xref: /linux/lib/vdso/datastore.c (revision 05988dba11791ccbb458254484826b32f17f4ad2)
1 // SPDX-License-Identifier: GPL-2.0-only
2 
3 #include <linux/gfp.h>
4 #include <linux/init.h>
5 #include <linux/mm.h>
6 #include <linux/time_namespace.h>
7 #include <linux/types.h>
8 #include <linux/vdso_datastore.h>
9 #include <vdso/datapage.h>
10 
11 static u8 vdso_initdata[VDSO_NR_PAGES * PAGE_SIZE] __aligned(PAGE_SIZE) __initdata = {};
12 
13 #ifdef CONFIG_GENERIC_GETTIMEOFDAY
14 struct vdso_time_data *vdso_k_time_data __refdata =
15 	(void *)&vdso_initdata[VDSO_TIME_PAGE_OFFSET * PAGE_SIZE];
16 
17 static_assert(sizeof(struct vdso_time_data) <= PAGE_SIZE);
18 #endif /* CONFIG_GENERIC_GETTIMEOFDAY */
19 
20 #ifdef CONFIG_VDSO_GETRANDOM
21 struct vdso_rng_data *vdso_k_rng_data __refdata =
22 	(void *)&vdso_initdata[VDSO_RNG_PAGE_OFFSET * PAGE_SIZE];
23 
24 static_assert(sizeof(struct vdso_rng_data) <= PAGE_SIZE);
25 #endif /* CONFIG_VDSO_GETRANDOM */
26 
27 #ifdef CONFIG_ARCH_HAS_VDSO_ARCH_DATA
28 struct vdso_arch_data *vdso_k_arch_data __refdata =
29 	(void *)&vdso_initdata[VDSO_ARCH_PAGES_START * PAGE_SIZE];
30 #endif /* CONFIG_ARCH_HAS_VDSO_ARCH_DATA */
31 
32 void __init vdso_setup_data_pages(void)
33 {
34 	unsigned int order = get_order(VDSO_NR_PAGES * PAGE_SIZE);
35 	struct page *pages;
36 
37 	/*
38 	 * Allocate the data pages dynamically. SPARC does not support mapping
39 	 * static pages to be mapped into userspace.
40 	 * It is also a requirement for mlockall() support.
41 	 *
42 	 * Do not use folios. In time namespaces the pages are mapped in a different order
43 	 * to userspace, which is not handled by the folio optimizations in finish_fault().
44 	 */
45 	pages = alloc_pages(GFP_KERNEL, order);
46 	if (!pages)
47 		panic("Unable to allocate VDSO storage pages");
48 
49 	/* The pages are mapped one-by-one into userspace and each one needs to be refcounted. */
50 	split_page(pages, order);
51 
52 	/* Move the data already written by other subsystems to the new pages */
53 	memcpy(page_address(pages), vdso_initdata, VDSO_NR_PAGES * PAGE_SIZE);
54 
55 	if (IS_ENABLED(CONFIG_GENERIC_GETTIMEOFDAY))
56 		vdso_k_time_data = page_address(pages + VDSO_TIME_PAGE_OFFSET);
57 
58 	if (IS_ENABLED(CONFIG_VDSO_GETRANDOM))
59 		vdso_k_rng_data = page_address(pages + VDSO_RNG_PAGE_OFFSET);
60 
61 	if (IS_ENABLED(CONFIG_ARCH_HAS_VDSO_ARCH_DATA))
62 		vdso_k_arch_data = page_address(pages + VDSO_ARCH_PAGES_START);
63 }
64 
65 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
66 			     struct vm_area_struct *vma, struct vm_fault *vmf)
67 {
68 	struct page *page, *timens_page;
69 
70 	timens_page = find_timens_vvar_page(vma);
71 
72 	switch (vmf->pgoff) {
73 	case VDSO_TIME_PAGE_OFFSET:
74 		if (!IS_ENABLED(CONFIG_GENERIC_GETTIMEOFDAY))
75 			return VM_FAULT_SIGBUS;
76 		page = virt_to_page(vdso_k_time_data);
77 		if (timens_page) {
78 			/*
79 			 * Fault in VVAR page too, since it will be accessed
80 			 * to get clock data anyway.
81 			 */
82 			unsigned long addr;
83 			vm_fault_t err;
84 
85 			addr = vmf->address + VDSO_TIMENS_PAGE_OFFSET * PAGE_SIZE;
86 			err = vmf_insert_page(vma, addr, page);
87 			if (unlikely(err & VM_FAULT_ERROR))
88 				return err;
89 			page = timens_page;
90 		}
91 		break;
92 	case VDSO_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 (!IS_ENABLED(CONFIG_TIME_NS) || !timens_page)
101 			return VM_FAULT_SIGBUS;
102 		page = virt_to_page(vdso_k_time_data);
103 		break;
104 	case VDSO_RNG_PAGE_OFFSET:
105 		if (!IS_ENABLED(CONFIG_VDSO_GETRANDOM))
106 			return VM_FAULT_SIGBUS;
107 		page = virt_to_page(vdso_k_rng_data);
108 		break;
109 	case VDSO_ARCH_PAGES_START ... VDSO_ARCH_PAGES_END:
110 		if (!IS_ENABLED(CONFIG_ARCH_HAS_VDSO_ARCH_DATA))
111 			return VM_FAULT_SIGBUS;
112 		page = virt_to_page(vdso_k_arch_data) + vmf->pgoff - VDSO_ARCH_PAGES_START;
113 		break;
114 	default:
115 		return VM_FAULT_SIGBUS;
116 	}
117 
118 	get_page(page);
119 	vmf->page = page;
120 	return 0;
121 }
122 
123 const struct vm_special_mapping vdso_vvar_mapping = {
124 	.name	= "[vvar]",
125 	.fault	= vvar_fault,
126 };
127 
128 struct vm_area_struct *vdso_install_vvar_mapping(struct mm_struct *mm, unsigned long addr)
129 {
130 	return _install_special_mapping(mm, addr, VDSO_NR_PAGES * PAGE_SIZE,
131 					VM_READ | VM_MAYREAD | VM_IO | VM_DONTDUMP |
132 					VM_MIXEDMAP | VM_SEALED_SYSMAP,
133 					&vdso_vvar_mapping);
134 }
135 
136 #ifdef CONFIG_TIME_NS
137 /*
138  * The vvar page layout depends on whether a task belongs to the root or
139  * non-root time namespace. Whenever a task changes its namespace, the VVAR
140  * page tables are cleared and then they will be re-faulted with a
141  * corresponding layout.
142  * See also the comment near timens_setup_vdso_clock_data() for details.
143  */
144 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
145 {
146 	struct mm_struct *mm = task->mm;
147 	struct vm_area_struct *vma;
148 	VMA_ITERATOR(vmi, mm, 0);
149 
150 	mmap_read_lock(mm);
151 	for_each_vma(vmi, vma) {
152 		if (vma_is_special_mapping(vma, &vdso_vvar_mapping))
153 			zap_vma_pages(vma);
154 	}
155 	mmap_read_unlock(mm);
156 
157 	return 0;
158 }
159 #endif
160