| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * Provide a pstore intermediate backend, organized into kernel memory |
| 4 | * allocated zones that are then mapped and flushed into a single |
| 5 | * contiguous region on a storage backend of some kind (block, mtd, etc). |
| 6 | */ |
| 7 | |
| 8 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
| 9 | |
| 10 | #include <linux/kernel.h> |
| 11 | #include <linux/module.h> |
| 12 | #include <linux/slab.h> |
| 13 | #include <linux/mount.h> |
| 14 | #include <linux/printk.h> |
| 15 | #include <linux/fs.h> |
| 16 | #include <linux/pstore_zone.h> |
| 17 | #include <linux/kdev_t.h> |
| 18 | #include <linux/device.h> |
| 19 | #include <linux/namei.h> |
| 20 | #include <linux/fcntl.h> |
| 21 | #include <linux/uio.h> |
| 22 | #include <linux/writeback.h> |
| 23 | #include "internal.h" |
| 24 | |
| 25 | /** |
| 26 | * struct psz_buffer - header of zone to flush to storage |
| 27 | * |
| 28 | * @sig: signature to indicate header (PSZ_SIG xor PSZONE-type value) |
| 29 | * @datalen: length of data in @data |
| 30 | * @start: offset into @data where the beginning of the stored bytes begin |
| 31 | * @data: zone data. |
| 32 | */ |
| 33 | struct psz_buffer { |
| 34 | #define PSZ_SIG (0x43474244) /* DBGC */ |
| 35 | uint32_t sig; |
| 36 | atomic_t datalen; |
| 37 | atomic_t start; |
| 38 | uint8_t data[]; |
| 39 | }; |
| 40 | |
| 41 | /** |
| 42 | * struct psz_kmsg_header - kmsg dump-specific header to flush to storage |
| 43 | * |
| 44 | * @magic: magic num for kmsg dump header |
| 45 | * @time: kmsg dump trigger time |
| 46 | * @compressed: whether compressed |
| 47 | * @counter: kmsg dump counter |
| 48 | * @reason: the kmsg dump reason (e.g. oops, panic, etc) |
| 49 | * @data: pointer to log data |
| 50 | * |
| 51 | * This is a sub-header for a kmsg dump, trailing after &psz_buffer. |
| 52 | */ |
| 53 | struct { |
| 54 | #define 0x4dfc3ae5 /* Just a random number */ |
| 55 | uint32_t ; |
| 56 | struct timespec64 ; |
| 57 | bool ; |
| 58 | uint32_t ; |
| 59 | enum kmsg_dump_reason ; |
| 60 | uint8_t []; |
| 61 | }; |
| 62 | |
| 63 | /** |
| 64 | * struct pstore_zone - single stored buffer |
| 65 | * |
| 66 | * @off: zone offset of storage |
| 67 | * @type: front-end type for this zone |
| 68 | * @name: front-end name for this zone |
| 69 | * @buffer: pointer to data buffer managed by this zone |
| 70 | * @oldbuf: pointer to old data buffer |
| 71 | * @buffer_size: bytes in @buffer->data |
| 72 | * @should_recover: whether this zone should recover from storage |
| 73 | * @dirty: whether the data in @buffer dirty |
| 74 | * |
| 75 | * zone structure in memory. |
| 76 | */ |
| 77 | struct pstore_zone { |
| 78 | loff_t off; |
| 79 | const char *name; |
| 80 | enum pstore_type_id type; |
| 81 | |
| 82 | struct psz_buffer *buffer; |
| 83 | struct psz_buffer *oldbuf; |
| 84 | size_t buffer_size; |
| 85 | bool should_recover; |
| 86 | atomic_t dirty; |
| 87 | }; |
| 88 | |
| 89 | /** |
| 90 | * struct psz_context - all about running state of pstore/zone |
| 91 | * |
| 92 | * @kpszs: kmsg dump storage zones |
| 93 | * @ppsz: pmsg storage zone |
| 94 | * @cpsz: console storage zone |
| 95 | * @fpszs: ftrace storage zones |
| 96 | * @kmsg_max_cnt: max count of @kpszs |
| 97 | * @kmsg_read_cnt: counter of total read kmsg dumps |
| 98 | * @kmsg_write_cnt: counter of total kmsg dump writes |
| 99 | * @pmsg_read_cnt: counter of total read pmsg zone |
| 100 | * @console_read_cnt: counter of total read console zone |
| 101 | * @ftrace_max_cnt: max count of @fpszs |
| 102 | * @ftrace_read_cnt: counter of max read ftrace zone |
| 103 | * @oops_counter: counter of oops dumps |
| 104 | * @panic_counter: counter of panic dumps |
| 105 | * @recovered: whether finished recovering data from storage |
| 106 | * @on_panic: whether panic is happening |
| 107 | * @pstore_zone_info_lock: lock to @pstore_zone_info |
| 108 | * @pstore_zone_info: information from backend |
| 109 | * @pstore: structure for pstore |
| 110 | */ |
| 111 | struct psz_context { |
| 112 | struct pstore_zone **kpszs; |
| 113 | struct pstore_zone *ppsz; |
| 114 | struct pstore_zone *cpsz; |
| 115 | struct pstore_zone **fpszs; |
| 116 | unsigned int kmsg_max_cnt; |
| 117 | unsigned int kmsg_read_cnt; |
| 118 | unsigned int kmsg_write_cnt; |
| 119 | unsigned int pmsg_read_cnt; |
| 120 | unsigned int console_read_cnt; |
| 121 | unsigned int ftrace_max_cnt; |
| 122 | unsigned int ftrace_read_cnt; |
| 123 | /* |
| 124 | * These counters should be calculated during recovery. |
| 125 | * It records the oops/panic times after crashes rather than boots. |
| 126 | */ |
| 127 | unsigned int oops_counter; |
| 128 | unsigned int panic_counter; |
| 129 | atomic_t recovered; |
| 130 | atomic_t on_panic; |
| 131 | |
| 132 | /* |
| 133 | * pstore_zone_info_lock protects this entire structure during calls |
| 134 | * to register_pstore_zone()/unregister_pstore_zone(). |
| 135 | */ |
| 136 | struct mutex pstore_zone_info_lock; |
| 137 | struct pstore_zone_info *pstore_zone_info; |
| 138 | struct pstore_info pstore; |
| 139 | }; |
| 140 | static struct psz_context pstore_zone_cxt; |
| 141 | |
| 142 | static void psz_flush_all_dirty_zones(struct work_struct *); |
| 143 | static DECLARE_DELAYED_WORK(psz_cleaner, psz_flush_all_dirty_zones); |
| 144 | |
| 145 | /** |
| 146 | * enum psz_flush_mode - flush mode for psz_zone_write() |
| 147 | * |
| 148 | * @FLUSH_NONE: do not flush to storage but update data on memory |
| 149 | * @FLUSH_PART: just flush part of data including meta data to storage |
| 150 | * @FLUSH_META: just flush meta data of zone to storage |
| 151 | * @FLUSH_ALL: flush all of zone |
| 152 | */ |
| 153 | enum psz_flush_mode { |
| 154 | FLUSH_NONE = 0, |
| 155 | FLUSH_PART, |
| 156 | FLUSH_META, |
| 157 | FLUSH_ALL, |
| 158 | }; |
| 159 | |
| 160 | static inline int buffer_datalen(struct pstore_zone *zone) |
| 161 | { |
| 162 | return atomic_read(v: &zone->buffer->datalen); |
| 163 | } |
| 164 | |
| 165 | static inline int buffer_start(struct pstore_zone *zone) |
| 166 | { |
| 167 | return atomic_read(v: &zone->buffer->start); |
| 168 | } |
| 169 | |
| 170 | static inline bool is_on_panic(void) |
| 171 | { |
| 172 | return atomic_read(v: &pstore_zone_cxt.on_panic); |
| 173 | } |
| 174 | |
