1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * vdso setup for s390
4 *
5 * Copyright IBM Corp. 2008
6 * Author(s): Martin Schwidefsky (schwidefsky@de.ibm.com)
7 */
8
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>
15 #include <linux/mm.h>
16 #include <linux/slab.h>
17 #include <linux/smp.h>
18 #include <linux/time_namespace.h>
19 #include <linux/random.h>
20 #include <vdso/datapage.h>
21 #include <asm/vdso.h>
22
23 extern char vdso64_start[], vdso64_end[];
24 extern char vdso32_start[], vdso32_end[];
25
26 static struct vm_special_mapping vvar_mapping;
27
28 static union {
29 struct vdso_data data[CS_BASES];
30 u8 page[PAGE_SIZE];
31 } vdso_data_store __page_aligned_data;
32
33 struct vdso_data *vdso_data = vdso_data_store.data;
34
35 enum vvar_pages {
36 VVAR_DATA_PAGE_OFFSET,
37 VVAR_TIMENS_PAGE_OFFSET,
38 VVAR_NR_PAGES,
39 };
40
41 #ifdef CONFIG_TIME_NS
arch_get_vdso_data(void * vvar_page)42 struct vdso_data *arch_get_vdso_data(void *vvar_page)
43 {
44 return (struct vdso_data *)(vvar_page);
45 }
46
47 /*
48 * The VVAR page layout depends on whether a task belongs to the root or
49 * non-root time namespace. Whenever a task changes its namespace, the VVAR
50 * page tables are cleared and then they will be re-faulted with a
51 * corresponding layout.
52 * See also the comment near timens_setup_vdso_data() for details.
53 */
vdso_join_timens(struct task_struct * task,struct time_namespace * ns)54 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
55 {
56 struct mm_struct *mm = task->mm;
57 VMA_ITERATOR(vmi, mm, 0);
58 struct vm_area_struct *vma;
59
60 mmap_read_lock(mm);
61 for_each_vma(vmi, vma) {
62 if (!vma_is_special_mapping(vma, &vvar_mapping))
63 continue;
64 zap_vma_pages(vma);
65 break;
66 }
67 mmap_read_unlock(mm);
68 return 0;
69 }
70 #endif
71
vvar_fault(const struct vm_special_mapping * sm,struct vm_area_struct * vma,struct vm_fault * vmf)72 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
73 struct vm_area_struct *vma, struct vm_fault *vmf)
74 {
75 struct page *timens_page = find_timens_vvar_page(vma);
76 unsigned long addr, pfn;
77 vm_fault_t err;
78
79 switch (vmf->pgoff) {
80 case VVAR_DATA_PAGE_OFFSET:
81 pfn = virt_to_pfn(vdso_data);
82 if (timens_page) {
83 /*
84 * Fault in VVAR page too, since it will be accessed
85 * to get clock data anyway.
86 */
87 addr = vmf->address + VVAR_TIMENS_PAGE_OFFSET * PAGE_SIZE;
88 err = vmf_insert_pfn(vma, addr, pfn);
89 if (unlikely(err & VM_FAULT_ERROR))
90 return err;
91 pfn = page_to_pfn(timens_page);
92 }
93 break;
94 #ifdef CONFIG_TIME_NS
95 case VVAR_TIMENS_PAGE_OFFSET:
96 /*
97 * If a task belongs to a time namespace then a namespace
98 * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
99 * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
100 * offset.
101 * See also the comment near timens_setup_vdso_data().
