1 // SPDX-License-Identifier: GPL-2.0-only
3 * Persistent Storage - platform driver interface parts.
5 * Copyright (C) 2007-2008 Google, Inc.
6 * Copyright (C) 2010 Intel Corporation <tony.luck@intel.com>
9 #define pr_fmt(fmt) "pstore: " fmt
11 #include <linux/atomic.h>
12 #include <linux/types.h>
13 #include <linux/errno.h>
14 #include <linux/init.h>
15 #include <linux/kmsg_dump.h>
16 #include <linux/console.h>
18 #include <linux/module.h>
19 #include <linux/pstore.h>
20 #include <linux/string.h>
21 #include <linux/timer.h>
22 #include <linux/slab.h>
23 #include <linux/uaccess.h>
24 #include <linux/jiffies.h>
25 #include <linux/vmalloc.h>
26 #include <linux/workqueue.h>
27 #include <linux/zlib.h>
32 * We defer making "oops" entries appear in pstore - see
33 * whether the system is actually still running well enough
34 * to let someone see the entry
36 static int pstore_update_ms = -1;
37 module_param_named(update_ms, pstore_update_ms, int, 0600);
38 MODULE_PARM_DESC(update_ms, "milliseconds before pstore updates its content "
39 "(default is -1, which means runtime updates are disabled; "
40 "enabling this option may not be safe; it may lead to further "
41 "corruption on Oopses)");
43 /* Names should be in the same order as the enum pstore_type_id */
44 static const char * const pstore_type_names[] = {
56 static int pstore_new_entry;
58 static void pstore_timefunc(struct timer_list *);
59 static DEFINE_TIMER(pstore_timer, pstore_timefunc);
61 static void pstore_dowork(struct work_struct *);
62 static DECLARE_WORK(pstore_work, pstore_dowork);
65 * psinfo_lock protects "psinfo" during calls to
66 * pstore_register(), pstore_unregister(), and
67 * the filesystem mount/unmount routines.
69 static DEFINE_MUTEX(psinfo_lock);
70 struct pstore_info *psinfo;
73 module_param(backend, charp, 0444);
74 MODULE_PARM_DESC(backend, "specific backend to use");
77 * pstore no longer implements compression via the crypto API, and only
78 * supports zlib deflate compression implemented using the zlib library
79 * interface. This removes additional complexity which is hard to justify for a
80 * diagnostic facility that has to operate in conditions where the system may
81 * have become unstable. Zlib deflate is comparatively small in terms of code
82 * size, and compresses ASCII text comparatively well. In terms of compression
83 * speed, deflate is not the best performer but for recording the log output on
84 * a kernel panic, this is not considered critical.
86 * The only remaining arguments supported by the compress= module parameter are
87 * 'deflate' and 'none'. To retain compatibility with existing installations,
88 * all other values are logged and replaced with 'deflate'.
90 static char *compress = "deflate";
91 module_param(compress, charp, 0444);
92 MODULE_PARM_DESC(compress, "compression to use");
94 /* How much of the kernel log to snapshot */
95 unsigned long kmsg_bytes = CONFIG_PSTORE_DEFAULT_KMSG_BYTES;
96 module_param(kmsg_bytes, ulong, 0444);
97 MODULE_PARM_DESC(kmsg_bytes, "amount of kernel log to snapshot (in bytes)");
99 static void *compress_workspace;
102 * Compression is only used for dmesg output, which consists of low-entropy
103 * ASCII text, and so we can assume worst-case 60%.