| 175 | static ssize_t psz_zone_read_buffer(struct pstore_zone *zone, char *buf, |
| 176 | size_t len, unsigned long off) |
| 177 | { |
| 178 | if (!buf || !zone || !zone->buffer) |
| 179 | return -EINVAL; |
| 180 | if (off > zone->buffer_size) |
| 181 | return -EINVAL; |
| 182 | len = min_t(size_t, len, zone->buffer_size - off); |
| 183 | memcpy(buf, zone->buffer->data + off, len); |
| 184 | return len; |
| 185 | } |
| 186 | |
| 187 | static int psz_zone_read_oldbuf(struct pstore_zone *zone, char *buf, |
| 188 | size_t len, unsigned long off) |
| 189 | { |
| 190 | if (!buf || !zone || !zone->oldbuf) |
| 191 | return -EINVAL; |
| 192 | if (off > zone->buffer_size) |
| 193 | return -EINVAL; |
| 194 | len = min_t(size_t, len, zone->buffer_size - off); |
| 195 | memcpy(buf, zone->oldbuf->data + off, len); |
| 196 | return 0; |
| 197 | } |
| 198 | |
| 199 | static int psz_zone_write(struct pstore_zone *zone, |
| 200 | enum psz_flush_mode flush_mode, const char *buf, |
| 201 | size_t len, unsigned long off) |
| 202 | { |
| 203 | struct pstore_zone_info *info = pstore_zone_cxt.pstore_zone_info; |
| 204 | ssize_t wcnt = 0; |
| 205 | ssize_t (*writeop)(const char *buf, size_t bytes, loff_t pos); |
| 206 | size_t wlen; |
| 207 | |
| 208 | if (off > zone->buffer_size) |
| 209 | return -EINVAL; |
| 210 | |
| 211 | wlen = min_t(size_t, len, zone->buffer_size - off); |
| 212 | if (buf && wlen) { |
| 213 | memcpy(zone->buffer->data + off, buf, wlen); |
| 214 | atomic_set(v: &zone->buffer->datalen, i: wlen + off); |
| 215 | } |
| 216 | |
| 217 | /* avoid damaging old records */ |
| 218 | if (!is_on_panic() && !atomic_read(v: &pstore_zone_cxt.recovered)) |
| 219 | goto dirty; |
| 220 | |
| 221 | writeop = is_on_panic() ? info->panic_write : info->write; |
| 222 | if (!writeop) |
| 223 | goto dirty; |
| 224 | |
| 225 | switch (flush_mode) { |
| 226 | case FLUSH_NONE: |
| 227 | if (unlikely(buf && wlen)) |
| 228 | goto dirty; |
| 229 | return 0; |
| 230 | case FLUSH_PART: |
| 231 | wcnt = writeop((const char *)zone->buffer->data + off, wlen, |
| 232 | zone->off + sizeof(*zone->buffer) + off); |
| 233 | if (wcnt != wlen) |
| 234 | goto dirty; |
| 235 | fallthrough; |
| 236 | case FLUSH_META: |
| 237 | wlen = sizeof(struct psz_buffer); |
| 238 | wcnt = writeop((const char *)zone->buffer, wlen, zone->off); |
| 239 | if (wcnt != wlen) |
| 240 | goto dirty; |
| 241 | break; |
| 242 | case FLUSH_ALL: |
| 243 | wlen = zone->buffer_size + sizeof(*zone->buffer); |
| 244 | wcnt = writeop((const char *)zone->buffer, wlen, zone->off); |
| 245 | if (wcnt != wlen) |
| 246 | goto dirty; |
| 247 | break; |
| 248 | } |
| 249 | |
| 250 | return 0; |
| 251 | dirty: |
| 252 | /* no need to mark it dirty if going to try next zone */ |
| 253 | if (wcnt == -ENOMSG) |
| 254 | return -ENOMSG; |
| 255 | atomic_set(v: &zone->dirty, i: true); |
| 256 | /* flush dirty zones nicely */ |
| 257 | if (wcnt == -EBUSY && !is_on_panic()) |
| 258 | schedule_delayed_work(dwork: &psz_cleaner, delay: msecs_to_jiffies(m: 500)); |
| 259 | return -EBUSY; |
| 260 | } |
| 261 | |
| 262 | static int psz_flush_dirty_zone(struct pstore_zone *zone) |
| 263 | { |
| 264 | int ret; |
| 265 | |
| 266 | if (unlikely(!zone)) |
| 267 | return -EINVAL; |
| 268 | |
| 269 | if (unlikely(!atomic_read(&pstore_zone_cxt.recovered))) |
| 270 | return -EBUSY; |
| 271 | |
| 272 | if (!atomic_xchg(v: &zone->dirty, new: false)) |
| 273 | return 0; |
| 274 | |
| 275 | ret = psz_zone_write(zone, flush_mode: FLUSH_ALL, NULL, len: 0, off: 0); |
| 276 | if (ret) |
| 277 | atomic_set(v: &zone->dirty, i: true); |
| 278 | return ret; |
| 279 | } |
| 280 | |
| 281 | static int psz_flush_dirty_zones(struct pstore_zone **zones, unsigned int cnt) |
| 282 | { |
| 283 | int i, ret; |
| 284 | struct pstore_zone *zone; |
| 285 | |
| 286 | if (!zones) |
| 287 | return -EINVAL; |
| 288 | |
| 289 | for (i = 0; i < cnt; i++) { |
| 290 | zone = zones[i]; |
| 291 | if (!zone) |
| 292 | return -EINVAL; |
| 293 | ret = psz_flush_dirty_zone(zone); |
| 294 | if (ret) |
| 295 | return ret; |
| 296 | } |
| 297 | return 0; |
| 298 | } |
| 299 | |
| 300 | static int psz_move_zone(struct pstore_zone *old, struct pstore_zone *new) |
| 301 | { |
| 302 | const char *data = (const char *)old->buffer->data; |
| 303 | int ret; |
| 304 | |
| 305 | ret = psz_zone_write(zone: new, flush_mode: FLUSH_ALL, buf: data, len: buffer_datalen(zone: old), off: 0); |
| 306 | if (ret) { |
| 307 | atomic_set(v: &new->buffer->datalen, i: 0); |
| 308 | atomic_set(v: &new->dirty, i: false); |
| 309 | return ret; |
| 310 | } |
| 311 | atomic_set(v: &old->buffer->datalen, i: 0); |
| 312 | return 0; |
| 313 | } |
| 314 | |
| 315 | static void psz_flush_all_dirty_zones(struct work_struct *work) |
| 316 | { |
| 317 | struct psz_context *cxt = &pstore_zone_cxt; |
| 318 | int ret = 0; |
| 319 | |
| 320 | if (cxt->ppsz) |
| 321 | ret |= psz_flush_dirty_zone(zone: cxt->ppsz); |
| 322 | if (cxt->cpsz) |
| 323 | ret |= psz_flush_dirty_zone(zone: cxt->cpsz); |
| 324 | if (cxt->kpszs) |
| 325 | ret |= psz_flush_dirty_zones(zones: cxt->kpszs, cnt: cxt->kmsg_max_cnt); |
| 326 | if (cxt->fpszs) |
| 327 | ret |= psz_flush_dirty_zones(zones: cxt->fpszs, cnt: cxt->ftrace_max_cnt); |
| 328 | if (ret && cxt->pstore_zone_info) |
| 329 | schedule_delayed_work(dwork: &psz_cleaner, delay: msecs_to_jiffies(m: 1000)); |
| 330 | } |
| 331 | |
| 332 | static int psz_kmsg_recover_data(struct psz_context *cxt) |
| 333 | { |
| 334 | struct pstore_zone_info *info = cxt->pstore_zone_info; |
| 335 | struct pstore_zone *zone = NULL; |
| 336 | struct psz_buffer *buf; |
| 337 | unsigned long i; |
| 338 | ssize_t rcnt; |
| 339 | |
| 340 | if (!info->read) |
| 341 | return -EINVAL; |
| 342 | |
| 343 | for (i = 0; i < cxt->kmsg_max_cnt; i++) { |
| 344 | zone = cxt->kpszs[i]; |
| 345 | if (unlikely(!zone)) |
| 346 | return -EINVAL; |
| 347 | if (atomic_read(v: &zone->dirty)) { |
| 348 | unsigned int wcnt = cxt->kmsg_write_cnt; |
| 349 | struct pstore_zone *new = cxt->kpszs[wcnt]; |
| 350 | int ret; |
| 351 | |
| 352 | ret = psz_move_zone(old: zone, new); |
| 353 | if (ret) { |
| 354 | pr_err("move zone from %lu to %d failed\n" , |
| 355 | i, wcnt); |
| 356 | return ret; |
| 357 | } |
| 358 | cxt->kmsg_write_cnt = (wcnt + 1) % cxt->kmsg_max_cnt; |
| 359 | } |
| 360 | if (!zone->should_recover) |
