xref: /linux/kernel/liveupdate/luo_core.c (revision d50569612c29215c5d1c64f47a65604ed265d2e9)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 /*
4  * Copyright (c) 2025, Google LLC.
5  * Pasha Tatashin <pasha.tatashin@soleen.com>
6  */
7 
8 /**
9  * DOC: Live Update Orchestrator (LUO)
10  *
11  * Live Update is a specialized, kexec-based reboot process that allows a
12  * running kernel to be updated from one version to another while preserving
13  * the state of selected resources and keeping designated hardware devices
14  * operational. For these devices, DMA activity may continue throughout the
15  * kernel transition.
16  *
17  * While the primary use case driving this work is supporting live updates of
18  * the Linux kernel when it is used as a hypervisor in cloud environments, the
19  * LUO framework itself is designed to be workload-agnostic. Live Update
20  * facilitates a full kernel version upgrade for any type of system.
21  *
22  * For example, a non-hypervisor system running an in-memory cache like
23  * memcached with many gigabytes of data can use LUO. The userspace service
24  * can place its cache into a memfd, have its state preserved by LUO, and
25  * restore it immediately after the kernel kexec.
26  *
27  * Whether the system is running virtual machines, containers, a
28  * high-performance database, or networking services, LUO's primary goal is to
29  * enable a full kernel update by preserving critical userspace state and
30  * keeping essential devices operational.
31  *
32  * The core of LUO is a mechanism that tracks the progress of a live update,
33  * along with a callback API that allows other kernel subsystems to participate
34  * in the process. Example subsystems that can hook into LUO include: kvm,
35  * iommu, interrupts, vfio, participating filesystems, and memory management.
36  *
37  * LUO uses Kexec Handover to transfer memory state from the current kernel to
38  * the next kernel. For more details see Documentation/core-api/kho/index.rst.
39  */
40 
41 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
42 
43 #include <linux/atomic.h>
44 #include <linux/errno.h>
45 #include <linux/file.h>
46 #include <linux/fs.h>
47 #include <linux/init.h>
48 #include <linux/io.h>
49 #include <linux/kernel.h>
50 #include <linux/kexec_handover.h>
51 #include <linux/kho/abi/luo.h>
52 #include <linux/kobject.h>
53 #include <linux/libfdt.h>
54 #include <linux/liveupdate.h>
55 #include <linux/miscdevice.h>
56 #include <linux/mm.h>
57 #include <linux/sizes.h>
58 #include <linux/string.h>
59 #include <linux/unaligned.h>
60 
61 #include "kexec_handover_internal.h"
62 #include "luo_internal.h"
63 
64 static struct {
65 	bool enabled;
66 	void *fdt_out;
67 	void *fdt_in;
68 	u64 liveupdate_num;
69 } luo_global;
70 
71 static int __init early_liveupdate_param(char *buf)
72 {
73 	return kstrtobool(buf, &luo_global.enabled);
74 }
75 early_param("liveupdate", early_liveupdate_param);
76 
77 static int __init luo_early_startup(void)
78 {
79 	phys_addr_t fdt_phys;
80 	int err, ln_size;
81 	const void *ptr;
82 
83 	if (!kho_is_enabled()) {
84 		if (liveupdate_enabled())
85 			pr_warn("Disabling liveupdate because KHO is disabled\n");
86 		luo_global.enabled = false;
87 		return 0;
88 	}
89 
90 	/* Retrieve LUO subtree, and verify its format. */
