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 /*
97  * During panic, heavy printk by other CPUs can delay the
98  * panic and risk deadlock on console resources.
99  */
100 static int __read_mostly suppress_panic_printk;
101 
102 #ifdef CONFIG_LOCKDEP
103 static struct lockdep_map console_lock_dep_map = {
104 	.name = "console_lock"
105 };
106 #endif
107 
108 enum devkmsg_log_bits {
109 	__DEVKMSG_LOG_BIT_ON = 0,
110 	__DEVKMSG_LOG_BIT_OFF,
111 	__DEVKMSG_LOG_BIT_LOCK,
112 };
113 
114 enum devkmsg_log_masks {
115 	DEVKMSG_LOG_MASK_ON             = BIT(__DEVKMSG_LOG_BIT_ON),
116 	DEVKMSG_LOG_MASK_OFF            = BIT(__DEVKMSG_LOG_BIT_OFF),
117 	DEVKMSG_LOG_MASK_LOCK           = BIT(__DEVKMSG_LOG_BIT_LOCK),
118 };
119 
120 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
121 #define DEVKMSG_LOG_MASK_DEFAULT	0
122 
123 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
124 
__control_devkmsg(char * str)125 static int __control_devkmsg(char *str)
126 {
127 	size_t len;
128 
129 	if (!str)
130 		return -EINVAL;
131 
132 	len = str_has_prefix(str, "on");
133 	if (len) {
134 		devkmsg_log = DEVKMSG_LOG_MASK_ON;
135 		return len;
136 	}
137 
138 	len = str_has_prefix(str, "off");
139 	if (len) {
140 		devkmsg_log = DEVKMSG_LOG_MASK_OFF;
141 		return len;
142 	}
143 
144 	len = str_has_prefix(str, "ratelimit");
145 	if (len) {
146 		devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
147 		return len;
148 	}
149 
150 	return -EINVAL;
151 }
152 
control_devkmsg(char * str)153 static int __init control_devkmsg(char *str)
154 {
155 	if (__control_devkmsg(str) < 0) {
156 		pr_warn("printk.devkmsg: bad option string '%s'\n", str);
157 		return 1;
158 	}
159 
160 	/*
161 	 * Set sysctl string accordingly:
162 	 */
163 	if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
164 		strcpy(devkmsg_log_str, "on");
165 	else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
166 		strcpy(devkmsg_log_str, "off");
167 	/* else "ratelimit" which is set by default. */
168 
169 	/*
170 	 * Sysctl cannot change it anymore. The kernel command line setting of
171 	 * this parameter is to force the setting to be permanent throughout the
172 	 * runtime of the system. This is a precation measure against userspace
173 	 * trying to be a smarta** and attempting to change it up on us.
174 	 */
175 	devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
176 
177 	return 1;
178 }
179 __setup("printk.devkmsg=", control_devkmsg);
180 
181 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
182 #if defined(CONFIG_PRINTK) && defined(CONFIG_SYSCTL)
devkmsg_sysctl_set_loglvl(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)183 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
184 			      void *buffer, size_t *lenp, loff_t *ppos)
185 {
186 	char old_str[DEVKMSG_STR_MAX_SIZE];
187 	unsigned int old;
188 	int err;
189 
190 	if (write) {
191 		if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
192 			return -EINVAL;
193 
194 		old = devkmsg_log;
195 		strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
196 	}
197 
198 	err = proc_dostring(table, write, buffer, lenp, ppos);
199 	if (err)
200 		return err;
201 
202 	if (write) {
203 		err = __control_devkmsg(devkmsg_log_str);
204 
205 		/*
206 		 * Do not accept an unknown string OR a known string with
207 		 * trailing crap...
208 		 */
209 		if (err < 0 || (err + 1 != *lenp)) {
210 
211 			/* ... and restore old setting. */
212 			devkmsg_log = old;
213 			strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
214 
215 			return -EINVAL;
216 		}
217 	}
218 
219 	return 0;
220 }
221 #endif /* CONFIG_PRINTK && CONFIG_SYSCTL */
222 
223 /*
224  * Helper macros to handle lockdep when locking/unlocking console_sem. We use
225  * macros instead of functions so that _RET_IP_ contains useful information.
226  */
227 #define down_console_sem() do { \
228 	down(&console_sem);\
229 	mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
230 } while (0)
231 
__down_trylock_console_sem(unsigned long ip)232 static int __down_trylock_console_sem(unsigned long ip)
233 {
234 	int lock_failed;
235 	unsigned long flags;
236 
237 	/*
238 	 * Here and in __up_console_sem() we need to be in safe mode,
239 	 * because spindump/WARN/etc from under console ->lock will
240 	 * deadlock in printk()->down_trylock_console_sem() otherwise.
241 	 */
242 	printk_safe_enter_irqsave(flags);
243 	lock_failed = down_trylock(&console_sem);
244 	printk_safe_exit_irqrestore(flags);
245 
246 	if (lock_failed)
247 		return 1;
248 	mutex_acquire(&console_lock_dep_map, 0, 1, ip);
249 	return 0;
250 }
251 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
252 
__up_console_sem(unsigned long ip)253 static void __up_console_sem(unsigned long ip)
254 {
255 	unsigned long flags;
256 
257 	mutex_release(&console_lock_dep_map, ip);
258 
259 	printk_safe_enter_irqsave(flags);
260 	up(&console_sem);
261 	printk_safe_exit_irqrestore(flags);
262 }
263 #define up_console_sem() __up_console_sem(_RET_IP_)
264 
panic_in_progress(void)265 static bool panic_in_progress(void)
266 {
267 	return unlikely(atomic_read(&panic_cpu) != PANIC_CPU_INVALID);
268 }
269 
270 /*
271  * This is used for debugging the mess that is the VT code by
272  * keeping track if we have the console semaphore held. It's
273  * definitely not the perfect debug tool (we don't know if _WE_
274  * hold it and are racing, but it helps tracking those weird code
275  * paths in the console code where we end up in places I want
276  * locked without the console semaphore held).
277  */
278 static int console_locked, console_suspended;
279 
280 /*
281  *	Array of consoles built from command line options (console=)
282  */
283 
284 #define MAX_CMDLINECONSOLES 8
285 
286 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
287 
288 static int preferred_console = -1;
289 int console_set_on_cmdline;
290 EXPORT_SYMBOL(console_set_on_cmdline);
291 
292 /* Flag: console code may call schedule() */
293 static int console_may_schedule;
294 
295 enum con_msg_format_flags {
296 	MSG_FORMAT_DEFAULT	= 0,
297 	MSG_FORMAT_SYSLOG	= (1 << 0),
298 };
299 
300 static int console_msg_format = MSG_FORMAT_DEFAULT;
301 
302 /*
303  * The printk log buffer consists of a sequenced collection of records, each
304  * containing variable length message text. Every record also contains its
305  * own meta-data (@info).
306  *
307  * Every record meta-data carries the timestamp in microseconds, as well as
308  * the standard userspace syslog level and syslog facility. The usual kernel
309  * messages use LOG_KERN; userspace-injected messages always carry a matching
310  * syslog facility, by default LOG_USER. The origin of every message can be
311  * reliably determined that way.
312  *
313  * The human readable log message of a record is available in @text, the
314  * length of the message text in @text_len. The stored message is not
315  * terminated.
316  *
317  * Optionally, a record can carry a dictionary of properties (key/value
318  * pairs), to provide userspace with a machine-readable message context.
319  *
320  * Examples for well-defined, commonly used property names are:
321  *   DEVICE=b12:8               device identifier
322  *                                b12:8         block dev_t
323  *                                c127:3        char dev_t
324  *                                n8            netdev ifindex
325  *                                +sound:card0  subsystem:devname
326  *   SUBSYSTEM=pci              driver-core subsystem name
327  *
328  * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
329  * and values are terminated by a '\0' character.
330  *
331  * Example of record values:
332  *   record.text_buf                = "it's a line" (unterminated)
333  *   record.info.seq                = 56
334  *   record.info.ts_nsec            = 36863
335  *   record.info.text_len           = 11
336  *   record.info.facility           = 0 (LOG_KERN)
337  *   record.info.flags              = 0
338  *   record.info.level              = 3 (LOG_ERR)
339  *   record.info.caller_id          = 299 (task 299)
340  *   record.info.dev_info.subsystem = "pci" (terminated)
341  *   record.info.dev_info.device    = "+pci:0000:00:01.0" (terminated)
342  *
343  * The 'struct printk_info' buffer must never be directly exported to
344  * userspace, it is a kernel-private implementation detail that might
345  * need to be changed in the future, when the requirements change.
346  *
347  * /dev/kmsg exports the structured data in the following line format:
348  *   "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
349  *
350  * Users of the export format should ignore possible additional values
351  * separated by ',', and find the message after the ';' character.
352  *
353  * The optional key/value pairs are attached as continuation lines starting
354  * with a space character and terminated by a newline. All possible
355  * non-prinatable characters are escaped in the "\xff" notation.
356  */
357 
358 /* syslog_lock protects syslog_* variables and write access to clear_seq. */
359 static DEFINE_MUTEX(syslog_lock);
360 
361 #ifdef CONFIG_PRINTK
362 DECLARE_WAIT_QUEUE_HEAD(log_wait);
363 /* All 3 protected by @syslog_lock. */
364 /* the next printk record to read by syslog(READ) or /proc/kmsg */
365 static u64 syslog_seq;
366 static size_t syslog_partial;
367 static bool syslog_time;
368 
369 struct latched_seq {
370 	seqcount_latch_t	latch;
371 	u64			val[2];
372 };
373 
374 /*
375  * The next printk record to read after the last 'clear' command. There are
376  * two copies (updated with seqcount_latch) so that reads can locklessly
377  * access a valid value. Writers are synchronized by @syslog_lock.
378  */
379 static struct latched_seq clear_seq = {
380 	.latch		= SEQCNT_LATCH_ZERO(clear_seq.latch),
381 	.val[0]		= 0,
382 	.val[1]		= 0,
383 };
384 
385 #ifdef CONFIG_PRINTK_CALLER
386 #define PREFIX_MAX		48
387 #else
388 #define PREFIX_MAX		32
389 #endif
390 
391 /* the maximum size of a formatted record (i.e. with prefix added per line) */
392 #define CONSOLE_LOG_MAX		1024
393 
394 /* the maximum size for a dropped text message */
395 #define DROPPED_TEXT_MAX	64
396 
397 /* the maximum size allowed to be reserved for a record */
398 #define LOG_LINE_MAX		(CONSOLE_LOG_MAX - PREFIX_MAX)
399 
400 #define LOG_LEVEL(v)		((v) & 0x07)
401 #define LOG_FACILITY(v)		((v) >> 3 & 0xff)
402 
403 /* record buffer */
404 #define LOG_ALIGN __alignof__(unsigned long)
405 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
406 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
407 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
408 static char *log_buf = __log_buf;
409 static u32 log_buf_len = __LOG_BUF_LEN;
410 
411 /*
412  * Define the average message size. This only affects the number of
413  * descriptors that will be available. Underestimating is better than
414  * overestimating (too many available descriptors is better than not enough).
415  */
416 #define PRB_AVGBITS 5	/* 32 character average length */
417 
418 #if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
419 #error CONFIG_LOG_BUF_SHIFT value too small.
420 #endif
421 _DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
422 		 PRB_AVGBITS, &__log_buf[0]);
423 
424 static struct printk_ringbuffer printk_rb_dynamic;
425 
426 static struct printk_ringbuffer *prb = &printk_rb_static;
427 
428 /*
429  * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
430  * per_cpu_areas are initialised. This variable is set to true when
431  * it's safe to access per-CPU data.
432  */
433 static bool __printk_percpu_data_ready __ro_after_init;
434 
printk_percpu_data_ready(void)435 bool printk_percpu_data_ready(void)
436 {
437 	return __printk_percpu_data_ready;
438 }
439 
440 /* Must be called under syslog_lock. */
latched_seq_write(struct latched_seq * ls,u64 val)441 static void latched_seq_write(struct latched_seq *ls, u64 val)
442 {
443 	raw_write_seqcount_latch(&ls->latch);
444 	ls->val[0] = val;
445 	raw_write_seqcount_latch(&ls->latch);
446 	ls->val[1] = val;
447 }
448 
449 /* Can be called from any context. */
latched_seq_read_nolock(struct latched_seq * ls)450 static u64 latched_seq_read_nolock(struct latched_seq *ls)
451 {
452 	unsigned int seq;
453 	unsigned int idx;
454 	u64 val;
455 
456 	do {
457 		seq = raw_read_seqcount_latch(&ls->latch);
458 		idx = seq & 0x1;
459 		val = ls->val[idx];
460 	} while (read_seqcount_latch_retry(&ls->latch, seq));
461 
462 	return val;
463 }
464 
465 /* Return log buffer address */
log_buf_addr_get(void)466 char *log_buf_addr_get(void)
467 {
468 	return log_buf;
469 }
470 
471 /* Return log buffer size */
log_buf_len_get(void)472 u32 log_buf_len_get(void)
473 {
474 	return log_buf_len;
475 }
476 
477 /*
478  * Define how much of the log buffer we could take at maximum. The value
479  * must be greater than two. Note that only half of the buffer is available
480  * when the index points to the middle.
481  */
482 #define MAX_LOG_TAKE_PART 4
483 static const char trunc_msg[] = "<truncated>";
484 
truncate_msg(u16 * text_len,u16 * trunc_msg_len)485 static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
486 {
487 	/*
488 	 * The message should not take the whole buffer. Otherwise, it might
489 	 * get removed too soon.
490 	 */
491 	u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
492 
493 	if (*text_len > max_text_len)
494 		*text_len = max_text_len;
495 
496 	/* enable the warning message (if there is room) */
497 	*trunc_msg_len = strlen(trunc_msg);
498 	if (*text_len >= *trunc_msg_len)
499 		*text_len -= *trunc_msg_len;
500 	else
501 		*trunc_msg_len = 0;
502 }
503 
504 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
505 
syslog_action_restricted(int type)506 static int syslog_action_restricted(int type)
507 {
508 	if (dmesg_restrict)
509 		return 1;
510 	/*
511 	 * Unless restricted, we allow "read all" and "get buffer size"
512 	 * for everybody.
513 	 */
514 	return type != SYSLOG_ACTION_READ_ALL &&
515 	       type != SYSLOG_ACTION_SIZE_BUFFER;
516 }
517 
check_syslog_permissions(int type,int source)518 static int check_syslog_permissions(int type, int source)
519 {
520 	/*
521 	 * If this is from /proc/kmsg and we've already opened it, then we've
522 	 * already done the capabilities checks at open time.
523 	 */
524 	if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
525 		goto ok;
526 
527 	if (syslog_action_restricted(type)) {
528 		if (capable(CAP_SYSLOG))
529 			goto ok;
530 		/*
531 		 * For historical reasons, accept CAP_SYS_ADMIN too, with
532 		 * a warning.
