| 1 | |
| 2 |
|
| 3 | |
| 4 | * Copyright (c) 2025, Google LLC.
|
| 5 | * Pasha Tatashin <[email protected]>
|
| 6 |
|
| 7 |
|
| 8 | |
| 9 | * DOC: LUO File Descriptors
|
| 10 | *
|
| 11 | * LUO provides the infrastructure to preserve specific, stateful file
|
| 12 | * descriptors across a kexec-based live update. The primary goal is to allow
|
| 13 | * workloads, such as virtual machines using vfio, memfd, or iommufd, to
|
| 14 | * retain access to their essential resources without interruption.
|
| 15 | *
|
| 16 | * The framework is built around a callback-based handler model and a well-
|
| 17 | * defined lifecycle for each preserved file.
|
| 18 | *
|
| 19 | * Handler Registration:
|
| 20 | * Kernel modules responsible for a specific file type (e.g., memfd, vfio)
|
| 21 | * register a &struct liveupdate_file_handler. This handler provides a set of
|
| 22 | * callbacks that LUO invokes at different stages of the update process, most
|
| 23 | * notably:
|
| 24 | *
|
| 25 | * - can_preserve(): A lightweight check to determine if the handler is
|
| 26 | * compatible with a given 'struct file'.
|
| 27 | * - preserve(): The heavyweight operation that saves the file's state and
|
| 28 | * returns an opaque u64 handle. This is typically performed while the
|
| 29 | * workload is still active to minimize the downtime during the
|
| 30 | * actual reboot transition.
|
| 31 | * - unpreserve(): Cleans up any resources allocated by .preserve(), called
|
| 32 | * if the preservation process is aborted before the reboot (i.e. session is
|
| 33 | * closed).
|
| 34 | * - freeze(): A final pre-reboot opportunity to prepare the state for kexec.
|
| 35 | * We are already in reboot syscall, and therefore userspace cannot mutate
|
| 36 | * the file anymore.
|
| 37 | * - unfreeze(): Undoes the actions of .freeze(), called if the live update
|
| 38 | * is aborted after the freeze phase.
|
| 39 | * - retrieve(): Reconstructs the file in the new kernel from the preserved
|
| 40 | * handle.
|
| 41 | * - finish(): Performs final check and cleanup in the new kernel. After
|
| 42 | * succesul finish call, LUO gives up ownership to this file.
|
| 43 | *
|
| 44 | * File Preservation Lifecycle happy path:
|
| 45 | *
|
| 46 | * 1. Preserve (Normal Operation): A userspace agent preserves files one by one
|
| 47 | * via an ioctl. For each file, luo_preserve_file() finds a compatible
|
| 48 | * handler, calls its .preserve() operation, and creates an internal &struct
|
| 49 | * luo_file to track the live state.
|
| 50 | *
|
| 51 | * 2. Freeze (Pre-Reboot): Just before the kexec, luo_file_freeze() is called.
|
| 52 | * It iterates through all preserved files, calls their respective .freeze()
|
| 53 | * operation, and serializes their final metadata (compatible string, token,
|
| 54 | * and data handle) into a contiguous memory block for KHO.
|
| 55 | *
|
| 56 | * 3. Deserialize: After kexec, luo_file_deserialize() runs when session gets
|
| 57 | * deserialized (which is when /dev/liveupdate is first opened). It reads the
|
| 58 | * serialized data from the KHO memory region and reconstructs the in-memory
|
| 59 | * list of &struct luo_file instances for the new kernel, linking them to
|
| 60 | * their corresponding handlers.
|
| 61 | *
|
| 62 | * 4. Retrieve (New Kernel - Userspace Ready): The userspace agent can now
|
| 63 | * restore file descriptors by providing a token. luo_retrieve_file()
|
| 64 | * searches for the matching token, calls the handler's .retrieve() op to
|
| 65 | * re-create the 'struct file', and returns a new FD. Files can be
|
| 66 | * retrieved in ANY order.
|
| 67 | *
|
| 68 | * 5. Finish (New Kernel - Cleanup): Once a session retrival is complete,
|
| 69 | * luo_file_finish() is called. It iterates through all files, invokes their
|
| 70 | * .finish() operations for final cleanup, and releases all associated kernel
|
| 71 | * resources.
|
| 72 | *
|
| 73 | * File Preservation Lifecycle unhappy paths:
|
| 74 | *
|
| 75 | * 1. Abort Before Reboot: If the userspace agent aborts the live update
|
| 76 | * process before calling reboot (e.g., by closing the session file
|
| 77 | * descriptor), the session's release handler calls
|
| 78 | * luo_file_unpreserve_files(). This invokes the .unpreserve() callback on
|
| 79 | * all preserved files, ensuring all allocated resources are cleaned up and
|
| 80 | * returning the system to a clean state.
|
| 81 | *
|
| 82 | * 2. Freeze Failure: During the reboot() syscall, if any handler's .freeze()
|
| 83 | * op fails, the .unfreeze() op is invoked on all previously *successful*
|
| 84 | * freezes to roll back their state. The reboot() syscall then returns an
|
| 85 | * error to userspace, canceling the live update.
|
| 86 | *
|
| 87 | * 3. Finish Failure: In the new kernel, if a handler's .finish() op fails,
|
| 88 | * the luo_file_finish() operation is aborted. LUO retains ownership of
|
| 89 | * all files within that session, including those that were not yet
|
| 90 | * processed. The userspace agent can attempt to call the finish operation
|
| 91 | * again later. If the issue cannot be resolved, these resources will be held
|
| 92 | * by LUO until the next live update cycle, at which point they will be
|
| 93 | * discarded.
