1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * kexec.c - kexec_load system call
4 * Copyright (C) 2002-2004 Eric Biederman <ebiederm@xmission.com>
5 */
6
7#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
9#include <linux/capability.h>
10#include <linux/mm.h>
11#include <linux/file.h>
12#include <linux/security.h>
13#include <linux/kexec.h>
14#include <linux/mutex.h>
15#include <linux/list.h>
16#include <linux/syscalls.h>
17#include <linux/vmalloc.h>
18#include <linux/slab.h>
19
20#include "kexec_internal.h"
21
22static int kimage_alloc_init(struct kimage **rimage, unsigned long entry,
23 unsigned long nr_segments,
24 struct kexec_segment *segments,
25 unsigned long flags)
26{
27 int ret;
28 struct kimage *image;
29 bool kexec_on_panic = flags & KEXEC_ON_CRASH;
30
31#ifdef CONFIG_CRASH_DUMP
32 if (kexec_on_panic) {
33 /* Verify we have a valid entry point */
34 if ((entry < phys_to_boot_phys(phys: crashk_res.start)) ||
35 (entry > phys_to_boot_phys(phys: crashk_res.end)))
36 return -EADDRNOTAVAIL;
37 }
38#endif
39
40 /* Allocate and initialize a controlling structure */
41 image = do_kimage_alloc_init();
42 if (!image)
43 return -ENOMEM;
44
45 image->start = entry;
46 image->nr_segments = nr_segments;
47 memcpy(to: image->segment, from: segments, len: nr_segments * sizeof(*segments));
48
49#ifdef CONFIG_CRASH_DUMP
50 if (kexec_on_panic) {
51 /* Enable special crash kernel control page alloc policy. */
52 image->control_page = crashk_res.start;
53 image->type = KEXEC_TYPE_CRASH;
54 }
55#endif
56
57 ret = sanity_check_segment_list(image);
58 if (ret)
59 goto out_free_image;
60
61 /*
62 * Find a location for the control code buffer, and add it
63 * the vector of segments so that it's pages will also be
64 * counted as destination pages.
65 */
66 ret = -ENOMEM;
67 image->control_code_page = kimage_alloc_control_pages(image,
68 order: get_order(KEXEC_CONTROL_PAGE_SIZE));
69 if (!image->control_code_page) {
70 pr_err("Could not allocate control_code_buffer\n");
71 goto out_free_image;
72 }
73
74 if (!kexec_on_panic) {
75 image->swap_page = kimage_alloc_control_pages(image, order: 0);
76 if (!image->swap_page) {
77 pr_err("Could not allocate swap buffer\n");
78 goto out_free_control_pages;
79 }
80 }
81
82 *rimage = image;
83 return 0;
84out_free_control_pages:
85 kimage_free_page_list(list: &image->control_pages);
86out_free_image:
87 kfree(objp: image);
88 return ret;
89}
90
91static int do_kexec_load(unsigned long entry, unsigned long nr_segments,
92 struct kexec_segment *segments, unsigned long flags)
93{
94 struct kimage **dest_image, *image;
95 unsigned long i;
96 int ret;
97
98 /*
99 * Because we write directly to the reserved memory region when loading
100 * crash kernels we need a serialization here to prevent multiple crash
101 * kernels from attempting to load simultaneously.
102 */
103 if (!kexec_trylock())
104 return -EBUSY;
105
106#ifdef CONFIG_CRASH_DUMP
107 if (flags & KEXEC_ON_CRASH) {
108 dest_image = &kexec_crash_image;
109 if (kexec_crash_image)
110 arch_kexec_unprotect_crashkres();
111 } else
112#endif
113 dest_image = &kexec_image;
114
115 if (nr_segments == 0) {
116 /* Uninstall image */
117 kimage_free(xchg(dest_image, NULL));
118 ret = 0;
119 goto out_unlock;
120 }
121 if (flags & KEXEC_ON_CRASH) {
122 /*
123 * Loading another kernel to switch to if this one
124 * crashes. Free any current crash dump kernel before
125 * we corrupt it.
126 */
127 kimage_free(xchg(&kexec_crash_image, NULL));
128 }
129
130 ret = kimage_alloc_init(rimage: &image, entry, nr_segments, segments, flags);
131 if (ret)
132 goto out_unlock;
133
134 if (flags & KEXEC_PRESERVE_CONTEXT)
135 image->preserve_context = 1;
136
137#ifdef CONFIG_CRASH_HOTPLUG
138 if ((flags & KEXEC_ON_CRASH) && arch_crash_hotplug_support(image, kexec_flags: flags))
139 image->hotplug_support = 1;
140#endif
141
142 ret = machine_kexec_prepare(image);
143 if (ret)
144 goto out;
145
146 /*
147 * Some architecture(like S390) may touch the crash memory before
148 * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
149 */
150 ret = kimage_crash_copy_vmcoreinfo(image);
151 if (ret)
152 goto out;
153
154 for (i = 0; i < nr_segments; i++) {
155 ret = kimage_load_segment(image, idx: i);
156 if (ret)
157 goto out;
158 }
159
160 kimage_terminate(image);
161
162 ret = machine_kexec_post_load(image);
163 if (ret)
164 goto out;
165
166 /* Install the new kernel and uninstall the old */
167 image = xchg(dest_image, image);
168
169out:
170#ifdef CONFIG_CRASH_DUMP
171 if ((flags & KEXEC_ON_CRASH) && kexec_crash_image)
172 arch_kexec_protect_crashkres();
173#endif
174
175 kimage_free(image);
176out_unlock:
177 kexec_unlock();
178 return ret;
179}
180
181/*
182 * Exec Kernel system call: for obvious reasons only root may call it.