533 		 */
534 		if (capable(CAP_SYS_ADMIN)) {
535 			pr_warn_once("%s (%d): Attempt to access syslog with "
536 				     "CAP_SYS_ADMIN but no CAP_SYSLOG "
537 				     "(deprecated).\n",
538 				 current->comm, task_pid_nr(current));
539 			goto ok;
540 		}
541 		return -EPERM;
542 	}
543 ok:
544 	return security_syslog(type);
545 }
546 
append_char(char ** pp,char * e,char c)547 static void append_char(char **pp, char *e, char c)
548 {
549 	if (*pp < e)
550 		*(*pp)++ = c;
551 }
552 
info_print_ext_header(char * buf,size_t size,struct printk_info * info)553 static ssize_t info_print_ext_header(char *buf, size_t size,
554 				     struct printk_info *info)
555 {
556 	u64 ts_usec = info->ts_nsec;
557 	char caller[20];
558 #ifdef CONFIG_PRINTK_CALLER
559 	u32 id = info->caller_id;
560 
561 	snprintf(caller, sizeof(caller), ",caller=%c%u",
562 		 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
563 #else
564 	caller[0] = '\0';
565 #endif
566 
567 	do_div(ts_usec, 1000);
568 
569 	return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
570 			 (info->facility << 3) | info->level, info->seq,
571 			 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
572 }
573 
msg_add_ext_text(char * buf,size_t size,const char * text,size_t text_len,unsigned char endc)574 static ssize_t msg_add_ext_text(char *buf, size_t size,
575 				const char *text, size_t text_len,
576 				unsigned char endc)
577 {
578 	char *p = buf, *e = buf + size;
579 	size_t i;
580 
581 	/* escape non-printable characters */
582 	for (i = 0; i < text_len; i++) {
583 		unsigned char c = text[i];
584 
585 		if (c < ' ' || c >= 127 || c == '\\')
586 			p += scnprintf(p, e - p, "\\x%02x", c);
587 		else
588 			append_char(&p, e, c);
589 	}
590 	append_char(&p, e, endc);
591 
592 	return p - buf;
593 }
594 
msg_add_dict_text(char * buf,size_t size,const char * key,const char * val)595 static ssize_t msg_add_dict_text(char *buf, size_t size,
596 				 const char *key, const char *val)
597 {
598 	size_t val_len = strlen(val);
599 	ssize_t len;
600 
601 	if (!val_len)
602 		return 0;
603 
604 	len = msg_add_ext_text(buf, size, "", 0, ' ');	/* dict prefix */
605 	len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
606 	len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
607 
608 	return len;
609 }
610 
msg_print_ext_body(char * buf,size_t size,char * text,size_t text_len,struct dev_printk_info * dev_info)611 static ssize_t msg_print_ext_body(char *buf, size_t size,
612 				  char *text, size_t text_len,
613 				  struct dev_printk_info *dev_info)
614 {
615 	ssize_t len;
616 
617 	len = msg_add_ext_text(buf, size, text, text_len, '\n');
618 
619 	if (!dev_info)
620 		goto out;
621 
622 	len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
623 				 dev_info->subsystem);
624 	len += msg_add_dict_text(buf + len, size - len, "DEVICE",
625 				 dev_info->device);
626 out:
627 	return len;
628 }
629 
630 /* /dev/kmsg - userspace message inject/listen interface */
631 struct devkmsg_user {
632 	atomic64_t seq;
633 	struct ratelimit_state rs;
634 	struct mutex lock;
635 	char buf[CONSOLE_EXT_LOG_MAX];
636 
637 	struct printk_info info;
638 	char text_buf[CONSOLE_EXT_LOG_MAX];
639 	struct printk_record record;
640 };
641 
642 static __printf(3, 4) __cold
devkmsg_emit(int facility,int level,const char * fmt,...)643 int devkmsg_emit(int facility, int level, const char *fmt, ...)
644 {
645 	va_list args;
646 	int r;
647 
648 	va_start(args, fmt);
649 	r = vprintk_emit(facility, level, NULL, fmt, args);
650 	va_end(args);
651 
652 	return r;
653 }
654 
devkmsg_write(struct kiocb * iocb,struct iov_iter * from)655 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
656 {
657 	char *buf, *line;
658 	int level = default_message_loglevel;
659 	int facility = 1;	/* LOG_USER */
660 	struct file *file = iocb->ki_filp;
661 	struct devkmsg_user *user = file->private_data;
662 	size_t len = iov_iter_count(from);
663 	ssize_t ret = len;
664 
665 	if (!user || len > LOG_LINE_MAX)
666 		return -EINVAL;
667 
668 	/* Ignore when user logging is disabled. */
669 	if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
670 		return len;
671 
672 	/* Ratelimit when not explicitly enabled. */
673 	if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
674 		if (!___ratelimit(&user->rs, current->comm))
675 			return ret;
676 	}
677 
678 	buf = kmalloc(len+1, GFP_KERNEL);
679 	if (buf == NULL)
680 		return -ENOMEM;
681 
682 	buf[len] = '\0';
683 	if (!copy_from_iter_full(buf, len, from)) {
684 		kfree(buf);
685 		return -EFAULT;
686 	}
687 
688 	/*
689 	 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
690 	 * the decimal value represents 32bit, the lower 3 bit are the log
691 	 * level, the rest are the log facility.
692 	 *
693 	 * If no prefix or no userspace facility is specified, we
694 	 * enforce LOG_USER, to be able to reliably distinguish
695 	 * kernel-generated messages from userspace-injected ones.
696 	 */
697 	line = buf;
698 	if (line[0] == '<') {
699 		char *endp = NULL;
700 		unsigned int u;
701 
702 		u = simple_strtoul(line + 1, &endp, 10);
703 		if (endp && endp[0] == '>') {
704 			level = LOG_LEVEL(u);
705 			if (LOG_FACILITY(u) != 0)
706 				facility = LOG_FACILITY(u);
707 			endp++;
708 			line = endp;
709 		}
710 	}
711 
712 	devkmsg_emit(facility, level, "%s", line);
713 	kfree(buf);
714 	return ret;
715 }
716 
devkmsg_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)717 static ssize_t devkmsg_read(struct file *file, char __user *buf,
718 			    size_t count, loff_t *ppos)
719 {
720 	struct devkmsg_user *user = file->private_data;
721 	struct printk_record *r = &user->record;
722 	size_t len;
723 	ssize_t ret;
724 
725 	if (!user)
726 		return -EBADF;
727 
728 	ret = mutex_lock_interruptible(&user->lock);
729 	if (ret)
730 		return ret;
731 
732 	if (!prb_read_valid(prb, atomic64_read(&user->seq), r)) {
733 		if (file->f_flags & O_NONBLOCK) {
734 			ret = -EAGAIN;
735 			goto out;
736 		}
737 
738 		/*
739 		 * Guarantee this task is visible on the waitqueue before
740 		 * checking the wake condition.
741 		 *
742 		 * The full memory barrier within set_current_state() of
743 		 * prepare_to_wait_event() pairs with the full memory barrier
744 		 * within wq_has_sleeper().
745 		 *
746 		 * This pairs with __wake_up_klogd:A.
747 		 */
748 		ret = wait_event_interruptible(log_wait,
749 				prb_read_valid(prb,
750 					atomic64_read(&user->seq), r)); /* LMM(devkmsg_read:A) */
751 		if (ret)
752 			goto out;
753 	}
754 
755 	if (r->info->seq != atomic64_read(&user->seq)) {
756 		/* our last seen message is gone, return error and reset */
757 		atomic64_set(&user->seq, r->info->seq);
758 		ret = -EPIPE;
759 		goto out;
760 	}
761 
762 	len = info_print_ext_header(user->buf, sizeof(user->buf), r->info);
763 	len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
764 				  &r->text_buf[0], r->info->text_len,
765 				  &r->info->dev_info);
766 
767 	atomic64_set(&user->seq, r->info->seq + 1);
768 
769 	if (len > count) {
770 		ret = -EINVAL;
771 		goto out;
772 	}
773 
774 	if (copy_to_user(buf, user->buf, len)) {
775 		ret = -EFAULT;
776 		goto out;
777 	}
778 	ret = len;
779 out:
780 	mutex_unlock(&user->lock);
781 	return ret;
782 }
783 
784 /*
785  * Be careful when modifying this function!!!
786  *
787  * Only few operations are supported because the device works only with the
788  * entire variable length messages (records). Non-standard values are
789  * returned in the other cases and has been this way for quite some time.
790  * User space applications might depend on this behavior.
791  */
devkmsg_llseek(struct file * file,loff_t offset,int whence)792 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
793 {
794 	struct devkmsg_user *user = file->private_data;
795 	loff_t ret = 0;
796 
797 	if (!user)
798 		return -EBADF;
799 	if (offset)
800 		return -ESPIPE;
801 
802 	switch (whence) {
803 	case SEEK_SET:
804 		/* the first record */
805 		atomic64_set(&user->seq, prb_first_valid_seq(prb));
806 		break;
807 	case SEEK_DATA:
808 		/*
809 		 * The first record after the last SYSLOG_ACTION_CLEAR,
810 		 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
811 		 * changes no global state, and does not clear anything.
812 		 */
813 		atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
814 		break;
815 	case SEEK_END:
816 		/* after the last record */
817 		atomic64_set(&user->seq, prb_next_seq(prb));
818 		break;
819 	default:
820 		ret = -EINVAL;
821 	}
822 	return ret;
823 }
824 
devkmsg_poll(struct file * file,poll_table * wait)825 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
826 {
827 	struct devkmsg_user *user = file->private_data;
828 	struct printk_info info;
829 	__poll_t ret = 0;
830 
831 	if (!user)
832 		return EPOLLERR|EPOLLNVAL;
833 
834 	poll_wait(file, &log_wait, wait);
835 
836 	if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
837 		/* return error when data has vanished underneath us */
838 		if (info.seq != atomic64_read(&user->seq))
839 			ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
840 		else
841 			ret = EPOLLIN|EPOLLRDNORM;
842 	}
843 
844 	return ret;
845 }
846 
devkmsg_open(struct inode * inode,struct file * file)847 static int devkmsg_open(struct inode *inode, struct file *file)
848 {
849 	struct devkmsg_user *user;
850 	int err;
851 
852 	if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
853 		return -EPERM;
854 
855 	/* write-only does not need any file context */
856 	if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
857 		err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
858 					       SYSLOG_FROM_READER);
859 		if (err)
860 			return err;
861 	}
862 
863 	user = kvmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
864 	if (!user)
865 		return -ENOMEM;
866 
867 	ratelimit_default_init(&user->rs);
868 	ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
869 
870 	mutex_init(&user->lock);
871 
872 	prb_rec_init_rd(&user->record, &user->info,
873 			&user->text_buf[0], sizeof(user->text_buf));
874 
875 	atomic64_set(&user->seq, prb_first_valid_seq(prb));
876 
877 	file->private_data = user;
878 	return 0;
879 }
880 
devkmsg_release(struct inode * inode,struct file * file)881 static int devkmsg_release(struct inode *inode, struct file *file)
882 {
883 	struct devkmsg_user *user = file->private_data;
884 
885 	if (!user)
886 		return 0;
887 
888 	ratelimit_state_exit(&user->rs);
889 
890 	mutex_destroy(&user->lock);
891 	kvfree(user);
892 	return 0;
893 }
894 
895 const struct file_operations kmsg_fops = {
896 	.open = devkmsg_open,
897 	.read = devkmsg_read,
898 	.write_iter = devkmsg_write,
899 	.llseek = devkmsg_llseek,
900 	.poll = devkmsg_poll,
901 	.release = devkmsg_release,
902 };
903 
904 #ifdef CONFIG_CRASH_CORE
905 /*
906  * This appends the listed symbols to /proc/vmcore
907  *
908  * /proc/vmcore is used by various utilities, like crash and makedumpfile to
909  * obtain access to symbols that are otherwise very difficult to locate.  These
910  * symbols are specifically used so that utilities can access and extract the
911  * dmesg log from a vmcore file after a crash.
912  */
log_buf_vmcoreinfo_setup(void)913 void log_buf_vmcoreinfo_setup(void)
914 {
915 	struct dev_printk_info *dev_info = NULL;
916 
917 	VMCOREINFO_SYMBOL(prb);
918 	VMCOREINFO_SYMBOL(printk_rb_static);
919 	VMCOREINFO_SYMBOL(clear_seq);
920 
921 	/*
922 	 * Export struct size and field offsets. User space tools can
923 	 * parse it and detect any changes to structure down the line.
924 	 */
925 
926 	VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
927 	VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
928 	VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
929 	VMCOREINFO_OFFSET(printk_ringbuffer, fail);
930 
931 	VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
932 	VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
933 	VMCOREINFO_OFFSET(prb_desc_ring, descs);
934 	VMCOREINFO_OFFSET(prb_desc_ring, infos);
935 	VMCOREINFO_OFFSET(prb_desc_ring, head_id);
936 	VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
937 
938 	VMCOREINFO_STRUCT_SIZE(prb_desc);
939 	VMCOREINFO_OFFSET(prb_desc, state_var);
940 	VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
941 
942 	VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
943 	VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
944 	VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
945 
946 	VMCOREINFO_STRUCT_SIZE(printk_info);
947 	VMCOREINFO_OFFSET(printk_info, seq);
948 	VMCOREINFO_OFFSET(printk_info, ts_nsec);
949 	VMCOREINFO_OFFSET(printk_info, text_len);
950 	VMCOREINFO_OFFSET(printk_info, caller_id);
951 	VMCOREINFO_OFFSET(printk_info, dev_info);
952 
953 	VMCOREINFO_STRUCT_SIZE(dev_printk_info);
954 	VMCOREINFO_OFFSET(dev_printk_info, subsystem);
955 	VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
956 	VMCOREINFO_OFFSET(dev_printk_info, device);
957 	VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
958 
959 	VMCOREINFO_STRUCT_SIZE(prb_data_ring);
960 	VMCOREINFO_OFFSET(prb_data_ring, size_bits);
961 	VMCOREINFO_OFFSET(prb_data_ring, data);
962 	VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
963 	VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
964 
965 	VMCOREINFO_SIZE(atomic_long_t);
966 	VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
967 
968 	VMCOREINFO_STRUCT_SIZE(latched_seq);
969 	VMCOREINFO_OFFSET(latched_seq, val);
970 }
971 #endif
972 
973 /* requested log_buf_len from kernel cmdline */
974 static unsigned long __initdata new_log_buf_len;
975 
976 /* we practice scaling the ring buffer by powers of 2 */
log_buf_len_update(u64 size)977 static void __init log_buf_len_update(u64 size)
978 {
979 	if (size > (u64)LOG_BUF_LEN_MAX) {
980 		size = (u64)LOG_BUF_LEN_MAX;
981 		pr_err("log_buf over 2G is not supported.\n");
982 	}
983 
984 	if (size)
985 		size = roundup_pow_of_two(size);
986 	if (size > log_buf_len)
987 		new_log_buf_len = (unsigned long)size;
988 }
989 
990 /* save requested log_buf_len since it's too early to process it */
log_buf_len_setup(char * str)991 static int __init log_buf_len_setup(char *str)
992 {
993 	u64 size;
994 
995 	if (!str)
996 		return -EINVAL;
997 
998 	size = memparse(str, &str);
999 
1000 	log_buf_len_update(size);
1001 
1002 	return 0;
1003 }
1004 early_param("log_buf_len", log_buf_len_setup);
1005 
1006 #ifdef CONFIG_SMP
1007 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1008 
log_buf_add_cpu(void)1009 static void __init log_buf_add_cpu(void)
1010 {
1011 	unsigned int cpu_extra;
1012 
1013 	/*
1014 	 * archs should set up cpu_possible_bits properly with
1015 	 * set_cpu_possible() after setup_arch() but just in
1016 	 * case lets ensure this is valid.
1017 	 */
1018 	if (num_possible_cpus() == 1)
1019 		return;
1020 
1021 	cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1022 
1023 	/* by default this will only continue through for large > 64 CPUs */
1024 	if (cpu_extra <= __LOG_BUF_LEN / 2)
1025 		return;
1026 
1027 	pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1028 		__LOG_CPU_MAX_BUF_LEN);
1029 	pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1030 		cpu_extra);
1031 	pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1032 
1033 	log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1034 }
1035 #else /* !CONFIG_SMP */
log_buf_add_cpu(void)1036 static inline void log_buf_add_cpu(void) {}
1037 #endif /* CONFIG_SMP */
1038 
set_percpu_data_ready(void)1039 static void __init set_percpu_data_ready(void)
1040 {
1041 	__printk_percpu_data_ready = true;
1042 }
1043 
add_to_rb(struct printk_ringbuffer * rb,struct printk_record * r)1044 static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1045 				     struct printk_record *r)
1046 {
1047 	struct prb_reserved_entry e;
1048 	struct printk_record dest_r;
1049 
1050 	prb_rec_init_wr(&dest_r, r->info->text_len);
1051 
1052 	if (!prb_reserve(&e, rb, &dest_r))
1053 		return 0;
1054 
1055 	memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1056 	dest_r.info->text_len = r->info->text_len;
1057 	dest_r.info->facility = r->info->facility;
1058 	dest_r.info->level = r->info->level;
1059 	dest_r.info->flags = r->info->flags;
1060 	dest_r.info->ts_nsec = r->info->ts_nsec;
1061 	dest_r.info->caller_id = r->info->caller_id;
1062 	memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1063 
1064 	prb_final_commit(&e);
1065 
1066 	return prb_record_text_space(&e);
1067 }
1068 
1069 static char setup_text_buf[LOG_LINE_MAX] __initdata;
1070 
setup_log_buf(int early)1071 void __init setup_log_buf(int early)
1072 {
1073 	struct printk_info *new_infos;
1074 	unsigned int new_descs_count;
1075 	struct prb_desc *new_descs;
1076 	struct printk_info info;
1077 	struct printk_record r;
1078 	unsigned int text_size;
1079 	size_t new_descs_size;
1080 	size_t new_infos_size;
1081 	unsigned long flags;
1082 	char *new_log_buf;
1083 	unsigned int free;
1084 	u64 seq;
1085 
1086 	/*
1087 	 * Some archs call setup_log_buf() multiple times - first is very
1088 	 * early, e.g. from setup_arch(), and second - when percpu_areas
1089 	 * are initialised.
