| 1 | /* |
| 2 | * Persistent Storage - platform driver interface parts. |
| 3 | * |
| 4 | * Copyright (C) 2007-2008 Google, Inc. |
| 5 | * Copyright (C) 2010 Intel Corporation <tony.luck@intel.com> |
| 6 | * |
| 7 | * This program is free software; you can redistribute it and/or modify |
| 8 | * it under the terms of the GNU General Public License version 2 as |
| 9 | * published by the Free Software Foundation. |
| 10 | * |
| 11 | * This program is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 14 | * GNU General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU General Public License |
| 17 | * along with this program; if not, write to the Free Software |
| 18 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
| 19 | */ |
| 20 | |
| 21 | #define pr_fmt(fmt) "pstore: " fmt |
| 22 | |
| 23 | #include <linux/atomic.h> |
| 24 | #include <linux/types.h> |
| 25 | #include <linux/errno.h> |
| 26 | #include <linux/init.h> |
| 27 | #include <linux/kmsg_dump.h> |
| 28 | #include <linux/console.h> |
| 29 | #include <linux/module.h> |
| 30 | #include <linux/pstore.h> |
| 31 | #include <linux/zlib.h> |
| 32 | #include <linux/string.h> |
| 33 | #include <linux/timer.h> |
| 34 | #include <linux/slab.h> |
| 35 | #include <linux/uaccess.h> |
| 36 | #include <linux/hardirq.h> |
| 37 | #include <linux/jiffies.h> |
| 38 | #include <linux/workqueue.h> |
| 39 | |
| 40 | #include "internal.h" |
| 41 | |
| 42 | /* |
| 43 | * We defer making "oops" entries appear in pstore - see |
| 44 | * whether the system is actually still running well enough |
| 45 | * to let someone see the entry |
| 46 | */ |
| 47 | static int pstore_update_ms = -1; |
| 48 | module_param_named(update_ms, pstore_update_ms, int, 0600); |
| 49 | MODULE_PARM_DESC(update_ms, "milliseconds before pstore updates its content " |
| 50 | "(default is -1, which means runtime updates are disabled; " |
| 51 | "enabling this option is not safe, it may lead to further " |
| 52 | "corruption on Oopses)"); |
| 53 | |
| 54 | static int pstore_new_entry; |
| 55 | |
| 56 | static void pstore_timefunc(unsigned long); |
| 57 | static DEFINE_TIMER(pstore_timer, pstore_timefunc, 0, 0); |
| 58 | |
| 59 | static void pstore_dowork(struct work_struct *); |
| 60 | static DECLARE_WORK(pstore_work, pstore_dowork); |
| 61 | |
| 62 | /* |
| 63 | * pstore_lock just protects "psinfo" during |
| 64 | * calls to pstore_register() |
| 65 | */ |
| 66 | static DEFINE_SPINLOCK(pstore_lock); |
| 67 | struct pstore_info *psinfo; |
| 68 | |
| 69 | static char *backend; |
| 70 | |
| 71 | /* Compression parameters */ |
| 72 | #define COMPR_LEVEL 6 |
| 73 | #define WINDOW_BITS 12 |
| 74 | #define MEM_LEVEL 4 |
| 75 | static struct z_stream_s stream; |
| 76 | |
| 77 | static char *big_oops_buf; |
| 78 | static size_t big_oops_buf_sz; |
| 79 | |
| 80 | /* How much of the console log to snapshot */ |
| 81 | static unsigned long kmsg_bytes = 10240; |
| 82 | |
| 83 | void pstore_set_kmsg_bytes(int bytes) |
| 84 | { |
| 85 | kmsg_bytes = bytes; |
| 86 | } |
| 87 | |
| 88 | /* Tag each group of saved records with a sequence number */ |
