xref: /openbmc/linux/kernel/printk/printk.c (revision 901181b7)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/kernel/printk.c
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  *
7  * Modified to make sys_syslog() more flexible: added commands to
8  * return the last 4k of kernel messages, regardless of whether
9  * they've been read or not.  Added option to suppress kernel printk's
10  * to the console.  Added hook for sending the console messages
11  * elsewhere, in preparation for a serial line console (someday).
12  * Ted Ts'o, 2/11/93.
13  * Modified for sysctl support, 1/8/97, Chris Horn.
14  * Fixed SMP synchronization, 08/08/99, Manfred Spraul
15  *     manfred@colorfullife.com
16  * Rewrote bits to get rid of console_lock
17  *	01Mar01 Andrew Morton
18  */
19 
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 
22 #include <linux/kernel.h>
23 #include <linux/mm.h>
24 #include <linux/tty.h>
25 #include <linux/tty_driver.h>
26 #include <linux/console.h>
27 #include <linux/init.h>
28 #include <linux/jiffies.h>
29 #include <linux/nmi.h>
30 #include <linux/module.h>
31 #include <linux/moduleparam.h>
32 #include <linux/delay.h>
33 #include <linux/smp.h>
34 #include <linux/security.h>
35 #include <linux/memblock.h>
36 #include <linux/syscalls.h>
37 #include <linux/crash_core.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/rculist.h>
43 #include <linux/poll.h>
44 #include <linux/irq_work.h>
45 #include <linux/ctype.h>
46 #include <linux/uio.h>
47 #include <linux/sched/clock.h>
48 #include <linux/sched/debug.h>
49 #include <linux/sched/task_stack.h>
50 
51 #include <linux/uaccess.h>
52 #include <asm/sections.h>
53 
54 #include <trace/events/initcall.h>
55 #define CREATE_TRACE_POINTS
56 #include <trace/events/printk.h>
57 
58 #include "printk_ringbuffer.h"
59 #include "console_cmdline.h"
60 #include "braille.h"
61 #include "internal.h"
62 
63 int console_printk[4] = {
64 	CONSOLE_LOGLEVEL_DEFAULT,	/* console_loglevel */
65 	MESSAGE_LOGLEVEL_DEFAULT,	/* default_message_loglevel */
66 	CONSOLE_LOGLEVEL_MIN,		/* minimum_console_loglevel */
67 	CONSOLE_LOGLEVEL_DEFAULT,	/* default_console_loglevel */
68 };
69 EXPORT_SYMBOL_GPL(console_printk);
70 
71 atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
72 EXPORT_SYMBOL(ignore_console_lock_warning);
73 
74 /*
75  * Low level drivers may need that to know if they can schedule in
76  * their unblank() callback or not. So let's export it.
77  */
78 int oops_in_progress;
79 EXPORT_SYMBOL(oops_in_progress);
80 
81 /*
82  * console_sem protects the console_drivers list, and also
83  * provides serialisation for access to the entire console
84  * driver system.
85  */
86 static DEFINE_SEMAPHORE(console_sem);
87 struct console *console_drivers;
88 EXPORT_SYMBOL_GPL(console_drivers);
89 
90 /*
91  * System may need to suppress printk message under certain
92  * circumstances, like after kernel panic happens.
93  */
94 int __read_mostly suppress_printk;
95 
96 #ifdef CONFIG_LOCKDEP
97 static struct lockdep_map console_lock_dep_map = {
98 	.name = "console_lock"
99 };
100 #endif
101 
102 enum devkmsg_log_bits {
103 	__DEVKMSG_LOG_BIT_ON = 0,
104 	__DEVKMSG_LOG_BIT_OFF,
105 	__DEVKMSG_LOG_BIT_LOCK,
106 };
107 
108 enum devkmsg_log_masks {
109 	DEVKMSG_LOG_MASK_ON             = BIT(__DEVKMSG_LOG_BIT_ON),
110 	DEVKMSG_LOG_MASK_OFF            = BIT(__DEVKMSG_LOG_BIT_OFF),
111 	DEVKMSG_LOG_MASK_LOCK           = BIT(__DEVKMSG_LOG_BIT_LOCK),
112 };
113 
114 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
115 #define DEVKMSG_LOG_MASK_DEFAULT	0
116 
117 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
118 
119 static int __control_devkmsg(char *str)
120 {
121 	size_t len;
122 
123 	if (!str)
124 		return -EINVAL;
125 
126 	len = str_has_prefix(str, "on");
127 	if (len) {
128 		devkmsg_log = DEVKMSG_LOG_MASK_ON;
129 		return len;
130 	}
131 
132 	len = str_has_prefix(str, "off");
133 	if (len) {
134 		devkmsg_log = DEVKMSG_LOG_MASK_OFF;
135 		return len;
136 	}
137 
138 	len = str_has_prefix(str, "ratelimit");
139 	if (len) {
140 		devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
141 		return len;
142 	}
143 
144 	return -EINVAL;
145 }
146 
147 static int __init control_devkmsg(char *str)
148 {
149 	if (__control_devkmsg(str) < 0)
150 		return 1;
151 
152 	/*
153 	 * Set sysctl string accordingly:
154 	 */
155 	if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
156 		strcpy(devkmsg_log_str, "on");
157 	else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
158 		strcpy(devkmsg_log_str, "off");
159 	/* else "ratelimit" which is set by default. */
160 
161 	/*
162 	 * Sysctl cannot change it anymore. The kernel command line setting of
163 	 * this parameter is to force the setting to be permanent throughout the
164 	 * runtime of the system. This is a precation measure against userspace
165 	 * trying to be a smarta** and attempting to change it up on us.
166 	 */
167 	devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
168 
169 	return 0;
170 }
171 __setup("printk.devkmsg=", control_devkmsg);
172 
173 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
174 
175 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
176 			      void *buffer, size_t *lenp, loff_t *ppos)
177 {
178 	char old_str[DEVKMSG_STR_MAX_SIZE];
179 	unsigned int old;
180 	int err;
181 
182 	if (write) {
183 		if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
184 			return -EINVAL;
185 
186 		old = devkmsg_log;
187 		strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
188 	}
189 
190 	err = proc_dostring(table, write, buffer, lenp, ppos);
191 	if (err)
192 		return err;
193 
194 	if (write) {
195 		err = __control_devkmsg(devkmsg_log_str);
196 
197 		/*
198 		 * Do not accept an unknown string OR a known string with
199 		 * trailing crap...
200 		 */
201 		if (err < 0 || (err + 1 != *lenp)) {
202 
203 			/* ... and restore old setting. */
204 			devkmsg_log = old;
205 			strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
206 
207 			return -EINVAL;
208 		}
209 	}
210 
211 	return 0;
212 }
213 
214 /* Number of registered extended console drivers. */
215 static int nr_ext_console_drivers;
216 
217 /*
218  * Helper macros to handle lockdep when locking/unlocking console_sem. We use
219  * macros instead of functions so that _RET_IP_ contains useful information.
220  */
221 #define down_console_sem() do { \
222 	down(&console_sem);\
223 	mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
224 } while (0)
225 
226 static int __down_trylock_console_sem(unsigned long ip)
227 {
228 	int lock_failed;
229 	unsigned long flags;
230 
231 	/*
232 	 * Here and in __up_console_sem() we need to be in safe mode,
233 	 * because spindump/WARN/etc from under console ->lock will
234 	 * deadlock in printk()->down_trylock_console_sem() otherwise.
235 	 */
236 	printk_safe_enter_irqsave(flags);
237 	lock_failed = down_trylock(&console_sem);
238 	printk_safe_exit_irqrestore(flags);
239 
240 	if (lock_failed)
241 		return 1;
242 	mutex_acquire(&console_lock_dep_map, 0, 1, ip);
243 	return 0;
244 }
245 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
246 
247 static void __up_console_sem(unsigned long ip)
248 {
249 	unsigned long flags;
250 
251 	mutex_release(&console_lock_dep_map, ip);
252 
253 	printk_safe_enter_irqsave(flags);
254 	up(&console_sem);
255 	printk_safe_exit_irqrestore(flags);
256 }
257 #define up_console_sem() __up_console_sem(_RET_IP_)
258 
259 /*
260  * This is used for debugging the mess that is the VT code by
261  * keeping track if we have the console semaphore held. It's
262  * definitely not the perfect debug tool (we don't know if _WE_
263  * hold it and are racing, but it helps tracking those weird code
264  * paths in the console code where we end up in places I want
265  * locked without the console semaphore held).
266  */
267 static int console_locked, console_suspended;
268 
269 /*
270  * If exclusive_console is non-NULL then only this console is to be printed to.
271  */
272 static struct console *exclusive_console;
273 
274 /*
275  *	Array of consoles built from command line options (console=)
276  */
277 
278 #define MAX_CMDLINECONSOLES 8
279 
280 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
281 
282 static int preferred_console = -1;
283 static bool has_preferred_console;
284 int console_set_on_cmdline;
285 EXPORT_SYMBOL(console_set_on_cmdline);
286 
287 /* Flag: console code may call schedule() */
288 static int console_may_schedule;
289 
290 enum con_msg_format_flags {
291 	MSG_FORMAT_DEFAULT	= 0,
292 	MSG_FORMAT_SYSLOG	= (1 << 0),
293 };
294 
295 static int console_msg_format = MSG_FORMAT_DEFAULT;
296 
297 /*
298  * The printk log buffer consists of a sequenced collection of records, each
299  * containing variable length message text. Every record also contains its
300  * own meta-data (@info).
301  *
302  * Every record meta-data carries the timestamp in microseconds, as well as
303  * the standard userspace syslog level and syslog facility. The usual kernel
304  * messages use LOG_KERN; userspace-injected messages always carry a matching
305  * syslog facility, by default LOG_USER. The origin of every message can be
306  * reliably determined that way.
307  *
308  * The human readable log message of a record is available in @text, the
309  * length of the message text in @text_len. The stored message is not
310  * terminated.
311  *
312  * Optionally, a record can carry a dictionary of properties (key/value
313  * pairs), to provide userspace with a machine-readable message context.
314  *
315  * Examples for well-defined, commonly used property names are:
316  *   DEVICE=b12:8               device identifier
317  *                                b12:8         block dev_t
318  *                                c127:3        char dev_t
319  *                                n8            netdev ifindex
320  *                                +sound:card0  subsystem:devname
321  *   SUBSYSTEM=pci              driver-core subsystem name
322  *
323  * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
324  * and values are terminated by a '\0' character.
325  *
326  * Example of record values:
327  *   record.text_buf                = "it's a line" (unterminated)
328  *   record.info.seq                = 56
329  *   record.info.ts_nsec            = 36863
330  *   record.info.text_len           = 11
331  *   record.info.facility           = 0 (LOG_KERN)
332  *   record.info.flags              = 0
333  *   record.info.level              = 3 (LOG_ERR)
334  *   record.info.caller_id          = 299 (task 299)
335  *   record.info.dev_info.subsystem = "pci" (terminated)
336  *   record.info.dev_info.device    = "+pci:0000:00:01.0" (terminated)
337  *
338  * The 'struct printk_info' buffer must never be directly exported to
339  * userspace, it is a kernel-private implementation detail that might
340  * need to be changed in the future, when the requirements change.
341  *
342  * /dev/kmsg exports the structured data in the following line format:
343  *   "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
344  *
345  * Users of the export format should ignore possible additional values
346  * separated by ',', and find the message after the ';' character.
347  *
348  * The optional key/value pairs are attached as continuation lines starting
349  * with a space character and terminated by a newline. All possible
350  * non-prinatable characters are escaped in the "\xff" notation.
351  */
352 
353 /* syslog_lock protects syslog_* variables and write access to clear_seq. */
354 static DEFINE_MUTEX(syslog_lock);
355 
356 #ifdef CONFIG_PRINTK
357 DECLARE_WAIT_QUEUE_HEAD(log_wait);
358 /* All 3 protected by @syslog_lock. */
359 /* the next printk record to read by syslog(READ) or /proc/kmsg */
360 static u64 syslog_seq;
361 static size_t syslog_partial;
362 static bool syslog_time;
363 
364 /* All 3 protected by @console_sem. */
365 /* the next printk record to write to the console */
366 static u64 console_seq;
367 static u64 exclusive_console_stop_seq;
368 static unsigned long console_dropped;
369 
370 struct latched_seq {
371 	seqcount_latch_t	latch;
372 	u64			val[2];
373 };
374 
375 /*
376  * The next printk record to read after the last 'clear' command. There are
377  * two copies (updated with seqcount_latch) so that reads can locklessly
378  * access a valid value. Writers are synchronized by @syslog_lock.
379  */
380 static struct latched_seq clear_seq = {
381 	.latch		= SEQCNT_LATCH_ZERO(clear_seq.latch),
382 	.val[0]		= 0,
383 	.val[1]		= 0,
384 };
385 
386 #ifdef CONFIG_PRINTK_CALLER
387 #define PREFIX_MAX		48
388 #else
389 #define PREFIX_MAX		32
390 #endif
391 
392 /* the maximum size of a formatted record (i.e. with prefix added per line) */
393 #define CONSOLE_LOG_MAX		1024
394 
395 /* the maximum size allowed to be reserved for a record */
396 #define LOG_LINE_MAX		(CONSOLE_LOG_MAX - PREFIX_MAX)
397 
398 #define LOG_LEVEL(v)		((v) & 0x07)
399 #define LOG_FACILITY(v)		((v) >> 3 & 0xff)
400 
401 /* record buffer */
402 #define LOG_ALIGN __alignof__(unsigned long)
403 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
404 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
405 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
406 static char *log_buf = __log_buf;
407 static u32 log_buf_len = __LOG_BUF_LEN;
408 
409 /*
410  * Define the average message size. This only affects the number of
411  * descriptors that will be available. Underestimating is better than
412  * overestimating (too many available descriptors is better than not enough).
413  */
414 #define PRB_AVGBITS 5	/* 32 character average length */
415 
416 #if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
417 #error CONFIG_LOG_BUF_SHIFT value too small.
418 #endif
419 _DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
420 		 PRB_AVGBITS, &__log_buf[0]);
421 
422 static struct printk_ringbuffer printk_rb_dynamic;
423 
424 static struct printk_ringbuffer *prb = &printk_rb_static;
425 
426 /*
427  * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
428  * per_cpu_areas are initialised. This variable is set to true when
429  * it's safe to access per-CPU data.
430  */
431 static bool __printk_percpu_data_ready __read_mostly;
432 
433 bool printk_percpu_data_ready(void)
434 {
435 	return __printk_percpu_data_ready;
436 }
437 
438 /* Must be called under syslog_lock. */
439 static void latched_seq_write(struct latched_seq *ls, u64 val)
440 {
441 	raw_write_seqcount_latch(&ls->latch);
442 	ls->val[0] = val;
443 	raw_write_seqcount_latch(&ls->latch);
444 	ls->val[1] = val;
445 }
446 
447 /* Can be called from any context. */
448 static u64 latched_seq_read_nolock(struct latched_seq *ls)
449 {
450 	unsigned int seq;
451 	unsigned int idx;
452 	u64 val;
453 
454 	do {
455 		seq = raw_read_seqcount_latch(&ls->latch);
456 		idx = seq & 0x1;
457 		val = ls->val[idx];
458 	} while (read_seqcount_latch_retry(&ls->latch, seq));
459 
460 	return val;
461 }
462 
463 /* Return log buffer address */
464 char *log_buf_addr_get(void)
465 {
466 	return log_buf;
467 }
468 
469 /* Return log buffer size */
470 u32 log_buf_len_get(void)
471 {
472 	return log_buf_len;
473 }
474 
475 /*
476  * Define how much of the log buffer we could take at maximum. The value
477  * must be greater than two. Note that only half of the buffer is available
478  * when the index points to the middle.
