xref: /linux/arch/s390/kernel/vdso.c (revision cbafa54aa2ae23939846e150ad4ba98c784f6395)
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 <vdso/datapage.h>
20 #include <asm/vdso.h>
21 
22 extern char vdso64_start[], vdso64_end[];
23 static unsigned int vdso_pages;
24 
25 static struct vm_special_mapping vvar_mapping;
26 
27 static union {
28 	struct vdso_data	data[CS_BASES];
29 	u8			page[PAGE_SIZE];
30 } vdso_data_store __page_aligned_data;
31 
32 struct vdso_data *vdso_data = vdso_data_store.data;
33 
34 enum vvar_pages {
35 	VVAR_DATA_PAGE_OFFSET,
36 	VVAR_TIMENS_PAGE_OFFSET,
37 	VVAR_NR_PAGES,
38 };
39 
40 unsigned int __read_mostly vdso_enabled = 1;
41 
42 static int __init vdso_setup(char *str)
43 {
44 	bool enabled;
45 
46 	if (!kstrtobool(str, &enabled))
47 		vdso_enabled = enabled;
48 	return 1;
49 }
50 __setup("vdso=", vdso_setup);
51 
52 #ifdef CONFIG_TIME_NS
53 struct vdso_data *arch_get_vdso_data(void *vvar_page)
54 {
55 	return (struct vdso_data *)(vvar_page);
56 }
57 
58 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
59 {
60 	if (likely(vma->vm_mm == current->mm))
61 		return current->nsproxy->time_ns->vvar_page;
62 	/*
63 	 * VM_PFNMAP | VM_IO protect .fault() handler from being called
64 	 * through interfaces like /proc/$pid/mem or
65 	 * process_vm_{readv,writev}() as long as there's no .access()
66 	 * in special_mapping_vmops().
67 	 * For more details check_vma_flags() and __access_remote_vm()
68 	 */
69 	WARN(1, "vvar_page accessed remotely");
70 	return NULL;
71 }
72 
73 /*
74  * The VVAR page layout depends on whether a task belongs to the root or
75  * non-root time namespace. Whenever a task changes its namespace, the VVAR
76  * page tables are cleared and then they will be re-faulted with a
77  * corresponding layout.
78  * See also the comment near timens_setup_vdso_data() for details.
79  */
80 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
81 {
82 	struct mm_struct *mm = task->mm;
83 	struct vm_area_struct *vma;
84 
85 	mmap_read_lock(mm);
86 	for (vma = mm->mmap; vma; vma = vma->vm_next) {
87 		unsigned long size = vma->vm_end - vma->vm_start;
88 
89 		if (!vma_is_special_mapping(vma, &vvar_mapping))
90 			continue;
91 		zap_page_range(vma, vma->vm_start, size);
92 		break;
93 	}
94 	mmap_read_unlock(mm);
95 	return 0;
96 }
97 #else
98 static inline struct page *find_timens_vvar_page(struct vm_area_struct *vma)
99 {
100 	return NULL;
101 }
102 #endif
103 
104 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
105 			     struct vm_area_struct *vma, struct vm_fault *vmf)
106 {
107 	struct page *timens_page = find_timens_vvar_page(vma);
108 	unsigned long addr, pfn;
109 	vm_fault_t err;
110 
111 	switch (vmf->pgoff) {
112 	case VVAR_DATA_PAGE_OFFSET:
113 		pfn = virt_to_pfn(vdso_data);
114 		if (timens_page) {
115 			/*
116 			 * Fault in VVAR page too, since it will be accessed
117 			 * to get clock data anyway.
118 			 */
119 			addr = vmf->address + VVAR_TIMENS_PAGE_OFFSET * PAGE_SIZE;
120 			err = vmf_insert_pfn(vma, addr, pfn);
121 			if (unlikely(err & VM_FAULT_ERROR))
122 				return err;
123 			pfn = page_to_pfn(timens_page);
124 		}
125 		break;
126 #ifdef CONFIG_TIME_NS
127 	case VVAR_TIMENS_PAGE_OFFSET:
128 		/*
129 		 * If a task belongs to a time namespace then a namespace
130 		 * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
131 		 * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
132 		 * offset.