|
| 94 |
|
| 95 |
|
| 96 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
|
| 97 |
|
| 98 | #include <linux/cleanup.h>
|
| 99 | #include <linux/compiler.h>
|
| 100 | #include <linux/err.h>
|
| 101 | #include <linux/errno.h>
|
| 102 | #include <linux/file.h>
|
| 103 | #include <linux/fs.h>
|
| 104 | #include <linux/io.h>
|
| 105 | #include <linux/kexec_handover.h>
|
| 106 | #include <linux/kho/abi/luo.h>
|
| 107 | #include <linux/list_private.h>
|
| 108 | #include <linux/liveupdate.h>
|
| 109 | #include <linux/module.h>
|
| 110 | #include <linux/sizes.h>
|
| 111 | #include <linux/xarray.h>
|
| 112 | #include <linux/slab.h>
|
| 113 | #include <linux/string.h>
|
| 114 | #include "luo_internal.h"
|
| 115 |
|
| 116 | static LIST_HEAD(luo_file_handler_list);
|
| 117 |
|
| 118 |
|
| 119 | static DEFINE_XARRAY(luo_preserved_files);
|
| 120 |
|
| 121 | |
| 122 | * struct luo_file - Represents a single preserved file instance.
|
| 123 | * @fh: Pointer to the &struct liveupdate_file_handler that manages
|
| 124 | * this type of file.
|
| 125 | * @file: Pointer to the kernel's &struct file that is being preserved.
|
| 126 | * This is NULL in the new kernel until the file is successfully
|
| 127 | * retrieved.
|
| 128 | * @serialized_data: The opaque u64 handle to the serialized state of the file.
|
| 129 | * This handle is passed back to the handler's .freeze(),
|
| 130 | * .retrieve(), and .finish() callbacks, allowing it to track
|
| 131 | * and update its serialized state across phases.
|
| 132 | * @private_data: Pointer to the private data for the file used to hold runtime
|
| 133 | * state that is not preserved. Set by the handler's .preserve()
|
| 134 | * callback, and must be freed in the handler's .unpreserve()
|
| 135 | * callback.
|
| 136 | * @retrieve_status: Status code indicating whether a user/kernel in the new kernel has
|
| 137 | * successfully called retrieve() on this file. This prevents
|
| 138 | * multiple retrieval attempts. A value of 0 means a retrieve()
|
| 139 | * has not been attempted, a positive value means the retrieve()
|
| 140 | * was successful, and a negative value means the retrieve()
|
| 141 | * failed, and the value is the error code of the call.
|
| 142 | * @mutex: A mutex that protects the fields of this specific instance
|
| 143 | * (e.g., @retrieved, @file), ensuring that operations like
|
| 144 | * retrieving or finishing a file are atomic.
|
| 145 | * @list: The list_head linking this instance into its parent
|
| 146 | * file_set's list of preserved files.
|
| 147 | * @token: The user-provided unique token used to identify this file.
|
| 148 | *
|
| 149 | * This structure is the core in-kernel representation of a single file being
|
| 150 | * managed through a live update. An instance is created by luo_preserve_file()
|
| 151 | * to link a 'struct file' to its corresponding handler, a user-provided token,
|
| 152 | * and the serialized state handle returned by the handler's .preserve()
|
| 153 | * operation.
|
| 154 | *
|
| 155 | * These instances are tracked in a per-file_set list. The @serialized_data
|
| 156 | * field, which holds a handle to the file's serialized state, may be updated
|
| 157 | * during the .freeze() callback before being serialized for the next kernel.
|
| 158 | * After reboot, these structures are recreated by luo_file_deserialize() and
|
| 159 | * are finally cleaned up by luo_file_finish().
|
| 160 |
|
| 161 | struct luo_file {
|
| 162 | struct liveupdate_file_handler *fh;
|
| 163 | struct file *file;
|
| 164 | u64 serialized_data;
|
| 165 | void *private_data;
|
| 166 | int retrieve_status;
|
| 167 | struct mutex mutex;
|
| 168 | struct list_head list;
|
| 169 | u64 token;
|
| 170 | };
|
| 171 |
|
| 172 | static unsigned long luo_get_id(struct liveupdate_file_handler *fh,
|
| 173 | struct file *file)
|
| 174 | {
|
| 175 | return fh->ops->get_id ? fh->ops->get_id(file) : (unsigned long)file;
|
| 176 | }
|
| 177 |
|
| 178 | static bool luo_token_is_used(struct luo_file_set *file_set, u64 token)
|
| 179 | {
|
| 180 | struct luo_file *iter;
|
| 181 |
|
| 182 | list_for_each_entry(iter, &file_set->files_list, list) {
|
| 183 | if (iter->token == token)
|
| 184 | return true;
|
| 185 | }
|
| 186 |
|
| 187 | return false;
|
| 188 | }
|
| 189 |
|
| 190 | |
| 191 | * luo_preserve_file - Initiate the preservation of a file descriptor.
|
| 192 | * @file_set: The file_set to which the preserved file will be added.
|
| 193 | * @token: A unique, user-provided identifier for the file.
|
| 194 | * @fd: The file descriptor to be preserved.
|
| 195 | *
|
| 196 | * This function orchestrates the first phase of preserving a file. Upon entry,
|
| 197 | * it takes a reference to the 'struct file' via fget(), effectively making LUO
|
| 198 | * a co-owner of the file. This reference is held until the file is either
|
| 199 | * unpreserved or successfully finished in the next kernel, preventing the file
|
| 200 | * from being prematurely destroyed.
|
| 201 | *
|
| 202 | * This function orchestrates the first phase of preserving a file. It performs
|
| 203 | * the following steps:
|
| 204 | *
|
| 205 | * 1. Validates that the @token is not already in use within the file_set.
|
| 206 | * 2. Ensures the file_set's memory for files serialization is allocated
|
| 207 | * (allocates if needed).
|
| 208 | * 3. Iterates through registered handlers, calling can_preserve() to find one
|
| 209 | * compatible with the given @fd.
