1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright 2007 Andi Kleen, SUSE Labs.
4 *
5 * This contains most of the x86 vDSO kernel-side code.
6 */
7 #include <linux/mm.h>
8 #include <linux/err.h>
9 #include <linux/sched.h>
10 #include <linux/sched/task_stack.h>
11 #include <linux/slab.h>
12 #include <linux/init.h>
13 #include <linux/random.h>
14 #include <linux/elf.h>
15 #include <linux/cpu.h>
16 #include <linux/ptrace.h>
17 #include <linux/time_namespace.h>
18
19 #include <asm/pvclock.h>
20 #include <asm/vgtod.h>
21 #include <asm/proto.h>
22 #include <asm/vdso.h>
23 #include <asm/vvar.h>
24 #include <asm/tlb.h>
25 #include <asm/page.h>
26 #include <asm/desc.h>
27 #include <asm/cpufeature.h>
28 #include <clocksource/hyperv_timer.h>
29
30 #undef _ASM_X86_VVAR_H
31 #define EMIT_VVAR(name, offset) \
32 const size_t name ## _offset = offset;
33 #include <asm/vvar.h>
34
arch_get_vdso_data(void * vvar_page)35 struct vdso_data *arch_get_vdso_data(void *vvar_page)
36 {
37 return (struct vdso_data *)(vvar_page + _vdso_data_offset);
38 }
39 #undef EMIT_VVAR
40
41 unsigned int vclocks_used __read_mostly;
42
43 #if defined(CONFIG_X86_64)
44 unsigned int __read_mostly vdso64_enabled = 1;
45 #endif
46
init_vdso_image(const struct vdso_image * image)47 void __init init_vdso_image(const struct vdso_image *image)
48 {
49 BUG_ON(image->size % PAGE_SIZE != 0);
50
51 apply_alternatives((struct alt_instr *)(image->data + image->alt),
52 (struct alt_instr *)(image->data + image->alt +
53 image->alt_len));
54 }
55
56 static const struct vm_special_mapping vvar_mapping;
57 struct linux_binprm;
58
vdso_fault(const struct vm_special_mapping * sm,struct vm_area_struct * vma,struct vm_fault * vmf)59 static vm_fault_t vdso_fault(const struct vm_special_mapping *sm,
60 struct vm_area_struct *vma, struct vm_fault *vmf)
61 {
62 const struct vdso_image *image = vma->vm_mm->context.vdso_image;
63
64 if (!image || (vmf->pgoff << PAGE_SHIFT) >= image->size)
65 return VM_FAULT_SIGBUS;
66
67 vmf->page = virt_to_page(image->data + (vmf->pgoff << PAGE_SHIFT));
68 get_page(vmf->page);
69 return 0;
70 }
71
vdso_fix_landing(const struct vdso_image * image,struct vm_area_struct * new_vma)72 static void vdso_fix_landing(const struct vdso_image *image,
73 struct vm_area_struct *new_vma)
74 {
75 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
76 if (in_ia32_syscall() && image == &vdso_image_32) {
77 struct pt_regs *regs = current_pt_regs();
78 unsigned long vdso_land = image->sym_int80_landing_pad;
79 unsigned long old_land_addr = vdso_land +
80 (unsigned long)current->mm->context.vdso;
81
82 /* Fixing userspace landing - look at do_fast_syscall_32 */
83 if (regs->ip == old_land_addr)
84 regs->ip = new_vma->vm_start + vdso_land;
85 }
86 #endif
87 }
88
vdso_mremap(const struct vm_special_mapping * sm,struct vm_area_struct * new_vma)89 static int vdso_mremap(const struct vm_special_mapping *sm,
90 struct vm_area_struct *new_vma)
91 {
92 unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
93 const struct vdso_image *image = current->mm->context.vdso_image;
94
95 if (image->size != new_size)
96 return -EINVAL;
97
98 vdso_fix_landing(image, new_vma);
99 current->mm->context.vdso = (void __user *)new_vma->vm_start;
100
101 return 0;
102 }
103
vvar_mremap(const struct vm_special_mapping * sm,struct vm_area_struct * new_vma)104 static int vvar_mremap(const struct vm_special_mapping *sm,
105 struct vm_area_struct *new_vma)
106 {
107 const struct vdso_image *image = new_vma->vm_mm->context.vdso_image;
108 unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
109
110 if (new_size != -image->sym_vvar_start)
111 return -EINVAL;
112
113 return 0;
114 }
115
116 #ifdef CONFIG_TIME_NS
find_timens_vvar_page(struct vm_area_struct * vma)117 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
118 {
119 if (likely(vma->vm_mm == current->mm))
120 return current->nsproxy->time_ns->vvar_page;
121
122 /*
123 * VM_PFNMAP | VM_IO protect .fault() handler from being called
124 * through interfaces like /proc/$pid/mem or
125 * process_vm_{readv,writev}() as long as there's no .access()
126 * in special_mapping_vmops().
