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/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
26extern char vdso64_start[], vdso64_end[];
27extern char vdso32_start[], vdso32_end[];
28
29static struct vm_special_mapping vvar_mapping;
30
31static union vdso_data_store vdso_data_store __page_aligned_data;
32
33struct vdso_data *vdso_data = vdso_data_store.data;
34
35#ifdef CONFIG_TIME_NS
36struct 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 */
48int 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
66static 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
108static 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
115static struct vm_special_mapping vvar_mapping = {
116 .name = "[vvar]",
117 .fault = vvar_fault,
118};
119
120static struct vm_special_mapping vdso64_mapping = {
121 .name = "[vdso]",
122 .mremap = vdso_mremap,
123};
124
125static struct vm_special_mapping vdso32_mapping = {
126 .name = "[vdso]",
127 .mremap = vdso_mremap,
128};
129
130int vdso_getcpu_init(void)
131{
132 set_tod_programmable_field(smp_processor_id());
133 return 0;
134}
135early_initcall(vdso_getcpu_init); /* Must be called before SMP init */
136
137static 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 }
180out:
181 mmap_write_unlock(mm);
182 return rc;
183}
184
185static 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
210unsigned 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
221unsigned long vdso_size(void)
222{
223 return vdso_text_size() + VVAR_NR_PAGES * PAGE_SIZE;
224}
225
226int 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
236static 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
250static 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
266static 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}
274arch_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/memblock.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_VDSO
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 vm_fault_t 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_VDSO
59 if (vma->vm_mm->context.compat_mm) {
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_VDSO
80 if (vma->vm_mm->context.compat_mm)
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 *str)
101{
102 bool enabled;
103
104 if (!kstrtobool(str, &enabled))
105 vdso_enabled = enabled;
106 return 1;
107}
108__setup("vdso=", vdso_setup);
109
110/*
111 * The vdso data page
112 */
113static union {
114 struct vdso_data data;
115 u8 page[PAGE_SIZE];
116} vdso_data_store __page_aligned_data;
117struct vdso_data *vdso_data = &vdso_data_store.data;
118
119/*
120 * Setup vdso data page.
121 */
122static void __init vdso_init_data(struct vdso_data *vd)
123{
124 vd->ectg_available = test_facility(31);
125}
126
127/*
128 * Allocate/free per cpu vdso data.
129 */
130#define SEGMENT_ORDER 2
131
132/*
133 * The initial vdso_data structure for the boot CPU. Eventually
134 * it is replaced with a properly allocated structure in vdso_init.
135 * This is necessary because a valid S390_lowcore.vdso_per_cpu_data
136 * pointer is required to be able to return from an interrupt or
137 * program check. See the exit paths in entry.S.
138 */
139struct vdso_data boot_vdso_data __initdata;
140
141void __init vdso_alloc_boot_cpu(struct lowcore *lowcore)
142{
143 lowcore->vdso_per_cpu_data = (unsigned long) &boot_vdso_data;
144}
145
146int vdso_alloc_per_cpu(struct lowcore *lowcore)
147{
148 unsigned long segment_table, page_table, page_frame;
149 struct vdso_per_cpu_data *vd;
150
151 segment_table = __get_free_pages(GFP_KERNEL, SEGMENT_ORDER);
152 page_table = get_zeroed_page(GFP_KERNEL);
153 page_frame = get_zeroed_page(GFP_KERNEL);
154 if (!segment_table || !page_table || !page_frame)
155 goto out;
156 arch_set_page_dat(virt_to_page(segment_table), SEGMENT_ORDER);
157 arch_set_page_dat(virt_to_page(page_table), 0);
158
159 /* Initialize per-cpu vdso data page */
160 vd = (struct vdso_per_cpu_data *) page_frame;
161 vd->cpu_nr = lowcore->cpu_nr;
162 vd->node_id = cpu_to_node(vd->cpu_nr);
163
164 /* Set up page table for the vdso address space */
165 memset64((u64 *)segment_table, _SEGMENT_ENTRY_EMPTY, _CRST_ENTRIES);
