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