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/security.h>
15#include <linux/kexec.h>
16#include <linux/mutex.h>
17#include <linux/list.h>
18#include <linux/syscalls.h>
19#include <linux/vmalloc.h>
20#include <linux/slab.h>
21
22#include "kexec_internal.h"
23
24static int copy_user_segment_list(struct kimage *image,
25 unsigned long nr_segments,
26 struct kexec_segment __user *segments)
27{
28 int ret;
29 size_t segment_bytes;
30
31 /* Read in the segments */
32 image->nr_segments = nr_segments;
33 segment_bytes = nr_segments * sizeof(*segments);
34 ret = copy_from_user(image->segment, segments, segment_bytes);
35 if (ret)
36 ret = -EFAULT;
37
38 return ret;
39}
40
41static int kimage_alloc_init(struct kimage **rimage, unsigned long entry,
42 unsigned long nr_segments,
43 struct kexec_segment __user *segments,
44 unsigned long flags)
45{
46 int ret;
47 struct kimage *image;
48 bool kexec_on_panic = flags & KEXEC_ON_CRASH;
49
50 if (kexec_on_panic) {
51 /* Verify we have a valid entry point */
52 if ((entry < phys_to_boot_phys(crashk_res.start)) ||
53 (entry > phys_to_boot_phys(crashk_res.end)))
54 return -EADDRNOTAVAIL;
55 }
56
57 /* Allocate and initialize a controlling structure */
58 image = do_kimage_alloc_init();
59 if (!image)
60 return -ENOMEM;
61
62 image->start = entry;
63
64 ret = copy_user_segment_list(image, nr_segments, segments);
65 if (ret)
66 goto out_free_image;
67
68 if (kexec_on_panic) {
69 /* Enable special crash kernel control page alloc policy. */
70 image->control_page = crashk_res.start;
71 image->type = KEXEC_TYPE_CRASH;
72 }
73
74 ret = sanity_check_segment_list(image);
75 if (ret)
76 goto out_free_image;
77
78 /*
79 * Find a location for the control code buffer, and add it
80 * the vector of segments so that it's pages will also be
81 * counted as destination pages.
82 */
83 ret = -ENOMEM;
84 image->control_code_page = kimage_alloc_control_pages(image,
85 get_order(KEXEC_CONTROL_PAGE_SIZE));
86 if (!image->control_code_page) {
87 pr_err("Could not allocate control_code_buffer\n");
88 goto out_free_image;
89 }
90
91 if (!kexec_on_panic) {
92 image->swap_page = kimage_alloc_control_pages(image, 0);
93 if (!image->swap_page) {
94 pr_err("Could not allocate swap buffer\n");
95 goto out_free_control_pages;
96 }
97 }
98
99 *rimage = image;
100 return 0;
101out_free_control_pages:
102 kimage_free_page_list(&image->control_pages);
103out_free_image:
104 kfree(image);
105 return ret;
106}
107
108static int do_kexec_load(unsigned long entry, unsigned long nr_segments,
109 struct kexec_segment __user *segments, unsigned long flags)
110{
111 struct kimage **dest_image, *image;
112 unsigned long i;
113 int ret;
114
115 if (flags & KEXEC_ON_CRASH) {
116 dest_image = &kexec_crash_image;
117 if (kexec_crash_image)
118 arch_kexec_unprotect_crashkres();
119 } else {
120 dest_image = &kexec_image;
121 }
122
123 if (nr_segments == 0) {
124 /* Uninstall image */
125 kimage_free(xchg(dest_image, NULL));
126 return 0;
127 }
128 if (flags & KEXEC_ON_CRASH) {
129 /*
130 * Loading another kernel to switch to if this one
131 * crashes. Free any current crash dump kernel before
132 * we corrupt it.
133 */
134 kimage_free(xchg(&kexec_crash_image, NULL));
135 }
136
137 ret = kimage_alloc_init(&image, entry, nr_segments, segments, flags);
138 if (ret)
139 return ret;
140
141 if (flags & KEXEC_PRESERVE_CONTEXT)
142 image->preserve_context = 1;
143
144 ret = machine_kexec_prepare(image);
145 if (ret)
146 goto out;
147
148 /*
149 * Some architecture(like S390) may touch the crash memory before
150 * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
151 */
152 ret = kimage_crash_copy_vmcoreinfo(image);
153 if (ret)
154 goto out;
155
156 for (i = 0; i < nr_segments; i++) {
157 ret = kimage_load_segment(image, &image->segment[i]);
158 if (ret)
159 goto out;
160 }
161
162 kimage_terminate(image);
163
164 /* Install the new kernel and uninstall the old */
165 image = xchg(dest_image, image);
166
167out:
168 if ((flags & KEXEC_ON_CRASH) && kexec_crash_image)
169 arch_kexec_protect_crashkres();
170
171 kimage_free(image);
172 return ret;
173}
174
175/*
176 * Exec Kernel system call: for obvious reasons only root may call it.