1090 	 */
1091 	if (!early)
1092 		set_percpu_data_ready();
1093 
1094 	if (log_buf != __log_buf)
1095 		return;
1096 
1097 	if (!early && !new_log_buf_len)
1098 		log_buf_add_cpu();
1099 
1100 	if (!new_log_buf_len)
1101 		return;
1102 
1103 	new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1104 	if (new_descs_count == 0) {
1105 		pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1106 		return;
1107 	}
1108 
1109 	new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1110 	if (unlikely(!new_log_buf)) {
1111 		pr_err("log_buf_len: %lu text bytes not available\n",
1112 		       new_log_buf_len);
1113 		return;
1114 	}
1115 
1116 	new_descs_size = new_descs_count * sizeof(struct prb_desc);
1117 	new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1118 	if (unlikely(!new_descs)) {
1119 		pr_err("log_buf_len: %zu desc bytes not available\n",
1120 		       new_descs_size);
1121 		goto err_free_log_buf;
1122 	}
1123 
1124 	new_infos_size = new_descs_count * sizeof(struct printk_info);
1125 	new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1126 	if (unlikely(!new_infos)) {
1127 		pr_err("log_buf_len: %zu info bytes not available\n",
1128 		       new_infos_size);
1129 		goto err_free_descs;
1130 	}
1131 
1132 	prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1133 
1134 	prb_init(&printk_rb_dynamic,
1135 		 new_log_buf, ilog2(new_log_buf_len),
1136 		 new_descs, ilog2(new_descs_count),
1137 		 new_infos);
1138 
1139 	local_irq_save(flags);
1140 
1141 	log_buf_len = new_log_buf_len;
1142 	log_buf = new_log_buf;
1143 	new_log_buf_len = 0;
1144 
1145 	free = __LOG_BUF_LEN;
1146 	prb_for_each_record(0, &printk_rb_static, seq, &r) {
1147 		text_size = add_to_rb(&printk_rb_dynamic, &r);
1148 		if (text_size > free)
1149 			free = 0;
1150 		else
1151 			free -= text_size;
1152 	}
1153 
1154 	prb = &printk_rb_dynamic;
1155 
1156 	local_irq_restore(flags);
1157 
1158 	/*
1159 	 * Copy any remaining messages that might have appeared from
1160 	 * NMI context after copying but before switching to the
1161 	 * dynamic buffer.
1162 	 */
1163 	prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1164 		text_size = add_to_rb(&printk_rb_dynamic, &r);
1165 		if (text_size > free)
1166 			free = 0;
1167 		else
1168 			free -= text_size;
1169 	}
1170 
1171 	if (seq != prb_next_seq(&printk_rb_static)) {
1172 		pr_err("dropped %llu messages\n",
1173 		       prb_next_seq(&printk_rb_static) - seq);
1174 	}
1175 
1176 	pr_info("log_buf_len: %u bytes\n", log_buf_len);
1177 	pr_info("early log buf free: %u(%u%%)\n",
1178 		free, (free * 100) / __LOG_BUF_LEN);
1179 	return;
1180 
1181 err_free_descs:
1182 	memblock_free(new_descs, new_descs_size);
1183 err_free_log_buf:
1184 	memblock_free(new_log_buf, new_log_buf_len);
1185 }
1186 
1187 static bool __read_mostly ignore_loglevel;
1188 
ignore_loglevel_setup(char * str)1189 static int __init ignore_loglevel_setup(char *str)
1190 {
1191 	ignore_loglevel = true;
1192 	pr_info("debug: ignoring loglevel setting.\n");
1193 
1194 	return 0;
1195 }
1196 
1197 early_param("ignore_loglevel", ignore_loglevel_setup);
1198 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1199 MODULE_PARM_DESC(ignore_loglevel,
1200 		 "ignore loglevel setting (prints all kernel messages to the console)");
1201 
suppress_message_printing(int level)1202 static bool suppress_message_printing(int level)
1203 {
1204 	return (level >= console_loglevel && !ignore_loglevel);
1205 }
1206 
1207 #ifdef CONFIG_BOOT_PRINTK_DELAY
1208 
1209 static int boot_delay; /* msecs delay after each printk during bootup */
1210 static unsigned long long loops_per_msec;	/* based on boot_delay */
1211 
boot_delay_setup(char * str)1212 static int __init boot_delay_setup(char *str)
1213 {
1214 	unsigned long lpj;
1215 
1216 	lpj = preset_lpj ? preset_lpj : 1000000;	/* some guess */
1217 	loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1218 
1219 	get_option(&str, &boot_delay);
1220 	if (boot_delay > 10 * 1000)
1221 		boot_delay = 0;
1222 
1223 	pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1224 		"HZ: %d, loops_per_msec: %llu\n",
1225 		boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1226 	return 0;
1227 }
1228 early_param("boot_delay", boot_delay_setup);
1229 
boot_delay_msec(int level)1230 static void boot_delay_msec(int level)
1231 {
1232 	unsigned long long k;
1233 	unsigned long timeout;
1234 
1235 	if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1236 		|| suppress_message_printing(level)) {
1237 		return;
1238 	}
1239 
1240 	k = (unsigned long long)loops_per_msec * boot_delay;
1241 
1242 	timeout = jiffies + msecs_to_jiffies(boot_delay);
1243 	while (k) {
1244 		k--;
1245 		cpu_relax();
1246 		/*
1247 		 * use (volatile) jiffies to prevent
1248 		 * compiler reduction; loop termination via jiffies
1249 		 * is secondary and may or may not happen.
1250 		 */
1251 		if (time_after(jiffies, timeout))
1252 			break;
1253 		touch_nmi_watchdog();
1254 	}
1255 }
1256 #else
boot_delay_msec(int level)1257 static inline void boot_delay_msec(int level)
1258 {
1259 }
1260 #endif
1261 
1262 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1263 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1264 
print_syslog(unsigned int level,char * buf)1265 static size_t print_syslog(unsigned int level, char *buf)
1266 {
1267 	return sprintf(buf, "<%u>", level);
1268 }
1269 
print_time(u64 ts,char * buf)1270 static size_t print_time(u64 ts, char *buf)
1271 {
1272 	unsigned long rem_nsec = do_div(ts, 1000000000);
1273 
1274 	return sprintf(buf, "[%5lu.%06lu]",
1275 		       (unsigned long)ts, rem_nsec / 1000);
1276 }
1277 
1278 #ifdef CONFIG_PRINTK_CALLER
print_caller(u32 id,char * buf)1279 static size_t print_caller(u32 id, char *buf)
1280 {
1281 	char caller[12];
1282 
1283 	snprintf(caller, sizeof(caller), "%c%u",
1284 		 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1285 	return sprintf(buf, "[%6s]", caller);
1286 }
1287 #else
1288 #define print_caller(id, buf) 0
1289 #endif
1290 
info_print_prefix(const struct printk_info * info,bool syslog,bool time,char * buf)1291 static size_t info_print_prefix(const struct printk_info  *info, bool syslog,
1292 				bool time, char *buf)
1293 {
1294 	size_t len = 0;
1295 
1296 	if (syslog)
1297 		len = print_syslog((info->facility << 3) | info->level, buf);
1298 
1299 	if (time)
1300 		len += print_time(info->ts_nsec, buf + len);
1301 
1302 	len += print_caller(info->caller_id, buf + len);
1303 
1304 	if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1305 		buf[len++] = ' ';
1306 		buf[len] = '\0';
1307 	}
1308 
1309 	return len;
1310 }
1311 
1312 /*
1313  * Prepare the record for printing. The text is shifted within the given
1314  * buffer to avoid a need for another one. The following operations are
1315  * done:
1316  *
1317  *   - Add prefix for each line.
1318  *   - Drop truncated lines that no longer fit into the buffer.
1319  *   - Add the trailing newline that has been removed in vprintk_store().
1320  *   - Add a string terminator.
1321  *
1322  * Since the produced string is always terminated, the maximum possible
1323  * return value is @r->text_buf_size - 1;
1324  *
1325  * Return: The length of the updated/prepared text, including the added
1326  * prefixes and the newline. The terminator is not counted. The dropped
1327  * line(s) are not counted.
1328  */
record_print_text(struct printk_record * r,bool syslog,bool time)1329 static size_t record_print_text(struct printk_record *r, bool syslog,
1330 				bool time)
1331 {
1332 	size_t text_len = r->info->text_len;
1333 	size_t buf_size = r->text_buf_size;
1334 	char *text = r->text_buf;
1335 	char prefix[PREFIX_MAX];
1336 	bool truncated = false;
1337 	size_t prefix_len;
1338 	size_t line_len;
1339 	size_t len = 0;
1340 	char *next;
1341 
1342 	/*
1343 	 * If the message was truncated because the buffer was not large
1344 	 * enough, treat the available text as if it were the full text.
1345 	 */
1346 	if (text_len > buf_size)
1347 		text_len = buf_size;
1348 
1349 	prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1350 
1351 	/*
1352 	 * @text_len: bytes of unprocessed text
1353 	 * @line_len: bytes of current line _without_ newline
1354 	 * @text:     pointer to beginning of current line
1355 	 * @len:      number of bytes prepared in r->text_buf
1356 	 */
1357 	for (;;) {
1358 		next = memchr(text, '\n', text_len);
1359 		if (next) {
1360 			line_len = next - text;
1361 		} else {
1362 			/* Drop truncated line(s). */
1363 			if (truncated)
1364 				break;
1365 			line_len = text_len;
1366 		}
1367 
1368 		/*
1369 		 * Truncate the text if there is not enough space to add the
1370 		 * prefix and a trailing newline and a terminator.
1371 		 */
1372 		if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1373 			/* Drop even the current line if no space. */
1374 			if (len + prefix_len + line_len + 1 + 1 > buf_size)
1375 				break;
1376 
1377 			text_len = buf_size - len - prefix_len - 1 - 1;
1378 			truncated = true;
1379 		}
1380 
1381 		memmove(text + prefix_len, text, text_len);
1382 		memcpy(text, prefix, prefix_len);
1383 
1384 		/*
1385 		 * Increment the prepared length to include the text and
1386 		 * prefix that were just moved+copied. Also increment for the
1387 		 * newline at the end of this line. If this is the last line,
1388 		 * there is no newline, but it will be added immediately below.
1389 		 */
1390 		len += prefix_len + line_len + 1;
1391 		if (text_len == line_len) {
1392 			/*
1393 			 * This is the last line. Add the trailing newline
1394 			 * removed in vprintk_store().
1395 			 */
1396 			text[prefix_len + line_len] = '\n';
1397 			break;
1398 		}
1399 
1400 		/*
1401 		 * Advance beyond the added prefix and the related line with
1402 		 * its newline.
1403 		 */
1404 		text += prefix_len + line_len + 1;
1405 
1406 		/*
1407 		 * The remaining text has only decreased by the line with its
1408 		 * newline.
1409 		 *
1410 		 * Note that @text_len can become zero. It happens when @text
1411 		 * ended with a newline (either due to truncation or the
1412 		 * original string ending with "\n\n"). The loop is correctly
1413 		 * repeated and (if not truncated) an empty line with a prefix
1414 		 * will be prepared.
1415 		 */
1416 		text_len -= line_len + 1;
1417 	}
1418 
1419 	/*
1420 	 * If a buffer was provided, it will be terminated. Space for the
1421 	 * string terminator is guaranteed to be available. The terminator is
1422 	 * not counted in the return value.
1423 	 */
1424 	if (buf_size > 0)
1425 		r->text_buf[len] = 0;
1426 
1427 	return len;
1428 }
1429 
get_record_print_text_size(struct printk_info * info,unsigned int line_count,bool syslog,bool time)1430 static size_t get_record_print_text_size(struct printk_info *info,
1431 					 unsigned int line_count,
1432 					 bool syslog, bool time)
1433 {
1434 	char prefix[PREFIX_MAX];
1435 	size_t prefix_len;
1436 
1437 	prefix_len = info_print_prefix(info, syslog, time, prefix);
1438 
1439 	/*
1440 	 * Each line will be preceded with a prefix. The intermediate
1441 	 * newlines are already within the text, but a final trailing
1442 	 * newline will be added.
1443 	 */
1444 	return ((prefix_len * line_count) + info->text_len + 1);
1445 }
1446 
1447 /*
1448  * Beginning with @start_seq, find the first record where it and all following
1449  * records up to (but not including) @max_seq fit into @size.
1450  *
1451  * @max_seq is simply an upper bound and does not need to exist. If the caller
1452  * does not require an upper bound, -1 can be used for @max_seq.
1453  */
find_first_fitting_seq(u64 start_seq,u64 max_seq,size_t size,bool syslog,bool time)1454 static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1455 				  bool syslog, bool time)
1456 {
1457 	struct printk_info info;
1458 	unsigned int line_count;
1459 	size_t len = 0;
1460 	u64 seq;
1461 
1462 	/* Determine the size of the records up to @max_seq. */
1463 	prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1464 		if (info.seq >= max_seq)
1465 			break;
1466 		len += get_record_print_text_size(&info, line_count, syslog, time);
1467 	}
1468 
1469 	/*
1470 	 * Adjust the upper bound for the next loop to avoid subtracting
1471 	 * lengths that were never added.
1472 	 */
1473 	if (seq < max_seq)
1474 		max_seq = seq;
1475 
1476 	/*
1477 	 * Move first record forward until length fits into the buffer. Ignore
1478 	 * newest messages that were not counted in the above cycle. Messages
1479 	 * might appear and get lost in the meantime. This is a best effort
1480 	 * that prevents an infinite loop that could occur with a retry.
1481 	 */
1482 	prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1483 		if (len <= size || info.seq >= max_seq)
1484 			break;
1485 		len -= get_record_print_text_size(&info, line_count, syslog, time);
1486 	}
1487 
1488 	return seq;
1489 }
1490 
1491 /* The caller is responsible for making sure @size is greater than 0. */
syslog_print(char __user * buf,int size)1492 static int syslog_print(char __user *buf, int size)
1493 {
1494 	struct printk_info info;
1495 	struct printk_record r;
1496 	char *text;
1497 	int len = 0;
1498 	u64 seq;
1499 
1500 	text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1501 	if (!text)
1502 		return -ENOMEM;
1503 
1504 	prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1505 
1506 	mutex_lock(&syslog_lock);
1507 
1508 	/*
1509 	 * Wait for the @syslog_seq record to be available. @syslog_seq may
1510 	 * change while waiting.
1511 	 */
1512 	do {
1513 		seq = syslog_seq;
1514 
1515 		mutex_unlock(&syslog_lock);
1516 		/*
1517 		 * Guarantee this task is visible on the waitqueue before
1518 		 * checking the wake condition.
1519 		 *
1520 		 * The full memory barrier within set_current_state() of
1521 		 * prepare_to_wait_event() pairs with the full memory barrier
1522 		 * within wq_has_sleeper().
1523 		 *
1524 		 * This pairs with __wake_up_klogd:A.
1525 		 */
1526 		len = wait_event_interruptible(log_wait,
1527 				prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
1528 		mutex_lock(&syslog_lock);
1529 
1530 		if (len)
1531 			goto out;
1532 	} while (syslog_seq != seq);
1533 
1534 	/*
1535 	 * Copy records that fit into the buffer. The above cycle makes sure
1536 	 * that the first record is always available.
1537 	 */
1538 	do {
1539 		size_t n;
1540 		size_t skip;
1541 		int err;
1542 
1543 		if (!prb_read_valid(prb, syslog_seq, &r))
1544 			break;
1545 
1546 		if (r.info->seq != syslog_seq) {
1547 			/* message is gone, move to next valid one */
1548 			syslog_seq = r.info->seq;
1549 			syslog_partial = 0;
1550 		}
1551 
1552 		/*
1553 		 * To keep reading/counting partial line consistent,
1554 		 * use printk_time value as of the beginning of a line.
1555 		 */
1556 		if (!syslog_partial)
1557 			syslog_time = printk_time;
1558 
1559 		skip = syslog_partial;
1560 		n = record_print_text(&r, true, syslog_time);
1561 		if (n - syslog_partial <= size) {
1562 			/* message fits into buffer, move forward */
1563 			syslog_seq = r.info->seq + 1;
1564 			n -= syslog_partial;
1565 			syslog_partial = 0;
1566 		} else if (!len){
1567 			/* partial read(), remember position */
1568 			n = size;
1569 			syslog_partial += n;
1570 		} else
1571 			n = 0;
1572 
1573 		if (!n)
1574 			break;
1575 
1576 		mutex_unlock(&syslog_lock);
1577 		err = copy_to_user(buf, text + skip, n);
1578 		mutex_lock(&syslog_lock);
1579 
1580 		if (err) {
1581 			if (!len)
1582 				len = -EFAULT;
1583 			break;
1584 		}
1585 
1586 		len += n;
1587 		size -= n;
1588 		buf += n;
1589 	} while (size);
1590 out:
1591 	mutex_unlock(&syslog_lock);
1592 	kfree(text);
1593 	return len;
1594 }
1595 
syslog_print_all(char __user * buf,int size,bool clear)1596 static int syslog_print_all(char __user *buf, int size, bool clear)
1597 {
1598 	struct printk_info info;
1599 	struct printk_record r;
1600 	char *text;
1601 	int len = 0;
1602 	u64 seq;
1603 	bool time;
1604 
1605 	text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1606 	if (!text)
1607 		return -ENOMEM;
1608 
1609 	time = printk_time;
1610 	/*
1611 	 * Find first record that fits, including all following records,
1612 	 * into the user-provided buffer for this dump.
