1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * VDSO implementations.
4 *
5 * Copyright (C) 2012 ARM Limited
6 *
7 * Author: Will Deacon <will.deacon@arm.com>
8 */
9
10 #include <linux/cache.h>
11 #include <linux/clocksource.h>
12 #include <linux/elf.h>
13 #include <linux/err.h>
14 #include <linux/errno.h>
15 #include <linux/gfp.h>
16 #include <linux/kernel.h>
17 #include <linux/mm.h>
18 #include <linux/sched.h>
19 #include <linux/signal.h>
20 #include <linux/slab.h>
21 #include <linux/time_namespace.h>
22 #include <linux/timekeeper_internal.h>
23 #include <linux/vmalloc.h>
24 #include <vdso/datapage.h>
25 #include <vdso/helpers.h>
26 #include <vdso/vsyscall.h>
27
28 #include <asm/cacheflush.h>
29 #include <asm/signal32.h>
30 #include <asm/vdso.h>
31
32 extern char vdso_start[], vdso_end[];
33 extern char vdso32_start[], vdso32_end[];
34
35 enum vdso_abi {
36 VDSO_ABI_AA64,
37 VDSO_ABI_AA32,
38 };
39
40 enum vvar_pages {
41 VVAR_DATA_PAGE_OFFSET,
42 VVAR_TIMENS_PAGE_OFFSET,
43 VVAR_NR_PAGES,
44 };
45
46 struct vdso_abi_info {
47 const char *name;
48 const char *vdso_code_start;
49 const char *vdso_code_end;
50 unsigned long vdso_pages;
51 /* Data Mapping */
52 struct vm_special_mapping *dm;
53 /* Code Mapping */
54 struct vm_special_mapping *cm;
55 };
56
57 static struct vdso_abi_info vdso_info[] __ro_after_init = {
58 [VDSO_ABI_AA64] = {
59 .name = "vdso",
60 .vdso_code_start = vdso_start,
61 .vdso_code_end = vdso_end,
62 },
63 #ifdef CONFIG_COMPAT_VDSO
64 [VDSO_ABI_AA32] = {
65 .name = "vdso32",
66 .vdso_code_start = vdso32_start,
67 .vdso_code_end = vdso32_end,
68 },
69 #endif /* CONFIG_COMPAT_VDSO */
70 };
71
72 /*
73 * The vDSO data page.
74 */
75 static union {
76 struct vdso_data data[CS_BASES];
77 u8 page[PAGE_SIZE];
78 } vdso_data_store __page_aligned_data;
79 struct vdso_data *vdso_data = vdso_data_store.data;
80
vdso_mremap(const struct vm_special_mapping * sm,struct vm_area_struct * new_vma)81 static int vdso_mremap(const struct vm_special_mapping *sm,
82 struct vm_area_struct *new_vma)
83 {
84 current->mm->context.vdso = (void *)new_vma->vm_start;
85
86 return 0;
87 }
88
__vdso_init(enum vdso_abi abi)89 static int __init __vdso_init(enum vdso_abi abi)
90 {
91 int i;
92 struct page **vdso_pagelist;
93 unsigned long pfn;
94
95 if (memcmp(vdso_info[abi].vdso_code_start, "\177ELF", 4)) {
96 pr_err("vDSO is not a valid ELF object!\n");
97 return -EINVAL;
98 }
99
100 vdso_info[abi].vdso_pages = (
101 vdso_info[abi].vdso_code_end -
102 vdso_info[abi].vdso_code_start) >>
103 PAGE_SHIFT;
104
105 vdso_pagelist = kcalloc(vdso_info[abi].vdso_pages,
106 sizeof(struct page *),
107 GFP_KERNEL);
108 if (vdso_pagelist == NULL)
109 return -ENOMEM;
110
111 /* Grab the vDSO code pages. */
112 pfn = sym_to_pfn(vdso_info[abi].vdso_code_start);
113
114 for (i = 0; i < vdso_info[abi].vdso_pages; i++)
115 vdso_pagelist[i] = pfn_to_page(pfn + i);
116
117 vdso_info[abi].cm->pages = vdso_pagelist;
118
119 return 0;
120 }
121
122 #ifdef CONFIG_TIME_NS
arch_get_vdso_data(void * vvar_page)123 struct vdso_data *arch_get_vdso_data(void *vvar_page)
124 {
125 return (struct vdso_data *)(vvar_page);
126 }
127
128 /*
129 * The vvar mapping contains data for a specific time namespace, so when a task
130 * changes namespace we must unmap its vvar data for the old namespace.
