xref: /linux/kernel/printk/printk.c (revision 2bc46b3ad3c15165f91459b07ff8682478683194)
1 /*
2  *  linux/kernel/printk.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  *
6  * Modified to make sys_syslog() more flexible: added commands to
7  * return the last 4k of kernel messages, regardless of whether
8  * they've been read or not.  Added option to suppress kernel printk's
9  * to the console.  Added hook for sending the console messages
10  * elsewhere, in preparation for a serial line console (someday).
11  * Ted Ts'o, 2/11/93.
12  * Modified for sysctl support, 1/8/97, Chris Horn.
13  * Fixed SMP synchronization, 08/08/99, Manfred Spraul
14  *     manfred@colorfullife.com
15  * Rewrote bits to get rid of console_lock
16  *	01Mar01 Andrew Morton
17  */
18 
19 #include <linux/kernel.h>
20 #include <linux/mm.h>
21 #include <linux/tty.h>
22 #include <linux/tty_driver.h>
23 #include <linux/console.h>
24 #include <linux/init.h>
25 #include <linux/jiffies.h>
26 #include <linux/nmi.h>
27 #include <linux/module.h>
28 #include <linux/moduleparam.h>
29 #include <linux/interrupt.h>			/* For in_interrupt() */
30 #include <linux/delay.h>
31 #include <linux/smp.h>
32 #include <linux/security.h>
33 #include <linux/bootmem.h>
34 #include <linux/memblock.h>
35 #include <linux/syscalls.h>
36 #include <linux/kexec.h>
37 #include <linux/kdb.h>
38 #include <linux/ratelimit.h>
39 #include <linux/kmsg_dump.h>
40 #include <linux/syslog.h>
41 #include <linux/cpu.h>
42 #include <linux/notifier.h>
43 #include <linux/rculist.h>
44 #include <linux/poll.h>
45 #include <linux/irq_work.h>
46 #include <linux/utsname.h>
47 #include <linux/ctype.h>
48 #include <linux/uio.h>
49 
50 #include <asm/uaccess.h>
51 #include <asm-generic/sections.h>
52 
53 #define CREATE_TRACE_POINTS
54 #include <trace/events/printk.h>
55 
56 #include "console_cmdline.h"
57 #include "braille.h"
58 #include "internal.h"
59 
60 int console_printk[4] = {
61 	CONSOLE_LOGLEVEL_DEFAULT,	/* console_loglevel */
62 	MESSAGE_LOGLEVEL_DEFAULT,	/* default_message_loglevel */
63 	CONSOLE_LOGLEVEL_MIN,		/* minimum_console_loglevel */
64 	CONSOLE_LOGLEVEL_DEFAULT,	/* default_console_loglevel */
65 };
66 
67 /*
68  * Low level drivers may need that to know if they can schedule in
69  * their unblank() callback or not. So let's export it.
70  */
71 int oops_in_progress;
72 EXPORT_SYMBOL(oops_in_progress);
73 
74 /*
75  * console_sem protects the console_drivers list, and also
76  * provides serialisation for access to the entire console
77  * driver system.
78  */
79 static DEFINE_SEMAPHORE(console_sem);
80 struct console *console_drivers;
81 EXPORT_SYMBOL_GPL(console_drivers);
82 
83 #ifdef CONFIG_LOCKDEP
84 static struct lockdep_map console_lock_dep_map = {
85 	.name = "console_lock"
86 };
87 #endif
88 
89 /*
90  * Number of registered extended console drivers.
91  *
92  * If extended consoles are present, in-kernel cont reassembly is disabled
93  * and each fragment is stored as a separate log entry with proper
94  * continuation flag so that every emitted message has full metadata.  This
95  * doesn't change the result for regular consoles or /proc/kmsg.  For
96  * /dev/kmsg, as long as the reader concatenates messages according to
97  * consecutive continuation flags, the end result should be the same too.
98  */
99 static int nr_ext_console_drivers;
100 
101 /*
102  * Helper macros to handle lockdep when locking/unlocking console_sem. We use
103  * macros instead of functions so that _RET_IP_ contains useful information.
104  */
105 #define down_console_sem() do { \
106 	down(&console_sem);\
107 	mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
108 } while (0)
109 
110 static int __down_trylock_console_sem(unsigned long ip)
111 {
112 	if (down_trylock(&console_sem))
113 		return 1;
114 	mutex_acquire(&console_lock_dep_map, 0, 1, ip);
115 	return 0;
116 }
117 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
118 
119 #define up_console_sem() do { \
120 	mutex_release(&console_lock_dep_map, 1, _RET_IP_);\
121 	up(&console_sem);\
122 } while (0)
123 
124 /*
125  * This is used for debugging the mess that is the VT code by
126  * keeping track if we have the console semaphore held. It's
127  * definitely not the perfect debug tool (we don't know if _WE_
128  * hold it and are racing, but it helps tracking those weird code
129  * paths in the console code where we end up in places I want
130  * locked without the console sempahore held).
131  */
132 static int console_locked, console_suspended;
133 
134 /*
135  * If exclusive_console is non-NULL then only this console is to be printed to.
136  */
137 static struct console *exclusive_console;
138 
139 /*
140  *	Array of consoles built from command line options (console=)
141  */
142 
143 #define MAX_CMDLINECONSOLES 8
144 
145 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
146 
147 static int selected_console = -1;
148 static int preferred_console = -1;
149 int console_set_on_cmdline;
150 EXPORT_SYMBOL(console_set_on_cmdline);
151 
152 /* Flag: console code may call schedule() */
153 static int console_may_schedule;
154 
155 /*
156  * The printk log buffer consists of a chain of concatenated variable
157  * length records. Every record starts with a record header, containing
158  * the overall length of the record.
159  *
160  * The heads to the first and last entry in the buffer, as well as the
161  * sequence numbers of these entries are maintained when messages are
162  * stored.
163  *
164  * If the heads indicate available messages, the length in the header
165  * tells the start next message. A length == 0 for the next message
166  * indicates a wrap-around to the beginning of the buffer.
167  *
168  * Every record carries the monotonic timestamp in microseconds, as well as
169  * the standard userspace syslog level and syslog facility. The usual
170  * kernel messages use LOG_KERN; userspace-injected messages always carry
171  * a matching syslog facility, by default LOG_USER. The origin of every
172  * message can be reliably determined that way.
173  *
174  * The human readable log message directly follows the message header. The
175  * length of the message text is stored in the header, the stored message
176  * is not terminated.
177  *
178  * Optionally, a message can carry a dictionary of properties (key/value pairs),
179  * to provide userspace with a machine-readable message context.
180  *
181  * Examples for well-defined, commonly used property names are:
182  *   DEVICE=b12:8               device identifier
183  *                                b12:8         block dev_t
184  *                                c127:3        char dev_t
185  *                                n8            netdev ifindex
186  *                                +sound:card0  subsystem:devname
187  *   SUBSYSTEM=pci              driver-core subsystem name
188  *
189  * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
190  * follows directly after a '=' character. Every property is terminated by
191  * a '\0' character. The last property is not terminated.
192  *
193  * Example of a message structure:
194  *   0000  ff 8f 00 00 00 00 00 00      monotonic time in nsec
195  *   0008  34 00                        record is 52 bytes long
196  *   000a        0b 00                  text is 11 bytes long
197  *   000c              1f 00            dictionary is 23 bytes long
198  *   000e                    03 00      LOG_KERN (facility) LOG_ERR (level)
199  *   0010  69 74 27 73 20 61 20 6c      "it's a l"
200  *         69 6e 65                     "ine"
201  *   001b           44 45 56 49 43      "DEVIC"
202  *         45 3d 62 38 3a 32 00 44      "E=b8:2\0D"
203  *         52 49 56 45 52 3d 62 75      "RIVER=bu"
204  *         67                           "g"
205  *   0032     00 00 00                  padding to next message header
206  *
207  * The 'struct printk_log' buffer header must never be directly exported to
208  * userspace, it is a kernel-private implementation detail that might
209  * need to be changed in the future, when the requirements change.
210  *
211  * /dev/kmsg exports the structured data in the following line format:
212  *   "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
213  *
214  * Users of the export format should ignore possible additional values
215  * separated by ',', and find the message after the ';' character.
216  *
217  * The optional key/value pairs are attached as continuation lines starting
218  * with a space character and terminated by a newline. All possible
219  * non-prinatable characters are escaped in the "\xff" notation.
220  */
221 
222 enum log_flags {
223 	LOG_NOCONS	= 1,	/* already flushed, do not print to console */
224 	LOG_NEWLINE	= 2,	/* text ended with a newline */
225 	LOG_PREFIX	= 4,	/* text started with a prefix */
226 	LOG_CONT	= 8,	/* text is a fragment of a continuation line */
227 };
228 
229 struct printk_log {
230 	u64 ts_nsec;		/* timestamp in nanoseconds */
231 	u16 len;		/* length of entire record */
232 	u16 text_len;		/* length of text buffer */
233 	u16 dict_len;		/* length of dictionary buffer */
234 	u8 facility;		/* syslog facility */
235 	u8 flags:5;		/* internal record flags */
236 	u8 level:3;		/* syslog level */
237 }
238 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
239 __packed __aligned(4)
240 #endif
241 ;
242 
243 /*
244  * The logbuf_lock protects kmsg buffer, indices, counters.  This can be taken
245  * within the scheduler's rq lock. It must be released before calling
246  * console_unlock() or anything else that might wake up a process.
247  */
248 DEFINE_RAW_SPINLOCK(logbuf_lock);
249 
250 #ifdef CONFIG_PRINTK
251 DECLARE_WAIT_QUEUE_HEAD(log_wait);
252 /* the next printk record to read by syslog(READ) or /proc/kmsg */
253 static u64 syslog_seq;
254 static u32 syslog_idx;
255 static enum log_flags syslog_prev;
256 static size_t syslog_partial;
257 
258 /* index and sequence number of the first record stored in the buffer */
259 static u64 log_first_seq;
260 static u32 log_first_idx;
261 
262 /* index and sequence number of the next record to store in the buffer */
263 static u64 log_next_seq;
264 static u32 log_next_idx;
265 
266 /* the next printk record to write to the console */
267 static u64 console_seq;
268 static u32 console_idx;
269 static enum log_flags console_prev;
270 
271 /* the next printk record to read after the last 'clear' command */
272 static u64 clear_seq;
273 static u32 clear_idx;
274 
275 #define PREFIX_MAX		32
276 #define LOG_LINE_MAX		(1024 - PREFIX_MAX)
277 
278 #define LOG_LEVEL(v)		((v) & 0x07)
279 #define LOG_FACILITY(v)		((v) >> 3 & 0xff)
280 
281 /* record buffer */
282 #define LOG_ALIGN __alignof__(struct printk_log)
283 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
284 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
285 static char *log_buf = __log_buf;
286 static u32 log_buf_len = __LOG_BUF_LEN;
287 
288 /* Return log buffer address */
289 char *log_buf_addr_get(void)
290 {
291 	return log_buf;
292 }
293 
294 /* Return log buffer size */
295 u32 log_buf_len_get(void)
296 {
297 	return log_buf_len;
298 }
299 
300 /* human readable text of the record */
301 static char *log_text(const struct printk_log *msg)
302 {
303 	return (char *)msg + sizeof(struct printk_log);
304 }
305 
306 /* optional key/value pair dictionary attached to the record */
307 static char *log_dict(const struct printk_log *msg)
308 {
309 	return (char *)msg + sizeof(struct printk_log) + msg->text_len;
310 }
311 
312 /* get record by index; idx must point to valid msg */
313 static struct printk_log *log_from_idx(u32 idx)
314 {
315 	struct printk_log *msg = (struct printk_log *)(log_buf + idx);
316 
317 	/*
318 	 * A length == 0 record is the end of buffer marker. Wrap around and
319 	 * read the message at the start of the buffer.
320 	 */
321 	if (!msg->len)
322 		return (struct printk_log *)log_buf;
323 	return msg;
324 }
325 
326 /* get next record; idx must point to valid msg */
327 static u32 log_next(u32 idx)
328 {
329 	struct printk_log *msg = (struct printk_log *)(log_buf + idx);
330 
331 	/* length == 0 indicates the end of the buffer; wrap */
332 	/*
333 	 * A length == 0 record is the end of buffer marker. Wrap around and
334 	 * read the message at the start of the buffer as *this* one, and
335 	 * return the one after that.
336 	 */
337 	if (!msg->len) {
338 		msg = (struct printk_log *)log_buf;
339 		return msg->len;
340 	}
341 	return idx + msg->len;
342 }
343 
344 /*
345  * Check whether there is enough free space for the given message.
346  *
347  * The same values of first_idx and next_idx mean that the buffer
348  * is either empty or full.
349  *
350  * If the buffer is empty, we must respect the position of the indexes.
351  * They cannot be reset to the beginning of the buffer.
352  */
353 static int logbuf_has_space(u32 msg_size, bool empty)
354 {
355 	u32 free;
356 
357 	if (log_next_idx > log_first_idx || empty)
358 		free = max(log_buf_len - log_next_idx, log_first_idx);
359 	else
360 		free = log_first_idx - log_next_idx;
361 
362 	/*
363 	 * We need space also for an empty header that signalizes wrapping
364 	 * of the buffer.