| 89 | static int oopscount; |
| 90 | |
| 91 | static const char *get_reason_str(enum kmsg_dump_reason reason) |
| 92 | { |
| 93 | switch (reason) { |
| 94 | case KMSG_DUMP_PANIC: |
| 95 | return "Panic"; |
| 96 | case KMSG_DUMP_OOPS: |
| 97 | return "Oops"; |
| 98 | case KMSG_DUMP_EMERG: |
| 99 | return "Emergency"; |
| 100 | case KMSG_DUMP_RESTART: |
| 101 | return "Restart"; |
| 102 | case KMSG_DUMP_HALT: |
| 103 | return "Halt"; |
| 104 | case KMSG_DUMP_POWEROFF: |
| 105 | return "Poweroff"; |
| 106 | default: |
| 107 | return "Unknown"; |
| 108 | } |
| 109 | } |
| 110 | |
| 111 | bool pstore_cannot_block_path(enum kmsg_dump_reason reason) |
| 112 | { |
| 113 | /* |
| 114 | * In case of NMI path, pstore shouldn't be blocked |
| 115 | * regardless of reason. |
| 116 | */ |
| 117 | if (in_nmi()) |
| 118 | return true; |
| 119 | |
| 120 | switch (reason) { |
| 121 | /* In panic case, other cpus are stopped by smp_send_stop(). */ |
| 122 | case KMSG_DUMP_PANIC: |
| 123 | /* Emergency restart shouldn't be blocked by spin lock. */ |
| 124 | case KMSG_DUMP_EMERG: |
| 125 | return true; |
| 126 | default: |
| 127 | return false; |
| 128 | } |
| 129 | } |
| 130 | EXPORT_SYMBOL_GPL(pstore_cannot_block_path); |
| 131 | |
| 132 | /* Derived from logfs_compress() */ |
| 133 | static int pstore_compress(const void *in, void *out, size_t inlen, |
| 134 | size_t outlen) |
| 135 | { |
| 136 | int err, ret; |
| 137 | |
| 138 | ret = -EIO; |
| 139 | err = zlib_deflateInit2(&stream, COMPR_LEVEL, Z_DEFLATED, WINDOW_BITS, |
| 140 | MEM_LEVEL, Z_DEFAULT_STRATEGY); |
| 141 | if (err != Z_OK) |
| 142 | goto error; |
| 143 | |
| 144 | stream.next_in = in; |
| 145 | stream.avail_in = inlen; |
| 146 | stream.total_in = 0; |
| 147 | stream.next_out = out; |
| 148 | stream.avail_out = outlen; |
| 149 | stream.total_out = 0; |
| 150 | |
| 151 | err = zlib_deflate(&stream, Z_FINISH); |
| 152 | if (err != Z_STREAM_END) |
| 153 | goto error; |
| 154 | |
| 155 | err = zlib_deflateEnd(&stream); |
| 156 | if (err != Z_OK) |
| 157 | goto error; |
| 158 | |
| 159 | if (stream.total_out >= stream.total_in) |
| 160 | goto error; |
| 161 | |
| 162 | ret = stream.total_out; |
| 163 | error: |
| 164 | return ret; |
| 165 | } |
| 166 | |
| 167 | /* Derived from logfs_uncompress */ |
| 168 | static int pstore_decompress(void *in, void *out, size_t inlen, size_t outlen) |
| 169 | { |
| 170 | int err, ret; |
| 171 | |
| 172 | ret = -EIO; |
| 173 | err = zlib_inflateInit2(&stream, WINDOW_BITS); |
| 174 | if (err != Z_OK) |
| 175 | goto error; |
| 176 | |
| 177 | stream.next_in = in; |
| 178 | stream.avail_in = inlen; |
| 179 | stream.total_in = 0; |
| 180 | stream.next_out = out; |
| 181 | stream.avail_out = outlen; |
| 182 | stream.total_out = 0; |
| 183 | |
| 184 | err = zlib_inflate(&stream, Z_FINISH); |
| 185 | if (err != Z_STREAM_END) |
| 186 | goto error; |
| 187 | |
| 188 | err = zlib_inflateEnd(&stream); |
| 189 | if (err != Z_OK) |
| 190 | goto error; |
| 191 | |
| 192 | ret = stream.total_out; |
| 193 | error: |
| 194 | return ret; |
| 195 | } |
| 196 | |