1613 	 */
1614 	seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1615 				     size, true, time);
1616 
1617 	prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1618 
1619 	len = 0;
1620 	prb_for_each_record(seq, prb, seq, &r) {
1621 		int textlen;
1622 
1623 		textlen = record_print_text(&r, true, time);
1624 
1625 		if (len + textlen > size) {
1626 			seq--;
1627 			break;
1628 		}
1629 
1630 		if (copy_to_user(buf + len, text, textlen))
1631 			len = -EFAULT;
1632 		else
1633 			len += textlen;
1634 
1635 		if (len < 0)
1636 			break;
1637 	}
1638 
1639 	if (clear) {
1640 		mutex_lock(&syslog_lock);
1641 		latched_seq_write(&clear_seq, seq);
1642 		mutex_unlock(&syslog_lock);
1643 	}
1644 
1645 	kfree(text);
1646 	return len;
1647 }
1648 
syslog_clear(void)1649 static void syslog_clear(void)
1650 {
1651 	mutex_lock(&syslog_lock);
1652 	latched_seq_write(&clear_seq, prb_next_seq(prb));
1653 	mutex_unlock(&syslog_lock);
1654 }
1655 
do_syslog(int type,char __user * buf,int len,int source)1656 int do_syslog(int type, char __user *buf, int len, int source)
1657 {
1658 	struct printk_info info;
1659 	bool clear = false;
1660 	static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1661 	int error;
1662 
1663 	error = check_syslog_permissions(type, source);
1664 	if (error)
1665 		return error;
1666 
1667 	switch (type) {
1668 	case SYSLOG_ACTION_CLOSE:	/* Close log */
1669 		break;
1670 	case SYSLOG_ACTION_OPEN:	/* Open log */
1671 		break;
1672 	case SYSLOG_ACTION_READ:	/* Read from log */
1673 		if (!buf || len < 0)
1674 			return -EINVAL;
1675 		if (!len)
1676 			return 0;
1677 		if (!access_ok(buf, len))
1678 			return -EFAULT;
1679 		error = syslog_print(buf, len);
1680 		break;
1681 	/* Read/clear last kernel messages */
1682 	case SYSLOG_ACTION_READ_CLEAR:
1683 		clear = true;
1684 		fallthrough;
1685 	/* Read last kernel messages */
1686 	case SYSLOG_ACTION_READ_ALL:
1687 		if (!buf || len < 0)
1688 			return -EINVAL;
1689 		if (!len)
1690 			return 0;
1691 		if (!access_ok(buf, len))
1692 			return -EFAULT;
1693 		error = syslog_print_all(buf, len, clear);
1694 		break;
1695 	/* Clear ring buffer */
1696 	case SYSLOG_ACTION_CLEAR:
1697 		syslog_clear();
1698 		break;
1699 	/* Disable logging to console */
1700 	case SYSLOG_ACTION_CONSOLE_OFF:
1701 		if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1702 			saved_console_loglevel = console_loglevel;
1703 		console_loglevel = minimum_console_loglevel;
1704 		break;
1705 	/* Enable logging to console */
1706 	case SYSLOG_ACTION_CONSOLE_ON:
1707 		if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1708 			console_loglevel = saved_console_loglevel;
1709 			saved_console_loglevel = LOGLEVEL_DEFAULT;
1710 		}
1711 		break;
1712 	/* Set level of messages printed to console */
1713 	case SYSLOG_ACTION_CONSOLE_LEVEL:
1714 		if (len < 1 || len > 8)
1715 			return -EINVAL;
1716 		if (len < minimum_console_loglevel)
1717 			len = minimum_console_loglevel;
1718 		console_loglevel = len;
1719 		/* Implicitly re-enable logging to console */
1720 		saved_console_loglevel = LOGLEVEL_DEFAULT;
1721 		break;
1722 	/* Number of chars in the log buffer */
1723 	case SYSLOG_ACTION_SIZE_UNREAD:
1724 		mutex_lock(&syslog_lock);
1725 		if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1726 			/* No unread messages. */
1727 			mutex_unlock(&syslog_lock);
1728 			return 0;
1729 		}
1730 		if (info.seq != syslog_seq) {
1731 			/* messages are gone, move to first one */
1732 			syslog_seq = info.seq;
1733 			syslog_partial = 0;
1734 		}
1735 		if (source == SYSLOG_FROM_PROC) {
1736 			/*
1737 			 * Short-cut for poll(/"proc/kmsg") which simply checks
1738 			 * for pending data, not the size; return the count of
1739 			 * records, not the length.
1740 			 */
1741 			error = prb_next_seq(prb) - syslog_seq;
1742 		} else {
1743 			bool time = syslog_partial ? syslog_time : printk_time;
1744 			unsigned int line_count;
1745 			u64 seq;
1746 
1747 			prb_for_each_info(syslog_seq, prb, seq, &info,
1748 					  &line_count) {
1749 				error += get_record_print_text_size(&info, line_count,
1750 								    true, time);
1751 				time = printk_time;
1752 			}
1753 			error -= syslog_partial;
1754 		}
1755 		mutex_unlock(&syslog_lock);
1756 		break;
1757 	/* Size of the log buffer */
1758 	case SYSLOG_ACTION_SIZE_BUFFER:
1759 		error = log_buf_len;
1760 		break;
1761 	default:
1762 		error = -EINVAL;
1763 		break;
1764 	}
1765 
1766 	return error;
1767 }
1768 
SYSCALL_DEFINE3(syslog,int,type,char __user *,buf,int,len)1769 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1770 {
1771 	return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1772 }
1773 
1774 /*
1775  * Special console_lock variants that help to reduce the risk of soft-lockups.
1776  * They allow to pass console_lock to another printk() call using a busy wait.
1777  */
1778 
1779 #ifdef CONFIG_LOCKDEP
1780 static struct lockdep_map console_owner_dep_map = {
1781 	.name = "console_owner"
1782 };
1783 #endif
1784 
1785 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1786 static struct task_struct *console_owner;
1787 static bool console_waiter;
1788 
1789 /**
1790  * console_lock_spinning_enable - mark beginning of code where another
1791  *	thread might safely busy wait
1792  *
1793  * This basically converts console_lock into a spinlock. This marks
1794  * the section where the console_lock owner can not sleep, because
1795  * there may be a waiter spinning (like a spinlock). Also it must be
1796  * ready to hand over the lock at the end of the section.
1797  */
console_lock_spinning_enable(void)1798 static void console_lock_spinning_enable(void)
1799 {
1800 	raw_spin_lock(&console_owner_lock);
1801 	console_owner = current;
1802 	raw_spin_unlock(&console_owner_lock);
1803 
1804 	/* The waiter may spin on us after setting console_owner */
1805 	spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1806 }
1807 
1808 /**
1809  * console_lock_spinning_disable_and_check - mark end of code where another
1810  *	thread was able to busy wait and check if there is a waiter
1811  *
1812  * This is called at the end of the section where spinning is allowed.
1813  * It has two functions. First, it is a signal that it is no longer
1814  * safe to start busy waiting for the lock. Second, it checks if
1815  * there is a busy waiter and passes the lock rights to her.
1816  *
1817  * Important: Callers lose the lock if there was a busy waiter.
1818  *	They must not touch items synchronized by console_lock
1819  *	in this case.
1820  *
1821  * Return: 1 if the lock rights were passed, 0 otherwise.
1822  */
console_lock_spinning_disable_and_check(void)1823 static int console_lock_spinning_disable_and_check(void)
1824 {
1825 	int waiter;
1826 
1827 	raw_spin_lock(&console_owner_lock);
1828 	waiter = READ_ONCE(console_waiter);
1829 	console_owner = NULL;
1830 	raw_spin_unlock(&console_owner_lock);
1831 
1832 	if (!waiter) {
1833 		spin_release(&console_owner_dep_map, _THIS_IP_);
1834 		return 0;
1835 	}
1836 
1837 	/* The waiter is now free to continue */
1838 	WRITE_ONCE(console_waiter, false);
1839 
1840 	spin_release(&console_owner_dep_map, _THIS_IP_);
1841 
1842 	/*
1843 	 * Hand off console_lock to waiter. The waiter will perform
1844 	 * the up(). After this, the waiter is the console_lock owner.
1845 	 */
1846 	mutex_release(&console_lock_dep_map, _THIS_IP_);
1847 	return 1;
1848 }
1849 
1850 /**
1851  * console_trylock_spinning - try to get console_lock by busy waiting
1852  *
1853  * This allows to busy wait for the console_lock when the current
1854  * owner is running in specially marked sections. It means that
1855  * the current owner is running and cannot reschedule until it
1856  * is ready to lose the lock.
1857  *
1858  * Return: 1 if we got the lock, 0 othrewise
1859  */
console_trylock_spinning(void)1860 static int console_trylock_spinning(void)
1861 {
1862 	struct task_struct *owner = NULL;
1863 	bool waiter;
1864 	bool spin = false;
1865 	unsigned long flags;
1866 
1867 	if (console_trylock())
1868 		return 1;
1869 
1870 	/*
1871 	 * It's unsafe to spin once a panic has begun. If we are the
1872 	 * panic CPU, we may have already halted the owner of the
1873 	 * console_sem. If we are not the panic CPU, then we should
1874 	 * avoid taking console_sem, so the panic CPU has a better
1875 	 * chance of cleanly acquiring it later.
1876 	 */
1877 	if (panic_in_progress())
1878 		return 0;
1879 
1880 	printk_safe_enter_irqsave(flags);
1881 
1882 	raw_spin_lock(&console_owner_lock);
1883 	owner = READ_ONCE(console_owner);
1884 	waiter = READ_ONCE(console_waiter);
1885 	if (!waiter && owner && owner != current) {
1886 		WRITE_ONCE(console_waiter, true);
1887 		spin = true;
1888 	}
1889 	raw_spin_unlock(&console_owner_lock);
1890 
1891 	/*
1892 	 * If there is an active printk() writing to the
1893 	 * consoles, instead of having it write our data too,
1894 	 * see if we can offload that load from the active
1895 	 * printer, and do some printing ourselves.
1896 	 * Go into a spin only if there isn't already a waiter
1897 	 * spinning, and there is an active printer, and
1898 	 * that active printer isn't us (recursive printk?).
1899 	 */
1900 	if (!spin) {
1901 		printk_safe_exit_irqrestore(flags);
1902 		return 0;
1903 	}
1904 
1905 	/* We spin waiting for the owner to release us */
1906 	spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1907 	/* Owner will clear console_waiter on hand off */
1908 	while (READ_ONCE(console_waiter))
1909 		cpu_relax();
1910 	spin_release(&console_owner_dep_map, _THIS_IP_);
1911 
1912 	printk_safe_exit_irqrestore(flags);
1913 	/*
1914 	 * The owner passed the console lock to us.
1915 	 * Since we did not spin on console lock, annotate
1916 	 * this as a trylock. Otherwise lockdep will
1917 	 * complain.
1918 	 */
1919 	mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1920 
1921 	return 1;
1922 }
1923 
1924 /*
1925  * Call the specified console driver, asking it to write out the specified
1926  * text and length. If @dropped_text is non-NULL and any records have been
1927  * dropped, a dropped message will be written out first.
1928  */
call_console_driver(struct console * con,const char * text,size_t len,char * dropped_text)1929 static void call_console_driver(struct console *con, const char *text, size_t len,
1930 				char *dropped_text)
1931 {
1932 	size_t dropped_len;
1933 
1934 	if (con->dropped && dropped_text) {
1935 		dropped_len = snprintf(dropped_text, DROPPED_TEXT_MAX,
1936 				       "** %lu printk messages dropped **\n",
1937 				       con->dropped);
1938 		con->dropped = 0;
1939 		con->write(con, dropped_text, dropped_len);
1940 	}
1941 
1942 	con->write(con, text, len);
1943 }
1944 
1945 /*
1946  * Recursion is tracked separately on each CPU. If NMIs are supported, an
1947  * additional NMI context per CPU is also separately tracked. Until per-CPU
1948  * is available, a separate "early tracking" is performed.
1949  */
1950 static DEFINE_PER_CPU(u8, printk_count);
1951 static u8 printk_count_early;
1952 #ifdef CONFIG_HAVE_NMI
1953 static DEFINE_PER_CPU(u8, printk_count_nmi);
1954 static u8 printk_count_nmi_early;
1955 #endif
1956 
1957 /*
1958  * Recursion is limited to keep the output sane. printk() should not require
1959  * more than 1 level of recursion (allowing, for example, printk() to trigger
1960  * a WARN), but a higher value is used in case some printk-internal errors
1961  * exist, such as the ringbuffer validation checks failing.
1962  */
1963 #define PRINTK_MAX_RECURSION 3
1964 
1965 /*
1966  * Return a pointer to the dedicated counter for the CPU+context of the
1967  * caller.
1968  */
__printk_recursion_counter(void)1969 static u8 *__printk_recursion_counter(void)
1970 {
1971 #ifdef CONFIG_HAVE_NMI
1972 	if (in_nmi()) {
1973 		if (printk_percpu_data_ready())
1974 			return this_cpu_ptr(&printk_count_nmi);
1975 		return &printk_count_nmi_early;
1976 	}
1977 #endif
1978 	if (printk_percpu_data_ready())
1979 		return this_cpu_ptr(&printk_count);
1980 	return &printk_count_early;
1981 }
1982 
1983 /*
1984  * Enter recursion tracking. Interrupts are disabled to simplify tracking.
1985  * The caller must check the boolean return value to see if the recursion is
1986  * allowed. On failure, interrupts are not disabled.
1987  *
1988  * @recursion_ptr must be a variable of type (u8 *) and is the same variable
1989  * that is passed to printk_exit_irqrestore().
1990  */
1991 #define printk_enter_irqsave(recursion_ptr, flags)	\
1992 ({							\
1993 	bool success = true;				\
1994 							\
1995 	typecheck(u8 *, recursion_ptr);			\
1996 	local_irq_save(flags);				\
1997 	(recursion_ptr) = __printk_recursion_counter();	\
1998 	if (*(recursion_ptr) > PRINTK_MAX_RECURSION) {	\
1999 		local_irq_restore(flags);		\
2000 		success = false;			\
2001 	} else {					\
2002 		(*(recursion_ptr))++;			\
2003 	}						\
2004 	success;					\
2005 })
2006 
2007 /* Exit recursion tracking, restoring interrupts. */
2008 #define printk_exit_irqrestore(recursion_ptr, flags)	\
2009 	do {						\
2010 		typecheck(u8 *, recursion_ptr);		\
2011 		(*(recursion_ptr))--;			\
2012 		local_irq_restore(flags);		\
2013 	} while (0)
2014 
2015 int printk_delay_msec __read_mostly;
2016 
printk_delay(int level)2017 static inline void printk_delay(int level)
2018 {
2019 	boot_delay_msec(level);
2020 
2021 	if (unlikely(printk_delay_msec)) {
2022 		int m = printk_delay_msec;
2023 
2024 		while (m--) {
2025 			mdelay(1);
2026 			touch_nmi_watchdog();
2027 		}
2028 	}
2029 }
2030 
printk_caller_id(void)2031 static inline u32 printk_caller_id(void)
2032 {
2033 	return in_task() ? task_pid_nr(current) :
2034 		0x80000000 + smp_processor_id();
2035 }
2036 
2037 /**
2038  * printk_parse_prefix - Parse level and control flags.
2039  *
2040  * @text:     The terminated text message.
2041  * @level:    A pointer to the current level value, will be updated.
2042  * @flags:    A pointer to the current printk_info flags, will be updated.
2043  *
2044  * @level may be NULL if the caller is not interested in the parsed value.
2045  * Otherwise the variable pointed to by @level must be set to
2046  * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2047  *
2048  * @flags may be NULL if the caller is not interested in the parsed value.
2049  * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2050  * value.
2051  *
2052  * Return: The length of the parsed level and control flags.
