xref: /linux/kernel/printk/printk.c (revision 42b16d3ac371a2fac9b6f08fd75f23f34ba3955a)
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/syscore_ops.h>
38 #include <linux/vmcore_info.h>
39 #include <linux/ratelimit.h>
40 #include <linux/kmsg_dump.h>
41 #include <linux/syslog.h>
42 #include <linux/cpu.h>
43 #include <linux/rculist.h>
44 #include <linux/poll.h>
45 #include <linux/irq_work.h>
46 #include <linux/ctype.h>
47 #include <linux/uio.h>
48 #include <linux/sched/clock.h>
49 #include <linux/sched/debug.h>
50 #include <linux/sched/task_stack.h>
51 
52 #include <linux/uaccess.h>
53 #include <asm/sections.h>
54 
55 #include <trace/events/initcall.h>
56 #define CREATE_TRACE_POINTS
57 #include <trace/events/printk.h>
58 
59 #include "printk_ringbuffer.h"
60 #include "console_cmdline.h"
61 #include "braille.h"
62 #include "internal.h"
63 
64 int console_printk[4] = {
65 	CONSOLE_LOGLEVEL_DEFAULT,	/* console_loglevel */
66 	MESSAGE_LOGLEVEL_DEFAULT,	/* default_message_loglevel */
67 	CONSOLE_LOGLEVEL_MIN,		/* minimum_console_loglevel */
68 	CONSOLE_LOGLEVEL_DEFAULT,	/* default_console_loglevel */
69 };
70 EXPORT_SYMBOL_GPL(console_printk);
71 
72 atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
73 EXPORT_SYMBOL(ignore_console_lock_warning);
74 
75 EXPORT_TRACEPOINT_SYMBOL_GPL(console);
76 
77 /*
78  * Low level drivers may need that to know if they can schedule in
79  * their unblank() callback or not. So let's export it.
80  */
81 int oops_in_progress;
82 EXPORT_SYMBOL(oops_in_progress);
83 
84 /*
85  * console_mutex protects console_list updates and console->flags updates.
86  * The flags are synchronized only for consoles that are registered, i.e.
87  * accessible via the console list.
88  */
89 static DEFINE_MUTEX(console_mutex);
90 
91 /*
92  * console_sem protects updates to console->seq
93  * and also provides serialization for console printing.
94  */
95 static DEFINE_SEMAPHORE(console_sem, 1);
96 HLIST_HEAD(console_list);
97 EXPORT_SYMBOL_GPL(console_list);
98 DEFINE_STATIC_SRCU(console_srcu);
99 
100 /*
101  * System may need to suppress printk message under certain
102  * circumstances, like after kernel panic happens.
103  */
104 int __read_mostly suppress_printk;
105 
106 #ifdef CONFIG_LOCKDEP
107 static struct lockdep_map console_lock_dep_map = {
108 	.name = "console_lock"
109 };
110 
lockdep_assert_console_list_lock_held(void)111 void lockdep_assert_console_list_lock_held(void)
112 {
113 	lockdep_assert_held(&console_mutex);
114 }
115 EXPORT_SYMBOL(lockdep_assert_console_list_lock_held);
116 #endif
117 
118 #ifdef CONFIG_DEBUG_LOCK_ALLOC
console_srcu_read_lock_is_held(void)119 bool console_srcu_read_lock_is_held(void)
120 {
121 	return srcu_read_lock_held(&console_srcu);
122 }
123 EXPORT_SYMBOL(console_srcu_read_lock_is_held);
124 #endif
125 
126 enum devkmsg_log_bits {
127 	__DEVKMSG_LOG_BIT_ON = 0,
128 	__DEVKMSG_LOG_BIT_OFF,
129 	__DEVKMSG_LOG_BIT_LOCK,
130 };
131 
132 enum devkmsg_log_masks {
133 	DEVKMSG_LOG_MASK_ON             = BIT(__DEVKMSG_LOG_BIT_ON),
134 	DEVKMSG_LOG_MASK_OFF            = BIT(__DEVKMSG_LOG_BIT_OFF),
135 	DEVKMSG_LOG_MASK_LOCK           = BIT(__DEVKMSG_LOG_BIT_LOCK),
136 };
137 
138 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
139 #define DEVKMSG_LOG_MASK_DEFAULT	0
140 
141 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
142 
__control_devkmsg(char * str)143 static int __control_devkmsg(char *str)
144 {
145 	size_t len;
146 
147 	if (!str)
148 		return -EINVAL;
149 
150 	len = str_has_prefix(str, "on");
151 	if (len) {
152 		devkmsg_log = DEVKMSG_LOG_MASK_ON;
153 		return len;
154 	}
155 
156 	len = str_has_prefix(str, "off");
157 	if (len) {
158 		devkmsg_log = DEVKMSG_LOG_MASK_OFF;
159 		return len;
160 	}
161 
162 	len = str_has_prefix(str, "ratelimit");
163 	if (len) {
164 		devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
165 		return len;
166 	}
167 
168 	return -EINVAL;
169 }
170 
control_devkmsg(char * str)171 static int __init control_devkmsg(char *str)
172 {
173 	if (__control_devkmsg(str) < 0) {
174 		pr_warn("printk.devkmsg: bad option string '%s'\n", str);
175 		return 1;
176 	}
177 
178 	/*
179 	 * Set sysctl string accordingly:
180 	 */
181 	if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
182 		strscpy(devkmsg_log_str, "on");
183 	else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
184 		strscpy(devkmsg_log_str, "off");
185 	/* else "ratelimit" which is set by default. */
186 
187 	/*
188 	 * Sysctl cannot change it anymore. The kernel command line setting of
189 	 * this parameter is to force the setting to be permanent throughout the
190 	 * runtime of the system. This is a precation measure against userspace
191 	 * trying to be a smarta** and attempting to change it up on us.