105 #define DMESG_COMP_PERCENT 60
107 static char *big_oops_buf;
108 static size_t max_compressed_size;
110 void pstore_set_kmsg_bytes(int bytes)
115 /* Tag each group of saved records with a sequence number */
116 static int oopscount;
118 const char *pstore_type_to_name(enum pstore_type_id type)
120 BUILD_BUG_ON(ARRAY_SIZE(pstore_type_names) != PSTORE_TYPE_MAX);
122 if (WARN_ON_ONCE(type >= PSTORE_TYPE_MAX))
125 return pstore_type_names[type];
127 EXPORT_SYMBOL_GPL(pstore_type_to_name);
129 enum pstore_type_id pstore_name_to_type(const char *name)
133 for (i = 0; i < PSTORE_TYPE_MAX; i++) {
134 if (!strcmp(pstore_type_names[i], name))
138 return PSTORE_TYPE_MAX;
140 EXPORT_SYMBOL_GPL(pstore_name_to_type);
142 static void pstore_timer_kick(void)
144 if (pstore_update_ms < 0)
147 mod_timer(&pstore_timer, jiffies + msecs_to_jiffies(pstore_update_ms));
150 static bool pstore_cannot_block_path(enum kmsg_dump_reason reason)
153 * In case of NMI path, pstore shouldn't be blocked
154 * regardless of reason.
160 /* In panic case, other cpus are stopped by smp_send_stop(). */
161 case KMSG_DUMP_PANIC:
163 * Emergency restart shouldn't be blocked by spinning on
164 * pstore_info::buf_lock.
166 case KMSG_DUMP_EMERG:
173 static int pstore_compress(const void *in, void *out,
174 unsigned int inlen, unsigned int outlen)
176 struct z_stream_s zstream = {
181 .workspace = compress_workspace,
185 if (!IS_ENABLED(CONFIG_PSTORE_COMPRESS))
188 ret = zlib_deflateInit2(&zstream, Z_DEFAULT_COMPRESSION, Z_DEFLATED,
189 -MAX_WBITS, DEF_MEM_LEVEL, Z_DEFAULT_STRATEGY);
193 ret = zlib_deflate(&zstream, Z_FINISH);
194 if (ret != Z_STREAM_END)
197 ret = zlib_deflateEnd(&zstream);
199 pr_warn_once("zlib_deflateEnd() failed: %d\n", ret);
201 return zstream.total_out;
204 static void allocate_buf_for_compression(void)
206 size_t compressed_size;
209 /* Skip if not built-in or compression disabled. */
210 if (!IS_ENABLED(CONFIG_PSTORE_COMPRESS) || !compress ||
211 !strcmp(compress, "none")) {
216 if (strcmp(compress, "deflate")) {
217 pr_err("Unsupported compression '%s', falling back to deflate\n",
219 compress = "deflate";
223 * The compression buffer only needs to be as large as the maximum
224 * uncompressed record size, since any record that would be expanded by
225 * compression is just stored uncompressed.
227 compressed_size = (psinfo->bufsize * 100) / DMESG_COMP_PERCENT;
228 buf = kvzalloc(compressed_size, GFP_KERNEL);
230 pr_err("Failed %zu byte compression buffer allocation for: %s\n",
231 psinfo->bufsize, compress);
236 vmalloc(zlib_deflate_workspacesize(MAX_WBITS, DEF_MEM_LEVEL));
237 if (!compress_workspace) {
238 pr_err("Failed to allocate zlib deflate workspace\n");
243 /* A non-NULL big_oops_buf indicates compression is available. */
245 max_compressed_size = compressed_size;
247 pr_info("Using crash dump compression: %s\n", compress);
250 static void free_buf_for_compression(void)
252 if (IS_ENABLED(CONFIG_PSTORE_COMPRESS) && compress_workspace) {
253 vfree(compress_workspace);
254 compress_workspace = NULL;
257 kvfree(big_oops_buf);
259 max_compressed_size = 0;
262 void pstore_record_init(struct pstore_record *record,
263 struct pstore_info *psinfo)
