vdso_size = &vdso32_end - &vdso32_start; 205 } else { 206 vdso_spec = &vdso64_spec; 207 vdso_size = &vdso64_end - &vdso64_start; 208 } 209 210 mappings_size = vdso_size + vvar_size; 211 mappings_size += (VDSO_ALIGNMENT - 1) & PAGE_MASK; 212 213 /* 214 * Pick a base address for the vDSO in process space. 215 * Add enough to the size so that the result can be aligned. 216 */ 217 vdso_base = get_unmapped_area(NULL, 0, mappings_size, 0, 0); 218 if (IS_ERR_VALUE(vdso_base)) 219 return vdso_base; 220 221 /* Add required alignment. */ 222 vdso_base = ALIGN(vdso_base, VDSO_ALIGNMENT); 223 224 /* 225 * Put vDSO base into mm struct. We need to do this before calling 226 * install_special_mapping or the perf counter mmap tracking code 227 * will fail to recognise it as a vDSO. 228 */ 229 mm->context.vdso = (void __user *)vdso_base + vvar_size; 230 231 vma = _install_special_mapping(mm, vdso_base, vvar_size, 232 VM_READ | VM_MAYREAD | VM_IO | 233 VM_DONTDUMP | VM_PFNMAP, &vvar_spec); 234 if (IS_ERR(vma)) 235 return PTR_ERR(vma); 236 237 /* 238 * our vma flags don't have VM_WRITE so by default, the process isn't 239 * allowed to write those pages. 240 * gdb can break that with ptrace interface, and thus trigger COW on 241 * those pages but it's then your responsibility to never do that on 242 * the "data" page of the vDSO or you'll stop getting kernel updates 243 * and your nice userland gettimeofday will be totally dead. 244 * It's fine to use that for setting breakpoints in the vDSO code 245 * pages though. 246 */ 247 vma = _install_special_mapping(mm, vdso_base + vvar_size, vdso_size, 248 VM_READ | VM_EXEC | VM_MAYREAD | 249 VM_MAYWRITE | VM_MAYEXEC, vdso_spec); 250 if (IS_ERR(vma)) 251 do_munmap(mm, vdso_base, vvar_size, NULL); 252 253 return PTR_ERR_OR_ZERO(vma); 254 } 255 256 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp) 257 { 258 struct mm_struct *mm = current->mm; 259 int rc; 260 261 mm->context.vdso = NULL; 262 263 if (mmap_write_lock_killable(mm)) 264 return -EINTR; 265 266 rc = __arch_setup_additional_pages(bprm, uses_interp); 267 if (rc) 268 mm->context.vdso = NULL; 269 270 mmap_write_unlock(mm); 271 return rc; 272 } 273 274 #define VDSO_DO_FIXUPS(type, value, bits, sec) do { \ 275 void *__start = (void *)VDSO##bits##_SYMBOL(&vdso##bits##_start, sec##_start); \ 276 void *__end = (void *)VDSO##bits##_SYMBOL(&vdso##bits##_start, sec##_end); \ 277 \ 278 do_##type##_fixups((value), __start, __end); \ 279 } while (0) 280 281 static void __init vdso_fixup_features(void) 282 { 283 #ifdef CONFIG_PPC64 284 VDSO_DO_FIXUPS(feature, cur_cpu_spec->cpu_features, 64, ftr_fixup); 285 VDSO_DO_FIXUPS(feature, cur_cpu_spec->mmu_features, 64, mmu_ftr_fixup); 286 VDSO_DO_FIXUPS(feature, powerpc_firmware_features, 64, fw_ftr_fixup); 287 VDSO_DO_FIXUPS(lwsync, cur_cpu_spec->cpu_features, 64, lwsync_fixup); 288 #endif /* CONFIG_PPC64 */ 289 290 #ifdef CONFIG_VDSO32 291 VDSO_DO_FIXUPS(feature, cur_cpu_spec->cpu_features, 32, ftr_fixup); 292 VDSO_DO_FIXUPS(feature, cur_cpu_spec->mmu_features, 32, mmu_ftr_fixup); 293 #ifdef CONFIG_PPC64 294 VDSO_DO_FIXUPS(feature, powerpc_firmware_features, 32, fw_ftr_fixup); 295 #endif /* CONFIG_PPC64 */ 296 