|
| 210 | * 4. Calls the handler's .preserve() operation, which saves the file's state
|
| 211 | * and returns an opaque private data handle.
|
| 212 | * 5. Adds the new instance to the file_set's internal list.
|
| 213 | *
|
| 214 | * On success, LUO takes a reference to the 'struct file' and considers it
|
| 215 | * under its management until it is unpreserved or finished.
|
| 216 | *
|
| 217 | * In case of any failure, all intermediate allocations (file reference, memory
|
| 218 | * for the 'luo_file' struct, etc.) are cleaned up before returning an error.
|
| 219 | *
|
| 220 | * Context: Can be called from an ioctl handler during normal system operation.
|
| 221 | * Return: 0 on success. Returns a negative errno on failure:
|
| 222 | * -EEXIST if the token is already used.
|
| 223 | * -EBUSY if the file descriptor is already preserved by another session.
|
| 224 | * -EBADF if the file descriptor is invalid.
|
| 225 | * -ENOSPC if the file_set is full.
|
| 226 | * -ENOENT if no compatible handler is found.
|
| 227 | * -ENOMEM on memory allocation failure.
|
| 228 | * Other erros might be returned by .preserve().
|
| 229 |
|
| 230 | int luo_preserve_file(struct luo_file_set *file_set, u64 token, int fd)
|
| 231 | {
|
| 232 | struct liveupdate_file_op_args args = {0};
|
| 233 | struct liveupdate_file_handler *fh;
|
| 234 | struct luo_file *luo_file;
|
| 235 | struct file *file;
|
| 236 | int err;
|
| 237 |
|
| 238 | if (luo_token_is_used(file_set, token))
|
| 239 | return -EEXIST;
|
| 240 |
|
| 241 | err = kho_block_set_grow(&file_set->block_set, file_set->count + 1);
|
| 242 | if (err)
|
| 243 | return err;
|
| 244 |
|
| 245 | file = fget(fd);
|
| 246 | if (!file) {
|
| 247 | err = -EBADF;
|
| 248 | goto err_shrink;
|
| 249 | }
|
| 250 |
|
| 251 | err = -ENOENT;
|
| 252 | down_read(&luo_register_rwlock);
|
| 253 | list_private_for_each_entry(fh, &luo_file_handler_list, list) {
|
| 254 | if (fh->ops->can_preserve(fh, file)) {
|
| 255 | if (try_module_get(fh->ops->owner))
|
| 256 | err = 0;
|
| 257 | break;
|
| 258 | }
|
| 259 | }
|
| 260 | up_read(&luo_register_rwlock);
|
| 261 |
|
| 262 |
|
| 263 | if (err)
|
| 264 | goto err_fput;
|
| 265 |
|
| 266 | err = xa_insert(&luo_preserved_files, luo_get_id(fh, file),
|
| 267 | file, GFP_KERNEL);
|
| 268 | if (err)
|
| 269 | goto err_module_put;
|
| 270 |
|
| 271 | err = luo_flb_file_preserve(fh);
|
| 272 | if (err)
|
| 273 | goto err_erase_xa;
|
| 274 |
|
| 275 | luo_file = kzalloc_obj(*luo_file);
|
| 276 | if (!luo_file) {
|
| 277 | err = -ENOMEM;
|
| 278 | goto err_flb_unpreserve;
|
| 279 | }
|
| 280 |
|
| 281 | luo_file->file = file;
|
| 282 | luo_file->fh = fh;
|
| 283 | luo_file->token = token;
|
| 284 | mutex_init(&luo_file->mutex);
|
| 285 |
|
| 286 | args.handler = fh;
|
| 287 | args.file = file;
|
| 288 | err = fh->ops->preserve(&args);
|
| 289 | if (err)
|
| 290 | goto err_kfree;
|
| 291 |
|
| 292 | luo_file->serialized_data = args.serialized_data;
|
| 293 | luo_file->private_data = args.private_data;
|
| 294 | list_add_tail(&luo_file->list, &file_set->files_list);
|
| 295 | file_set->count++;
|
| 296 |
|
| 297 | return 0;
|
| 298 |
|
| 299 | err_kfree:
|
| 300 | kfree(luo_file);
|
| 301 | err_flb_unpreserve:
|
| 302 | luo_flb_file_unpreserve(fh);
|
| 303 | err_erase_xa:
|
| 304 | xa_erase(&luo_preserved_files, luo_get_id(fh, file));
|
| 305 | err_module_put:
|
| 306 | module_put(fh->ops->owner);
|
| 307 | err_fput:
|
| 308 | fput(file);
|
| 309 | err_shrink:
|
| 310 | kho_block_set_shrink(&file_set->block_set, file_set->count);
|
| 311 |
|
| 312 | return err;
|
| 313 | }
|
| 314 |
|
| 315 | |
| 316 | * luo_file_unpreserve_files - Unpreserves all files from a file_set.
|
| 317 | * @file_set: The files to be cleaned up.
|
| 318 | *
|
| 319 | * This function serves as the primary cleanup path for a file_set. It is
|
| 320 | * invoked when the userspace agent closes the file_set's file descriptor.
|
| 321 | *
|
| 322 | * For each file, it performs the following cleanup actions:
|
| 323 | * 1. Calls the handler's .unpreserve() callback to allow the handler to
|
| 324 | * release any resources it allocated.
|
| 325 | * 2. Removes the file from the file_set's internal tracking list.
|
| 326 | * 3. Releases the reference to the 'struct file' that was taken by
|
| 327 | * luo_preserve_file() via fput(), returning ownership.
|
| 328 | * 4. Frees the memory associated with the internal 'struct luo_file'.
|
| 329 | *
|
| 330 | * After all individual files are unpreserved, it frees the contiguous memory
|
| 331 | * block that was allocated to hold their serialization data.