265 memset(record, 0, sizeof(*record));
267 record->psi = psinfo;
269 /* Report zeroed timestamp if called before timekeeping has resumed. */
270 record->time = ns_to_timespec64(ktime_get_real_fast_ns());
274 * callback from kmsg_dump. Save as much as we can (up to kmsg_bytes) from the
277 static void pstore_dump(struct kmsg_dumper *dumper,
278 enum kmsg_dump_reason reason)
280 struct kmsg_dump_iter iter;
281 unsigned long total = 0;
283 unsigned int part = 1;
284 unsigned long flags = 0;
288 why = kmsg_dump_reason_str(reason);
290 if (pstore_cannot_block_path(reason)) {
291 if (!spin_trylock_irqsave(&psinfo->buf_lock, flags)) {
292 pr_err("dump skipped in %s path because of concurrent dump\n",
293 in_nmi() ? "NMI" : why);
297 spin_lock_irqsave(&psinfo->buf_lock, flags);
300 kmsg_dump_rewind(&iter);
303 while (total < kmsg_bytes) {
309 struct pstore_record record;
311 pstore_record_init(&record, psinfo);
312 record.type = PSTORE_TYPE_DMESG;
313 record.count = oopscount;
314 record.reason = reason;
316 record.buf = psinfo->buf;
318 dst = big_oops_buf ?: psinfo->buf;
319 dst_size = max_compressed_size ?: psinfo->bufsize;
321 /* Write dump header. */
322 header_size = snprintf(dst, dst_size, "%s#%d Part%u\n", why,
324 dst_size -= header_size;
326 /* Write dump contents. */
327 if (!kmsg_dump_get_buffer(&iter, true, dst + header_size,
328 dst_size, &dump_size))
332 zipped_len = pstore_compress(dst, psinfo->buf,
333 header_size + dump_size,
336 if (zipped_len > 0) {
337 record.compressed = true;
338 record.size = zipped_len;
341 * Compression failed, so the buffer is most
342 * likely filled with binary data that does not
343 * compress as well as ASCII text. Copy as much
344 * of the uncompressed data as possible into
345 * the pstore record, and discard the rest.
347 record.size = psinfo->bufsize;
348 memcpy(psinfo->buf, dst, psinfo->bufsize);
351 record.size = header_size + dump_size;
354 ret = psinfo->write(&record);
355 if (ret == 0 && reason == KMSG_DUMP_OOPS) {
356 pstore_new_entry = 1;
359 /* Preserve only the first non-zero returned value. */
364 total += record.size;
367 spin_unlock_irqrestore(&psinfo->buf_lock, flags);
370 pr_err_once("backend (%s) writing error (%d)\n", psinfo->name,
375 static struct kmsg_dumper pstore_dumper = {
380 * Register with kmsg_dump to save last part of console log on panic.
382 static void pstore_register_kmsg(void)
384 kmsg_dump_register(&pstore_dumper);
387 static void pstore_unregister_kmsg(void)
389 kmsg_dump_unregister(&pstore_dumper);
392 #ifdef CONFIG_PSTORE_CONSOLE
393 static void pstore_console_write(struct console *con, const char *s, unsigned c)
395 struct pstore_record record;
400 pstore_record_init(&record, psinfo);
401 record.type = PSTORE_TYPE_CONSOLE;
403 record.buf = (char *)s;
405 psinfo->write(&record);
408 static struct console pstore_console = {
409 .write = pstore_console_write,
413 static void pstore_register_console(void)
415 /* Show which backend is going to get console writes. */
416 strscpy(pstore_console.name, psinfo->name,
417 sizeof(pstore_console.name));
419 * Always initialize flags here since prior unregister_console()
420 * calls may have changed settings (specifically CON_ENABLED).