166 memset64((u64 *)page_table, _PAGE_INVALID, PTRS_PER_PTE);
167
168 *(unsigned long *) segment_table = _SEGMENT_ENTRY + page_table;
169 *(unsigned long *) page_table = _PAGE_PROTECT + page_frame;
170
171 lowcore->vdso_asce = segment_table +
172 _ASCE_TABLE_LENGTH + _ASCE_USER_BITS + _ASCE_TYPE_SEGMENT;
173 lowcore->vdso_per_cpu_data = page_frame;
174
175 return 0;
176
177out:
178 free_page(page_frame);
179 free_page(page_table);
180 free_pages(segment_table, SEGMENT_ORDER);
181 return -ENOMEM;
182}
183
184void vdso_free_per_cpu(struct lowcore *lowcore)
185{
186 unsigned long segment_table, page_table, page_frame;
187
188 segment_table = lowcore->vdso_asce & PAGE_MASK;
189 page_table = *(unsigned long *) segment_table;
190 page_frame = *(unsigned long *) page_table;
191
192 free_page(page_frame);
193 free_page(page_table);
194 free_pages(segment_table, SEGMENT_ORDER);
195}
196
197/*
198 * This is called from binfmt_elf, we create the special vma for the
199 * vDSO and insert it into the mm struct tree
200 */
201int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
202{
203 struct mm_struct *mm = current->mm;
204 struct vm_area_struct *vma;
205 unsigned long vdso_pages;
206 unsigned long vdso_base;
207 int rc;
208
209 if (!vdso_enabled)
210 return 0;
211
212 vdso_pages = vdso64_pages;
213#ifdef CONFIG_COMPAT_VDSO
214 mm->context.compat_mm = is_compat_task();
215 if (mm->context.compat_mm)
216 vdso_pages = vdso32_pages;
217#endif
218 /*
219 * vDSO has a problem and was disabled, just don't "enable" it for
220 * the process
221 */
222 if (vdso_pages == 0)
223 return 0;
224
225 /*
226 * pick a base address for the vDSO in process space. We try to put
227 * it at vdso_base which is the "natural" base for it, but we might
228 * fail and end up putting it elsewhere.
229 */
230 if (down_write_killable(&mm->mmap_sem))
231 return -EINTR;
232 vdso_base = get_unmapped_area(NULL, 0, vdso_pages << PAGE_SHIFT, 0, 0);
233 if (IS_ERR_VALUE(vdso_base)) {
234 rc = vdso_base;
235 goto out_up;
236 }
237
238 /*
239 * our vma flags don't have VM_WRITE so by default, the process
240 * isn't allowed to write those pages.
241 * gdb can break that with ptrace interface, and thus trigger COW
242 * on those pages but it's then your responsibility to never do that
243 * on the "data" page of the vDSO or you'll stop getting kernel
244 * updates and your nice userland gettimeofday will be totally dead.
245 * It's fine to use that for setting breakpoints in the vDSO code
246 * pages though.
247 */
248 vma = _install_special_mapping(mm, vdso_base, vdso_pages << PAGE_SHIFT,
249 VM_READ|VM_EXEC|
250 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
251 &vdso_mapping);
252 if (IS_ERR(vma)) {
253 rc = PTR_ERR(vma);
254 goto out_up;
255 }
256
257 current->mm->context.vdso_base = vdso_base;
258 rc = 0;
259
260out_up:
261 up_write(&mm->mmap_sem);
262 return rc;
263}
264
265static int __init vdso_init(void)
266{
267 int i;
268
269 vdso_init_data(vdso_data);
270#ifdef CONFIG_COMPAT_VDSO
271 /* Calculate the size of the 32 bit vDSO */
272 vdso32_pages = ((&vdso32_end - &vdso32_start
273 + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
274
275 /* Make sure pages are in the correct state */
276 vdso32_pagelist = kcalloc(vdso32_pages + 1, sizeof(struct page *),
277 GFP_KERNEL);
278 BUG_ON(vdso32_pagelist == NULL);
279 for (i = 0; i < vdso32_pages - 1; i++) {
280 struct page *pg = virt_to_page(vdso32_kbase + i*PAGE_SIZE);
281 get_page(pg);
282 vdso32_pagelist[i] = pg;
283 }
284 vdso32_pagelist[vdso32_pages - 1] = virt_to_page(vdso_data);
285 vdso32_pagelist[vdso32_pages] = NULL;
286#endif
287
288 /* Calculate the size of the 64 bit vDSO */
289 vdso64_pages = ((&vdso64_end - &vdso64_start
290 + PAGE_SIZE - 1) >> PAGE_SHIFT) + 1;
291
292 /* Make sure pages are in the correct state */
293 vdso64_pagelist = kcalloc(vdso64_pages + 1, sizeof(struct page *),
294 GFP_KERNEL);
295 BUG_ON(vdso64_pagelist == NULL);
296 for (i = 0; i < vdso64_pages - 1; i++) {
297 struct page *pg = virt_to_page(vdso64_kbase + i*PAGE_SIZE);
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
306 get_page(virt_to_page(vdso_data));
307
308 return 0;
309}
310early_initcall(vdso_init);