127 * For more details check_vma_flags() and __access_remote_vm()
128 */
129
130 WARN(1, "vvar_page accessed remotely");
131
132 return NULL;
133 }
134
135 /*
136 * The vvar page layout depends on whether a task belongs to the root or
137 * non-root time namespace. Whenever a task changes its namespace, the VVAR
138 * page tables are cleared and then they will re-faulted with a
139 * corresponding layout.
140 * See also the comment near timens_setup_vdso_data() for details.
141 */
vdso_join_timens(struct task_struct * task,struct time_namespace * ns)142 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
143 {
144 struct mm_struct *mm = task->mm;
145 struct vm_area_struct *vma;
146
147 mmap_read_lock(mm);
148
149 for (vma = mm->mmap; vma; vma = vma->vm_next) {
150 unsigned long size = vma->vm_end - vma->vm_start;
151
152 if (vma_is_special_mapping(vma, &vvar_mapping))
153 zap_page_range(vma, vma->vm_start, size);
154 }
155
156 mmap_read_unlock(mm);
157 return 0;
158 }
159 #else
find_timens_vvar_page(struct vm_area_struct * vma)160 static inline struct page *find_timens_vvar_page(struct vm_area_struct *vma)
161 {
162 return NULL;
163 }
164 #endif
165
vvar_fault(const struct vm_special_mapping * sm,struct vm_area_struct * vma,struct vm_fault * vmf)166 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
167 struct vm_area_struct *vma, struct vm_fault *vmf)
168 {
169 const struct vdso_image *image = vma->vm_mm->context.vdso_image;
170 unsigned long pfn;
171 long sym_offset;
172
173 if (!image)
174 return VM_FAULT_SIGBUS;
175
176 sym_offset = (long)(vmf->pgoff << PAGE_SHIFT) +
177 image->sym_vvar_start;
178
179 /*
180 * Sanity check: a symbol offset of zero means that the page
181 * does not exist for this vdso image, not that the page is at
182 * offset zero relative to the text mapping. This should be
183 * impossible here, because sym_offset should only be zero for
184 * the page past the end of the vvar mapping.
185 */
186 if (sym_offset == 0)
187 return VM_FAULT_SIGBUS;
188
189 if (sym_offset == image->sym_vvar_page) {
190 struct page *timens_page = find_timens_vvar_page(vma);
191
192 pfn = __pa_symbol(&__vvar_page) >> PAGE_SHIFT;
193
194 /*
195 * If a task belongs to a time namespace then a namespace
196 * specific VVAR is mapped with the sym_vvar_page offset and
197 * the real VVAR page is mapped with the sym_timens_page
198 * offset.
199 * See also the comment near timens_setup_vdso_data().
200 */
201 if (timens_page) {
202 unsigned long addr;
203 vm_fault_t err;
204
205 /*
206 * Optimization: inside time namespace pre-fault
207 * VVAR page too. As on timens page there are only
208 * offsets for clocks on VVAR, it'll be faulted
209 * shortly by VDSO code.