131 * Subsequent faults will map in data for the new namespace.
132 *
133 * For more details see timens_setup_vdso_data().
134 */
vdso_join_timens(struct task_struct * task,struct time_namespace * ns)135 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
136 {
137 struct mm_struct *mm = task->mm;
138 struct vm_area_struct *vma;
139
140 mmap_read_lock(mm);
141
142 for (vma = mm->mmap; vma; vma = vma->vm_next) {
143 unsigned long size = vma->vm_end - vma->vm_start;
144
145 if (vma_is_special_mapping(vma, vdso_info[VDSO_ABI_AA64].dm))
146 zap_page_range(vma, vma->vm_start, size);
147 #ifdef CONFIG_COMPAT_VDSO
148 if (vma_is_special_mapping(vma, vdso_info[VDSO_ABI_AA32].dm))
149 zap_page_range(vma, vma->vm_start, size);
150 #endif
151 }
152
153 mmap_read_unlock(mm);
154 return 0;
155 }
156
find_timens_vvar_page(struct vm_area_struct * vma)157 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
158 {
159 if (likely(vma->vm_mm == current->mm))
160 return current->nsproxy->time_ns->vvar_page;
161
162 /*
163 * VM_PFNMAP | VM_IO protect .fault() handler from being called
164 * through interfaces like /proc/$pid/mem or
165 * process_vm_{readv,writev}() as long as there's no .access()
166 * in special_mapping_vmops.
167 * For more details check_vma_flags() and __access_remote_vm()
168 */
169 WARN(1, "vvar_page accessed remotely");
170
171 return NULL;
172 }
173 #else
find_timens_vvar_page(struct vm_area_struct * vma)174 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
175 {
176 return NULL;
177 }
178 #endif
179
vvar_fault(const struct vm_special_mapping * sm,struct vm_area_struct * vma,struct vm_fault * vmf)180 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
181 struct vm_area_struct *vma, struct vm_fault *vmf)
182 {
183 struct page *timens_page = find_timens_vvar_page(vma);
184 unsigned long pfn;
185
186 switch (vmf->pgoff) {
187 case VVAR_DATA_PAGE_OFFSET:
188 if (timens_page)
189 pfn = page_to_pfn(timens_page);
190 else
191 pfn = sym_to_pfn(vdso_data);
192 break;
193 #ifdef CONFIG_TIME_NS
194 case VVAR_TIMENS_PAGE_OFFSET:
195 /*
196 * If a task belongs to a time namespace then a namespace
197 * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
198 * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
199 * offset.
200 * See also the comment near timens_setup_vdso_data().
201 */
202 if (!timens_page)
203 return VM_FAULT_SIGBUS;
204 pfn = sym_to_pfn(vdso_data);
205 break;
206 #endif /* CONFIG_TIME_NS */
207 default:
208 return VM_FAULT_SIGBUS;
209 }
210
211 return vmf_insert_pfn(vma, vmf->address, pfn);
212 }
213
__setup_additional_pages(enum vdso_abi abi,struct mm_struct * mm,struct linux_binprm * bprm,int uses_interp)214 static int __setup_additional_pages(enum vdso_abi abi,
215 struct mm_struct *mm,
216 struct linux_binprm *bprm,
217 int uses_interp)
218 {
219 unsigned long vdso_base, vdso_text_len, vdso_mapping_len;
220 unsigned long gp_flags = 0;
221 void *ret;
222
223 BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES);
224
225 vdso_text_len = vdso_info[abi].vdso_pages << PAGE_SHIFT;
226 /* Be sure to map the data page */
227 vdso_mapping_len = vdso_text_len + VVAR_NR_PAGES * PAGE_SIZE;
228
229 vdso_base = get_unmapped_area(NULL, 0, vdso_mapping_len, 0, 0);
230 if (IS_ERR_VALUE(vdso_base)) {
231 ret = ERR_PTR(vdso_base);
232 goto up_fail;
233 }
234
235 ret = _install_special_mapping(mm, vdso_base, VVAR_NR_PAGES * PAGE_SIZE,
236 VM_READ|VM_MAYREAD|VM_PFNMAP,
237 vdso_info[abi].dm);
238 if (IS_ERR(ret))
239 goto up_fail;
240
241 if (IS_ENABLED(CONFIG_ARM64_BTI_KERNEL) && system_supports_bti())
242 gp_flags = VM_ARM64_BTI;
243
244 vdso_base += VVAR_NR_PAGES * PAGE_SIZE;
245 mm->context.vdso = (void *)vdso_base;
246 ret = _install_special_mapping(mm, vdso_base, vdso_text_len,
247 VM_READ|VM_EXEC|gp_flags|
248 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
249 vdso_info[abi].cm);
250 if (IS_ERR(ret))
251 goto up_fail;
252
253 return 0;
254
255 up_fail:
256 mm->context.vdso = NULL;
257 return PTR_ERR(ret);
258 }
259
260 #ifdef CONFIG_COMPAT
261 /*
262 * Create and map the vectors page for AArch32 tasks.
