xref: /openbmc/linux/kernel/kexec.c (revision eb3fcf00)
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 #include <linux/capability.h>
10 #include <linux/mm.h>
11 #include <linux/file.h>
12 #include <linux/kexec.h>
13 #include <linux/mutex.h>
14 #include <linux/list.h>
15 #include <linux/syscalls.h>
16 #include <linux/vmalloc.h>
17 #include <linux/slab.h>
18 
19 #include "kexec_internal.h"
20 
21 static int copy_user_segment_list(struct kimage *image,
22 				  unsigned long nr_segments,
23 				  struct kexec_segment __user *segments)
24 {
25 	int ret;
26 	size_t segment_bytes;
27 
28 	/* Read in the segments */
29 	image->nr_segments = nr_segments;
30 	segment_bytes = nr_segments * sizeof(*segments);
31 	ret = copy_from_user(image->segment, segments, segment_bytes);
32 	if (ret)
33 		ret = -EFAULT;
34 
35 	return ret;
36 }
37 
38 static int kimage_alloc_init(struct kimage **rimage, unsigned long entry,
39 			     unsigned long nr_segments,
40 			     struct kexec_segment __user *segments,
41 			     unsigned long flags)
42 {
43 	int ret;
44 	struct kimage *image;
45 	bool kexec_on_panic = flags & KEXEC_ON_CRASH;
46 
47 	if (kexec_on_panic) {
48 		/* Verify we have a valid entry point */
49 		if ((entry < crashk_res.start) || (entry > crashk_res.end))
50 			return -EADDRNOTAVAIL;
51 	}
52 
53 	/* Allocate and initialize a controlling structure */
54 	image = do_kimage_alloc_init();
55 	if (!image)
56 		return -ENOMEM;
57 
58 	image->start = entry;
59 
60 	ret = copy_user_segment_list(image, nr_segments, segments);
61 	if (ret)
62 		goto out_free_image;
63 
64 	ret = sanity_check_segment_list(image);
65 	if (ret)
66 		goto out_free_image;
67 
68 	 /* Enable the special crash kernel control page allocation policy. */
69 	if (kexec_on_panic) {
70 		image->control_page = crashk_res.start;
71 		image->type = KEXEC_TYPE_CRASH;
72 	}
73 
74 	/*
75 	 * Find a location for the control code buffer, and add it
76 	 * the vector of segments so that it's pages will also be
77 	 * counted as destination pages.
78 	 */
79 	ret = -ENOMEM;
80 	image->control_code_page = kimage_alloc_control_pages(image,
81 					   get_order(KEXEC_CONTROL_PAGE_SIZE));
82 	if (!image->control_code_page) {
83 		pr_err("Could not allocate control_code_buffer\n");
84 		goto out_free_image;
85 	}
86 
87 	if (!kexec_on_panic) {
88 		image->swap_page = kimage_alloc_control_pages(image, 0);
89 		if (!image->swap_page) {
90 			pr_err("Could not allocate swap buffer\n");
91 			goto out_free_control_pages;
92 		}
93 	}
94 
95 	*rimage = image;
96 	return 0;
97 out_free_control_pages:
98 	kimage_free_page_list(&image->control_pages);
99 out_free_image:
100 	kfree(image);
101 	return ret;
102 }
103 
104 /*
105  * Exec Kernel system call: for obvious reasons only root may call it.
106  *
107  * This call breaks up into three pieces.
108  * - A generic part which loads the new kernel from the current
109  *   address space, and very carefully places the data in the
110  *   allocated pages.
111  *
112  * - A generic part that interacts with the kernel and tells all of
113  *   the devices to shut down.  Preventing on-going dmas, and placing
114  *   the devices in a consistent state so a later kernel can
115  *   reinitialize them.
116  *
117  * - A machine specific part that includes the syscall number
118  *   and then copies the image to it's final destination.  And
119  *   jumps into the image at entry.
120  *
121  * kexec does not sync, or unmount filesystems so if you need
122  * that to happen you need to do that yourself.
123  */
124 
125 SYSCALL_DEFINE4(kexec_load, unsigned long, entry, unsigned long, nr_segments,
126 		struct kexec_segment __user *, segments, unsigned long, flags)
127 {
128 	struct kimage **dest_image, *image;
129 	int result;
130 
131 	/* We only trust the superuser with rebooting the system. */
132 	if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
133 		return -EPERM;
134 
135 	/*
136 	 * Verify we have a legal set of flags
137 	 * This leaves us room for future extensions.
