Loading...
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
23extern char vdso64_start[], vdso64_end[];
24extern char vdso32_start[], vdso32_end[];
25
26static struct vm_special_mapping vvar_mapping;
27
28static union {
29 struct vdso_data data[CS_BASES];
30 u8 page[PAGE_SIZE];
31} vdso_data_store __page_aligned_data;
32
33struct vdso_data *vdso_data = vdso_data_store.data;
34
35enum vvar_pages {
36 VVAR_DATA_PAGE_OFFSET,
37 VVAR_TIMENS_PAGE_OFFSET,
38 VVAR_NR_PAGES,
39};
40
41#ifdef CONFIG_TIME_NS
42struct 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 */
54int 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 unsigned long size = vma->vm_end - vma->vm_start;
63
64 if (!vma_is_special_mapping(vma, &vvar_mapping))
65 continue;
66 zap_page_range(vma, vma->vm_start, size);
67 break;
68 }
69 mmap_read_unlock(mm);
70 return 0;
71}
72#endif
73
74static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
75 struct vm_area_struct *vma, struct vm_fault *vmf)
76{
77 struct page *timens_page = find_timens_vvar_page(vma);
78 unsigned long addr, pfn;
79 vm_fault_t err;
80
81 switch (vmf->pgoff) {
82 case VVAR_DATA_PAGE_OFFSET:
83 pfn = virt_to_pfn(vdso_data);
84 if (timens_page) {
85 /*
86 * Fault in VVAR page too, since it will be accessed
87 * to get clock data anyway.
88 */
89 addr = vmf->address + VVAR_TIMENS_PAGE_OFFSET * PAGE_SIZE;
90 err = vmf_insert_pfn(vma, addr, pfn);
91 if (unlikely(err & VM_FAULT_ERROR))
92 return err;
93 pfn = page_to_pfn(timens_page);
94 }
95 break;
96#ifdef CONFIG_TIME_NS
97 case VVAR_TIMENS_PAGE_OFFSET:
98 /*
99 * If a task belongs to a time namespace then a namespace
100 * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
101 * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
102 * offset.
103 * See also the comment near timens_setup_vdso_data().
104 */
105 if (!timens_page)
106 return VM_FAULT_SIGBUS;
107 pfn = virt_to_pfn(vdso_data);
108 break;
109#endif /* CONFIG_TIME_NS */
110 default:
111 return VM_FAULT_SIGBUS;
112 }
113 return vmf_insert_pfn(vma, vmf->address, pfn);
114}
115
116static int vdso_mremap(const struct vm_special_mapping *sm,
117 struct vm_area_struct *vma)
118{
119 current->mm->context.vdso_base = vma->vm_start;
120 return 0;
121}
122
123static struct vm_special_mapping vvar_mapping = {
124 .name = "[vvar]",
125 .fault = vvar_fault,
126};
127
128static struct vm_special_mapping vdso64_mapping = {
129 .name = "[vdso]",
130 .mremap = vdso_mremap,
131};
132
133static struct vm_special_mapping vdso32_mapping = {
134 .name = "[vdso]",
135 .mremap = vdso_mremap,
136};
137
138int vdso_getcpu_init(void)
139{
140 set_tod_programmable_field(smp_processor_id());
141 return 0;
142}
143early_initcall(vdso_getcpu_init); /* Must be called before SMP init */
144
145static int map_vdso(unsigned long addr, unsigned long vdso_mapping_len)
146{
147 unsigned long vvar_start, vdso_text_start, vdso_text_len;
148 struct vm_special_mapping *vdso_mapping;
149 struct mm_struct *mm = current->mm;
150 struct vm_area_struct *vma;
151 int rc;
152
153 BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES);
154 if (mmap_write_lock_killable(mm))
155 return -EINTR;
156
157 if (is_compat_task()) {
158 vdso_text_len = vdso32_end - vdso32_start;
159 vdso_mapping = &vdso32_mapping;
160 } else {
161 vdso_text_len = vdso64_end - vdso64_start;
162 vdso_mapping = &vdso64_mapping;
163 }
164 vvar_start = get_unmapped_area(NULL, addr, vdso_mapping_len, 0, 0);
165 rc = vvar_start;
166 if (IS_ERR_VALUE(vvar_start))
167 goto out;
168 vma = _install_special_mapping(mm, vvar_start, VVAR_NR_PAGES*PAGE_SIZE,
169 VM_READ|VM_MAYREAD|VM_IO|VM_DONTDUMP|
170 VM_PFNMAP,
171 &vvar_mapping);
172 rc = PTR_ERR(vma);
173 if (IS_ERR(vma))
174 goto out;
175 vdso_text_start = vvar_start + VVAR_NR_PAGES * PAGE_SIZE;
176 /* VM_MAYWRITE for COW so gdb can set breakpoints */
177 vma = _install_special_mapping(mm, vdso_text_start, vdso_text_len,
178 VM_READ|VM_EXEC|
179 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
180 vdso_mapping);
181 if (IS_ERR(vma)) {
182 do_munmap(mm, vvar_start, PAGE_SIZE, NULL);
183 rc = PTR_ERR(vma);
184 } else {
185 current->mm->context.vdso_base = vdso_text_start;
186 rc = 0;
187 }
188out:
189 mmap_write_unlock(mm);
190 return rc;
191}
192
193static unsigned long vdso_addr(unsigned long start, unsigned long len)
194{
195 unsigned long addr, end, offset;
196
197 /*
198 * Round up the start address. It can start out unaligned as a result
199 * of stack start randomization.