| 197 | static void allocate_buf_for_compression(void) |
| 198 | { |
| 199 | size_t size; |
| 200 | size_t cmpr; |
| 201 | |
| 202 | switch (psinfo->bufsize) { |
| 203 | /* buffer range for efivars */ |
| 204 | case 1000 ... 2000: |
| 205 | cmpr = 56; |
| 206 | break; |
| 207 | case 2001 ... 3000: |
| 208 | cmpr = 54; |
| 209 | break; |
| 210 | case 3001 ... 3999: |
| 211 | cmpr = 52; |
| 212 | break; |
| 213 | /* buffer range for nvram, erst */ |
| 214 | case 4000 ... 10000: |
| 215 | cmpr = 45; |
| 216 | break; |
| 217 | default: |
| 218 | cmpr = 60; |
| 219 | break; |
| 220 | } |
| 221 | |
| 222 | big_oops_buf_sz = (psinfo->bufsize * 100) / cmpr; |
| 223 | big_oops_buf = kmalloc(big_oops_buf_sz, GFP_KERNEL); |
| 224 | if (big_oops_buf) { |
| 225 | size = max(zlib_deflate_workspacesize(WINDOW_BITS, MEM_LEVEL), |
| 226 | zlib_inflate_workspacesize()); |
| 227 | stream.workspace = kmalloc(size, GFP_KERNEL); |
| 228 | if (!stream.workspace) { |
| 229 | pr_err("No memory for compression workspace; skipping compression\n"); |
| 230 | kfree(big_oops_buf); |
| 231 | big_oops_buf = NULL; |
| 232 | } |
| 233 | } else { |
| 234 | pr_err("No memory for uncompressed data; skipping compression\n"); |
| 235 | stream.workspace = NULL; |
| 236 | } |
| 237 | |
| 238 | } |
| 239 | |
| 240 | static void free_buf_for_compression(void) |
| 241 | { |
| 242 | kfree(stream.workspace); |
| 243 | stream.workspace = NULL; |
| 244 | kfree(big_oops_buf); |
| 245 | big_oops_buf = NULL; |
| 246 | } |
| 247 | |
| 248 | /* |
| 249 | * Called when compression fails, since the printk buffer |
| 250 | * would be fetched for compression calling it again when |
| 251 | * compression fails would have moved the iterator of |
| 252 | * printk buffer which results in fetching old contents. |
| 253 | * Copy the recent messages from big_oops_buf to psinfo->buf |
| 254 | */ |
| 255 | static size_t copy_kmsg_to_buffer(int hsize, size_t len) |
| 256 | { |
| 257 | size_t total_len; |
| 258 | size_t diff; |
| 259 | |
| 260 | total_len = hsize + len; |
| 261 | |
| 262 | if (total_len > psinfo->bufsize) { |
| 263 | diff = total_len - psinfo->bufsize + hsize; |
| 264 | memcpy(psinfo->buf, big_oops_buf, hsize); |
| 265 | memcpy(psinfo->buf + hsize, big_oops_buf + diff, |
| 266 | psinfo->bufsize - hsize); |
| 267 | total_len = psinfo->bufsize; |
| 268 | } else |
| 269 | memcpy(psinfo->buf, big_oops_buf, total_len); |
| 270 | |
| 271 | return total_len; |
| 272 | } |
| 273 | |
| 274 | /* |
| 275 | * callback from kmsg_dump. (s2,l2) has the most recently |
| 276 | * written bytes, older bytes are in (s1,l1). Save as much |
| 277 | * as we can from the end of the buffer. |
| 278 | */ |
| 279 | static void pstore_dump(struct kmsg_dumper *dumper, |
| 280 | enum kmsg_dump_reason reason) |
| 281 | { |
| 282 | unsigned long total = 0; |
| 283 | const char *why; |
| 284 | u64 id; |
| 285 | unsigned int part = 1; |
| 286 | unsigned long flags = 0; |
| 287 | int is_locked = 0; |
| 288 | int ret; |
| 289 | |
| 290 | why = get_reason_str(reason); |
| 291 | |
| 292 | if (pstore_cannot_block_path(reason)) { |