VDSO_DO_FIXUPS(lwsync, cur_cpu_spec->cpu_features, 32, lwsync_fixup); 297 #endif 298 } 299 300 /* 301 * Called from setup_arch to initialize the bitmap of available 302 * syscalls in the systemcfg page 303 */ 304 static void __init vdso_setup_syscall_map(void) 305 { 306 unsigned int i; 307 308 for (i = 0; i < NR_syscalls; i++) { 309 if (sys_call_table[i] != (void *)&sys_ni_syscall) 310 vdso_data->syscall_map[i >> 5] |= 0x80000000UL >> (i & 0x1f); 311 if (IS_ENABLED(CONFIG_COMPAT) && 312 compat_sys_call_table[i] != (void *)&sys_ni_syscall) 313 vdso_data->compat_syscall_map[i >> 5] |= 0x80000000UL >> (i & 0x1f); 314 } 315 } 316 317 #ifdef CONFIG_PPC64 318 int vdso_getcpu_init(void) 319 { 320 unsigned long cpu, node, val; 321 322 /* 323 * SPRG_VDSO contains the CPU in the bottom 16 bits and the NUMA node 324 * in the next 16 bits. The VDSO uses this to implement getcpu(). 325 */ 326 cpu = get_cpu(); 327 WARN_ON_ONCE(cpu > 0xffff); 328 329 node = cpu_to_node(cpu); 330 WARN_ON_ONCE(node > 0xffff); 331 332 val = (cpu & 0xffff) | ((node & 0xffff) << 16); 333 mtspr(SPRN_SPRG_VDSO_WRITE, val); 334 get_paca()->sprg_vdso = val; 335 336 put_cpu(); 337 338 return 0; 339 } 340 /* We need to call this before SMP init */ 341 early_initcall(vdso_getcpu_init); 342 #endif 343 344 static struct page ** __init vdso_setup_pages(void *start, void *end) 345 { 346 int i; 347 struct page **pagelist; 348 int pages = (end - start) >> PAGE_SHIFT; 349 350 pagelist = kcalloc(pages + 1, sizeof(struct page *), GFP_KERNEL); 351 if (!pagelist) 352 panic("%s: Cannot allocate page list for VDSO", __func__); 353 354 for (i = 0; i < pages; i++) 355 pagelist[i] = virt_to_page(start + i * PAGE_SIZE); 356 357 return pagelist; 358 } 359 360 static int __init vdso_init(void) 361 { 362 #ifdef CONFIG_PPC64 363 /* 364 * Fill up the "systemcfg" stuff for backward compatibility 365 */ 366 strcpy((char *)vdso_data->eye_catcher, "SYSTEMCFG:PPC64"); 367 vdso_data->version.major = SYSTEMCFG_MAJOR; 368 vdso_data->version.minor = SYSTEMCFG_MINOR; 369 vdso_data->processor = mfspr(SPRN_PVR); 370 /* 371 * Fake the old platform number for pSeries and add 372 * in LPAR bit if necessary 373 */ 374 vdso_data->platform = 0x100; 375 if (firmware_has_feature(FW_FEATURE_LPAR)) 376 vdso_data->platform |= 1; 377 vdso_data->physicalMemorySize = memblock_phys_mem_size(); 378 vdso_data->dcache_size = ppc64_caches.l1d.size; 379 vdso_data->dcache_line_size = ppc64_caches.l1d.line_size; 380 vdso_data->icache_size = ppc64_caches.l1i.size; 381 vdso_data->icache_line_size = ppc64_caches.l1i.line_size; 382 vdso_data->dcache_block_size = ppc64_caches.l1d.block_size; 383 vdso_data->icache_block_size = ppc64_caches.l1i.block_size; 384 vdso_data->dcache_log_block_size = ppc64_caches.l1d.log_block_size; 385 vdso_data->icache_log_block_size = ppc64_caches.l1i.log_block_size; 386 #endif /* CONFIG_PPC64 */ 387 388 vdso_setup_syscall_map(); 389 390 vdso_fixup_features(); 391 392 if (IS_ENABLED(CONFIG_VDSO32)) 393 vdso32_spec.pages = vdso_setup_pages(&vdso32_start, &vdso32_end); 394 395 if (IS_ENABLED(CONFIG_PPC64)) 396 vdso64_spec.pages = vdso_setup_pages(&vdso64_start, &vdso64_end); 397 398 smp_wmb(); 399 400 return 0; 401 } 402 arch_initcall(vdso_init); 403