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/module.h>
16 #include <linux/mm.h>
17 #include <linux/slab.h>
18 #include <linux/smp.h>
19 #include <linux/time_namespace.h>
20 #include <linux/random.h>
21 #include <vdso/datapage.h>
22 #include <asm/vdso/vsyscall.h>
23 #include <asm/alternative.h>
24 #include <asm/vdso.h>
25
26 extern char vdso64_start[], vdso64_end[];
27 extern char vdso32_start[], vdso32_end[];
28
29 static struct vm_special_mapping vvar_mapping;
30
31 static union vdso_data_store vdso_data_store __page_aligned_data;
32
33 struct vdso_data *vdso_data = vdso_data_store.data;
34
35 #ifdef CONFIG_TIME_NS
arch_get_vdso_data(void * vvar_page)36 struct vdso_data *arch_get_vdso_data(void *vvar_page)
37 {
38 return (struct vdso_data *)(vvar_page);
39 }
40
41 /*
42 * The VVAR page layout depends on whether a task belongs to the root or
43 * non-root time namespace. Whenever a task changes its namespace, the VVAR
44 * page tables are cleared and then they will be re-faulted with a
45 * corresponding layout.
46 * See also the comment near timens_setup_vdso_data() for details.
47 */
vdso_join_timens(struct task_struct * task,struct time_namespace * ns)48 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
49 {
50 struct mm_struct *mm = task->mm;
51 VMA_ITERATOR(vmi, mm, 0);
52 struct vm_area_struct *vma;
53
54 mmap_read_lock(mm);
55 for_each_vma(vmi, vma) {
56 if (!vma_is_special_mapping(vma, &vvar_mapping))
57 continue;
58 zap_vma_pages(vma);
59 break;
60 }
61 mmap_read_unlock(mm);
62 return 0;
63 }
64 #endif
65
vvar_fault(const struct vm_special_mapping * sm,struct vm_area_struct * vma,struct vm_fault * vmf)66 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
67 struct vm_area_struct *vma, struct vm_fault *vmf)
68 {
69 struct page *timens_page = find_timens_vvar_page(vma);
70 unsigned long addr, pfn;
71 vm_fault_t err;
72
73 switch (vmf->pgoff) {
74 case VVAR_DATA_PAGE_OFFSET:
75 pfn = virt_to_pfn(vdso_data);
76 if (timens_page) {
77 /*
78 * Fault in VVAR page too, since it will be accessed
79 * to get clock data anyway.
80 */
81 addr = vmf->address + VVAR_TIMENS_PAGE_OFFSET * PAGE_SIZE;
82 err = vmf_insert_pfn(vma, addr, pfn);
83 if (unlikely(err & VM_FAULT_ERROR))
84 return err;
85 pfn = page_to_pfn(timens_page);
86 }
87 break;
88 #ifdef CONFIG_TIME_NS
89 case VVAR_TIMENS_PAGE_OFFSET:
90 /*
91 * If a task belongs to a time namespace then a namespace
92 * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
93 * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
94 * offset.
95 * See also the comment near timens_setup_vdso_data().
96 */
97 if (!timens_page)
98 return VM_FAULT_SIGBUS;
99 pfn = virt_to_pfn(vdso_data);
100 break;
101 #endif /* CONFIG_TIME_NS */
102 default:
103 return VM_FAULT_SIGBUS;
104 }
105 return vmf_insert_pfn(vma, vmf->address, pfn);
106 }
107
vdso_mremap(const struct vm_special_mapping * sm,struct vm_area_struct * vma)108 static int vdso_mremap(const struct vm_special_mapping *sm,
109 struct vm_area_struct *vma)
110 {
111 current->mm->context.vdso_base = vma->vm_start;
112 return 0;
113 }
114
115 static struct vm_special_mapping vvar_mapping = {
116 .name = "[vvar]",
117 .fault = vvar_fault,
118 };
119
120 static struct vm_special_mapping vdso64_mapping = {
121 .name = "[vdso]",
122 .mremap = vdso_mremap,
123 };
124
125 static struct vm_special_mapping vdso32_mapping = {
126 .name = "[vdso]",
127 .mremap = vdso_mremap,
128 };
129
vdso_getcpu_init(void)130 int vdso_getcpu_init(void)
131 {
132 set_tod_programmable_field(smp_processor_id());
133 return 0;
134 }
135 early_initcall(vdso_getcpu_init); /* Must be called before SMP init */
136
map_vdso(unsigned long addr,unsigned long vdso_mapping_len)137 static int map_vdso(unsigned long addr, unsigned long vdso_mapping_len)
138 {
139 unsigned long vvar_start, vdso_text_start, vdso_text_len;
140 struct vm_special_mapping *vdso_mapping;
141 struct mm_struct *mm = current->mm;
142 struct vm_area_struct *vma;