365 	 */
366 	return free >= msg_size + sizeof(struct printk_log);
367 }
368 
369 static int log_make_free_space(u32 msg_size)
370 {
371 	while (log_first_seq < log_next_seq &&
372 	       !logbuf_has_space(msg_size, false)) {
373 		/* drop old messages until we have enough contiguous space */
374 		log_first_idx = log_next(log_first_idx);
375 		log_first_seq++;
376 	}
377 
378 	if (clear_seq < log_first_seq) {
379 		clear_seq = log_first_seq;
380 		clear_idx = log_first_idx;
381 	}
382 
383 	/* sequence numbers are equal, so the log buffer is empty */
384 	if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
385 		return 0;
386 
387 	return -ENOMEM;
388 }
389 
390 /* compute the message size including the padding bytes */
391 static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
392 {
393 	u32 size;
394 
395 	size = sizeof(struct printk_log) + text_len + dict_len;
396 	*pad_len = (-size) & (LOG_ALIGN - 1);
397 	size += *pad_len;
398 
399 	return size;
400 }
401 
402 /*
403  * Define how much of the log buffer we could take at maximum. The value
404  * must be greater than two. Note that only half of the buffer is available
405  * when the index points to the middle.
406  */
407 #define MAX_LOG_TAKE_PART 4
408 static const char trunc_msg[] = "<truncated>";
409 
410 static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
411 			u16 *dict_len, u32 *pad_len)
412 {
413 	/*
414 	 * The message should not take the whole buffer. Otherwise, it might
415 	 * get removed too soon.
416 	 */
417 	u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
418 	if (*text_len > max_text_len)
419 		*text_len = max_text_len;
420 	/* enable the warning message */
421 	*trunc_msg_len = strlen(trunc_msg);
422 	/* disable the "dict" completely */
423 	*dict_len = 0;
424 	/* compute the size again, count also the warning message */
425 	return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
426 }
427 
428 /* insert record into the buffer, discard old ones, update heads */
429 static int log_store(int facility, int level,
430 		     enum log_flags flags, u64 ts_nsec,
431 		     const char *dict, u16 dict_len,
432 		     const char *text, u16 text_len)
433 {
434 	struct printk_log *msg;
435 	u32 size, pad_len;
436 	u16 trunc_msg_len = 0;
437 
438 	/* number of '\0' padding bytes to next message */
439 	size = msg_used_size(text_len, dict_len, &pad_len);
440 
441 	if (log_make_free_space(size)) {
442 		/* truncate the message if it is too long for empty buffer */
443 		size = truncate_msg(&text_len, &trunc_msg_len,
444 				    &dict_len, &pad_len);
445 		/* survive when the log buffer is too small for trunc_msg */
446 		if (log_make_free_space(size))
447 			return 0;
448 	}
449 
450 	if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
451 		/*
452 		 * This message + an additional empty header does not fit
453 		 * at the end of the buffer. Add an empty header with len == 0
454 		 * to signify a wrap around.
455 		 */
456 		memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
457 		log_next_idx = 0;
458 	}
459 
460 	/* fill message */
461 	msg = (struct printk_log *)(log_buf + log_next_idx);
462 	memcpy(log_text(msg), text, text_len);
463 	msg->text_len = text_len;
464 	if (trunc_msg_len) {
465 		memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
466 		msg->text_len += trunc_msg_len;
467 	}
468 	memcpy(log_dict(msg), dict, dict_len);
469 	msg->dict_len = dict_len;
470 	msg->facility = facility;
471 	msg->level = level & 7;
472 	msg->flags = flags & 0x1f;
473 	if (ts_nsec > 0)
474 		msg->ts_nsec = ts_nsec;
475 	else
476 		msg->ts_nsec = local_clock();
477 	memset(log_dict(msg) + dict_len, 0, pad_len);
478 	msg->len = size;
479 
480 	/* insert message */
481 	log_next_idx += msg->len;
482 	log_next_seq++;
483 
484 	return msg->text_len;
485 }
486 
487 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
488 
489 static int syslog_action_restricted(int type)
490 {
491 	if (dmesg_restrict)
492 		return 1;
493 	/*
494 	 * Unless restricted, we allow "read all" and "get buffer size"
495 	 * for everybody.
496 	 */
497 	return type != SYSLOG_ACTION_READ_ALL &&
498 	       type != SYSLOG_ACTION_SIZE_BUFFER;
499 }
500 
501 int check_syslog_permissions(int type, int source)
502 {
503 	/*
504 	 * If this is from /proc/kmsg and we've already opened it, then we've
505 	 * already done the capabilities checks at open time.
506 	 */
507 	if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
508 		goto ok;
509 
510 	if (syslog_action_restricted(type)) {
511 		if (capable(CAP_SYSLOG))
512 			goto ok;
513 		/*
514 		 * For historical reasons, accept CAP_SYS_ADMIN too, with
515 		 * a warning.
516 		 */
517 		if (capable(CAP_SYS_ADMIN)) {
518 			pr_warn_once("%s (%d): Attempt to access syslog with "
519 				     "CAP_SYS_ADMIN but no CAP_SYSLOG "
520 				     "(deprecated).\n",
521 				 current->comm, task_pid_nr(current));
522 			goto ok;
523 		}
524 		return -EPERM;
525 	}
526 ok:
527 	return security_syslog(type);
528 }
529 EXPORT_SYMBOL_GPL(check_syslog_permissions);
530 
531 static void append_char(char **pp, char *e, char c)
532 {
533 	if (*pp < e)
534 		*(*pp)++ = c;
535 }
536 
537 static ssize_t msg_print_ext_header(char *buf, size_t size,
538 				    struct printk_log *msg, u64 seq,
539 				    enum log_flags prev_flags)
540 {
541 	u64 ts_usec = msg->ts_nsec;
542 	char cont = '-';
543 
544 	do_div(ts_usec, 1000);
545 
546 	/*
547 	 * If we couldn't merge continuation line fragments during the print,
548 	 * export the stored flags to allow an optional external merge of the
549 	 * records. Merging the records isn't always neccessarily correct, like
550 	 * when we hit a race during printing. In most cases though, it produces
551 	 * better readable output. 'c' in the record flags mark the first
552 	 * fragment of a line, '+' the following.
553 	 */
554 	if (msg->flags & LOG_CONT && !(prev_flags & LOG_CONT))
555 		cont = 'c';
556 	else if ((msg->flags & LOG_CONT) ||
557 		 ((prev_flags & LOG_CONT) && !(msg->flags & LOG_PREFIX)))
558 		cont = '+';
559 
560 	return scnprintf(buf, size, "%u,%llu,%llu,%c;",
561 		       (msg->facility << 3) | msg->level, seq, ts_usec, cont);
562 }
563 
564 static ssize_t msg_print_ext_body(char *buf, size_t size,
565 				  char *dict, size_t dict_len,
566 				  char *text, size_t text_len)
567 {
568 	char *p = buf, *e = buf + size;
569 	size_t i;
570 
571 	/* escape non-printable characters */
572 	for (i = 0; i < text_len; i++) {
573 		unsigned char c = text[i];
574 
575 		if (c < ' ' || c >= 127 || c == '\\')
576 			p += scnprintf(p, e - p, "\\x%02x", c);
577 		else
578 			append_char(&p, e, c);
579 	}
580 	append_char(&p, e, '\n');
581 
582 	if (dict_len) {
583 		bool line = true;
584 
585 		for (i = 0; i < dict_len; i++) {
586 			unsigned char c = dict[i];
587 
588 			if (line) {
589 				append_char(&p, e, ' ');
590 				line = false;
591 			}
592 
593 			if (c == '\0') {
594 				append_char(&p, e, '\n');
595 				line = true;
596 				continue;
597 			}
598 
599 			if (c < ' ' || c >= 127 || c == '\\') {
600 				p += scnprintf(p, e - p, "\\x%02x", c);
601 				continue;
602 			}
603 
604 			append_char(&p, e, c);
605 		}
606 		append_char(&p, e, '\n');
607 	}
608 
609 	return p - buf;
610 }
611 
612 /* /dev/kmsg - userspace message inject/listen interface */
613 struct devkmsg_user {
614 	u64 seq;
615 	u32 idx;
616 	enum log_flags prev;
617 	struct mutex lock;
618 	char buf[CONSOLE_EXT_LOG_MAX];
619 };
620 
621 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
622 {
623 	char *buf, *line;
624 	int level = default_message_loglevel;
625 	int facility = 1;	/* LOG_USER */
626 	size_t len = iov_iter_count(from);
627 	ssize_t ret = len;
628 
629 	if (len > LOG_LINE_MAX)
630 		return -EINVAL;
631 	buf = kmalloc(len+1, GFP_KERNEL);
632 	if (buf == NULL)
633 		return -ENOMEM;
634 
635 	buf[len] = '\0';
636 	if (copy_from_iter(buf, len, from) != len) {
637 		kfree(buf);
638 		return -EFAULT;
639 	}
640 
641 	/*
642 	 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
643 	 * the decimal value represents 32bit, the lower 3 bit are the log
644 	 * level, the rest are the log facility.
645 	 *
646 	 * If no prefix or no userspace facility is specified, we
647 	 * enforce LOG_USER, to be able to reliably distinguish
648 	 * kernel-generated messages from userspace-injected ones.
649 	 */
650 	line = buf;
651 	if (line[0] == '<') {
652 		char *endp = NULL;
653 		unsigned int u;
654 
655 		u = simple_strtoul(line + 1, &endp, 10);
656 		if (endp && endp[0] == '>') {
657 			level = LOG_LEVEL(u);
658 			if (LOG_FACILITY(u) != 0)
659 				facility = LOG_FACILITY(u);
660 			endp++;
661 			len -= endp - line;
662 			line = endp;
663 		}
664 	}
665 
666 	printk_emit(facility, level, NULL, 0, "%s", line);
667 	kfree(buf);
668 	return ret;
669 }
670 
671 static ssize_t devkmsg_read(struct file *file, char __user *buf,
672 			    size_t count, loff_t *ppos)
673 {
674 	struct devkmsg_user *user = file->private_data;
675 	struct printk_log *msg;
676 	size_t len;
677 	ssize_t ret;
678 
679 	if (!user)
680 		return -EBADF;
681 
682 	ret = mutex_lock_interruptible(&user->lock);
683 	if (ret)
684 		return ret;
685 	raw_spin_lock_irq(&logbuf_lock);
686 	while (user->seq == log_next_seq) {
687 		if (file->f_flags & O_NONBLOCK) {
688 			ret = -EAGAIN;
689 			raw_spin_unlock_irq(&logbuf_lock);
690 			goto out;
691 		}
692 
693 		raw_spin_unlock_irq(&logbuf_lock);
694 		ret = wait_event_interruptible(log_wait,
695 					       user->seq != log_next_seq);
696 		if (ret)
697 			goto out;
698 		raw_spin_lock_irq(&logbuf_lock);
699 	}
700 
701 	if (user->seq < log_first_seq) {
702 		/* our last seen message is gone, return error and reset */
703 		user->idx = log_first_idx;
704 		user->seq = log_first_seq;
705 		ret = -EPIPE;
706 		raw_spin_unlock_irq(&logbuf_lock);
707 		goto out;
708 	}
709 
710 	msg = log_from_idx(user->idx);
711 	len = msg_print_ext_header(user->buf, sizeof(user->buf),
712 				   msg, user->seq, user->prev);
713 	len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
714 				  log_dict(msg), msg->dict_len,
715 				  log_text(msg), msg->text_len);
716 
717 	user->prev = msg->flags;
718 	user->idx = log_next(user->idx);
719 	user->seq++;
720 	raw_spin_unlock_irq(&logbuf_lock);
721 
722 	if (len > count) {
723 		ret = -EINVAL;
724 		goto out;
725 	}
726 
727 	if (copy_to_user(buf, user->buf, len)) {
728 		ret = -EFAULT;
729 		goto out;
730 	}
731 	ret = len;
732 out:
733 	mutex_unlock(&user->lock);
734 	return ret;
735 }
736 
737 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
738 {
739 	struct devkmsg_user *user = file->private_data;
740 	loff_t ret = 0;
741 
742 	if (!user)
743 		return -EBADF;
744 	if (offset)
745 		return -ESPIPE;
746 
747 	raw_spin_lock_irq(&logbuf_lock);
748 	switch (whence) {
749 	case SEEK_SET:
750 		/* the first record */
751 		user->idx = log_first_idx;
752 		user->seq = log_first_seq;
753 		break;
754 	case SEEK_DATA:
755 		/*
756 		 * The first record after the last SYSLOG_ACTION_CLEAR,
757 		 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
758 		 * changes no global state, and does not clear anything.
