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// 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/init.h>
10#include <linux/errno.h>
11#include <linux/sched.h>
12#include <linux/kernel.h>
13#include <linux/mm.h>
14#include <linux/smp.h>
15#include <linux/stddef.h>
16#include <linux/unistd.h>
17#include <linux/slab.h>
18#include <linux/user.h>
19#include <linux/elf.h>
20#include <linux/security.h>
21#include <linux/bootmem.h>
22#include <linux/compat.h>
23#include <asm/asm-offsets.h>
24#include <asm/pgtable.h>
25#include <asm/processor.h>
26#include <asm/mmu.h>
27#include <asm/mmu_context.h>
28#include <asm/sections.h>
29#include <asm/vdso.h>
30#include <asm/facility.h>
31
32#ifdef CONFIG_COMPAT
33extern char vdso32_start, vdso32_end;
34static void *vdso32_kbase = &vdso32_start;
35static unsigned int vdso32_pages;
36static struct page **vdso32_pagelist;
37#endif
38
39extern char vdso64_start, vdso64_end;
40static void *vdso64_kbase = &vdso64_start;
41static unsigned int vdso64_pages;
42static struct page **vdso64_pagelist;
43
44/*
45 * Should the kernel map a VDSO page into processes and pass its
46 * address down to glibc upon exec()?
47 */
48unsigned int __read_mostly vdso_enabled = 1;
49
50static int vdso_fault(const struct vm_special_mapping *sm,
51 struct vm_area_struct *vma, struct vm_fault *vmf)
52{
53 struct page **vdso_pagelist;
54 unsigned long vdso_pages;
55
56 vdso_pagelist = vdso64_pagelist;
57 vdso_pages = vdso64_pages;
58#ifdef CONFIG_COMPAT
59 if (is_compat_task()) {
60 vdso_pagelist = vdso32_pagelist;
61 vdso_pages = vdso32_pages;
62 }
63#endif
64
65 if (vmf->pgoff >= vdso_pages)
66 return VM_FAULT_SIGBUS;
67
68 vmf->page = vdso_pagelist[vmf->pgoff];
69 get_page(vmf->page);
70 return 0;
71}
72
73static int vdso_mremap(const struct vm_special_mapping *sm,
74 struct vm_area_struct *vma)
75{
76 unsigned long vdso_pages;
77
78 vdso_pages = vdso64_pages;
79#ifdef CONFIG_COMPAT
80 if (is_compat_task())
81 vdso_pages = vdso32_pages;
82#endif
83
84 if ((vdso_pages << PAGE_SHIFT) != vma->vm_end - vma->vm_start)
85 return -EINVAL;
86
87 if (WARN_ON_ONCE(current->mm != vma->vm_mm))
88 return -EFAULT;
89
90 current->mm->context.vdso_base = vma->vm_start;
91 return 0;
92}
93
94static const struct vm_special_mapping vdso_mapping = {
95 .name = "[vdso]",
96 .fault = vdso_fault,
97 .mremap = vdso_mremap,
98};
99
100static int __init vdso_setup(char *s)
101{
102 unsigned long val;
103 int rc;
104
105 rc = 0;
106 if (strncmp(s, "on", 3) == 0)
107 vdso_enabled = 1;
108 else if (strncmp(s, "off", 4) == 0)
109 vdso_enabled = 0;
110 else {
111 rc = kstrtoul(s, 0, &val);
112 vdso_enabled = rc ? 0 : !!val;
113 }
114 return !rc;
115}
116__setup("vdso=", vdso_setup);
117
118/*
119 * The vdso data page
120 */
121static union {
122 struct vdso_data data;
123 u8 page[PAGE_SIZE];
124} vdso_data_store __page_aligned_data;
125struct vdso_data *vdso_data = &vdso_data_store.data;
126
127/*
128 * Setup vdso data page.
129 */
130static void __init vdso_init_data(struct vdso_data *vd)
131{
132 vd->ectg_available = test_facility(31);
133}
134
135/*
136 * Allocate/free per cpu vdso data.
137 */
138#define SEGMENT_ORDER 2
139
140/*
141 * The initial vdso_data structure for the boot CPU. Eventually
142 * it is replaced with a properly allocated structure in vdso_init.
143 * This is necessary because a valid S390_lowcore.vdso_per_cpu_data
144 * pointer is required to be able to return from an interrupt or
145 * program check. See the exit paths in entry.S.