| 293 | is_locked = spin_trylock_irqsave(&psinfo->buf_lock, flags); |
| 294 | if (!is_locked) { |
| 295 | pr_err("pstore dump routine blocked in %s path, may corrupt error record\n" |
| 296 | , in_nmi() ? "NMI" : why); |
| 297 | } |
| 298 | } else |
| 299 | spin_lock_irqsave(&psinfo->buf_lock, flags); |
| 300 | oopscount++; |
| 301 | while (total < kmsg_bytes) { |
| 302 | char *dst; |
| 303 | unsigned long size; |
| 304 | int hsize; |
| 305 | int zipped_len = -1; |
| 306 | size_t len; |
| 307 | bool compressed; |
| 308 | size_t total_len; |
| 309 | |
| 310 | if (big_oops_buf && is_locked) { |
| 311 | dst = big_oops_buf; |
| 312 | hsize = sprintf(dst, "%s#%d Part%u\n", why, |
| 313 | oopscount, part); |
| 314 | size = big_oops_buf_sz - hsize; |
| 315 | |
| 316 | if (!kmsg_dump_get_buffer(dumper, true, dst + hsize, |
| 317 | size, &len)) |
| 318 | break; |
| 319 | |
| 320 | zipped_len = pstore_compress(dst, psinfo->buf, |
| 321 | hsize + len, psinfo->bufsize); |
| 322 | |
| 323 | if (zipped_len > 0) { |
| 324 | compressed = true; |
| 325 | total_len = zipped_len; |
| 326 | } else { |
| 327 | compressed = false; |
| 328 | total_len = copy_kmsg_to_buffer(hsize, len); |
| 329 | } |
| 330 | } else { |
| 331 | dst = psinfo->buf; |
| 332 | hsize = sprintf(dst, "%s#%d Part%u\n", why, oopscount, |
| 333 | part); |
| 334 | size = psinfo->bufsize - hsize; |
| 335 | dst += hsize; |
| 336 | |
| 337 | if (!kmsg_dump_get_buffer(dumper, true, dst, |
| 338 | size, &len)) |
| 339 | break; |
| 340 | |
| 341 | compressed = false; |
| 342 | total_len = hsize + len; |
| 343 | } |
| 344 | |
| 345 | ret = psinfo->write(PSTORE_TYPE_DMESG, reason, &id, part, |
| 346 | oopscount, compressed, total_len, psinfo); |
| 347 | if (ret == 0 && reason == KMSG_DUMP_OOPS && pstore_is_mounted()) |
| 348 | pstore_new_entry = 1; |
| 349 | |
| 350 | total += total_len; |
| 351 | part++; |
| 352 | } |
| 353 | if (pstore_cannot_block_path(reason)) { |
| 354 | if (is_locked) |
| 355 | spin_unlock_irqrestore(&psinfo->buf_lock, flags); |
| 356 | } else |
| 357 | spin_unlock_irqrestore(&psinfo->buf_lock, flags); |
| 358 | } |
| 359 | |
| 360 | static struct kmsg_dumper pstore_dumper = { |
| 361 | .dump = pstore_dump, |
| 362 | }; |
| 363 | |
| 364 | /* |
| 365 | * Register with kmsg_dump to save last part of console log on panic. |
| 366 | */ |
| 367 | static void pstore_register_kmsg(void) |
| 368 | { |
| 369 | kmsg_dump_register(&pstore_dumper); |
| 370 | } |
| 371 | |
| 372 | static void pstore_unregister_kmsg(void) |
| 373 | { |
| 374 | kmsg_dump_unregister(&pstore_dumper); |
| 375 | } |
| 376 | |
| 377 | #ifdef CONFIG_PSTORE_CONSOLE |
| 378 | static void pstore_console_write(struct console *con, const char *s, unsigned c) |
| 379 | { |
| 380 | const char *e = s + c; |
| 381 | |
| 382 | while (s < e) { |
| 383 | unsigned long flags; |
| 384 | u64 id; |
| 385 | |
| 386 | if (c > psinfo->bufsize) |
| 387 | c = psinfo->bufsize; |
| 388 | |
| 389 | if (oops_in_progress) { |
| 390 | if (!spin_trylock_irqsave(&psinfo->buf_lock, flags)) |
| 391 | break; |
| 392 | } else { |
