xref: /linux/mm/vma_init.c (revision 18fbf5151d2c0bfe433c7428eef03cabf5fdb2fa)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 
3 /*
4  * Functions for initializing, allocating, freeing and duplicating VMAs. Shared
5  * between CONFIG_MMU and non-CONFIG_MMU kernel configurations.
6  */
7 
8 /*
9  * To allow for userland testing we place internal dependencies in
10  * vma_internal.h and external VMA API declarations in vma.h.
11  */
12 #include "vma_internal.h"
13 #include "vma.h"
14 
15 /* SLAB cache for vm_area_struct structures */
16 static struct kmem_cache *vm_area_cachep;
17 
18 void __init vma_state_init(void)
19 {
20 	struct kmem_cache_args args = {
21 		.use_freeptr_offset = true,
22 		.freeptr_offset = offsetof(struct vm_area_struct, vm_freeptr),
23 		.sheaf_capacity = 32,
24 	};
25 
26 	vm_area_cachep = kmem_cache_create("vm_area_struct",
27 			sizeof(struct vm_area_struct), &args,
28 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
29 			SLAB_ACCOUNT);
30 }
31 
32 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
33 {
34 	struct vm_area_struct *vma;
35 
36 	vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
37 	if (!vma)
38 		return NULL;
39 
40 	vma_init(vma, mm);
41 
42 	return vma;
43 }
44 
45 static void vm_area_init_from(const struct vm_area_struct *src,
46 			      struct vm_area_struct *dest)
47 {
48 	dest->vm_mm = src->vm_mm;
49 	dest->vm_ops = src->vm_ops;
50 	dest->vm_start = src->vm_start;
51 	dest->vm_end = src->vm_end;
52 	dest->anon_vma = src->anon_vma;
53 	dest->vm_pgoff = vma_start_pgoff(src);
54 	__vma_set_anon_pgoff(dest, vma_start_anon_pgoff(src));
55 	dest->vm_file = src->vm_file;
56 	dest->vm_private_data = src->vm_private_data;
57 	vm_flags_init(dest, src->vm_flags);
58 	memcpy(&dest->vm_page_prot, &src->vm_page_prot,
59 	       sizeof(dest->vm_page_prot));
60 	/*
61 	 * src->shared.rb may be modified concurrently when called from
62 	 * dup_mmap(), but the clone will reinitialize it.
63 	 */
64 	data_race(memcpy(&dest->shared, &src->shared, sizeof(dest->shared)));
65 	memcpy(&dest->vm_userfaultfd_ctx, &src->vm_userfaultfd_ctx,
66 	       sizeof(dest->vm_userfaultfd_ctx));
67 #ifdef CONFIG_ANON_VMA_NAME
68 	dest->anon_name = src->anon_name;
69 #endif
70 #ifdef CONFIG_SWAP
71 	memcpy(&dest->swap_readahead_info, &src->swap_readahead_info,
72 	       sizeof(dest->swap_readahead_info));
73 #endif
74 #ifndef CONFIG_MMU
75 	dest->vm_region = src->vm_region;
76 #endif
77 #ifdef CONFIG_NUMA
78 	dest->vm_policy = src->vm_policy;
79 #endif
80 #ifdef __HAVE_PFNMAP_TRACKING
81 	dest->pfnmap_track_ctx = NULL;
82 #endif
83 }
84 
85 #ifdef __HAVE_PFNMAP_TRACKING
86 static inline int vma_pfnmap_track_ctx_dup(struct vm_area_struct *orig,
87 		struct vm_area_struct *new)
88 {
89 	struct pfnmap_track_ctx *ctx = orig->pfnmap_track_ctx;
90 
91 	if (likely(!ctx))
92 		return 0;
93 
94 	/*
95 	 * We don't expect to ever hit this. If ever required, we would have
96 	 * to duplicate the tracking.
97 	 */
98 	if (unlikely(kref_read(&ctx->kref) >= REFCOUNT_MAX))
99 		return -ENOMEM;
100 	kref_get(&ctx->kref);
101 	new->pfnmap_track_ctx = ctx;
102 	return 0;
103 }
104 
105 static inline void vma_pfnmap_track_ctx_release(struct vm_area_struct *vma)
106 {
107 	struct pfnmap_track_ctx *ctx = vma->pfnmap_track_ctx;
108 
109 	if (likely(!ctx))
110 		return;
111 
112 	kref_put(&ctx->kref, pfnmap_track_ctx_release);
113 	vma->pfnmap_track_ctx = NULL;
114 }
115 #else
116 static inline int vma_pfnmap_track_ctx_dup(struct vm_area_struct *orig,
117 		struct vm_area_struct *new)
118 {
119 	return 0;
120 }
121 static inline void vma_pfnmap_track_ctx_release(struct vm_area_struct *vma)
122 {
123 }
124 #endif
125 
126 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
127 {
128 	struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
129 
130 	if (!new)
131 		return NULL;
132 
133 	ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
134 	ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
135 	vm_area_init_from(orig, new);
136 
137 	if (vma_pfnmap_track_ctx_dup(orig, new)) {
138 		kmem_cache_free(vm_area_cachep, new);
139 		return NULL;
140 	}
141 	vma_lock_init(new, true);
142 	INIT_LIST_HEAD(&new->anon_vma_chain);
143 	vma_numab_state_init(new);
144 	dup_anon_vma_name(orig, new);
145 
146 	return new;
147 }
148 
149 void vm_area_free(struct vm_area_struct *vma)
150 {
151 	/* The vma should be detached while being destroyed. */
152 	vma_assert_detached(vma);
153 	vma_numab_state_free(vma);
154 	free_anon_vma_name(vma);
155 	vma_pfnmap_track_ctx_release(vma);
156 	kmem_cache_free(vm_area_cachep, vma);
157 }
158