91 	err = kho_retrieve_subtree(LUO_FDT_KHO_ENTRY_NAME, &fdt_phys);
92 	if (err) {
93 		if (err != -ENOENT) {
94 			pr_err("failed to retrieve FDT '%s' from KHO: %pe\n",
95 			       LUO_FDT_KHO_ENTRY_NAME, ERR_PTR(err));
96 			return err;
97 		}
98 
99 		return 0;
100 	}
101 
102 	luo_global.fdt_in = phys_to_virt(fdt_phys);
103 	err = fdt_node_check_compatible(luo_global.fdt_in, 0,
104 					LUO_FDT_COMPATIBLE);
105 	if (err) {
106 		pr_err("FDT '%s' is incompatible with '%s' [%d]\n",
107 		       LUO_FDT_KHO_ENTRY_NAME, LUO_FDT_COMPATIBLE, err);
108 
109 		return -EINVAL;
110 	}
111 
112 	ln_size = 0;
113 	ptr = fdt_getprop(luo_global.fdt_in, 0, LUO_FDT_LIVEUPDATE_NUM,
114 			  &ln_size);
115 	if (!ptr || ln_size != sizeof(luo_global.liveupdate_num)) {
116 		pr_err("Unable to get live update number '%s' [%d]\n",
117 		       LUO_FDT_LIVEUPDATE_NUM, ln_size);
118 
119 		return -EINVAL;
120 	}
121 
122 	luo_global.liveupdate_num = get_unaligned((u64 *)ptr);
123 	pr_info("Retrieved live update data, liveupdate number: %lld\n",
124 		luo_global.liveupdate_num);
125 
126 	err = luo_session_setup_incoming(luo_global.fdt_in);
127 	if (err)
128 		return err;
129 
130 	err = luo_flb_setup_incoming(luo_global.fdt_in);
131 
132 	return err;
133 }
134 
135 static int __init liveupdate_early_init(void)
136 {
137 	int err;
138 
139 	err = luo_early_startup();
140 	if (err) {
141 		luo_global.enabled = false;
142 		luo_restore_fail("The incoming tree failed to initialize properly [%pe], disabling live update\n",
143 				 ERR_PTR(err));
144 	}
145 
146 	return err;
147 }
148 early_initcall(liveupdate_early_init);
149 
150 /* Called during boot to create outgoing LUO fdt tree */
151 static int __init luo_fdt_setup(void)
152 {
153 	const u64 ln = luo_global.liveupdate_num + 1;
154 	void *fdt_out;
155 	int err;
156 
157 	fdt_out = kho_alloc_preserve(LUO_FDT_SIZE);
158 	if (IS_ERR(fdt_out)) {
159 		pr_err("failed to allocate/preserve FDT memory\n");
160 		return PTR_ERR(fdt_out);
161 	}
162 
163 	err = fdt_create(fdt_out, LUO_FDT_SIZE);
164 	err |= fdt_finish_reservemap(fdt_out);
165 	err |= fdt_begin_node(fdt_out, "");
166 	err |= fdt_property_string(fdt_out, "compatible", LUO_FDT_COMPATIBLE);
167 	err |= fdt_property(fdt_out, LUO_FDT_LIVEUPDATE_NUM, &ln, sizeof(ln));
168 	err |= luo_session_setup_outgoing(fdt_out);
169 	err |= luo_flb_setup_outgoing(fdt_out);
170 	err |= fdt_end_node(fdt_out);
171 	err |= fdt_finish(fdt_out);
172 	if (err)
173 		goto exit_free;
174 
175 	err = kho_add_subtree(LUO_FDT_KHO_ENTRY_NAME, fdt_out);
176 	if (err)
177 		goto exit_free;
178 	luo_global.fdt_out = fdt_out;
179 
180 	return 0;
181 
182 exit_free:
183 	kho_unpreserve_free(fdt_out);
184 	pr_err("failed to prepare LUO FDT: %d\n", err);
185 
186 	return err;
187 }
188 
189 /*
190  * late initcall because it initializes the outgoing tree that is needed only
191  * once userspace starts using /dev/liveupdate.
192  */
193 static int __init luo_late_startup(void)
194 {
195 	int err;
196 
197 	if (!liveupdate_enabled())
198 		return 0;
199 
200 	err = luo_fdt_setup();
201 	if (err)
202 		luo_global.enabled = false;
203 
204 	return err;
205 }
206 late_initcall(luo_late_startup);
207 
208 /* Public Functions */
209 
210 /**
211  * liveupdate_reboot() - Kernel reboot notifier for live update final
212  * serialization.