133 		 * See also the comment near timens_setup_vdso_data().
134 		 */
135 		if (!timens_page)
136 			return VM_FAULT_SIGBUS;
137 		pfn = virt_to_pfn(vdso_data);
138 		break;
139 #endif /* CONFIG_TIME_NS */
140 	default:
141 		return VM_FAULT_SIGBUS;
142 	}
143 	return vmf_insert_pfn(vma, vmf->address, pfn);
144 }
145 
146 static int vdso_mremap(const struct vm_special_mapping *sm,
147 		       struct vm_area_struct *vma)
148 {
149 	current->mm->context.vdso_base = vma->vm_start;
150 	return 0;
151 }
152 
153 static struct vm_special_mapping vvar_mapping = {
154 	.name = "[vvar]",
155 	.fault = vvar_fault,
156 };
157 
158 static struct vm_special_mapping vdso_mapping = {
159 	.name = "[vdso]",
160 	.mremap = vdso_mremap,
161 };
162 
163 int vdso_getcpu_init(void)
164 {
165 	set_tod_programmable_field(smp_processor_id());
166 	return 0;
167 }
168 early_initcall(vdso_getcpu_init); /* Must be called before SMP init */
169 
170 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
171 {
172 	unsigned long vdso_text_len, vdso_mapping_len;
173 	unsigned long vvar_start, vdso_text_start;
174 	struct mm_struct *mm = current->mm;
175 	struct vm_area_struct *vma;
176 	int rc;
177 
178 	BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES);
179 	if (!vdso_enabled || is_compat_task())
180 		return 0;
181 	if (mmap_write_lock_killable(mm))
182 		return -EINTR;
183 	vdso_text_len = vdso_pages << PAGE_SHIFT;
184 	vdso_mapping_len = vdso_text_len + VVAR_NR_PAGES * PAGE_SIZE;
185 	vvar_start = get_unmapped_area(NULL, 0, vdso_mapping_len, 0, 0);
186 	rc = vvar_start;
187 	if (IS_ERR_VALUE(vvar_start))
188 		goto out;
189 	vma = _install_special_mapping(mm, vvar_start, VVAR_NR_PAGES*PAGE_SIZE,
190 				       VM_READ|VM_MAYREAD|VM_IO|VM_DONTDUMP|
191 				       VM_PFNMAP,
192 				       &vvar_mapping);
193 	rc = PTR_ERR(vma);
194 	if (IS_ERR(vma))
195 		goto out;
196 	vdso_text_start = vvar_start + VVAR_NR_PAGES * PAGE_SIZE;
197 	/* VM_MAYWRITE for COW so gdb can set breakpoints */
198 	vma = _install_special_mapping(mm, vdso_text_start, vdso_text_len,
199 				       VM_READ|VM_EXEC|
200 				       VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
201 				       &vdso_mapping);
202 	if (IS_ERR(vma)) {
203 		do_munmap(mm, vvar_start, PAGE_SIZE, NULL);
204 		rc = PTR_ERR(vma);
205 	} else {
206 		current->mm->context.vdso_base = vdso_text_start;
207 		rc = 0;
208 	}
209 out:
210 	mmap_write_unlock(mm);
211 	return rc;
212 }
213 
214 static int __init vdso_init(void)
215 {
216 	struct page **pages;
217 	int i;
218 
219 	vdso_pages = (vdso64_end - vdso64_start) >> PAGE_SHIFT;
220 	pages = kcalloc(vdso_pages + 1, sizeof(struct page *), GFP_KERNEL);
221 	if (!pages) {
222 		vdso_enabled = 0;
223 		return -ENOMEM;
224 	}
225 	for (i = 0; i < vdso_pages; i++)
226 		pages[i] = virt_to_page(vdso64_start + i * PAGE_SIZE);
227 	pages[vdso_pages] = NULL;
228 	vdso_mapping.pages = pages;
229 	return 0;
230 }
231 arch_initcall(vdso_init);
232