| 393 | spin_lock_irqsave(&psinfo->buf_lock, flags); |
| 394 | } |
| 395 | memcpy(psinfo->buf, s, c); |
| 396 | psinfo->write(PSTORE_TYPE_CONSOLE, 0, &id, 0, 0, 0, c, psinfo); |
| 397 | spin_unlock_irqrestore(&psinfo->buf_lock, flags); |
| 398 | s += c; |
| 399 | c = e - s; |
| 400 | } |
| 401 | } |
| 402 | |
| 403 | static struct console pstore_console = { |
| 404 | .name = "pstore", |
| 405 | .write = pstore_console_write, |
| 406 | .flags = CON_PRINTBUFFER | CON_ENABLED | CON_ANYTIME, |
| 407 | .index = -1, |
| 408 | }; |
| 409 | |
| 410 | static void pstore_register_console(void) |
| 411 | { |
| 412 | register_console(&pstore_console); |
| 413 | } |
| 414 | |
| 415 | static void pstore_unregister_console(void) |
| 416 | { |
| 417 | unregister_console(&pstore_console); |
| 418 | } |
| 419 | #else |
| 420 | static void pstore_register_console(void) {} |
| 421 | static void pstore_unregister_console(void) {} |
| 422 | #endif |
| 423 | |
| 424 | static int pstore_write_compat(enum pstore_type_id type, |
| 425 | enum kmsg_dump_reason reason, |
| 426 | u64 *id, unsigned int part, int count, |
| 427 | bool compressed, size_t size, |
| 428 | struct pstore_info *psi) |
| 429 | { |
| 430 | return psi->write_buf(type, reason, id, part, psinfo->buf, compressed, |
| 431 | size, psi); |
| 432 | } |
| 433 | |
| 434 | /* |
| 435 | * platform specific persistent storage driver registers with |
| 436 | * us here. If pstore is already mounted, call the platform |
| 437 | * read function right away to populate the file system. If not |
| 438 | * then the pstore mount code will call us later to fill out |
| 439 | * the file system. |
| 440 | */ |
| 441 | int pstore_register(struct pstore_info *psi) |
| 442 | { |
| 443 | struct module *owner = psi->owner; |
| 444 | |
| 445 | if (backend && strcmp(backend, psi->name)) |
| 446 | return -EPERM; |
| 447 | |
| 448 | spin_lock(&pstore_lock); |
| 449 | if (psinfo) { |
| 450 | spin_unlock(&pstore_lock); |
| 451 | return -EBUSY; |
| 452 | } |
| 453 | |
| 454 | if (!psi->write) |
| 455 | psi->write = pstore_write_compat; |
| 456 | psinfo = psi; |
| 457 | mutex_init(&psinfo->read_mutex); |
| 458 | spin_unlock(&pstore_lock); |
| 459 | |
| 460 | if (owner && !try_module_get(owner)) { |
| 461 | psinfo = NULL; |
| 462 | return -EINVAL; |
| 463 | } |
| 464 | |
| 465 | allocate_buf_for_compression(); |
| 466 | |
| 467 | if (pstore_is_mounted()) |
| 468 | pstore_get_records(0); |
| 469 | |
| 470 | pstore_register_kmsg(); |
| 471 | |
| 472 | if ((psi->flags & PSTORE_FLAGS_FRAGILE) == 0) { |
| 473 | pstore_register_console(); |
| 474 | pstore_register_ftrace(); |
| 475 | pstore_register_pmsg(); |
| 476 | } |
| 477 | |
| 478 | if (pstore_update_ms >= 0) { |
| 479 | pstore_timer.expires = jiffies + |
| 480 | msecs_to_jiffies(pstore_update_ms); |
| 481 | add_timer(&pstore_timer); |
| 482 | } |
| 483 | |
| 484 | /* |
| 485 | * Update the module parameter backend, so it is visible |
| 486 | * through /sys/module/pstore/parameters/backend |
| 487 | */ |
| 488 | backend = psi->name; |
| 489 | |
| 490 | module_put(owner); |
| 491 | |