479  */
480 #define MAX_LOG_TAKE_PART 4
481 static const char trunc_msg[] = "<truncated>";
482 
483 static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
484 {
485 	/*
486 	 * The message should not take the whole buffer. Otherwise, it might
487 	 * get removed too soon.
488 	 */
489 	u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
490 
491 	if (*text_len > max_text_len)
492 		*text_len = max_text_len;
493 
494 	/* enable the warning message (if there is room) */
495 	*trunc_msg_len = strlen(trunc_msg);
496 	if (*text_len >= *trunc_msg_len)
497 		*text_len -= *trunc_msg_len;
498 	else
499 		*trunc_msg_len = 0;
500 }
501 
502 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
503 
504 static int syslog_action_restricted(int type)
505 {
506 	if (dmesg_restrict)
507 		return 1;
508 	/*
509 	 * Unless restricted, we allow "read all" and "get buffer size"
510 	 * for everybody.
511 	 */
512 	return type != SYSLOG_ACTION_READ_ALL &&
513 	       type != SYSLOG_ACTION_SIZE_BUFFER;
514 }
515 
516 static int check_syslog_permissions(int type, int source)
517 {
518 	/*
519 	 * If this is from /proc/kmsg and we've already opened it, then we've
520 	 * already done the capabilities checks at open time.
521 	 */
522 	if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
523 		goto ok;
524 
525 	if (syslog_action_restricted(type)) {
526 		if (capable(CAP_SYSLOG))
527 			goto ok;
528 		/*
529 		 * For historical reasons, accept CAP_SYS_ADMIN too, with
530 		 * a warning.
531 		 */
532 		if (capable(CAP_SYS_ADMIN)) {
533 			pr_warn_once("%s (%d): Attempt to access syslog with "
534 				     "CAP_SYS_ADMIN but no CAP_SYSLOG "
535 				     "(deprecated).\n",
536 				 current->comm, task_pid_nr(current));
537 			goto ok;
538 		}
539 		return -EPERM;
540 	}
541 ok:
542 	return security_syslog(type);
543 }
544 
545 static void append_char(char **pp, char *e, char c)
546 {
547 	if (*pp < e)
548 		*(*pp)++ = c;
549 }
550 
551 static ssize_t info_print_ext_header(char *buf, size_t size,
552 				     struct printk_info *info)
553 {
554 	u64 ts_usec = info->ts_nsec;
555 	char caller[20];
556 #ifdef CONFIG_PRINTK_CALLER
557 	u32 id = info->caller_id;
558 
559 	snprintf(caller, sizeof(caller), ",caller=%c%u",
560 		 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
561 #else
562 	caller[0] = '\0';
563 #endif
564 
565 	do_div(ts_usec, 1000);
566 
567 	return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
568 			 (info->facility << 3) | info->level, info->seq,
569 			 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
570 }
571 
572 static ssize_t msg_add_ext_text(char *buf, size_t size,
573 				const char *text, size_t text_len,
574 				unsigned char endc)
575 {
576 	char *p = buf, *e = buf + size;
577 	size_t i;
578 
579 	/* escape non-printable characters */
580 	for (i = 0; i < text_len; i++) {
581 		unsigned char c = text[i];
582 
583 		if (c < ' ' || c >= 127 || c == '\\')
584 			p += scnprintf(p, e - p, "\\x%02x", c);
585 		else
586 			append_char(&p, e, c);
587 	}
588 	append_char(&p, e, endc);
589 
590 	return p - buf;
591 }
592 
593 static ssize_t msg_add_dict_text(char *buf, size_t size,
594 				 const char *key, const char *val)
595 {
596 	size_t val_len = strlen(val);
597 	ssize_t len;
598 
599 	if (!val_len)
600 		return 0;
601 
602 	len = msg_add_ext_text(buf, size, "", 0, ' ');	/* dict prefix */
603 	len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
604 	len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
605 
606 	return len;
607 }
608 
609 static ssize_t msg_print_ext_body(char *buf, size_t size,
610 				  char *text, size_t text_len,
611 				  struct dev_printk_info *dev_info)
612 {
613 	ssize_t len;
614 
615 	len = msg_add_ext_text(buf, size, text, text_len, '\n');
616 
617 	if (!dev_info)
618 		goto out;
619 
620 	len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
621 				 dev_info->subsystem);
622 	len += msg_add_dict_text(buf + len, size - len, "DEVICE",
623 				 dev_info->device);
624 out:
625 	return len;
626 }
627 
628 /* /dev/kmsg - userspace message inject/listen interface */
629 struct devkmsg_user {
630 	atomic64_t seq;
631 	struct ratelimit_state rs;
632 	struct mutex lock;
633 	char buf[CONSOLE_EXT_LOG_MAX];
634 
635 	struct printk_info info;
636 	char text_buf[CONSOLE_EXT_LOG_MAX];
637 	struct printk_record record;
638 };
639 
640 static __printf(3, 4) __cold
641 int devkmsg_emit(int facility, int level, const char *fmt, ...)
642 {
643 	va_list args;
644 	int r;
645 
646 	va_start(args, fmt);
647 	r = vprintk_emit(facility, level, NULL, fmt, args);
648 	va_end(args);
649 
650 	return r;
651 }
652 
653 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
654 {
655 	char *buf, *line;
656 	int level = default_message_loglevel;
657 	int facility = 1;	/* LOG_USER */
658 	struct file *file = iocb->ki_filp;
659 	struct devkmsg_user *user = file->private_data;
660 	size_t len = iov_iter_count(from);
661 	ssize_t ret = len;
662 
663 	if (!user || len > LOG_LINE_MAX)
664 		return -EINVAL;
665 
666 	/* Ignore when user logging is disabled. */
667 	if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
668 		return len;
669 
670 	/* Ratelimit when not explicitly enabled. */
671 	if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
672 		if (!___ratelimit(&user->rs, current->comm))
673 			return ret;
674 	}
675 
676 	buf = kmalloc(len+1, GFP_KERNEL);
677 	if (buf == NULL)
678 		return -ENOMEM;
679 
680 	buf[len] = '\0';
681 	if (!copy_from_iter_full(buf, len, from)) {
682 		kfree(buf);
683 		return -EFAULT;
684 	}
685 
686 	/*
687 	 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
688 	 * the decimal value represents 32bit, the lower 3 bit are the log
689 	 * level, the rest are the log facility.
690 	 *
691 	 * If no prefix or no userspace facility is specified, we
692 	 * enforce LOG_USER, to be able to reliably distinguish
693 	 * kernel-generated messages from userspace-injected ones.
694 	 */
695 	line = buf;
696 	if (line[0] == '<') {
697 		char *endp = NULL;
698 		unsigned int u;
699 
700 		u = simple_strtoul(line + 1, &endp, 10);
701 		if (endp && endp[0] == '>') {
702 			level = LOG_LEVEL(u);
703 			if (LOG_FACILITY(u) != 0)
704 				facility = LOG_FACILITY(u);
705 			endp++;
706 			line = endp;
707 		}
708 	}
709 
710 	devkmsg_emit(facility, level, "%s", line);
711 	kfree(buf);
712 	return ret;
713 }
714 
715 static ssize_t devkmsg_read(struct file *file, char __user *buf,
716 			    size_t count, loff_t *ppos)
717 {
718 	struct devkmsg_user *user = file->private_data;
719 	struct printk_record *r = &user->record;
720 	size_t len;
721 	ssize_t ret;
722 
723 	if (!user)
724 		return -EBADF;
725 
726 	ret = mutex_lock_interruptible(&user->lock);
727 	if (ret)
728 		return ret;
729 
730 	if (!prb_read_valid(prb, atomic64_read(&user->seq), r)) {
731 		if (file->f_flags & O_NONBLOCK) {
732 			ret = -EAGAIN;
733 			goto out;
734 		}
735 
736 		ret = wait_event_interruptible(log_wait,
737 				prb_read_valid(prb, atomic64_read(&user->seq), r));
738 		if (ret)
739 			goto out;
740 	}
741 
742 	if (r->info->seq != atomic64_read(&user->seq)) {
743 		/* our last seen message is gone, return error and reset */
744 		atomic64_set(&user->seq, r->info->seq);
745 		ret = -EPIPE;
746 		goto out;
747 	}
748 
749 	len = info_print_ext_header(user->buf, sizeof(user->buf), r->info);
750 	len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
751 				  &r->text_buf[0], r->info->text_len,
752 				  &r->info->dev_info);
753 
754 	atomic64_set(&user->seq, r->info->seq + 1);
755 
756 	if (len > count) {
757 		ret = -EINVAL;
758 		goto out;
759 	}
760 
761 	if (copy_to_user(buf, user->buf, len)) {
762 		ret = -EFAULT;
763 		goto out;
764 	}
765 	ret = len;
766 out:
767 	mutex_unlock(&user->lock);
768 	return ret;
769 }
770 
771 /*
772  * Be careful when modifying this function!!!
773  *
774  * Only few operations are supported because the device works only with the
775  * entire variable length messages (records). Non-standard values are
776  * returned in the other cases and has been this way for quite some time.
777  * User space applications might depend on this behavior.
778  */
779 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
780 {
781 	struct devkmsg_user *user = file->private_data;
782 	loff_t ret = 0;
783 
784 	if (!user)
785 		return -EBADF;
786 	if (offset)
787 		return -ESPIPE;
788 
789 	switch (whence) {
790 	case SEEK_SET:
791 		/* the first record */
792 		atomic64_set(&user->seq, prb_first_valid_seq(prb));
793 		break;
794 	case SEEK_DATA:
795 		/*
796 		 * The first record after the last SYSLOG_ACTION_CLEAR,
797 		 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
798 		 * changes no global state, and does not clear anything.
799 		 */
800 		atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
801 		break;
802 	case SEEK_END:
803 		/* after the last record */
804 		atomic64_set(&user->seq, prb_next_seq(prb));
805 		break;
806 	default:
807 		ret = -EINVAL;
808 	}
809 	return ret;
810 }
811 
812 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
813 {
814 	struct devkmsg_user *user = file->private_data;
815 	struct printk_info info;
816 	__poll_t ret = 0;
817 
818 	if (!user)
819 		return EPOLLERR|EPOLLNVAL;
820 
821 	poll_wait(file, &log_wait, wait);
822 
823 	if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
824 		/* return error when data has vanished underneath us */
825 		if (info.seq != atomic64_read(&user->seq))
826 			ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
827 		else
828 			ret = EPOLLIN|EPOLLRDNORM;
829 	}
830 
831 	return ret;
832 }
833 
834 static int devkmsg_open(struct inode *inode, struct file *file)
835 {
836 	struct devkmsg_user *user;
837 	int err;
838 
839 	if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
840 		return -EPERM;
841 
842 	/* write-only does not need any file context */
843 	if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
844 		err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
845 					       SYSLOG_FROM_READER);
846 		if (err)
847 			return err;
848 	}
849 
850 	user = kvmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
851 	if (!user)
852 		return -ENOMEM;
853 
854 	ratelimit_default_init(&user->rs);
855 	ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
856 
857 	mutex_init(&user->lock);
858 
859 	prb_rec_init_rd(&user->record, &user->info,
860 			&user->text_buf[0], sizeof(user->text_buf));
861 
862 	atomic64_set(&user->seq, prb_first_valid_seq(prb));
863 
864 	file->private_data = user;
865 	return 0;
866 }
867 
868 static int devkmsg_release(struct inode *inode, struct file *file)
869 {
870 	struct devkmsg_user *user = file->private_data;
871 
872 	if (!user)
873 		return 0;
874 
875 	ratelimit_state_exit(&user->rs);
876 
877 	mutex_destroy(&user->lock);
878 	kvfree(user);
879 	return 0;
880 }
881 
882 const struct file_operations kmsg_fops = {
883 	.open = devkmsg_open,
884 	.read = devkmsg_read,
885 	.write_iter = devkmsg_write,
886 	.llseek = devkmsg_llseek,
887 	.poll = devkmsg_poll,
888 	.release = devkmsg_release,
889 };
890 
891 #ifdef CONFIG_CRASH_CORE
892 /*
893  * This appends the listed symbols to /proc/vmcore
894  *
895  * /proc/vmcore is used by various utilities, like crash and makedumpfile to
896  * obtain access to symbols that are otherwise very difficult to locate.  These
897  * symbols are specifically used so that utilities can access and extract the
898  * dmesg log from a vmcore file after a crash.
899  */
900 void log_buf_vmcoreinfo_setup(void)
901 {
902 	struct dev_printk_info *dev_info = NULL;
903 
904 	VMCOREINFO_SYMBOL(prb);
905 	VMCOREINFO_SYMBOL(printk_rb_static);
906 	VMCOREINFO_SYMBOL(clear_seq);
907 
908 	/*
909 	 * Export struct size and field offsets. User space tools can
910 	 * parse it and detect any changes to structure down the line.
911 	 */
912 
913 	VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
914 	VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
915 	VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
916 	VMCOREINFO_OFFSET(printk_ringbuffer, fail);
917 
918 	VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
919 	VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
920 	VMCOREINFO_OFFSET(prb_desc_ring, descs);
921 	VMCOREINFO_OFFSET(prb_desc_ring, infos);
922 	VMCOREINFO_OFFSET(prb_desc_ring, head_id);
923 	VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
924 
925 	VMCOREINFO_STRUCT_SIZE(prb_desc);
926 	VMCOREINFO_OFFSET(prb_desc, state_var);
927 	VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
928 
929 	VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
930 	VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
931 	VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
932 
933 	VMCOREINFO_STRUCT_SIZE(printk_info);
934 	VMCOREINFO_OFFSET(printk_info, seq);
935 	VMCOREINFO_OFFSET(printk_info, ts_nsec);
936 	VMCOREINFO_OFFSET(printk_info, text_len);
937 	VMCOREINFO_OFFSET(printk_info, caller_id);
938 	VMCOREINFO_OFFSET(printk_info, dev_info);
939 
940 	VMCOREINFO_STRUCT_SIZE(dev_printk_info);
941 	VMCOREINFO_OFFSET(dev_printk_info, subsystem);
942 	VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
943 	VMCOREINFO_OFFSET(dev_printk_info, device);
944 	VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
945 
946 	VMCOREINFO_STRUCT_SIZE(prb_data_ring);
947 	VMCOREINFO_OFFSET(prb_data_ring, size_bits);
948 	VMCOREINFO_OFFSET(prb_data_ring, data);
949 	VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
950 	VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
951 
952 	VMCOREINFO_SIZE(atomic_long_t);
953 	VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
954 
955 	VMCOREINFO_STRUCT_SIZE(latched_seq);
956 	VMCOREINFO_OFFSET(latched_seq, val);
957 }
958 #endif
959 
960 /* requested log_buf_len from kernel cmdline */
961 static unsigned long __initdata new_log_buf_len;
962 
963 /* we practice scaling the ring buffer by powers of 2 */
964 static void __init log_buf_len_update(u64 size)
965 {
966 	if (size > (u64)LOG_BUF_LEN_MAX) {
967 		size = (u64)LOG_BUF_LEN_MAX;
968 		pr_err("log_buf over 2G is not supported.\n");
969 	}
970 
971 	if (size)
972 		size = roundup_pow_of_two(size);
973 	if (size > log_buf_len)
974 		new_log_buf_len = (unsigned long)size;
975 }
976 
977 /* save requested log_buf_len since it's too early to process it */
978 static int __init log_buf_len_setup(char *str)
979 {
980 	u64 size;
981 
982 	if (!str)
983 		return -EINVAL;
984 
985 	size = memparse(str, &str);
986 
987 	log_buf_len_update(size);
988 
989 	return 0;
990 }
991 early_param("log_buf_len", log_buf_len_setup);
992 
993 #ifdef CONFIG_SMP
994 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
995 
996 static void __init log_buf_add_cpu(void)
997 {
998 	unsigned int cpu_extra;
999 
1000 	/*
1001 	 * archs should set up cpu_possible_bits properly with
1002 	 * set_cpu_possible() after setup_arch() but just in
1003 	 * case lets ensure this is valid.