| 361 | continue; |
| 362 | buf = zone->buffer; |
| 363 | rcnt = info->read((char *)buf, zone->buffer_size + sizeof(*buf), |
| 364 | zone->off); |
| 365 | if (rcnt != zone->buffer_size + sizeof(*buf)) |
| 366 | return rcnt < 0 ? rcnt : -EIO; |
| 367 | } |
| 368 | return 0; |
| 369 | } |
| 370 | |
| 371 | static int psz_kmsg_recover_meta(struct psz_context *cxt) |
| 372 | { |
| 373 | struct pstore_zone_info *info = cxt->pstore_zone_info; |
| 374 | struct pstore_zone *zone; |
| 375 | ssize_t rcnt, len; |
| 376 | struct psz_buffer *buf; |
| 377 | struct psz_kmsg_header *hdr; |
| 378 | struct timespec64 time = { }; |
| 379 | unsigned long i; |
| 380 | /* |
| 381 | * Recover may happen on panic, we can't allocate any memory by kmalloc. |
| 382 | * So, we use local array instead. |
| 383 | */ |
| 384 | char [sizeof(*buf) + sizeof(*hdr)] = {0}; |
| 385 | |
| 386 | if (!info->read) |
| 387 | return -EINVAL; |
| 388 | |
| 389 | len = sizeof(*buf) + sizeof(*hdr); |
| 390 | buf = (struct psz_buffer *)buffer_header; |
| 391 | for (i = 0; i < cxt->kmsg_max_cnt; i++) { |
| 392 | zone = cxt->kpszs[i]; |
| 393 | if (unlikely(!zone)) |
| 394 | return -EINVAL; |
| 395 | |
| 396 | rcnt = info->read((char *)buf, len, zone->off); |
| 397 | if (rcnt == -ENOMSG) { |
| 398 | pr_debug("%s with id %lu may be broken, skip\n" , |
| 399 | zone->name, i); |
| 400 | continue; |
| 401 | } else if (rcnt != len) { |
| 402 | pr_err("read %s with id %lu failed\n" , zone->name, i); |
| 403 | return rcnt < 0 ? rcnt : -EIO; |
| 404 | } |
| 405 | |
| 406 | if (buf->sig != zone->buffer->sig) { |
| 407 | pr_debug("no valid data in kmsg dump zone %lu\n" , i); |
| 408 | continue; |
| 409 | } |
| 410 | |
| 411 | if (zone->buffer_size < atomic_read(v: &buf->datalen)) { |
| 412 | pr_info("found overtop zone: %s: id %lu, off %lld, size %zu\n" , |
| 413 | zone->name, i, zone->off, |
| 414 | zone->buffer_size); |
| 415 | continue; |
| 416 | } |
| 417 | |
| 418 | hdr = (struct psz_kmsg_header *)buf->data; |
| 419 | if (hdr->magic != PSTORE_KMSG_HEADER_MAGIC) { |
| 420 | pr_info("found invalid zone: %s: id %lu, off %lld, size %zu\n" , |
| 421 | zone->name, i, zone->off, |
| 422 | zone->buffer_size); |
| 423 | continue; |
| 424 | } |
| 425 | |
| 426 | /* |
| 427 | * we get the newest zone, and the next one must be the oldest |
| 428 | * or unused zone, because we do write one by one like a circle. |
| 429 | */ |
| 430 | if (hdr->time.tv_sec >= time.tv_sec) { |
| 431 | time.tv_sec = hdr->time.tv_sec; |
| 432 | cxt->kmsg_write_cnt = (i + 1) % cxt->kmsg_max_cnt; |
| 433 | } |
| 434 | |
| 435 | if (hdr->reason == KMSG_DUMP_OOPS) |
| 436 | cxt->oops_counter = |
| 437 | max(cxt->oops_counter, hdr->counter); |
| 438 | else if (hdr->reason == KMSG_DUMP_PANIC) |
| 439 | cxt->panic_counter = |
| 440 | max(cxt->panic_counter, hdr->counter); |
| 441 | |
| 442 | if (!atomic_read(v: &buf->datalen)) { |
| 443 | pr_debug("found erased zone: %s: id %lu, off %lld, size %zu, datalen %d\n" , |
| 444 | zone->name, i, zone->off, |
| 445 | zone->buffer_size, |
| 446 | atomic_read(&buf->datalen)); |
| 447 | continue; |
| 448 | } |
| 449 | |
| 450 | if (!is_on_panic()) |
| 451 | zone->should_recover = true; |
| 452 | pr_debug("found nice zone: %s: id %lu, off %lld, size %zu, datalen %d\n" , |
| 453 | zone->name, i, zone->off, |
| 454 | zone->buffer_size, atomic_read(&buf->datalen)); |
| 455 | } |
| 456 | |
| 457 | return 0; |
| 458 | } |
| 459 | |
| 460 | static int psz_kmsg_recover(struct psz_context *cxt) |
| 461 | { |
| 462 | int ret; |
| 463 | |
| 464 | if (!cxt->kpszs) |
| 465 | return 0; |
| 466 | |
| 467 | ret = psz_kmsg_recover_meta(cxt); |
| 468 | if (ret) |
| 469 | goto recover_fail; |
| 470 | |
| 471 | ret = psz_kmsg_recover_data(cxt); |
| 472 | if (ret) |
| 473 | goto recover_fail; |
| 474 | |
| 475 | return 0; |
| 476 | recover_fail: |
| 477 | pr_debug("psz_recover_kmsg failed\n" ); |
| 478 | return ret; |
| 479 | } |
| 480 | |
| 481 | static int psz_recover_zone(struct psz_context *cxt, struct pstore_zone *zone) |
| 482 | { |
| 483 | struct pstore_zone_info *info = cxt->pstore_zone_info; |
| 484 | struct psz_buffer *oldbuf, tmpbuf; |
| 485 | int ret = 0; |
| 486 | char *buf; |
| 487 | ssize_t rcnt, len, start, off; |
| 488 | |
| 489 | if (!zone || zone->oldbuf) |
| 490 | return 0; |
| 491 | |
| 492 | if (is_on_panic()) { |
| 493 | /* save data as much as possible */ |
| 494 | psz_flush_dirty_zone(zone); |
| 495 | return 0; |
| 496 | } |
| 497 | |
| 498 | if (unlikely(!info->read)) |
| 499 | return -EINVAL; |
| 500 | |
| 501 | len = sizeof(struct psz_buffer); |
| 502 | rcnt = info->read((char *)&tmpbuf, len, zone->off); |
| 503 | if (rcnt != len) { |
| 504 | pr_debug("read zone %s failed\n" , zone->name); |
| 505 | return rcnt < 0 ? rcnt : -EIO; |
| 506 | } |
| 507 | |
| 508 | if (tmpbuf.sig != zone->buffer->sig) { |
| 509 | pr_debug("no valid data in zone %s\n" , zone->name); |
| 510 | return 0; |
| 511 | } |
| 512 | |
| 513 | if (zone->buffer_size < atomic_read(v: &tmpbuf.datalen) || |
| 514 | zone->buffer_size < atomic_read(v: &tmpbuf.start)) { |
| 515 | pr_info("found overtop zone: %s: off %lld, size %zu\n" , |
| 516 | zone->name, zone->off, zone->buffer_size); |
| 517 | /* just keep going */ |
| 518 | return 0; |
| 519 | } |
| 520 | |
| 521 | if (!atomic_read(v: &tmpbuf.datalen)) { |
| 522 | pr_debug("found erased zone: %s: off %lld, size %zu, datalen %d\n" , |
| 523 | zone->name, zone->off, zone->buffer_size, |
| 524 | atomic_read(&tmpbuf.datalen)); |
| 525 | return 0; |
| 526 | } |
| 527 | |
| 528 | pr_debug("found nice zone: %s: off %lld, size %zu, datalen %d\n" , |
| 529 | zone->name, zone->off, zone->buffer_size, |
| 530 | atomic_read(&tmpbuf.datalen)); |
| 531 | |
| 532 | len = atomic_read(v: &tmpbuf.datalen) + sizeof(*oldbuf); |
| 533 | oldbuf = kzalloc(len, GFP_KERNEL); |
| 534 | if (!oldbuf) |
| 535 | return -ENOMEM; |
| 536 | |
| 537 | memcpy(oldbuf, &tmpbuf, sizeof(*oldbuf)); |
| 538 | buf = (char *)oldbuf + sizeof(*oldbuf); |
| 539 | len = atomic_read(v: &oldbuf->datalen); |
| 540 | start = atomic_read(v: &oldbuf->start); |
| 541 | off = zone->off + sizeof(*oldbuf); |
| 542 | |
| 543 | /* get part of data */ |
| 544 | rcnt = info->read(buf, len - start, off + start); |
| 545 | if (rcnt != len - start) { |