|
| 332 |
|
| 333 | void luo_file_unpreserve_files(struct luo_file_set *file_set)
|
| 334 | {
|
| 335 | struct luo_file *luo_file;
|
| 336 |
|
| 337 | while (!list_empty(&file_set->files_list)) {
|
| 338 | struct liveupdate_file_op_args args = {0};
|
| 339 |
|
| 340 | luo_file = list_last_entry(&file_set->files_list,
|
| 341 | struct luo_file, list);
|
| 342 |
|
| 343 | args.handler = luo_file->fh;
|
| 344 | args.file = luo_file->file;
|
| 345 | args.serialized_data = luo_file->serialized_data;
|
| 346 | args.private_data = luo_file->private_data;
|
| 347 | luo_file->fh->ops->unpreserve(&args);
|
| 348 | luo_flb_file_unpreserve(luo_file->fh);
|
| 349 |
|
| 350 | xa_erase(&luo_preserved_files,
|
| 351 | luo_get_id(luo_file->fh, luo_file->file));
|
| 352 | module_put(luo_file->fh->ops->owner);
|
| 353 |
|
| 354 | list_del(&luo_file->list);
|
| 355 | file_set->count--;
|
| 356 | kho_block_set_shrink(&file_set->block_set, file_set->count);
|
| 357 |
|
| 358 | fput(luo_file->file);
|
| 359 | mutex_destroy(&luo_file->mutex);
|
| 360 | kfree(luo_file);
|
| 361 | }
|
| 362 |
|
| 363 | kho_block_set_destroy(&file_set->block_set);
|
| 364 | }
|
| 365 |
|
| 366 | static int luo_file_freeze_one(struct luo_file_set *file_set,
|
| 367 | struct luo_file *luo_file)
|
| 368 | {
|
| 369 | int err = 0;
|
| 370 |
|
| 371 | guard(mutex)(&luo_file->mutex);
|
| 372 |
|
| 373 | if (luo_file->fh->ops->freeze) {
|
| 374 | struct liveupdate_file_op_args args = {0};
|
| 375 |
|
| 376 | args.handler = luo_file->fh;
|
| 377 | args.file = luo_file->file;
|
| 378 | args.serialized_data = luo_file->serialized_data;
|
| 379 | args.private_data = luo_file->private_data;
|
| 380 |
|
| 381 | err = luo_file->fh->ops->freeze(&args);
|
| 382 | if (!err)
|
| 383 | luo_file->serialized_data = args.serialized_data;
|
| 384 | }
|
| 385 |
|
| 386 | return err;
|
| 387 | }
|
| 388 |
|
| 389 | static void luo_file_unfreeze_one(struct luo_file_set *file_set,
|
| 390 | struct luo_file *luo_file)
|
| 391 | {
|
| 392 | guard(mutex)(&luo_file->mutex);
|
| 393 |
|
| 394 | if (luo_file->fh->ops->unfreeze) {
|
| 395 | struct liveupdate_file_op_args args = {0};
|
| 396 |
|
| 397 | args.handler = luo_file->fh;
|
| 398 | args.file = luo_file->file;
|
| 399 | args.serialized_data = luo_file->serialized_data;
|
| 400 | args.private_data = luo_file->private_data;
|
| 401 |
|
| 402 | luo_file->fh->ops->unfreeze(&args);
|
| 403 | }
|
| 404 | }
|
| 405 |
|
| 406 | static void __luo_file_unfreeze(struct luo_file_set *file_set,
|
| 407 | struct luo_file *failed_entry)
|
| 408 | {
|
| 409 | struct list_head *files_list = &file_set->files_list;
|
| 410 | struct luo_file *luo_file;
|
| 411 |
|
| 412 | list_for_each_entry(luo_file, files_list, list) {
|
| 413 | if (luo_file == failed_entry)
|
| 414 | break;
|
| 415 |
|
| 416 | luo_file_unfreeze_one(file_set, luo_file);
|
| 417 | }
|
| 418 |
|
| 419 | kho_block_set_clear(&file_set->block_set);
|
| 420 | }
|
| 421 |
|
| 422 | |
| 423 | * luo_file_freeze - Freezes all preserved files and serializes their metadata.
|
| 424 | * @file_set: The file_set whose files are to be frozen.
|
| 425 | * @file_set_ser: Where to put the serialized file_set.
|
| 426 | *
|
| 427 | * This function is called from the reboot() syscall path, just before the
|
| 428 | * kernel transitions to the new image via kexec. Its purpose is to perform the
|
| 429 | * final preparation and serialization of all preserved files in the file_set.
|
| 430 | *
|
| 431 | * It iterates through each preserved file in FIFO order (the order of
|
| 432 | * preservation) and performs two main actions:
|
| 433 | *
|
| 434 | * 1. Freezes the File: It calls the handler's .freeze() callback for each
|
| 435 | * file. This gives the handler a final opportunity to quiesce the device or
|
| 436 | * prepare its state for the upcoming reboot. The handler may update its
|
| 437 | * private data handle during this step.
|
| 438 | *
|
| 439 | * 2. Serializes Metadata: After a successful freeze, it copies the final file
|
| 440 | * metadata—the handler's compatible string, the user token, and the final
|
| 441 | * private data handle—into the pre-allocated contiguous memory buffer
|
| 442 | * (file_set->files) that will be handed over to the next kernel via KHO.
|
| 443 | *
|
| 444 | * Error Handling (Rollback):
|
| 445 | * This function is atomic. If any handler's .freeze() operation fails, the
|
| 446 | * entire live update is aborted. The __luo_file_unfreeze() helper is
|
| 447 | * immediately called to invoke the .unfreeze() op on all files that were
|
| 448 | * successfully frozen before the point of failure, rolling them back to a
|
| 449 | * running state. The function then returns an error, causing the reboot()
|
| 450 | * syscall to fail.