200 */
201 start = PAGE_ALIGN(start);
202
203 /* Round the lowest possible end address up to a PMD boundary. */
204 end = (start + len + PMD_SIZE - 1) & PMD_MASK;
205 if (end >= VDSO_BASE)
206 end = VDSO_BASE;
207 end -= len;
208
209 if (end > start) {
210 offset = get_random_u32_below(((end - start) >> PAGE_SHIFT) + 1);
211 addr = start + (offset << PAGE_SHIFT);
212 } else {
213 addr = start;
214 }
215 return addr;
216}
217
218unsigned long vdso_size(void)
219{
220 unsigned long size = VVAR_NR_PAGES * PAGE_SIZE;
221
222 if (is_compat_task())
223 size += vdso32_end - vdso32_start;
224 else
225 size += vdso64_end - vdso64_start;
226 return PAGE_ALIGN(size);
227}
228
229int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
230{
231 unsigned long addr = VDSO_BASE;
232 unsigned long size = vdso_size();
233
234 if (current->flags & PF_RANDOMIZE)
235 addr = vdso_addr(current->mm->start_stack + PAGE_SIZE, size);
236 return map_vdso(addr, size);
237}
238
239static struct page ** __init vdso_setup_pages(void *start, void *end)
240{
241 int pages = (end - start) >> PAGE_SHIFT;
242 struct page **pagelist;
243 int i;
244
245 pagelist = kcalloc(pages + 1, sizeof(struct page *), GFP_KERNEL);
246 if (!pagelist)
247 panic("%s: Cannot allocate page list for VDSO", __func__);
248 for (i = 0; i < pages; i++)
249 pagelist[i] = virt_to_page(start + i * PAGE_SIZE);
250 return pagelist;
251}
252
253static int __init vdso_init(void)
254{
255 vdso64_mapping.pages = vdso_setup_pages(vdso64_start, vdso64_end);
256 if (IS_ENABLED(CONFIG_COMPAT))
257 vdso32_mapping.pages = vdso_setup_pages(vdso32_start, vdso32_end);
258 return 0;
259}
260arch_initcall(vdso_init);
1/*
2 * vdso setup for s390
3 *
4 * Copyright IBM Corp. 2008
5 * Author(s): Martin Schwidefsky (schwidefsky@de.ibm.com)
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License (version 2 only)
9 * as published by the Free Software Foundation.
10 */
11
12#include <linux/module.h>
13#include <linux/errno.h>
14#include <linux/sched.h>
15#include <linux/kernel.h>
16#include <linux/mm.h>
17#include <linux/smp.h>
18#include <linux/stddef.h>
19#include <linux/unistd.h>
20#include <linux/slab.h>
21#include <linux/user.h>
22#include <linux/elf.h>
23#include <linux/security.h>
24#include <linux/bootmem.h>
25#include <linux/compat.h>
26#include <asm/asm-offsets.h>
27#include <asm/pgtable.h>
28#include <asm/processor.h>
29#include <asm/mmu.h>
30#include <asm/mmu_context.h>
31#include <asm/sections.h>
32#include <asm/vdso.h>
33#include <asm/facility.h>
34
35#ifdef CONFIG_COMPAT
36extern char vdso32_start, vdso32_end;
37static void *vdso32_kbase = &vdso32_start;
38static unsigned int vdso32_pages;
39static struct page **vdso32_pagelist;
40#endif
41
42extern char vdso64_start, vdso64_end;
43static void *vdso64_kbase = &vdso64_start;
44static unsigned int vdso64_pages;
45static struct page **vdso64_pagelist;
46
47/*
48 * Should the kernel map a VDSO page into processes and pass its
49 * address down to glibc upon exec()?