213  *
214  * This function is invoked directly from the reboot() syscall pathway
215  * if kexec is in progress.
216  *
217  * If any callback fails, this function aborts KHO, undoes the freeze()
218  * callbacks, and returns an error.
219  */
220 int liveupdate_reboot(void)
221 {
222 	int err;
223 
224 	if (!liveupdate_enabled())
225 		return 0;
226 
227 	err = luo_session_serialize();
228 	if (err)
229 		return err;
230 
231 	luo_flb_serialize();
232 
233 	return 0;
234 }
235 
236 /**
237  * liveupdate_enabled - Check if the live update feature is enabled.
238  *
239  * This function returns the state of the live update feature flag, which
240  * can be controlled via the ``liveupdate`` kernel command-line parameter.
241  *
242  * @return true if live update is enabled, false otherwise.
243  */
244 bool liveupdate_enabled(void)
245 {
246 	return luo_global.enabled;
247 }
248 
249 /**
250  * DOC: LUO ioctl Interface
251  *
252  * The IOCTL user-space control interface for the LUO subsystem.
253  * It registers a character device, typically found at ``/dev/liveupdate``,
254  * which allows a userspace agent to manage the LUO state machine and its
255  * associated resources, such as preservable file descriptors.
256  *
257  * To ensure that the state machine is controlled by a single entity, access
258  * to this device is exclusive: only one process is permitted to have
259  * ``/dev/liveupdate`` open at any given time. Subsequent open attempts will
260  * fail with -EBUSY until the first process closes its file descriptor.
261  * This singleton model simplifies state management by preventing conflicting
262  * commands from multiple userspace agents.
263  */
264 
265 struct luo_device_state {
266 	struct miscdevice miscdev;
267 	atomic_t in_use;
268 };
269 
270 static int luo_ioctl_create_session(struct luo_ucmd *ucmd)
271 {
272 	struct liveupdate_ioctl_create_session *argp = ucmd->cmd;
273 	struct file *file;
274 	int err;
275 
276 	argp->fd = get_unused_fd_flags(O_CLOEXEC);
277 	if (argp->fd < 0)
278 		return argp->fd;
279 
280 	err = luo_session_create(argp->name, &file);
281 	if (err)
282 		goto err_put_fd;
283 
284 	err = luo_ucmd_respond(ucmd, sizeof(*argp));
285 	if (err)
286 		goto err_put_file;
287 
288 	fd_install(argp->fd, file);
289 
290 	return 0;
291 
292 err_put_file:
293 	fput(file);
294 err_put_fd:
295 	put_unused_fd(argp->fd);
296 
297 	return err;
298 }
299 
300 static int luo_ioctl_retrieve_session(struct luo_ucmd *ucmd)
301 {
302 	struct liveupdate_ioctl_retrieve_session *argp = ucmd->cmd;
303 	struct file *file;
304 	int err;
305 
306 	argp->fd = get_unused_fd_flags(O_CLOEXEC);
307 	if (argp->fd < 0)
308 		return argp->fd;
309 
310 	err = luo_session_retrieve(argp->name, &file);
311 	if (err < 0)
312 		goto err_put_fd;
313 
314 	err = luo_ucmd_respond(ucmd, sizeof(*argp));
315 	if (err)
316 		goto err_put_file;
317 
318 	fd_install(argp->fd, file);
319 
320 	return 0;
321 
322 err_put_file:
323 	fput(file);
324 err_put_fd:
325 	put_unused_fd(argp->fd);
326 
327 	return err;
328 }
329 
330 static int luo_open(struct inode *inodep, struct file *filep)
331 {
332 	struct luo_device_state *ldev = container_of(filep->private_data,
333 						     struct luo_device_state,