759 		 */
760 		user->idx = clear_idx;
761 		user->seq = clear_seq;
762 		break;
763 	case SEEK_END:
764 		/* after the last record */
765 		user->idx = log_next_idx;
766 		user->seq = log_next_seq;
767 		break;
768 	default:
769 		ret = -EINVAL;
770 	}
771 	raw_spin_unlock_irq(&logbuf_lock);
772 	return ret;
773 }
774 
775 static unsigned int devkmsg_poll(struct file *file, poll_table *wait)
776 {
777 	struct devkmsg_user *user = file->private_data;
778 	int ret = 0;
779 
780 	if (!user)
781 		return POLLERR|POLLNVAL;
782 
783 	poll_wait(file, &log_wait, wait);
784 
785 	raw_spin_lock_irq(&logbuf_lock);
786 	if (user->seq < log_next_seq) {
787 		/* return error when data has vanished underneath us */
788 		if (user->seq < log_first_seq)
789 			ret = POLLIN|POLLRDNORM|POLLERR|POLLPRI;
790 		else
791 			ret = POLLIN|POLLRDNORM;
792 	}
793 	raw_spin_unlock_irq(&logbuf_lock);
794 
795 	return ret;
796 }
797 
798 static int devkmsg_open(struct inode *inode, struct file *file)
799 {
800 	struct devkmsg_user *user;
801 	int err;
802 
803 	/* write-only does not need any file context */
804 	if ((file->f_flags & O_ACCMODE) == O_WRONLY)
805 		return 0;
806 
807 	err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
808 				       SYSLOG_FROM_READER);
809 	if (err)
810 		return err;
811 
812 	user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
813 	if (!user)
814 		return -ENOMEM;
815 
816 	mutex_init(&user->lock);
817 
818 	raw_spin_lock_irq(&logbuf_lock);
819 	user->idx = log_first_idx;
820 	user->seq = log_first_seq;
821 	raw_spin_unlock_irq(&logbuf_lock);
822 
823 	file->private_data = user;
824 	return 0;
825 }
826 
827 static int devkmsg_release(struct inode *inode, struct file *file)
828 {
829 	struct devkmsg_user *user = file->private_data;
830 
831 	if (!user)
832 		return 0;
833 
834 	mutex_destroy(&user->lock);
835 	kfree(user);
836 	return 0;
837 }
838 
839 const struct file_operations kmsg_fops = {
840 	.open = devkmsg_open,
841 	.read = devkmsg_read,
842 	.write_iter = devkmsg_write,
843 	.llseek = devkmsg_llseek,
844 	.poll = devkmsg_poll,
845 	.release = devkmsg_release,
846 };
847 
848 #ifdef CONFIG_KEXEC_CORE
849 /*
850  * This appends the listed symbols to /proc/vmcore
851  *
852  * /proc/vmcore is used by various utilities, like crash and makedumpfile to
853  * obtain access to symbols that are otherwise very difficult to locate.  These
854  * symbols are specifically used so that utilities can access and extract the
855  * dmesg log from a vmcore file after a crash.
856  */
857 void log_buf_kexec_setup(void)
858 {
859 	VMCOREINFO_SYMBOL(log_buf);
860 	VMCOREINFO_SYMBOL(log_buf_len);
861 	VMCOREINFO_SYMBOL(log_first_idx);
862 	VMCOREINFO_SYMBOL(clear_idx);
863 	VMCOREINFO_SYMBOL(log_next_idx);
864 	/*
865 	 * Export struct printk_log size and field offsets. User space tools can
866 	 * parse it and detect any changes to structure down the line.
867 	 */
868 	VMCOREINFO_STRUCT_SIZE(printk_log);
869 	VMCOREINFO_OFFSET(printk_log, ts_nsec);
870 	VMCOREINFO_OFFSET(printk_log, len);
871 	VMCOREINFO_OFFSET(printk_log, text_len);
872 	VMCOREINFO_OFFSET(printk_log, dict_len);
873 }
874 #endif
875 
876 /* requested log_buf_len from kernel cmdline */
877 static unsigned long __initdata new_log_buf_len;
878 
879 /* we practice scaling the ring buffer by powers of 2 */
880 static void __init log_buf_len_update(unsigned size)
881 {
882 	if (size)
883 		size = roundup_pow_of_two(size);
884 	if (size > log_buf_len)
885 		new_log_buf_len = size;
886 }
887 
888 /* save requested log_buf_len since it's too early to process it */
889 static int __init log_buf_len_setup(char *str)
890 {
891 	unsigned size = memparse(str, &str);
892 
893 	log_buf_len_update(size);
894 
895 	return 0;
896 }
897 early_param("log_buf_len", log_buf_len_setup);
898 
899 #ifdef CONFIG_SMP
900 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
901 
902 static void __init log_buf_add_cpu(void)
903 {
904 	unsigned int cpu_extra;
905 
906 	/*
907 	 * archs should set up cpu_possible_bits properly with
908 	 * set_cpu_possible() after setup_arch() but just in
909 	 * case lets ensure this is valid.
910 	 */
911 	if (num_possible_cpus() == 1)
912 		return;
913 
914 	cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
915 
916 	/* by default this will only continue through for large > 64 CPUs */
917 	if (cpu_extra <= __LOG_BUF_LEN / 2)
918 		return;
919 
920 	pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
921 		__LOG_CPU_MAX_BUF_LEN);
922 	pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
923 		cpu_extra);
924 	pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
925 
926 	log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
927 }
928 #else /* !CONFIG_SMP */
929 static inline void log_buf_add_cpu(void) {}
930 #endif /* CONFIG_SMP */
931 
932 void __init setup_log_buf(int early)
933 {
934 	unsigned long flags;
935 	char *new_log_buf;
936 	int free;
937 
938 	if (log_buf != __log_buf)
939 		return;
940 
941 	if (!early && !new_log_buf_len)
942 		log_buf_add_cpu();
943 
944 	if (!new_log_buf_len)
945 		return;
946 
947 	if (early) {
948 		new_log_buf =
949 			memblock_virt_alloc(new_log_buf_len, LOG_ALIGN);
950 	} else {
951 		new_log_buf = memblock_virt_alloc_nopanic(new_log_buf_len,
952 							  LOG_ALIGN);
953 	}
954 
955 	if (unlikely(!new_log_buf)) {
956 		pr_err("log_buf_len: %ld bytes not available\n",
957 			new_log_buf_len);
958 		return;
959 	}
960 
961 	raw_spin_lock_irqsave(&logbuf_lock, flags);
962 	log_buf_len = new_log_buf_len;
963 	log_buf = new_log_buf;
964 	new_log_buf_len = 0;
965 	free = __LOG_BUF_LEN - log_next_idx;
966 	memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
967 	raw_spin_unlock_irqrestore(&logbuf_lock, flags);
968 
969 	pr_info("log_buf_len: %d bytes\n", log_buf_len);
970 	pr_info("early log buf free: %d(%d%%)\n",
971 		free, (free * 100) / __LOG_BUF_LEN);
972 }
973 
974 static bool __read_mostly ignore_loglevel;
975 
976 static int __init ignore_loglevel_setup(char *str)
977 {
978 	ignore_loglevel = true;
979 	pr_info("debug: ignoring loglevel setting.\n");
980 
981 	return 0;
982 }
983 
984 early_param("ignore_loglevel", ignore_loglevel_setup);
985 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
986 MODULE_PARM_DESC(ignore_loglevel,
987 		 "ignore loglevel setting (prints all kernel messages to the console)");
988 
989 #ifdef CONFIG_BOOT_PRINTK_DELAY
990 
991 static int boot_delay; /* msecs delay after each printk during bootup */
992 static unsigned long long loops_per_msec;	/* based on boot_delay */
993 
994 static int __init boot_delay_setup(char *str)
995 {
996 	unsigned long lpj;
997 
998 	lpj = preset_lpj ? preset_lpj : 1000000;	/* some guess */
999 	loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1000 
1001 	get_option(&str, &boot_delay);
1002 	if (boot_delay > 10 * 1000)
1003 		boot_delay = 0;
1004 
1005 	pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1006 		"HZ: %d, loops_per_msec: %llu\n",
1007 		boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1008 	return 0;
1009 }
1010 early_param("boot_delay", boot_delay_setup);
1011 
1012 static void boot_delay_msec(int level)
1013 {
1014 	unsigned long long k;
1015 	unsigned long timeout;
1016 
1017 	if ((boot_delay == 0 || system_state != SYSTEM_BOOTING)
1018 		|| (level >= console_loglevel && !ignore_loglevel)) {
1019 		return;
1020 	}
1021 
1022 	k = (unsigned long long)loops_per_msec * boot_delay;
1023 
1024 	timeout = jiffies + msecs_to_jiffies(boot_delay);
1025 	while (k) {
1026 		k--;
1027 		cpu_relax();
1028 		/*
1029 		 * use (volatile) jiffies to prevent
1030 		 * compiler reduction; loop termination via jiffies
1031 		 * is secondary and may or may not happen.
1032 		 */
1033 		if (time_after(jiffies, timeout))
1034 			break;
1035 		touch_nmi_watchdog();
1036 	}
1037 }
1038 #else
1039 static inline void boot_delay_msec(int level)
1040 {
1041 }
1042 #endif
1043 
1044 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1045 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1046 
1047 static size_t print_time(u64 ts, char *buf)
1048 {
1049 	unsigned long rem_nsec;
1050 
1051 	if (!printk_time)
1052 		return 0;
1053 
1054 	rem_nsec = do_div(ts, 1000000000);
1055 
1056 	if (!buf)
1057 		return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
1058 
1059 	return sprintf(buf, "[%5lu.%06lu] ",
1060 		       (unsigned long)ts, rem_nsec / 1000);
1061 }
1062 
1063 static size_t print_prefix(const struct printk_log *msg, bool syslog, char *buf)
1064 {
1065 	size_t len = 0;
1066 	unsigned int prefix = (msg->facility << 3) | msg->level;
1067 
1068 	if (syslog) {
1069 		if (buf) {
1070 			len += sprintf(buf, "<%u>", prefix);
1071 		} else {
1072 			len += 3;
1073 			if (prefix > 999)
1074 				len += 3;
1075 			else if (prefix > 99)
1076 				len += 2;
1077 			else if (prefix > 9)
1078 				len++;
1079 		}
1080 	}
1081 
1082 	len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
1083 	return len;
1084 }
1085 
1086 static size_t msg_print_text(const struct printk_log *msg, enum log_flags prev,
1087 			     bool syslog, char *buf, size_t size)
1088 {
1089 	const char *text = log_text(msg);
1090 	size_t text_size = msg->text_len;
1091 	bool prefix = true;
1092 	bool newline = true;
1093 	size_t len = 0;
1094 
1095 	if ((prev & LOG_CONT) && !(msg->flags & LOG_PREFIX))
1096 		prefix = false;
1097 
1098 	if (msg->flags & LOG_CONT) {
1099 		if ((prev & LOG_CONT) && !(prev & LOG_NEWLINE))
1100 			prefix = false;
1101 
1102 		if (!(msg->flags & LOG_NEWLINE))
1103 			newline = false;
1104 	}
1105 
1106 	do {
1107 		const char *next = memchr(text, '\n', text_size);
1108 		size_t text_len;
1109 
1110 		if (next) {
1111 			text_len = next - text;
1112 			next++;
1113 			text_size -= next - text;
1114 		} else {
1115 			text_len = text_size;
1116 		}
1117 
1118 		if (buf) {
1119 			if (print_prefix(msg, syslog, NULL) +
1120 			    text_len + 1 >= size - len)
1121 				break;
1122 
1123 			if (prefix)
1124 				len += print_prefix(msg, syslog, buf + len);
1125 			memcpy(buf + len, text, text_len);
1126 			len += text_len;
1127 			if (next || newline)
1128 				buf[len++] = '\n';
1129 		} else {
1130 			/* SYSLOG_ACTION_* buffer size only calculation */
1131 			if (prefix)
1132 				len += print_prefix(msg, syslog, NULL);
1133 			len += text_len;
1134 			if (next || newline)
1135 				len++;
1136 		}
1137 
1138 		prefix = true;
1139 		text = next;
1140 	} while (text);
1141 
1142 	return len;
1143 }
1144 
1145 static int syslog_print(char __user *buf, int size)
1146 {
1147 	char *text;
1148 	struct printk_log *msg;
1149 	int len = 0;
1150 
1151 	text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1152 	if (!text)
1153 		return -ENOMEM;
1154 
1155 	while (size > 0) {
1156 		size_t n;
1157 		size_t skip;
1158 
1159 		raw_spin_lock_irq(&logbuf_lock);
1160 		if (syslog_seq < log_first_seq) {
1161 			/* messages are gone, move to first one */
1162 			syslog_seq = log_first_seq;
1163 			syslog_idx = log_first_idx;
1164 			syslog_prev = 0;
1165 			syslog_partial = 0;
1166 		}
1167 		if (syslog_seq == log_next_seq) {
1168 			raw_spin_unlock_irq(&logbuf_lock);
1169 			break;
1170 		}
1171 
1172 		skip = syslog_partial;
1173 		msg = log_from_idx(syslog_idx);
1174 		n = msg_print_text(msg, syslog_prev, true, text,
1175 				   LOG_LINE_MAX + PREFIX_MAX);
1176 		if (n - syslog_partial <= size) {
1177 			/* message fits into buffer, move forward */
1178 			syslog_idx = log_next(syslog_idx);
1179 			syslog_seq++;
1180 			syslog_prev = msg->flags;
1181 			n -= syslog_partial;
1182 			syslog_partial = 0;
1183 		} else if (!len){
1184 			/* partial read(), remember position */
1185 			n = size;
1186 			syslog_partial += n;
1187 		} else
1188 			n = 0;
1189 		raw_spin_unlock_irq(&logbuf_lock);
1190 
1191 		if (!n)
1192 			break;
1193 
1194 		if (copy_to_user(buf, text + skip, n)) {
1195 			if (!len)
1196 				len = -EFAULT;
1197 			break;
1198 		}
1199 
1200 		len += n;
1201 		size -= n;
1202 		buf += n;
1203 	}
1204 
1205 	kfree(text);
1206 	return len;
1207 }
1208 
1209 static int syslog_print_all(char __user *buf, int size, bool clear)
1210 {
1211 	char *text;
1212 	int len = 0;
1213 
1214 	text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1215 	if (!text)
1216 		return -ENOMEM;
1217 
1218 	raw_spin_lock_irq(&logbuf_lock);
1219 	if (buf) {
1220 		u64 next_seq;
1221 		u64 seq;
1222 		u32 idx;
1223 		enum log_flags prev;
1224 
1225 		/*
1226 		 * Find first record that fits, including all following records,
1227 		 * into the user-provided buffer for this dump.