2053  */
printk_parse_prefix(const char * text,int * level,enum printk_info_flags * flags)2054 u16 printk_parse_prefix(const char *text, int *level,
2055 			enum printk_info_flags *flags)
2056 {
2057 	u16 prefix_len = 0;
2058 	int kern_level;
2059 
2060 	while (*text) {
2061 		kern_level = printk_get_level(text);
2062 		if (!kern_level)
2063 			break;
2064 
2065 		switch (kern_level) {
2066 		case '0' ... '7':
2067 			if (level && *level == LOGLEVEL_DEFAULT)
2068 				*level = kern_level - '0';
2069 			break;
2070 		case 'c':	/* KERN_CONT */
2071 			if (flags)
2072 				*flags |= LOG_CONT;
2073 		}
2074 
2075 		prefix_len += 2;
2076 		text += 2;
2077 	}
2078 
2079 	return prefix_len;
2080 }
2081 
2082 __printf(5, 0)
printk_sprint(char * text,u16 size,int facility,enum printk_info_flags * flags,const char * fmt,va_list args)2083 static u16 printk_sprint(char *text, u16 size, int facility,
2084 			 enum printk_info_flags *flags, const char *fmt,
2085 			 va_list args)
2086 {
2087 	u16 text_len;
2088 
2089 	text_len = vscnprintf(text, size, fmt, args);
2090 
2091 	/* Mark and strip a trailing newline. */
2092 	if (text_len && text[text_len - 1] == '\n') {
2093 		text_len--;
2094 		*flags |= LOG_NEWLINE;
2095 	}
2096 
2097 	/* Strip log level and control flags. */
2098 	if (facility == 0) {
2099 		u16 prefix_len;
2100 
2101 		prefix_len = printk_parse_prefix(text, NULL, NULL);
2102 		if (prefix_len) {
2103 			text_len -= prefix_len;
2104 			memmove(text, text + prefix_len, text_len);
2105 		}
2106 	}
2107 
2108 	trace_console_rcuidle(text, text_len);
2109 
2110 	return text_len;
2111 }
2112 
2113 __printf(4, 0)
vprintk_store(int facility,int level,const struct dev_printk_info * dev_info,const char * fmt,va_list args)2114 int vprintk_store(int facility, int level,
2115 		  const struct dev_printk_info *dev_info,
2116 		  const char *fmt, va_list args)
2117 {
2118 	struct prb_reserved_entry e;
2119 	enum printk_info_flags flags = 0;
2120 	struct printk_record r;
2121 	unsigned long irqflags;
2122 	u16 trunc_msg_len = 0;
2123 	char prefix_buf[8];
2124 	u8 *recursion_ptr;
2125 	u16 reserve_size;
2126 	va_list args2;
2127 	u32 caller_id;
2128 	u16 text_len;
2129 	int ret = 0;
2130 	u64 ts_nsec;
2131 
2132 	if (!printk_enter_irqsave(recursion_ptr, irqflags))
2133 		return 0;
2134 
2135 	/*
2136 	 * Since the duration of printk() can vary depending on the message
2137 	 * and state of the ringbuffer, grab the timestamp now so that it is
2138 	 * close to the call of printk(). This provides a more deterministic
2139 	 * timestamp with respect to the caller.
2140 	 */
2141 	ts_nsec = local_clock();
2142 
2143 	caller_id = printk_caller_id();
2144 
2145 	/*
2146 	 * The sprintf needs to come first since the syslog prefix might be
2147 	 * passed in as a parameter. An extra byte must be reserved so that
2148 	 * later the vscnprintf() into the reserved buffer has room for the
2149 	 * terminating '\0', which is not counted by vsnprintf().
2150 	 */
2151 	va_copy(args2, args);
2152 	reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2153 	va_end(args2);
2154 
2155 	if (reserve_size > LOG_LINE_MAX)
2156 		reserve_size = LOG_LINE_MAX;
2157 
2158 	/* Extract log level or control flags. */
2159 	if (facility == 0)
2160 		printk_parse_prefix(&prefix_buf[0], &level, &flags);
2161 
2162 	if (level == LOGLEVEL_DEFAULT)
2163 		level = default_message_loglevel;
2164 
2165 	if (dev_info)
2166 		flags |= LOG_NEWLINE;
2167 
2168 	if (flags & LOG_CONT) {
2169 		prb_rec_init_wr(&r, reserve_size);
2170 		if (prb_reserve_in_last(&e, prb, &r, caller_id, LOG_LINE_MAX)) {
2171 			text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2172 						 facility, &flags, fmt, args);
2173 			r.info->text_len += text_len;
2174 
2175 			if (flags & LOG_NEWLINE) {
2176 				r.info->flags |= LOG_NEWLINE;
2177 				prb_final_commit(&e);
2178 			} else {
2179 				prb_commit(&e);
2180 			}
2181 
2182 			ret = text_len;
2183 			goto out;
2184 		}
2185 	}
2186 
2187 	/*
2188 	 * Explicitly initialize the record before every prb_reserve() call.
2189 	 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2190 	 * structure when they fail.
2191 	 */
2192 	prb_rec_init_wr(&r, reserve_size);
2193 	if (!prb_reserve(&e, prb, &r)) {
2194 		/* truncate the message if it is too long for empty buffer */
2195 		truncate_msg(&reserve_size, &trunc_msg_len);
2196 
2197 		prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2198 		if (!prb_reserve(&e, prb, &r))
2199 			goto out;
2200 	}
2201 
2202 	/* fill message */
2203 	text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2204 	if (trunc_msg_len)
2205 		memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2206 	r.info->text_len = text_len + trunc_msg_len;
2207 	r.info->facility = facility;
2208 	r.info->level = level & 7;
2209 	r.info->flags = flags & 0x1f;
2210 	r.info->ts_nsec = ts_nsec;
2211 	r.info->caller_id = caller_id;
2212 	if (dev_info)
2213 		memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2214 
2215 	/* A message without a trailing newline can be continued. */
2216 	if (!(flags & LOG_NEWLINE))
2217 		prb_commit(&e);
2218 	else
2219 		prb_final_commit(&e);
2220 
2221 	ret = text_len + trunc_msg_len;
2222 out:
2223 	printk_exit_irqrestore(recursion_ptr, irqflags);
2224 	return ret;
2225 }
2226 
vprintk_emit(int facility,int level,const struct dev_printk_info * dev_info,const char * fmt,va_list args)2227 asmlinkage int vprintk_emit(int facility, int level,
2228 			    const struct dev_printk_info *dev_info,
2229 			    const char *fmt, va_list args)
2230 {
2231 	int printed_len;
2232 	bool in_sched = false;
2233 
2234 	/* Suppress unimportant messages after panic happens */
2235 	if (unlikely(suppress_printk))
2236 		return 0;
2237 
2238 	if (unlikely(suppress_panic_printk) &&
2239 	    atomic_read(&panic_cpu) != raw_smp_processor_id())
2240 		return 0;
2241 
2242 	if (level == LOGLEVEL_SCHED) {
2243 		level = LOGLEVEL_DEFAULT;
2244 		in_sched = true;
2245 	}
2246 
2247 	printk_delay(level);
2248 
2249 	printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2250 
2251 	/* If called from the scheduler, we can not call up(). */
2252 	if (!in_sched) {
2253 		/*
2254 		 * The caller may be holding system-critical or
2255 		 * timing-sensitive locks. Disable preemption during
2256 		 * printing of all remaining records to all consoles so that
2257 		 * this context can return as soon as possible. Hopefully
2258 		 * another printk() caller will take over the printing.
2259 		 */
2260 		preempt_disable();
2261 		/*
2262 		 * Try to acquire and then immediately release the console
2263 		 * semaphore. The release will print out buffers. With the
2264 		 * spinning variant, this context tries to take over the
2265 		 * printing from another printing context.
2266 		 */
2267 		if (console_trylock_spinning())
2268 			console_unlock();
2269 		preempt_enable();
2270 	}
2271 
2272 	wake_up_klogd();
2273 	return printed_len;
2274 }
2275 EXPORT_SYMBOL(vprintk_emit);
2276 
vprintk_default(const char * fmt,va_list args)2277 int vprintk_default(const char *fmt, va_list args)
2278 {
2279 	return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2280 }
2281 EXPORT_SYMBOL_GPL(vprintk_default);
2282 
_printk(const char * fmt,...)2283 asmlinkage __visible int _printk(const char *fmt, ...)
2284 {
2285 	va_list args;
2286 	int r;
2287 
2288 	va_start(args, fmt);
2289 	r = vprintk(fmt, args);
2290 	va_end(args);
2291 
2292 	return r;
2293 }
2294 EXPORT_SYMBOL(_printk);
2295 
2296 static bool pr_flush(int timeout_ms, bool reset_on_progress);
2297 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress);
2298 
2299 #else /* CONFIG_PRINTK */
2300 
2301 #define CONSOLE_LOG_MAX		0
2302 #define DROPPED_TEXT_MAX	0
2303 #define printk_time		false
2304 
2305 #define prb_read_valid(rb, seq, r)	false
2306 #define prb_first_valid_seq(rb)		0
2307 #define prb_next_seq(rb)		0
2308 
2309 static u64 syslog_seq;
2310 
record_print_text(const struct printk_record * r,bool syslog,bool time)2311 static size_t record_print_text(const struct printk_record *r,
2312 				bool syslog, bool time)
2313 {
2314 	return 0;
2315 }
info_print_ext_header(char * buf,size_t size,struct printk_info * info)2316 static ssize_t info_print_ext_header(char *buf, size_t size,
2317 				     struct printk_info *info)
2318 {
2319 	return 0;
2320 }
msg_print_ext_body(char * buf,size_t size,char * text,size_t text_len,struct dev_printk_info * dev_info)2321 static ssize_t msg_print_ext_body(char *buf, size_t size,
2322 				  char *text, size_t text_len,
2323 				  struct dev_printk_info *dev_info) { return 0; }
console_lock_spinning_enable(void)2324 static void console_lock_spinning_enable(void) { }
console_lock_spinning_disable_and_check(void)2325 static int console_lock_spinning_disable_and_check(void) { return 0; }
call_console_driver(struct console * con,const char * text,size_t len,char * dropped_text)2326 static void call_console_driver(struct console *con, const char *text, size_t len,
2327 				char *dropped_text)
2328 {
2329 }
suppress_message_printing(int level)2330 static bool suppress_message_printing(int level) { return false; }
pr_flush(int timeout_ms,bool reset_on_progress)2331 static bool pr_flush(int timeout_ms, bool reset_on_progress) { return true; }
__pr_flush(struct console * con,int timeout_ms,bool reset_on_progress)2332 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress) { return true; }
2333 
2334 #endif /* CONFIG_PRINTK */
2335 
2336 #ifdef CONFIG_EARLY_PRINTK
2337 struct console *early_console;
2338 
early_printk(const char * fmt,...)2339 asmlinkage __visible void early_printk(const char *fmt, ...)
2340 {
2341 	va_list ap;
2342 	char buf[512];
2343 	int n;
2344 
2345 	if (!early_console)
2346 		return;
2347 
2348 	va_start(ap, fmt);
2349 	n = vscnprintf(buf, sizeof(buf), fmt, ap);
2350 	va_end(ap);
2351 
2352 	early_console->write(early_console, buf, n);
2353 }
2354 #endif
2355 
set_user_specified(struct console_cmdline * c,bool user_specified)2356 static void set_user_specified(struct console_cmdline *c, bool user_specified)
2357 {
2358 	if (!user_specified)
2359 		return;
2360 
2361 	/*
2362 	 * @c console was defined by the user on the command line.
2363 	 * Do not clear when added twice also by SPCR or the device tree.
2364 	 */
2365 	c->user_specified = true;
2366 	/* At least one console defined by the user on the command line. */
2367 	console_set_on_cmdline = 1;
2368 }
2369 
__add_preferred_console(char * name,int idx,char * options,char * brl_options,bool user_specified)2370 static int __add_preferred_console(char *name, int idx, char *options,
2371 				   char *brl_options, bool user_specified)
2372 {
2373 	struct console_cmdline *c;
2374 	int i;
2375 
2376 	/*
2377 	 *	See if this tty is not yet registered, and
2378 	 *	if we have a slot free.
2379 	 */
2380 	for (i = 0, c = console_cmdline;
2381 	     i < MAX_CMDLINECONSOLES && c->name[0];
2382 	     i++, c++) {
2383 		if (strcmp(c->name, name) == 0 && c->index == idx) {
2384 			if (!brl_options)
2385 				preferred_console = i;
2386 			set_user_specified(c, user_specified);
2387 			return 0;
2388 		}
2389 	}
2390 	if (i == MAX_CMDLINECONSOLES)
2391 		return -E2BIG;
2392 	if (!brl_options)
2393 		preferred_console = i;
2394 	strlcpy(c->name, name, sizeof(c->name));
2395 	c->options = options;
2396 	set_user_specified(c, user_specified);
2397 	braille_set_options(c, brl_options);
2398 
2399 	c->index = idx;
2400 	return 0;
2401 }
2402 
console_msg_format_setup(char * str)2403 static int __init console_msg_format_setup(char *str)
2404 {
2405 	if (!strcmp(str, "syslog"))
2406 		console_msg_format = MSG_FORMAT_SYSLOG;
2407 	if (!strcmp(str, "default"))
2408 		console_msg_format = MSG_FORMAT_DEFAULT;
2409 	return 1;
2410 }
2411 __setup("console_msg_format=", console_msg_format_setup);
2412 
2413 /*
2414  * Set up a console.  Called via do_early_param() in init/main.c
2415  * for each "console=" parameter in the boot command line.
2416  */
console_setup(char * str)2417 static int __init console_setup(char *str)
2418 {
2419 	char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2420 	char *s, *options, *brl_options = NULL;
2421 	int idx;
2422 
2423 	/*
2424 	 * console="" or console=null have been suggested as a way to
2425 	 * disable console output. Use ttynull that has been created
2426 	 * for exactly this purpose.
2427 	 */
2428 	if (str[0] == 0 || strcmp(str, "null") == 0) {
2429 		__add_preferred_console("ttynull", 0, NULL, NULL, true);
2430 		return 1;
2431 	}
2432 
2433 	if (_braille_console_setup(&str, &brl_options))
2434 		return 1;
2435 
2436 	/*
2437 	 * Decode str into name, index, options.
2438 	 */
2439 	if (str[0] >= '0' && str[0] <= '9') {
2440 		strcpy(buf, "ttyS");
2441 		strncpy(buf + 4, str, sizeof(buf) - 5);
2442 	} else {
2443 		strncpy(buf, str, sizeof(buf) - 1);
2444 	}
2445 	buf[sizeof(buf) - 1] = 0;
2446 	options = strchr(str, ',');
2447 	if (options)
2448 		*(options++) = 0;
2449 #ifdef __sparc__
2450 	if (!strcmp(str, "ttya"))
2451 		strcpy(buf, "ttyS0");
2452 	if (!strcmp(str, "ttyb"))
2453 		strcpy(buf, "ttyS1");
2454 #endif
2455 	for (s = buf; *s; s++)
2456 		if (isdigit(*s) || *s == ',')
2457 			break;
2458 	idx = simple_strtoul(s, NULL, 10);
2459 	*s = 0;
2460 
2461 	__add_preferred_console(buf, idx, options, brl_options, true);
2462 	return 1;
2463 }
2464 __setup("console=", console_setup);
2465 
2466 /**
2467  * add_preferred_console - add a device to the list of preferred consoles.
2468  * @name: device name
2469  * @idx: device index
2470  * @options: options for this console
2471  *
2472  * The last preferred console added will be used for kernel messages
2473  * and stdin/out/err for init.  Normally this is used by console_setup
2474  * above to handle user-supplied console arguments; however it can also
2475  * be used by arch-specific code either to override the user or more
2476  * commonly to provide a default console (ie from PROM variables) when
2477  * the user has not supplied one.
2478  */
add_preferred_console(char * name,int idx,char * options)2479 int add_preferred_console(char *name, int idx, char *options)
2480 {
2481 	return __add_preferred_console(name, idx, options, NULL, false);
2482 }
2483 
2484 bool console_suspend_enabled = true;
2485 EXPORT_SYMBOL(console_suspend_enabled);
2486 
console_suspend_disable(char * str)2487 static int __init console_suspend_disable(char *str)
2488 {
2489 	console_suspend_enabled = false;
2490 	return 1;
2491 }
2492 __setup("no_console_suspend", console_suspend_disable);
2493 module_param_named(console_suspend, console_suspend_enabled,
2494 		bool, S_IRUGO | S_IWUSR);
2495 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2496 	" and hibernate operations");
2497 
2498 static bool printk_console_no_auto_verbose;
2499 
console_verbose(void)2500 void console_verbose(void)
2501 {
2502 	if (console_loglevel && !printk_console_no_auto_verbose)
2503 		console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2504 }
2505 EXPORT_SYMBOL_GPL(console_verbose);
2506 
2507 module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2508 MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2509 
2510 /**
2511  * suspend_console - suspend the console subsystem
2512  *
2513  * This disables printk() while we go into suspend states
2514  */
suspend_console(void)2515 void suspend_console(void)
2516 {
2517 	if (!console_suspend_enabled)
2518 		return;
2519 	pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2520 	pr_flush(1000, true);
2521 	console_lock();
2522 	console_suspended = 1;
2523 	up_console_sem();
2524 }
2525 
resume_console(void)2526 void resume_console(void)
2527 {
2528 	if (!console_suspend_enabled)
2529 		return;
2530 	down_console_sem();
2531 	console_suspended = 0;
2532 	console_unlock();
2533 	pr_flush(1000, true);
2534 }
2535 
2536 /**
2537  * console_cpu_notify - print deferred console messages after CPU hotplug
2538  * @cpu: unused
2539  *
2540  * If printk() is called from a CPU that is not online yet, the messages
2541  * will be printed on the console only if there are CON_ANYTIME consoles.