102 */
103 if (!timens_page)
104 return VM_FAULT_SIGBUS;
105 pfn = virt_to_pfn(vdso_data);
106 break;
107 #endif /* CONFIG_TIME_NS */
108 default:
109 return VM_FAULT_SIGBUS;
110 }
111 return vmf_insert_pfn(vma, vmf->address, pfn);
112 }
113
vdso_mremap(const struct vm_special_mapping * sm,struct vm_area_struct * vma)114 static int vdso_mremap(const struct vm_special_mapping *sm,
115 struct vm_area_struct *vma)
116 {
117 current->mm->context.vdso_base = vma->vm_start;
118 return 0;
119 }
120
121 static struct vm_special_mapping vvar_mapping = {
122 .name = "[vvar]",
123 .fault = vvar_fault,
124 };
125
126 static struct vm_special_mapping vdso64_mapping = {
127 .name = "[vdso]",
128 .mremap = vdso_mremap,
129 };
130
131 static struct vm_special_mapping vdso32_mapping = {
132 .name = "[vdso]",
133 .mremap = vdso_mremap,
134 };
135
vdso_getcpu_init(void)136 int vdso_getcpu_init(void)
137 {
138 set_tod_programmable_field(smp_processor_id());
139 return 0;
140 }
141 early_initcall(vdso_getcpu_init); /* Must be called before SMP init */
142
map_vdso(unsigned long addr,unsigned long vdso_mapping_len)143 static int map_vdso(unsigned long addr, unsigned long vdso_mapping_len)
144 {
145 unsigned long vvar_start, vdso_text_start, vdso_text_len;
146 struct vm_special_mapping *vdso_mapping;
147 struct mm_struct *mm = current->mm;
148 struct vm_area_struct *vma;
149 int rc;
150
151 BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES);
152 if (mmap_write_lock_killable(mm))
153 return -EINTR;
154
155 if (is_compat_task()) {
156 vdso_text_len = vdso32_end - vdso32_start;
157 vdso_mapping = &vdso32_mapping;
158 } else {
159 vdso_text_len = vdso64_end - vdso64_start;
160 vdso_mapping = &vdso64_mapping;
161 }
162 vvar_start = get_unmapped_area(NULL, addr, vdso_mapping_len, 0, 0);
163 rc = vvar_start;
164 if (IS_ERR_VALUE(vvar_start))
165 goto out;
166 vma = _install_special_mapping(mm, vvar_start, VVAR_NR_PAGES*PAGE_SIZE,
167 VM_READ|VM_MAYREAD|VM_IO|VM_DONTDUMP|
168 VM_PFNMAP,
169 &vvar_mapping);
170 rc = PTR_ERR(vma);
171 if (IS_ERR(vma))
172 goto out;
173 vdso_text_start = vvar_start + VVAR_NR_PAGES * PAGE_SIZE;
174 /* VM_MAYWRITE for COW so gdb can set breakpoints */
175 vma = _install_special_mapping(mm, vdso_text_start, vdso_text_len,
176 VM_READ|VM_EXEC|
177 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
178 vdso_mapping);
179 if (IS_ERR(vma)) {
180 do_munmap(mm, vvar_start, PAGE_SIZE, NULL);
181 rc = PTR_ERR(vma);
182 } else {
183 current->mm->context.vdso_base = vdso_text_start;
184 rc = 0;
185 }
186 out:
187 mmap_write_unlock(mm);
188 return rc;
189 }
190
vdso_addr(unsigned long start,unsigned long len)191 static unsigned long vdso_addr(unsigned long start, unsigned long len)
192 {
193 unsigned long addr, end, offset;
194
195 /*
196 * Round up the start address. It can start out unaligned as a result
197 * of stack start randomization.
198 */
199 start = PAGE_ALIGN(start);
200
201 /* Round the lowest possible end address up to a PMD boundary. */
202 end = (start + len + PMD_SIZE - 1) & PMD_MASK;
203 if (end >= VDSO_BASE)
204 end = VDSO_BASE;
205 end -= len;
206
207 if (end > start) {
208 offset = get_random_u32_below(((end - start) >> PAGE_SHIFT) + 1);
209 addr = start + (offset << PAGE_SHIFT);
210 } else {
211 addr = start;
212 }
213 return addr;
214 }
215
vdso_size(void)216 unsigned long vdso_size(void)
217 {
218 unsigned long size = VVAR_NR_PAGES * PAGE_SIZE;
219
220 if (is_compat_task())
221 size += vdso32_end - vdso32_start;
222 else
223 size += vdso64_end - vdso64_start;
224 return PAGE_ALIGN(size);
225 }
226
arch_setup_additional_pages(struct linux_binprm * bprm,int uses_interp)227 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
228 {
229 unsigned long addr = VDSO_BASE;
230 unsigned long size = vdso_size();
231
232 if (current->flags & PF_RANDOMIZE)
233 addr = vdso_addr(current->mm->start_stack + PAGE_SIZE, size);
234 return map_vdso(addr, size);
235 }
236
vdso_setup_pages(void * start,void * end)237 static struct page ** __init vdso_setup_pages(void *start, void *end)
238 {
239 int pages = (end - start) >> PAGE_SHIFT;
240 struct page **pagelist;
241 int i;
242
243 pagelist = kcalloc(pages + 1, sizeof(struct page *), GFP_KERNEL);
244 if (!pagelist)
245 panic("%s: Cannot allocate page list for VDSO", __func__);
246 for (i = 0; i < pages; i++)
247 pagelist[i] = virt_to_page(start + i * PAGE_SIZE);
248 return pagelist;
249 }
250
vdso_init(void)251 static int __init vdso_init(void)
252 {
253 vdso64_mapping.pages = vdso_setup_pages(vdso64_start, vdso64_end);
254 if (IS_ENABLED(CONFIG_COMPAT))
255 vdso32_mapping.pages = vdso_setup_pages(vdso32_start, vdso32_end);
256 return 0;
257 }
258 arch_initcall(vdso_init);
259