1004 	 */
1005 	if (num_possible_cpus() == 1)
1006 		return;
1007 
1008 	cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1009 
1010 	/* by default this will only continue through for large > 64 CPUs */
1011 	if (cpu_extra <= __LOG_BUF_LEN / 2)
1012 		return;
1013 
1014 	pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1015 		__LOG_CPU_MAX_BUF_LEN);
1016 	pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1017 		cpu_extra);
1018 	pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1019 
1020 	log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1021 }
1022 #else /* !CONFIG_SMP */
1023 static inline void log_buf_add_cpu(void) {}
1024 #endif /* CONFIG_SMP */
1025 
1026 static void __init set_percpu_data_ready(void)
1027 {
1028 	__printk_percpu_data_ready = true;
1029 }
1030 
1031 static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1032 				     struct printk_record *r)
1033 {
1034 	struct prb_reserved_entry e;
1035 	struct printk_record dest_r;
1036 
1037 	prb_rec_init_wr(&dest_r, r->info->text_len);
1038 
1039 	if (!prb_reserve(&e, rb, &dest_r))
1040 		return 0;
1041 
1042 	memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1043 	dest_r.info->text_len = r->info->text_len;
1044 	dest_r.info->facility = r->info->facility;
1045 	dest_r.info->level = r->info->level;
1046 	dest_r.info->flags = r->info->flags;
1047 	dest_r.info->ts_nsec = r->info->ts_nsec;
1048 	dest_r.info->caller_id = r->info->caller_id;
1049 	memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1050 
1051 	prb_final_commit(&e);
1052 
1053 	return prb_record_text_space(&e);
1054 }
1055 
1056 static char setup_text_buf[LOG_LINE_MAX] __initdata;
1057 
1058 void __init setup_log_buf(int early)
1059 {
1060 	struct printk_info *new_infos;
1061 	unsigned int new_descs_count;
1062 	struct prb_desc *new_descs;
1063 	struct printk_info info;
1064 	struct printk_record r;
1065 	unsigned int text_size;
1066 	size_t new_descs_size;
1067 	size_t new_infos_size;
1068 	unsigned long flags;
1069 	char *new_log_buf;
1070 	unsigned int free;
1071 	u64 seq;
1072 
1073 	/*
1074 	 * Some archs call setup_log_buf() multiple times - first is very
1075 	 * early, e.g. from setup_arch(), and second - when percpu_areas
1076 	 * are initialised.
1077 	 */
1078 	if (!early)
1079 		set_percpu_data_ready();
1080 
1081 	if (log_buf != __log_buf)
1082 		return;
1083 
1084 	if (!early && !new_log_buf_len)
1085 		log_buf_add_cpu();
1086 
1087 	if (!new_log_buf_len)
1088 		return;
1089 
1090 	new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1091 	if (new_descs_count == 0) {
1092 		pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1093 		return;
1094 	}
1095 
1096 	new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1097 	if (unlikely(!new_log_buf)) {
1098 		pr_err("log_buf_len: %lu text bytes not available\n",
1099 		       new_log_buf_len);
1100 		return;
1101 	}
1102 
1103 	new_descs_size = new_descs_count * sizeof(struct prb_desc);
1104 	new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1105 	if (unlikely(!new_descs)) {
1106 		pr_err("log_buf_len: %zu desc bytes not available\n",
1107 		       new_descs_size);
1108 		goto err_free_log_buf;
1109 	}
1110 
1111 	new_infos_size = new_descs_count * sizeof(struct printk_info);
1112 	new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1113 	if (unlikely(!new_infos)) {
1114 		pr_err("log_buf_len: %zu info bytes not available\n",
1115 		       new_infos_size);
1116 		goto err_free_descs;
1117 	}
1118 
1119 	prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1120 
1121 	prb_init(&printk_rb_dynamic,
1122 		 new_log_buf, ilog2(new_log_buf_len),
1123 		 new_descs, ilog2(new_descs_count),
1124 		 new_infos);
1125 
1126 	local_irq_save(flags);
1127 
1128 	log_buf_len = new_log_buf_len;
1129 	log_buf = new_log_buf;
1130 	new_log_buf_len = 0;
1131 
1132 	free = __LOG_BUF_LEN;
1133 	prb_for_each_record(0, &printk_rb_static, seq, &r) {
1134 		text_size = add_to_rb(&printk_rb_dynamic, &r);
1135 		if (text_size > free)
1136 			free = 0;
1137 		else
1138 			free -= text_size;
1139 	}
1140 
1141 	prb = &printk_rb_dynamic;
1142 
1143 	local_irq_restore(flags);
1144 
1145 	/*
1146 	 * Copy any remaining messages that might have appeared from
1147 	 * NMI context after copying but before switching to the
1148 	 * dynamic buffer.
1149 	 */
1150 	prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1151 		text_size = add_to_rb(&printk_rb_dynamic, &r);
1152 		if (text_size > free)
1153 			free = 0;
1154 		else
1155 			free -= text_size;
1156 	}
1157 
1158 	if (seq != prb_next_seq(&printk_rb_static)) {
1159 		pr_err("dropped %llu messages\n",
1160 		       prb_next_seq(&printk_rb_static) - seq);
1161 	}
1162 
1163 	pr_info("log_buf_len: %u bytes\n", log_buf_len);
1164 	pr_info("early log buf free: %u(%u%%)\n",
1165 		free, (free * 100) / __LOG_BUF_LEN);
1166 	return;
1167 
1168 err_free_descs:
1169 	memblock_free(new_descs, new_descs_size);
1170 err_free_log_buf:
1171 	memblock_free(new_log_buf, new_log_buf_len);
1172 }
1173 
1174 static bool __read_mostly ignore_loglevel;
1175 
1176 static int __init ignore_loglevel_setup(char *str)
1177 {
1178 	ignore_loglevel = true;
1179 	pr_info("debug: ignoring loglevel setting.\n");
1180 
1181 	return 0;
1182 }
1183 
1184 early_param("ignore_loglevel", ignore_loglevel_setup);
1185 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1186 MODULE_PARM_DESC(ignore_loglevel,
1187 		 "ignore loglevel setting (prints all kernel messages to the console)");
1188 
1189 static bool suppress_message_printing(int level)
1190 {
1191 	return (level >= console_loglevel && !ignore_loglevel);
1192 }
1193 
1194 #ifdef CONFIG_BOOT_PRINTK_DELAY
1195 
1196 static int boot_delay; /* msecs delay after each printk during bootup */
1197 static unsigned long long loops_per_msec;	/* based on boot_delay */
1198 
1199 static int __init boot_delay_setup(char *str)
1200 {
1201 	unsigned long lpj;
1202 
1203 	lpj = preset_lpj ? preset_lpj : 1000000;	/* some guess */
1204 	loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1205 
1206 	get_option(&str, &boot_delay);
1207 	if (boot_delay > 10 * 1000)
1208 		boot_delay = 0;
1209 
1210 	pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1211 		"HZ: %d, loops_per_msec: %llu\n",
1212 		boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1213 	return 0;
1214 }
1215 early_param("boot_delay", boot_delay_setup);
1216 
1217 static void boot_delay_msec(int level)
1218 {
1219 	unsigned long long k;
1220 	unsigned long timeout;
1221 
1222 	if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1223 		|| suppress_message_printing(level)) {
1224 		return;
1225 	}
1226 
1227 	k = (unsigned long long)loops_per_msec * boot_delay;
1228 
1229 	timeout = jiffies + msecs_to_jiffies(boot_delay);
1230 	while (k) {
1231 		k--;
1232 		cpu_relax();
1233 		/*
1234 		 * use (volatile) jiffies to prevent
1235 		 * compiler reduction; loop termination via jiffies
1236 		 * is secondary and may or may not happen.
1237 		 */
1238 		if (time_after(jiffies, timeout))
1239 			break;
1240 		touch_nmi_watchdog();
1241 	}
1242 }
1243 #else
1244 static inline void boot_delay_msec(int level)
1245 {
1246 }
1247 #endif
1248 
1249 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1250 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1251 
1252 static size_t print_syslog(unsigned int level, char *buf)
1253 {
1254 	return sprintf(buf, "<%u>", level);
1255 }
1256 
1257 static size_t print_time(u64 ts, char *buf)
1258 {
1259 	unsigned long rem_nsec = do_div(ts, 1000000000);
1260 
1261 	return sprintf(buf, "[%5lu.%06lu]",
1262 		       (unsigned long)ts, rem_nsec / 1000);
1263 }
1264 
1265 #ifdef CONFIG_PRINTK_CALLER
1266 static size_t print_caller(u32 id, char *buf)
1267 {
1268 	char caller[12];
1269 
1270 	snprintf(caller, sizeof(caller), "%c%u",
1271 		 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1272 	return sprintf(buf, "[%6s]", caller);
1273 }
1274 #else
1275 #define print_caller(id, buf) 0
1276 #endif
1277 
1278 static size_t info_print_prefix(const struct printk_info  *info, bool syslog,
1279 				bool time, char *buf)
1280 {
1281 	size_t len = 0;
1282 
1283 	if (syslog)
1284 		len = print_syslog((info->facility << 3) | info->level, buf);
1285 
1286 	if (time)
1287 		len += print_time(info->ts_nsec, buf + len);
1288 
1289 	len += print_caller(info->caller_id, buf + len);
1290 
1291 	if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1292 		buf[len++] = ' ';
1293 		buf[len] = '\0';
1294 	}
1295 
1296 	return len;
1297 }
1298 
1299 /*
1300  * Prepare the record for printing. The text is shifted within the given
1301  * buffer to avoid a need for another one. The following operations are
1302  * done:
1303  *
1304  *   - Add prefix for each line.
1305  *   - Drop truncated lines that no longer fit into the buffer.
1306  *   - Add the trailing newline that has been removed in vprintk_store().
1307  *   - Add a string terminator.
1308  *
1309  * Since the produced string is always terminated, the maximum possible
1310  * return value is @r->text_buf_size - 1;
1311  *
1312  * Return: The length of the updated/prepared text, including the added
1313  * prefixes and the newline. The terminator is not counted. The dropped
1314  * line(s) are not counted.
1315  */
1316 static size_t record_print_text(struct printk_record *r, bool syslog,
1317 				bool time)
1318 {
1319 	size_t text_len = r->info->text_len;
1320 	size_t buf_size = r->text_buf_size;
1321 	char *text = r->text_buf;
1322 	char prefix[PREFIX_MAX];
1323 	bool truncated = false;
1324 	size_t prefix_len;
1325 	size_t line_len;
1326 	size_t len = 0;
1327 	char *next;
1328 
1329 	/*
1330 	 * If the message was truncated because the buffer was not large
1331 	 * enough, treat the available text as if it were the full text.
1332 	 */
1333 	if (text_len > buf_size)
1334 		text_len = buf_size;
1335 
1336 	prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1337 
1338 	/*
1339 	 * @text_len: bytes of unprocessed text
1340 	 * @line_len: bytes of current line _without_ newline
1341 	 * @text:     pointer to beginning of current line
1342 	 * @len:      number of bytes prepared in r->text_buf
1343 	 */
1344 	for (;;) {
1345 		next = memchr(text, '\n', text_len);
1346 		if (next) {
1347 			line_len = next - text;
1348 		} else {
1349 			/* Drop truncated line(s). */
1350 			if (truncated)
1351 				break;
1352 			line_len = text_len;
1353 		}
1354 
1355 		/*
1356 		 * Truncate the text if there is not enough space to add the
1357 		 * prefix and a trailing newline and a terminator.
1358 		 */
1359 		if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1360 			/* Drop even the current line if no space. */
1361 			if (len + prefix_len + line_len + 1 + 1 > buf_size)
1362 				break;
1363 
1364 			text_len = buf_size - len - prefix_len - 1 - 1;
1365 			truncated = true;
1366 		}
1367 
1368 		memmove(text + prefix_len, text, text_len);
1369 		memcpy(text, prefix, prefix_len);
1370 
1371 		/*
1372 		 * Increment the prepared length to include the text and
1373 		 * prefix that were just moved+copied. Also increment for the
1374 		 * newline at the end of this line. If this is the last line,
1375 		 * there is no newline, but it will be added immediately below.
1376 		 */
1377 		len += prefix_len + line_len + 1;
1378 		if (text_len == line_len) {
1379 			/*
1380 			 * This is the last line. Add the trailing newline
1381 			 * removed in vprintk_store().
1382 			 */
1383 			text[prefix_len + line_len] = '\n';
1384 			break;
1385 		}
1386 
1387 		/*
1388 		 * Advance beyond the added prefix and the related line with
1389 		 * its newline.
1390 		 */
1391 		text += prefix_len + line_len + 1;
1392 
1393 		/*
1394 		 * The remaining text has only decreased by the line with its
1395 		 * newline.
1396 		 *
1397 		 * Note that @text_len can become zero. It happens when @text
1398 		 * ended with a newline (either due to truncation or the
1399 		 * original string ending with "\n\n"). The loop is correctly
1400 		 * repeated and (if not truncated) an empty line with a prefix
1401 		 * will be prepared.
1402 		 */
1403 		text_len -= line_len + 1;
1404 	}
1405 
1406 	/*
1407 	 * If a buffer was provided, it will be terminated. Space for the
1408 	 * string terminator is guaranteed to be available. The terminator is
1409 	 * not counted in the return value.
1410 	 */
1411 	if (buf_size > 0)
1412 		r->text_buf[len] = 0;
1413 
1414 	return len;
1415 }
1416 
1417 static size_t get_record_print_text_size(struct printk_info *info,
1418 					 unsigned int line_count,
1419 					 bool syslog, bool time)
1420 {
1421 	char prefix[PREFIX_MAX];
1422 	size_t prefix_len;
1423 
1424 	prefix_len = info_print_prefix(info, syslog, time, prefix);
1425 
1426 	/*
1427 	 * Each line will be preceded with a prefix. The intermediate
1428 	 * newlines are already within the text, but a final trailing
1429 	 * newline will be added.
1430 	 */
1431 	return ((prefix_len * line_count) + info->text_len + 1);
1432 }
1433 
1434 /*
1435  * Beginning with @start_seq, find the first record where it and all following
1436  * records up to (but not including) @max_seq fit into @size.
