xref: /openbmc/linux/kernel/kexec.c (revision f9834f18)
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 
22 static int copy_user_segment_list(struct kimage *image,
23 				  unsigned long nr_segments,
24 				  struct kexec_segment __user *segments)
25 {
26 	int ret;
27 	size_t segment_bytes;
28 
29 	/* Read in the segments */
30 	image->nr_segments = nr_segments;
31 	segment_bytes = nr_segments * sizeof(*segments);
32 	ret = copy_from_user(image->segment, segments, segment_bytes);
33 	if (ret)
34 		ret = -EFAULT;
35 
36 	return ret;
37 }
38 
39 static int kimage_alloc_init(struct kimage **rimage, unsigned long entry,
40 			     unsigned long nr_segments,
41 			     struct kexec_segment __user *segments,
42 			     unsigned long flags)
43 {
44 	int ret;
45 	struct kimage *image;
46 	bool kexec_on_panic = flags & KEXEC_ON_CRASH;
47 
48 	if (kexec_on_panic) {
49 		/* Verify we have a valid entry point */
50 		if ((entry < phys_to_boot_phys(crashk_res.start)) ||
51 		    (entry > phys_to_boot_phys(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;
99 out_free_control_pages:
100 	kimage_free_page_list(&image->control_pages);
101 out_free_image:
102 	kfree(image);
103 	return ret;
104 }
105 
106 static int do_kexec_load(unsigned long entry, unsigned long nr_segments,
107 		struct kexec_segment __user *segments, unsigned long flags)
108 {
109 	struct kimage **dest_image, *image;
110 	unsigned long i;
111 	int ret;
112 
113 	if (flags & KEXEC_ON_CRASH) {
114 		dest_image = &kexec_crash_image;
115 		if (kexec_crash_image)
116 			arch_kexec_unprotect_crashkres();
117 	} else {
118 		dest_image = &kexec_image;
119 	}
120 
121 	if (nr_segments == 0) {
122 		/* Uninstall image */
123 		kimage_free(xchg(dest_image, NULL));
124 		return 0;
125 	}
126 	if (flags & KEXEC_ON_CRASH) {
127 		/*
128 		 * Loading another kernel to switch to if this one
129 		 * crashes.  Free any current crash dump kernel before
130 		 * we corrupt it.
131 		 */
132 		kimage_free(xchg(&kexec_crash_image, NULL));
133 	}
134 
135 	ret = kimage_alloc_init(&image, entry, nr_segments, segments, flags);
136 	if (ret)
137 		return ret;
138 
139 	if (flags & KEXEC_PRESERVE_CONTEXT)
140 		image->preserve_context = 1;
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, &image->segment[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 
169 out:
170 	if ((flags & KEXEC_ON_CRASH) && kexec_crash_image)
171 		arch_kexec_protect_crashkres();
172 
173 	kimage_free(image);
174 	return ret;
175 }
176 
177 /*
178  * Exec Kernel system call: for obvious reasons only root may call it.
179  *
180  * This call breaks up into three pieces.
181  * - A generic part which loads the new kernel from the current
182  *   address space, and very carefully places the data in the
183  *   allocated pages.
184  *
185  * - A generic part that interacts with the kernel and tells all of
186  *   the devices to shut down.  Preventing on-going dmas, and placing
187  *   the devices in a consistent state so a later kernel can
188  *   reinitialize them.
189  *
190  * - A machine specific part that includes the syscall number
191  *   and then copies the image to it's final destination.  And
192  *   jumps into the image at entry.
193  *
194  * kexec does not sync, or unmount filesystems so if you need
195  * that to happen you need to do that yourself.