210 */
211 addr = vmf->address + (image->sym_timens_page - sym_offset);
212 err = vmf_insert_pfn(vma, addr, pfn);
213 if (unlikely(err & VM_FAULT_ERROR))
214 return err;
215
216 pfn = page_to_pfn(timens_page);
217 }
218
219 return vmf_insert_pfn(vma, vmf->address, pfn);
220 } else if (sym_offset == image->sym_pvclock_page) {
221 struct pvclock_vsyscall_time_info *pvti =
222 pvclock_get_pvti_cpu0_va();
223 if (pvti && vclock_was_used(VDSO_CLOCKMODE_PVCLOCK)) {
224 return vmf_insert_pfn_prot(vma, vmf->address,
225 __pa(pvti) >> PAGE_SHIFT,
226 pgprot_decrypted(vma->vm_page_prot));
227 }
228 } else if (sym_offset == image->sym_hvclock_page) {
229 struct ms_hyperv_tsc_page *tsc_pg = hv_get_tsc_page();
230
231 if (tsc_pg && vclock_was_used(VDSO_CLOCKMODE_HVCLOCK))
232 return vmf_insert_pfn(vma, vmf->address,
233 virt_to_phys(tsc_pg) >> PAGE_SHIFT);
234 } else if (sym_offset == image->sym_timens_page) {
235 struct page *timens_page = find_timens_vvar_page(vma);
236
237 if (!timens_page)
238 return VM_FAULT_SIGBUS;
239
240 pfn = __pa_symbol(&__vvar_page) >> PAGE_SHIFT;
241 return vmf_insert_pfn(vma, vmf->address, pfn);
242 }
243
244 return VM_FAULT_SIGBUS;
245 }
246
247 static const struct vm_special_mapping vdso_mapping = {
248 .name = "[vdso]",
249 .fault = vdso_fault,
250 .mremap = vdso_mremap,
251 };
252 static const struct vm_special_mapping vvar_mapping = {
253 .name = "[vvar]",
254 .fault = vvar_fault,
255 .mremap = vvar_mremap,
256 };
257
258 /*
259 * Add vdso and vvar mappings to current process.
260 * @image - blob to map
261 * @addr - request a specific address (zero to map at free addr)
262 */
map_vdso(const struct vdso_image * image,unsigned long addr)263 static int map_vdso(const struct vdso_image *image, unsigned long addr)
264 {
265 struct mm_struct *mm = current->mm;
266 struct vm_area_struct *vma;
267 unsigned long text_start;
268 int ret = 0;
269
270 if (mmap_write_lock_killable(mm))
271 return -EINTR;
272
273 addr = get_unmapped_area(NULL, addr,
274 image->size - image->sym_vvar_start, 0, 0);
275 if (IS_ERR_VALUE(addr)) {
276 ret = addr;
277 goto up_fail;
278 }
279
280 text_start = addr - image->sym_vvar_start;
281
282 /*
283 * MAYWRITE to allow gdb to COW and set breakpoints
284 */
285 vma = _install_special_mapping(mm,
286 text_start,
287 image->size,
288 VM_READ|VM_EXEC|
289 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
290 &vdso_mapping);
291
292 if (IS_ERR(vma)) {
293 ret = PTR_ERR(vma);
294 goto up_fail;
295 }
296
297 vma = _install_special_mapping(mm,
298 addr,
299 -image->sym_vvar_start,
300 VM_READ|VM_MAYREAD|VM_IO|VM_DONTDUMP|
301 VM_PFNMAP,
302 &vvar_mapping);
303
304 if (IS_ERR(vma)) {
305 ret = PTR_ERR(vma);
306 do_munmap(mm, text_start, image->size, NULL);
307 } else {
308 current->mm->context.vdso = (void __user *)text_start;
309 current->mm->context.vdso_image = image;
310 }
311
312 up_fail:
313 mmap_write_unlock(mm);
314 return ret;
315 }
316
317 #ifdef CONFIG_X86_64
318 /*
319 * Put the vdso above the (randomized) stack with another randomized
320 * offset. This way there is no hole in the middle of address space.
321 * To save memory make sure it is still in the same PTE as the stack
322 * top. This doesn't give that many random bits.
323 *
324 * Note that this algorithm is imperfect: the distribution of the vdso
325 * start address within a PMD is biased toward the end.
326 *
327 * Only used for the 64-bit and x32 vdsos.
328 */
vdso_addr(unsigned long start,unsigned len)329 static unsigned long vdso_addr(unsigned long start, unsigned len)
330 {
331 unsigned long addr, end;
332 unsigned offset;
333
334 /*
335 * Round up the start address. It can start out unaligned as a result
336 * of stack start randomization.