1437  *
1438  * @max_seq is simply an upper bound and does not need to exist. If the caller
1439  * does not require an upper bound, -1 can be used for @max_seq.
1440  */
1441 static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1442 				  bool syslog, bool time)
1443 {
1444 	struct printk_info info;
1445 	unsigned int line_count;
1446 	size_t len = 0;
1447 	u64 seq;
1448 
1449 	/* Determine the size of the records up to @max_seq. */
1450 	prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1451 		if (info.seq >= max_seq)
1452 			break;
1453 		len += get_record_print_text_size(&info, line_count, syslog, time);
1454 	}
1455 
1456 	/*
1457 	 * Adjust the upper bound for the next loop to avoid subtracting
1458 	 * lengths that were never added.
1459 	 */
1460 	if (seq < max_seq)
1461 		max_seq = seq;
1462 
1463 	/*
1464 	 * Move first record forward until length fits into the buffer. Ignore
1465 	 * newest messages that were not counted in the above cycle. Messages
1466 	 * might appear and get lost in the meantime. This is a best effort
1467 	 * that prevents an infinite loop that could occur with a retry.
1468 	 */
1469 	prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1470 		if (len <= size || info.seq >= max_seq)
1471 			break;
1472 		len -= get_record_print_text_size(&info, line_count, syslog, time);
1473 	}
1474 
1475 	return seq;
1476 }
1477 
1478 /* The caller is responsible for making sure @size is greater than 0. */
1479 static int syslog_print(char __user *buf, int size)
1480 {
1481 	struct printk_info info;
1482 	struct printk_record r;
1483 	char *text;
1484 	int len = 0;
1485 	u64 seq;
1486 
1487 	text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1488 	if (!text)
1489 		return -ENOMEM;
1490 
1491 	prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1492 
1493 	mutex_lock(&syslog_lock);
1494 
1495 	/*
1496 	 * Wait for the @syslog_seq record to be available. @syslog_seq may
1497 	 * change while waiting.
1498 	 */
1499 	do {
1500 		seq = syslog_seq;
1501 
1502 		mutex_unlock(&syslog_lock);
1503 		len = wait_event_interruptible(log_wait, prb_read_valid(prb, seq, NULL));
1504 		mutex_lock(&syslog_lock);
1505 
1506 		if (len)
1507 			goto out;
1508 	} while (syslog_seq != seq);
1509 
1510 	/*
1511 	 * Copy records that fit into the buffer. The above cycle makes sure
1512 	 * that the first record is always available.
1513 	 */
1514 	do {
1515 		size_t n;
1516 		size_t skip;
1517 		int err;
1518 
1519 		if (!prb_read_valid(prb, syslog_seq, &r))
1520 			break;
1521 
1522 		if (r.info->seq != syslog_seq) {
1523 			/* message is gone, move to next valid one */
1524 			syslog_seq = r.info->seq;
1525 			syslog_partial = 0;
1526 		}
1527 
1528 		/*
1529 		 * To keep reading/counting partial line consistent,
1530 		 * use printk_time value as of the beginning of a line.
1531 		 */
1532 		if (!syslog_partial)
1533 			syslog_time = printk_time;
1534 
1535 		skip = syslog_partial;
1536 		n = record_print_text(&r, true, syslog_time);
1537 		if (n - syslog_partial <= size) {
1538 			/* message fits into buffer, move forward */
1539 			syslog_seq = r.info->seq + 1;
1540 			n -= syslog_partial;
1541 			syslog_partial = 0;
1542 		} else if (!len){
1543 			/* partial read(), remember position */
1544 			n = size;
1545 			syslog_partial += n;
1546 		} else
1547 			n = 0;
1548 
1549 		if (!n)
1550 			break;
1551 
1552 		mutex_unlock(&syslog_lock);
1553 		err = copy_to_user(buf, text + skip, n);
1554 		mutex_lock(&syslog_lock);
1555 
1556 		if (err) {
1557 			if (!len)
1558 				len = -EFAULT;
1559 			break;
1560 		}
1561 
1562 		len += n;
1563 		size -= n;
1564 		buf += n;
1565 	} while (size);
1566 out:
1567 	mutex_unlock(&syslog_lock);
1568 	kfree(text);
1569 	return len;
1570 }
1571 
1572 static int syslog_print_all(char __user *buf, int size, bool clear)
1573 {
1574 	struct printk_info info;
1575 	struct printk_record r;
1576 	char *text;
1577 	int len = 0;
1578 	u64 seq;
1579 	bool time;
1580 
1581 	text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1582 	if (!text)
1583 		return -ENOMEM;
1584 
1585 	time = printk_time;
1586 	/*
1587 	 * Find first record that fits, including all following records,
1588 	 * into the user-provided buffer for this dump.
1589 	 */
1590 	seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1591 				     size, true, time);
1592 
1593 	prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1594 
1595 	len = 0;
1596 	prb_for_each_record(seq, prb, seq, &r) {
1597 		int textlen;
1598 
1599 		textlen = record_print_text(&r, true, time);
1600 
1601 		if (len + textlen > size) {
1602 			seq--;
1603 			break;
1604 		}
1605 
1606 		if (copy_to_user(buf + len, text, textlen))
1607 			len = -EFAULT;
1608 		else
1609 			len += textlen;
1610 
1611 		if (len < 0)
1612 			break;
1613 	}
1614 
1615 	if (clear) {
1616 		mutex_lock(&syslog_lock);
1617 		latched_seq_write(&clear_seq, seq);
1618 		mutex_unlock(&syslog_lock);
1619 	}
1620 
1621 	kfree(text);
1622 	return len;
1623 }
1624 
1625 static void syslog_clear(void)
1626 {
1627 	mutex_lock(&syslog_lock);
1628 	latched_seq_write(&clear_seq, prb_next_seq(prb));
1629 	mutex_unlock(&syslog_lock);
1630 }
1631 
1632 int do_syslog(int type, char __user *buf, int len, int source)
1633 {
1634 	struct printk_info info;
1635 	bool clear = false;
1636 	static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1637 	int error;
1638 
1639 	error = check_syslog_permissions(type, source);
1640 	if (error)
1641 		return error;
1642 
1643 	switch (type) {
1644 	case SYSLOG_ACTION_CLOSE:	/* Close log */
1645 		break;
1646 	case SYSLOG_ACTION_OPEN:	/* Open log */
1647 		break;
1648 	case SYSLOG_ACTION_READ:	/* Read from log */
1649 		if (!buf || len < 0)
1650 			return -EINVAL;
1651 		if (!len)
1652 			return 0;
1653 		if (!access_ok(buf, len))
1654 			return -EFAULT;
1655 		error = syslog_print(buf, len);
1656 		break;
1657 	/* Read/clear last kernel messages */
1658 	case SYSLOG_ACTION_READ_CLEAR:
1659 		clear = true;
1660 		fallthrough;
1661 	/* Read last kernel messages */
1662 	case SYSLOG_ACTION_READ_ALL:
1663 		if (!buf || len < 0)
1664 			return -EINVAL;
1665 		if (!len)
1666 			return 0;
1667 		if (!access_ok(buf, len))
1668 			return -EFAULT;
1669 		error = syslog_print_all(buf, len, clear);
1670 		break;
1671 	/* Clear ring buffer */
1672 	case SYSLOG_ACTION_CLEAR:
1673 		syslog_clear();
1674 		break;
1675 	/* Disable logging to console */
1676 	case SYSLOG_ACTION_CONSOLE_OFF:
1677 		if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1678 			saved_console_loglevel = console_loglevel;
1679 		console_loglevel = minimum_console_loglevel;
1680 		break;
1681 	/* Enable logging to console */
1682 	case SYSLOG_ACTION_CONSOLE_ON:
1683 		if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1684 			console_loglevel = saved_console_loglevel;
1685 			saved_console_loglevel = LOGLEVEL_DEFAULT;
1686 		}
1687 		break;
1688 	/* Set level of messages printed to console */
1689 	case SYSLOG_ACTION_CONSOLE_LEVEL:
1690 		if (len < 1 || len > 8)
1691 			return -EINVAL;
1692 		if (len < minimum_console_loglevel)
1693 			len = minimum_console_loglevel;
1694 		console_loglevel = len;
1695 		/* Implicitly re-enable logging to console */
1696 		saved_console_loglevel = LOGLEVEL_DEFAULT;
1697 		break;
1698 	/* Number of chars in the log buffer */
1699 	case SYSLOG_ACTION_SIZE_UNREAD:
1700 		mutex_lock(&syslog_lock);
1701 		if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1702 			/* No unread messages. */
1703 			mutex_unlock(&syslog_lock);
1704 			return 0;
1705 		}
1706 		if (info.seq != syslog_seq) {
1707 			/* messages are gone, move to first one */
1708 			syslog_seq = info.seq;
1709 			syslog_partial = 0;
1710 		}
1711 		if (source == SYSLOG_FROM_PROC) {
1712 			/*
1713 			 * Short-cut for poll(/"proc/kmsg") which simply checks
1714 			 * for pending data, not the size; return the count of
1715 			 * records, not the length.
1716 			 */
1717 			error = prb_next_seq(prb) - syslog_seq;
1718 		} else {
1719 			bool time = syslog_partial ? syslog_time : printk_time;
1720 			unsigned int line_count;
1721 			u64 seq;
1722 
1723 			prb_for_each_info(syslog_seq, prb, seq, &info,
1724 					  &line_count) {
1725 				error += get_record_print_text_size(&info, line_count,
1726 								    true, time);
1727 				time = printk_time;
1728 			}
1729 			error -= syslog_partial;
1730 		}
1731 		mutex_unlock(&syslog_lock);
1732 		break;
1733 	/* Size of the log buffer */
1734 	case SYSLOG_ACTION_SIZE_BUFFER:
1735 		error = log_buf_len;
1736 		break;
1737 	default:
1738 		error = -EINVAL;
1739 		break;
1740 	}
1741 
1742 	return error;
1743 }
1744 
1745 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1746 {
1747 	return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1748 }
1749 
1750 /*
1751  * Special console_lock variants that help to reduce the risk of soft-lockups.
1752  * They allow to pass console_lock to another printk() call using a busy wait.
1753  */
1754 
1755 #ifdef CONFIG_LOCKDEP
1756 static struct lockdep_map console_owner_dep_map = {
1757 	.name = "console_owner"
1758 };
1759 #endif
1760 
1761 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1762 static struct task_struct *console_owner;
1763 static bool console_waiter;
1764 
1765 /**
1766  * console_lock_spinning_enable - mark beginning of code where another
1767  *	thread might safely busy wait
1768  *
1769  * This basically converts console_lock into a spinlock. This marks
1770  * the section where the console_lock owner can not sleep, because
1771  * there may be a waiter spinning (like a spinlock). Also it must be
1772  * ready to hand over the lock at the end of the section.
1773  */
1774 static void console_lock_spinning_enable(void)
1775 {
1776 	raw_spin_lock(&console_owner_lock);
1777 	console_owner = current;
1778 	raw_spin_unlock(&console_owner_lock);
1779 
1780 	/* The waiter may spin on us after setting console_owner */
1781 	spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1782 }
1783 
1784 /**
1785  * console_lock_spinning_disable_and_check - mark end of code where another
1786  *	thread was able to busy wait and check if there is a waiter
1787  *
1788  * This is called at the end of the section where spinning is allowed.
1789  * It has two functions. First, it is a signal that it is no longer
1790  * safe to start busy waiting for the lock. Second, it checks if
1791  * there is a busy waiter and passes the lock rights to her.
1792  *
1793  * Important: Callers lose the lock if there was a busy waiter.
1794  *	They must not touch items synchronized by console_lock
1795  *	in this case.
1796  *
1797  * Return: 1 if the lock rights were passed, 0 otherwise.
1798  */
1799 static int console_lock_spinning_disable_and_check(void)
1800 {
1801 	int waiter;
1802 
1803 	raw_spin_lock(&console_owner_lock);
1804 	waiter = READ_ONCE(console_waiter);
1805 	console_owner = NULL;
1806 	raw_spin_unlock(&console_owner_lock);
1807 
1808 	if (!waiter) {
1809 		spin_release(&console_owner_dep_map, _THIS_IP_);
1810 		return 0;
1811 	}
1812 
1813 	/* The waiter is now free to continue */
1814 	WRITE_ONCE(console_waiter, false);
1815 
1816 	spin_release(&console_owner_dep_map, _THIS_IP_);
1817 
1818 	/*
1819 	 * Hand off console_lock to waiter. The waiter will perform
1820 	 * the up(). After this, the waiter is the console_lock owner.
1821 	 */
1822 	mutex_release(&console_lock_dep_map, _THIS_IP_);
1823 	return 1;
1824 }
1825 
1826 /**
1827  * console_trylock_spinning - try to get console_lock by busy waiting
1828  *
1829  * This allows to busy wait for the console_lock when the current
1830  * owner is running in specially marked sections. It means that
1831  * the current owner is running and cannot reschedule until it
1832  * is ready to lose the lock.
1833  *
1834  * Return: 1 if we got the lock, 0 othrewise
1835  */
1836 static int console_trylock_spinning(void)
1837 {
1838 	struct task_struct *owner = NULL;
1839 	bool waiter;
1840 	bool spin = false;
1841 	unsigned long flags;
1842 
1843 	if (console_trylock())
1844 		return 1;
1845 
1846 	printk_safe_enter_irqsave(flags);
1847 
1848 	raw_spin_lock(&console_owner_lock);
1849 	owner = READ_ONCE(console_owner);
1850 	waiter = READ_ONCE(console_waiter);
1851 	if (!waiter && owner && owner != current) {
1852 		WRITE_ONCE(console_waiter, true);
1853 		spin = true;
1854 	}
1855 	raw_spin_unlock(&console_owner_lock);
1856 
1857 	/*
1858 	 * If there is an active printk() writing to the
1859 	 * consoles, instead of having it write our data too,
1860 	 * see if we can offload that load from the active
1861 	 * printer, and do some printing ourselves.
1862 	 * Go into a spin only if there isn't already a waiter
1863 	 * spinning, and there is an active printer, and
1864 	 * that active printer isn't us (recursive printk?).
1865 	 */
1866 	if (!spin) {
1867 		printk_safe_exit_irqrestore(flags);
1868 		return 0;
1869 	}
1870 
1871 	/* We spin waiting for the owner to release us */
1872 	spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1873 	/* Owner will clear console_waiter on hand off */
1874 	while (READ_ONCE(console_waiter))
1875 		cpu_relax();
1876 	spin_release(&console_owner_dep_map, _THIS_IP_);
1877 
1878 	printk_safe_exit_irqrestore(flags);
1879 	/*
1880 	 * The owner passed the console lock to us.
1881 	 * Since we did not spin on console lock, annotate
1882 	 * this as a trylock. Otherwise lockdep will
1883 	 * complain.
1884 	 */
1885 	mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1886 
1887 	return 1;
1888 }
1889 
1890 /*
1891  * Call the console drivers, asking them to write out
1892  * log_buf[start] to log_buf[end - 1].
1893  * The console_lock must be held.