422 pstore_console.flags = CON_PRINTBUFFER | CON_ENABLED | CON_ANYTIME;
423 register_console(&pstore_console);
426 static void pstore_unregister_console(void)
428 unregister_console(&pstore_console);
431 static void pstore_register_console(void) {}
432 static void pstore_unregister_console(void) {}
435 static int pstore_write_user_compat(struct pstore_record *record,
436 const char __user *buf)
443 record->buf = vmemdup_user(buf, record->size);
444 if (IS_ERR(record->buf)) {
445 ret = PTR_ERR(record->buf);
449 ret = record->psi->write(record);
455 return unlikely(ret < 0) ? ret : record->size;
459 * platform specific persistent storage driver registers with
460 * us here. If pstore is already mounted, call the platform
461 * read function right away to populate the file system. If not
462 * then the pstore mount code will call us later to fill out
465 int pstore_register(struct pstore_info *psi)
467 if (backend && strcmp(backend, psi->name)) {
468 pr_warn("backend '%s' already in use: ignoring '%s'\n",
473 /* Sanity check flags. */
475 pr_warn("backend '%s' must support at least one frontend\n",
480 /* Check for required functions. */
481 if (!psi->read || !psi->write) {
482 pr_warn("backend '%s' must implement read() and write()\n",
487 mutex_lock(&psinfo_lock);
489 pr_warn("backend '%s' already loaded: ignoring '%s'\n",
490 psinfo->name, psi->name);
491 mutex_unlock(&psinfo_lock);
495 if (!psi->write_user)
496 psi->write_user = pstore_write_user_compat;
498 mutex_init(&psinfo->read_mutex);
499 spin_lock_init(&psinfo->buf_lock);
501 if (psi->flags & PSTORE_FLAGS_DMESG)
502 allocate_buf_for_compression();
504 pstore_get_records(0);
506 if (psi->flags & PSTORE_FLAGS_DMESG) {
507 pstore_dumper.max_reason = psinfo->max_reason;
508 pstore_register_kmsg();
510 if (psi->flags & PSTORE_FLAGS_CONSOLE)
511 pstore_register_console();
512 if (psi->flags & PSTORE_FLAGS_FTRACE)
513 pstore_register_ftrace();
514 if (psi->flags & PSTORE_FLAGS_PMSG)
515 pstore_register_pmsg();
517 /* Start watching for new records, if desired. */
521 * Update the module parameter backend, so it is visible
522 * through /sys/module/pstore/parameters/backend
524 backend = kstrdup(psi->name, GFP_KERNEL);
526 pr_info("Registered %s as persistent store backend\n", psi->name);
528 mutex_unlock(&psinfo_lock);
531 EXPORT_SYMBOL_GPL(pstore_register);
533 void pstore_unregister(struct pstore_info *psi)
535 /* It's okay to unregister nothing. */
539 mutex_lock(&psinfo_lock);
541 /* Only one backend can be registered at a time. */
542 if (WARN_ON(psi != psinfo)) {
543 mutex_unlock(&psinfo_lock);
547 /* Unregister all callbacks. */
548 if (psi->flags & PSTORE_FLAGS_PMSG)
549 pstore_unregister_pmsg();
550 if (psi->flags & PSTORE_FLAGS_FTRACE)
551 pstore_unregister_ftrace();
552 if (psi->flags & PSTORE_FLAGS_CONSOLE)
553 pstore_unregister_console();
554 if (psi->flags & PSTORE_FLAGS_DMESG)
555 pstore_unregister_kmsg();
557 /* Stop timer and make sure all work has finished. */
558 del_timer_sync(&pstore_timer);
559 flush_work(&pstore_work);
561 /* Remove all backend records from filesystem tree. */
562 pstore_put_backend_records(psi);
564 free_buf_for_compression();
570 pr_info("Unregistered %s as persistent store backend\n", psi->name);
571 mutex_unlock(&psinfo_lock);
573 EXPORT_SYMBOL_GPL(pstore_unregister);
575 static void decompress_record(struct pstore_record *record,
576 struct z_stream_s *zstream)
580 char *unzipped, *workspace;
581 size_t max_uncompressed_size;
583 if (!IS_ENABLED(CONFIG_PSTORE_COMPRESS) || !record->compressed)
