xref: /linux/kernel/liveupdate/luo_core.c (revision 509d3f45847627f4c5cdce004c3ec79262b5239c)
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
39  * Documentation/core-api/kho/concepts.rst.
40  */
41 
42 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
43 
44 #include <linux/atomic.h>
45 #include <linux/errno.h>
46 #include <linux/file.h>
47 #include <linux/fs.h>
48 #include <linux/init.h>
49 #include <linux/io.h>
50 #include <linux/kernel.h>
51 #include <linux/kexec_handover.h>
52 #include <linux/kho/abi/luo.h>
53 #include <linux/kobject.h>
54 #include <linux/libfdt.h>
55 #include <linux/liveupdate.h>
56 #include <linux/miscdevice.h>
57 #include <linux/mm.h>
58 #include <linux/sizes.h>
59 #include <linux/string.h>
60 #include <linux/unaligned.h>
61 
62 #include "kexec_handover_internal.h"
63 #include "luo_internal.h"
64 
65 static struct {
66 	bool enabled;
67 	void *fdt_out;
68 	void *fdt_in;
69 	u64 liveupdate_num;
70 } luo_global;
71 
early_liveupdate_param(char * buf)72 static int __init early_liveupdate_param(char *buf)
73 {
74 	return kstrtobool(buf, &luo_global.enabled);
75 }
76 early_param("liveupdate", early_liveupdate_param);
77 
luo_early_startup(void)78 static int __init luo_early_startup(void)
79 {
80 	phys_addr_t fdt_phys;
81 	int err, ln_size;
82 	const void *ptr;
83 
84 	if (!kho_is_enabled()) {
85 		if (liveupdate_enabled())
86 			pr_warn("Disabling liveupdate because KHO is disabled\n");
87 		luo_global.enabled = false;
88 		return 0;
89 	}
90 
91 	/* Retrieve LUO subtree, and verify its format. */
92 	err = kho_retrieve_subtree(LUO_FDT_KHO_ENTRY_NAME, &fdt_phys);
93 	if (err) {
94 		if (err != -ENOENT) {
95 			pr_err("failed to retrieve FDT '%s' from KHO: %pe\n",
96 			       LUO_FDT_KHO_ENTRY_NAME, ERR_PTR(err));
97 			return err;
98 		}
99 
100 		return 0;
101 	}
102 
103 	luo_global.fdt_in = phys_to_virt(fdt_phys);
104 	err = fdt_node_check_compatible(luo_global.fdt_in, 0,
105 					LUO_FDT_COMPATIBLE);
106 	if (err) {
107 		pr_err("FDT '%s' is incompatible with '%s' [%d]\n",
108 		       LUO_FDT_KHO_ENTRY_NAME, LUO_FDT_COMPATIBLE, err);
109 
110 		return -EINVAL;
111 	}
112 
113 	ln_size = 0;
114 	ptr = fdt_getprop(luo_global.fdt_in, 0, LUO_FDT_LIVEUPDATE_NUM,
115 			  &ln_size);
116 	if (!ptr || ln_size != sizeof(luo_global.liveupdate_num)) {
117 		pr_err("Unable to get live update number '%s' [%d]\n",
118 		       LUO_FDT_LIVEUPDATE_NUM, ln_size);
119 
120 		return -EINVAL;
121 	}
122 
123 	luo_global.liveupdate_num = get_unaligned((u64 *)ptr);
124 	pr_info("Retrieved live update data, liveupdate number: %lld\n",
125 		luo_global.liveupdate_num);
126 
127 	err = luo_session_setup_incoming(luo_global.fdt_in);
128 	if (err)
129 		return err;
130 
131 	return 0;
132 }
133 
liveupdate_early_init(void)134 static int __init liveupdate_early_init(void)
135 {
136 	int err;
137 
138 	err = luo_early_startup();
139 	if (err) {
140 		luo_global.enabled = false;
141 		luo_restore_fail("The incoming tree failed to initialize properly [%pe], disabling live update\n",
142 				 ERR_PTR(err));
143 	}
144 
145 	return err;
146 }
147 early_initcall(liveupdate_early_init);
148 
149 /* Called during boot to create outgoing LUO fdt tree */
luo_fdt_setup(void)150 static int __init luo_fdt_setup(void)
151 {
152 	const u64 ln = luo_global.liveupdate_num + 1;
153 	void *fdt_out;
154 	int err;
155 
156 	fdt_out = kho_alloc_preserve(LUO_FDT_SIZE);
157 	if (IS_ERR(fdt_out)) {
158 		pr_err("failed to allocate/preserve FDT memory\n");
159 		return PTR_ERR(fdt_out);
160 	}
161 
162 	err = fdt_create(fdt_out, LUO_FDT_SIZE);
163 	err |= fdt_finish_reservemap(fdt_out);
164 	err |= fdt_begin_node(fdt_out, "");
165 	err |= fdt_property_string(fdt_out, "compatible", LUO_FDT_COMPATIBLE);
166 	err |= fdt_property(fdt_out, LUO_FDT_LIVEUPDATE_NUM, &ln, sizeof(ln));
167 	err |= luo_session_setup_outgoing(fdt_out);
168 	err |= fdt_end_node(fdt_out);
169 	err |= fdt_finish(fdt_out);
170 	if (err)
171 		goto exit_free;
172 
173 	err = kho_add_subtree(LUO_FDT_KHO_ENTRY_NAME, fdt_out);
174 	if (err)
175 		goto exit_free;
176 	luo_global.fdt_out = fdt_out;
177 
178 	return 0;
179 
180 exit_free:
181 	kho_unpreserve_free(fdt_out);
182 	pr_err("failed to prepare LUO FDT: %d\n", err);
183 
184 	return err;
185 }
186 
187 /*
188  * late initcall because it initializes the outgoing tree that is needed only
189  * once userspace starts using /dev/liveupdate.