143 int rc;
144
145 BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES);
146 if (mmap_write_lock_killable(mm))
147 return -EINTR;
148
149 if (is_compat_task()) {
150 vdso_text_len = vdso32_end - vdso32_start;
151 vdso_mapping = &vdso32_mapping;
152 } else {
153 vdso_text_len = vdso64_end - vdso64_start;
154 vdso_mapping = &vdso64_mapping;
155 }
156 vvar_start = get_unmapped_area(NULL, addr, vdso_mapping_len, 0, 0);
157 rc = vvar_start;
158 if (IS_ERR_VALUE(vvar_start))
159 goto out;
160 vma = _install_special_mapping(mm, vvar_start, VVAR_NR_PAGES*PAGE_SIZE,
161 VM_READ|VM_MAYREAD|VM_IO|VM_DONTDUMP|
162 VM_PFNMAP,
163 &vvar_mapping);
164 rc = PTR_ERR(vma);
165 if (IS_ERR(vma))
166 goto out;
167 vdso_text_start = vvar_start + VVAR_NR_PAGES * PAGE_SIZE;
168 /* VM_MAYWRITE for COW so gdb can set breakpoints */
169 vma = _install_special_mapping(mm, vdso_text_start, vdso_text_len,
170 VM_READ|VM_EXEC|
171 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
172 vdso_mapping);
173 if (IS_ERR(vma)) {
174 do_munmap(mm, vvar_start, PAGE_SIZE, NULL);
175 rc = PTR_ERR(vma);
176 } else {
177 current->mm->context.vdso_base = vdso_text_start;
178 rc = 0;
179 }
180 out:
181 mmap_write_unlock(mm);
182 return rc;
183 }
184
vdso_addr(unsigned long start,unsigned long len)185 static unsigned long vdso_addr(unsigned long start, unsigned long len)
186 {
187 unsigned long addr, end, offset;
188
189 /*
190 * Round up the start address. It can start out unaligned as a result
191 * of stack start randomization.
192 */
193 start = PAGE_ALIGN(start);
194
195 /* Round the lowest possible end address up to a PMD boundary. */
196 end = (start + len + PMD_SIZE - 1) & PMD_MASK;
197 if (end >= VDSO_BASE)
198 end = VDSO_BASE;
199 end -= len;
200
201 if (end > start) {
202 offset = get_random_u32_below(((end - start) >> PAGE_SHIFT) + 1);
203 addr = start + (offset << PAGE_SHIFT);
204 } else {
205 addr = start;
206 }
207 return addr;
208 }
209
vdso_text_size(void)210 unsigned long vdso_text_size(void)
211 {
212 unsigned long size;
213
214 if (is_compat_task())
215 size = vdso32_end - vdso32_start;
216 else
217 size = vdso64_end - vdso64_start;
218 return PAGE_ALIGN(size);
219 }
220
vdso_size(void)221 unsigned long vdso_size(void)
222 {
223 return vdso_text_size() + VVAR_NR_PAGES * PAGE_SIZE;
224 }
225
arch_setup_additional_pages(struct linux_binprm * bprm,int uses_interp)226 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
227 {
228 unsigned long addr = VDSO_BASE;
229 unsigned long size = vdso_size();
230
231 if (current->flags & PF_RANDOMIZE)
232 addr = vdso_addr(current->mm->start_stack + PAGE_SIZE, size);
233 return map_vdso(addr, size);
234 }
235
vdso_setup_pages(void * start,void * end)236 static struct page ** __init vdso_setup_pages(void *start, void *end)
237 {
238 int pages = (end - start) >> PAGE_SHIFT;
239 struct page **pagelist;
240 int i;
241
242 pagelist = kcalloc(pages + 1, sizeof(struct page *), GFP_KERNEL);
243 if (!pagelist)
244 panic("%s: Cannot allocate page list for VDSO", __func__);
245 for (i = 0; i < pages; i++)
246 pagelist[i] = virt_to_page(start + i * PAGE_SIZE);
247 return pagelist;
248 }
249
vdso_apply_alternatives(void)250 static void vdso_apply_alternatives(void)
251 {
252 const struct elf64_shdr *alt, *shdr;
253 struct alt_instr *start, *end;
254 const struct elf64_hdr *hdr;
255
256 hdr = (struct elf64_hdr *)vdso64_start;
257 shdr = (void *)hdr + hdr->e_shoff;
258 alt = find_section(hdr, shdr, ".altinstructions");
259 if (!alt)
260 return;
261 start = (void *)hdr + alt->sh_offset;
262 end = (void *)hdr + alt->sh_offset + alt->sh_size;
263 apply_alternatives(start, end);
264 }
265
vdso_init(void)266 static int __init vdso_init(void)
267 {
268 vdso_apply_alternatives();
269 vdso64_mapping.pages = vdso_setup_pages(vdso64_start, vdso64_end);
270 if (IS_ENABLED(CONFIG_COMPAT))
271 vdso32_mapping.pages = vdso_setup_pages(vdso32_start, vdso32_end);
272 return 0;
273 }
274 arch_initcall(vdso_init);
275