586 /* Only PSTORE_TYPE_DMESG support compression. */
587 if (record->type != PSTORE_TYPE_DMESG) {
588 pr_warn("ignored compressed record type %d\n", record->type);
592 /* Missing compression buffer means compression was not initialized. */
593 if (!zstream->workspace) {
594 pr_warn("no decompression method initialized!\n");
598 ret = zlib_inflateReset(zstream);
600 pr_err("zlib_inflateReset() failed, ret = %d!\n", ret);
604 /* Allocate enough space to hold max decompression and ECC. */
605 max_uncompressed_size = 3 * psinfo->bufsize;
606 workspace = kvzalloc(max_uncompressed_size + record->ecc_notice_size,
611 zstream->next_in = record->buf;
612 zstream->avail_in = record->size;
613 zstream->next_out = workspace;
614 zstream->avail_out = max_uncompressed_size;
616 ret = zlib_inflate(zstream, Z_FINISH);
617 if (ret != Z_STREAM_END) {
618 pr_err_ratelimited("zlib_inflate() failed, ret = %d!\n", ret);
623 unzipped_len = zstream->total_out;
625 /* Append ECC notice to decompressed buffer. */
626 memcpy(workspace + unzipped_len, record->buf + record->size,
627 record->ecc_notice_size);
629 /* Copy decompressed contents into an minimum-sized allocation. */
630 unzipped = kvmemdup(workspace, unzipped_len + record->ecc_notice_size,
636 /* Swap out compressed contents with decompressed contents. */
638 record->buf = unzipped;
639 record->size = unzipped_len;
640 record->compressed = false;
644 * Read all the records from one persistent store backend. Create
645 * files in our filesystem. Don't warn about -EEXIST errors
646 * when we are re-scanning the backing store looking to add new
649 void pstore_get_backend_records(struct pstore_info *psi,
650 struct dentry *root, int quiet)
653 unsigned int stop_loop = 65536;
654 struct z_stream_s zstream = {};
659 if (IS_ENABLED(CONFIG_PSTORE_COMPRESS) && compress) {
660 zstream.workspace = kvmalloc(zlib_inflate_workspacesize(),
662 zlib_inflateInit2(&zstream, -DEF_WBITS);
665 mutex_lock(&psi->read_mutex);
666 if (psi->open && psi->open(psi))
670 * Backend callback read() allocates record.buf. decompress_record()
671 * may reallocate record.buf. On success, pstore_mkfile() will keep
672 * the record.buf, so free it only on failure.
674 for (; stop_loop; stop_loop--) {
675 struct pstore_record *record;
678 record = kzalloc(sizeof(*record), GFP_KERNEL);
680 pr_err("out of memory creating record\n");
683 pstore_record_init(record, psi);
685 record->size = psi->read(record);
687 /* No more records left in backend? */
688 if (record->size <= 0) {
693 decompress_record(record, &zstream);
694 rc = pstore_mkfile(root, record);
696 /* pstore_mkfile() did not take record, so free it. */
700 if (rc != -EEXIST || !quiet)
707 mutex_unlock(&psi->read_mutex);
709 if (IS_ENABLED(CONFIG_PSTORE_COMPRESS) && compress) {
710 if (zlib_inflateEnd(&zstream) != Z_OK)
711 pr_warn("zlib_inflateEnd() failed\n");
712 kvfree(zstream.workspace);
716 pr_warn("failed to create %d record(s) from '%s'\n",
719 pr_err("looping? Too many records seen from '%s'\n",
723 static void pstore_dowork(struct work_struct *work)
725 pstore_get_records(1);
728 static void pstore_timefunc(struct timer_list *unused)
730 if (pstore_new_entry) {
731 pstore_new_entry = 0;
732 schedule_work(&pstore_work);
738 static int __init pstore_init(void)
742 ret = pstore_init_fs();
744 free_buf_for_compression();
748 late_initcall(pstore_init);
750 static void __exit pstore_exit(void)
754 module_exit(pstore_exit)
756 MODULE_AUTHOR("Tony Luck <tony.luck@intel.com>");
757 MODULE_LICENSE("GPL");