2542  * This function is called when a new CPU comes online (or fails to come
2543  * up) or goes offline.
2544  */
console_cpu_notify(unsigned int cpu)2545 static int console_cpu_notify(unsigned int cpu)
2546 {
2547 	if (!cpuhp_tasks_frozen) {
2548 		/* If trylock fails, someone else is doing the printing */
2549 		if (console_trylock())
2550 			console_unlock();
2551 	}
2552 	return 0;
2553 }
2554 
2555 /**
2556  * console_lock - lock the console system for exclusive use.
2557  *
2558  * Acquires a lock which guarantees that the caller has
2559  * exclusive access to the console system and the console_drivers list.
2560  *
2561  * Can sleep, returns nothing.
2562  */
console_lock(void)2563 void console_lock(void)
2564 {
2565 	might_sleep();
2566 
2567 	down_console_sem();
2568 	if (console_suspended)
2569 		return;
2570 	console_locked = 1;
2571 	console_may_schedule = 1;
2572 }
2573 EXPORT_SYMBOL(console_lock);
2574 
2575 /**
2576  * console_trylock - try to lock the console system for exclusive use.
2577  *
2578  * Try to acquire a lock which guarantees that the caller has exclusive
2579  * access to the console system and the console_drivers list.
2580  *
2581  * returns 1 on success, and 0 on failure to acquire the lock.
2582  */
console_trylock(void)2583 int console_trylock(void)
2584 {
2585 	if (down_trylock_console_sem())
2586 		return 0;
2587 	if (console_suspended) {
2588 		up_console_sem();
2589 		return 0;
2590 	}
2591 	console_locked = 1;
2592 	console_may_schedule = 0;
2593 	return 1;
2594 }
2595 EXPORT_SYMBOL(console_trylock);
2596 
is_console_locked(void)2597 int is_console_locked(void)
2598 {
2599 	return console_locked;
2600 }
2601 EXPORT_SYMBOL(is_console_locked);
2602 
2603 /*
2604  * Return true when this CPU should unlock console_sem without pushing all
2605  * messages to the console. This reduces the chance that the console is
2606  * locked when the panic CPU tries to use it.
2607  */
abandon_console_lock_in_panic(void)2608 static bool abandon_console_lock_in_panic(void)
2609 {
2610 	if (!panic_in_progress())
2611 		return false;
2612 
2613 	/*
2614 	 * We can use raw_smp_processor_id() here because it is impossible for
2615 	 * the task to be migrated to the panic_cpu, or away from it. If
2616 	 * panic_cpu has already been set, and we're not currently executing on
2617 	 * that CPU, then we never will be.
2618 	 */
2619 	return atomic_read(&panic_cpu) != raw_smp_processor_id();
2620 }
2621 
2622 /*
2623  * Check if the given console is currently capable and allowed to print
2624  * records.
2625  *
2626  * Requires the console_lock.
2627  */
console_is_usable(struct console * con)2628 static inline bool console_is_usable(struct console *con)
2629 {
2630 	if (!(con->flags & CON_ENABLED))
2631 		return false;
2632 
2633 	if (!con->write)
2634 		return false;
2635 
2636 	/*
2637 	 * Console drivers may assume that per-cpu resources have been
2638 	 * allocated. So unless they're explicitly marked as being able to
2639 	 * cope (CON_ANYTIME) don't call them until this CPU is officially up.
2640 	 */
2641 	if (!cpu_online(raw_smp_processor_id()) &&
2642 	    !(con->flags & CON_ANYTIME))
2643 		return false;
2644 
2645 	return true;
2646 }
2647 
__console_unlock(void)2648 static void __console_unlock(void)
2649 {
2650 	console_locked = 0;
2651 	up_console_sem();
2652 }
2653 
2654 /*
2655  * Print one record for the given console. The record printed is whatever
2656  * record is the next available record for the given console.
2657  *
2658  * @text is a buffer of size CONSOLE_LOG_MAX.
2659  *
2660  * If extended messages should be printed, @ext_text is a buffer of size
2661  * CONSOLE_EXT_LOG_MAX. Otherwise @ext_text must be NULL.
2662  *
2663  * If dropped messages should be printed, @dropped_text is a buffer of size
2664  * DROPPED_TEXT_MAX. Otherwise @dropped_text must be NULL.
2665  *
2666  * @handover will be set to true if a printk waiter has taken over the
2667  * console_lock, in which case the caller is no longer holding the
2668  * console_lock. Otherwise it is set to false.
2669  *
2670  * Returns false if the given console has no next record to print, otherwise
2671  * true.
2672  *
2673  * Requires the console_lock.
2674  */
console_emit_next_record(struct console * con,char * text,char * ext_text,char * dropped_text,bool * handover)2675 static bool console_emit_next_record(struct console *con, char *text, char *ext_text,
2676 				     char *dropped_text, bool *handover)
2677 {
2678 	static int panic_console_dropped;
2679 	struct printk_info info;
2680 	struct printk_record r;
2681 	unsigned long flags;
2682 	char *write_text;
2683 	size_t len;
2684 
2685 	prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
2686 
2687 	*handover = false;
2688 
2689 	if (!prb_read_valid(prb, con->seq, &r))
2690 		return false;
2691 
2692 	if (con->seq != r.info->seq) {
2693 		con->dropped += r.info->seq - con->seq;
2694 		con->seq = r.info->seq;
2695 		if (panic_in_progress() && panic_console_dropped++ > 10) {
2696 			suppress_panic_printk = 1;
2697 			pr_warn_once("Too many dropped messages. Suppress messages on non-panic CPUs to prevent livelock.\n");
2698 		}
2699 	}
2700 
2701 	/* Skip record that has level above the console loglevel. */
2702 	if (suppress_message_printing(r.info->level)) {
2703 		con->seq++;
2704 		goto skip;
2705 	}
2706 
2707 	if (ext_text) {
2708 		write_text = ext_text;
2709 		len = info_print_ext_header(ext_text, CONSOLE_EXT_LOG_MAX, r.info);
2710 		len += msg_print_ext_body(ext_text + len, CONSOLE_EXT_LOG_MAX - len,
2711 					  &r.text_buf[0], r.info->text_len, &r.info->dev_info);
2712 	} else {
2713 		write_text = text;
2714 		len = record_print_text(&r, console_msg_format & MSG_FORMAT_SYSLOG, printk_time);
2715 	}
2716 
2717 	/*
2718 	 * While actively printing out messages, if another printk()
2719 	 * were to occur on another CPU, it may wait for this one to
2720 	 * finish. This task can not be preempted if there is a
2721 	 * waiter waiting to take over.
2722 	 *
2723 	 * Interrupts are disabled because the hand over to a waiter
2724 	 * must not be interrupted until the hand over is completed
2725 	 * (@console_waiter is cleared).
2726 	 */
2727 	printk_safe_enter_irqsave(flags);
2728 	console_lock_spinning_enable();
2729 
2730 	stop_critical_timings();	/* don't trace print latency */
2731 	call_console_driver(con, write_text, len, dropped_text);
2732 	start_critical_timings();
2733 
2734 	con->seq++;
2735 
2736 	*handover = console_lock_spinning_disable_and_check();
2737 	printk_safe_exit_irqrestore(flags);
2738 skip:
2739 	return true;
2740 }
2741 
2742 /*
2743  * Print out all remaining records to all consoles.
2744  *
2745  * @do_cond_resched is set by the caller. It can be true only in schedulable
2746  * context.
2747  *
2748  * @next_seq is set to the sequence number after the last available record.
2749  * The value is valid only when this function returns true. It means that all
2750  * usable consoles are completely flushed.
2751  *
2752  * @handover will be set to true if a printk waiter has taken over the
2753  * console_lock, in which case the caller is no longer holding the
2754  * console_lock. Otherwise it is set to false.
2755  *
2756  * Returns true when there was at least one usable console and all messages
2757  * were flushed to all usable consoles. A returned false informs the caller
2758  * that everything was not flushed (either there were no usable consoles or
2759  * another context has taken over printing or it is a panic situation and this
2760  * is not the panic CPU). Regardless the reason, the caller should assume it
2761  * is not useful to immediately try again.
2762  *
2763  * Requires the console_lock.
2764  */
console_flush_all(bool do_cond_resched,u64 * next_seq,bool * handover)2765 static bool console_flush_all(bool do_cond_resched, u64 *next_seq, bool *handover)
2766 {
2767 	static char dropped_text[DROPPED_TEXT_MAX];
2768 	static char ext_text[CONSOLE_EXT_LOG_MAX];
2769 	static char text[CONSOLE_LOG_MAX];
2770 	bool any_usable = false;
2771 	struct console *con;
2772 	bool any_progress;
2773 
2774 	*next_seq = 0;
2775 	*handover = false;
2776 
2777 	do {
2778 		any_progress = false;
2779 
2780 		for_each_console(con) {
2781 			bool progress;
2782 
2783 			if (!console_is_usable(con))
2784 				continue;
2785 			any_usable = true;
2786 
2787 			if (con->flags & CON_EXTENDED) {
2788 				/* Extended consoles do not print "dropped messages". */
2789 				progress = console_emit_next_record(con, &text[0],
2790 								    &ext_text[0], NULL,
2791 								    handover);
2792 			} else {
2793 				progress = console_emit_next_record(con, &text[0],
2794 								    NULL, &dropped_text[0],
2795 								    handover);
2796 			}
2797 			if (*handover)
2798 				return false;
2799 
2800 			/* Track the next of the highest seq flushed. */
2801 			if (con->seq > *next_seq)
2802 				*next_seq = con->seq;
2803 
2804 			if (!progress)
2805 				continue;
2806 			any_progress = true;
2807 
2808 			/* Allow panic_cpu to take over the consoles safely. */
2809 			if (abandon_console_lock_in_panic())
2810 				return false;
2811 
2812 			if (do_cond_resched)
2813 				cond_resched();
2814 		}
2815 	} while (any_progress);
2816 
2817 	return any_usable;
2818 }
2819 
2820 /**
2821  * console_unlock - unlock the console system
2822  *
2823  * Releases the console_lock which the caller holds on the console system
2824  * and the console driver list.
2825  *
2826  * While the console_lock was held, console output may have been buffered
2827  * by printk().  If this is the case, console_unlock(); emits
2828  * the output prior to releasing the lock.
2829  *
2830  * console_unlock(); may be called from any context.
2831  */
console_unlock(void)2832 void console_unlock(void)
2833 {
2834 	bool do_cond_resched;
2835 	bool handover;
2836 	bool flushed;
2837 	u64 next_seq;
2838 
2839 	if (console_suspended) {
2840 		up_console_sem();
2841 		return;
2842 	}
2843 
2844 	/*
2845 	 * Console drivers are called with interrupts disabled, so
2846 	 * @console_may_schedule should be cleared before; however, we may
2847 	 * end up dumping a lot of lines, for example, if called from
2848 	 * console registration path, and should invoke cond_resched()
2849 	 * between lines if allowable.  Not doing so can cause a very long
2850 	 * scheduling stall on a slow console leading to RCU stall and
2851 	 * softlockup warnings which exacerbate the issue with more
2852 	 * messages practically incapacitating the system. Therefore, create
2853 	 * a local to use for the printing loop.
2854 	 */
2855 	do_cond_resched = console_may_schedule;
2856 
2857 	do {
2858 		console_may_schedule = 0;
2859 
2860 		flushed = console_flush_all(do_cond_resched, &next_seq, &handover);
2861 		if (!handover)
2862 			__console_unlock();
2863 
2864 		/*
2865 		 * Abort if there was a failure to flush all messages to all
2866 		 * usable consoles. Either it is not possible to flush (in
2867 		 * which case it would be an infinite loop of retrying) or
2868 		 * another context has taken over printing.
2869 		 */
2870 		if (!flushed)
2871 			break;
2872 
2873 		/*
2874 		 * Some context may have added new records after
2875 		 * console_flush_all() but before unlocking the console.
2876 		 * Re-check if there is a new record to flush. If the trylock
2877 		 * fails, another context is already handling the printing.
2878 		 */
2879 	} while (prb_read_valid(prb, next_seq, NULL) && console_trylock());
2880 }
2881 EXPORT_SYMBOL(console_unlock);
2882 
2883 /**
2884  * console_conditional_schedule - yield the CPU if required
2885  *
2886  * If the console code is currently allowed to sleep, and
2887  * if this CPU should yield the CPU to another task, do
2888  * so here.
2889  *
2890  * Must be called within console_lock();.
2891  */
console_conditional_schedule(void)2892 void __sched console_conditional_schedule(void)
2893 {
2894 	if (console_may_schedule)
2895 		cond_resched();
2896 }
2897 EXPORT_SYMBOL(console_conditional_schedule);
2898 
console_unblank(void)2899 void console_unblank(void)
2900 {
2901 	struct console *c;
2902 
2903 	/*
2904 	 * console_unblank can no longer be called in interrupt context unless
2905 	 * oops_in_progress is set to 1..
2906 	 */
2907 	if (oops_in_progress) {
2908 		if (down_trylock_console_sem() != 0)
2909 			return;
2910 	} else
2911 		console_lock();
2912 
2913 	console_locked = 1;
2914 	console_may_schedule = 0;
2915 	for_each_console(c)
2916 		if ((c->flags & CON_ENABLED) && c->unblank)
2917 			c->unblank();
2918 	console_unlock();
2919 
2920 	if (!oops_in_progress)
2921 		pr_flush(1000, true);
2922 }
2923 
2924 /**
2925  * console_flush_on_panic - flush console content on panic
2926  * @mode: flush all messages in buffer or just the pending ones
2927  *
2928  * Immediately output all pending messages no matter what.
2929  */
console_flush_on_panic(enum con_flush_mode mode)2930 void console_flush_on_panic(enum con_flush_mode mode)
2931 {
2932 	/*
2933 	 * If someone else is holding the console lock, trylock will fail
2934 	 * and may_schedule may be set.  Ignore and proceed to unlock so
2935 	 * that messages are flushed out.  As this can be called from any
2936 	 * context and we don't want to get preempted while flushing,
2937 	 * ensure may_schedule is cleared.
2938 	 */
2939 	console_trylock();
2940 	console_may_schedule = 0;
2941 
2942 	if (mode == CONSOLE_REPLAY_ALL) {
2943 		struct console *c;
2944 		u64 seq;
2945 
2946 		seq = prb_first_valid_seq(prb);
2947 		for_each_console(c)
2948 			c->seq = seq;
2949 	}
2950 	console_unlock();
2951 }
2952 
2953 /*
2954  * Return the console tty driver structure and its associated index
2955  */
console_device(int * index)2956 struct tty_driver *console_device(int *index)
2957 {
2958 	struct console *c;
2959 	struct tty_driver *driver = NULL;
2960 
2961 	console_lock();
2962 	for_each_console(c) {
2963 		if (!c->device)
2964 			continue;
2965 		driver = c->device(c, index);
2966 		if (driver)
2967 			break;
2968 	}
2969 	console_unlock();
2970 	return driver;
2971 }
2972 
2973 /*
2974  * Prevent further output on the passed console device so that (for example)
2975  * serial drivers can disable console output before suspending a port, and can
2976  * re-enable output afterwards.
2977  */
console_stop(struct console * console)2978 void console_stop(struct console *console)
2979 {
2980 	__pr_flush(console, 1000, true);
2981 	console_lock();
2982 	console->flags &= ~CON_ENABLED;
2983 	console_unlock();
2984 }
2985 EXPORT_SYMBOL(console_stop);
2986 
console_start(struct console * console)2987 void console_start(struct console *console)
2988 {
2989 	console_lock();
2990 	console->flags |= CON_ENABLED;
2991 	console_unlock();
2992 	__pr_flush(console, 1000, true);
2993 }
2994 EXPORT_SYMBOL(console_start);
2995 
2996 static int __read_mostly keep_bootcon;
2997 
keep_bootcon_setup(char * str)2998 static int __init keep_bootcon_setup(char *str)
2999 {
3000 	keep_bootcon = 1;
3001 	pr_info("debug: skip boot console de-registration.\n");
3002 
3003 	return 0;
3004 }
3005 
3006 early_param("keep_bootcon", keep_bootcon_setup);
3007 
3008 /*
3009  * This is called by register_console() to try to match
3010  * the newly registered console with any of the ones selected
3011  * by either the command line or add_preferred_console() and
3012  * setup/enable it.