| 546 | pr_err("read zone %s failed\n" , zone->name); |
| 547 | ret = rcnt < 0 ? rcnt : -EIO; |
| 548 | goto free_oldbuf; |
| 549 | } |
| 550 | |
| 551 | /* get the rest of data */ |
| 552 | rcnt = info->read(buf + len - start, start, off); |
| 553 | if (rcnt != start) { |
| 554 | pr_err("read zone %s failed\n" , zone->name); |
| 555 | ret = rcnt < 0 ? rcnt : -EIO; |
| 556 | goto free_oldbuf; |
| 557 | } |
| 558 | |
| 559 | zone->oldbuf = oldbuf; |
| 560 | psz_flush_dirty_zone(zone); |
| 561 | return 0; |
| 562 | |
| 563 | free_oldbuf: |
| 564 | kfree(objp: oldbuf); |
| 565 | return ret; |
| 566 | } |
| 567 | |
| 568 | static int psz_recover_zones(struct psz_context *cxt, |
| 569 | struct pstore_zone **zones, unsigned int cnt) |
| 570 | { |
| 571 | int ret; |
| 572 | unsigned int i; |
| 573 | struct pstore_zone *zone; |
| 574 | |
| 575 | if (!zones) |
| 576 | return 0; |
| 577 | |
| 578 | for (i = 0; i < cnt; i++) { |
| 579 | zone = zones[i]; |
| 580 | if (unlikely(!zone)) |
| 581 | continue; |
| 582 | ret = psz_recover_zone(cxt, zone); |
| 583 | if (ret) |
| 584 | goto recover_fail; |
| 585 | } |
| 586 | |
| 587 | return 0; |
| 588 | recover_fail: |
| 589 | pr_debug("recover %s[%u] failed\n" , zone->name, i); |
| 590 | return ret; |
| 591 | } |
| 592 | |
| 593 | /** |
| 594 | * psz_recovery() - recover data from storage |
| 595 | * @cxt: the context of pstore/zone |
| 596 | * |
| 597 | * recovery means reading data back from storage after rebooting |
| 598 | * |
| 599 | * Return: 0 on success, others on failure. |
| 600 | */ |
| 601 | static inline int psz_recovery(struct psz_context *cxt) |
| 602 | { |
| 603 | int ret; |
| 604 | |
| 605 | if (atomic_read(v: &cxt->recovered)) |
| 606 | return 0; |
| 607 | |
| 608 | ret = psz_kmsg_recover(cxt); |
| 609 | if (ret) |
| 610 | goto out; |
| 611 | |
| 612 | ret = psz_recover_zone(cxt, zone: cxt->ppsz); |
| 613 | if (ret) |
| 614 | goto out; |
| 615 | |
| 616 | ret = psz_recover_zone(cxt, zone: cxt->cpsz); |
| 617 | if (ret) |
| 618 | goto out; |
| 619 | |
| 620 | ret = psz_recover_zones(cxt, zones: cxt->fpszs, cnt: cxt->ftrace_max_cnt); |
| 621 | |
| 622 | out: |
| 623 | if (unlikely(ret)) |
| 624 | pr_err("recover failed\n" ); |
| 625 | else { |
| 626 | pr_debug("recover end!\n" ); |
| 627 | atomic_set(v: &cxt->recovered, i: 1); |
| 628 | } |
| 629 | return ret; |
| 630 | } |
| 631 | |
| 632 | static int psz_pstore_open(struct pstore_info *psi) |
| 633 | { |
| 634 | struct psz_context *cxt = psi->data; |
| 635 | |
| 636 | cxt->kmsg_read_cnt = 0; |
| 637 | cxt->pmsg_read_cnt = 0; |
| 638 | cxt->console_read_cnt = 0; |
| 639 | cxt->ftrace_read_cnt = 0; |
| 640 | return 0; |
| 641 | } |
| 642 | |
| 643 | static inline bool psz_old_ok(struct pstore_zone *zone) |
| 644 | { |
| 645 | if (zone && zone->oldbuf && atomic_read(v: &zone->oldbuf->datalen)) |
| 646 | return true; |
| 647 | return false; |
| 648 | } |
| 649 | |
| 650 | static inline bool psz_ok(struct pstore_zone *zone) |
| 651 | { |
| 652 | if (zone && zone->buffer && buffer_datalen(zone)) |
| 653 | return true; |
| 654 | return false; |
| 655 | } |
| 656 | |
| 657 | static inline int psz_kmsg_erase(struct psz_context *cxt, |
| 658 | struct pstore_zone *zone, struct pstore_record *record) |
| 659 | { |
| 660 | struct psz_buffer *buffer = zone->buffer; |
| 661 | struct psz_kmsg_header *hdr = |
| 662 | (struct psz_kmsg_header *)buffer->data; |
| 663 | size_t size; |
| 664 | |
| 665 | if (unlikely(!psz_ok(zone))) |
| 666 | return 0; |
| 667 | |
| 668 | /* this zone is already updated, no need to erase */ |
| 669 | if (record->count != hdr->counter) |
| 670 | return 0; |
| 671 | |
| 672 | size = buffer_datalen(zone) + sizeof(*zone->buffer); |
| 673 | atomic_set(v: &zone->buffer->datalen, i: 0); |
| 674 | if (cxt->pstore_zone_info->erase) |
| 675 | return cxt->pstore_zone_info->erase(size, zone->off); |
| 676 | else |
| 677 | return psz_zone_write(zone, flush_mode: FLUSH_META, NULL, len: 0, off: 0); |
| 678 | } |
| 679 | |
| 680 | static inline int psz_record_erase(struct psz_context *cxt, |
| 681 | struct pstore_zone *zone) |
| 682 | { |
| 683 | if (unlikely(!psz_old_ok(zone))) |
| 684 | return 0; |
| 685 | |
| 686 | kfree(objp: zone->oldbuf); |
| 687 | zone->oldbuf = NULL; |
| 688 | /* |
| 689 | * if there are new data in zone buffer, that means the old data |
| 690 | * are already invalid. It is no need to flush 0 (erase) to |
| 691 | * block device. |
| 692 | */ |
| 693 | if (!buffer_datalen(zone)) |
| 694 | return psz_zone_write(zone, flush_mode: FLUSH_META, NULL, len: 0, off: 0); |
| 695 | psz_flush_dirty_zone(zone); |
| 696 | return 0; |
| 697 | } |
| 698 | |
| 699 | static int psz_pstore_erase(struct pstore_record *record) |
| 700 | { |
| 701 | struct psz_context *cxt = record->psi->data; |
| 702 | |
| 703 | switch (record->type) { |
| 704 | case PSTORE_TYPE_DMESG: |
| 705 | if (record->id >= cxt->kmsg_max_cnt) |
| 706 | return -EINVAL; |
| 707 | return psz_kmsg_erase(cxt, zone: cxt->kpszs[record->id], record); |
| 708 | case PSTORE_TYPE_PMSG: |
| 709 | return psz_record_erase(cxt, zone: cxt->ppsz); |
| 710 | case PSTORE_TYPE_CONSOLE: |
| 711 | return psz_record_erase(cxt, zone: cxt->cpsz); |
| 712 | case PSTORE_TYPE_FTRACE: |
| 713 | if (record->id >= cxt->ftrace_max_cnt) |
| 714 | return -EINVAL; |
| 715 | return psz_record_erase(cxt, zone: cxt->fpszs[record->id]); |
| 716 | default: return -EINVAL; |
| 717 | } |
| 718 | } |
| 719 | |
| 720 | static void psz_write_kmsg_hdr(struct pstore_zone *zone, |
| 721 | struct pstore_record *record) |
| 722 | { |
| 723 | struct psz_context *cxt = record->psi->data; |
| 724 | struct psz_buffer *buffer = zone->buffer; |
| 725 | struct psz_kmsg_header *hdr = |
| 726 | (struct psz_kmsg_header *)buffer->data; |
| 727 | |
| 728 | hdr->magic = PSTORE_KMSG_HEADER_MAGIC; |
| 729 | hdr->compressed = record->compressed; |
| 730 | hdr->time.tv_sec = record->time.tv_sec; |
| 731 | hdr->time.tv_nsec = record->time.tv_nsec; |
| 732 | hdr->reason = record->reason; |
| 733 | if (hdr->reason == KMSG_DUMP_OOPS) |
| 734 | hdr->counter = ++cxt->oops_counter; |
| 735 | else if (hdr->reason == KMSG_DUMP_PANIC) |
| 736 | hdr->counter = ++cxt->panic_counter; |
| 737 | else |
| 738 | hdr->counter = 0; |
| 739 | } |
| 740 | |
| 741 | /* |
| 742 | * In case zone is broken, which may occur to MTD device, we try each zones, |