1228 		 */
1229 		seq = clear_seq;
1230 		idx = clear_idx;
1231 		prev = 0;
1232 		while (seq < log_next_seq) {
1233 			struct printk_log *msg = log_from_idx(idx);
1234 
1235 			len += msg_print_text(msg, prev, true, NULL, 0);
1236 			prev = msg->flags;
1237 			idx = log_next(idx);
1238 			seq++;
1239 		}
1240 
1241 		/* move first record forward until length fits into the buffer */
1242 		seq = clear_seq;
1243 		idx = clear_idx;
1244 		prev = 0;
1245 		while (len > size && seq < log_next_seq) {
1246 			struct printk_log *msg = log_from_idx(idx);
1247 
1248 			len -= msg_print_text(msg, prev, true, NULL, 0);
1249 			prev = msg->flags;
1250 			idx = log_next(idx);
1251 			seq++;
1252 		}
1253 
1254 		/* last message fitting into this dump */
1255 		next_seq = log_next_seq;
1256 
1257 		len = 0;
1258 		while (len >= 0 && seq < next_seq) {
1259 			struct printk_log *msg = log_from_idx(idx);
1260 			int textlen;
1261 
1262 			textlen = msg_print_text(msg, prev, true, text,
1263 						 LOG_LINE_MAX + PREFIX_MAX);
1264 			if (textlen < 0) {
1265 				len = textlen;
1266 				break;
1267 			}
1268 			idx = log_next(idx);
1269 			seq++;
1270 			prev = msg->flags;
1271 
1272 			raw_spin_unlock_irq(&logbuf_lock);
1273 			if (copy_to_user(buf + len, text, textlen))
1274 				len = -EFAULT;
1275 			else
1276 				len += textlen;
1277 			raw_spin_lock_irq(&logbuf_lock);
1278 
1279 			if (seq < log_first_seq) {
1280 				/* messages are gone, move to next one */
1281 				seq = log_first_seq;
1282 				idx = log_first_idx;
1283 				prev = 0;
1284 			}
1285 		}
1286 	}
1287 
1288 	if (clear) {
1289 		clear_seq = log_next_seq;
1290 		clear_idx = log_next_idx;
1291 	}
1292 	raw_spin_unlock_irq(&logbuf_lock);
1293 
1294 	kfree(text);
1295 	return len;
1296 }
1297 
1298 int do_syslog(int type, char __user *buf, int len, int source)
1299 {
1300 	bool clear = false;
1301 	static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1302 	int error;
1303 
1304 	error = check_syslog_permissions(type, source);
1305 	if (error)
1306 		goto out;
1307 
1308 	switch (type) {
1309 	case SYSLOG_ACTION_CLOSE:	/* Close log */
1310 		break;
1311 	case SYSLOG_ACTION_OPEN:	/* Open log */
1312 		break;
1313 	case SYSLOG_ACTION_READ:	/* Read from log */
1314 		error = -EINVAL;
1315 		if (!buf || len < 0)
1316 			goto out;
1317 		error = 0;
1318 		if (!len)
1319 			goto out;
1320 		if (!access_ok(VERIFY_WRITE, buf, len)) {
1321 			error = -EFAULT;
1322 			goto out;
1323 		}
1324 		error = wait_event_interruptible(log_wait,
1325 						 syslog_seq != log_next_seq);
1326 		if (error)
1327 			goto out;
1328 		error = syslog_print(buf, len);
1329 		break;
1330 	/* Read/clear last kernel messages */
1331 	case SYSLOG_ACTION_READ_CLEAR:
1332 		clear = true;
1333 		/* FALL THRU */
1334 	/* Read last kernel messages */
1335 	case SYSLOG_ACTION_READ_ALL:
1336 		error = -EINVAL;
1337 		if (!buf || len < 0)
1338 			goto out;
1339 		error = 0;
1340 		if (!len)
1341 			goto out;
1342 		if (!access_ok(VERIFY_WRITE, buf, len)) {
1343 			error = -EFAULT;
1344 			goto out;
1345 		}
1346 		error = syslog_print_all(buf, len, clear);
1347 		break;
1348 	/* Clear ring buffer */
1349 	case SYSLOG_ACTION_CLEAR:
1350 		syslog_print_all(NULL, 0, true);
1351 		break;
1352 	/* Disable logging to console */
1353 	case SYSLOG_ACTION_CONSOLE_OFF:
1354 		if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1355 			saved_console_loglevel = console_loglevel;
1356 		console_loglevel = minimum_console_loglevel;
1357 		break;
1358 	/* Enable logging to console */
1359 	case SYSLOG_ACTION_CONSOLE_ON:
1360 		if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1361 			console_loglevel = saved_console_loglevel;
1362 			saved_console_loglevel = LOGLEVEL_DEFAULT;
1363 		}
1364 		break;
1365 	/* Set level of messages printed to console */
1366 	case SYSLOG_ACTION_CONSOLE_LEVEL:
1367 		error = -EINVAL;
1368 		if (len < 1 || len > 8)
1369 			goto out;
1370 		if (len < minimum_console_loglevel)
1371 			len = minimum_console_loglevel;
1372 		console_loglevel = len;
1373 		/* Implicitly re-enable logging to console */
1374 		saved_console_loglevel = LOGLEVEL_DEFAULT;
1375 		error = 0;
1376 		break;
1377 	/* Number of chars in the log buffer */
1378 	case SYSLOG_ACTION_SIZE_UNREAD:
1379 		raw_spin_lock_irq(&logbuf_lock);
1380 		if (syslog_seq < log_first_seq) {
1381 			/* messages are gone, move to first one */
1382 			syslog_seq = log_first_seq;
1383 			syslog_idx = log_first_idx;
1384 			syslog_prev = 0;
1385 			syslog_partial = 0;
1386 		}
1387 		if (source == SYSLOG_FROM_PROC) {
1388 			/*
1389 			 * Short-cut for poll(/"proc/kmsg") which simply checks
1390 			 * for pending data, not the size; return the count of
1391 			 * records, not the length.
1392 			 */
1393 			error = log_next_seq - syslog_seq;
1394 		} else {
1395 			u64 seq = syslog_seq;
1396 			u32 idx = syslog_idx;
1397 			enum log_flags prev = syslog_prev;
1398 
1399 			error = 0;
1400 			while (seq < log_next_seq) {
1401 				struct printk_log *msg = log_from_idx(idx);
1402 
1403 				error += msg_print_text(msg, prev, true, NULL, 0);
1404 				idx = log_next(idx);
1405 				seq++;
1406 				prev = msg->flags;
1407 			}
1408 			error -= syslog_partial;
1409 		}
1410 		raw_spin_unlock_irq(&logbuf_lock);
1411 		break;
1412 	/* Size of the log buffer */
1413 	case SYSLOG_ACTION_SIZE_BUFFER:
1414 		error = log_buf_len;
1415 		break;
1416 	default:
1417 		error = -EINVAL;
1418 		break;
1419 	}
1420 out:
1421 	return error;
1422 }
1423 
1424 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1425 {
1426 	return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1427 }
1428 
1429 /*
1430  * Call the console drivers, asking them to write out
1431  * log_buf[start] to log_buf[end - 1].
1432  * The console_lock must be held.
1433  */
1434 static void call_console_drivers(int level,
1435 				 const char *ext_text, size_t ext_len,
1436 				 const char *text, size_t len)
1437 {
1438 	struct console *con;
1439 
1440 	trace_console(text, len);
1441 
1442 	if (level >= console_loglevel && !ignore_loglevel)
1443 		return;
1444 	if (!console_drivers)
1445 		return;
1446 
1447 	for_each_console(con) {
1448 		if (exclusive_console && con != exclusive_console)
1449 			continue;
1450 		if (!(con->flags & CON_ENABLED))
1451 			continue;
1452 		if (!con->write)
1453 			continue;
1454 		if (!cpu_online(smp_processor_id()) &&
1455 		    !(con->flags & CON_ANYTIME))
1456 			continue;
1457 		if (con->flags & CON_EXTENDED)
1458 			con->write(con, ext_text, ext_len);
1459 		else
1460 			con->write(con, text, len);
1461 	}
1462 }
1463 
1464 /*
1465  * Zap console related locks when oopsing.
1466  * To leave time for slow consoles to print a full oops,
1467  * only zap at most once every 30 seconds.
1468  */
1469 static void zap_locks(void)
1470 {
1471 	static unsigned long oops_timestamp;
1472 
1473 	if (time_after_eq(jiffies, oops_timestamp) &&
1474 	    !time_after(jiffies, oops_timestamp + 30 * HZ))
1475 		return;
1476 
1477 	oops_timestamp = jiffies;
1478 
1479 	debug_locks_off();
1480 	/* If a crash is occurring, make sure we can't deadlock */
1481 	raw_spin_lock_init(&logbuf_lock);
1482 	/* And make sure that we print immediately */
1483 	sema_init(&console_sem, 1);
1484 }
1485 
1486 int printk_delay_msec __read_mostly;
1487 
1488 static inline void printk_delay(void)
1489 {
1490 	if (unlikely(printk_delay_msec)) {
1491 		int m = printk_delay_msec;
1492 
1493 		while (m--) {
1494 			mdelay(1);
1495 			touch_nmi_watchdog();
1496 		}
1497 	}
1498 }
1499 
1500 /*
1501  * Continuation lines are buffered, and not committed to the record buffer
1502  * until the line is complete, or a race forces it. The line fragments
1503  * though, are printed immediately to the consoles to ensure everything has
1504  * reached the console in case of a kernel crash.
1505  */
1506 static struct cont {
1507 	char buf[LOG_LINE_MAX];
1508 	size_t len;			/* length == 0 means unused buffer */
1509 	size_t cons;			/* bytes written to console */
1510 	struct task_struct *owner;	/* task of first print*/
1511 	u64 ts_nsec;			/* time of first print */
1512 	u8 level;			/* log level of first message */
1513 	u8 facility;			/* log facility of first message */
1514 	enum log_flags flags;		/* prefix, newline flags */
1515 	bool flushed:1;			/* buffer sealed and committed */
1516 } cont;
1517 
1518 static void cont_flush(enum log_flags flags)
1519 {
1520 	if (cont.flushed)
1521 		return;
1522 	if (cont.len == 0)
1523 		return;
1524 
1525 	if (cont.cons) {
1526 		/*
1527 		 * If a fragment of this line was directly flushed to the
1528 		 * console; wait for the console to pick up the rest of the
1529 		 * line. LOG_NOCONS suppresses a duplicated output.
1530 		 */
1531 		log_store(cont.facility, cont.level, flags | LOG_NOCONS,
1532 			  cont.ts_nsec, NULL, 0, cont.buf, cont.len);
1533 		cont.flags = flags;
1534 		cont.flushed = true;
1535 	} else {
1536 		/*
1537 		 * If no fragment of this line ever reached the console,
1538 		 * just submit it to the store and free the buffer.
1539 		 */
1540 		log_store(cont.facility, cont.level, flags, 0,
1541 			  NULL, 0, cont.buf, cont.len);
1542 		cont.len = 0;
1543 	}
1544 }
1545 
1546 static bool cont_add(int facility, int level, const char *text, size_t len)
1547 {
1548 	if (cont.len && cont.flushed)
1549 		return false;
1550 
1551 	/*
1552 	 * If ext consoles are present, flush and skip in-kernel
1553 	 * continuation.  See nr_ext_console_drivers definition.  Also, if
1554 	 * the line gets too long, split it up in separate records.
1555 	 */
1556 	if (nr_ext_console_drivers || cont.len + len > sizeof(cont.buf)) {
1557 		cont_flush(LOG_CONT);
1558 		return false;
1559 	}
1560 
1561 	if (!cont.len) {
1562 		cont.facility = facility;
1563 		cont.level = level;
1564 		cont.owner = current;
1565 		cont.ts_nsec = local_clock();
1566 		cont.flags = 0;
1567 		cont.cons = 0;
1568 		cont.flushed = false;
1569 	}
1570 
1571 	memcpy(cont.buf + cont.len, text, len);
1572 	cont.len += len;
1573 
1574 	if (cont.len > (sizeof(cont.buf) * 80) / 100)
1575 		cont_flush(LOG_CONT);
1576 
1577 	return true;
1578 }
1579 
1580 static size_t cont_print_text(char *text, size_t size)
1581 {
1582 	size_t textlen = 0;
1583 	size_t len;
1584 
1585 	if (cont.cons == 0 && (console_prev & LOG_NEWLINE)) {
1586 		textlen += print_time(cont.ts_nsec, text);
1587 		size -= textlen;
1588 	}
1589 
1590 	len = cont.len - cont.cons;
1591 	if (len > 0) {
1592 		if (len+1 > size)
1593 			len = size-1;
1594 		memcpy(text + textlen, cont.buf + cont.cons, len);
1595 		textlen += len;
1596 		cont.cons = cont.len;
1597 	}
1598 
1599 	if (cont.flushed) {
1600 		if (cont.flags & LOG_NEWLINE)
1601 			text[textlen++] = '\n';
1602 		/* got everything, release buffer */
1603 		cont.len = 0;
1604 	}
1605 	return textlen;
1606 }
1607 
1608 asmlinkage int vprintk_emit(int facility, int level,
1609 			    const char *dict, size_t dictlen,
1610 			    const char *fmt, va_list args)
1611 {
1612 	static bool recursion_bug;
1613 	static char textbuf[LOG_LINE_MAX];
1614 	char *text = textbuf;
1615 	size_t text_len = 0;
1616 	enum log_flags lflags = 0;
1617 	unsigned long flags;
1618 	int this_cpu;
1619 	int printed_len = 0;
1620 	int nmi_message_lost;
1621 	bool in_sched = false;
1622 	/* cpu currently holding logbuf_lock in this function */
1623 	static unsigned int logbuf_cpu = UINT_MAX;
1624 
1625 	if (level == LOGLEVEL_SCHED) {
1626 		level = LOGLEVEL_DEFAULT;
1627 		in_sched = true;
1628 	}
1629 
1630 	boot_delay_msec(level);
1631 	printk_delay();
1632 
1633 	local_irq_save(flags);
1634 	this_cpu = smp_processor_id();
1635 
1636 	/*
1637 	 * Ouch, printk recursed into itself!
