1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright IBM Corp. 2008
6 * Author(s): Martin Schwidefsky (schwidefsky@de.ibm.com)
9 #include <linux/binfmts.h>
10 #include <linux/compat.h>
11 #include <linux/elf.h>
12 #include <linux/errno.h>
13 #include <linux/init.h>
14 #include <linux/kernel.h>
16 #include <linux/slab.h>
17 #include <linux/smp.h>
18 #include <linux/time_namespace.h>
19 #include <vdso/datapage.h>
22 extern char vdso64_start[], vdso64_end[];
23 extern char vdso32_start[], vdso32_end[];
25 static struct vm_special_mapping vvar_mapping;
28 struct vdso_data data[CS_BASES];
30 } vdso_data_store __page_aligned_data;
32 struct vdso_data *vdso_data = vdso_data_store.data;
35 VVAR_DATA_PAGE_OFFSET,
36 VVAR_TIMENS_PAGE_OFFSET,
41 struct vdso_data *arch_get_vdso_data(void *vvar_page)
43 return (struct vdso_data *)(vvar_page);
46 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
48 if (likely(vma->vm_mm == current->mm))
49 return current->nsproxy->time_ns->vvar_page;
51 * VM_PFNMAP | VM_IO protect .fault() handler from being called
52 * through interfaces like /proc/$pid/mem or
53 * process_vm_{readv,writev}() as long as there's no .access()
54 * in special_mapping_vmops().
55 * For more details check_vma_flags() and __access_remote_vm()
57 WARN(1, "vvar_page accessed remotely");
62 * The VVAR page layout depends on whether a task belongs to the root or
63 * non-root time namespace. Whenever a task changes its namespace, the VVAR
64 * page tables are cleared and then they will be re-faulted with a
65 * corresponding layout.
66 * See also the comment near timens_setup_vdso_data() for details.
68 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
70 struct mm_struct *mm = task->mm;
71 struct vm_area_struct *vma;
74 for (vma = mm->mmap; vma; vma = vma->vm_next) {
75 unsigned long size = vma->vm_end - vma->vm_start;
77 if (!vma_is_special_mapping(vma, &vvar_mapping))
79 zap_page_range(vma, vma->vm_start, size);
86 static inline struct page *find_timens_vvar_page(struct vm_area_struct *vma)
92 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
93 struct vm_area_struct *vma, struct vm_fault *vmf)
95 struct page *timens_page = find_timens_vvar_page(vma);
96 unsigned long addr, pfn;
100 case VVAR_DATA_PAGE_OFFSET:
101 pfn = virt_to_pfn(vdso_data);
104 * Fault in VVAR page too, since it will be accessed
105 * to get clock data anyway.
107 addr = vmf->address + VVAR_TIMENS_PAGE_OFFSET * PAGE_SIZE;
108 err = vmf_insert_pfn(vma, addr, pfn);
109 if (unlikely(err & VM_FAULT_ERROR))
111 pfn = page_to_pfn(timens_page);
114 #ifdef CONFIG_TIME_NS
115 case VVAR_TIMENS_PAGE_OFFSET:
117 * If a task belongs to a time namespace then a namespace
118 * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
119 * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
121 * See also the comment near timens_setup_vdso_data().
124 return VM_FAULT_SIGBUS;
125 pfn = virt_to_pfn(vdso_data);
127 #endif /* CONFIG_TIME_NS */
129 return VM_FAULT_SIGBUS;
131 return vmf_insert_pfn(vma, vmf->address, pfn);
134 static int vdso_mremap(const struct vm_special_mapping *sm,
135 struct vm_area_struct *vma)
137 current->mm->context.vdso_base = vma->vm_start;
141 static struct vm_special_mapping vvar_mapping = {
146 static struct vm_special_mapping vdso64_mapping = {
148 .mremap = vdso_mremap,
151 static struct vm_special_mapping vdso32_mapping = {
153 .mremap = vdso_mremap,
156 int vdso_getcpu_init(void)
158 set_tod_programmable_field(smp_processor_id());
161 early_initcall(vdso_getcpu_init); /* Must be called before SMP init */
163 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
165 unsigned long vdso_text_len, vdso_mapping_len;
166 unsigned long vvar_start, vdso_text_start;
167 struct vm_special_mapping *vdso_mapping;
168 struct mm_struct *mm = current->mm;
169 struct vm_area_struct *vma;
172 BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES);
173 if (mmap_write_lock_killable(mm))
176 if (is_compat_task()) {
177 vdso_text_len = vdso32_end - vdso32_start;
178 vdso_mapping = &vdso32_mapping;
180 vdso_text_len = vdso64_end - vdso64_start;
181 vdso_mapping = &vdso64_mapping;
183 vdso_mapping_len = vdso_text_len + VVAR_NR_PAGES * PAGE_SIZE;
184 vvar_start = get_unmapped_area(NULL, 0, vdso_mapping_len, 0, 0);
186 if (IS_ERR_VALUE(vvar_start))
188 vma = _install_special_mapping(mm, vvar_start, VVAR_NR_PAGES*PAGE_SIZE,
189 VM_READ|VM_MAYREAD|VM_IO|VM_DONTDUMP|
195 vdso_text_start = vvar_start + VVAR_NR_PAGES * PAGE_SIZE;
196 /* VM_MAYWRITE for COW so gdb can set breakpoints */
197 vma = _install_special_mapping(mm, vdso_text_start, vdso_text_len,
199 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
202 do_munmap(mm, vvar_start, PAGE_SIZE, NULL);
205 current->mm->context.vdso_base = vdso_text_start;
209 mmap_write_unlock(mm);
213 static struct page ** __init vdso_setup_pages(void *start, void *end)
215 int pages = (end - start) >> PAGE_SHIFT;
216 struct page **pagelist;
219 pagelist = kcalloc(pages + 1, sizeof(struct page *), GFP_KERNEL);
221 panic("%s: Cannot allocate page list for VDSO", __func__);
222 for (i = 0; i < pages; i++)
223 pagelist[i] = virt_to_page(start + i * PAGE_SIZE);
227 static int __init vdso_init(void)
229 vdso64_mapping.pages = vdso_setup_pages(vdso64_start, vdso64_end);
230 if (IS_ENABLED(CONFIG_COMPAT))
231 vdso32_mapping.pages = vdso_setup_pages(vdso32_start, vdso32_end);
234 arch_initcall(vdso_init);