| 743 | * start at cxt->kmsg_write_cnt. |
| 744 | */ |
| 745 | static inline int notrace psz_kmsg_write_record(struct psz_context *cxt, |
| 746 | struct pstore_record *record) |
| 747 | { |
| 748 | size_t size, hlen; |
| 749 | struct pstore_zone *zone; |
| 750 | unsigned int i; |
| 751 | |
| 752 | for (i = 0; i < cxt->kmsg_max_cnt; i++) { |
| 753 | unsigned int zonenum, len; |
| 754 | int ret; |
| 755 | |
| 756 | zonenum = (cxt->kmsg_write_cnt + i) % cxt->kmsg_max_cnt; |
| 757 | zone = cxt->kpszs[zonenum]; |
| 758 | if (unlikely(!zone)) |
| 759 | return -ENOSPC; |
| 760 | |
| 761 | /* avoid destroying old data, allocate a new one */ |
| 762 | len = zone->buffer_size + sizeof(*zone->buffer); |
| 763 | zone->oldbuf = zone->buffer; |
| 764 | zone->buffer = kzalloc(len, GFP_ATOMIC); |
| 765 | if (!zone->buffer) { |
| 766 | zone->buffer = zone->oldbuf; |
| 767 | return -ENOMEM; |
| 768 | } |
| 769 | zone->buffer->sig = zone->oldbuf->sig; |
| 770 | |
| 771 | pr_debug("write %s to zone id %d\n" , zone->name, zonenum); |
| 772 | psz_write_kmsg_hdr(zone, record); |
| 773 | hlen = sizeof(struct psz_kmsg_header); |
| 774 | size = min_t(size_t, record->size, zone->buffer_size - hlen); |
| 775 | ret = psz_zone_write(zone, flush_mode: FLUSH_ALL, buf: record->buf, len: size, off: hlen); |
| 776 | if (likely(!ret || ret != -ENOMSG)) { |
| 777 | cxt->kmsg_write_cnt = zonenum + 1; |
| 778 | cxt->kmsg_write_cnt %= cxt->kmsg_max_cnt; |
| 779 | /* no need to try next zone, free last zone buffer */ |
| 780 | kfree(objp: zone->oldbuf); |
| 781 | zone->oldbuf = NULL; |
| 782 | return ret; |
| 783 | } |
| 784 | |
| 785 | pr_debug("zone %u may be broken, try next dmesg zone\n" , |
| 786 | zonenum); |
| 787 | kfree(objp: zone->buffer); |
| 788 | zone->buffer = zone->oldbuf; |
| 789 | zone->oldbuf = NULL; |
| 790 | } |
| 791 | |
| 792 | return -EBUSY; |
| 793 | } |
| 794 | |
| 795 | static int notrace psz_kmsg_write(struct psz_context *cxt, |
| 796 | struct pstore_record *record) |
| 797 | { |
| 798 | int ret; |
| 799 | |
| 800 | /* |
| 801 | * Explicitly only take the first part of any new crash. |
| 802 | * If our buffer is larger than kmsg_bytes, this can never happen, |
| 803 | * and if our buffer is smaller than kmsg_bytes, we don't want the |
| 804 | * report split across multiple records. |
| 805 | */ |
| 806 | if (record->part != 1) |
| 807 | return -ENOSPC; |
| 808 | |
| 809 | if (!cxt->kpszs) |
| 810 | return -ENOSPC; |
| 811 | |
| 812 | ret = psz_kmsg_write_record(cxt, record); |
| 813 | if (!ret && is_on_panic()) { |
| 814 | /* ensure all data are flushed to storage when panic */ |
| 815 | pr_debug("try to flush other dirty zones\n" ); |
| 816 | psz_flush_all_dirty_zones(NULL); |
| 817 | } |
| 818 | |
| 819 | /* always return 0 as we had handled it on buffer */ |
| 820 | return 0; |
| 821 | } |
| 822 | |
| 823 | static int notrace psz_record_write(struct pstore_zone *zone, |
| 824 | struct pstore_record *record) |
| 825 | { |
| 826 | size_t start, rem; |
| 827 | bool is_full_data = false; |
| 828 | char *buf; |
| 829 | int cnt; |
| 830 | |
| 831 | if (!zone || !record) |
| 832 | return -ENOSPC; |
| 833 | |
| 834 | if (atomic_read(v: &zone->buffer->datalen) >= zone->buffer_size) |
| 835 | is_full_data = true; |
| 836 | |
| 837 | cnt = record->size; |
| 838 | buf = record->buf; |
| 839 | if (unlikely(cnt > zone->buffer_size)) { |
| 840 | buf += cnt - zone->buffer_size; |
| 841 | cnt = zone->buffer_size; |
| 842 | } |
| 843 | |
| 844 | start = buffer_start(zone); |
| 845 | rem = zone->buffer_size - start; |
| 846 | if (unlikely(rem < cnt)) { |
| 847 | psz_zone_write(zone, flush_mode: FLUSH_PART, buf, len: rem, off: start); |
| 848 | buf += rem; |
| 849 | cnt -= rem; |
| 850 | start = 0; |
| 851 | is_full_data = true; |
| 852 | } |
| 853 | |
| 854 | atomic_set(v: &zone->buffer->start, i: cnt + start); |
| 855 | psz_zone_write(zone, flush_mode: FLUSH_PART, buf, len: cnt, off: start); |
| 856 | |
| 857 | /** |
| 858 | * psz_zone_write will set datalen as start + cnt. |
| 859 | * It works if actual data length is lesser than buffer size. |
| 860 | * If data length is greater than buffer size, pmsg will rewrite to |
| 861 | * the beginning of the zone, which makes buffer->datalen wrong. |
| 862 | * So we should reset datalen as buffer size once actual data length |
| 863 | * is greater than buffer size. |
| 864 | */ |
| 865 | if (is_full_data) { |
| 866 | atomic_set(v: &zone->buffer->datalen, i: zone->buffer_size); |
| 867 | psz_zone_write(zone, flush_mode: FLUSH_META, NULL, len: 0, off: 0); |
| 868 | } |
| 869 | return 0; |
| 870 | } |
| 871 | |
| 872 | static int notrace psz_pstore_write(struct pstore_record *record) |
| 873 | { |
| 874 | struct psz_context *cxt = record->psi->data; |
| 875 | |
| 876 | if (record->type == PSTORE_TYPE_DMESG && |
| 877 | record->reason == KMSG_DUMP_PANIC) |
| 878 | atomic_set(v: &cxt->on_panic, i: 1); |
| 879 | |
| 880 | /* |
| 881 | * If on panic, do not write anything except panic records. |
| 882 | * Fix the case when panic_write prints log that wakes up |
| 883 | * console backend. |
| 884 | */ |
| 885 | if (is_on_panic() && record->type != PSTORE_TYPE_DMESG) |
| 886 | return -EBUSY; |
| 887 | |
| 888 | switch (record->type) { |
| 889 | case PSTORE_TYPE_DMESG: |
| 890 | return psz_kmsg_write(cxt, record); |
| 891 | case PSTORE_TYPE_CONSOLE: |
| 892 | return psz_record_write(zone: cxt->cpsz, record); |
| 893 | case PSTORE_TYPE_PMSG: |
| 894 | return psz_record_write(zone: cxt->ppsz, record); |
| 895 | case PSTORE_TYPE_FTRACE: { |
| 896 | int zonenum = smp_processor_id(); |
| 897 | |
| 898 | if (!cxt->fpszs) |
| 899 | return -ENOSPC; |
| 900 | return psz_record_write(zone: cxt->fpszs[zonenum], record); |
| 901 | } |
| 902 | default: |
| 903 | return -EINVAL; |
| 904 | } |
| 905 | } |
| 906 | |
| 907 | static struct pstore_zone *psz_read_next_zone(struct psz_context *cxt) |
| 908 | { |
| 909 | struct pstore_zone *zone = NULL; |
| 910 | |
| 911 | while (cxt->kmsg_read_cnt < cxt->kmsg_max_cnt) { |
| 912 | zone = cxt->kpszs[cxt->kmsg_read_cnt++]; |
| 913 | if (psz_ok(zone)) |
| 914 | return zone; |
| 915 | } |
| 916 | |
| 917 | if (cxt->ftrace_read_cnt < cxt->ftrace_max_cnt) |
| 918 | /* |
| 919 | * No need psz_old_ok(). Let psz_ftrace_read() do so for |