1638 	 */
1639 	if (unlikely(logbuf_cpu == this_cpu)) {
1640 		/*
1641 		 * If a crash is occurring during printk() on this CPU,
1642 		 * then try to get the crash message out but make sure
1643 		 * we can't deadlock. Otherwise just return to avoid the
1644 		 * recursion and return - but flag the recursion so that
1645 		 * it can be printed at the next appropriate moment:
1646 		 */
1647 		if (!oops_in_progress && !lockdep_recursing(current)) {
1648 			recursion_bug = true;
1649 			local_irq_restore(flags);
1650 			return 0;
1651 		}
1652 		zap_locks();
1653 	}
1654 
1655 	lockdep_off();
1656 	/* This stops the holder of console_sem just where we want him */
1657 	raw_spin_lock(&logbuf_lock);
1658 	logbuf_cpu = this_cpu;
1659 
1660 	if (unlikely(recursion_bug)) {
1661 		static const char recursion_msg[] =
1662 			"BUG: recent printk recursion!";
1663 
1664 		recursion_bug = false;
1665 		/* emit KERN_CRIT message */
1666 		printed_len += log_store(0, 2, LOG_PREFIX|LOG_NEWLINE, 0,
1667 					 NULL, 0, recursion_msg,
1668 					 strlen(recursion_msg));
1669 	}
1670 
1671 	nmi_message_lost = get_nmi_message_lost();
1672 	if (unlikely(nmi_message_lost)) {
1673 		text_len = scnprintf(textbuf, sizeof(textbuf),
1674 				     "BAD LUCK: lost %d message(s) from NMI context!",
1675 				     nmi_message_lost);
1676 		printed_len += log_store(0, 2, LOG_PREFIX|LOG_NEWLINE, 0,
1677 					 NULL, 0, textbuf, text_len);
1678 	}
1679 
1680 	/*
1681 	 * The printf needs to come first; we need the syslog
1682 	 * prefix which might be passed-in as a parameter.
1683 	 */
1684 	text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1685 
1686 	/* mark and strip a trailing newline */
1687 	if (text_len && text[text_len-1] == '\n') {
1688 		text_len--;
1689 		lflags |= LOG_NEWLINE;
1690 	}
1691 
1692 	/* strip kernel syslog prefix and extract log level or control flags */
1693 	if (facility == 0) {
1694 		int kern_level = printk_get_level(text);
1695 
1696 		if (kern_level) {
1697 			const char *end_of_header = printk_skip_level(text);
1698 			switch (kern_level) {
1699 			case '0' ... '7':
1700 				if (level == LOGLEVEL_DEFAULT)
1701 					level = kern_level - '0';
1702 				/* fallthrough */
1703 			case 'd':	/* KERN_DEFAULT */
1704 				lflags |= LOG_PREFIX;
1705 			}
1706 			/*
1707 			 * No need to check length here because vscnprintf
1708 			 * put '\0' at the end of the string. Only valid and
1709 			 * newly printed level is detected.
1710 			 */
1711 			text_len -= end_of_header - text;
1712 			text = (char *)end_of_header;
1713 		}
1714 	}
1715 
1716 	if (level == LOGLEVEL_DEFAULT)
1717 		level = default_message_loglevel;
1718 
1719 	if (dict)
1720 		lflags |= LOG_PREFIX|LOG_NEWLINE;
1721 
1722 	if (!(lflags & LOG_NEWLINE)) {
1723 		/*
1724 		 * Flush the conflicting buffer. An earlier newline was missing,
1725 		 * or another task also prints continuation lines.
1726 		 */
1727 		if (cont.len && (lflags & LOG_PREFIX || cont.owner != current))
1728 			cont_flush(LOG_NEWLINE);
1729 
1730 		/* buffer line if possible, otherwise store it right away */
1731 		if (cont_add(facility, level, text, text_len))
1732 			printed_len += text_len;
1733 		else
1734 			printed_len += log_store(facility, level,
1735 						 lflags | LOG_CONT, 0,
1736 						 dict, dictlen, text, text_len);
1737 	} else {
1738 		bool stored = false;
1739 
1740 		/*
1741 		 * If an earlier newline was missing and it was the same task,
1742 		 * either merge it with the current buffer and flush, or if
1743 		 * there was a race with interrupts (prefix == true) then just
1744 		 * flush it out and store this line separately.
1745 		 * If the preceding printk was from a different task and missed
1746 		 * a newline, flush and append the newline.
1747 		 */
1748 		if (cont.len) {
1749 			if (cont.owner == current && !(lflags & LOG_PREFIX))
1750 				stored = cont_add(facility, level, text,
1751 						  text_len);
1752 			cont_flush(LOG_NEWLINE);
1753 		}
1754 
1755 		if (stored)
1756 			printed_len += text_len;
1757 		else
1758 			printed_len += log_store(facility, level, lflags, 0,
1759 						 dict, dictlen, text, text_len);
1760 	}
1761 
1762 	logbuf_cpu = UINT_MAX;
1763 	raw_spin_unlock(&logbuf_lock);
1764 	lockdep_on();
1765 	local_irq_restore(flags);
1766 
1767 	/* If called from the scheduler, we can not call up(). */
1768 	if (!in_sched) {
1769 		lockdep_off();
1770 		/*
1771 		 * Try to acquire and then immediately release the console
1772 		 * semaphore.  The release will print out buffers and wake up
1773 		 * /dev/kmsg and syslog() users.
1774 		 */
1775 		if (console_trylock())
1776 			console_unlock();
1777 		lockdep_on();
1778 	}
1779 
1780 	return printed_len;
1781 }
1782 EXPORT_SYMBOL(vprintk_emit);
1783 
1784 asmlinkage int vprintk(const char *fmt, va_list args)
1785 {
1786 	return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1787 }
1788 EXPORT_SYMBOL(vprintk);
1789 
1790 asmlinkage int printk_emit(int facility, int level,
1791 			   const char *dict, size_t dictlen,
1792 			   const char *fmt, ...)
1793 {
1794 	va_list args;
1795 	int r;
1796 
1797 	va_start(args, fmt);
1798 	r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1799 	va_end(args);
1800 
1801 	return r;
1802 }
1803 EXPORT_SYMBOL(printk_emit);
1804 
1805 int vprintk_default(const char *fmt, va_list args)
1806 {
1807 	int r;
1808 
1809 #ifdef CONFIG_KGDB_KDB
1810 	if (unlikely(kdb_trap_printk)) {
1811 		r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
1812 		return r;
1813 	}
1814 #endif
1815 	r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1816 
1817 	return r;
1818 }
1819 EXPORT_SYMBOL_GPL(vprintk_default);
1820 
1821 /**
1822  * printk - print a kernel message
1823  * @fmt: format string
1824  *
1825  * This is printk(). It can be called from any context. We want it to work.
1826  *
1827  * We try to grab the console_lock. If we succeed, it's easy - we log the
1828  * output and call the console drivers.  If we fail to get the semaphore, we
1829  * place the output into the log buffer and return. The current holder of
1830  * the console_sem will notice the new output in console_unlock(); and will
1831  * send it to the consoles before releasing the lock.
1832  *
1833  * One effect of this deferred printing is that code which calls printk() and
1834  * then changes console_loglevel may break. This is because console_loglevel
1835  * is inspected when the actual printing occurs.
1836  *
1837  * See also:
1838  * printf(3)
1839  *
1840  * See the vsnprintf() documentation for format string extensions over C99.
1841  */
1842 asmlinkage __visible int printk(const char *fmt, ...)
1843 {
1844 	va_list args;
1845 	int r;
1846 
1847 	va_start(args, fmt);
1848 	r = vprintk_func(fmt, args);
1849 	va_end(args);
1850 
1851 	return r;
1852 }
1853 EXPORT_SYMBOL(printk);
1854 
1855 #else /* CONFIG_PRINTK */
1856 
1857 #define LOG_LINE_MAX		0
1858 #define PREFIX_MAX		0
1859 
1860 static u64 syslog_seq;
1861 static u32 syslog_idx;
1862 static u64 console_seq;
1863 static u32 console_idx;
1864 static enum log_flags syslog_prev;
1865 static u64 log_first_seq;
1866 static u32 log_first_idx;
1867 static u64 log_next_seq;
1868 static enum log_flags console_prev;
1869 static struct cont {
1870 	size_t len;
1871 	size_t cons;
1872 	u8 level;
1873 	bool flushed:1;
1874 } cont;
1875 static char *log_text(const struct printk_log *msg) { return NULL; }
1876 static char *log_dict(const struct printk_log *msg) { return NULL; }
1877 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
1878 static u32 log_next(u32 idx) { return 0; }
1879 static ssize_t msg_print_ext_header(char *buf, size_t size,
1880 				    struct printk_log *msg, u64 seq,
1881 				    enum log_flags prev_flags) { return 0; }
1882 static ssize_t msg_print_ext_body(char *buf, size_t size,
1883 				  char *dict, size_t dict_len,
1884 				  char *text, size_t text_len) { return 0; }
1885 static void call_console_drivers(int level,
1886 				 const char *ext_text, size_t ext_len,
1887 				 const char *text, size_t len) {}
1888 static size_t msg_print_text(const struct printk_log *msg, enum log_flags prev,
1889 			     bool syslog, char *buf, size_t size) { return 0; }
1890 static size_t cont_print_text(char *text, size_t size) { return 0; }
1891 
1892 /* Still needs to be defined for users */
1893 DEFINE_PER_CPU(printk_func_t, printk_func);
1894 
1895 #endif /* CONFIG_PRINTK */
1896 
1897 #ifdef CONFIG_EARLY_PRINTK
1898 struct console *early_console;
1899 
1900 asmlinkage __visible void early_printk(const char *fmt, ...)
1901 {
1902 	va_list ap;
1903 	char buf[512];
1904 	int n;
1905 
1906 	if (!early_console)
1907 		return;
1908 
1909 	va_start(ap, fmt);
1910 	n = vscnprintf(buf, sizeof(buf), fmt, ap);
1911 	va_end(ap);
1912 
1913 	early_console->write(early_console, buf, n);
1914 }
1915 #endif
1916 
1917 static int __add_preferred_console(char *name, int idx, char *options,
1918 				   char *brl_options)
1919 {
1920 	struct console_cmdline *c;
1921 	int i;
1922 
1923 	/*
1924 	 *	See if this tty is not yet registered, and
1925 	 *	if we have a slot free.
1926 	 */
1927 	for (i = 0, c = console_cmdline;
1928 	     i < MAX_CMDLINECONSOLES && c->name[0];
1929 	     i++, c++) {
1930 		if (strcmp(c->name, name) == 0 && c->index == idx) {
1931 			if (!brl_options)
1932 				selected_console = i;
1933 			return 0;
1934 		}
1935 	}
1936 	if (i == MAX_CMDLINECONSOLES)
1937 		return -E2BIG;
1938 	if (!brl_options)
1939 		selected_console = i;
1940 	strlcpy(c->name, name, sizeof(c->name));
1941 	c->options = options;
1942 	braille_set_options(c, brl_options);
1943 
1944 	c->index = idx;
1945 	return 0;
1946 }
1947 /*
1948  * Set up a console.  Called via do_early_param() in init/main.c
1949  * for each "console=" parameter in the boot command line.
1950  */
1951 static int __init console_setup(char *str)
1952 {
1953 	char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
1954 	char *s, *options, *brl_options = NULL;
1955 	int idx;
1956 
1957 	if (_braille_console_setup(&str, &brl_options))
1958 		return 1;
1959 
1960 	/*
1961 	 * Decode str into name, index, options.
1962 	 */
1963 	if (str[0] >= '0' && str[0] <= '9') {
1964 		strcpy(buf, "ttyS");
1965 		strncpy(buf + 4, str, sizeof(buf) - 5);
1966 	} else {
1967 		strncpy(buf, str, sizeof(buf) - 1);
1968 	}
1969 	buf[sizeof(buf) - 1] = 0;
1970 	options = strchr(str, ',');
1971 	if (options)
1972 		*(options++) = 0;
1973 #ifdef __sparc__
1974 	if (!strcmp(str, "ttya"))
1975 		strcpy(buf, "ttyS0");
1976 	if (!strcmp(str, "ttyb"))
1977 		strcpy(buf, "ttyS1");
1978 #endif
1979 	for (s = buf; *s; s++)
1980 		if (isdigit(*s) || *s == ',')
1981 			break;
1982 	idx = simple_strtoul(s, NULL, 10);
1983 	*s = 0;
1984 
1985 	__add_preferred_console(buf, idx, options, brl_options);
1986 	console_set_on_cmdline = 1;
1987 	return 1;
1988 }
1989 __setup("console=", console_setup);
1990 
1991 /**
1992  * add_preferred_console - add a device to the list of preferred consoles.
