1 // SPDX-License-Identifier: GPL-2.0-only
3 * VDSO implementations.
5 * Copyright (C) 2012 ARM Limited
7 * Author: Will Deacon <will.deacon@arm.com>
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>
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>
28 #include <asm/cacheflush.h>
29 #include <asm/signal32.h>
32 extern char vdso_start[], vdso_end[];
33 extern char vdso32_start[], vdso32_end[];
41 VVAR_DATA_PAGE_OFFSET,
42 VVAR_TIMENS_PAGE_OFFSET,
46 struct vdso_abi_info {
48 const char *vdso_code_start;
49 const char *vdso_code_end;
50 unsigned long vdso_pages;
52 struct vm_special_mapping *dm;
54 struct vm_special_mapping *cm;
57 static struct vdso_abi_info vdso_info[] __ro_after_init = {
60 .vdso_code_start = vdso_start,
61 .vdso_code_end = vdso_end,
63 #ifdef CONFIG_COMPAT_VDSO
66 .vdso_code_start = vdso32_start,
67 .vdso_code_end = vdso32_end,
69 #endif /* CONFIG_COMPAT_VDSO */
76 struct vdso_data data[CS_BASES];
78 } vdso_data_store __page_aligned_data;
79 struct vdso_data *vdso_data = vdso_data_store.data;
81 static int vdso_mremap(const struct vm_special_mapping *sm,
82 struct vm_area_struct *new_vma)
84 current->mm->context.vdso = (void *)new_vma->vm_start;
89 static int __vdso_init(enum vdso_abi abi)
92 struct page **vdso_pagelist;
95 if (memcmp(vdso_info[abi].vdso_code_start, "\177ELF", 4)) {
96 pr_err("vDSO is not a valid ELF object!\n");
100 vdso_info[abi].vdso_pages = (
101 vdso_info[abi].vdso_code_end -
102 vdso_info[abi].vdso_code_start) >>
105 vdso_pagelist = kcalloc(vdso_info[abi].vdso_pages,
106 sizeof(struct page *),
108 if (vdso_pagelist == NULL)
111 /* Grab the vDSO code pages. */
112 pfn = sym_to_pfn(vdso_info[abi].vdso_code_start);
114 for (i = 0; i < vdso_info[abi].vdso_pages; i++)
115 vdso_pagelist[i] = pfn_to_page(pfn + i);
117 vdso_info[abi].cm->pages = vdso_pagelist;
122 #ifdef CONFIG_TIME_NS
123 struct vdso_data *arch_get_vdso_data(void *vvar_page)
125 return (struct vdso_data *)(vvar_page);
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.
133 * For more details see timens_setup_vdso_data().
135 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
137 struct mm_struct *mm = task->mm;
138 struct vm_area_struct *vma;
142 for (vma = mm->mmap; vma; vma = vma->vm_next) {
143 unsigned long size = vma->vm_end - vma->vm_start;
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);
153 mmap_read_unlock(mm);
157 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
159 if (likely(vma->vm_mm == current->mm))
160 return current->nsproxy->time_ns->vvar_page;
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()
169 WARN(1, "vvar_page accessed remotely");
174 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
180 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
181 struct vm_area_struct *vma, struct vm_fault *vmf)
183 struct page *timens_page = find_timens_vvar_page(vma);
186 switch (vmf->pgoff) {
187 case VVAR_DATA_PAGE_OFFSET:
189 pfn = page_to_pfn(timens_page);
191 pfn = sym_to_pfn(vdso_data);
193 #ifdef CONFIG_TIME_NS
194 case VVAR_TIMENS_PAGE_OFFSET:
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
200 * See also the comment near timens_setup_vdso_data().
203 return VM_FAULT_SIGBUS;
204 pfn = sym_to_pfn(vdso_data);
206 #endif /* CONFIG_TIME_NS */
208 return VM_FAULT_SIGBUS;
211 return vmf_insert_pfn(vma, vmf->address, pfn);
214 static int __setup_additional_pages(enum vdso_abi abi,
215 struct mm_struct *mm,
216 struct linux_binprm *bprm,
219 unsigned long vdso_base, vdso_text_len, vdso_mapping_len;
220 unsigned long gp_flags = 0;
223 BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES);
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;
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);
235 ret = _install_special_mapping(mm, vdso_base, VVAR_NR_PAGES * PAGE_SIZE,
236 VM_READ|VM_MAYREAD|VM_PFNMAP,
241 if (IS_ENABLED(CONFIG_ARM64_BTI_KERNEL) && system_supports_bti())
242 gp_flags = VM_ARM64_BTI;
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,
256 mm->context.vdso = NULL;
262 * Create and map the vectors page for AArch32 tasks.
