Linux Audio

Check our new training course

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 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
 39static 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;
 99out_free_control_pages:
100	kimage_free_page_list(&image->control_pages);
101out_free_image:
102	kfree(image);
103	return ret;
104}
105
106static 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
169out:
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
198static 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
236SYSCALL_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
269COMPAT_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
  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