1993  * @name: device name
1994  * @idx: device index
1995  * @options: options for this console
1996  *
1997  * The last preferred console added will be used for kernel messages
1998  * and stdin/out/err for init.  Normally this is used by console_setup
1999  * above to handle user-supplied console arguments; however it can also
2000  * be used by arch-specific code either to override the user or more
2001  * commonly to provide a default console (ie from PROM variables) when
2002  * the user has not supplied one.
2003  */
2004 int add_preferred_console(char *name, int idx, char *options)
2005 {
2006 	return __add_preferred_console(name, idx, options, NULL);
2007 }
2008 
2009 bool console_suspend_enabled = true;
2010 EXPORT_SYMBOL(console_suspend_enabled);
2011 
2012 static int __init console_suspend_disable(char *str)
2013 {
2014 	console_suspend_enabled = false;
2015 	return 1;
2016 }
2017 __setup("no_console_suspend", console_suspend_disable);
2018 module_param_named(console_suspend, console_suspend_enabled,
2019 		bool, S_IRUGO | S_IWUSR);
2020 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2021 	" and hibernate operations");
2022 
2023 /**
2024  * suspend_console - suspend the console subsystem
2025  *
2026  * This disables printk() while we go into suspend states
2027  */
2028 void suspend_console(void)
2029 {
2030 	if (!console_suspend_enabled)
2031 		return;
2032 	printk("Suspending console(s) (use no_console_suspend to debug)\n");
2033 	console_lock();
2034 	console_suspended = 1;
2035 	up_console_sem();
2036 }
2037 
2038 void resume_console(void)
2039 {
2040 	if (!console_suspend_enabled)
2041 		return;
2042 	down_console_sem();
2043 	console_suspended = 0;
2044 	console_unlock();
2045 }
2046 
2047 /**
2048  * console_cpu_notify - print deferred console messages after CPU hotplug
2049  * @self: notifier struct
2050  * @action: CPU hotplug event
2051  * @hcpu: unused
2052  *
2053  * If printk() is called from a CPU that is not online yet, the messages
2054  * will be spooled but will not show up on the console.  This function is
2055  * called when a new CPU comes online (or fails to come up), and ensures
2056  * that any such output gets printed.
2057  */
2058 static int console_cpu_notify(struct notifier_block *self,
2059 	unsigned long action, void *hcpu)
2060 {
2061 	switch (action) {
2062 	case CPU_ONLINE:
2063 	case CPU_DEAD:
2064 	case CPU_DOWN_FAILED:
2065 	case CPU_UP_CANCELED:
2066 		console_lock();
2067 		console_unlock();
2068 	}
2069 	return NOTIFY_OK;
2070 }
2071 
2072 /**
2073  * console_lock - lock the console system for exclusive use.
2074  *
2075  * Acquires a lock which guarantees that the caller has
2076  * exclusive access to the console system and the console_drivers list.
2077  *
2078  * Can sleep, returns nothing.
2079  */
2080 void console_lock(void)
2081 {
2082 	might_sleep();
2083 
2084 	down_console_sem();
2085 	if (console_suspended)
2086 		return;
2087 	console_locked = 1;
2088 	console_may_schedule = 1;
2089 }
2090 EXPORT_SYMBOL(console_lock);
2091 
2092 /**
2093  * console_trylock - try to lock the console system for exclusive use.
2094  *
2095  * Try to acquire a lock which guarantees that the caller has exclusive
2096  * access to the console system and the console_drivers list.
2097  *
2098  * returns 1 on success, and 0 on failure to acquire the lock.
2099  */
2100 int console_trylock(void)
2101 {
2102 	if (down_trylock_console_sem())
2103 		return 0;
2104 	if (console_suspended) {
2105 		up_console_sem();
2106 		return 0;
2107 	}
2108 	console_locked = 1;
2109 	/*
2110 	 * When PREEMPT_COUNT disabled we can't reliably detect if it's
2111 	 * safe to schedule (e.g. calling printk while holding a spin_lock),
2112 	 * because preempt_disable()/preempt_enable() are just barriers there
2113 	 * and preempt_count() is always 0.
2114 	 *
2115 	 * RCU read sections have a separate preemption counter when
2116 	 * PREEMPT_RCU enabled thus we must take extra care and check
2117 	 * rcu_preempt_depth(), otherwise RCU read sections modify
2118 	 * preempt_count().
2119 	 */
2120 	console_may_schedule = !oops_in_progress &&
2121 			preemptible() &&
2122 			!rcu_preempt_depth();
2123 	return 1;
2124 }
2125 EXPORT_SYMBOL(console_trylock);
2126 
2127 int is_console_locked(void)
2128 {
2129 	return console_locked;
2130 }
2131 
2132 /*
2133  * Check if we have any console that is capable of printing while cpu is
2134  * booting or shutting down. Requires console_sem.
2135  */
2136 static int have_callable_console(void)
2137 {
2138 	struct console *con;
2139 
2140 	for_each_console(con)
2141 		if ((con->flags & CON_ENABLED) &&
2142 				(con->flags & CON_ANYTIME))
2143 			return 1;
2144 
2145 	return 0;
2146 }
2147 
2148 /*
2149  * Can we actually use the console at this time on this cpu?
2150  *
2151  * Console drivers may assume that per-cpu resources have been allocated. So
2152  * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2153  * call them until this CPU is officially up.
2154  */
2155 static inline int can_use_console(void)
2156 {
2157 	return cpu_online(raw_smp_processor_id()) || have_callable_console();
2158 }
2159 
2160 static void console_cont_flush(char *text, size_t size)
2161 {
2162 	unsigned long flags;
2163 	size_t len;
2164 
2165 	raw_spin_lock_irqsave(&logbuf_lock, flags);
2166 
2167 	if (!cont.len)
2168 		goto out;
2169 
2170 	/*
2171 	 * We still queue earlier records, likely because the console was
2172 	 * busy. The earlier ones need to be printed before this one, we
2173 	 * did not flush any fragment so far, so just let it queue up.
2174 	 */
2175 	if (console_seq < log_next_seq && !cont.cons)
2176 		goto out;
2177 
2178 	len = cont_print_text(text, size);
2179 	raw_spin_unlock(&logbuf_lock);
2180 	stop_critical_timings();
2181 	call_console_drivers(cont.level, NULL, 0, text, len);
2182 	start_critical_timings();
2183 	local_irq_restore(flags);
2184 	return;
2185 out:
2186 	raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2187 }
2188 
2189 /**
2190  * console_unlock - unlock the console system
2191  *
2192  * Releases the console_lock which the caller holds on the console system
2193  * and the console driver list.
2194  *
2195  * While the console_lock was held, console output may have been buffered
2196  * by printk().  If this is the case, console_unlock(); emits
2197  * the output prior to releasing the lock.
2198  *
2199  * If there is output waiting, we wake /dev/kmsg and syslog() users.
2200  *
2201  * console_unlock(); may be called from any context.
2202  */
2203 void console_unlock(void)
2204 {
2205 	static char ext_text[CONSOLE_EXT_LOG_MAX];
2206 	static char text[LOG_LINE_MAX + PREFIX_MAX];
2207 	static u64 seen_seq;
2208 	unsigned long flags;
2209 	bool wake_klogd = false;
2210 	bool do_cond_resched, retry;
2211 
2212 	if (console_suspended) {
2213 		up_console_sem();
2214 		return;
2215 	}
2216 
2217 	/*
2218 	 * Console drivers are called under logbuf_lock, so
2219 	 * @console_may_schedule should be cleared before; however, we may
2220 	 * end up dumping a lot of lines, for example, if called from
2221 	 * console registration path, and should invoke cond_resched()
2222 	 * between lines if allowable.  Not doing so can cause a very long
2223 	 * scheduling stall on a slow console leading to RCU stall and
2224 	 * softlockup warnings which exacerbate the issue with more
2225 	 * messages practically incapacitating the system.
2226 	 */
2227 	do_cond_resched = console_may_schedule;
2228 	console_may_schedule = 0;
2229 
2230 again:
2231 	/*
2232 	 * We released the console_sem lock, so we need to recheck if
2233 	 * cpu is online and (if not) is there at least one CON_ANYTIME
2234 	 * console.
2235 	 */
2236 	if (!can_use_console()) {
2237 		console_locked = 0;
2238 		up_console_sem();
2239 		return;
2240 	}
2241 
2242 	/* flush buffered message fragment immediately to console */
2243 	console_cont_flush(text, sizeof(text));
2244 
2245 	for (;;) {
2246 		struct printk_log *msg;
2247 		size_t ext_len = 0;
2248 		size_t len;
2249 		int level;
2250 
2251 		raw_spin_lock_irqsave(&logbuf_lock, flags);
2252 		if (seen_seq != log_next_seq) {
2253 			wake_klogd = true;
2254 			seen_seq = log_next_seq;
2255 		}
2256 
2257 		if (console_seq < log_first_seq) {
2258 			len = sprintf(text, "** %u printk messages dropped ** ",
2259 				      (unsigned)(log_first_seq - console_seq));
2260 
2261 			/* messages are gone, move to first one */
2262 			console_seq = log_first_seq;
2263 			console_idx = log_first_idx;
2264 			console_prev = 0;
2265 		} else {
2266 			len = 0;
2267 		}
2268 skip:
2269 		if (console_seq == log_next_seq)
2270 			break;
2271 
2272 		msg = log_from_idx(console_idx);
2273 		if (msg->flags & LOG_NOCONS) {
2274 			/*
2275 			 * Skip record we have buffered and already printed
2276 			 * directly to the console when we received it.
2277 			 */
2278 			console_idx = log_next(console_idx);
2279 			console_seq++;
2280 			/*
2281 			 * We will get here again when we register a new
2282 			 * CON_PRINTBUFFER console. Clear the flag so we
2283 			 * will properly dump everything later.
2284 			 */
2285 			msg->flags &= ~LOG_NOCONS;
2286 			console_prev = msg->flags;
2287 			goto skip;
2288 		}
2289 
2290 		level = msg->level;
2291 		len += msg_print_text(msg, console_prev, false,
2292 				      text + len, sizeof(text) - len);
2293 		if (nr_ext_console_drivers) {
2294 			ext_len = msg_print_ext_header(ext_text,
2295 						sizeof(ext_text),
2296 						msg, console_seq, console_prev);
2297 			ext_len += msg_print_ext_body(ext_text + ext_len,
2298 						sizeof(ext_text) - ext_len,
2299 						log_dict(msg), msg->dict_len,
2300 						log_text(msg), msg->text_len);
2301 		}
2302 		console_idx = log_next(console_idx);
2303 		console_seq++;
2304 		console_prev = msg->flags;
2305 		raw_spin_unlock(&logbuf_lock);
2306 
2307 		stop_critical_timings();	/* don't trace print latency */
2308 		call_console_drivers(level, ext_text, ext_len, text, len);
2309 		start_critical_timings();
2310 		local_irq_restore(flags);
2311 
2312 		if (do_cond_resched)
2313 			cond_resched();
2314 	}
2315 	console_locked = 0;
2316 
2317 	/* Release the exclusive_console once it is used */
2318 	if (unlikely(exclusive_console))
2319 		exclusive_console = NULL;
2320 
2321 	raw_spin_unlock(&logbuf_lock);
2322 
2323 	up_console_sem();
2324 
2325 	/*
2326 	 * Someone could have filled up the buffer again, so re-check if there's
2327 	 * something to flush. In case we cannot trylock the console_sem again,
2328 	 * there's a new owner and the console_unlock() from them will do the
2329 	 * flush, no worries.
2330 	 */
2331 	raw_spin_lock(&logbuf_lock);
2332 	retry = console_seq != log_next_seq;
2333 	raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2334 
2335 	if (retry && console_trylock())
2336 		goto again;
2337 
2338 	if (wake_klogd)
2339 		wake_up_klogd();
2340 }
2341 EXPORT_SYMBOL(console_unlock);
2342 
2343 /**
2344  * console_conditional_schedule - yield the CPU if required
2345  *
2346  * If the console code is currently allowed to sleep, and
2347  * if this CPU should yield the CPU to another task, do
2348  * so here.
2349  *
2350  * Must be called within console_lock();.
2351  */
2352 void __sched console_conditional_schedule(void)
2353 {
2354 	if (console_may_schedule)
2355 		cond_resched();
2356 }
2357 EXPORT_SYMBOL(console_conditional_schedule);
2358 
2359 void console_unblank(void)
2360 {
2361 	struct console *c;
2362 
2363 	/*
2364 	 * console_unblank can no longer be called in interrupt context unless
2365 	 * oops_in_progress is set to 1..
2366 	 */
2367 	if (oops_in_progress) {
2368 		if (down_trylock_console_sem() != 0)
2369 			return;
2370 	} else
2371 		console_lock();
2372 
2373 	console_locked = 1;
2374 	console_may_schedule = 0;
2375 	for_each_console(c)
2376 		if ((c->flags & CON_ENABLED) && c->unblank)
2377 			c->unblank();
2378 	console_unlock();
2379 }
2380 
2381 /**
2382  * console_flush_on_panic - flush console content on panic
2383  *
2384  * Immediately output all pending messages no matter what.
2385  */
2386 void console_flush_on_panic(void)
2387 {
2388 	/*
2389 	 * If someone else is holding the console lock, trylock will fail
2390 	 * and may_schedule may be set.  Ignore and proceed to unlock so
2391 	 * that messages are flushed out.  As this can be called from any
2392 	 * context and we don't want to get preempted while flushing,
2393 	 * ensure may_schedule is cleared.