177 *
178 * This call breaks up into three pieces.
179 * - A generic part which loads the new kernel from the current
180 * address space, and very carefully places the data in the
181 * allocated pages.
182 *
183 * - A generic part that interacts with the kernel and tells all of
184 * the devices to shut down. Preventing on-going dmas, and placing
185 * the devices in a consistent state so a later kernel can
186 * reinitialize them.
187 *
188 * - A machine specific part that includes the syscall number
189 * and then copies the image to it's final destination. And
190 * jumps into the image at entry.
191 *
192 * kexec does not sync, or unmount filesystems so if you need
193 * that to happen you need to do that yourself.
194 */
195
196static inline int kexec_load_check(unsigned long nr_segments,
197 unsigned long flags)
198{
199 int result;
200
201 /* We only trust the superuser with rebooting the system. */
202 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
203 return -EPERM;
204
205 /* Permit LSMs and IMA to fail the kexec */
206 result = security_kernel_load_data(LOADING_KEXEC_IMAGE);
207 if (result < 0)
208 return result;
209
210 /*
211 * Verify we have a legal set of flags
212 * This leaves us room for future extensions.
213 */
214 if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK))
215 return -EINVAL;
216
217 /* Put an artificial cap on the number
218 * of segments passed to kexec_load.
219 */
220 if (nr_segments > KEXEC_SEGMENT_MAX)
221 return -EINVAL;
222
223 return 0;
224}
225
226SYSCALL_DEFINE4(kexec_load, unsigned long, entry, unsigned long, nr_segments,
227 struct kexec_segment __user *, segments, unsigned long, flags)
228{
229 int result;
230
231 result = kexec_load_check(nr_segments, flags);
232 if (result)
233 return result;
234
235 /* Verify we are on the appropriate architecture */
236 if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) &&
237 ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))
238 return -EINVAL;
239
240 /* Because we write directly to the reserved memory
241 * region when loading crash kernels we need a mutex here to
242 * prevent multiple crash kernels from attempting to load
243 * simultaneously, and to prevent a crash kernel from loading
244 * over the top of a in use crash kernel.
245 *
246 * KISS: always take the mutex.
247 */
248 if (!mutex_trylock(&kexec_mutex))
249 return -EBUSY;
250
251 result = do_kexec_load(entry, nr_segments, segments, flags);
252
253 mutex_unlock(&kexec_mutex);
254
255 return result;
256}
257
258#ifdef CONFIG_COMPAT
259COMPAT_SYSCALL_DEFINE4(kexec_load, compat_ulong_t, entry,
260 compat_ulong_t, nr_segments,
261 struct compat_kexec_segment __user *, segments,
262 compat_ulong_t, flags)
263{
264 struct compat_kexec_segment in;
265 struct kexec_segment out, __user *ksegments;
266 unsigned long i, result;
267
268 result = kexec_load_check(nr_segments, flags);
269 if (result)
270 return result;
271
272 /* Don't allow clients that don't understand the native
273 * architecture to do anything.
274 */
275 if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT)
276 return -EINVAL;
277
278 ksegments = compat_alloc_user_space(nr_segments * sizeof(out));
279 for (i = 0; i < nr_segments; i++) {
280 result = copy_from_user(&in, &segments[i], sizeof(in));
281 if (result)
282 return -EFAULT;
283
284 out.buf = compat_ptr(in.buf);
285 out.bufsz = in.bufsz;
286 out.mem = in.mem;
287 out.memsz = in.memsz;
288
289 result = copy_to_user(&ksegments[i], &out, sizeof(out));
290 if (result)
291 return -EFAULT;
292 }
293
294 /* Because we write directly to the reserved memory
295 * region when loading crash kernels we need a mutex here to
296 * prevent multiple crash kernels from attempting to load
297 * simultaneously, and to prevent a crash kernel from loading
298 * over the top of a in use crash kernel.
299 *
300 * KISS: always take the mutex.
301 */
302 if (!mutex_trylock(&kexec_mutex))
303 return -EBUSY;
304
305 result = do_kexec_load(entry, nr_segments, ksegments, flags);
306
307 mutex_unlock(&kexec_mutex);
308
309 return result;
310}
311#endif
312