3013  *
3014  * Care need to be taken with consoles that are statically
3015  * enabled such as netconsole
3016  */
try_enable_preferred_console(struct console * newcon,bool user_specified)3017 static int try_enable_preferred_console(struct console *newcon,
3018 					bool user_specified)
3019 {
3020 	struct console_cmdline *c;
3021 	int i, err;
3022 
3023 	for (i = 0, c = console_cmdline;
3024 	     i < MAX_CMDLINECONSOLES && c->name[0];
3025 	     i++, c++) {
3026 		if (c->user_specified != user_specified)
3027 			continue;
3028 		if (!newcon->match ||
3029 		    newcon->match(newcon, c->name, c->index, c->options) != 0) {
3030 			/* default matching */
3031 			BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
3032 			if (strcmp(c->name, newcon->name) != 0)
3033 				continue;
3034 			if (newcon->index >= 0 &&
3035 			    newcon->index != c->index)
3036 				continue;
3037 			if (newcon->index < 0)
3038 				newcon->index = c->index;
3039 
3040 			if (_braille_register_console(newcon, c))
3041 				return 0;
3042 
3043 			if (newcon->setup &&
3044 			    (err = newcon->setup(newcon, c->options)) != 0)
3045 				return err;
3046 		}
3047 		newcon->flags |= CON_ENABLED;
3048 		if (i == preferred_console)
3049 			newcon->flags |= CON_CONSDEV;
3050 		return 0;
3051 	}
3052 
3053 	/*
3054 	 * Some consoles, such as pstore and netconsole, can be enabled even
3055 	 * without matching. Accept the pre-enabled consoles only when match()
3056 	 * and setup() had a chance to be called.
3057 	 */
3058 	if (newcon->flags & CON_ENABLED && c->user_specified ==	user_specified)
3059 		return 0;
3060 
3061 	return -ENOENT;
3062 }
3063 
3064 /* Try to enable the console unconditionally */
try_enable_default_console(struct console * newcon)3065 static void try_enable_default_console(struct console *newcon)
3066 {
3067 	if (newcon->index < 0)
3068 		newcon->index = 0;
3069 
3070 	if (newcon->setup && newcon->setup(newcon, NULL) != 0)
3071 		return;
3072 
3073 	newcon->flags |= CON_ENABLED;
3074 
3075 	if (newcon->device)
3076 		newcon->flags |= CON_CONSDEV;
3077 }
3078 
3079 #define con_printk(lvl, con, fmt, ...)			\
3080 	printk(lvl pr_fmt("%sconsole [%s%d] " fmt),	\
3081 	       (con->flags & CON_BOOT) ? "boot" : "",	\
3082 	       con->name, con->index, ##__VA_ARGS__)
3083 
3084 /*
3085  * The console driver calls this routine during kernel initialization
3086  * to register the console printing procedure with printk() and to
3087  * print any messages that were printed by the kernel before the
3088  * console driver was initialized.
3089  *
3090  * This can happen pretty early during the boot process (because of
3091  * early_printk) - sometimes before setup_arch() completes - be careful
3092  * of what kernel features are used - they may not be initialised yet.
3093  *
3094  * There are two types of consoles - bootconsoles (early_printk) and
3095  * "real" consoles (everything which is not a bootconsole) which are
3096  * handled differently.
3097  *  - Any number of bootconsoles can be registered at any time.
3098  *  - As soon as a "real" console is registered, all bootconsoles
3099  *    will be unregistered automatically.
3100  *  - Once a "real" console is registered, any attempt to register a
3101  *    bootconsoles will be rejected
3102  */
register_console(struct console * newcon)3103 void register_console(struct console *newcon)
3104 {
3105 	struct console *con;
3106 	bool bootcon_enabled = false;
3107 	bool realcon_enabled = false;
3108 	int err;
3109 
3110 	for_each_console(con) {
3111 		if (WARN(con == newcon, "console '%s%d' already registered\n",
3112 					 con->name, con->index))
3113 			return;
3114 	}
3115 
3116 	for_each_console(con) {
3117 		if (con->flags & CON_BOOT)
3118 			bootcon_enabled = true;
3119 		else
3120 			realcon_enabled = true;
3121 	}
3122 
3123 	/* Do not register boot consoles when there already is a real one. */
3124 	if (newcon->flags & CON_BOOT && realcon_enabled) {
3125 		pr_info("Too late to register bootconsole %s%d\n",
3126 			newcon->name, newcon->index);
3127 		return;
3128 	}
3129 
3130 	/*
3131 	 * See if we want to enable this console driver by default.
3132 	 *
3133 	 * Nope when a console is preferred by the command line, device
3134 	 * tree, or SPCR.
3135 	 *
3136 	 * The first real console with tty binding (driver) wins. More
3137 	 * consoles might get enabled before the right one is found.
3138 	 *
3139 	 * Note that a console with tty binding will have CON_CONSDEV
3140 	 * flag set and will be first in the list.
3141 	 */
3142 	if (preferred_console < 0) {
3143 		if (!console_drivers || !console_drivers->device ||
3144 		    console_drivers->flags & CON_BOOT) {
3145 			try_enable_default_console(newcon);
3146 		}
3147 	}
3148 
3149 	/* See if this console matches one we selected on the command line */
3150 	err = try_enable_preferred_console(newcon, true);
3151 
3152 	/* If not, try to match against the platform default(s) */
3153 	if (err == -ENOENT)
3154 		err = try_enable_preferred_console(newcon, false);
3155 
3156 	/* printk() messages are not printed to the Braille console. */
3157 	if (err || newcon->flags & CON_BRL)
3158 		return;
3159 
3160 	/*
3161 	 * If we have a bootconsole, and are switching to a real console,
3162 	 * don't print everything out again, since when the boot console, and
3163 	 * the real console are the same physical device, it's annoying to
3164 	 * see the beginning boot messages twice
3165 	 */
3166 	if (bootcon_enabled &&
3167 	    ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
3168 		newcon->flags &= ~CON_PRINTBUFFER;
3169 	}
3170 
3171 	/*
3172 	 *	Put this console in the list - keep the
3173 	 *	preferred driver at the head of the list.
3174 	 */
3175 	console_lock();
3176 	if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
3177 		newcon->next = console_drivers;
3178 		console_drivers = newcon;
3179 		if (newcon->next)
3180 			newcon->next->flags &= ~CON_CONSDEV;
3181 		/* Ensure this flag is always set for the head of the list */
3182 		newcon->flags |= CON_CONSDEV;
3183 	} else {
3184 		newcon->next = console_drivers->next;
3185 		console_drivers->next = newcon;
3186 	}
3187 
3188 	newcon->dropped = 0;
3189 	if (newcon->flags & CON_PRINTBUFFER) {
3190 		/* Get a consistent copy of @syslog_seq. */
3191 		mutex_lock(&syslog_lock);
3192 		newcon->seq = syslog_seq;
3193 		mutex_unlock(&syslog_lock);
3194 	} else {
3195 		/* Begin with next message. */
3196 		newcon->seq = prb_next_seq(prb);
3197 	}
3198 	console_unlock();
3199 	console_sysfs_notify();
3200 
3201 	/*
3202 	 * By unregistering the bootconsoles after we enable the real console
3203 	 * we get the "console xxx enabled" message on all the consoles -
3204 	 * boot consoles, real consoles, etc - this is to ensure that end
3205 	 * users know there might be something in the kernel's log buffer that
3206 	 * went to the bootconsole (that they do not see on the real console)
3207 	 */
3208 	con_printk(KERN_INFO, newcon, "enabled\n");
3209 	if (bootcon_enabled &&
3210 	    ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
3211 	    !keep_bootcon) {
3212 		for_each_console(con)
3213 			if (con->flags & CON_BOOT)
3214 				unregister_console(con);
3215 	}
3216 }
3217 EXPORT_SYMBOL(register_console);
3218 
unregister_console(struct console * console)3219 int unregister_console(struct console *console)
3220 {
3221 	struct console *con;
3222 	int res;
3223 
3224 	con_printk(KERN_INFO, console, "disabled\n");
3225 
3226 	res = _braille_unregister_console(console);
3227 	if (res < 0)
3228 		return res;
3229 	if (res > 0)
3230 		return 0;
3231 
3232 	res = -ENODEV;
3233 	console_lock();
3234 	if (console_drivers == console) {
3235 		console_drivers=console->next;
3236 		res = 0;
3237 	} else {
3238 		for_each_console(con) {
3239 			if (con->next == console) {
3240 				con->next = console->next;
3241 				res = 0;
3242 				break;
3243 			}
3244 		}
3245 	}
3246 
3247 	if (res)
3248 		goto out_disable_unlock;
3249 
3250 	/*
3251 	 * If this isn't the last console and it has CON_CONSDEV set, we
3252 	 * need to set it on the next preferred console.
3253 	 */
3254 	if (console_drivers != NULL && console->flags & CON_CONSDEV)
3255 		console_drivers->flags |= CON_CONSDEV;
3256 
3257 	console->flags &= ~CON_ENABLED;
3258 	console_unlock();
3259 	console_sysfs_notify();
3260 
3261 	if (console->exit)
3262 		res = console->exit(console);
3263 
3264 	return res;
3265 
3266 out_disable_unlock:
3267 	console->flags &= ~CON_ENABLED;
3268 	console_unlock();
3269 
3270 	return res;
3271 }
3272 EXPORT_SYMBOL(unregister_console);
3273 
3274 /*
3275  * Initialize the console device. This is called *early*, so
3276  * we can't necessarily depend on lots of kernel help here.
3277  * Just do some early initializations, and do the complex setup
3278  * later.
3279  */
console_init(void)3280 void __init console_init(void)
3281 {
3282 	int ret;
3283 	initcall_t call;
3284 	initcall_entry_t *ce;
3285 
3286 	/* Setup the default TTY line discipline. */
3287 	n_tty_init();
3288 
3289 	/*
3290 	 * set up the console device so that later boot sequences can
3291 	 * inform about problems etc..
3292 	 */
3293 	ce = __con_initcall_start;
3294 	trace_initcall_level("console");
3295 	while (ce < __con_initcall_end) {
3296 		call = initcall_from_entry(ce);
3297 		trace_initcall_start(call);
3298 		ret = call();
3299 		trace_initcall_finish(call, ret);
3300 		ce++;
3301 	}
3302 }
3303 
3304 /*
3305  * Some boot consoles access data that is in the init section and which will
3306  * be discarded after the initcalls have been run. To make sure that no code
3307  * will access this data, unregister the boot consoles in a late initcall.
3308  *
3309  * If for some reason, such as deferred probe or the driver being a loadable
3310  * module, the real console hasn't registered yet at this point, there will
3311  * be a brief interval in which no messages are logged to the console, which
3312  * makes it difficult to diagnose problems that occur during this time.
3313  *
3314  * To mitigate this problem somewhat, only unregister consoles whose memory
3315  * intersects with the init section. Note that all other boot consoles will
3316  * get unregistered when the real preferred console is registered.
3317  */
printk_late_init(void)3318 static int __init printk_late_init(void)
3319 {
3320 	struct console *con;
3321 	int ret;
3322 
3323 	for_each_console(con) {
3324 		if (!(con->flags & CON_BOOT))
3325 			continue;
3326 
3327 		/* Check addresses that might be used for enabled consoles. */
3328 		if (init_section_intersects(con, sizeof(*con)) ||
3329 		    init_section_contains(con->write, 0) ||
3330 		    init_section_contains(con->read, 0) ||
3331 		    init_section_contains(con->device, 0) ||
3332 		    init_section_contains(con->unblank, 0) ||
3333 		    init_section_contains(con->data, 0)) {
3334 			/*
3335 			 * Please, consider moving the reported consoles out
3336 			 * of the init section.
3337 			 */
3338 			pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3339 				con->name, con->index);
3340 			unregister_console(con);
3341 		}
3342 	}
3343 	ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3344 					console_cpu_notify);
3345 	WARN_ON(ret < 0);
3346 	ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3347 					console_cpu_notify, NULL);
3348 	WARN_ON(ret < 0);
3349 	printk_sysctl_init();
3350 	return 0;
3351 }
3352 late_initcall(printk_late_init);
3353 
3354 #if defined CONFIG_PRINTK
3355 /* If @con is specified, only wait for that console. Otherwise wait for all. */
__pr_flush(struct console * con,int timeout_ms,bool reset_on_progress)3356 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
3357 {
3358 	int remaining = timeout_ms;
3359 	struct console *c;
3360 	u64 last_diff = 0;
3361 	u64 printk_seq;
3362 	u64 diff;
3363 	u64 seq;
3364 
3365 	might_sleep();
3366 
3367 	seq = prb_next_seq(prb);
3368 
3369 	for (;;) {
3370 		diff = 0;
3371 
3372 		console_lock();
3373 
3374 		for_each_console(c) {
3375 			if (con && con != c)
3376 				continue;
3377 			if (!console_is_usable(c))
3378 				continue;
3379 			printk_seq = c->seq;
3380 			if (printk_seq < seq)
3381 				diff += seq - printk_seq;
3382 		}
3383 
3384 		/*
3385 		 * If consoles are suspended, it cannot be expected that they
3386 		 * make forward progress, so timeout immediately. @diff is
3387 		 * still used to return a valid flush status.
3388 		 */
3389 		if (console_suspended)
3390 			remaining = 0;
3391 		else if (diff != last_diff && reset_on_progress)
3392 			remaining = timeout_ms;
3393 
3394 		console_unlock();
3395 
3396 		if (diff == 0 || remaining == 0)
3397 			break;
3398 
3399 		if (remaining < 0) {
3400 			/* no timeout limit */
3401 			msleep(100);
3402 		} else if (remaining < 100) {
3403 			msleep(remaining);
3404 			remaining = 0;
3405 		} else {
3406 			msleep(100);
3407 			remaining -= 100;
3408 		}
3409 
3410 		last_diff = diff;
3411 	}
3412 
3413 	return (diff == 0);
3414 }
3415 
3416 /**
3417  * pr_flush() - Wait for printing threads to catch up.
3418  *
3419  * @timeout_ms:        The maximum time (in ms) to wait.
3420  * @reset_on_progress: Reset the timeout if forward progress is seen.
3421  *
3422  * A value of 0 for @timeout_ms means no waiting will occur. A value of -1
3423  * represents infinite waiting.
3424  *
3425  * If @reset_on_progress is true, the timeout will be reset whenever any
3426  * printer has been seen to make some forward progress.
3427  *
3428  * Context: Process context. May sleep while acquiring console lock.
3429  * Return: true if all enabled printers are caught up.
3430  */
pr_flush(int timeout_ms,bool reset_on_progress)3431 static bool pr_flush(int timeout_ms, bool reset_on_progress)
3432 {
3433 	return __pr_flush(NULL, timeout_ms, reset_on_progress);
3434 }
3435 
3436 /*
3437  * Delayed printk version, for scheduler-internal messages:
3438  */
3439 #define PRINTK_PENDING_WAKEUP	0x01
3440 #define PRINTK_PENDING_OUTPUT	0x02
3441 
3442 static DEFINE_PER_CPU(int, printk_pending);
3443 
wake_up_klogd_work_func(struct irq_work * irq_work)3444 static void wake_up_klogd_work_func(struct irq_work *irq_work)
3445 {
3446 	int pending = this_cpu_xchg(printk_pending, 0);
3447 
3448 	if (pending & PRINTK_PENDING_OUTPUT) {
3449 		/* If trylock fails, someone else is doing the printing */
3450 		if (console_trylock())
3451 			console_unlock();
3452 	}
3453 
3454 	if (pending & PRINTK_PENDING_WAKEUP)
3455 		wake_up_interruptible(&log_wait);
3456 }
3457 
3458 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3459 	IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3460 
__wake_up_klogd(int val)3461 static void __wake_up_klogd(int val)
3462 {
3463 	if (!printk_percpu_data_ready())
3464 		return;
3465 
3466 	preempt_disable();
3467 	/*
3468 	 * Guarantee any new records can be seen by tasks preparing to wait
3469 	 * before this context checks if the wait queue is empty.
3470 	 *
3471 	 * The full memory barrier within wq_has_sleeper() pairs with the full
3472 	 * memory barrier within set_current_state() of
3473 	 * prepare_to_wait_event(), which is called after ___wait_event() adds
3474 	 * the waiter but before it has checked the wait condition.
3475 	 *
3476 	 * This pairs with devkmsg_read:A and syslog_print:A.