334 						     miscdev);
335 
336 	if (atomic_cmpxchg(&ldev->in_use, 0, 1))
337 		return -EBUSY;
338 
339 	/* Always return -EIO to user if deserialization fail */
340 	if (luo_session_deserialize()) {
341 		atomic_set(&ldev->in_use, 0);
342 		return -EIO;
343 	}
344 
345 	return 0;
346 }
347 
348 static int luo_release(struct inode *inodep, struct file *filep)
349 {
350 	struct luo_device_state *ldev = container_of(filep->private_data,
351 						     struct luo_device_state,
352 						     miscdev);
353 	atomic_set(&ldev->in_use, 0);
354 
355 	return 0;
356 }
357 
358 union ucmd_buffer {
359 	struct liveupdate_ioctl_create_session create;
360 	struct liveupdate_ioctl_retrieve_session retrieve;
361 };
362 
363 struct luo_ioctl_op {
364 	unsigned int size;
365 	unsigned int min_size;
366 	unsigned int ioctl_num;
367 	int (*execute)(struct luo_ucmd *ucmd);
368 };
369 
370 #define IOCTL_OP(_ioctl, _fn, _struct, _last)                                  \
371 	[_IOC_NR(_ioctl) - LIVEUPDATE_CMD_BASE] = {                            \
372 		.size = sizeof(_struct) +                                      \
373 			BUILD_BUG_ON_ZERO(sizeof(union ucmd_buffer) <          \
374 					  sizeof(_struct)),                    \
375 		.min_size = offsetofend(_struct, _last),                       \
376 		.ioctl_num = _ioctl,                                           \
377 		.execute = _fn,                                                \
378 	}
379 
380 static const struct luo_ioctl_op luo_ioctl_ops[] = {
381 	IOCTL_OP(LIVEUPDATE_IOCTL_CREATE_SESSION, luo_ioctl_create_session,
382 		 struct liveupdate_ioctl_create_session, name),
383 	IOCTL_OP(LIVEUPDATE_IOCTL_RETRIEVE_SESSION, luo_ioctl_retrieve_session,
384 		 struct liveupdate_ioctl_retrieve_session, name),
385 };
386 
387 static long luo_ioctl(struct file *filep, unsigned int cmd, unsigned long arg)
388 {
389 	const struct luo_ioctl_op *op;
390 	struct luo_ucmd ucmd = {};
391 	union ucmd_buffer buf;
392 	unsigned int nr;
393 	int err;
394 
395 	nr = _IOC_NR(cmd);
396 	if (nr - LIVEUPDATE_CMD_BASE >= ARRAY_SIZE(luo_ioctl_ops))
397 		return -EINVAL;
398 
399 	ucmd.ubuffer = (void __user *)arg;
400 	err = get_user(ucmd.user_size, (u32 __user *)ucmd.ubuffer);
401 	if (err)
402 		return err;
403 
404 	op = &luo_ioctl_ops[nr - LIVEUPDATE_CMD_BASE];
405 	if (op->ioctl_num != cmd)
406 		return -ENOIOCTLCMD;
407 	if (ucmd.user_size < op->min_size)
408 		return -EINVAL;
409 
410 	ucmd.cmd = &buf;
411 	err = copy_struct_from_user(ucmd.cmd, op->size, ucmd.ubuffer,
412 				    ucmd.user_size);
413 	if (err)
414 		return err;
415 
416 	return op->execute(&ucmd);
417 }
418 
419 static const struct file_operations luo_fops = {
420 	.owner		= THIS_MODULE,
421 	.open		= luo_open,
422 	.release	= luo_release,
423 	.unlocked_ioctl	= luo_ioctl,
424 };
425 
426 static struct luo_device_state luo_dev = {
427 	.miscdev = {
428 		.minor = MISC_DYNAMIC_MINOR,
429 		.name  = "liveupdate",
430 		.fops  = &luo_fops,
431 	},
432 	.in_use = ATOMIC_INIT(0),
433 };
434 
435 static int __init liveupdate_ioctl_init(void)
436 {
437 	if (!liveupdate_enabled())
438 		return 0;
439 
440 	return misc_register(&luo_dev.miscdev);
441 }
442 late_initcall(liveupdate_ioctl_init);
443