2394 	 */
2395 	console_trylock();
2396 	console_may_schedule = 0;
2397 	console_unlock();
2398 }
2399 
2400 /*
2401  * Return the console tty driver structure and its associated index
2402  */
2403 struct tty_driver *console_device(int *index)
2404 {
2405 	struct console *c;
2406 	struct tty_driver *driver = NULL;
2407 
2408 	console_lock();
2409 	for_each_console(c) {
2410 		if (!c->device)
2411 			continue;
2412 		driver = c->device(c, index);
2413 		if (driver)
2414 			break;
2415 	}
2416 	console_unlock();
2417 	return driver;
2418 }
2419 
2420 /*
2421  * Prevent further output on the passed console device so that (for example)
2422  * serial drivers can disable console output before suspending a port, and can
2423  * re-enable output afterwards.
2424  */
2425 void console_stop(struct console *console)
2426 {
2427 	console_lock();
2428 	console->flags &= ~CON_ENABLED;
2429 	console_unlock();
2430 }
2431 EXPORT_SYMBOL(console_stop);
2432 
2433 void console_start(struct console *console)
2434 {
2435 	console_lock();
2436 	console->flags |= CON_ENABLED;
2437 	console_unlock();
2438 }
2439 EXPORT_SYMBOL(console_start);
2440 
2441 static int __read_mostly keep_bootcon;
2442 
2443 static int __init keep_bootcon_setup(char *str)
2444 {
2445 	keep_bootcon = 1;
2446 	pr_info("debug: skip boot console de-registration.\n");
2447 
2448 	return 0;
2449 }
2450 
2451 early_param("keep_bootcon", keep_bootcon_setup);
2452 
2453 /*
2454  * The console driver calls this routine during kernel initialization
2455  * to register the console printing procedure with printk() and to
2456  * print any messages that were printed by the kernel before the
2457  * console driver was initialized.
2458  *
2459  * This can happen pretty early during the boot process (because of
2460  * early_printk) - sometimes before setup_arch() completes - be careful
2461  * of what kernel features are used - they may not be initialised yet.
2462  *
2463  * There are two types of consoles - bootconsoles (early_printk) and
2464  * "real" consoles (everything which is not a bootconsole) which are
2465  * handled differently.
2466  *  - Any number of bootconsoles can be registered at any time.
2467  *  - As soon as a "real" console is registered, all bootconsoles
2468  *    will be unregistered automatically.
2469  *  - Once a "real" console is registered, any attempt to register a
2470  *    bootconsoles will be rejected
2471  */
2472 void register_console(struct console *newcon)
2473 {
2474 	int i;
2475 	unsigned long flags;
2476 	struct console *bcon = NULL;
2477 	struct console_cmdline *c;
2478 
2479 	if (console_drivers)
2480 		for_each_console(bcon)
2481 			if (WARN(bcon == newcon,
2482 					"console '%s%d' already registered\n",
2483 					bcon->name, bcon->index))
2484 				return;
2485 
2486 	/*
2487 	 * before we register a new CON_BOOT console, make sure we don't
2488 	 * already have a valid console
2489 	 */
2490 	if (console_drivers && newcon->flags & CON_BOOT) {
2491 		/* find the last or real console */
2492 		for_each_console(bcon) {
2493 			if (!(bcon->flags & CON_BOOT)) {
2494 				pr_info("Too late to register bootconsole %s%d\n",
2495 					newcon->name, newcon->index);
2496 				return;
2497 			}
2498 		}
2499 	}
2500 
2501 	if (console_drivers && console_drivers->flags & CON_BOOT)
2502 		bcon = console_drivers;
2503 
2504 	if (preferred_console < 0 || bcon || !console_drivers)
2505 		preferred_console = selected_console;
2506 
2507 	/*
2508 	 *	See if we want to use this console driver. If we
2509 	 *	didn't select a console we take the first one
2510 	 *	that registers here.
2511 	 */
2512 	if (preferred_console < 0) {
2513 		if (newcon->index < 0)
2514 			newcon->index = 0;
2515 		if (newcon->setup == NULL ||
2516 		    newcon->setup(newcon, NULL) == 0) {
2517 			newcon->flags |= CON_ENABLED;
2518 			if (newcon->device) {
2519 				newcon->flags |= CON_CONSDEV;
2520 				preferred_console = 0;
2521 			}
2522 		}
2523 	}
2524 
2525 	/*
2526 	 *	See if this console matches one we selected on
2527 	 *	the command line.
2528 	 */
2529 	for (i = 0, c = console_cmdline;
2530 	     i < MAX_CMDLINECONSOLES && c->name[0];
2531 	     i++, c++) {
2532 		if (!newcon->match ||
2533 		    newcon->match(newcon, c->name, c->index, c->options) != 0) {
2534 			/* default matching */
2535 			BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2536 			if (strcmp(c->name, newcon->name) != 0)
2537 				continue;
2538 			if (newcon->index >= 0 &&
2539 			    newcon->index != c->index)
2540 				continue;
2541 			if (newcon->index < 0)
2542 				newcon->index = c->index;
2543 
2544 			if (_braille_register_console(newcon, c))
2545 				return;
2546 
2547 			if (newcon->setup &&
2548 			    newcon->setup(newcon, c->options) != 0)
2549 				break;
2550 		}
2551 
2552 		newcon->flags |= CON_ENABLED;
2553 		if (i == selected_console) {
2554 			newcon->flags |= CON_CONSDEV;
2555 			preferred_console = selected_console;
2556 		}
2557 		break;
2558 	}
2559 
2560 	if (!(newcon->flags & CON_ENABLED))
2561 		return;
2562 
2563 	/*
2564 	 * If we have a bootconsole, and are switching to a real console,
2565 	 * don't print everything out again, since when the boot console, and
2566 	 * the real console are the same physical device, it's annoying to
2567 	 * see the beginning boot messages twice
2568 	 */
2569 	if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2570 		newcon->flags &= ~CON_PRINTBUFFER;
2571 
2572 	/*
2573 	 *	Put this console in the list - keep the
2574 	 *	preferred driver at the head of the list.
2575 	 */
2576 	console_lock();
2577 	if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2578 		newcon->next = console_drivers;
2579 		console_drivers = newcon;
2580 		if (newcon->next)
2581 			newcon->next->flags &= ~CON_CONSDEV;
2582 	} else {
2583 		newcon->next = console_drivers->next;
2584 		console_drivers->next = newcon;
2585 	}
2586 
2587 	if (newcon->flags & CON_EXTENDED)
2588 		if (!nr_ext_console_drivers++)
2589 			pr_info("printk: continuation disabled due to ext consoles, expect more fragments in /dev/kmsg\n");
2590 
2591 	if (newcon->flags & CON_PRINTBUFFER) {
2592 		/*
2593 		 * console_unlock(); will print out the buffered messages
2594 		 * for us.
2595 		 */
2596 		raw_spin_lock_irqsave(&logbuf_lock, flags);
2597 		console_seq = syslog_seq;
2598 		console_idx = syslog_idx;
2599 		console_prev = syslog_prev;
2600 		raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2601 		/*
2602 		 * We're about to replay the log buffer.  Only do this to the
2603 		 * just-registered console to avoid excessive message spam to
2604 		 * the already-registered consoles.
2605 		 */
2606 		exclusive_console = newcon;
2607 	}
2608 	console_unlock();
2609 	console_sysfs_notify();
2610 
2611 	/*
2612 	 * By unregistering the bootconsoles after we enable the real console
2613 	 * we get the "console xxx enabled" message on all the consoles -
2614 	 * boot consoles, real consoles, etc - this is to ensure that end
2615 	 * users know there might be something in the kernel's log buffer that
2616 	 * went to the bootconsole (that they do not see on the real console)
2617 	 */
2618 	pr_info("%sconsole [%s%d] enabled\n",
2619 		(newcon->flags & CON_BOOT) ? "boot" : "" ,
2620 		newcon->name, newcon->index);
2621 	if (bcon &&
2622 	    ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2623 	    !keep_bootcon) {
2624 		/* We need to iterate through all boot consoles, to make
2625 		 * sure we print everything out, before we unregister them.
2626 		 */
2627 		for_each_console(bcon)
2628 			if (bcon->flags & CON_BOOT)
2629 				unregister_console(bcon);
2630 	}
2631 }
2632 EXPORT_SYMBOL(register_console);
2633 
2634 int unregister_console(struct console *console)
2635 {
2636         struct console *a, *b;
2637 	int res;
2638 
2639 	pr_info("%sconsole [%s%d] disabled\n",
2640 		(console->flags & CON_BOOT) ? "boot" : "" ,
2641 		console->name, console->index);
2642 
2643 	res = _braille_unregister_console(console);
2644 	if (res)
2645 		return res;
2646 
2647 	res = 1;
2648 	console_lock();
2649 	if (console_drivers == console) {
2650 		console_drivers=console->next;
2651 		res = 0;
2652 	} else if (console_drivers) {
2653 		for (a=console_drivers->next, b=console_drivers ;
2654 		     a; b=a, a=b->next) {
2655 			if (a == console) {
2656 				b->next = a->next;
2657 				res = 0;
2658 				break;
2659 			}
2660 		}
2661 	}
2662 
2663 	if (!res && (console->flags & CON_EXTENDED))
2664 		nr_ext_console_drivers--;
2665 
2666 	/*
2667 	 * If this isn't the last console and it has CON_CONSDEV set, we
2668 	 * need to set it on the next preferred console.
2669 	 */
2670 	if (console_drivers != NULL && console->flags & CON_CONSDEV)
2671 		console_drivers->flags |= CON_CONSDEV;
2672 
2673 	console->flags &= ~CON_ENABLED;
2674 	console_unlock();
2675 	console_sysfs_notify();
2676 	return res;
2677 }
2678 EXPORT_SYMBOL(unregister_console);
2679 
2680 /*
2681  * Some boot consoles access data that is in the init section and which will
2682  * be discarded after the initcalls have been run. To make sure that no code
2683  * will access this data, unregister the boot consoles in a late initcall.
2684  *
2685  * If for some reason, such as deferred probe or the driver being a loadable
2686  * module, the real console hasn't registered yet at this point, there will
2687  * be a brief interval in which no messages are logged to the console, which
2688  * makes it difficult to diagnose problems that occur during this time.
2689  *
2690  * To mitigate this problem somewhat, only unregister consoles whose memory
2691  * intersects with the init section. Note that code exists elsewhere to get
2692  * rid of the boot console as soon as the proper console shows up, so there
2693  * won't be side-effects from postponing the removal.
2694  */
2695 static int __init printk_late_init(void)
2696 {
2697 	struct console *con;
2698 
2699 	for_each_console(con) {
2700 		if (!keep_bootcon && con->flags & CON_BOOT) {
2701 			/*
2702 			 * Make sure to unregister boot consoles whose data
2703 			 * resides in the init section before the init section
2704 			 * is discarded. Boot consoles whose data will stick
2705 			 * around will automatically be unregistered when the
2706 			 * proper console replaces them.
2707 			 */
2708 			if (init_section_intersects(con, sizeof(*con)))
2709 				unregister_console(con);
2710 		}
2711 	}
2712 	hotcpu_notifier(console_cpu_notify, 0);
2713 	return 0;
2714 }
2715 late_initcall(printk_late_init);
2716 
2717 #if defined CONFIG_PRINTK
2718 /*
2719  * Delayed printk version, for scheduler-internal messages:
2720  */
2721 #define PRINTK_PENDING_WAKEUP	0x01
2722 #define PRINTK_PENDING_OUTPUT	0x02
2723 
2724 static DEFINE_PER_CPU(int, printk_pending);
2725 
2726 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2727 {
2728 	int pending = __this_cpu_xchg(printk_pending, 0);
2729 
2730 	if (pending & PRINTK_PENDING_OUTPUT) {
2731 		/* If trylock fails, someone else is doing the printing */
2732 		if (console_trylock())
2733 			console_unlock();
2734 	}
2735 
2736 	if (pending & PRINTK_PENDING_WAKEUP)
2737 		wake_up_interruptible(&log_wait);
2738 }
2739 
2740 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2741 	.func = wake_up_klogd_work_func,
2742 	.flags = IRQ_WORK_LAZY,
2743 };
2744 
2745 void wake_up_klogd(void)
2746 {
2747 	preempt_disable();
2748 	if (waitqueue_active(&log_wait)) {
2749 		this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2750 		irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2751 	}
2752 	preempt_enable();
2753 }
2754 
2755 int printk_deferred(const char *fmt, ...)
2756 {
2757 	va_list args;
2758 	int r;
2759 
2760 	preempt_disable();
2761 	va_start(args, fmt);
2762 	r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2763 	va_end(args);
2764 
2765 	__this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2766 	irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2767 	preempt_enable();
2768 
2769 	return r;
2770 }
2771 
2772 /*
2773  * printk rate limiting, lifted from the networking subsystem.
2774  *
2775  * This enforces a rate limit: not more than 10 kernel messages
2776  * every 5s to make a denial-of-service attack impossible.
2777  */
2778 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2779 
2780 int __printk_ratelimit(const char *func)
2781 {
2782 	return ___ratelimit(&printk_ratelimit_state, func);
2783 }
2784 EXPORT_SYMBOL(__printk_ratelimit);
2785 
2786 /**
2787  * printk_timed_ratelimit - caller-controlled printk ratelimiting
2788  * @caller_jiffies: pointer to caller's state
2789  * @interval_msecs: minimum interval between prints
2790  *
2791  * printk_timed_ratelimit() returns true if more than @interval_msecs
2792  * milliseconds have elapsed since the last time printk_timed_ratelimit()
2793  * returned true.