|
| 451 | *
|
| 452 | * Context: Called only from the liveupdate_reboot() path.
|
| 453 | * Return: 0 on success, or a negative errno on failure.
|
| 454 |
|
| 455 | int luo_file_freeze(struct luo_file_set *file_set,
|
| 456 | struct luo_file_set_ser *file_set_ser)
|
| 457 | {
|
| 458 | struct luo_file *luo_file;
|
| 459 | struct kho_block_set_it it;
|
| 460 | int err;
|
| 461 |
|
| 462 | if (!file_set->count)
|
| 463 | return 0;
|
| 464 |
|
| 465 | kho_block_set_it_init(&it, &file_set->block_set);
|
| 466 |
|
| 467 | list_for_each_entry(luo_file, &file_set->files_list, list) {
|
| 468 | struct luo_file_ser *file_ser = kho_block_set_it_reserve_entry(&it);
|
| 469 |
|
| 470 |
|
| 471 | if (WARN_ON_ONCE(!file_ser)) {
|
| 472 | err = -ENOSPC;
|
| 473 | goto err_unfreeze;
|
| 474 | }
|
| 475 |
|
| 476 | err = luo_file_freeze_one(file_set, luo_file);
|
| 477 | if (err < 0) {
|
| 478 | pr_warn("Freeze failed for token[%#0llx] handler[%s] err[%pe]\n",
|
| 479 | luo_file->token, luo_file->fh->compatible,
|
| 480 | ERR_PTR(err));
|
| 481 | goto err_unfreeze;
|
| 482 | }
|
| 483 |
|
| 484 | strscpy(file_ser->compatible, luo_file->fh->compatible,
|
| 485 | sizeof(file_ser->compatible));
|
| 486 | file_ser->data = luo_file->serialized_data;
|
| 487 | file_ser->token = luo_file->token;
|
| 488 | }
|
| 489 |
|
| 490 | file_set_ser->count = file_set->count;
|
| 491 | file_set_ser->files = kho_block_set_head_pa(&file_set->block_set);
|
| 492 |
|
| 493 | return 0;
|
| 494 |
|
| 495 | err_unfreeze:
|
| 496 | __luo_file_unfreeze(file_set, luo_file);
|
| 497 |
|
| 498 | return err;
|
| 499 | }
|
| 500 |
|
| 501 | |
| 502 | * luo_file_unfreeze - Unfreezes all files in a file_set and clear serialization
|
| 503 | * @file_set: The file_set whose files are to be unfrozen.
|
| 504 | * @file_set_ser: Serialized file_set.
|
| 505 | *
|
| 506 | * This function rolls back the state of all files in a file_set after the
|
| 507 | * freeze phase has begun but must be aborted. It is the counterpart to
|
| 508 | * luo_file_freeze().
|
| 509 | *
|
| 510 | * It invokes the __luo_file_unfreeze() helper with a NULL argument, which
|
| 511 | * signals the helper to iterate through all files in the file_set and call
|
| 512 | * their respective .unfreeze() handler callbacks.
|
| 513 | *
|
| 514 | * Context: This is called when the live update is aborted during
|
| 515 | * the reboot() syscall, after luo_file_freeze() has been called.
|
| 516 |
|
| 517 | void luo_file_unfreeze(struct luo_file_set *file_set,
|
| 518 | struct luo_file_set_ser *file_set_ser)
|
| 519 | {
|
| 520 | if (!file_set->count)
|
| 521 | return;
|
| 522 |
|
| 523 | __luo_file_unfreeze(file_set, NULL);
|
| 524 | memset(file_set_ser, 0, sizeof(*file_set_ser));
|
| 525 | }
|
| 526 |
|
| 527 | |
| 528 | * luo_retrieve_file - Restores a preserved file from a file_set by its token.
|
| 529 | * @file_set: The file_set from which to retrieve the file.
|
| 530 | * @token: The unique token identifying the file to be restored.
|
| 531 | * @filep: Output parameter; on success, this is populated with a pointer
|
| 532 | * to the newly retrieved 'struct file'.
|
| 533 | *
|
| 534 | * This function is the primary mechanism for recreating a file in the new
|
| 535 | * kernel after a live update. It searches the file_set's list of deserialized
|
| 536 | * files for an entry matching the provided @token.
|
| 537 | *
|
| 538 | * The operation is idempotent: if a file has already been successfully
|
| 539 | * retrieved, this function will simply return a pointer to the existing
|
| 540 | * 'struct file' and report success without re-executing the retrieve
|
| 541 | * operation. This is handled by checking the 'retrieved' flag under a lock.
|
| 542 | *
|
| 543 | * File retrieval can happen in any order; it is not bound by the order of
|
| 544 | * preservation.
|
| 545 | *
|
| 546 | * Context: Can be called from an ioctl or other in-kernel code in the new
|
| 547 | * kernel.
|
| 548 | * Return: 0 on success. Returns a negative errno on failure:
|
| 549 | * -ENOENT if no file with the matching token is found.
|
| 550 | * Any error code returned by the handler's .retrieve() op.
|
| 551 |
|
| 552 | int luo_retrieve_file(struct luo_file_set *file_set, u64 token,
|
| 553 | struct file **filep)
|
| 554 | {
|
| 555 | struct liveupdate_file_op_args args = {0};
|
| 556 | struct luo_file *luo_file;
|
| 557 | bool found = false;
|
| 558 | int err;
|
| 559 |
|
| 560 | if (list_empty(&file_set->files_list))
|
| 561 | return -ENOENT;
|
| 562 |
|
| 563 | list_for_each_entry(luo_file, &file_set->files_list, list) {
|
| 564 | if (luo_file->token == token) {
|
| 565 | found = true;
|
| 566 | break;
|
| 567 | }
|
| 568 | }
|
| 569 |
|
| 570 | if (!found)
|
| 571 | return -ENOENT;
|
| 572 |
|
| 573 | guard(mutex)(&luo_file->mutex);
|
| 574 | if (luo_file->retrieve_status < 0) {
|
| 575 |
|
| 576 | return luo_file->retrieve_status;
|
| 577 | }
|
| 578 |
|
| 579 | if (luo_file->retrieve_status > 0) {
|
| 580 | |
| 581 | * Someone is asking for this file again, so get a reference
|
| 582 | * for them.