263 */
264 enum aarch32_map {
265 AA32_MAP_VECTORS, /* kuser helpers */
266 AA32_MAP_SIGPAGE,
267 AA32_MAP_VVAR,
268 AA32_MAP_VDSO,
269 };
270
271 static struct page *aarch32_vectors_page __ro_after_init;
272 static struct page *aarch32_sig_page __ro_after_init;
273
aarch32_sigpage_mremap(const struct vm_special_mapping * sm,struct vm_area_struct * new_vma)274 static int aarch32_sigpage_mremap(const struct vm_special_mapping *sm,
275 struct vm_area_struct *new_vma)
276 {
277 current->mm->context.sigpage = (void *)new_vma->vm_start;
278
279 return 0;
280 }
281
282 static struct vm_special_mapping aarch32_vdso_maps[] = {
283 [AA32_MAP_VECTORS] = {
284 .name = "[vectors]", /* ABI */
285 .pages = &aarch32_vectors_page,
286 },
287 [AA32_MAP_SIGPAGE] = {
288 .name = "[sigpage]", /* ABI */
289 .pages = &aarch32_sig_page,
290 .mremap = aarch32_sigpage_mremap,
291 },
292 [AA32_MAP_VVAR] = {
293 .name = "[vvar]",
294 .fault = vvar_fault,
295 },
296 [AA32_MAP_VDSO] = {
297 .name = "[vdso]",
298 .mremap = vdso_mremap,
299 },
300 };
301
aarch32_alloc_kuser_vdso_page(void)302 static int aarch32_alloc_kuser_vdso_page(void)
303 {
304 extern char __kuser_helper_start[], __kuser_helper_end[];
305 int kuser_sz = __kuser_helper_end - __kuser_helper_start;
306 unsigned long vdso_page;
307
308 if (!IS_ENABLED(CONFIG_KUSER_HELPERS))
309 return 0;
310
311 vdso_page = get_zeroed_page(GFP_KERNEL);
312 if (!vdso_page)
313 return -ENOMEM;
314
315 memcpy((void *)(vdso_page + 0x1000 - kuser_sz), __kuser_helper_start,
316 kuser_sz);
317 aarch32_vectors_page = virt_to_page((void *)vdso_page);
318 return 0;
319 }
320
321 #define COMPAT_SIGPAGE_POISON_WORD 0xe7fddef1
aarch32_alloc_sigpage(void)322 static int aarch32_alloc_sigpage(void)
323 {
324 extern char __aarch32_sigret_code_start[], __aarch32_sigret_code_end[];
325 int sigret_sz = __aarch32_sigret_code_end - __aarch32_sigret_code_start;
326 __le32 poison = cpu_to_le32(COMPAT_SIGPAGE_POISON_WORD);
327 void *sigpage;
328
329 sigpage = (void *)__get_free_page(GFP_KERNEL);
330 if (!sigpage)
331 return -ENOMEM;
332
333 memset32(sigpage, (__force u32)poison, PAGE_SIZE / sizeof(poison));
334 memcpy(sigpage, __aarch32_sigret_code_start, sigret_sz);
335 aarch32_sig_page = virt_to_page(sigpage);
336 return 0;
337 }
338
__aarch32_alloc_vdso_pages(void)339 static int __init __aarch32_alloc_vdso_pages(void)
340 {
341
342 if (!IS_ENABLED(CONFIG_COMPAT_VDSO))
343 return 0;
344
345 vdso_info[VDSO_ABI_AA32].dm = &aarch32_vdso_maps[AA32_MAP_VVAR];
346 vdso_info[VDSO_ABI_AA32].cm = &aarch32_vdso_maps[AA32_MAP_VDSO];
347
348 return __vdso_init(VDSO_ABI_AA32);
349 }
350
aarch32_alloc_vdso_pages(void)351 static int __init aarch32_alloc_vdso_pages(void)
352 {
353 int ret;
354
355 ret = __aarch32_alloc_vdso_pages();
356 if (ret)
357 return ret;
358
359 ret = aarch32_alloc_sigpage();
360 if (ret)
361 return ret;
362
363 return aarch32_alloc_kuser_vdso_page();
364 }
365 arch_initcall(aarch32_alloc_vdso_pages);
366
aarch32_kuser_helpers_setup(struct mm_struct * mm)367 static int aarch32_kuser_helpers_setup(struct mm_struct *mm)
368 {
369 void *ret;
370
371 if (!IS_ENABLED(CONFIG_KUSER_HELPERS))
372 return 0;
373
374 /*
375 * Avoid VM_MAYWRITE for compatibility with arch/arm/, where it's
376 * not safe to CoW the page containing the CPU exception vectors.