| 920 | * combination. psz_ftrace_read() should traverse over |
| 921 | * all zones in case of some zone without data. |
| 922 | */ |
| 923 | return cxt->fpszs[cxt->ftrace_read_cnt++]; |
| 924 | |
| 925 | if (cxt->pmsg_read_cnt == 0) { |
| 926 | cxt->pmsg_read_cnt++; |
| 927 | zone = cxt->ppsz; |
| 928 | if (psz_old_ok(zone)) |
| 929 | return zone; |
| 930 | } |
| 931 | |
| 932 | if (cxt->console_read_cnt == 0) { |
| 933 | cxt->console_read_cnt++; |
| 934 | zone = cxt->cpsz; |
| 935 | if (psz_old_ok(zone)) |
| 936 | return zone; |
| 937 | } |
| 938 | |
| 939 | return NULL; |
| 940 | } |
| 941 | |
| 942 | static int psz_kmsg_read_hdr(struct pstore_zone *zone, |
| 943 | struct pstore_record *record) |
| 944 | { |
| 945 | struct psz_buffer *buffer = zone->buffer; |
| 946 | struct psz_kmsg_header *hdr = |
| 947 | (struct psz_kmsg_header *)buffer->data; |
| 948 | |
| 949 | if (hdr->magic != PSTORE_KMSG_HEADER_MAGIC) |
| 950 | return -EINVAL; |
| 951 | record->compressed = hdr->compressed; |
| 952 | record->time.tv_sec = hdr->time.tv_sec; |
| 953 | record->time.tv_nsec = hdr->time.tv_nsec; |
| 954 | record->reason = hdr->reason; |
| 955 | record->count = hdr->counter; |
| 956 | return 0; |
| 957 | } |
| 958 | |
| 959 | static ssize_t psz_kmsg_read(struct pstore_zone *zone, |
| 960 | struct pstore_record *record) |
| 961 | { |
| 962 | ssize_t size, hlen = 0; |
| 963 | |
| 964 | size = buffer_datalen(zone); |
| 965 | /* Clear and skip this kmsg dump record if it has no valid header */ |
| 966 | if (psz_kmsg_read_hdr(zone, record)) { |
| 967 | atomic_set(v: &zone->buffer->datalen, i: 0); |
| 968 | atomic_set(v: &zone->dirty, i: 0); |
| 969 | return -ENOMSG; |
| 970 | } |
| 971 | size -= sizeof(struct psz_kmsg_header); |
| 972 | |
| 973 | if (!record->compressed) { |
| 974 | char *buf = kasprintf(GFP_KERNEL, fmt: "%s: Total %d times\n" , |
| 975 | kmsg_dump_reason_str(reason: record->reason), |
| 976 | record->count); |
| 977 | if (!buf) |
| 978 | return -ENOMEM; |
| 979 | hlen = strlen(buf); |
| 980 | record->buf = krealloc(buf, hlen + size, GFP_KERNEL); |
| 981 | if (!record->buf) { |
| 982 | kfree(objp: buf); |
| 983 | return -ENOMEM; |
| 984 | } |
| 985 | } else { |
| 986 | record->buf = kmalloc(size, GFP_KERNEL); |
| 987 | if (!record->buf) |
| 988 | return -ENOMEM; |
| 989 | } |
| 990 | |
| 991 | size = psz_zone_read_buffer(zone, buf: record->buf + hlen, len: size, |
| 992 | off: sizeof(struct psz_kmsg_header)); |
| 993 | if (unlikely(size < 0)) { |
| 994 | kfree(objp: record->buf); |
| 995 | return -ENOMSG; |
| 996 | } |
| 997 | |
| 998 | return size + hlen; |
| 999 | } |
| 1000 | |
| 1001 | /* try to combine all ftrace zones */ |
| 1002 | static ssize_t psz_ftrace_read(struct pstore_zone *zone, |
| 1003 | struct pstore_record *record) |
| 1004 | { |
| 1005 | struct psz_context *cxt; |
| 1006 | struct psz_buffer *buf; |
| 1007 | int ret; |
| 1008 | |
| 1009 | if (!zone || !record) |
| 1010 | return -ENOSPC; |
| 1011 | |
| 1012 | if (!psz_old_ok(zone)) |
| 1013 | goto out; |
| 1014 | |
| 1015 | buf = (struct psz_buffer *)zone->oldbuf; |
| 1016 | if (!buf) |
| 1017 | return -ENOMSG; |
| 1018 | |
| 1019 | ret = pstore_ftrace_combine_log(dest_log: &record->buf, dest_log_size: &record->size, |
| 1020 | src_log: (char *)buf->data, src_log_size: atomic_read(v: &buf->datalen)); |
| 1021 | if (unlikely(ret)) |
| 1022 | return ret; |
| 1023 | |
| 1024 | out: |
| 1025 | cxt = record->psi->data; |
| 1026 | if (cxt->ftrace_read_cnt < cxt->ftrace_max_cnt) |
| 1027 | /* then, read next ftrace zone */ |
| 1028 | return -ENOMSG; |
| 1029 | record->id = 0; |
| 1030 | return record->size ? record->size : -ENOMSG; |
| 1031 | } |
| 1032 | |
| 1033 | static ssize_t psz_record_read(struct pstore_zone *zone, |
| 1034 | struct pstore_record *record) |
| 1035 | { |
| 1036 | size_t len; |
| 1037 | struct psz_buffer *buf; |
| 1038 | |
| 1039 | if (!zone || !record) |
| 1040 | return -ENOSPC; |
| 1041 | |
| 1042 | buf = (struct psz_buffer *)zone->oldbuf; |
| 1043 | if (!buf) |
| 1044 | return -ENOMSG; |
| 1045 | |
| 1046 | len = atomic_read(v: &buf->datalen); |
| 1047 | record->buf = kmalloc(len, GFP_KERNEL); |
| 1048 | if (!record->buf) |
| 1049 | return -ENOMEM; |
| 1050 | |
| 1051 | if (unlikely(psz_zone_read_oldbuf(zone, record->buf, len, 0))) { |
| 1052 | kfree(objp: record->buf); |
| 1053 | return -ENOMSG; |
| 1054 | } |
| 1055 | |
| 1056 | return len; |
| 1057 | } |
| 1058 | |
| 1059 | static ssize_t psz_pstore_read(struct pstore_record *record) |
| 1060 | { |
| 1061 | struct psz_context *cxt = record->psi->data; |
| 1062 | ssize_t (*readop)(struct pstore_zone *zone, |
| 1063 | struct pstore_record *record); |
| 1064 | struct pstore_zone *zone; |
| 1065 | ssize_t ret; |
| 1066 | |
| 1067 | /* before read, we must recover from storage */ |
| 1068 | ret = psz_recovery(cxt); |
| 1069 | if (ret) |
| 1070 | return ret; |
| 1071 | |
| 1072 | next_zone: |
| 1073 | zone = psz_read_next_zone(cxt); |
| 1074 | if (!zone) |
| 1075 | return 0; |
| 1076 | |
| 1077 | record->type = zone->type; |
| 1078 | switch (record->type) { |
| 1079 | case PSTORE_TYPE_DMESG: |
| 1080 | readop = psz_kmsg_read; |
| 1081 | record->id = cxt->kmsg_read_cnt - 1; |
| 1082 | break; |
| 1083 | case PSTORE_TYPE_FTRACE: |
| 1084 | readop = psz_ftrace_read; |
| 1085 | break; |
| 1086 | case PSTORE_TYPE_CONSOLE: |
| 1087 | case PSTORE_TYPE_PMSG: |
| 1088 | readop = psz_record_read; |
| 1089 | break; |
| 1090 | default: |
| 1091 | goto next_zone; |
| 1092 | } |
| 1093 | |
| 1094 | ret = readop(zone, record); |
| 1095 | if (ret == -ENOMSG) |
| 1096 | goto next_zone; |
| 1097 | return ret; |
| 1098 | } |
| 1099 | |
| 1100 | static struct psz_context pstore_zone_cxt = { |
| 1101 | .pstore_zone_info_lock = |
| 1102 | __MUTEX_INITIALIZER(pstore_zone_cxt.pstore_zone_info_lock), |
| 1103 | .recovered = ATOMIC_INIT(0), |
| 1104 | .on_panic = ATOMIC_INIT(0), |
| 1105 | .pstore = { |
| 1106 | .owner = THIS_MODULE, |
| 1107 | .open = psz_pstore_open, |
| 1108 | .read = psz_pstore_read, |
| 1109 | .write = psz_pstore_write, |
| 1110 | .erase = psz_pstore_erase, |
| 1111 | }, |
| 1112 | }; |
| 1113 | |
| 1114 | static void psz_free_zone(struct pstore_zone **pszone) |
| 1115 | { |
| 1116 | struct pstore_zone *zone = *pszone; |
| 1117 | |
| 1118 | if (!zone) |
