1// SPDX-License-Identifier: GPL-2.0-only
2
3/*
4 * Functions explicitly implemented for exec functionality which however are
5 * explicitly VMA-only logic.
6 */
7
8#include "vma_internal.h"
9#include "vma.h"
10
11/*
12 * Relocate a VMA downwards by shift bytes. There cannot be any VMAs between
13 * this VMA and its relocated range, which will now reside at [vma->vm_start -
14 * shift, vma->vm_end - shift).
15 *
16 * This function is almost certainly NOT what you want for anything other than
17 * early executable temporary stack relocation.
18 */
19int relocate_vma_down(struct vm_area_struct *vma, unsigned long shift)
20{
21 /*
22 * The process proceeds as follows:
23 *
24 * 1) Use shift to calculate the new vma endpoints.
25 * 2) Extend vma to cover both the old and new ranges. This ensures the
26 * arguments passed to subsequent functions are consistent.
27 * 3) Move vma's page tables to the new range.
28 * 4) Free up any cleared pgd range.
29 * 5) Shrink the vma to cover only the new range.
30 */
31
32 struct mm_struct *mm = vma->vm_mm;
33 unsigned long old_start = vma->vm_start;
34 unsigned long old_end = vma->vm_end;
35 unsigned long length = old_end - old_start;
36 unsigned long new_start = old_start - shift;
37 unsigned long new_end = old_end - shift;
38 VMA_ITERATOR(vmi, mm, new_start);
39 VMG_STATE(vmg, mm, &vmi, new_start, old_end, 0, vma->vm_pgoff);
40 struct vm_area_struct *next;
41 struct mmu_gather tlb;
42 PAGETABLE_MOVE(pmc, vma, vma, old_start, new_start, length);
43
44 BUG_ON(new_start > new_end);
45
46 /*
47 * ensure there are no vmas between where we want to go
48 * and where we are
49 */
50 if (vma != vma_next(vmi: &vmi))
51 return -EFAULT;
52
53 vma_iter_prev_range(vmi: &vmi);
54 /*
55 * cover the whole range: [new_start, old_end)
56 */
57 vmg.target = vma;
58 if (vma_expand(vmg: &vmg))
59 return -ENOMEM;
60
61 /*
62 * move the page tables downwards, on failure we rely on
63 * process cleanup to remove whatever mess we made.
64 */
65 pmc.for_stack = true;
66 if (length != move_page_tables(pmc: &pmc))
67 return -ENOMEM;
68
69 tlb_gather_mmu(tlb: &tlb, mm);
70 next = vma_next(vmi: &vmi);
71 if (new_end > old_start) {
72 /*
73 * when the old and new regions overlap clear from new_end.
74 */
75 free_pgd_range(tlb: &tlb, addr: new_end, end: old_end, floor: new_end,
76 ceiling: next ? next->vm_start : USER_PGTABLES_CEILING);
77 } else {
78 /*
79 * otherwise, clean from old_start; this is done to not touch
80 * the address space in [new_end, old_start) some architectures
81 * have constraints on va-space that make this illegal (IA64) -
82 * for the others its just a little faster.
83 */
84 free_pgd_range(tlb: &tlb, addr: old_start, end: old_end, floor: new_end,
85 ceiling: next ? next->vm_start : USER_PGTABLES_CEILING);
86 }
87 tlb_finish_mmu(tlb: &tlb);
88
89 vma_prev(vmi: &vmi);
90 /* Shrink the vma to just the new range */
91 return vma_shrink(vmi: &vmi, vma, start: new_start, end: new_end, pgoff: vma->vm_pgoff);
92}
93
94/*
95 * Establish the stack VMA in an execve'd process, located temporarily at the
96 * maximum stack address provided by the architecture.
97 *
98 * We later relocate this downwards in relocate_vma_down().
99 *
100 * This function is almost certainly NOT what you want for anything other than
101 * early executable initialisation.
102 *
103 * On success, returns 0 and sets *vmap to the stack VMA and *top_mem_p to the
104 * maximum addressable location in the stack (that is capable of storing a
105 * system word of data).
106 */
107int create_init_stack_vma(struct mm_struct *mm, struct vm_area_struct **vmap,
108 unsigned long *top_mem_p)
109{
110 int err;
111 struct vm_area_struct *vma = vm_area_alloc(mm);
112
113 if (!vma)
114 return -ENOMEM;
115
116 vma_set_anonymous(vma);
117
118 if (mmap_write_lock_killable(mm)) {
119 err = -EINTR;
120 goto err_free;
121 }
122
123 /*
124 * Need to be called with mmap write lock
125 * held, to avoid race with ksmd.
126 */
127 err = ksm_execve(mm);
128 if (err)
129 goto err_ksm;
130
131 /*
132 * Place the stack at the largest stack address the architecture
133 * supports. Later, we'll move this to an appropriate place. We don't
134 * use STACK_TOP because that can depend on attributes which aren't
135 * configured yet.
136 */
137 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
138 vma->vm_end = STACK_TOP_MAX;
139 vma->vm_start = vma->vm_end - PAGE_SIZE;
140 vm_flags_init(vma, VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP);
141 vma->vm_page_prot = vm_get_page_prot(vm_flags: vma->vm_flags);
142
143 err = insert_vm_struct(mm, vma);
144 if (err)
145 goto err;
146
147 mm->stack_vm = mm->total_vm = 1;
148 mmap_write_unlock(mm);
149 *vmap = vma;
150 *top_mem_p = vma->vm_end - sizeof(void *);
151 return 0;
152
153err:
154 ksm_exit(mm);
155err_ksm:
156 mmap_write_unlock(mm);
157err_free:
158 *vmap = NULL;
159 vm_area_free(vma);
160 return err;
161}
162