2794  */
2795 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
2796 			unsigned int interval_msecs)
2797 {
2798 	unsigned long elapsed = jiffies - *caller_jiffies;
2799 
2800 	if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
2801 		return false;
2802 
2803 	*caller_jiffies = jiffies;
2804 	return true;
2805 }
2806 EXPORT_SYMBOL(printk_timed_ratelimit);
2807 
2808 static DEFINE_SPINLOCK(dump_list_lock);
2809 static LIST_HEAD(dump_list);
2810 
2811 /**
2812  * kmsg_dump_register - register a kernel log dumper.
2813  * @dumper: pointer to the kmsg_dumper structure
2814  *
2815  * Adds a kernel log dumper to the system. The dump callback in the
2816  * structure will be called when the kernel oopses or panics and must be
2817  * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
2818  */
2819 int kmsg_dump_register(struct kmsg_dumper *dumper)
2820 {
2821 	unsigned long flags;
2822 	int err = -EBUSY;
2823 
2824 	/* The dump callback needs to be set */
2825 	if (!dumper->dump)
2826 		return -EINVAL;
2827 
2828 	spin_lock_irqsave(&dump_list_lock, flags);
2829 	/* Don't allow registering multiple times */
2830 	if (!dumper->registered) {
2831 		dumper->registered = 1;
2832 		list_add_tail_rcu(&dumper->list, &dump_list);
2833 		err = 0;
2834 	}
2835 	spin_unlock_irqrestore(&dump_list_lock, flags);
2836 
2837 	return err;
2838 }
2839 EXPORT_SYMBOL_GPL(kmsg_dump_register);
2840 
2841 /**
2842  * kmsg_dump_unregister - unregister a kmsg dumper.
2843  * @dumper: pointer to the kmsg_dumper structure
2844  *
2845  * Removes a dump device from the system. Returns zero on success and
2846  * %-EINVAL otherwise.
2847  */
2848 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
2849 {
2850 	unsigned long flags;
2851 	int err = -EINVAL;
2852 
2853 	spin_lock_irqsave(&dump_list_lock, flags);
2854 	if (dumper->registered) {
2855 		dumper->registered = 0;
2856 		list_del_rcu(&dumper->list);
2857 		err = 0;
2858 	}
2859 	spin_unlock_irqrestore(&dump_list_lock, flags);
2860 	synchronize_rcu();
2861 
2862 	return err;
2863 }
2864 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
2865 
2866 static bool always_kmsg_dump;
2867 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
2868 
2869 /**
2870  * kmsg_dump - dump kernel log to kernel message dumpers.
2871  * @reason: the reason (oops, panic etc) for dumping
2872  *
2873  * Call each of the registered dumper's dump() callback, which can
2874  * retrieve the kmsg records with kmsg_dump_get_line() or
2875  * kmsg_dump_get_buffer().
2876  */
2877 void kmsg_dump(enum kmsg_dump_reason reason)
2878 {
2879 	struct kmsg_dumper *dumper;
2880 	unsigned long flags;
2881 
2882 	if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
2883 		return;
2884 
2885 	rcu_read_lock();
2886 	list_for_each_entry_rcu(dumper, &dump_list, list) {
2887 		if (dumper->max_reason && reason > dumper->max_reason)
2888 			continue;
2889 
2890 		/* initialize iterator with data about the stored records */
2891 		dumper->active = true;
2892 
2893 		raw_spin_lock_irqsave(&logbuf_lock, flags);
2894 		dumper->cur_seq = clear_seq;
2895 		dumper->cur_idx = clear_idx;
2896 		dumper->next_seq = log_next_seq;
2897 		dumper->next_idx = log_next_idx;
2898 		raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2899 
2900 		/* invoke dumper which will iterate over records */
2901 		dumper->dump(dumper, reason);
2902 
2903 		/* reset iterator */
2904 		dumper->active = false;
2905 	}
2906 	rcu_read_unlock();
2907 }
2908 
2909 /**
2910  * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
2911  * @dumper: registered kmsg dumper
2912  * @syslog: include the "<4>" prefixes
2913  * @line: buffer to copy the line to
2914  * @size: maximum size of the buffer
2915  * @len: length of line placed into buffer
2916  *
2917  * Start at the beginning of the kmsg buffer, with the oldest kmsg
2918  * record, and copy one record into the provided buffer.
2919  *
2920  * Consecutive calls will return the next available record moving
2921  * towards the end of the buffer with the youngest messages.
2922  *
2923  * A return value of FALSE indicates that there are no more records to
2924  * read.
2925  *
2926  * The function is similar to kmsg_dump_get_line(), but grabs no locks.
2927  */
2928 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
2929 			       char *line, size_t size, size_t *len)
2930 {
2931 	struct printk_log *msg;
2932 	size_t l = 0;
2933 	bool ret = false;
2934 
2935 	if (!dumper->active)
2936 		goto out;
2937 
2938 	if (dumper->cur_seq < log_first_seq) {
2939 		/* messages are gone, move to first available one */
2940 		dumper->cur_seq = log_first_seq;
2941 		dumper->cur_idx = log_first_idx;
2942 	}
2943 
2944 	/* last entry */
2945 	if (dumper->cur_seq >= log_next_seq)
2946 		goto out;
2947 
2948 	msg = log_from_idx(dumper->cur_idx);
2949 	l = msg_print_text(msg, 0, syslog, line, size);
2950 
2951 	dumper->cur_idx = log_next(dumper->cur_idx);
2952 	dumper->cur_seq++;
2953 	ret = true;
2954 out:
2955 	if (len)
2956 		*len = l;
2957 	return ret;
2958 }
2959 
2960 /**
2961  * kmsg_dump_get_line - retrieve one kmsg log line
2962  * @dumper: registered kmsg dumper
2963  * @syslog: include the "<4>" prefixes
2964  * @line: buffer to copy the line to
2965  * @size: maximum size of the buffer
2966  * @len: length of line placed into buffer
2967  *
2968  * Start at the beginning of the kmsg buffer, with the oldest kmsg
2969  * record, and copy one record into the provided buffer.
2970  *
2971  * Consecutive calls will return the next available record moving
2972  * towards the end of the buffer with the youngest messages.
2973  *
2974  * A return value of FALSE indicates that there are no more records to
2975  * read.
2976  */
2977 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
2978 			char *line, size_t size, size_t *len)
2979 {
2980 	unsigned long flags;
2981 	bool ret;
2982 
2983 	raw_spin_lock_irqsave(&logbuf_lock, flags);
2984 	ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
2985 	raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2986 
2987 	return ret;
2988 }
2989 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
2990 
2991 /**
2992  * kmsg_dump_get_buffer - copy kmsg log lines
2993  * @dumper: registered kmsg dumper
2994  * @syslog: include the "<4>" prefixes
2995  * @buf: buffer to copy the line to
2996  * @size: maximum size of the buffer
2997  * @len: length of line placed into buffer
2998  *
2999  * Start at the end of the kmsg buffer and fill the provided buffer
3000  * with as many of the the *youngest* kmsg records that fit into it.
3001  * If the buffer is large enough, all available kmsg records will be
3002  * copied with a single call.
3003  *
3004  * Consecutive calls will fill the buffer with the next block of
3005  * available older records, not including the earlier retrieved ones.
3006  *
3007  * A return value of FALSE indicates that there are no more records to
3008  * read.
3009  */
3010 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3011 			  char *buf, size_t size, size_t *len)
3012 {
3013 	unsigned long flags;
3014 	u64 seq;
3015 	u32 idx;
3016 	u64 next_seq;
3017 	u32 next_idx;
3018 	enum log_flags prev;
3019 	size_t l = 0;
3020 	bool ret = false;
3021 
3022 	if (!dumper->active)
3023 		goto out;
3024 
3025 	raw_spin_lock_irqsave(&logbuf_lock, flags);
3026 	if (dumper->cur_seq < log_first_seq) {
3027 		/* messages are gone, move to first available one */
3028 		dumper->cur_seq = log_first_seq;
3029 		dumper->cur_idx = log_first_idx;
3030 	}
3031 
3032 	/* last entry */
3033 	if (dumper->cur_seq >= dumper->next_seq) {
3034 		raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3035 		goto out;
3036 	}
3037 
3038 	/* calculate length of entire buffer */
3039 	seq = dumper->cur_seq;
3040 	idx = dumper->cur_idx;
3041 	prev = 0;
3042 	while (seq < dumper->next_seq) {
3043 		struct printk_log *msg = log_from_idx(idx);
3044 
3045 		l += msg_print_text(msg, prev, true, NULL, 0);
3046 		idx = log_next(idx);
3047 		seq++;
3048 		prev = msg->flags;
3049 	}
3050 
3051 	/* move first record forward until length fits into the buffer */
3052 	seq = dumper->cur_seq;
3053 	idx = dumper->cur_idx;
3054 	prev = 0;
3055 	while (l > size && seq < dumper->next_seq) {
3056 		struct printk_log *msg = log_from_idx(idx);
3057 
3058 		l -= msg_print_text(msg, prev, true, NULL, 0);
3059 		idx = log_next(idx);
3060 		seq++;
3061 		prev = msg->flags;
3062 	}
3063 
3064 	/* last message in next interation */
3065 	next_seq = seq;
3066 	next_idx = idx;
3067 
3068 	l = 0;
3069 	while (seq < dumper->next_seq) {
3070 		struct printk_log *msg = log_from_idx(idx);
3071 
3072 		l += msg_print_text(msg, prev, syslog, buf + l, size - l);
3073 		idx = log_next(idx);
3074 		seq++;
3075 		prev = msg->flags;
3076 	}
3077 
3078 	dumper->next_seq = next_seq;
3079 	dumper->next_idx = next_idx;
3080 	ret = true;
3081 	raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3082 out:
3083 	if (len)
3084 		*len = l;
3085 	return ret;
3086 }
3087 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3088 
3089 /**
3090  * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3091  * @dumper: registered kmsg dumper
3092  *
3093  * Reset the dumper's iterator so that kmsg_dump_get_line() and
3094  * kmsg_dump_get_buffer() can be called again and used multiple
3095  * times within the same dumper.dump() callback.
3096  *
3097  * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3098  */
3099 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3100 {
3101 	dumper->cur_seq = clear_seq;
3102 	dumper->cur_idx = clear_idx;
3103 	dumper->next_seq = log_next_seq;
3104 	dumper->next_idx = log_next_idx;
3105 }
3106 
3107 /**
3108  * kmsg_dump_rewind - reset the interator
3109  * @dumper: registered kmsg dumper
3110  *
3111  * Reset the dumper's iterator so that kmsg_dump_get_line() and
3112  * kmsg_dump_get_buffer() can be called again and used multiple
3113  * times within the same dumper.dump() callback.
3114  */
3115 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3116 {
3117 	unsigned long flags;
3118 
3119 	raw_spin_lock_irqsave(&logbuf_lock, flags);
3120 	kmsg_dump_rewind_nolock(dumper);
3121 	raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3122 }
3123 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3124 
3125 static char dump_stack_arch_desc_str[128];
3126 
3127 /**
3128  * dump_stack_set_arch_desc - set arch-specific str to show with task dumps
3129  * @fmt: printf-style format string
3130  * @...: arguments for the format string
3131  *
3132  * The configured string will be printed right after utsname during task
3133  * dumps.  Usually used to add arch-specific system identifiers.  If an
3134  * arch wants to make use of such an ID string, it should initialize this
3135  * as soon as possible during boot.
3136  */
3137 void __init dump_stack_set_arch_desc(const char *fmt, ...)
3138 {
3139 	va_list args;
3140 
3141 	va_start(args, fmt);
3142 	vsnprintf(dump_stack_arch_desc_str, sizeof(dump_stack_arch_desc_str),
3143 		  fmt, args);
3144 	va_end(args);
3145 }
3146 
3147 /**
3148  * dump_stack_print_info - print generic debug info for dump_stack()
3149  * @log_lvl: log level
3150  *
3151  * Arch-specific dump_stack() implementations can use this function to
3152  * print out the same debug information as the generic dump_stack().
3153  */
3154 void dump_stack_print_info(const char *log_lvl)
3155 {
3156 	printk("%sCPU: %d PID: %d Comm: %.20s %s %s %.*s\n",
3157 	       log_lvl, raw_smp_processor_id(), current->pid, current->comm,
3158 	       print_tainted(), init_utsname()->release,
3159 	       (int)strcspn(init_utsname()->version, " "),
3160 	       init_utsname()->version);
3161 
3162 	if (dump_stack_arch_desc_str[0] != '\0')
3163 		printk("%sHardware name: %s\n",
3164 		       log_lvl, dump_stack_arch_desc_str);
3165 
3166 	print_worker_info(log_lvl, current);
3167 }
3168 
3169 /**
3170  * show_regs_print_info - print generic debug info for show_regs()
3171  * @log_lvl: log level
3172  *
3173  * show_regs() implementations can use this function to print out generic
3174  * debug information.
3175  */
3176 void show_regs_print_info(const char *log_lvl)
3177 {
3178 	dump_stack_print_info(log_lvl);
3179 
3180 	printk("%stask: %p ti: %p task.ti: %p\n",
3181 	       log_lvl, current, current_thread_info(),
3182 	       task_thread_info(current));
3183 }
3184 
3185 #endif
3186