| 1119 | return; |
| 1120 | |
| 1121 | kfree(objp: zone->buffer); |
| 1122 | kfree(objp: zone); |
| 1123 | *pszone = NULL; |
| 1124 | } |
| 1125 | |
| 1126 | static void psz_free_zones(struct pstore_zone ***pszones, unsigned int *cnt) |
| 1127 | { |
| 1128 | struct pstore_zone **zones = *pszones; |
| 1129 | |
| 1130 | if (!zones) |
| 1131 | return; |
| 1132 | |
| 1133 | while (*cnt > 0) { |
| 1134 | (*cnt)--; |
| 1135 | psz_free_zone(pszone: &(zones[*cnt])); |
| 1136 | } |
| 1137 | kfree(objp: zones); |
| 1138 | *pszones = NULL; |
| 1139 | } |
| 1140 | |
| 1141 | static void psz_free_all_zones(struct psz_context *cxt) |
| 1142 | { |
| 1143 | if (cxt->kpszs) |
| 1144 | psz_free_zones(pszones: &cxt->kpszs, cnt: &cxt->kmsg_max_cnt); |
| 1145 | if (cxt->ppsz) |
| 1146 | psz_free_zone(pszone: &cxt->ppsz); |
| 1147 | if (cxt->cpsz) |
| 1148 | psz_free_zone(pszone: &cxt->cpsz); |
| 1149 | if (cxt->fpszs) |
| 1150 | psz_free_zones(pszones: &cxt->fpszs, cnt: &cxt->ftrace_max_cnt); |
| 1151 | } |
| 1152 | |
| 1153 | static struct pstore_zone *psz_init_zone(enum pstore_type_id type, |
| 1154 | loff_t *off, size_t size) |
| 1155 | { |
| 1156 | struct pstore_zone_info *info = pstore_zone_cxt.pstore_zone_info; |
| 1157 | struct pstore_zone *zone; |
| 1158 | const char *name = pstore_type_to_name(type); |
| 1159 | |
| 1160 | if (!size) |
| 1161 | return NULL; |
| 1162 | |
| 1163 | if (*off + size > info->total_size) { |
| 1164 | pr_err("no room for %s (0x%zx@0x%llx over 0x%lx)\n" , |
| 1165 | name, size, *off, info->total_size); |
| 1166 | return ERR_PTR(error: -ENOMEM); |
| 1167 | } |
| 1168 | |
| 1169 | zone = kzalloc(sizeof(struct pstore_zone), GFP_KERNEL); |
| 1170 | if (!zone) |
| 1171 | return ERR_PTR(error: -ENOMEM); |
| 1172 | |
| 1173 | zone->buffer = kmalloc(size, GFP_KERNEL); |
| 1174 | if (!zone->buffer) { |
| 1175 | kfree(objp: zone); |
| 1176 | return ERR_PTR(error: -ENOMEM); |
| 1177 | } |
| 1178 | memset(zone->buffer, 0xFF, size); |
| 1179 | zone->off = *off; |
| 1180 | zone->name = name; |
| 1181 | zone->type = type; |
| 1182 | zone->buffer_size = size - sizeof(struct psz_buffer); |
| 1183 | zone->buffer->sig = type ^ PSZ_SIG; |
| 1184 | zone->oldbuf = NULL; |
| 1185 | atomic_set(v: &zone->dirty, i: 0); |
| 1186 | atomic_set(v: &zone->buffer->datalen, i: 0); |
| 1187 | atomic_set(v: &zone->buffer->start, i: 0); |
| 1188 | |
| 1189 | *off += size; |
| 1190 | |
| 1191 | pr_debug("pszone %s: off 0x%llx, %zu header, %zu data\n" , zone->name, |
| 1192 | zone->off, sizeof(*zone->buffer), zone->buffer_size); |
| 1193 | return zone; |
| 1194 | } |
| 1195 | |
| 1196 | static struct pstore_zone **psz_init_zones(enum pstore_type_id type, |
| 1197 | loff_t *off, size_t total_size, ssize_t record_size, |
| 1198 | unsigned int *cnt) |
| 1199 | { |
| 1200 | struct pstore_zone_info *info = pstore_zone_cxt.pstore_zone_info; |
| 1201 | struct pstore_zone **zones, *zone; |
| 1202 | const char *name = pstore_type_to_name(type); |
| 1203 | int c, i; |
| 1204 | |
| 1205 | *cnt = 0; |
| 1206 | if (!total_size || !record_size) |
| 1207 | return NULL; |
| 1208 | |
| 1209 | if (*off + total_size > info->total_size) { |
| 1210 | pr_err("no room for zones %s (0x%zx@0x%llx over 0x%lx)\n" , |
| 1211 | name, total_size, *off, info->total_size); |
| 1212 | return ERR_PTR(error: -ENOMEM); |
| 1213 | } |
| 1214 | |
| 1215 | c = total_size / record_size; |
| 1216 | if (unlikely(!c)) { |
| 1217 | pr_err("zone %s total_size too small\n" , name); |
| 1218 | return ERR_PTR(error: -EINVAL); |
| 1219 | } |
| 1220 | |
| 1221 | zones = kcalloc(c, sizeof(*zones), GFP_KERNEL); |
| 1222 | if (!zones) { |
| 1223 | pr_err("allocate for zones %s failed\n" , name); |
| 1224 | return ERR_PTR(error: -ENOMEM); |
| 1225 | } |
| 1226 | |
| 1227 | for (i = 0; i < c; i++) { |
| 1228 | zone = psz_init_zone(type, off, size: record_size); |
| 1229 | if (!zone || IS_ERR(ptr: zone)) { |
| 1230 | pr_err("initialize zones %s failed\n" , name); |
| 1231 | psz_free_zones(pszones: &zones, cnt: &i); |
| 1232 | return (void *)zone; |
| 1233 | } |
| 1234 | zones[i] = zone; |
| 1235 | } |
| 1236 | |
| 1237 | *cnt = c; |
| 1238 | return zones; |
| 1239 | } |
| 1240 | |
| 1241 | static int psz_alloc_zones(struct psz_context *cxt) |
| 1242 | { |
| 1243 | struct pstore_zone_info *info = cxt->pstore_zone_info; |
| 1244 | loff_t off = 0; |
| 1245 | int err; |
| 1246 | size_t off_size = 0; |
| 1247 | |
| 1248 | off_size += info->pmsg_size; |
| 1249 | cxt->ppsz = psz_init_zone(type: PSTORE_TYPE_PMSG, off: &off, size: info->pmsg_size); |
| 1250 | if (IS_ERR(ptr: cxt->ppsz)) { |
| 1251 | err = PTR_ERR(ptr: cxt->ppsz); |
| 1252 | cxt->ppsz = NULL; |
| 1253 | goto free_out; |
| 1254 | } |
| 1255 | |
| 1256 | off_size += info->console_size; |
| 1257 | cxt->cpsz = psz_init_zone(type: PSTORE_TYPE_CONSOLE, off: &off, |
| 1258 | size: info->console_size); |
| 1259 | if (IS_ERR(ptr: cxt->cpsz)) { |
| 1260 | err = PTR_ERR(ptr: cxt->cpsz); |
| 1261 | cxt->cpsz = NULL; |
| 1262 | goto free_out; |
| 1263 | } |
| 1264 | |
| 1265 | off_size += info->ftrace_size; |
| 1266 | cxt->fpszs = psz_init_zones(type: PSTORE_TYPE_FTRACE, off: &off, |
| 1267 | total_size: info->ftrace_size, |
| 1268 | record_size: info->ftrace_size / nr_cpu_ids, |
| 1269 | cnt: &cxt->ftrace_max_cnt); |
| 1270 | if (IS_ERR(ptr: cxt->fpszs)) { |
| 1271 | err = PTR_ERR(ptr: cxt->fpszs); |
| 1272 | cxt->fpszs = NULL; |
| 1273 | goto free_out; |
| 1274 | } |
| 1275 | |
| 1276 | cxt->kpszs = psz_init_zones(type: PSTORE_TYPE_DMESG, off: &off, |
| 1277 | total_size: info->total_size - off_size, |
| 1278 | record_size: info->kmsg_size, cnt: &cxt->kmsg_max_cnt); |
| 1279 | if (IS_ERR(ptr: cxt->kpszs)) { |
| 1280 | err = PTR_ERR(ptr: cxt->kpszs); |
| 1281 | cxt->kpszs = NULL; |
| 1282 | goto free_out; |
| 1283 | } |
| 1284 | |
| 1285 | return 0; |
| 1286 | free_out: |
| 1287 | psz_free_all_zones(cxt); |
| 1288 | return err; |
| 1289 | } |
| 1290 | |
| 1291 | /** |
| 1292 | * register_pstore_zone() - register to pstore/zone |
| 1293 | * |
| 1294 | * @info: back-end driver information. See &struct pstore_zone_info. |
| 1295 | * |
| 1296 | * Only one back-end at one time. |
| 1297 | * |
| 1298 | * Return: 0 on success, others on failure. |
| 1299 | */ |
| 1300 | int register_pstore_zone(struct pstore_zone_info *info) |
| 1301 | { |