50 */
51unsigned int __read_mostly vdso_enabled = 1;
52
53static int __init vdso_setup(char *s)
54{
55 unsigned long val;
56 int rc;
57
58 rc = 0;
59 if (strncmp(s, "on", 3) == 0)
60 vdso_enabled = 1;
61 else if (strncmp(s, "off", 4) == 0)
62 vdso_enabled = 0;
63 else {
64 rc = kstrtoul(s, 0, &val);
65 vdso_enabled = rc ? 0 : !!val;
66 }
67 return !rc;
68}
69__setup("vdso=", vdso_setup);
70
71/*
72 * The vdso data page
73 */
74static union {
75 struct vdso_data data;
76 u8 page[PAGE_SIZE];
77} vdso_data_store __page_aligned_data;
78struct vdso_data *vdso_data = &vdso_data_store.data;
79
80/*
81 * Setup vdso data page.
82 */
83static void __init vdso_init_data(struct vdso_data *vd)
84{
85 vd->ectg_available = test_facility(31);
86}
87
88/*
89 * Allocate/free per cpu vdso data.
90 */
91#define SEGMENT_ORDER 2
92
93int vdso_alloc_per_cpu(struct lowcore *lowcore)
94{
95 unsigned long segment_table, page_table, page_frame;
96 struct vdso_per_cpu_data *vd;
97 u32 *psal, *aste;
98 int i;
99
100 lowcore->vdso_per_cpu_data = __LC_PASTE;
101
102 if (!vdso_enabled)
103 return 0;
104
105 segment_table = __get_free_pages(GFP_KERNEL, SEGMENT_ORDER);
106 page_table = get_zeroed_page(GFP_KERNEL | GFP_DMA);
107 page_frame = get_zeroed_page(GFP_KERNEL);
108 if (!segment_table || !page_table || !page_frame)
109 goto out;
110
111 /* Initialize per-cpu vdso data page */
112 vd = (struct vdso_per_cpu_data *) page_frame;
113 vd->cpu_nr = lowcore->cpu_nr;
114 vd->node_id = cpu_to_node(vd->cpu_nr);
115
116 /* Set up access register mode page table */
117 clear_table((unsigned long *) segment_table, _SEGMENT_ENTRY_EMPTY,
118 PAGE_SIZE << SEGMENT_ORDER);
119 clear_table((unsigned long *) page_table, _PAGE_INVALID,
120 256*sizeof(unsigned long));
121
122 *(unsigned long *) segment_table = _SEGMENT_ENTRY + page_table;
123 *(unsigned long *) page_table = _PAGE_PROTECT + page_frame;
124
125 psal = (u32 *) (page_table + 256*sizeof(unsigned long));
126 aste = psal + 32;
127
128 for (i = 4; i < 32; i += 4)
129 psal[i] = 0x80000000;
130
131 lowcore->paste[4] = (u32)(addr_t) psal;
132 psal[0] = 0x02000000;
133 psal[2] = (u32)(addr_t) aste;
134 *(unsigned long *) (aste + 2) = segment_table +
135 _ASCE_TABLE_LENGTH + _ASCE_USER_BITS + _ASCE_TYPE_SEGMENT;
136 aste[4] = (u32)(addr_t) psal;
137 lowcore->vdso_per_cpu_data = page_frame;
138
139 return 0;
140
141out:
142 free_page(page_frame);
143 free_page(page_table);
144 free_pages(segment_table, SEGMENT_ORDER);
145 return -ENOMEM;
146}
147
148void vdso_free_per_cpu(struct lowcore *lowcore)
149{
150 unsigned long segment_table, page_table, page_frame;
151 u32 *psal, *aste;
152
153 if (!vdso_enabled)
154 return;
155
156 psal = (u32 *)(addr_t) lowcore->paste[4];
157 aste = (u32 *)(addr_t) psal[2];
158 segment_table = *(unsigned long *)(aste + 2) & PAGE_MASK;
159 page_table = *(unsigned long *) segment_table;
160 page_frame = *(unsigned long *) page_table;
161
162 free_page(page_frame);
163 free_page(page_table);
164 free_pages(segment_table, SEGMENT_ORDER);
165}
166
167static void vdso_init_cr5(void)
168{
169 unsigned long cr5;
170
171 if (!vdso_enabled)
172 return;
173 cr5 = offsetof(struct lowcore, paste);
174 __ctl_load(cr5, 5, 5);
175}
176
177/*
178 * This is called from binfmt_elf, we create the special vma for the
179 * vDSO and insert it into the mm struct tree
180 */
181int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
182{
183 struct mm_struct *mm = current->mm;
184 struct page **vdso_pagelist;
185 unsigned long vdso_pages;
186 unsigned long vdso_base;
187 int rc;
188
189 if (!vdso_enabled)
190 return 0;
191 /*
192 * Only map the vdso for dynamically linked elf binaries.
