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