| 492 | pr_info("Registered %s as persistent store backend\n", psi->name); |
| 493 | |
| 494 | return 0; |
| 495 | } |
| 496 | EXPORT_SYMBOL_GPL(pstore_register); |
| 497 | |
| 498 | void pstore_unregister(struct pstore_info *psi) |
| 499 | { |
| 500 | pstore_unregister_pmsg(); |
| 501 | pstore_unregister_ftrace(); |
| 502 | pstore_unregister_console(); |
| 503 | pstore_unregister_kmsg(); |
| 504 | |
| 505 | free_buf_for_compression(); |
| 506 | |
| 507 | psinfo = NULL; |
| 508 | backend = NULL; |
| 509 | } |
| 510 | EXPORT_SYMBOL_GPL(pstore_unregister); |
| 511 | |
| 512 | /* |
| 513 | * Read all the records from the persistent store. Create |
| 514 | * files in our filesystem. Don't warn about -EEXIST errors |
| 515 | * when we are re-scanning the backing store looking to add new |
| 516 | * error records. |
| 517 | */ |
| 518 | void pstore_get_records(int quiet) |
| 519 | { |
| 520 | struct pstore_info *psi = psinfo; |
| 521 | char *buf = NULL; |
| 522 | ssize_t size; |
| 523 | u64 id; |
| 524 | int count; |
| 525 | enum pstore_type_id type; |
| 526 | struct timespec time; |
| 527 | int failed = 0, rc; |
| 528 | bool compressed; |
| 529 | int unzipped_len = -1; |
| 530 | |
| 531 | if (!psi) |
| 532 | return; |
| 533 | |
| 534 | mutex_lock(&psi->read_mutex); |
| 535 | if (psi->open && psi->open(psi)) |
| 536 | goto out; |
| 537 | |
| 538 | while ((size = psi->read(&id, &type, &count, &time, &buf, &compressed, |
| 539 | psi)) > 0) { |
| 540 | if (compressed && (type == PSTORE_TYPE_DMESG)) { |
| 541 | if (big_oops_buf) |
| 542 | unzipped_len = pstore_decompress(buf, |
| 543 | big_oops_buf, size, |
| 544 | big_oops_buf_sz); |
| 545 | |
| 546 | if (unzipped_len > 0) { |
| 547 | kfree(buf); |
| 548 | buf = big_oops_buf; |
| 549 | size = unzipped_len; |
| 550 | compressed = false; |
| 551 | } else { |
| 552 | pr_err("decompression failed;returned %d\n", |
| 553 | unzipped_len); |
| 554 | compressed = true; |
| 555 | } |
| 556 | } |
| 557 | rc = pstore_mkfile(type, psi->name, id, count, buf, |
| 558 | compressed, (size_t)size, time, psi); |
| 559 | if (unzipped_len < 0) { |
| 560 | /* Free buffer other than big oops */ |
| 561 | kfree(buf); |
| 562 | buf = NULL; |
| 563 | } else |
| 564 | unzipped_len = -1; |
| 565 | if (rc && (rc != -EEXIST || !quiet)) |
| 566 | failed++; |
| 567 | } |
| 568 | if (psi->close) |
| 569 | psi->close(psi); |
| 570 | out: |
| 571 | mutex_unlock(&psi->read_mutex); |
| 572 | |
| 573 | if (failed) |
| 574 | pr_warn("failed to load %d record(s) from '%s'\n", |
| 575 | failed, psi->name); |
| 576 | } |
| 577 | |
| 578 | static void pstore_dowork(struct work_struct *work) |
| 579 | { |
| 580 | pstore_get_records(1); |
| 581 | } |
| 582 | |
| 583 | static void pstore_timefunc(unsigned long dummy) |
| 584 | { |
| 585 | if (pstore_new_entry) { |
| 586 | pstore_new_entry = 0; |
| 587 | schedule_work(&pstore_work); |
| 588 | } |
| 589 | |
| 590 | mod_timer(&pstore_timer, jiffies + msecs_to_jiffies(pstore_update_ms)); |
| 591 | } |
| 592 | |
| 593 | module_param(backend, charp, 0444); |
| 594 | MODULE_PARM_DESC(backend, "Pstore backend to use"); |