377 */
378 ret = _install_special_mapping(mm, AARCH32_VECTORS_BASE, PAGE_SIZE,
379 VM_READ | VM_EXEC |
380 VM_MAYREAD | VM_MAYEXEC,
381 &aarch32_vdso_maps[AA32_MAP_VECTORS]);
382
383 return PTR_ERR_OR_ZERO(ret);
384 }
385
aarch32_sigreturn_setup(struct mm_struct * mm)386 static int aarch32_sigreturn_setup(struct mm_struct *mm)
387 {
388 unsigned long addr;
389 void *ret;
390
391 addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
392 if (IS_ERR_VALUE(addr)) {
393 ret = ERR_PTR(addr);
394 goto out;
395 }
396
397 /*
398 * VM_MAYWRITE is required to allow gdb to Copy-on-Write and
399 * set breakpoints.
400 */
401 ret = _install_special_mapping(mm, addr, PAGE_SIZE,
402 VM_READ | VM_EXEC | VM_MAYREAD |
403 VM_MAYWRITE | VM_MAYEXEC,
404 &aarch32_vdso_maps[AA32_MAP_SIGPAGE]);
405 if (IS_ERR(ret))
406 goto out;
407
408 mm->context.sigpage = (void *)addr;
409
410 out:
411 return PTR_ERR_OR_ZERO(ret);
412 }
413
aarch32_setup_additional_pages(struct linux_binprm * bprm,int uses_interp)414 int aarch32_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
415 {
416 struct mm_struct *mm = current->mm;
417 int ret;
418
419 if (mmap_write_lock_killable(mm))
420 return -EINTR;
421
422 ret = aarch32_kuser_helpers_setup(mm);
423 if (ret)
424 goto out;
425
426 if (IS_ENABLED(CONFIG_COMPAT_VDSO)) {
427 ret = __setup_additional_pages(VDSO_ABI_AA32, mm, bprm,
428 uses_interp);
429 if (ret)
430 goto out;
431 }
432
433 ret = aarch32_sigreturn_setup(mm);
434 out:
435 mmap_write_unlock(mm);
436 return ret;
437 }
438 #endif /* CONFIG_COMPAT */
439
440 enum aarch64_map {
441 AA64_MAP_VVAR,
442 AA64_MAP_VDSO,
443 };
444
445 static struct vm_special_mapping aarch64_vdso_maps[] __ro_after_init = {
446 [AA64_MAP_VVAR] = {
447 .name = "[vvar]",
448 .fault = vvar_fault,
449 },
450 [AA64_MAP_VDSO] = {
451 .name = "[vdso]",
452 .mremap = vdso_mremap,
453 },
454 };
455
vdso_init(void)456 static int __init vdso_init(void)
457 {
458 vdso_info[VDSO_ABI_AA64].dm = &aarch64_vdso_maps[AA64_MAP_VVAR];
459 vdso_info[VDSO_ABI_AA64].cm = &aarch64_vdso_maps[AA64_MAP_VDSO];
460
461 return __vdso_init(VDSO_ABI_AA64);
462 }
463 arch_initcall(vdso_init);
464
arch_setup_additional_pages(struct linux_binprm * bprm,int uses_interp)465 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
466 {
467 struct mm_struct *mm = current->mm;
468 int ret;
469
470 if (mmap_write_lock_killable(mm))
471 return -EINTR;
472
473 ret = __setup_additional_pages(VDSO_ABI_AA64, mm, bprm, uses_interp);
474 mmap_write_unlock(mm);
475
476 return ret;
477 }
478