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