337 */
338 start = PAGE_ALIGN(start);
339
340 /* Round the lowest possible end address up to a PMD boundary. */
341 end = (start + len + PMD_SIZE - 1) & PMD_MASK;
342 if (end >= DEFAULT_MAP_WINDOW)
343 end = DEFAULT_MAP_WINDOW;
344 end -= len;
345
346 if (end > start) {
347 offset = get_random_int() % (((end - start) >> PAGE_SHIFT) + 1);
348 addr = start + (offset << PAGE_SHIFT);
349 } else {
350 addr = start;
351 }
352
353 /*
354 * Forcibly align the final address in case we have a hardware
355 * issue that requires alignment for performance reasons.
356 */
357 addr = align_vdso_addr(addr);
358
359 return addr;
360 }
361
map_vdso_randomized(const struct vdso_image * image)362 static int map_vdso_randomized(const struct vdso_image *image)
363 {
364 unsigned long addr = vdso_addr(current->mm->start_stack, image->size-image->sym_vvar_start);
365
366 return map_vdso(image, addr);
367 }
368 #endif
369
map_vdso_once(const struct vdso_image * image,unsigned long addr)370 int map_vdso_once(const struct vdso_image *image, unsigned long addr)
371 {
372 struct mm_struct *mm = current->mm;
373 struct vm_area_struct *vma;
374
375 mmap_write_lock(mm);
376 /*
377 * Check if we have already mapped vdso blob - fail to prevent
378 * abusing from userspace install_speciall_mapping, which may
379 * not do accounting and rlimit right.
380 * We could search vma near context.vdso, but it's a slowpath,
381 * so let's explicitly check all VMAs to be completely sure.
382 */
383 for (vma = mm->mmap; vma; vma = vma->vm_next) {
384 if (vma_is_special_mapping(vma, &vdso_mapping) ||
385 vma_is_special_mapping(vma, &vvar_mapping)) {
386 mmap_write_unlock(mm);
387 return -EEXIST;
388 }
389 }
390 mmap_write_unlock(mm);
391
392 return map_vdso(image, addr);
393 }
394
395 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
load_vdso32(void)396 static int load_vdso32(void)
397 {
398 if (vdso32_enabled != 1) /* Other values all mean "disabled" */
399 return 0;
400
401 return map_vdso(&vdso_image_32, 0);
402 }
403 #endif
404
405 #ifdef CONFIG_X86_64
arch_setup_additional_pages(struct linux_binprm * bprm,int uses_interp)406 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
407 {
408 if (!vdso64_enabled)
409 return 0;
410
411 return map_vdso_randomized(&vdso_image_64);
412 }
413
414 #ifdef CONFIG_COMPAT
compat_arch_setup_additional_pages(struct linux_binprm * bprm,int uses_interp)415 int compat_arch_setup_additional_pages(struct linux_binprm *bprm,
416 int uses_interp)
417 {
418 #ifdef CONFIG_X86_X32_ABI
419 if (test_thread_flag(TIF_X32)) {
420 if (!vdso64_enabled)
421 return 0;
422 return map_vdso_randomized(&vdso_image_x32);
423 }
424 #endif
425 #ifdef CONFIG_IA32_EMULATION
426 return load_vdso32();
427 #else
428 return 0;
429 #endif
430 }
431 #endif
432 #else
arch_setup_additional_pages(struct linux_binprm * bprm,int uses_interp)433 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
434 {
435 return load_vdso32();
436 }
437 #endif
438
439 #ifdef CONFIG_X86_64
vdso_setup(char * s)440 static __init int vdso_setup(char *s)
441 {
442 vdso64_enabled = simple_strtoul(s, NULL, 0);
443 return 1;
444 }
445 __setup("vdso=", vdso_setup);
446
init_vdso(void)447 static int __init init_vdso(void)
448 {
449 BUILD_BUG_ON(VDSO_CLOCKMODE_MAX >= 32);
450
451 init_vdso_image(&vdso_image_64);
452
453 #ifdef CONFIG_X86_X32_ABI
454 init_vdso_image(&vdso_image_x32);
455 #endif
456
457 return 0;
458 }
459 subsys_initcall(init_vdso);
460 #endif /* CONFIG_X86_64 */
461