3477 	 */
3478 	if (wq_has_sleeper(&log_wait) || /* LMM(__wake_up_klogd:A) */
3479 	    (val & PRINTK_PENDING_OUTPUT)) {
3480 		this_cpu_or(printk_pending, val);
3481 		irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3482 	}
3483 	preempt_enable();
3484 }
3485 
wake_up_klogd(void)3486 void wake_up_klogd(void)
3487 {
3488 	__wake_up_klogd(PRINTK_PENDING_WAKEUP);
3489 }
3490 
defer_console_output(void)3491 void defer_console_output(void)
3492 {
3493 	/*
3494 	 * New messages may have been added directly to the ringbuffer
3495 	 * using vprintk_store(), so wake any waiters as well.
3496 	 */
3497 	__wake_up_klogd(PRINTK_PENDING_WAKEUP | PRINTK_PENDING_OUTPUT);
3498 }
3499 
printk_trigger_flush(void)3500 void printk_trigger_flush(void)
3501 {
3502 	defer_console_output();
3503 }
3504 
vprintk_deferred(const char * fmt,va_list args)3505 int vprintk_deferred(const char *fmt, va_list args)
3506 {
3507 	int r;
3508 
3509 	r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3510 	defer_console_output();
3511 
3512 	return r;
3513 }
3514 
_printk_deferred(const char * fmt,...)3515 int _printk_deferred(const char *fmt, ...)
3516 {
3517 	va_list args;
3518 	int r;
3519 
3520 	va_start(args, fmt);
3521 	r = vprintk_deferred(fmt, args);
3522 	va_end(args);
3523 
3524 	return r;
3525 }
3526 
3527 /*
3528  * printk rate limiting, lifted from the networking subsystem.
3529  *
3530  * This enforces a rate limit: not more than 10 kernel messages
3531  * every 5s to make a denial-of-service attack impossible.
3532  */
3533 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3534 
__printk_ratelimit(const char * func)3535 int __printk_ratelimit(const char *func)
3536 {
3537 	return ___ratelimit(&printk_ratelimit_state, func);
3538 }
3539 EXPORT_SYMBOL(__printk_ratelimit);
3540 
3541 /**
3542  * printk_timed_ratelimit - caller-controlled printk ratelimiting
3543  * @caller_jiffies: pointer to caller's state
3544  * @interval_msecs: minimum interval between prints
3545  *
3546  * printk_timed_ratelimit() returns true if more than @interval_msecs
3547  * milliseconds have elapsed since the last time printk_timed_ratelimit()
3548  * returned true.
3549  */
printk_timed_ratelimit(unsigned long * caller_jiffies,unsigned int interval_msecs)3550 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3551 			unsigned int interval_msecs)
3552 {
3553 	unsigned long elapsed = jiffies - *caller_jiffies;
3554 
3555 	if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3556 		return false;
3557 
3558 	*caller_jiffies = jiffies;
3559 	return true;
3560 }
3561 EXPORT_SYMBOL(printk_timed_ratelimit);
3562 
3563 static DEFINE_SPINLOCK(dump_list_lock);
3564 static LIST_HEAD(dump_list);
3565 
3566 /**
3567  * kmsg_dump_register - register a kernel log dumper.
3568  * @dumper: pointer to the kmsg_dumper structure
3569  *
3570  * Adds a kernel log dumper to the system. The dump callback in the
3571  * structure will be called when the kernel oopses or panics and must be
3572  * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3573  */
kmsg_dump_register(struct kmsg_dumper * dumper)3574 int kmsg_dump_register(struct kmsg_dumper *dumper)
3575 {
3576 	unsigned long flags;
3577 	int err = -EBUSY;
3578 
3579 	/* The dump callback needs to be set */
3580 	if (!dumper->dump)
3581 		return -EINVAL;
3582 
3583 	spin_lock_irqsave(&dump_list_lock, flags);
3584 	/* Don't allow registering multiple times */
3585 	if (!dumper->registered) {
3586 		dumper->registered = 1;
3587 		list_add_tail_rcu(&dumper->list, &dump_list);
3588 		err = 0;
3589 	}
3590 	spin_unlock_irqrestore(&dump_list_lock, flags);
3591 
3592 	return err;
3593 }
3594 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3595 
3596 /**
3597  * kmsg_dump_unregister - unregister a kmsg dumper.
3598  * @dumper: pointer to the kmsg_dumper structure
3599  *
3600  * Removes a dump device from the system. Returns zero on success and
3601  * %-EINVAL otherwise.
3602  */
kmsg_dump_unregister(struct kmsg_dumper * dumper)3603 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3604 {
3605 	unsigned long flags;
3606 	int err = -EINVAL;
3607 
3608 	spin_lock_irqsave(&dump_list_lock, flags);
3609 	if (dumper->registered) {
3610 		dumper->registered = 0;
3611 		list_del_rcu(&dumper->list);
3612 		err = 0;
3613 	}
3614 	spin_unlock_irqrestore(&dump_list_lock, flags);
3615 	synchronize_rcu();
3616 
3617 	return err;
3618 }
3619 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3620 
3621 static bool always_kmsg_dump;
3622 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3623 
kmsg_dump_reason_str(enum kmsg_dump_reason reason)3624 const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
3625 {
3626 	switch (reason) {
3627 	case KMSG_DUMP_PANIC:
3628 		return "Panic";
3629 	case KMSG_DUMP_OOPS:
3630 		return "Oops";
3631 	case KMSG_DUMP_EMERG:
3632 		return "Emergency";
3633 	case KMSG_DUMP_SHUTDOWN:
3634 		return "Shutdown";
3635 	default:
3636 		return "Unknown";
3637 	}
3638 }
3639 EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
3640 
3641 /**
3642  * kmsg_dump - dump kernel log to kernel message dumpers.
3643  * @reason: the reason (oops, panic etc) for dumping
3644  *
3645  * Call each of the registered dumper's dump() callback, which can
3646  * retrieve the kmsg records with kmsg_dump_get_line() or
3647  * kmsg_dump_get_buffer().
3648  */
kmsg_dump(enum kmsg_dump_reason reason)3649 void kmsg_dump(enum kmsg_dump_reason reason)
3650 {
3651 	struct kmsg_dumper *dumper;
3652 
3653 	rcu_read_lock();
3654 	list_for_each_entry_rcu(dumper, &dump_list, list) {
3655 		enum kmsg_dump_reason max_reason = dumper->max_reason;
3656 
3657 		/*
3658 		 * If client has not provided a specific max_reason, default
3659 		 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
3660 		 */
3661 		if (max_reason == KMSG_DUMP_UNDEF) {
3662 			max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
3663 							KMSG_DUMP_OOPS;
3664 		}
3665 		if (reason > max_reason)
3666 			continue;
3667 
3668 		/* invoke dumper which will iterate over records */
3669 		dumper->dump(dumper, reason);
3670 	}
3671 	rcu_read_unlock();
3672 }
3673 
3674 /**
3675  * kmsg_dump_get_line - retrieve one kmsg log line
3676  * @iter: kmsg dump iterator
3677  * @syslog: include the "<4>" prefixes
3678  * @line: buffer to copy the line to
3679  * @size: maximum size of the buffer
3680  * @len: length of line placed into buffer
3681  *
3682  * Start at the beginning of the kmsg buffer, with the oldest kmsg
3683  * record, and copy one record into the provided buffer.
3684  *
3685  * Consecutive calls will return the next available record moving
3686  * towards the end of the buffer with the youngest messages.
3687  *
3688  * A return value of FALSE indicates that there are no more records to
3689  * read.
3690  */
kmsg_dump_get_line(struct kmsg_dump_iter * iter,bool syslog,char * line,size_t size,size_t * len)3691 bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
3692 			char *line, size_t size, size_t *len)
3693 {
3694 	u64 min_seq = latched_seq_read_nolock(&clear_seq);
3695 	struct printk_info info;
3696 	unsigned int line_count;
3697 	struct printk_record r;
3698 	size_t l = 0;
3699 	bool ret = false;
3700 
3701 	if (iter->cur_seq < min_seq)
3702 		iter->cur_seq = min_seq;
3703 
3704 	prb_rec_init_rd(&r, &info, line, size);
3705 
3706 	/* Read text or count text lines? */
3707 	if (line) {
3708 		if (!prb_read_valid(prb, iter->cur_seq, &r))
3709 			goto out;
3710 		l = record_print_text(&r, syslog, printk_time);
3711 	} else {
3712 		if (!prb_read_valid_info(prb, iter->cur_seq,
3713 					 &info, &line_count)) {
3714 			goto out;
3715 		}
3716 		l = get_record_print_text_size(&info, line_count, syslog,
3717 					       printk_time);
3718 
3719 	}
3720 
3721 	iter->cur_seq = r.info->seq + 1;
3722 	ret = true;
3723 out:
3724 	if (len)
3725 		*len = l;
3726 	return ret;
3727 }
3728 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3729 
3730 /**
3731  * kmsg_dump_get_buffer - copy kmsg log lines
3732  * @iter: kmsg dump iterator
3733  * @syslog: include the "<4>" prefixes
3734  * @buf: buffer to copy the line to
3735  * @size: maximum size of the buffer
3736  * @len_out: length of line placed into buffer
3737  *
3738  * Start at the end of the kmsg buffer and fill the provided buffer
3739  * with as many of the *youngest* kmsg records that fit into it.
3740  * If the buffer is large enough, all available kmsg records will be
3741  * copied with a single call.
3742  *
3743  * Consecutive calls will fill the buffer with the next block of
3744  * available older records, not including the earlier retrieved ones.
3745  *
3746  * A return value of FALSE indicates that there are no more records to
3747  * read.
3748  */
kmsg_dump_get_buffer(struct kmsg_dump_iter * iter,bool syslog,char * buf,size_t size,size_t * len_out)3749 bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
3750 			  char *buf, size_t size, size_t *len_out)
3751 {
3752 	u64 min_seq = latched_seq_read_nolock(&clear_seq);
3753 	struct printk_info info;
3754 	struct printk_record r;
3755 	u64 seq;
3756 	u64 next_seq;
3757 	size_t len = 0;
3758 	bool ret = false;
3759 	bool time = printk_time;
3760 
3761 	if (!buf || !size)
3762 		goto out;
3763 
3764 	if (iter->cur_seq < min_seq)
3765 		iter->cur_seq = min_seq;
3766 
3767 	if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
3768 		if (info.seq != iter->cur_seq) {
3769 			/* messages are gone, move to first available one */
3770 			iter->cur_seq = info.seq;
3771 		}
3772 	}
3773 
3774 	/* last entry */
3775 	if (iter->cur_seq >= iter->next_seq)
3776 		goto out;
3777 
3778 	/*
3779 	 * Find first record that fits, including all following records,
3780 	 * into the user-provided buffer for this dump. Pass in size-1
3781 	 * because this function (by way of record_print_text()) will
3782 	 * not write more than size-1 bytes of text into @buf.
3783 	 */
3784 	seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
3785 				     size - 1, syslog, time);
3786 
3787 	/*
3788 	 * Next kmsg_dump_get_buffer() invocation will dump block of
3789 	 * older records stored right before this one.
3790 	 */
3791 	next_seq = seq;
3792 
3793 	prb_rec_init_rd(&r, &info, buf, size);
3794 
3795 	len = 0;
3796 	prb_for_each_record(seq, prb, seq, &r) {
3797 		if (r.info->seq >= iter->next_seq)
3798 			break;
3799 
3800 		len += record_print_text(&r, syslog, time);
3801 
3802 		/* Adjust record to store to remaining buffer space. */
3803 		prb_rec_init_rd(&r, &info, buf + len, size - len);
3804 	}
3805 
3806 	iter->next_seq = next_seq;
3807 	ret = true;
3808 out:
3809 	if (len_out)
3810 		*len_out = len;
3811 	return ret;
3812 }
3813 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3814 
3815 /**
3816  * kmsg_dump_rewind - reset the iterator
3817  * @iter: kmsg dump iterator
3818  *
3819  * Reset the dumper's iterator so that kmsg_dump_get_line() and
3820  * kmsg_dump_get_buffer() can be called again and used multiple
3821  * times within the same dumper.dump() callback.
3822  */
kmsg_dump_rewind(struct kmsg_dump_iter * iter)3823 void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
3824 {
3825 	iter->cur_seq = latched_seq_read_nolock(&clear_seq);
3826 	iter->next_seq = prb_next_seq(prb);
3827 }
3828 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3829 
3830 #endif
3831 
3832 #ifdef CONFIG_SMP
3833 static atomic_t printk_cpu_sync_owner = ATOMIC_INIT(-1);
3834 static atomic_t printk_cpu_sync_nested = ATOMIC_INIT(0);
3835 
3836 /**
3837  * __printk_cpu_sync_wait() - Busy wait until the printk cpu-reentrant
3838  *                            spinning lock is not owned by any CPU.
3839  *
3840  * Context: Any context.
3841  */
__printk_cpu_sync_wait(void)3842 void __printk_cpu_sync_wait(void)
3843 {
3844 	do {
3845 		cpu_relax();
3846 	} while (atomic_read(&printk_cpu_sync_owner) != -1);
3847 }
3848 EXPORT_SYMBOL(__printk_cpu_sync_wait);
3849 
3850 /**
3851  * __printk_cpu_sync_try_get() - Try to acquire the printk cpu-reentrant
3852  *                               spinning lock.
3853  *
3854  * If no processor has the lock, the calling processor takes the lock and
3855  * becomes the owner. If the calling processor is already the owner of the
3856  * lock, this function succeeds immediately.
3857  *
3858  * Context: Any context. Expects interrupts to be disabled.
3859  * Return: 1 on success, otherwise 0.
3860  */
__printk_cpu_sync_try_get(void)3861 int __printk_cpu_sync_try_get(void)
3862 {
3863 	int cpu;
3864 	int old;
3865 
3866 	cpu = smp_processor_id();
3867 
3868 	/*
3869 	 * Guarantee loads and stores from this CPU when it is the lock owner
3870 	 * are _not_ visible to the previous lock owner. This pairs with
3871 	 * __printk_cpu_sync_put:B.
3872 	 *
3873 	 * Memory barrier involvement:
3874 	 *
3875 	 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
3876 	 * then __printk_cpu_sync_put:A can never read from
3877 	 * __printk_cpu_sync_try_get:B.
3878 	 *
3879 	 * Relies on:
3880 	 *
3881 	 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
3882 	 * of the previous CPU
3883 	 *    matching
3884 	 * ACQUIRE from __printk_cpu_sync_try_get:A to
3885 	 * __printk_cpu_sync_try_get:B of this CPU
3886 	 */
3887 	old = atomic_cmpxchg_acquire(&printk_cpu_sync_owner, -1,
3888 				     cpu); /* LMM(__printk_cpu_sync_try_get:A) */
3889 	if (old == -1) {
3890 		/*
3891 		 * This CPU is now the owner and begins loading/storing
3892 		 * data: LMM(__printk_cpu_sync_try_get:B)
3893 		 */
3894 		return 1;
3895 
3896 	} else if (old == cpu) {
3897 		/* This CPU is already the owner. */
3898 		atomic_inc(&printk_cpu_sync_nested);
3899 		return 1;
3900 	}
3901 
3902 	return 0;
3903 }
3904 EXPORT_SYMBOL(__printk_cpu_sync_try_get);
3905 
3906 /**
3907  * __printk_cpu_sync_put() - Release the printk cpu-reentrant spinning lock.
3908  *
3909  * The calling processor must be the owner of the lock.
3910  *
3911  * Context: Any context. Expects interrupts to be disabled.
3912  */
__printk_cpu_sync_put(void)3913 void __printk_cpu_sync_put(void)
3914 {
3915 	if (atomic_read(&printk_cpu_sync_nested)) {
3916 		atomic_dec(&printk_cpu_sync_nested);
3917 		return;
3918 	}
3919 
3920 	/*
3921 	 * This CPU is finished loading/storing data:
3922 	 * LMM(__printk_cpu_sync_put:A)
3923 	 */
3924 
3925 	/*
3926 	 * Guarantee loads and stores from this CPU when it was the
3927 	 * lock owner are visible to the next lock owner. This pairs
3928 	 * with __printk_cpu_sync_try_get:A.
3929 	 *
3930 	 * Memory barrier involvement:
3931 	 *
3932 	 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
3933 	 * then __printk_cpu_sync_try_get:B reads from __printk_cpu_sync_put:A.
3934 	 *
3935 	 * Relies on:
3936 	 *
3937 	 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
3938 	 * of this CPU
3939 	 *    matching
3940 	 * ACQUIRE from __printk_cpu_sync_try_get:A to
3941 	 * __printk_cpu_sync_try_get:B of the next CPU
3942 	 */
3943 	atomic_set_release(&printk_cpu_sync_owner,
3944 			   -1); /* LMM(__printk_cpu_sync_put:B) */
3945 }
3946 EXPORT_SYMBOL(__printk_cpu_sync_put);
3947 #endif /* CONFIG_SMP */
3948