193 */
194 if (!uses_interp)
195 return 0;
196
197 vdso_pagelist = vdso64_pagelist;
198 vdso_pages = vdso64_pages;
199#ifdef CONFIG_COMPAT
200 if (is_compat_task()) {
201 vdso_pagelist = vdso32_pagelist;
202 vdso_pages = vdso32_pages;
203 }
204#endif
205 /*
206 * vDSO has a problem and was disabled, just don't "enable" it for
207 * the process
208 */
209 if (vdso_pages == 0)
210 return 0;
211
212 current->mm->context.vdso_base = 0;
213
214 /*
215 * pick a base address for the vDSO in process space. We try to put
216 * it at vdso_base which is the "natural" base for it, but we might
217 * fail and end up putting it elsewhere.
218 */
219 down_write(&mm->mmap_sem);
220 vdso_base = get_unmapped_area(NULL, 0, vdso_pages << PAGE_SHIFT, 0, 0);
221 if (IS_ERR_VALUE(vdso_base)) {
222 rc = vdso_base;
223 goto out_up;
224 }
225
226 /*
227 * Put vDSO base into mm struct. We need to do this before calling
228 * install_special_mapping or the perf counter mmap tracking code
229 * will fail to recognise it as a vDSO (since arch_vma_name fails).
230 */
231 current->mm->context.vdso_base = vdso_base;
232
233 /*
234 * our vma flags don't have VM_WRITE so by default, the process
235 * isn't allowed to write those pages.
236 * gdb can break that with ptrace interface, and thus trigger COW
237 * on those pages but it's then your responsibility to never do that
238 * on the "data" page of the vDSO or you'll stop getting kernel
239 * updates and your nice userland gettimeofday will be totally dead.
240 * It's fine to use that for setting breakpoints in the vDSO code
241 * pages though.
242 */
243 rc = install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
244 VM_READ|VM_EXEC|
245 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
246 vdso_pagelist);
247 if (rc)
248 current->mm->context.vdso_base = 0;
249out_up:
250 up_write(&mm->mmap_sem);
251 return rc;
252}
253
254const char *arch_vma_name(struct vm_area_struct *vma)
255{
256 if (vma->vm_mm && vma->vm_start == vma->vm_mm->context.vdso_base)
257 return "[vdso]";
258 return NULL;
259}
260
261static int __init vdso_init(void)
262{
263 int i;
264
265 if (!vdso_enabled)
266 return 0;
267 vdso_init_data(vdso_data);
268#ifdef CONFIG_COMPAT
269 /* Calculate the size of the 32 bit vDSO */
270 vdso32_pages = ((&vdso32_end - &vdso32_start
271 + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
272
273 /* Make sure pages are in the correct state */
274 vdso32_pagelist = kzalloc(sizeof(struct page *) * (vdso32_pages + 1),
275 GFP_KERNEL);
276 BUG_ON(vdso32_pagelist == NULL);
277 for (i = 0; i < vdso32_pages - 1; i++) {
278 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
279 ClearPageReserved(pg);
280 get_page(pg);
281 vdso32_pagelist[i] = pg;
282 }
283 vdso32_pagelist[vdso32_pages - 1] = virt_to_page(vdso_data);
284 vdso32_pagelist[vdso32_pages] = NULL;
285#endif
286
287 /* Calculate the size of the 64 bit vDSO */
288 vdso64_pages = ((&vdso64_end - &vdso64_start
289 + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
290
291 /* Make sure pages are in the correct state */
292 vdso64_pagelist = kzalloc(sizeof(struct page *) * (vdso64_pages + 1),
293 GFP_KERNEL);
294 BUG_ON(vdso64_pagelist == NULL);
295 for (i = 0; i < vdso64_pages - 1; i++) {
296 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
297 ClearPageReserved(pg);
298 get_page(pg);
299 vdso64_pagelist[i] = pg;
300 }
301 vdso64_pagelist[vdso64_pages - 1] = virt_to_page(vdso_data);
302 vdso64_pagelist[vdso64_pages] = NULL;
303 if (vdso_alloc_per_cpu(&S390_lowcore))
304 BUG();
305 vdso_init_cr5();
306
307 get_page(virt_to_page(vdso_data));
308
309 return 0;
310}
311early_initcall(vdso_init);