183 *
184 * This call breaks up into three pieces.
185 * - A generic part which loads the new kernel from the current
186 * address space, and very carefully places the data in the
187 * allocated pages.
188 *
189 * - A generic part that interacts with the kernel and tells all of
190 * the devices to shut down. Preventing on-going dmas, and placing
191 * the devices in a consistent state so a later kernel can
192 * reinitialize them.
193 *
194 * - A machine specific part that includes the syscall number
195 * and then copies the image to it's final destination. And
196 * jumps into the image at entry.
197 *
198 * kexec does not sync, or unmount filesystems so if you need
199 * that to happen you need to do that yourself.
200 */
201
202static inline int kexec_load_check(unsigned long nr_segments,
203 unsigned long flags)
204{
205 int image_type = (flags & KEXEC_ON_CRASH) ?
206 KEXEC_TYPE_CRASH : KEXEC_TYPE_DEFAULT;
207 int result;
208
209 /* We only trust the superuser with rebooting the system. */
210 if (!kexec_load_permitted(kexec_image_type: image_type))
211 return -EPERM;
212
213 /* Permit LSMs and IMA to fail the kexec */
214 result = security_kernel_load_data(id: LOADING_KEXEC_IMAGE, contents: false);
215 if (result < 0)
216 return result;
217
218 /*
219 * kexec can be used to circumvent module loading restrictions, so
220 * prevent loading in that case
221 */
222 result = security_locked_down(what: LOCKDOWN_KEXEC);
223 if (result)
224 return result;
225
226 /*
227 * Verify we have a legal set of flags
228 * This leaves us room for future extensions.
229 */
230 if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK))
231 return -EINVAL;
232
233 /* Put an artificial cap on the number
234 * of segments passed to kexec_load.
235 */
236 if (nr_segments > KEXEC_SEGMENT_MAX)
237 return -EINVAL;
238
239 return 0;
240}
241
242SYSCALL_DEFINE4(kexec_load, unsigned long, entry, unsigned long, nr_segments,
243 struct kexec_segment __user *, segments, unsigned long, flags)
244{
245 struct kexec_segment *ksegments;
246 unsigned long result;
247
248 result = kexec_load_check(nr_segments, flags);
249 if (result)
250 return result;
251
252 /* Verify we are on the appropriate architecture */
253 if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) &&
254 ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))
255 return -EINVAL;
256
257 ksegments = memdup_array_user(src: segments, n: nr_segments, size: sizeof(ksegments[0]));
258 if (IS_ERR(ptr: ksegments))
259 return PTR_ERR(ptr: ksegments);
260
261 result = do_kexec_load(entry, nr_segments, segments: ksegments, flags);
262 kfree(objp: ksegments);
263
264 return result;
265}
266
267#ifdef CONFIG_COMPAT
268COMPAT_SYSCALL_DEFINE4(kexec_load, compat_ulong_t, entry,
269 compat_ulong_t, nr_segments,
270 struct compat_kexec_segment __user *, segments,
271 compat_ulong_t, flags)
272{
273 struct compat_kexec_segment in;
274 struct kexec_segment *ksegments;
275 unsigned long i, result;
276
277 result = kexec_load_check(nr_segments, flags);
278 if (result)
279 return result;
280
281 /* Don't allow clients that don't understand the native
282 * architecture to do anything.
283 */
284 if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT)
285 return -EINVAL;
286
287 ksegments = kmalloc_array(nr_segments, sizeof(ksegments[0]),
288 GFP_KERNEL);
289 if (!ksegments)
290 return -ENOMEM;
291
292 for (i = 0; i < nr_segments; i++) {
293 result = copy_from_user(to: &in, from: &segments[i], n: sizeof(in));
294 if (result)
295 goto fail;
296
297 ksegments[i].buf = compat_ptr(uptr: in.buf);
298 ksegments[i].bufsz = in.bufsz;
299 ksegments[i].mem = in.mem;
300 ksegments[i].memsz = in.memsz;
301 }
302
303 result = do_kexec_load(entry, nr_segments, segments: ksegments, flags);
304
305fail:
306 kfree(objp: ksegments);
307 return result;
308}
309#endif
310