|
| 583 |
|
| 584 | get_file(luo_file->file);
|
| 585 | *filep = luo_file->file;
|
| 586 | return 0;
|
| 587 | }
|
| 588 |
|
| 589 | args.handler = luo_file->fh;
|
| 590 | args.serialized_data = luo_file->serialized_data;
|
| 591 | err = luo_file->fh->ops->retrieve(&args);
|
| 592 | if (err) {
|
| 593 |
|
| 594 | luo_file->retrieve_status = err;
|
| 595 | return err;
|
| 596 | }
|
| 597 |
|
| 598 | luo_file->file = args.file;
|
| 599 |
|
| 600 | get_file(luo_file->file);
|
| 601 |
|
| 602 | WARN_ON(xa_insert(&luo_preserved_files,
|
| 603 | luo_get_id(luo_file->fh, luo_file->file),
|
| 604 | luo_file->file, GFP_KERNEL));
|
| 605 |
|
| 606 | *filep = luo_file->file;
|
| 607 | luo_file->retrieve_status = 1;
|
| 608 |
|
| 609 | return 0;
|
| 610 | }
|
| 611 |
|
| 612 | static int luo_file_can_finish_one(struct luo_file_set *file_set,
|
| 613 | struct luo_file *luo_file)
|
| 614 | {
|
| 615 | bool can_finish = true;
|
| 616 |
|
| 617 | guard(mutex)(&luo_file->mutex);
|
| 618 |
|
| 619 | if (luo_file->fh->ops->can_finish) {
|
| 620 | struct liveupdate_file_op_args args = {0};
|
| 621 |
|
| 622 | args.handler = luo_file->fh;
|
| 623 | args.file = luo_file->file;
|
| 624 | args.serialized_data = luo_file->serialized_data;
|
| 625 | args.retrieve_status = luo_file->retrieve_status;
|
| 626 | can_finish = luo_file->fh->ops->can_finish(&args);
|
| 627 | }
|
| 628 |
|
| 629 | return can_finish ? 0 : -EBUSY;
|
| 630 | }
|
| 631 |
|
| 632 | static void luo_file_finish_one(struct luo_file_set *file_set,
|
| 633 | struct luo_file *luo_file)
|
| 634 | {
|
| 635 | struct liveupdate_file_op_args args = {0};
|
| 636 |
|
| 637 | guard(mutex)(&luo_file->mutex);
|
| 638 |
|
| 639 | args.handler = luo_file->fh;
|
| 640 | args.file = luo_file->file;
|
| 641 | args.serialized_data = luo_file->serialized_data;
|
| 642 | args.retrieve_status = luo_file->retrieve_status;
|
| 643 |
|
| 644 | luo_file->fh->ops->finish(&args);
|
| 645 | luo_flb_file_finish(luo_file->fh);
|
| 646 | }
|
| 647 |
|
| 648 | |
| 649 | * luo_file_finish - Completes the lifecycle for all files in a file_set.
|
| 650 | * @file_set: The file_set to be finalized.
|
| 651 | *
|
| 652 | * This function orchestrates the final teardown of a live update file_set in
|
| 653 | * the new kernel. It should be called after all necessary files have been
|
| 654 | * retrieved and the userspace agent is ready to release the preserved state.
|
| 655 | *
|
| 656 | * The function iterates through all tracked files. For each file, it performs
|
| 657 | * the following sequence of cleanup actions:
|
| 658 | *
|
| 659 | * 1. If file is not yet retrieved, retrieves it, and calls can_finish() on
|
| 660 | * every file in the file_set. If all can_finish return true, continue to
|
| 661 | * finish.
|
| 662 | * 2. Calls the handler's .finish() callback (via luo_file_finish_one) to
|
| 663 | * allow for final resource cleanup within the handler.
|
| 664 | * 3. Releases LUO's ownership reference on the 'struct file' via fput(). This
|
| 665 | * is the counterpart to the get_file() call in luo_retrieve_file().
|
| 666 | * 4. Removes the 'struct luo_file' from the file_set's internal list.
|
| 667 | * 5. Frees the memory for the 'struct luo_file' instance itself.
|
| 668 | *
|
| 669 | * After successfully finishing all individual files, it frees the
|
| 670 | * contiguous memory block that was used to transfer the serialized metadata
|
| 671 | * from the previous kernel.
|
| 672 | *
|
| 673 | * Error Handling (Atomic Failure):
|
| 674 | * This operation is atomic. If any handler's .can_finish() op fails, the entire
|
| 675 | * function aborts immediately and returns an error.
|
| 676 | *
|
| 677 | * Context: Can be called from an ioctl handler in the new kernel.
|
| 678 | * Return: 0 on success, or a negative errno on failure.