1894  */
1895 static void call_console_drivers(const char *ext_text, size_t ext_len,
1896 				 const char *text, size_t len)
1897 {
1898 	static char dropped_text[64];
1899 	size_t dropped_len = 0;
1900 	struct console *con;
1901 
1902 	trace_console_rcuidle(text, len);
1903 
1904 	if (!console_drivers)
1905 		return;
1906 
1907 	if (console_dropped) {
1908 		dropped_len = snprintf(dropped_text, sizeof(dropped_text),
1909 				       "** %lu printk messages dropped **\n",
1910 				       console_dropped);
1911 		console_dropped = 0;
1912 	}
1913 
1914 	for_each_console(con) {
1915 		if (exclusive_console && con != exclusive_console)
1916 			continue;
1917 		if (!(con->flags & CON_ENABLED))
1918 			continue;
1919 		if (!con->write)
1920 			continue;
1921 		if (!cpu_online(smp_processor_id()) &&
1922 		    !(con->flags & CON_ANYTIME))
1923 			continue;
1924 		if (con->flags & CON_EXTENDED)
1925 			con->write(con, ext_text, ext_len);
1926 		else {
1927 			if (dropped_len)
1928 				con->write(con, dropped_text, dropped_len);
1929 			con->write(con, text, len);
1930 		}
1931 	}
1932 }
1933 
1934 /*
1935  * Recursion is tracked separately on each CPU. If NMIs are supported, an
1936  * additional NMI context per CPU is also separately tracked. Until per-CPU
1937  * is available, a separate "early tracking" is performed.
1938  */
1939 static DEFINE_PER_CPU(u8, printk_count);
1940 static u8 printk_count_early;
1941 #ifdef CONFIG_HAVE_NMI
1942 static DEFINE_PER_CPU(u8, printk_count_nmi);
1943 static u8 printk_count_nmi_early;
1944 #endif
1945 
1946 /*
1947  * Recursion is limited to keep the output sane. printk() should not require
1948  * more than 1 level of recursion (allowing, for example, printk() to trigger
1949  * a WARN), but a higher value is used in case some printk-internal errors
1950  * exist, such as the ringbuffer validation checks failing.
1951  */
1952 #define PRINTK_MAX_RECURSION 3
1953 
1954 /*
1955  * Return a pointer to the dedicated counter for the CPU+context of the
1956  * caller.
1957  */
1958 static u8 *__printk_recursion_counter(void)
1959 {
1960 #ifdef CONFIG_HAVE_NMI
1961 	if (in_nmi()) {
1962 		if (printk_percpu_data_ready())
1963 			return this_cpu_ptr(&printk_count_nmi);
1964 		return &printk_count_nmi_early;
1965 	}
1966 #endif
1967 	if (printk_percpu_data_ready())
1968 		return this_cpu_ptr(&printk_count);
1969 	return &printk_count_early;
1970 }
1971 
1972 /*
1973  * Enter recursion tracking. Interrupts are disabled to simplify tracking.
1974  * The caller must check the boolean return value to see if the recursion is
1975  * allowed. On failure, interrupts are not disabled.
1976  *
1977  * @recursion_ptr must be a variable of type (u8 *) and is the same variable
1978  * that is passed to printk_exit_irqrestore().
1979  */
1980 #define printk_enter_irqsave(recursion_ptr, flags)	\
1981 ({							\
1982 	bool success = true;				\
1983 							\
1984 	typecheck(u8 *, recursion_ptr);			\
1985 	local_irq_save(flags);				\
1986 	(recursion_ptr) = __printk_recursion_counter();	\
1987 	if (*(recursion_ptr) > PRINTK_MAX_RECURSION) {	\
1988 		local_irq_restore(flags);		\
1989 		success = false;			\
1990 	} else {					\
1991 		(*(recursion_ptr))++;			\
1992 	}						\
1993 	success;					\
1994 })
1995 
1996 /* Exit recursion tracking, restoring interrupts. */
1997 #define printk_exit_irqrestore(recursion_ptr, flags)	\
1998 	do {						\
1999 		typecheck(u8 *, recursion_ptr);		\
2000 		(*(recursion_ptr))--;			\
2001 		local_irq_restore(flags);		\
2002 	} while (0)
2003 
2004 int printk_delay_msec __read_mostly;
2005 
2006 static inline void printk_delay(void)
2007 {
2008 	if (unlikely(printk_delay_msec)) {
2009 		int m = printk_delay_msec;
2010 
2011 		while (m--) {
2012 			mdelay(1);
2013 			touch_nmi_watchdog();
2014 		}
2015 	}
2016 }
2017 
2018 static inline u32 printk_caller_id(void)
2019 {
2020 	return in_task() ? task_pid_nr(current) :
2021 		0x80000000 + raw_smp_processor_id();
2022 }
2023 
2024 /**
2025  * printk_parse_prefix - Parse level and control flags.
2026  *
2027  * @text:     The terminated text message.
2028  * @level:    A pointer to the current level value, will be updated.
2029  * @flags:    A pointer to the current printk_info flags, will be updated.
2030  *
2031  * @level may be NULL if the caller is not interested in the parsed value.
2032  * Otherwise the variable pointed to by @level must be set to
2033  * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2034  *
2035  * @flags may be NULL if the caller is not interested in the parsed value.
2036  * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2037  * value.
2038  *
2039  * Return: The length of the parsed level and control flags.
2040  */
2041 u16 printk_parse_prefix(const char *text, int *level,
2042 			enum printk_info_flags *flags)
2043 {
2044 	u16 prefix_len = 0;
2045 	int kern_level;
2046 
2047 	while (*text) {
2048 		kern_level = printk_get_level(text);
2049 		if (!kern_level)
2050 			break;
2051 
2052 		switch (kern_level) {
2053 		case '0' ... '7':
2054 			if (level && *level == LOGLEVEL_DEFAULT)
2055 				*level = kern_level - '0';
2056 			break;
2057 		case 'c':	/* KERN_CONT */
2058 			if (flags)
2059 				*flags |= LOG_CONT;
2060 		}
2061 
2062 		prefix_len += 2;
2063 		text += 2;
2064 	}
2065 
2066 	return prefix_len;
2067 }
2068 
2069 __printf(5, 0)
2070 static u16 printk_sprint(char *text, u16 size, int facility,
2071 			 enum printk_info_flags *flags, const char *fmt,
2072 			 va_list args)
2073 {
2074 	u16 text_len;
2075 
2076 	text_len = vscnprintf(text, size, fmt, args);
2077 
2078 	/* Mark and strip a trailing newline. */
2079 	if (text_len && text[text_len - 1] == '\n') {
2080 		text_len--;
2081 		*flags |= LOG_NEWLINE;
2082 	}
2083 
2084 	/* Strip log level and control flags. */
2085 	if (facility == 0) {
2086 		u16 prefix_len;
2087 
2088 		prefix_len = printk_parse_prefix(text, NULL, NULL);
2089 		if (prefix_len) {
2090 			text_len -= prefix_len;
2091 			memmove(text, text + prefix_len, text_len);
2092 		}
2093 	}
2094 
2095 	return text_len;
2096 }
2097 
2098 __printf(4, 0)
2099 int vprintk_store(int facility, int level,
2100 		  const struct dev_printk_info *dev_info,
2101 		  const char *fmt, va_list args)
2102 {
2103 	const u32 caller_id = printk_caller_id();
2104 	struct prb_reserved_entry e;
2105 	enum printk_info_flags flags = 0;
2106 	struct printk_record r;
2107 	unsigned long irqflags;
2108 	u16 trunc_msg_len = 0;
2109 	char prefix_buf[8];
2110 	u8 *recursion_ptr;
2111 	u16 reserve_size;
2112 	va_list args2;
2113 	u16 text_len;
2114 	int ret = 0;
2115 	u64 ts_nsec;
2116 
2117 	/*
2118 	 * Since the duration of printk() can vary depending on the message
2119 	 * and state of the ringbuffer, grab the timestamp now so that it is
2120 	 * close to the call of printk(). This provides a more deterministic
2121 	 * timestamp with respect to the caller.
2122 	 */
2123 	ts_nsec = local_clock();
2124 
2125 	if (!printk_enter_irqsave(recursion_ptr, irqflags))
2126 		return 0;
2127 
2128 	/*
2129 	 * The sprintf needs to come first since the syslog prefix might be
2130 	 * passed in as a parameter. An extra byte must be reserved so that
2131 	 * later the vscnprintf() into the reserved buffer has room for the
2132 	 * terminating '\0', which is not counted by vsnprintf().
2133 	 */
2134 	va_copy(args2, args);
2135 	reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2136 	va_end(args2);
2137 
2138 	if (reserve_size > LOG_LINE_MAX)
2139 		reserve_size = LOG_LINE_MAX;
2140 
2141 	/* Extract log level or control flags. */
2142 	if (facility == 0)
2143 		printk_parse_prefix(&prefix_buf[0], &level, &flags);
2144 
2145 	if (level == LOGLEVEL_DEFAULT)
2146 		level = default_message_loglevel;
2147 
2148 	if (dev_info)
2149 		flags |= LOG_NEWLINE;
2150 
2151 	if (flags & LOG_CONT) {
2152 		prb_rec_init_wr(&r, reserve_size);
2153 		if (prb_reserve_in_last(&e, prb, &r, caller_id, LOG_LINE_MAX)) {
2154 			text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2155 						 facility, &flags, fmt, args);
2156 			r.info->text_len += text_len;
2157 
2158 			if (flags & LOG_NEWLINE) {
2159 				r.info->flags |= LOG_NEWLINE;
2160 				prb_final_commit(&e);
2161 			} else {
2162 				prb_commit(&e);
2163 			}
2164 
2165 			ret = text_len;
2166 			goto out;
2167 		}
2168 	}
2169 
2170 	/*
2171 	 * Explicitly initialize the record before every prb_reserve() call.
2172 	 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2173 	 * structure when they fail.
2174 	 */
2175 	prb_rec_init_wr(&r, reserve_size);
2176 	if (!prb_reserve(&e, prb, &r)) {
2177 		/* truncate the message if it is too long for empty buffer */
2178 		truncate_msg(&reserve_size, &trunc_msg_len);
2179 
2180 		prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2181 		if (!prb_reserve(&e, prb, &r))
2182 			goto out;
2183 	}
2184 
2185 	/* fill message */
2186 	text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2187 	if (trunc_msg_len)
2188 		memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2189 	r.info->text_len = text_len + trunc_msg_len;
2190 	r.info->facility = facility;
2191 	r.info->level = level & 7;
2192 	r.info->flags = flags & 0x1f;
2193 	r.info->ts_nsec = ts_nsec;
2194 	r.info->caller_id = caller_id;
2195 	if (dev_info)
2196 		memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2197 
2198 	/* A message without a trailing newline can be continued. */
2199 	if (!(flags & LOG_NEWLINE))
2200 		prb_commit(&e);
2201 	else
2202 		prb_final_commit(&e);
2203 
2204 	ret = text_len + trunc_msg_len;
2205 out:
2206 	printk_exit_irqrestore(recursion_ptr, irqflags);
2207 	return ret;
2208 }
2209 
2210 asmlinkage int vprintk_emit(int facility, int level,
2211 			    const struct dev_printk_info *dev_info,
2212 			    const char *fmt, va_list args)
2213 {
2214 	int printed_len;
2215 	bool in_sched = false;
2216 
2217 	/* Suppress unimportant messages after panic happens */
2218 	if (unlikely(suppress_printk))
2219 		return 0;
2220 
2221 	if (level == LOGLEVEL_SCHED) {
2222 		level = LOGLEVEL_DEFAULT;
2223 		in_sched = true;
2224 	}
2225 
2226 	boot_delay_msec(level);
2227 	printk_delay();
2228 
2229 	printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2230 
2231 	/* If called from the scheduler, we can not call up(). */
2232 	if (!in_sched) {
2233 		/*
2234 		 * Disable preemption to avoid being preempted while holding
2235 		 * console_sem which would prevent anyone from printing to
2236 		 * console
2237 		 */
2238 		preempt_disable();
2239 		/*
2240 		 * Try to acquire and then immediately release the console
2241 		 * semaphore.  The release will print out buffers and wake up
2242 		 * /dev/kmsg and syslog() users.
2243 		 */
2244 		if (console_trylock_spinning())
2245 			console_unlock();
2246 		preempt_enable();
2247 	}
2248 
2249 	wake_up_klogd();
2250 	return printed_len;
2251 }
2252 EXPORT_SYMBOL(vprintk_emit);
2253 
2254 int vprintk_default(const char *fmt, va_list args)
2255 {
2256 	return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2257 }
2258 EXPORT_SYMBOL_GPL(vprintk_default);
2259 
2260 asmlinkage __visible int _printk(const char *fmt, ...)
2261 {
2262 	va_list args;
2263 	int r;
2264 
2265 	va_start(args, fmt);
2266 	r = vprintk(fmt, args);
2267 	va_end(args);
2268 
2269 	return r;
2270 }
2271 EXPORT_SYMBOL(_printk);
2272 
2273 #else /* CONFIG_PRINTK */
2274 
2275 #define CONSOLE_LOG_MAX		0
2276 #define printk_time		false
2277 
2278 #define prb_read_valid(rb, seq, r)	false
2279 #define prb_first_valid_seq(rb)		0
2280 
2281 static u64 syslog_seq;
2282 static u64 console_seq;
2283 static u64 exclusive_console_stop_seq;
2284 static unsigned long console_dropped;
2285 
2286 static size_t record_print_text(const struct printk_record *r,
2287 				bool syslog, bool time)
2288 {
2289 	return 0;
2290 }
2291 static ssize_t info_print_ext_header(char *buf, size_t size,
2292 				     struct printk_info *info)
2293 {
2294 	return 0;
2295 }
2296 static ssize_t msg_print_ext_body(char *buf, size_t size,
2297 				  char *text, size_t text_len,
2298 				  struct dev_printk_info *dev_info) { return 0; }
2299 static void console_lock_spinning_enable(void) { }
2300 static int console_lock_spinning_disable_and_check(void) { return 0; }
2301 static void call_console_drivers(const char *ext_text, size_t ext_len,
2302 				 const char *text, size_t len) {}
2303 static bool suppress_message_printing(int level) { return false; }
2304 
2305 #endif /* CONFIG_PRINTK */
2306 
2307 #ifdef CONFIG_EARLY_PRINTK
2308 struct console *early_console;
2309 
2310 asmlinkage __visible void early_printk(const char *fmt, ...)
2311 {
2312 	va_list ap;
2313 	char buf[512];
2314 	int n;
2315 
2316 	if (!early_console)
2317 		return;
2318 
2319 	va_start(ap, fmt);
2320 	n = vscnprintf(buf, sizeof(buf), fmt, ap);
2321 	va_end(ap);
2322 
2323 	early_console->write(early_console, buf, n);
2324 }
2325 #endif
2326 
2327 static int __add_preferred_console(char *name, int idx, char *options,
2328 				   char *brl_options, bool user_specified)
2329 {
2330 	struct console_cmdline *c;
2331 	int i;
2332 
2333 	/*
2334 	 *	See if this tty is not yet registered, and
2335 	 *	if we have a slot free.
