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 
kimage_alloc_init(struct kimage ** rimage,unsigned long entry,unsigned long nr_segments,struct kexec_segment * segments,unsigned long flags)22 static 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;
80 out_free_control_pages:
81 	kimage_free_page_list(&image->control_pages);
82 out_free_image:
83 	kfree(image);
84 	return ret;
85 }
86 
do_kexec_load(unsigned long entry,unsigned long nr_segments,struct kexec_segment * segments,unsigned long flags)87 static 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 
159 out:
160 	if ((flags & KEXEC_ON_CRASH) && kexec_crash_image)
161 		arch_kexec_protect_crashkres();
162 
163 	kimage_free(image);
164 out_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 
kexec_load_check(unsigned long nr_segments,unsigned long flags)190 static 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 
SYSCALL_DEFINE4(kexec_load,unsigned long,entry,unsigned long,nr_segments,struct kexec_segment __user *,segments,unsigned long,flags)228 SYSCALL_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
COMPAT_SYSCALL_DEFINE4(kexec_load,compat_ulong_t,entry,compat_ulong_t,nr_segments,struct compat_kexec_segment __user *,segments,compat_ulong_t,flags)254 COMPAT_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 
291 fail:
292 	kfree(ksegments);
293 	return result;
294 }
295 #endif
296