138 	 */
139 	if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK))
140 		return -EINVAL;
141 
142 	/* Verify we are on the appropriate architecture */
143 	if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) &&
144 		((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))
145 		return -EINVAL;
146 
147 	/* Put an artificial cap on the number
148 	 * of segments passed to kexec_load.
149 	 */
150 	if (nr_segments > KEXEC_SEGMENT_MAX)
151 		return -EINVAL;
152 
153 	image = NULL;
154 	result = 0;
155 
156 	/* Because we write directly to the reserved memory
157 	 * region when loading crash kernels we need a mutex here to
158 	 * prevent multiple crash  kernels from attempting to load
159 	 * simultaneously, and to prevent a crash kernel from loading
160 	 * over the top of a in use crash kernel.
161 	 *
162 	 * KISS: always take the mutex.
163 	 */
164 	if (!mutex_trylock(&kexec_mutex))
165 		return -EBUSY;
166 
167 	dest_image = &kexec_image;
168 	if (flags & KEXEC_ON_CRASH)
169 		dest_image = &kexec_crash_image;
170 	if (nr_segments > 0) {
171 		unsigned long i;
172 
173 		if (flags & KEXEC_ON_CRASH) {
174 			/*
175 			 * Loading another kernel to switch to if this one
176 			 * crashes.  Free any current crash dump kernel before
177 			 * we corrupt it.
178 			 */
179 
180 			kimage_free(xchg(&kexec_crash_image, NULL));
181 			result = kimage_alloc_init(&image, entry, nr_segments,
182 						   segments, flags);
183 			crash_map_reserved_pages();
184 		} else {
185 			/* Loading another kernel to reboot into. */
186 
187 			result = kimage_alloc_init(&image, entry, nr_segments,
188 						   segments, flags);
189 		}
190 		if (result)
191 			goto out;
192 
193 		if (flags & KEXEC_PRESERVE_CONTEXT)
194 			image->preserve_context = 1;
195 		result = machine_kexec_prepare(image);
196 		if (result)
197 			goto out;
198 
199 		for (i = 0; i < nr_segments; i++) {
200 			result = kimage_load_segment(image, &image->segment[i]);
201 			if (result)
202 				goto out;
203 		}
204 		kimage_terminate(image);
205 		if (flags & KEXEC_ON_CRASH)
206 			crash_unmap_reserved_pages();
207 	}
208 	/* Install the new kernel, and  Uninstall the old */
209 	image = xchg(dest_image, image);
210 
211 out:
212 	mutex_unlock(&kexec_mutex);
213 	kimage_free(image);
214 
215 	return result;
216 }
217 
218 #ifdef CONFIG_COMPAT
219 COMPAT_SYSCALL_DEFINE4(kexec_load, compat_ulong_t, entry,
220 		       compat_ulong_t, nr_segments,
221 		       struct compat_kexec_segment __user *, segments,
222 		       compat_ulong_t, flags)
223 {
224 	struct compat_kexec_segment in;
225 	struct kexec_segment out, __user *ksegments;
226 	unsigned long i, result;
227 
228 	/* Don't allow clients that don't understand the native
229 	 * architecture to do anything.
230 	 */
231 	if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT)
232 		return -EINVAL;
233 
234 	if (nr_segments > KEXEC_SEGMENT_MAX)
235 		return -EINVAL;
236 
237 	ksegments = compat_alloc_user_space(nr_segments * sizeof(out));
238 	for (i = 0; i < nr_segments; i++) {
239 		result = copy_from_user(&in, &segments[i], sizeof(in));
240 		if (result)
241 			return -EFAULT;
242 
243 		out.buf   = compat_ptr(in.buf);
244 		out.bufsz = in.bufsz;
245 		out.mem   = in.mem;
246 		out.memsz = in.memsz;
247 
248 		result = copy_to_user(&ksegments[i], &out, sizeof(out));
249 		if (result)
250 			return -EFAULT;
251 	}
252 
253 	return sys_kexec_load(entry, nr_segments, ksegments, flags);
254 }
255 #endif
256