2336 	 */
2337 	for (i = 0, c = console_cmdline;
2338 	     i < MAX_CMDLINECONSOLES && c->name[0];
2339 	     i++, c++) {
2340 		if (strcmp(c->name, name) == 0 && c->index == idx) {
2341 			if (!brl_options)
2342 				preferred_console = i;
2343 			if (user_specified)
2344 				c->user_specified = true;
2345 			return 0;
2346 		}
2347 	}
2348 	if (i == MAX_CMDLINECONSOLES)
2349 		return -E2BIG;
2350 	if (!brl_options)
2351 		preferred_console = i;
2352 	strlcpy(c->name, name, sizeof(c->name));
2353 	c->options = options;
2354 	c->user_specified = user_specified;
2355 	braille_set_options(c, brl_options);
2356 
2357 	c->index = idx;
2358 	return 0;
2359 }
2360 
2361 static int __init console_msg_format_setup(char *str)
2362 {
2363 	if (!strcmp(str, "syslog"))
2364 		console_msg_format = MSG_FORMAT_SYSLOG;
2365 	if (!strcmp(str, "default"))
2366 		console_msg_format = MSG_FORMAT_DEFAULT;
2367 	return 1;
2368 }
2369 __setup("console_msg_format=", console_msg_format_setup);
2370 
2371 /*
2372  * Set up a console.  Called via do_early_param() in init/main.c
2373  * for each "console=" parameter in the boot command line.
2374  */
2375 static int __init console_setup(char *str)
2376 {
2377 	char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2378 	char *s, *options, *brl_options = NULL;
2379 	int idx;
2380 
2381 	/*
2382 	 * console="" or console=null have been suggested as a way to
2383 	 * disable console output. Use ttynull that has been created
2384 	 * for exactly this purpose.
2385 	 */
2386 	if (str[0] == 0 || strcmp(str, "null") == 0) {
2387 		__add_preferred_console("ttynull", 0, NULL, NULL, true);
2388 		return 1;
2389 	}
2390 
2391 	if (_braille_console_setup(&str, &brl_options))
2392 		return 1;
2393 
2394 	/*
2395 	 * Decode str into name, index, options.
2396 	 */
2397 	if (str[0] >= '0' && str[0] <= '9') {
2398 		strcpy(buf, "ttyS");
2399 		strncpy(buf + 4, str, sizeof(buf) - 5);
2400 	} else {
2401 		strncpy(buf, str, sizeof(buf) - 1);
2402 	}
2403 	buf[sizeof(buf) - 1] = 0;
2404 	options = strchr(str, ',');
2405 	if (options)
2406 		*(options++) = 0;
2407 #ifdef __sparc__
2408 	if (!strcmp(str, "ttya"))
2409 		strcpy(buf, "ttyS0");
2410 	if (!strcmp(str, "ttyb"))
2411 		strcpy(buf, "ttyS1");
2412 #endif
2413 	for (s = buf; *s; s++)
2414 		if (isdigit(*s) || *s == ',')
2415 			break;
2416 	idx = simple_strtoul(s, NULL, 10);
2417 	*s = 0;
2418 
2419 	__add_preferred_console(buf, idx, options, brl_options, true);
2420 	console_set_on_cmdline = 1;
2421 	return 1;
2422 }
2423 __setup("console=", console_setup);
2424 
2425 /**
2426  * add_preferred_console - add a device to the list of preferred consoles.
2427  * @name: device name
2428  * @idx: device index
2429  * @options: options for this console
2430  *
2431  * The last preferred console added will be used for kernel messages
2432  * and stdin/out/err for init.  Normally this is used by console_setup
2433  * above to handle user-supplied console arguments; however it can also
2434  * be used by arch-specific code either to override the user or more
2435  * commonly to provide a default console (ie from PROM variables) when
2436  * the user has not supplied one.
2437  */
2438 int add_preferred_console(char *name, int idx, char *options)
2439 {
2440 	return __add_preferred_console(name, idx, options, NULL, false);
2441 }
2442 
2443 bool console_suspend_enabled = true;
2444 EXPORT_SYMBOL(console_suspend_enabled);
2445 
2446 static int __init console_suspend_disable(char *str)
2447 {
2448 	console_suspend_enabled = false;
2449 	return 1;
2450 }
2451 __setup("no_console_suspend", console_suspend_disable);
2452 module_param_named(console_suspend, console_suspend_enabled,
2453 		bool, S_IRUGO | S_IWUSR);
2454 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2455 	" and hibernate operations");
2456 
2457 static bool printk_console_no_auto_verbose;
2458 
2459 void console_verbose(void)
2460 {
2461 	if (console_loglevel && !printk_console_no_auto_verbose)
2462 		console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2463 }
2464 EXPORT_SYMBOL_GPL(console_verbose);
2465 
2466 module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2467 MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2468 
2469 /**
2470  * suspend_console - suspend the console subsystem
2471  *
2472  * This disables printk() while we go into suspend states
2473  */
2474 void suspend_console(void)
2475 {
2476 	if (!console_suspend_enabled)
2477 		return;
2478 	pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2479 	console_lock();
2480 	console_suspended = 1;
2481 	up_console_sem();
2482 }
2483 
2484 void resume_console(void)
2485 {
2486 	if (!console_suspend_enabled)
2487 		return;
2488 	down_console_sem();
2489 	console_suspended = 0;
2490 	console_unlock();
2491 }
2492 
2493 /**
2494  * console_cpu_notify - print deferred console messages after CPU hotplug
2495  * @cpu: unused
2496  *
2497  * If printk() is called from a CPU that is not online yet, the messages
2498  * will be printed on the console only if there are CON_ANYTIME consoles.
2499  * This function is called when a new CPU comes online (or fails to come
2500  * up) or goes offline.
2501  */
2502 static int console_cpu_notify(unsigned int cpu)
2503 {
2504 	if (!cpuhp_tasks_frozen) {
2505 		/* If trylock fails, someone else is doing the printing */
2506 		if (console_trylock())
2507 			console_unlock();
2508 	}
2509 	return 0;
2510 }
2511 
2512 /**
2513  * console_lock - lock the console system for exclusive use.
2514  *
2515  * Acquires a lock which guarantees that the caller has
2516  * exclusive access to the console system and the console_drivers list.
2517  *
2518  * Can sleep, returns nothing.
2519  */
2520 void console_lock(void)
2521 {
2522 	might_sleep();
2523 
2524 	down_console_sem();
2525 	if (console_suspended)
2526 		return;
2527 	console_locked = 1;
2528 	console_may_schedule = 1;
2529 }
2530 EXPORT_SYMBOL(console_lock);
2531 
2532 /**
2533  * console_trylock - try to lock the console system for exclusive use.
2534  *
2535  * Try to acquire a lock which guarantees that the caller has exclusive
2536  * access to the console system and the console_drivers list.
2537  *
2538  * returns 1 on success, and 0 on failure to acquire the lock.
2539  */
2540 int console_trylock(void)
2541 {
2542 	if (down_trylock_console_sem())
2543 		return 0;
2544 	if (console_suspended) {
2545 		up_console_sem();
2546 		return 0;
2547 	}
2548 	console_locked = 1;
2549 	console_may_schedule = 0;
2550 	return 1;
2551 }
2552 EXPORT_SYMBOL(console_trylock);
2553 
2554 int is_console_locked(void)
2555 {
2556 	return console_locked;
2557 }
2558 EXPORT_SYMBOL(is_console_locked);
2559 
2560 /*
2561  * Check if we have any console that is capable of printing while cpu is
2562  * booting or shutting down. Requires console_sem.
2563  */
2564 static int have_callable_console(void)
2565 {
2566 	struct console *con;
2567 
2568 	for_each_console(con)
2569 		if ((con->flags & CON_ENABLED) &&
2570 				(con->flags & CON_ANYTIME))
2571 			return 1;
2572 
2573 	return 0;
2574 }
2575 
2576 /*
2577  * Can we actually use the console at this time on this cpu?
2578  *
2579  * Console drivers may assume that per-cpu resources have been allocated. So
2580  * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2581  * call them until this CPU is officially up.
2582  */
2583 static inline int can_use_console(void)
2584 {
2585 	return cpu_online(raw_smp_processor_id()) || have_callable_console();
2586 }
2587 
2588 /**
2589  * console_unlock - unlock the console system
2590  *
2591  * Releases the console_lock which the caller holds on the console system
2592  * and the console driver list.
2593  *
2594  * While the console_lock was held, console output may have been buffered
2595  * by printk().  If this is the case, console_unlock(); emits
2596  * the output prior to releasing the lock.
2597  *
2598  * If there is output waiting, we wake /dev/kmsg and syslog() users.
2599  *
2600  * console_unlock(); may be called from any context.
2601  */
2602 void console_unlock(void)
2603 {
2604 	static char ext_text[CONSOLE_EXT_LOG_MAX];
2605 	static char text[CONSOLE_LOG_MAX];
2606 	unsigned long flags;
2607 	bool do_cond_resched, retry;
2608 	struct printk_info info;
2609 	struct printk_record r;
2610 	u64 __maybe_unused next_seq;
2611 
2612 	if (console_suspended) {
2613 		up_console_sem();
2614 		return;
2615 	}
2616 
2617 	prb_rec_init_rd(&r, &info, text, sizeof(text));
2618 
2619 	/*
2620 	 * Console drivers are called with interrupts disabled, so
2621 	 * @console_may_schedule should be cleared before; however, we may
2622 	 * end up dumping a lot of lines, for example, if called from
2623 	 * console registration path, and should invoke cond_resched()
2624 	 * between lines if allowable.  Not doing so can cause a very long
2625 	 * scheduling stall on a slow console leading to RCU stall and
2626 	 * softlockup warnings which exacerbate the issue with more
2627 	 * messages practically incapacitating the system.
2628 	 *
2629 	 * console_trylock() is not able to detect the preemptive
2630 	 * context reliably. Therefore the value must be stored before
2631 	 * and cleared after the "again" goto label.
2632 	 */
2633 	do_cond_resched = console_may_schedule;
2634 again:
2635 	console_may_schedule = 0;
2636 
2637 	/*
2638 	 * We released the console_sem lock, so we need to recheck if
2639 	 * cpu is online and (if not) is there at least one CON_ANYTIME
2640 	 * console.
2641 	 */
2642 	if (!can_use_console()) {
2643 		console_locked = 0;
2644 		up_console_sem();
2645 		return;
2646 	}
2647 
2648 	for (;;) {
2649 		size_t ext_len = 0;
2650 		int handover;
2651 		size_t len;
2652 
2653 skip:
2654 		if (!prb_read_valid(prb, console_seq, &r))
2655 			break;
2656 
2657 		if (console_seq != r.info->seq) {
2658 			console_dropped += r.info->seq - console_seq;
2659 			console_seq = r.info->seq;
2660 		}
2661 
2662 		if (suppress_message_printing(r.info->level)) {
2663 			/*
2664 			 * Skip record we have buffered and already printed
2665 			 * directly to the console when we received it, and
2666 			 * record that has level above the console loglevel.
2667 			 */
2668 			console_seq++;
2669 			goto skip;
2670 		}
2671 
2672 		/* Output to all consoles once old messages replayed. */
2673 		if (unlikely(exclusive_console &&
2674 			     console_seq >= exclusive_console_stop_seq)) {
2675 			exclusive_console = NULL;
2676 		}
2677 
2678 		/*
2679 		 * Handle extended console text first because later
2680 		 * record_print_text() will modify the record buffer in-place.
2681 		 */
2682 		if (nr_ext_console_drivers) {
2683 			ext_len = info_print_ext_header(ext_text,
2684 						sizeof(ext_text),
2685 						r.info);
2686 			ext_len += msg_print_ext_body(ext_text + ext_len,
2687 						sizeof(ext_text) - ext_len,
2688 						&r.text_buf[0],
2689 						r.info->text_len,
2690 						&r.info->dev_info);
2691 		}
2692 		len = record_print_text(&r,
2693 				console_msg_format & MSG_FORMAT_SYSLOG,
2694 				printk_time);
2695 		console_seq++;
2696 
2697 		/*
2698 		 * While actively printing out messages, if another printk()
2699 		 * were to occur on another CPU, it may wait for this one to
2700 		 * finish. This task can not be preempted if there is a
2701 		 * waiter waiting to take over.
2702 		 *
2703 		 * Interrupts are disabled because the hand over to a waiter
2704 		 * must not be interrupted until the hand over is completed
2705 		 * (@console_waiter is cleared).
2706 		 */
2707 		printk_safe_enter_irqsave(flags);
2708 		console_lock_spinning_enable();
2709 
2710 		stop_critical_timings();	/* don't trace print latency */
2711 		call_console_drivers(ext_text, ext_len, text, len);
2712 		start_critical_timings();
2713 
2714 		handover = console_lock_spinning_disable_and_check();
2715 		printk_safe_exit_irqrestore(flags);
2716 		if (handover)
2717 			return;
2718 
2719 		if (do_cond_resched)
2720 			cond_resched();
2721 	}
2722 
2723 	/* Get consistent value of the next-to-be-used sequence number. */
2724 	next_seq = console_seq;
2725 
2726 	console_locked = 0;
2727 	up_console_sem();
2728 
2729 	/*
2730 	 * Someone could have filled up the buffer again, so re-check if there's
2731 	 * something to flush. In case we cannot trylock the console_sem again,
2732 	 * there's a new owner and the console_unlock() from them will do the
2733 	 * flush, no worries.
2734 	 */
2735 	retry = prb_read_valid(prb, next_seq, NULL);
2736 	if (retry && console_trylock())
2737 		goto again;
2738 }
2739 EXPORT_SYMBOL(console_unlock);
2740 
2741 /**
2742  * console_conditional_schedule - yield the CPU if required
2743  *
2744  * If the console code is currently allowed to sleep, and
2745  * if this CPU should yield the CPU to another task, do
2746  * so here.
2747  *
2748  * Must be called within console_lock();.
2749  */
2750 void __sched console_conditional_schedule(void)
2751 {
2752 	if (console_may_schedule)
2753 		cond_resched();
2754 }
2755 EXPORT_SYMBOL(console_conditional_schedule);
2756 
2757 void console_unblank(void)
2758 {
2759 	struct console *c;
2760 
2761 	/*
2762 	 * console_unblank can no longer be called in interrupt context unless
2763 	 * oops_in_progress is set to 1..
2764 	 */
2765 	if (oops_in_progress) {
2766 		if (down_trylock_console_sem() != 0)
2767 			return;
2768 	} else
2769 		console_lock();
2770 
2771 	console_locked = 1;
2772 	console_may_schedule = 0;
2773 	for_each_console(c)
2774 		if ((c->flags & CON_ENABLED) && c->unblank)
2775 			c->unblank();
2776 	console_unlock();
2777 }
2778 
2779 /**
2780  * console_flush_on_panic - flush console content on panic
2781  * @mode: flush all messages in buffer or just the pending ones
2782  *
2783  * Immediately output all pending messages no matter what.
2784  */
2785 void console_flush_on_panic(enum con_flush_mode mode)
2786 {
2787 	/*
2788 	 * If someone else is holding the console lock, trylock will fail
2789 	 * and may_schedule may be set.  Ignore and proceed to unlock so
2790 	 * that messages are flushed out.  As this can be called from any
2791 	 * context and we don't want to get preempted while flushing,
2792 	 * ensure may_schedule is cleared.