|
| 679 |
|
| 680 | int luo_file_finish(struct luo_file_set *file_set)
|
| 681 | {
|
| 682 | struct list_head *files_list = &file_set->files_list;
|
| 683 | struct luo_file *luo_file;
|
| 684 | int err;
|
| 685 |
|
| 686 | if (!file_set->count)
|
| 687 | return 0;
|
| 688 |
|
| 689 | list_for_each_entry(luo_file, files_list, list) {
|
| 690 | err = luo_file_can_finish_one(file_set, luo_file);
|
| 691 | if (err)
|
| 692 | return err;
|
| 693 | }
|
| 694 |
|
| 695 | while (!list_empty(&file_set->files_list)) {
|
| 696 | luo_file = list_last_entry(&file_set->files_list,
|
| 697 | struct luo_file, list);
|
| 698 |
|
| 699 | luo_file_finish_one(file_set, luo_file);
|
| 700 |
|
| 701 | if (luo_file->file) {
|
| 702 | xa_erase(&luo_preserved_files,
|
| 703 | luo_get_id(luo_file->fh, luo_file->file));
|
| 704 | fput(luo_file->file);
|
| 705 | }
|
| 706 | module_put(luo_file->fh->ops->owner);
|
| 707 | list_del(&luo_file->list);
|
| 708 | file_set->count--;
|
| 709 | kho_block_set_shrink(&file_set->block_set, file_set->count);
|
| 710 | mutex_destroy(&luo_file->mutex);
|
| 711 | kfree(luo_file);
|
| 712 | }
|
| 713 |
|
| 714 | kho_block_set_destroy(&file_set->block_set);
|
| 715 |
|
| 716 | return 0;
|
| 717 | }
|
| 718 |
|
| 719 | static int luo_file_deserialize_one(struct luo_file_set *file_set,
|
| 720 | struct luo_file_ser *ser)
|
| 721 | {
|
| 722 | struct liveupdate_file_handler *fh;
|
| 723 | bool handler_found = false;
|
| 724 | struct luo_file *luo_file;
|
| 725 |
|
| 726 | down_read(&luo_register_rwlock);
|
| 727 | list_private_for_each_entry(fh, &luo_file_handler_list, list) {
|
| 728 | if (!strcmp(fh->compatible, ser->compatible)) {
|
| 729 | if (try_module_get(fh->ops->owner))
|
| 730 | handler_found = true;
|
| 731 | break;
|
| 732 | }
|
| 733 | }
|
| 734 | up_read(&luo_register_rwlock);
|
| 735 |
|
| 736 | if (!handler_found) {
|
| 737 | pr_warn("No registered handler for compatible '%.*s'\n",
|
| 738 | (int)sizeof(ser->compatible),
|
| 739 | ser->compatible);
|
| 740 | return -ENOENT;
|
| 741 | }
|
| 742 |
|
| 743 | luo_file = kzalloc_obj(*luo_file);
|
| 744 | if (!luo_file) {
|
| 745 | module_put(fh->ops->owner);
|
| 746 | return -ENOMEM;
|
| 747 | }
|
| 748 |
|
| 749 | luo_file->fh = fh;
|
| 750 | luo_file->file = NULL;
|
| 751 | luo_file->serialized_data = ser->data;
|
| 752 | luo_file->token = ser->token;
|
| 753 | mutex_init(&luo_file->mutex);
|
| 754 | list_add_tail(&luo_file->list, &file_set->files_list);
|
| 755 |
|
| 756 | return 0;
|
| 757 | }
|
| 758 |
|
| 759 | |
| 760 | * luo_file_deserialize - Reconstructs the list of preserved files in the new kernel.
|
| 761 | * @file_set: The incoming file_set to fill with deserialized data.
|
| 762 | * @file_set_ser: Serialized KHO file_set data from the previous kernel.
|
| 763 | *
|
| 764 | * This function is called during the early boot process of the new kernel. It
|
| 765 | * takes the raw, contiguous memory block of 'struct luo_file_ser' entries,
|
| 766 | * provided by the previous kernel, and transforms it back into a live,
|
| 767 | * in-memory linked list of 'struct luo_file' instances.
|
| 768 | *
|
| 769 | * For each serialized entry, it performs the following steps:
|
| 770 | * 1. Reads the 'compatible' string.
|
| 771 | * 2. Searches the global list of registered file handlers for one that
|
| 772 | * matches the compatible string.
|
| 773 | * 3. Allocates a new 'struct luo_file'.
|
| 774 | * 4. Populates the new structure with the deserialized data (token, private
|
| 775 | * data handle) and links it to the found handler. The 'file' pointer is
|
| 776 | * initialized to NULL, as the file has not been retrieved yet.
|
| 777 | * 5. Adds the new 'struct luo_file' to the file_set's files_list.
|
| 778 | *
|
| 779 | * This prepares the file_set for userspace, which can later call
|
| 780 | * luo_retrieve_file() to restore the actual file descriptors.
|
| 781 | *
|
| 782 | * Context: Called from session deserialization.
|
| 783 |
|
| 784 | int luo_file_deserialize(struct luo_file_set *file_set,
|
| 785 | struct luo_file_set_ser *file_set_ser)
|
| 786 | {
|
| 787 | struct luo_file_ser *file_ser;
|
| 788 | struct kho_block_set_it it;
|
| 789 | int err;
|
| 790 |
|
| 791 | if (!file_set_ser->files) {
|
| 792 | WARN_ON(file_set_ser->count);
|
| 793 | return 0;
|
| 794 | }
|
| 795 |
|
| 796 | file_set->count = 0;
|
| 797 | err = kho_block_set_restore(&file_set->block_set, file_set_ser->files);
|
| 798 | if (err)
|
| 799 | return err;
|
| 800 |
|
| 801 | |
| 802 | * Note on error handling:
|
| 803 | *
|
| 804 | * If deserialization fails (e.g., allocation failure or corrupt data),
|
| 805 | * we intentionally skip cleanup of files that were already restored.
|
| 806 | *
|
| 807 | * A partial failure leaves the preserved state inconsistent.
|
| 808 | * Implementing a safe "undo" to unwind complex dependencies (sessions,
|
| 809 | * files, hardware state) is error-prone and provides little value, as
|
| 810 | * the system is effectively in a broken state.