190  */
luo_late_startup(void)191 static int __init luo_late_startup(void)
192 {
193 	int err;
194 
195 	if (!liveupdate_enabled())
196 		return 0;
197 
198 	err = luo_fdt_setup();
199 	if (err)
200 		luo_global.enabled = false;
201 
202 	return err;
203 }
204 late_initcall(luo_late_startup);
205 
206 /* Public Functions */
207 
208 /**
209  * liveupdate_reboot() - Kernel reboot notifier for live update final
210  * serialization.
211  *
212  * This function is invoked directly from the reboot() syscall pathway
213  * if kexec is in progress.
214  *
215  * If any callback fails, this function aborts KHO, undoes the freeze()
216  * callbacks, and returns an error.
217  */
liveupdate_reboot(void)218 int liveupdate_reboot(void)
219 {
220 	int err;
221 
222 	if (!liveupdate_enabled())
223 		return 0;
224 
225 	err = luo_session_serialize();
226 	if (err)
227 		return err;
228 
229 	err = kho_finalize();
230 	if (err) {
231 		pr_err("kho_finalize failed %d\n", err);
232 		/*
233 		 * kho_finalize() may return libfdt errors, to aboid passing to
234 		 * userspace unknown errors, change this to EAGAIN.
235 		 */
236 		err = -EAGAIN;
237 	}
238 
239 	return err;
240 }
241 
242 /**
243  * liveupdate_enabled - Check if the live update feature is enabled.
244  *
245  * This function returns the state of the live update feature flag, which
246  * can be controlled via the ``liveupdate`` kernel command-line parameter.
247  *
248  * @return true if live update is enabled, false otherwise.
249  */
liveupdate_enabled(void)250 bool liveupdate_enabled(void)
251 {
252 	return luo_global.enabled;
253 }
254 
255 /**
256  * DOC: LUO ioctl Interface
257  *
258  * The IOCTL user-space control interface for the LUO subsystem.
259  * It registers a character device, typically found at ``/dev/liveupdate``,
260  * which allows a userspace agent to manage the LUO state machine and its
261  * associated resources, such as preservable file descriptors.
262  *
263  * To ensure that the state machine is controlled by a single entity, access
264  * to this device is exclusive: only one process is permitted to have
265  * ``/dev/liveupdate`` open at any given time. Subsequent open attempts will
266  * fail with -EBUSY until the first process closes its file descriptor.
267  * This singleton model simplifies state management by preventing conflicting
268  * commands from multiple userspace agents.
269  */
270 
271 struct luo_device_state {
272 	struct miscdevice miscdev;
273 	atomic_t in_use;
274 };
275 
luo_ioctl_create_session(struct luo_ucmd * ucmd)276 static int luo_ioctl_create_session(struct luo_ucmd *ucmd)
277 {
278 	struct liveupdate_ioctl_create_session *argp = ucmd->cmd;
279 	struct file *file;
280 	int err;
281 
282 	argp->fd = get_unused_fd_flags(O_CLOEXEC);
283 	if (argp->fd < 0)
284 		return argp->fd;
285 
286 	err = luo_session_create(argp->name, &file);
287 	if (err)
288 		goto err_put_fd;
289 
290 	err = luo_ucmd_respond(ucmd, sizeof(*argp));
291 	if (err)
292 		goto err_put_file;
293 
294 	fd_install(argp->fd, file);
295 
296 	return 0;
297 
298 err_put_file:
299 	fput(file);
300 err_put_fd:
301 	put_unused_fd(argp->fd);
302 
303 	return err;
304 }
305 
luo_ioctl_retrieve_session(struct luo_ucmd * ucmd)306 static int luo_ioctl_retrieve_session(struct luo_ucmd *ucmd)
307 {
308 	struct liveupdate_ioctl_retrieve_session *argp = ucmd->cmd;
309 	struct file *file;
310 	int err;
311 
312 	argp->fd = get_unused_fd_flags(O_CLOEXEC);
313 	if (argp->fd < 0)
314 		return argp->fd;
315 
316 	err = luo_session_retrieve(argp->name, &file);
317 	if (err < 0)
318 		goto err_put_fd;
319 
320 	err = luo_ucmd_respond(ucmd, sizeof(*argp));
321 	if (err)
322 		goto err_put_file;
323 
324 	fd_install(argp->fd, file);
325 
326 	return 0;
327 