196  */
197 
198 static inline int kexec_load_check(unsigned long nr_segments,
199 				   unsigned long flags)
200 {
201 	int result;
202 
203 	/* We only trust the superuser with rebooting the system. */
204 	if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
205 		return -EPERM;
206 
207 	/* Permit LSMs and IMA to fail the kexec */
208 	result = security_kernel_load_data(LOADING_KEXEC_IMAGE);
209 	if (result < 0)
210 		return result;
211 
212 	/*
213 	 * kexec can be used to circumvent module loading restrictions, so
214 	 * prevent loading in that case
215 	 */
216 	result = security_locked_down(LOCKDOWN_KEXEC);
217 	if (result)
218 		return result;
219 
220 	/*
221 	 * Verify we have a legal set of flags
222 	 * This leaves us room for future extensions.
223 	 */
224 	if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK))
225 		return -EINVAL;
226 
227 	/* Put an artificial cap on the number
228 	 * of segments passed to kexec_load.
229 	 */
230 	if (nr_segments > KEXEC_SEGMENT_MAX)
231 		return -EINVAL;
232 
233 	return 0;
234 }
235 
236 SYSCALL_DEFINE4(kexec_load, unsigned long, entry, unsigned long, nr_segments,
237 		struct kexec_segment __user *, segments, unsigned long, flags)
238 {
239 	int result;
240 
241 	result = kexec_load_check(nr_segments, flags);
242 	if (result)
243 		return result;
244 
245 	/* Verify we are on the appropriate architecture */
246 	if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) &&
247 		((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))
248 		return -EINVAL;
249 
250 	/* Because we write directly to the reserved memory
251 	 * region when loading crash kernels we need a mutex here to
252 	 * prevent multiple crash  kernels from attempting to load
253 	 * simultaneously, and to prevent a crash kernel from loading
254 	 * over the top of a in use crash kernel.
255 	 *
256 	 * KISS: always take the mutex.
257 	 */
258 	if (!mutex_trylock(&kexec_mutex))
259 		return -EBUSY;
260 
261 	result = do_kexec_load(entry, nr_segments, segments, flags);
262 
263 	mutex_unlock(&kexec_mutex);
264 
265 	return result;
266 }
267 
268 #ifdef CONFIG_COMPAT
269 COMPAT_SYSCALL_DEFINE4(kexec_load, compat_ulong_t, entry,
270 		       compat_ulong_t, nr_segments,
271 		       struct compat_kexec_segment __user *, segments,
272 		       compat_ulong_t, flags)
273 {
274 	struct compat_kexec_segment in;
275 	struct kexec_segment out, __user *ksegments;
276 	unsigned long i, result;
277 
278 	result = kexec_load_check(nr_segments, flags);
279 	if (result)
280 		return result;
281 
282 	/* Don't allow clients that don't understand the native
283 	 * architecture to do anything.
284 	 */
285 	if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT)
286 		return -EINVAL;
287 
288 	ksegments = compat_alloc_user_space(nr_segments * sizeof(out));
289 	for (i = 0; i < nr_segments; i++) {
290 		result = copy_from_user(&in, &segments[i], sizeof(in));
291 		if (result)
292 			return -EFAULT;
293 
294 		out.buf   = compat_ptr(in.buf);
295 		out.bufsz = in.bufsz;
296 		out.mem   = in.mem;
297 		out.memsz = in.memsz;
298 
299 		result = copy_to_user(&ksegments[i], &out, sizeof(out));
300 		if (result)
301 			return -EFAULT;
302 	}
303 
304 	/* Because we write directly to the reserved memory
305 	 * region when loading crash kernels we need a mutex here to
306 	 * prevent multiple crash  kernels from attempting to load
307 	 * simultaneously, and to prevent a crash kernel from loading
308 	 * over the top of a in use crash kernel.
309 	 *
310 	 * KISS: always take the mutex.
311 	 */
312 	if (!mutex_trylock(&kexec_mutex))
313 		return -EBUSY;
314 
315 	result = do_kexec_load(entry, nr_segments, ksegments, flags);
316 
317 	mutex_unlock(&kexec_mutex);
318 
319 	return result;
320 }
321 #endif
322