265 AA32_MAP_VECTORS, /* kuser helpers */
271 static struct page *aarch32_vectors_page __ro_after_init;
272 static struct page *aarch32_sig_page __ro_after_init;
274 static struct vm_special_mapping aarch32_vdso_maps[] = {
275 [AA32_MAP_VECTORS] = {
276 .name = "[vectors]", /* ABI */
277 .pages = &aarch32_vectors_page,
279 [AA32_MAP_SIGPAGE] = {
280 .name = "[sigpage]", /* ABI */
281 .pages = &aarch32_sig_page,
289 .mremap = vdso_mremap,
293 static int aarch32_alloc_kuser_vdso_page(void)
295 extern char __kuser_helper_start[], __kuser_helper_end[];
296 int kuser_sz = __kuser_helper_end - __kuser_helper_start;
297 unsigned long vdso_page;
299 if (!IS_ENABLED(CONFIG_KUSER_HELPERS))
302 vdso_page = get_zeroed_page(GFP_ATOMIC);
306 memcpy((void *)(vdso_page + 0x1000 - kuser_sz), __kuser_helper_start,
308 aarch32_vectors_page = virt_to_page(vdso_page);
309 flush_dcache_page(aarch32_vectors_page);
313 static int aarch32_alloc_sigpage(void)
315 extern char __aarch32_sigret_code_start[], __aarch32_sigret_code_end[];
316 int sigret_sz = __aarch32_sigret_code_end - __aarch32_sigret_code_start;
317 unsigned long sigpage;
319 sigpage = get_zeroed_page(GFP_ATOMIC);
323 memcpy((void *)sigpage, __aarch32_sigret_code_start, sigret_sz);
324 aarch32_sig_page = virt_to_page(sigpage);
325 flush_dcache_page(aarch32_sig_page);
329 static int __aarch32_alloc_vdso_pages(void)
332 if (!IS_ENABLED(CONFIG_COMPAT_VDSO))
335 vdso_info[VDSO_ABI_AA32].dm = &aarch32_vdso_maps[AA32_MAP_VVAR];
336 vdso_info[VDSO_ABI_AA32].cm = &aarch32_vdso_maps[AA32_MAP_VDSO];
338 return __vdso_init(VDSO_ABI_AA32);
341 static int __init aarch32_alloc_vdso_pages(void)
345 ret = __aarch32_alloc_vdso_pages();
349 ret = aarch32_alloc_sigpage();
353 return aarch32_alloc_kuser_vdso_page();
355 arch_initcall(aarch32_alloc_vdso_pages);
357 static int aarch32_kuser_helpers_setup(struct mm_struct *mm)
361 if (!IS_ENABLED(CONFIG_KUSER_HELPERS))
365 * Avoid VM_MAYWRITE for compatibility with arch/arm/, where it's
366 * not safe to CoW the page containing the CPU exception vectors.
368 ret = _install_special_mapping(mm, AARCH32_VECTORS_BASE, PAGE_SIZE,
370 VM_MAYREAD | VM_MAYEXEC,
371 &aarch32_vdso_maps[AA32_MAP_VECTORS]);
373 return PTR_ERR_OR_ZERO(ret);
376 static int aarch32_sigreturn_setup(struct mm_struct *mm)
381 addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
382 if (IS_ERR_VALUE(addr)) {
388 * VM_MAYWRITE is required to allow gdb to Copy-on-Write and
391 ret = _install_special_mapping(mm, addr, PAGE_SIZE,
392 VM_READ | VM_EXEC | VM_MAYREAD |
393 VM_MAYWRITE | VM_MAYEXEC,
394 &aarch32_vdso_maps[AA32_MAP_SIGPAGE]);
398 mm->context.sigpage = (void *)addr;
401 return PTR_ERR_OR_ZERO(ret);
404 int aarch32_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
406 struct mm_struct *mm = current->mm;
409 if (mmap_write_lock_killable(mm))
412 ret = aarch32_kuser_helpers_setup(mm);
416 if (IS_ENABLED(CONFIG_COMPAT_VDSO)) {
417 ret = __setup_additional_pages(VDSO_ABI_AA32, mm, bprm,
423 ret = aarch32_sigreturn_setup(mm);
425 mmap_write_unlock(mm);
428 #endif /* CONFIG_COMPAT */
435 static struct vm_special_mapping aarch64_vdso_maps[] __ro_after_init = {
442 .mremap = vdso_mremap,
446 static int __init vdso_init(void)
448 vdso_info[VDSO_ABI_AA64].dm = &aarch64_vdso_maps[AA64_MAP_VVAR];
449 vdso_info[VDSO_ABI_AA64].cm = &aarch64_vdso_maps[AA64_MAP_VDSO];
451 return __vdso_init(VDSO_ABI_AA64);
453 arch_initcall(vdso_init);
455 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
457 struct mm_struct *mm = current->mm;
460 if (mmap_write_lock_killable(mm))
463 ret = __setup_additional_pages(VDSO_ABI_AA64, mm, bprm, uses_interp);
464 mmap_write_unlock(mm);