2793 	 */
2794 	console_trylock();
2795 	console_may_schedule = 0;
2796 
2797 	if (mode == CONSOLE_REPLAY_ALL)
2798 		console_seq = prb_first_valid_seq(prb);
2799 	console_unlock();
2800 }
2801 
2802 /*
2803  * Return the console tty driver structure and its associated index
2804  */
2805 struct tty_driver *console_device(int *index)
2806 {
2807 	struct console *c;
2808 	struct tty_driver *driver = NULL;
2809 
2810 	console_lock();
2811 	for_each_console(c) {
2812 		if (!c->device)
2813 			continue;
2814 		driver = c->device(c, index);
2815 		if (driver)
2816 			break;
2817 	}
2818 	console_unlock();
2819 	return driver;
2820 }
2821 
2822 /*
2823  * Prevent further output on the passed console device so that (for example)
2824  * serial drivers can disable console output before suspending a port, and can
2825  * re-enable output afterwards.
2826  */
2827 void console_stop(struct console *console)
2828 {
2829 	console_lock();
2830 	console->flags &= ~CON_ENABLED;
2831 	console_unlock();
2832 }
2833 EXPORT_SYMBOL(console_stop);
2834 
2835 void console_start(struct console *console)
2836 {
2837 	console_lock();
2838 	console->flags |= CON_ENABLED;
2839 	console_unlock();
2840 }
2841 EXPORT_SYMBOL(console_start);
2842 
2843 static int __read_mostly keep_bootcon;
2844 
2845 static int __init keep_bootcon_setup(char *str)
2846 {
2847 	keep_bootcon = 1;
2848 	pr_info("debug: skip boot console de-registration.\n");
2849 
2850 	return 0;
2851 }
2852 
2853 early_param("keep_bootcon", keep_bootcon_setup);
2854 
2855 /*
2856  * This is called by register_console() to try to match
2857  * the newly registered console with any of the ones selected
2858  * by either the command line or add_preferred_console() and
2859  * setup/enable it.
2860  *
2861  * Care need to be taken with consoles that are statically
2862  * enabled such as netconsole
2863  */
2864 static int try_enable_new_console(struct console *newcon, bool user_specified)
2865 {
2866 	struct console_cmdline *c;
2867 	int i, err;
2868 
2869 	for (i = 0, c = console_cmdline;
2870 	     i < MAX_CMDLINECONSOLES && c->name[0];
2871 	     i++, c++) {
2872 		if (c->user_specified != user_specified)
2873 			continue;
2874 		if (!newcon->match ||
2875 		    newcon->match(newcon, c->name, c->index, c->options) != 0) {
2876 			/* default matching */
2877 			BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2878 			if (strcmp(c->name, newcon->name) != 0)
2879 				continue;
2880 			if (newcon->index >= 0 &&
2881 			    newcon->index != c->index)
2882 				continue;
2883 			if (newcon->index < 0)
2884 				newcon->index = c->index;
2885 
2886 			if (_braille_register_console(newcon, c))
2887 				return 0;
2888 
2889 			if (newcon->setup &&
2890 			    (err = newcon->setup(newcon, c->options)) != 0)
2891 				return err;
2892 		}
2893 		newcon->flags |= CON_ENABLED;
2894 		if (i == preferred_console) {
2895 			newcon->flags |= CON_CONSDEV;
2896 			has_preferred_console = true;
2897 		}
2898 		return 0;
2899 	}
2900 
2901 	/*
2902 	 * Some consoles, such as pstore and netconsole, can be enabled even
2903 	 * without matching. Accept the pre-enabled consoles only when match()
2904 	 * and setup() had a chance to be called.
2905 	 */
2906 	if (newcon->flags & CON_ENABLED && c->user_specified ==	user_specified)
2907 		return 0;
2908 
2909 	return -ENOENT;
2910 }
2911 
2912 /*
2913  * The console driver calls this routine during kernel initialization
2914  * to register the console printing procedure with printk() and to
2915  * print any messages that were printed by the kernel before the
2916  * console driver was initialized.
2917  *
2918  * This can happen pretty early during the boot process (because of
2919  * early_printk) - sometimes before setup_arch() completes - be careful
2920  * of what kernel features are used - they may not be initialised yet.
2921  *
2922  * There are two types of consoles - bootconsoles (early_printk) and
2923  * "real" consoles (everything which is not a bootconsole) which are
2924  * handled differently.
2925  *  - Any number of bootconsoles can be registered at any time.
2926  *  - As soon as a "real" console is registered, all bootconsoles
2927  *    will be unregistered automatically.
2928  *  - Once a "real" console is registered, any attempt to register a
2929  *    bootconsoles will be rejected
2930  */
2931 void register_console(struct console *newcon)
2932 {
2933 	struct console *bcon = NULL;
2934 	int err;
2935 
2936 	for_each_console(bcon) {
2937 		if (WARN(bcon == newcon, "console '%s%d' already registered\n",
2938 					 bcon->name, bcon->index))
2939 			return;
2940 	}
2941 
2942 	/*
2943 	 * before we register a new CON_BOOT console, make sure we don't
2944 	 * already have a valid console
2945 	 */
2946 	if (newcon->flags & CON_BOOT) {
2947 		for_each_console(bcon) {
2948 			if (!(bcon->flags & CON_BOOT)) {
2949 				pr_info("Too late to register bootconsole %s%d\n",
2950 					newcon->name, newcon->index);
2951 				return;
2952 			}
2953 		}
2954 	}
2955 
2956 	if (console_drivers && console_drivers->flags & CON_BOOT)
2957 		bcon = console_drivers;
2958 
2959 	if (!has_preferred_console || bcon || !console_drivers)
2960 		has_preferred_console = preferred_console >= 0;
2961 
2962 	/*
2963 	 *	See if we want to use this console driver. If we
2964 	 *	didn't select a console we take the first one
2965 	 *	that registers here.
2966 	 */
2967 	if (!has_preferred_console) {
2968 		if (newcon->index < 0)
2969 			newcon->index = 0;
2970 		if (newcon->setup == NULL ||
2971 		    newcon->setup(newcon, NULL) == 0) {
2972 			newcon->flags |= CON_ENABLED;
2973 			if (newcon->device) {
2974 				newcon->flags |= CON_CONSDEV;
2975 				has_preferred_console = true;
2976 			}
2977 		}
2978 	}
2979 
2980 	/* See if this console matches one we selected on the command line */
2981 	err = try_enable_new_console(newcon, true);
2982 
2983 	/* If not, try to match against the platform default(s) */
2984 	if (err == -ENOENT)
2985 		err = try_enable_new_console(newcon, false);
2986 
2987 	/* printk() messages are not printed to the Braille console. */
2988 	if (err || newcon->flags & CON_BRL)
2989 		return;
2990 
2991 	/*
2992 	 * If we have a bootconsole, and are switching to a real console,
2993 	 * don't print everything out again, since when the boot console, and
2994 	 * the real console are the same physical device, it's annoying to
2995 	 * see the beginning boot messages twice
2996 	 */
2997 	if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2998 		newcon->flags &= ~CON_PRINTBUFFER;
2999 
3000 	/*
3001 	 *	Put this console in the list - keep the
3002 	 *	preferred driver at the head of the list.
3003 	 */
3004 	console_lock();
3005 	if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
3006 		newcon->next = console_drivers;
3007 		console_drivers = newcon;
3008 		if (newcon->next)
3009 			newcon->next->flags &= ~CON_CONSDEV;
3010 		/* Ensure this flag is always set for the head of the list */
3011 		newcon->flags |= CON_CONSDEV;
3012 	} else {
3013 		newcon->next = console_drivers->next;
3014 		console_drivers->next = newcon;
3015 	}
3016 
3017 	if (newcon->flags & CON_EXTENDED)
3018 		nr_ext_console_drivers++;
3019 
3020 	if (newcon->flags & CON_PRINTBUFFER) {
3021 		/*
3022 		 * console_unlock(); will print out the buffered messages
3023 		 * for us.
3024 		 *
3025 		 * We're about to replay the log buffer.  Only do this to the
3026 		 * just-registered console to avoid excessive message spam to
3027 		 * the already-registered consoles.
3028 		 *
3029 		 * Set exclusive_console with disabled interrupts to reduce
3030 		 * race window with eventual console_flush_on_panic() that
3031 		 * ignores console_lock.
3032 		 */
3033 		exclusive_console = newcon;
3034 		exclusive_console_stop_seq = console_seq;
3035 
3036 		/* Get a consistent copy of @syslog_seq. */
3037 		mutex_lock(&syslog_lock);
3038 		console_seq = syslog_seq;
3039 		mutex_unlock(&syslog_lock);
3040 	}
3041 	console_unlock();
3042 	console_sysfs_notify();
3043 
3044 	/*
3045 	 * By unregistering the bootconsoles after we enable the real console
3046 	 * we get the "console xxx enabled" message on all the consoles -
3047 	 * boot consoles, real consoles, etc - this is to ensure that end
3048 	 * users know there might be something in the kernel's log buffer that
3049 	 * went to the bootconsole (that they do not see on the real console)
3050 	 */
3051 	pr_info("%sconsole [%s%d] enabled\n",
3052 		(newcon->flags & CON_BOOT) ? "boot" : "" ,
3053 		newcon->name, newcon->index);
3054 	if (bcon &&
3055 	    ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
3056 	    !keep_bootcon) {
3057 		/* We need to iterate through all boot consoles, to make
3058 		 * sure we print everything out, before we unregister them.
3059 		 */
3060 		for_each_console(bcon)
3061 			if (bcon->flags & CON_BOOT)
3062 				unregister_console(bcon);
3063 	}
3064 }
3065 EXPORT_SYMBOL(register_console);
3066 
3067 int unregister_console(struct console *console)
3068 {
3069 	struct console *con;
3070 	int res;
3071 
3072 	pr_info("%sconsole [%s%d] disabled\n",
3073 		(console->flags & CON_BOOT) ? "boot" : "" ,
3074 		console->name, console->index);
3075 
3076 	res = _braille_unregister_console(console);
3077 	if (res < 0)
3078 		return res;
3079 	if (res > 0)
3080 		return 0;
3081 
3082 	res = -ENODEV;
3083 	console_lock();
3084 	if (console_drivers == console) {
3085 		console_drivers=console->next;
3086 		res = 0;
3087 	} else {
3088 		for_each_console(con) {
3089 			if (con->next == console) {
3090 				con->next = console->next;
3091 				res = 0;
3092 				break;
3093 			}
3094 		}
3095 	}
3096 
3097 	if (res)
3098 		goto out_disable_unlock;
3099 
3100 	if (console->flags & CON_EXTENDED)
3101 		nr_ext_console_drivers--;
3102 
3103 	/*
3104 	 * If this isn't the last console and it has CON_CONSDEV set, we
3105 	 * need to set it on the next preferred console.
3106 	 */
3107 	if (console_drivers != NULL && console->flags & CON_CONSDEV)
3108 		console_drivers->flags |= CON_CONSDEV;
3109 
3110 	console->flags &= ~CON_ENABLED;
3111 	console_unlock();
3112 	console_sysfs_notify();
3113 
3114 	if (console->exit)
3115 		res = console->exit(console);
3116 
3117 	return res;
3118 
3119 out_disable_unlock:
3120 	console->flags &= ~CON_ENABLED;
3121 	console_unlock();
3122 
3123 	return res;
3124 }
3125 EXPORT_SYMBOL(unregister_console);
3126 
3127 /*
3128  * Initialize the console device. This is called *early*, so
3129  * we can't necessarily depend on lots of kernel help here.
3130  * Just do some early initializations, and do the complex setup
3131  * later.
3132  */
3133 void __init console_init(void)
3134 {
3135 	int ret;
3136 	initcall_t call;
3137 	initcall_entry_t *ce;
3138 
3139 	/* Setup the default TTY line discipline. */
3140 	n_tty_init();
3141 
3142 	/*
3143 	 * set up the console device so that later boot sequences can
3144 	 * inform about problems etc..
3145 	 */
3146 	ce = __con_initcall_start;
3147 	trace_initcall_level("console");
3148 	while (ce < __con_initcall_end) {
3149 		call = initcall_from_entry(ce);
3150 		trace_initcall_start(call);
3151 		ret = call();
3152 		trace_initcall_finish(call, ret);
3153 		ce++;
3154 	}
3155 }
3156 
3157 /*
3158  * Some boot consoles access data that is in the init section and which will
3159  * be discarded after the initcalls have been run. To make sure that no code
3160  * will access this data, unregister the boot consoles in a late initcall.
3161  *
3162  * If for some reason, such as deferred probe or the driver being a loadable
3163  * module, the real console hasn't registered yet at this point, there will
3164  * be a brief interval in which no messages are logged to the console, which
3165  * makes it difficult to diagnose problems that occur during this time.
3166  *
3167  * To mitigate this problem somewhat, only unregister consoles whose memory
3168  * intersects with the init section. Note that all other boot consoles will
3169  * get unregistered when the real preferred console is registered.
3170  */
3171 static int __init printk_late_init(void)
3172 {
3173 	struct console *con;
3174 	int ret;
3175 
3176 	for_each_console(con) {
3177 		if (!(con->flags & CON_BOOT))
3178 			continue;
3179 
3180 		/* Check addresses that might be used for enabled consoles. */
3181 		if (init_section_intersects(con, sizeof(*con)) ||
3182 		    init_section_contains(con->write, 0) ||
3183 		    init_section_contains(con->read, 0) ||
3184 		    init_section_contains(con->device, 0) ||
3185 		    init_section_contains(con->unblank, 0) ||
3186 		    init_section_contains(con->data, 0)) {
3187 			/*
3188 			 * Please, consider moving the reported consoles out
3189 			 * of the init section.
3190 			 */
3191 			pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3192 				con->name, con->index);
3193 			unregister_console(con);
3194 		}
3195 	}
3196 	ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3197 					console_cpu_notify);
3198 	WARN_ON(ret < 0);
3199 	ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3200 					console_cpu_notify, NULL);
3201 	WARN_ON(ret < 0);
3202 	return 0;
3203 }
3204 late_initcall(printk_late_init);
3205 
3206 #if defined CONFIG_PRINTK
3207 /*
3208  * Delayed printk version, for scheduler-internal messages:
3209  */
3210 #define PRINTK_PENDING_WAKEUP	0x01
3211 #define PRINTK_PENDING_OUTPUT	0x02
3212 
3213 static DEFINE_PER_CPU(int, printk_pending);
3214 
3215 static void wake_up_klogd_work_func(struct irq_work *irq_work)
3216 {
3217 	int pending = __this_cpu_xchg(printk_pending, 0);
3218 
3219 	if (pending & PRINTK_PENDING_OUTPUT) {
3220 		/* If trylock fails, someone else is doing the printing */
3221 		if (console_trylock())
3222 			console_unlock();
3223 	}
3224 
3225 	if (pending & PRINTK_PENDING_WAKEUP)
3226 		wake_up_interruptible(&log_wait);
3227 }
3228 
3229 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3230 	IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3231 
3232 void wake_up_klogd(void)
3233 {
3234 	if (!printk_percpu_data_ready())
3235 		return;
3236 
3237 	preempt_disable();
3238 	if (waitqueue_active(&log_wait)) {
3239 		this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
3240 		irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3241 	}
3242 	preempt_enable();
3243 }
3244 
3245 void defer_console_output(void)
3246 {
3247 	if (!printk_percpu_data_ready())
3248 		return;
3249 
3250 	preempt_disable();
3251 	__this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
3252 	irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3253 	preempt_enable();
3254 }
3255 
3256 int vprintk_deferred(const char *fmt, va_list args)
3257 {
3258 	int r;
3259 
3260 	r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3261 	defer_console_output();
3262 
3263 	return r;
3264 }
3265 
3266 int _printk_deferred(const char *fmt, ...)