146 */
147struct vdso_data boot_vdso_data __initdata;
148
149void __init vdso_alloc_boot_cpu(struct lowcore *lowcore)
150{
151 lowcore->vdso_per_cpu_data = (unsigned long) &boot_vdso_data;
152}
153
154int vdso_alloc_per_cpu(struct lowcore *lowcore)
155{
156 unsigned long segment_table, page_table, page_frame;
157 struct vdso_per_cpu_data *vd;
158
159 segment_table = __get_free_pages(GFP_KERNEL, SEGMENT_ORDER);
160 page_table = get_zeroed_page(GFP_KERNEL);
161 page_frame = get_zeroed_page(GFP_KERNEL);
162 if (!segment_table || !page_table || !page_frame)
163 goto out;
164 arch_set_page_dat(virt_to_page(segment_table), SEGMENT_ORDER);
165 arch_set_page_dat(virt_to_page(page_table), 0);
166
167 /* Initialize per-cpu vdso data page */
168 vd = (struct vdso_per_cpu_data *) page_frame;
169 vd->cpu_nr = lowcore->cpu_nr;
170 vd->node_id = cpu_to_node(vd->cpu_nr);
171
172 /* Set up page table for the vdso address space */
173 memset64((u64 *)segment_table, _SEGMENT_ENTRY_EMPTY, _CRST_ENTRIES);
174 memset64((u64 *)page_table, _PAGE_INVALID, PTRS_PER_PTE);
175
176 *(unsigned long *) segment_table = _SEGMENT_ENTRY + page_table;
177 *(unsigned long *) page_table = _PAGE_PROTECT + page_frame;
178
179 lowcore->vdso_asce = segment_table +
180 _ASCE_TABLE_LENGTH + _ASCE_USER_BITS + _ASCE_TYPE_SEGMENT;
181 lowcore->vdso_per_cpu_data = page_frame;
182
183 return 0;
184
185out:
186 free_page(page_frame);
187 free_page(page_table);
188 free_pages(segment_table, SEGMENT_ORDER);
189 return -ENOMEM;
190}
191
192void vdso_free_per_cpu(struct lowcore *lowcore)
193{
194 unsigned long segment_table, page_table, page_frame;
195
196 segment_table = lowcore->vdso_asce & PAGE_MASK;
197 page_table = *(unsigned long *) segment_table;
198 page_frame = *(unsigned long *) page_table;
199
200 free_page(page_frame);
201 free_page(page_table);
202 free_pages(segment_table, SEGMENT_ORDER);
203}
204
205/*
206 * This is called from binfmt_elf, we create the special vma for the
207 * vDSO and insert it into the mm struct tree
208 */
209int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
210{
211 struct mm_struct *mm = current->mm;
212 struct vm_area_struct *vma;
213 unsigned long vdso_pages;
214 unsigned long vdso_base;
215 int rc;
216
217 if (!vdso_enabled)
218 return 0;
219 /*
220 * Only map the vdso for dynamically linked elf binaries.
221 */
222 if (!uses_interp)
223 return 0;
224
225 vdso_pages = vdso64_pages;
226#ifdef CONFIG_COMPAT
227 if (is_compat_task())
228 vdso_pages = vdso32_pages;
229#endif
230 /*
231 * vDSO has a problem and was disabled, just don't "enable" it for
232 * the process
233 */
234 if (vdso_pages == 0)
235 return 0;
236
237 /*
238 * pick a base address for the vDSO in process space. We try to put
239 * it at vdso_base which is the "natural" base for it, but we might
240 * fail and end up putting it elsewhere.
241 */
242 if (down_write_killable(&mm->mmap_sem))
243 return -EINTR;
244 vdso_base = get_unmapped_area(NULL, 0, vdso_pages << PAGE_SHIFT, 0, 0);
245 if (IS_ERR_VALUE(vdso_base)) {
246 rc = vdso_base;
247 goto out_up;
248 }
249
250 /*
251 * our vma flags don't have VM_WRITE so by default, the process
252 * isn't allowed to write those pages.
253 * gdb can break that with ptrace interface, and thus trigger COW
254 * on those pages but it's then your responsibility to never do that
255 * on the "data" page of the vDSO or you'll stop getting kernel
256 * updates and your nice userland gettimeofday will be totally dead.
257 * It's fine to use that for setting breakpoints in the vDSO code
258 * pages though.
259 */
260 vma = _install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
261 VM_READ|VM_EXEC|
262 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
263 &vdso_mapping);
264 if (IS_ERR(vma)) {
265 rc = PTR_ERR(vma);
266 goto out_up;
267 }
268
269 current->mm->context.vdso_base = vdso_base;
270 rc = 0;
271
272out_up:
273 up_write(&mm->mmap_sem);
274 return rc;
275}
276
277static int __init vdso_init(void)
278{
279 int i;
280
281 vdso_init_data(vdso_data);
282#ifdef CONFIG_COMPAT
283 /* Calculate the size of the 32 bit vDSO */
284 vdso32_pages = ((&vdso32_end - &vdso32_start
285 + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
286
287 /* Make sure pages are in the correct state */
288 vdso32_pagelist = kzalloc(sizeof(struct page *) * (vdso32_pages + 1),
289 GFP_KERNEL);
290 BUG_ON(vdso32_pagelist == NULL);
291 for (i = 0; i < vdso32_pages - 1; i++) {
292 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
293 ClearPageReserved(pg);
294 get_page(pg);
295 vdso32_pagelist[i] = pg;
296 }
297 vdso32_pagelist[vdso32_pages - 1] = virt_to_page(vdso_data);
298 vdso32_pagelist[vdso32_pages] = NULL;
299#endif
300
301 /* Calculate the size of the 64 bit vDSO */
302 vdso64_pages = ((&vdso64_end - &vdso64_start
303 + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
304
305 /* Make sure pages are in the correct state */
306 vdso64_pagelist = kzalloc(sizeof(struct page *) * (vdso64_pages + 1),
307 GFP_KERNEL);
308 BUG_ON(vdso64_pagelist == NULL);
309 for (i = 0; i < vdso64_pages - 1; i++) {
310 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
311 ClearPageReserved(pg);
312 get_page(pg);
313 vdso64_pagelist[i] = pg;
314 }
315 vdso64_pagelist[vdso64_pages - 1] = virt_to_page(vdso_data);
316 vdso64_pagelist[vdso64_pages] = NULL;
317 if (vdso_alloc_per_cpu(&S390_lowcore))
318 BUG();
319
320 get_page(virt_to_page(vdso_data));
321
322 return 0;
323}
324early_initcall(vdso_init);