328 err_put_file:
329 	fput(file);
330 err_put_fd:
331 	put_unused_fd(argp->fd);
332 
333 	return err;
334 }
335 
luo_open(struct inode * inodep,struct file * filep)336 static int luo_open(struct inode *inodep, struct file *filep)
337 {
338 	struct luo_device_state *ldev = container_of(filep->private_data,
339 						     struct luo_device_state,
340 						     miscdev);
341 
342 	if (atomic_cmpxchg(&ldev->in_use, 0, 1))
343 		return -EBUSY;
344 
345 	/* Always return -EIO to user if deserialization fail */
346 	if (luo_session_deserialize()) {
347 		atomic_set(&ldev->in_use, 0);
348 		return -EIO;
349 	}
350 
351 	return 0;
352 }
353 
luo_release(struct inode * inodep,struct file * filep)354 static int luo_release(struct inode *inodep, struct file *filep)
355 {
356 	struct luo_device_state *ldev = container_of(filep->private_data,
357 						     struct luo_device_state,
358 						     miscdev);
359 	atomic_set(&ldev->in_use, 0);
360 
361 	return 0;
362 }
363 
364 union ucmd_buffer {
365 	struct liveupdate_ioctl_create_session create;
366 	struct liveupdate_ioctl_retrieve_session retrieve;
367 };
368 
369 struct luo_ioctl_op {
370 	unsigned int size;
371 	unsigned int min_size;
372 	unsigned int ioctl_num;
373 	int (*execute)(struct luo_ucmd *ucmd);
374 };
375 
376 #define IOCTL_OP(_ioctl, _fn, _struct, _last)                                  \
377 	[_IOC_NR(_ioctl) - LIVEUPDATE_CMD_BASE] = {                            \
378 		.size = sizeof(_struct) +                                      \
379 			BUILD_BUG_ON_ZERO(sizeof(union ucmd_buffer) <          \
380 					  sizeof(_struct)),                    \
381 		.min_size = offsetofend(_struct, _last),                       \
382 		.ioctl_num = _ioctl,                                           \
383 		.execute = _fn,                                                \
384 	}
385 
386 static const struct luo_ioctl_op luo_ioctl_ops[] = {
387 	IOCTL_OP(LIVEUPDATE_IOCTL_CREATE_SESSION, luo_ioctl_create_session,
388 		 struct liveupdate_ioctl_create_session, name),
389 	IOCTL_OP(LIVEUPDATE_IOCTL_RETRIEVE_SESSION, luo_ioctl_retrieve_session,
390 		 struct liveupdate_ioctl_retrieve_session, name),
391 };
392 
luo_ioctl(struct file * filep,unsigned int cmd,unsigned long arg)393 static long luo_ioctl(struct file *filep, unsigned int cmd, unsigned long arg)
394 {
395 	const struct luo_ioctl_op *op;
396 	struct luo_ucmd ucmd = {};
397 	union ucmd_buffer buf;
398 	unsigned int nr;
399 	int err;
400 
401 	nr = _IOC_NR(cmd);
402 	if (nr < LIVEUPDATE_CMD_BASE ||
403 	    (nr - LIVEUPDATE_CMD_BASE) >= ARRAY_SIZE(luo_ioctl_ops)) {
404 		return -EINVAL;
405 	}
406 
407 	ucmd.ubuffer = (void __user *)arg;
408 	err = get_user(ucmd.user_size, (u32 __user *)ucmd.ubuffer);
409 	if (err)
410 		return err;
411 
412 	op = &luo_ioctl_ops[nr - LIVEUPDATE_CMD_BASE];
413 	if (op->ioctl_num != cmd)
414 		return -ENOIOCTLCMD;
415 	if (ucmd.user_size < op->min_size)
416 		return -EINVAL;
417 
418 	ucmd.cmd = &buf;
419 	err = copy_struct_from_user(ucmd.cmd, op->size, ucmd.ubuffer,
420 				    ucmd.user_size);
421 	if (err)
422 		return err;
423 
424 	return op->execute(&ucmd);
425 }
426 
427 static const struct file_operations luo_fops = {
428 	.owner		= THIS_MODULE,
429 	.open		= luo_open,
430 	.release	= luo_release,
431 	.unlocked_ioctl	= luo_ioctl,
432 };
433 
434 static struct luo_device_state luo_dev = {
435 	.miscdev = {
436 		.minor = MISC_DYNAMIC_MINOR,
437 		.name  = "liveupdate",
438 		.fops  = &luo_fops,
439 	},
440 	.in_use = ATOMIC_INIT(0),
441 };
442 
liveupdate_ioctl_init(void)443 static int __init liveupdate_ioctl_init(void)
444 {
445 	if (!liveupdate_enabled())
446 		return 0;
447 
448 	return misc_register(&luo_dev.miscdev);
449 }
450 late_initcall(liveupdate_ioctl_init);
451