3267 {
3268 	va_list args;
3269 	int r;
3270 
3271 	va_start(args, fmt);
3272 	r = vprintk_deferred(fmt, args);
3273 	va_end(args);
3274 
3275 	return r;
3276 }
3277 
3278 /*
3279  * printk rate limiting, lifted from the networking subsystem.
3280  *
3281  * This enforces a rate limit: not more than 10 kernel messages
3282  * every 5s to make a denial-of-service attack impossible.
3283  */
3284 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3285 
3286 int __printk_ratelimit(const char *func)
3287 {
3288 	return ___ratelimit(&printk_ratelimit_state, func);
3289 }
3290 EXPORT_SYMBOL(__printk_ratelimit);
3291 
3292 /**
3293  * printk_timed_ratelimit - caller-controlled printk ratelimiting
3294  * @caller_jiffies: pointer to caller's state
3295  * @interval_msecs: minimum interval between prints
3296  *
3297  * printk_timed_ratelimit() returns true if more than @interval_msecs
3298  * milliseconds have elapsed since the last time printk_timed_ratelimit()
3299  * returned true.
3300  */
3301 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3302 			unsigned int interval_msecs)
3303 {
3304 	unsigned long elapsed = jiffies - *caller_jiffies;
3305 
3306 	if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3307 		return false;
3308 
3309 	*caller_jiffies = jiffies;
3310 	return true;
3311 }
3312 EXPORT_SYMBOL(printk_timed_ratelimit);
3313 
3314 static DEFINE_SPINLOCK(dump_list_lock);
3315 static LIST_HEAD(dump_list);
3316 
3317 /**
3318  * kmsg_dump_register - register a kernel log dumper.
3319  * @dumper: pointer to the kmsg_dumper structure
3320  *
3321  * Adds a kernel log dumper to the system. The dump callback in the
3322  * structure will be called when the kernel oopses or panics and must be
3323  * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3324  */
3325 int kmsg_dump_register(struct kmsg_dumper *dumper)
3326 {
3327 	unsigned long flags;
3328 	int err = -EBUSY;
3329 
3330 	/* The dump callback needs to be set */
3331 	if (!dumper->dump)
3332 		return -EINVAL;
3333 
3334 	spin_lock_irqsave(&dump_list_lock, flags);
3335 	/* Don't allow registering multiple times */
3336 	if (!dumper->registered) {
3337 		dumper->registered = 1;
3338 		list_add_tail_rcu(&dumper->list, &dump_list);
3339 		err = 0;
3340 	}
3341 	spin_unlock_irqrestore(&dump_list_lock, flags);
3342 
3343 	return err;
3344 }
3345 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3346 
3347 /**
3348  * kmsg_dump_unregister - unregister a kmsg dumper.
3349  * @dumper: pointer to the kmsg_dumper structure
3350  *
3351  * Removes a dump device from the system. Returns zero on success and
3352  * %-EINVAL otherwise.
3353  */
3354 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3355 {
3356 	unsigned long flags;
3357 	int err = -EINVAL;
3358 
3359 	spin_lock_irqsave(&dump_list_lock, flags);
3360 	if (dumper->registered) {
3361 		dumper->registered = 0;
3362 		list_del_rcu(&dumper->list);
3363 		err = 0;
3364 	}
3365 	spin_unlock_irqrestore(&dump_list_lock, flags);
3366 	synchronize_rcu();
3367 
3368 	return err;
3369 }
3370 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3371 
3372 static bool always_kmsg_dump;
3373 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3374 
3375 const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
3376 {
3377 	switch (reason) {
3378 	case KMSG_DUMP_PANIC:
3379 		return "Panic";
3380 	case KMSG_DUMP_OOPS:
3381 		return "Oops";
3382 	case KMSG_DUMP_EMERG:
3383 		return "Emergency";
3384 	case KMSG_DUMP_SHUTDOWN:
3385 		return "Shutdown";
3386 	default:
3387 		return "Unknown";
3388 	}
3389 }
3390 EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
3391 
3392 /**
3393  * kmsg_dump - dump kernel log to kernel message dumpers.
3394  * @reason: the reason (oops, panic etc) for dumping
3395  *
3396  * Call each of the registered dumper's dump() callback, which can
3397  * retrieve the kmsg records with kmsg_dump_get_line() or
3398  * kmsg_dump_get_buffer().
3399  */
3400 void kmsg_dump(enum kmsg_dump_reason reason)
3401 {
3402 	struct kmsg_dumper *dumper;
3403 
3404 	rcu_read_lock();
3405 	list_for_each_entry_rcu(dumper, &dump_list, list) {
3406 		enum kmsg_dump_reason max_reason = dumper->max_reason;
3407 
3408 		/*
3409 		 * If client has not provided a specific max_reason, default
3410 		 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
3411 		 */
3412 		if (max_reason == KMSG_DUMP_UNDEF) {
3413 			max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
3414 							KMSG_DUMP_OOPS;
3415 		}
3416 		if (reason > max_reason)
3417 			continue;
3418 
3419 		/* invoke dumper which will iterate over records */
3420 		dumper->dump(dumper, reason);
3421 	}
3422 	rcu_read_unlock();
3423 }
3424 
3425 /**
3426  * kmsg_dump_get_line - retrieve one kmsg log line
3427  * @iter: kmsg dump iterator
3428  * @syslog: include the "<4>" prefixes
3429  * @line: buffer to copy the line to
3430  * @size: maximum size of the buffer
3431  * @len: length of line placed into buffer
3432  *
3433  * Start at the beginning of the kmsg buffer, with the oldest kmsg
3434  * record, and copy one record into the provided buffer.
3435  *
3436  * Consecutive calls will return the next available record moving
3437  * towards the end of the buffer with the youngest messages.
3438  *
3439  * A return value of FALSE indicates that there are no more records to
3440  * read.
3441  */
3442 bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
3443 			char *line, size_t size, size_t *len)
3444 {
3445 	u64 min_seq = latched_seq_read_nolock(&clear_seq);
3446 	struct printk_info info;
3447 	unsigned int line_count;
3448 	struct printk_record r;
3449 	size_t l = 0;
3450 	bool ret = false;
3451 
3452 	if (iter->cur_seq < min_seq)
3453 		iter->cur_seq = min_seq;
3454 
3455 	prb_rec_init_rd(&r, &info, line, size);
3456 
3457 	/* Read text or count text lines? */
3458 	if (line) {
3459 		if (!prb_read_valid(prb, iter->cur_seq, &r))
3460 			goto out;
3461 		l = record_print_text(&r, syslog, printk_time);
3462 	} else {
3463 		if (!prb_read_valid_info(prb, iter->cur_seq,
3464 					 &info, &line_count)) {
3465 			goto out;
3466 		}
3467 		l = get_record_print_text_size(&info, line_count, syslog,
3468 					       printk_time);
3469 
3470 	}
3471 
3472 	iter->cur_seq = r.info->seq + 1;
3473 	ret = true;
3474 out:
3475 	if (len)
3476 		*len = l;
3477 	return ret;
3478 }
3479 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3480 
3481 /**
3482  * kmsg_dump_get_buffer - copy kmsg log lines
3483  * @iter: kmsg dump iterator
3484  * @syslog: include the "<4>" prefixes
3485  * @buf: buffer to copy the line to
3486  * @size: maximum size of the buffer
3487  * @len_out: length of line placed into buffer
3488  *
3489  * Start at the end of the kmsg buffer and fill the provided buffer
3490  * with as many of the *youngest* kmsg records that fit into it.
3491  * If the buffer is large enough, all available kmsg records will be
3492  * copied with a single call.
3493  *
3494  * Consecutive calls will fill the buffer with the next block of
3495  * available older records, not including the earlier retrieved ones.
3496  *
3497  * A return value of FALSE indicates that there are no more records to
3498  * read.
3499  */
3500 bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
3501 			  char *buf, size_t size, size_t *len_out)
3502 {
3503 	u64 min_seq = latched_seq_read_nolock(&clear_seq);
3504 	struct printk_info info;
3505 	struct printk_record r;
3506 	u64 seq;
3507 	u64 next_seq;
3508 	size_t len = 0;
3509 	bool ret = false;
3510 	bool time = printk_time;
3511 
3512 	if (!buf || !size)
3513 		goto out;
3514 
3515 	if (iter->cur_seq < min_seq)
3516 		iter->cur_seq = min_seq;
3517 
3518 	if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
3519 		if (info.seq != iter->cur_seq) {
3520 			/* messages are gone, move to first available one */
3521 			iter->cur_seq = info.seq;
3522 		}
3523 	}
3524 
3525 	/* last entry */
3526 	if (iter->cur_seq >= iter->next_seq)
3527 		goto out;
3528 
3529 	/*
3530 	 * Find first record that fits, including all following records,
3531 	 * into the user-provided buffer for this dump. Pass in size-1
3532 	 * because this function (by way of record_print_text()) will
3533 	 * not write more than size-1 bytes of text into @buf.
3534 	 */
3535 	seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
3536 				     size - 1, syslog, time);
3537 
3538 	/*
3539 	 * Next kmsg_dump_get_buffer() invocation will dump block of
3540 	 * older records stored right before this one.
3541 	 */
3542 	next_seq = seq;
3543 
3544 	prb_rec_init_rd(&r, &info, buf, size);
3545 
3546 	len = 0;
3547 	prb_for_each_record(seq, prb, seq, &r) {
3548 		if (r.info->seq >= iter->next_seq)
3549 			break;
3550 
3551 		len += record_print_text(&r, syslog, time);
3552 
3553 		/* Adjust record to store to remaining buffer space. */
3554 		prb_rec_init_rd(&r, &info, buf + len, size - len);
3555 	}
3556 
3557 	iter->next_seq = next_seq;
3558 	ret = true;
3559 out:
3560 	if (len_out)
3561 		*len_out = len;
3562 	return ret;
3563 }
3564 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3565 
3566 /**
3567  * kmsg_dump_rewind - reset the iterator
3568  * @iter: kmsg dump iterator
3569  *
3570  * Reset the dumper's iterator so that kmsg_dump_get_line() and
3571  * kmsg_dump_get_buffer() can be called again and used multiple
3572  * times within the same dumper.dump() callback.
3573  */
3574 void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
3575 {
3576 	iter->cur_seq = latched_seq_read_nolock(&clear_seq);
3577 	iter->next_seq = prb_next_seq(prb);
3578 }
3579 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3580 
3581 #endif
3582 
3583 #ifdef CONFIG_SMP
3584 static atomic_t printk_cpulock_owner = ATOMIC_INIT(-1);
3585 static atomic_t printk_cpulock_nested = ATOMIC_INIT(0);
3586 
3587 /**
3588  * __printk_wait_on_cpu_lock() - Busy wait until the printk cpu-reentrant
3589  *                               spinning lock is not owned by any CPU.
3590  *
3591  * Context: Any context.
3592  */
3593 void __printk_wait_on_cpu_lock(void)
3594 {
3595 	do {
3596 		cpu_relax();
3597 	} while (atomic_read(&printk_cpulock_owner) != -1);
3598 }
3599 EXPORT_SYMBOL(__printk_wait_on_cpu_lock);
3600 
3601 /**
3602  * __printk_cpu_trylock() - Try to acquire the printk cpu-reentrant
3603  *                          spinning lock.
3604  *
3605  * If no processor has the lock, the calling processor takes the lock and
3606  * becomes the owner. If the calling processor is already the owner of the
3607  * lock, this function succeeds immediately.
3608  *
3609  * Context: Any context. Expects interrupts to be disabled.
3610  * Return: 1 on success, otherwise 0.
3611  */
3612 int __printk_cpu_trylock(void)
3613 {
3614 	int cpu;
3615 	int old;
3616 
3617 	cpu = smp_processor_id();
3618 
3619 	/*
3620 	 * Guarantee loads and stores from this CPU when it is the lock owner
3621 	 * are _not_ visible to the previous lock owner. This pairs with
3622 	 * __printk_cpu_unlock:B.
3623 	 *
3624 	 * Memory barrier involvement:
3625 	 *
3626 	 * If __printk_cpu_trylock:A reads from __printk_cpu_unlock:B, then
3627 	 * __printk_cpu_unlock:A can never read from __printk_cpu_trylock:B.
3628 	 *
3629 	 * Relies on:
3630 	 *
3631 	 * RELEASE from __printk_cpu_unlock:A to __printk_cpu_unlock:B
3632 	 * of the previous CPU
3633 	 *    matching
3634 	 * ACQUIRE from __printk_cpu_trylock:A to __printk_cpu_trylock:B
3635 	 * of this CPU
3636 	 */
3637 	old = atomic_cmpxchg_acquire(&printk_cpulock_owner, -1,
3638 				     cpu); /* LMM(__printk_cpu_trylock:A) */
3639 	if (old == -1) {
3640 		/*
3641 		 * This CPU is now the owner and begins loading/storing
3642 		 * data: LMM(__printk_cpu_trylock:B)
3643 		 */
3644 		return 1;
3645 
3646 	} else if (old == cpu) {
3647 		/* This CPU is already the owner. */
3648 		atomic_inc(&printk_cpulock_nested);
3649 		return 1;
3650 	}
3651 
3652 	return 0;
3653 }
3654 EXPORT_SYMBOL(__printk_cpu_trylock);
3655 
3656 /**
3657  * __printk_cpu_unlock() - Release the printk cpu-reentrant spinning lock.
3658  *
3659  * The calling processor must be the owner of the lock.
3660  *
3661  * Context: Any context. Expects interrupts to be disabled.
3662  */
3663 void __printk_cpu_unlock(void)
3664 {
3665 	if (atomic_read(&printk_cpulock_nested)) {
3666 		atomic_dec(&printk_cpulock_nested);
3667 		return;
3668 	}
3669 
3670 	/*
3671 	 * This CPU is finished loading/storing data:
3672 	 * LMM(__printk_cpu_unlock:A)
3673 	 */
3674 
3675 	/*
3676 	 * Guarantee loads and stores from this CPU when it was the
3677 	 * lock owner are visible to the next lock owner. This pairs
3678 	 * with __printk_cpu_trylock:A.
3679 	 *
3680 	 * Memory barrier involvement:
3681 	 *
3682 	 * If __printk_cpu_trylock:A reads from __printk_cpu_unlock:B,
3683 	 * then __printk_cpu_trylock:B reads from __printk_cpu_unlock:A.
3684 	 *
3685 	 * Relies on:
3686 	 *
3687 	 * RELEASE from __printk_cpu_unlock:A to __printk_cpu_unlock:B
3688 	 * of this CPU
3689 	 *    matching
3690 	 * ACQUIRE from __printk_cpu_trylock:A to __printk_cpu_trylock:B
3691 	 * of the next CPU
3692 	 */
3693 	atomic_set_release(&printk_cpulock_owner,
3694 			   -1); /* LMM(__printk_cpu_unlock:B) */
3695 }
3696 EXPORT_SYMBOL(__printk_cpu_unlock);
3697 #endif /* CONFIG_SMP */
3698