|
| 811 | *
|
| 812 | * We treat these resources as leaked. The expected recovery path is for
|
| 813 | * userspace to detect the failure and trigger a reboot, which will
|
| 814 | * reliably reset devices and reclaim memory.
|
| 815 |
|
| 816 | kho_block_set_it_init(&it, &file_set->block_set);
|
| 817 | while ((file_ser = kho_block_set_it_read_entry(&it))) {
|
| 818 | err = luo_file_deserialize_one(file_set, file_ser);
|
| 819 | if (err)
|
| 820 | goto err_destroy_blocks;
|
| 821 | file_set->count++;
|
| 822 | }
|
| 823 |
|
| 824 | if (file_set->count != file_set_ser->count) {
|
| 825 | pr_warn("File count mismatch: expected %llu, found %llu\n",
|
| 826 | file_set_ser->count, file_set->count);
|
| 827 | err = -EINVAL;
|
| 828 | goto err_destroy_blocks;
|
| 829 | }
|
| 830 |
|
| 831 | return 0;
|
| 832 |
|
| 833 | err_destroy_blocks:
|
| 834 | while (!list_empty(&file_set->files_list)) {
|
| 835 | struct luo_file *luo_file;
|
| 836 |
|
| 837 | luo_file = list_first_entry(&file_set->files_list,
|
| 838 | struct luo_file, list);
|
| 839 | list_del(&luo_file->list);
|
| 840 | module_put(luo_file->fh->ops->owner);
|
| 841 | mutex_destroy(&luo_file->mutex);
|
| 842 | kfree(luo_file);
|
| 843 | }
|
| 844 | file_set->count = 0;
|
| 845 | kho_block_set_destroy(&file_set->block_set);
|
| 846 | return err;
|
| 847 | }
|
| 848 |
|
| 849 | void luo_file_set_init(struct luo_file_set *file_set)
|
| 850 | {
|
| 851 | INIT_LIST_HEAD(&file_set->files_list);
|
| 852 | kho_block_set_init(&file_set->block_set, sizeof(struct luo_file_ser));
|
| 853 | }
|
| 854 |
|
| 855 | void luo_file_set_destroy(struct luo_file_set *file_set)
|
| 856 | {
|
| 857 | WARN_ON(file_set->count);
|
| 858 | WARN_ON(!list_empty(&file_set->files_list));
|
| 859 | WARN_ON(!kho_block_set_is_empty(&file_set->block_set));
|
| 860 | }
|
| 861 |
|
| 862 | |
| 863 | * liveupdate_register_file_handler - Register a file handler with LUO.
|
| 864 | * @fh: Pointer to a caller-allocated &struct liveupdate_file_handler.
|
| 865 | * The caller must initialize this structure, including a unique
|
| 866 | * 'compatible' string and a valid 'fh' callbacks. This function adds the
|
| 867 | * handler to the global list of supported file handlers.
|
| 868 | *
|
| 869 | * Context: Typically called during module initialization for file types that
|
| 870 | * support live update preservation.
|
| 871 | *
|
| 872 | * Return: 0 on success. Negative errno on failure.
|
| 873 |
|
| 874 | int liveupdate_register_file_handler(struct liveupdate_file_handler *fh)
|
| 875 | {
|
| 876 | struct liveupdate_file_handler *fh_iter;
|
| 877 | int err;
|
| 878 |
|
| 879 | if (!liveupdate_enabled())
|
| 880 | return -EOPNOTSUPP;
|
| 881 |
|
| 882 |
|
| 883 | if (!fh->ops->preserve || !fh->ops->unpreserve || !fh->ops->retrieve ||
|
| 884 | !fh->ops->finish || !fh->ops->can_preserve) {
|
| 885 | return -EINVAL;
|
| 886 | }
|
| 887 |
|
| 888 | down_write(&luo_register_rwlock);
|
| 889 |
|
| 890 | list_private_for_each_entry(fh_iter, &luo_file_handler_list, list) {
|
| 891 | if (!strcmp(fh_iter->compatible, fh->compatible)) {
|
| 892 | pr_err("File handler registration failed: Compatible string '%s' already registered.\n",
|
| 893 | fh->compatible);
|
| 894 | err = -EEXIST;
|
| 895 | goto err_unlock;
|
| 896 | }
|
| 897 | }
|
| 898 |
|
| 899 | INIT_LIST_HEAD(&ACCESS_PRIVATE(fh, flb_list));
|
| 900 | INIT_LIST_HEAD(&ACCESS_PRIVATE(fh, list));
|
| 901 | list_add_tail(&ACCESS_PRIVATE(fh, list), &luo_file_handler_list);
|
| 902 | up_write(&luo_register_rwlock);
|
| 903 |
|
| 904 | liveupdate_test_register(fh);
|
| 905 |
|
| 906 | return 0;
|
| 907 |
|
| 908 | err_unlock:
|
| 909 | up_write(&luo_register_rwlock);
|
| 910 | return err;
|
| 911 | }
|
| 912 |
|
| 913 | |
| 914 | * liveupdate_unregister_file_handler - Unregister a liveupdate file handler
|
| 915 | * @fh: The file handler to unregister
|
| 916 | *
|
| 917 | * Unregisters the file handler from the liveupdate core. This function
|
| 918 | * reverses the operations of liveupdate_register_file_handler().
|
| 919 |
|
| 920 | void liveupdate_unregister_file_handler(struct liveupdate_file_handler *fh)
|
| 921 | {
|
| 922 | if (!liveupdate_enabled())
|
| 923 | return;
|
| 924 |
|
| 925 | guard(rwsem_write)(&luo_register_rwlock);
|
| 926 | luo_flb_unregister_all(fh);
|
| 927 | list_del(&ACCESS_PRIVATE(fh, list));
|
| 928 | }
|