| 1302 | int err = -EINVAL; |
| 1303 | struct psz_context *cxt = &pstore_zone_cxt; |
| 1304 | |
| 1305 | if (info->total_size < 4096) { |
| 1306 | pr_warn("total_size must be >= 4096\n" ); |
| 1307 | return -EINVAL; |
| 1308 | } |
| 1309 | if (info->total_size > SZ_128M) { |
| 1310 | pr_warn("capping size to 128MiB\n" ); |
| 1311 | info->total_size = SZ_128M; |
| 1312 | } |
| 1313 | |
| 1314 | if (!info->kmsg_size && !info->pmsg_size && !info->console_size && |
| 1315 | !info->ftrace_size) { |
| 1316 | pr_warn("at least one record size must be non-zero\n" ); |
| 1317 | return -EINVAL; |
| 1318 | } |
| 1319 | |
| 1320 | if (!info->name || !info->name[0]) |
| 1321 | return -EINVAL; |
| 1322 | |
| 1323 | #define check_size(name, size) { \ |
| 1324 | if (info->name > 0 && info->name < (size)) { \ |
| 1325 | pr_err(#name " must be over %d\n", (size)); \ |
| 1326 | return -EINVAL; \ |
| 1327 | } \ |
| 1328 | if (info->name & (size - 1)) { \ |
| 1329 | pr_err(#name " must be a multiple of %d\n", \ |
| 1330 | (size)); \ |
| 1331 | return -EINVAL; \ |
| 1332 | } \ |
| 1333 | } |
| 1334 | |
| 1335 | check_size(total_size, 4096); |
| 1336 | check_size(kmsg_size, SECTOR_SIZE); |
| 1337 | check_size(pmsg_size, SECTOR_SIZE); |
| 1338 | check_size(console_size, SECTOR_SIZE); |
| 1339 | check_size(ftrace_size, SECTOR_SIZE); |
| 1340 | |
| 1341 | #undef check_size |
| 1342 | |
| 1343 | /* |
| 1344 | * the @read and @write must be applied. |
| 1345 | * if no @read, pstore may mount failed. |
| 1346 | * if no @write, pstore do not support to remove record file. |
| 1347 | */ |
| 1348 | if (!info->read || !info->write) { |
| 1349 | pr_err("no valid general read/write interface\n" ); |
| 1350 | return -EINVAL; |
| 1351 | } |
| 1352 | |
| 1353 | mutex_lock(&cxt->pstore_zone_info_lock); |
| 1354 | if (cxt->pstore_zone_info) { |
| 1355 | pr_warn("'%s' already loaded: ignoring '%s'\n" , |
| 1356 | cxt->pstore_zone_info->name, info->name); |
| 1357 | mutex_unlock(lock: &cxt->pstore_zone_info_lock); |
| 1358 | return -EBUSY; |
| 1359 | } |
| 1360 | cxt->pstore_zone_info = info; |
| 1361 | |
| 1362 | pr_debug("register %s with properties:\n" , info->name); |
| 1363 | pr_debug("\ttotal size : %ld Bytes\n" , info->total_size); |
| 1364 | pr_debug("\tkmsg size : %ld Bytes\n" , info->kmsg_size); |
| 1365 | pr_debug("\tpmsg size : %ld Bytes\n" , info->pmsg_size); |
| 1366 | pr_debug("\tconsole size : %ld Bytes\n" , info->console_size); |
| 1367 | pr_debug("\tftrace size : %ld Bytes\n" , info->ftrace_size); |
| 1368 | |
| 1369 | err = psz_alloc_zones(cxt); |
| 1370 | if (err) { |
| 1371 | pr_err("alloc zones failed\n" ); |
| 1372 | goto fail_out; |
| 1373 | } |
| 1374 | |
| 1375 | if (info->kmsg_size) { |
| 1376 | cxt->pstore.bufsize = cxt->kpszs[0]->buffer_size - |
| 1377 | sizeof(struct psz_kmsg_header); |
| 1378 | cxt->pstore.buf = kzalloc(cxt->pstore.bufsize, GFP_KERNEL); |
| 1379 | if (!cxt->pstore.buf) { |
| 1380 | err = -ENOMEM; |
| 1381 | goto fail_free; |
| 1382 | } |
| 1383 | } |
| 1384 | cxt->pstore.data = cxt; |
| 1385 | |
| 1386 | pr_info("registered %s as backend for" , info->name); |
| 1387 | cxt->pstore.max_reason = info->max_reason; |
| 1388 | cxt->pstore.name = info->name; |
| 1389 | if (info->kmsg_size) { |
| 1390 | cxt->pstore.flags |= PSTORE_FLAGS_DMESG; |
| 1391 | pr_cont(" kmsg(%s" , |
| 1392 | kmsg_dump_reason_str(cxt->pstore.max_reason)); |
| 1393 | if (cxt->pstore_zone_info->panic_write) |
| 1394 | pr_cont(",panic_write" ); |
| 1395 | pr_cont(")" ); |
| 1396 | } |
| 1397 | if (info->pmsg_size) { |
| 1398 | cxt->pstore.flags |= PSTORE_FLAGS_PMSG; |
| 1399 | pr_cont(" pmsg" ); |
| 1400 | } |
| 1401 | if (info->console_size) { |
| 1402 | cxt->pstore.flags |= PSTORE_FLAGS_CONSOLE; |
| 1403 | pr_cont(" console" ); |
| 1404 | } |
| 1405 | if (info->ftrace_size) { |
| 1406 | cxt->pstore.flags |= PSTORE_FLAGS_FTRACE; |
| 1407 | pr_cont(" ftrace" ); |
| 1408 | } |
| 1409 | pr_cont("\n" ); |
| 1410 | |
| 1411 | err = pstore_register(&cxt->pstore); |
| 1412 | if (err) { |
| 1413 | pr_err("registering with pstore failed\n" ); |
| 1414 | goto fail_free; |
| 1415 | } |
| 1416 | mutex_unlock(lock: &pstore_zone_cxt.pstore_zone_info_lock); |
| 1417 | |
| 1418 | return 0; |
| 1419 | |
| 1420 | fail_free: |
| 1421 | kfree(objp: cxt->pstore.buf); |
| 1422 | cxt->pstore.buf = NULL; |
| 1423 | cxt->pstore.bufsize = 0; |
| 1424 | psz_free_all_zones(cxt); |
| 1425 | fail_out: |
| 1426 | pstore_zone_cxt.pstore_zone_info = NULL; |
| 1427 | mutex_unlock(lock: &pstore_zone_cxt.pstore_zone_info_lock); |
| 1428 | return err; |
| 1429 | } |
| 1430 | EXPORT_SYMBOL_GPL(register_pstore_zone); |
| 1431 | |
| 1432 | /** |
| 1433 | * unregister_pstore_zone() - unregister to pstore/zone |
| 1434 | * |
| 1435 | * @info: back-end driver information. See struct pstore_zone_info. |
| 1436 | */ |
| 1437 | void unregister_pstore_zone(struct pstore_zone_info *info) |
| 1438 | { |
| 1439 | struct psz_context *cxt = &pstore_zone_cxt; |
| 1440 | |
| 1441 | mutex_lock(&cxt->pstore_zone_info_lock); |
| 1442 | if (!cxt->pstore_zone_info) { |
| 1443 | mutex_unlock(lock: &cxt->pstore_zone_info_lock); |
| 1444 | return; |
| 1445 | } |
| 1446 | |
| 1447 | /* Stop incoming writes from pstore. */ |
| 1448 | pstore_unregister(&cxt->pstore); |
| 1449 | |
| 1450 | /* Flush any pending writes. */ |
| 1451 | psz_flush_all_dirty_zones(NULL); |
| 1452 | flush_delayed_work(dwork: &psz_cleaner); |
| 1453 | |
| 1454 | /* Clean up allocations. */ |
| 1455 | kfree(objp: cxt->pstore.buf); |
| 1456 | cxt->pstore.buf = NULL; |
| 1457 | cxt->pstore.bufsize = 0; |
| 1458 | cxt->pstore_zone_info = NULL; |
| 1459 | |
| 1460 | psz_free_all_zones(cxt); |
| 1461 | |
| 1462 | /* Clear counters and zone state. */ |
| 1463 | cxt->oops_counter = 0; |
| 1464 | cxt->panic_counter = 0; |
| 1465 | atomic_set(v: &cxt->recovered, i: 0); |
| 1466 | atomic_set(v: &cxt->on_panic, i: 0); |
| 1467 | |
| 1468 | mutex_unlock(lock: &cxt->pstore_zone_info_lock); |
| 1469 | } |
| 1470 | EXPORT_SYMBOL_GPL(unregister_pstore_zone); |
| 1471 | |
| 1472 | MODULE_LICENSE("GPL" ); |
| 1473 | MODULE_AUTHOR("WeiXiong Liao <liaoweixiong@allwinnertech.com>" ); |
| 1474 | MODULE_AUTHOR("Kees Cook <keescook@chromium.org>" ); |
| 1475 | MODULE_DESCRIPTION("Storage Manager for pstore/blk" ); |
| 1476 | |