192 	 */
193 	devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
194 
195 	return 1;
196 }
197 __setup("printk.devkmsg=", control_devkmsg);
198 
199 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
200 #if defined(CONFIG_PRINTK) && defined(CONFIG_SYSCTL)
devkmsg_sysctl_set_loglvl(const struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)201 int devkmsg_sysctl_set_loglvl(const struct ctl_table *table, int write,
202 			      void *buffer, size_t *lenp, loff_t *ppos)
203 {
204 	char old_str[DEVKMSG_STR_MAX_SIZE];
205 	unsigned int old;
206 	int err;
207 
208 	if (write) {
209 		if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
210 			return -EINVAL;
211 
212 		old = devkmsg_log;
213 		strscpy(old_str, devkmsg_log_str);
214 	}
215 
216 	err = proc_dostring(table, write, buffer, lenp, ppos);
217 	if (err)
218 		return err;
219 
220 	if (write) {
221 		err = __control_devkmsg(devkmsg_log_str);
222 
223 		/*
224 		 * Do not accept an unknown string OR a known string with
225 		 * trailing crap...
226 		 */
227 		if (err < 0 || (err + 1 != *lenp)) {
228 
229 			/* ... and restore old setting. */
230 			devkmsg_log = old;
231 			strscpy(devkmsg_log_str, old_str);
232 
233 			return -EINVAL;
234 		}
235 	}
236 
237 	return 0;
238 }
239 #endif /* CONFIG_PRINTK && CONFIG_SYSCTL */
240 
241 /**
242  * console_list_lock - Lock the console list
243  *
244  * For console list or console->flags updates
245  */
console_list_lock(void)246 void console_list_lock(void)
247 {
248 	/*
249 	 * In unregister_console() and console_force_preferred_locked(),
250 	 * synchronize_srcu() is called with the console_list_lock held.
251 	 * Therefore it is not allowed that the console_list_lock is taken
252 	 * with the srcu_lock held.
253 	 *
254 	 * Detecting if this context is really in the read-side critical
255 	 * section is only possible if the appropriate debug options are
256 	 * enabled.
257 	 */
258 	WARN_ON_ONCE(debug_lockdep_rcu_enabled() &&
259 		     srcu_read_lock_held(&console_srcu));
260 
261 	mutex_lock(&console_mutex);
262 }
263 EXPORT_SYMBOL(console_list_lock);
264 
265 /**
266  * console_list_unlock - Unlock the console list
267  *
268  * Counterpart to console_list_lock()
269  */
console_list_unlock(void)270 void console_list_unlock(void)
271 {
272 	mutex_unlock(&console_mutex);
273 }
274 EXPORT_SYMBOL(console_list_unlock);
275 
276 /**
277  * console_srcu_read_lock - Register a new reader for the
278  *	SRCU-protected console list
279  *
280  * Use for_each_console_srcu() to iterate the console list
281  *
282  * Context: Any context.
283  * Return: A cookie to pass to console_srcu_read_unlock().
284  */
console_srcu_read_lock(void)285 int console_srcu_read_lock(void)
286 	__acquires(&console_srcu)
287 {
288 	return srcu_read_lock_nmisafe(&console_srcu);
289 }
290 EXPORT_SYMBOL(console_srcu_read_lock);
291 
292 /**
293  * console_srcu_read_unlock - Unregister an old reader from
294  *	the SRCU-protected console list
295  * @cookie: cookie returned from console_srcu_read_lock()
296  *
297  * Counterpart to console_srcu_read_lock()
298  */
console_srcu_read_unlock(int cookie)299 void console_srcu_read_unlock(int cookie)
300 	__releases(&console_srcu)
301 {
302 	srcu_read_unlock_nmisafe(&console_srcu, cookie);
303 }
304 EXPORT_SYMBOL(console_srcu_read_unlock);
305 
306 /*
307  * Helper macros to handle lockdep when locking/unlocking console_sem. We use
308  * macros instead of functions so that _RET_IP_ contains useful information.
309  */
310 #define down_console_sem() do { \
311 	down(&console_sem);\
312 	mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
313 } while (0)
314 
__down_trylock_console_sem(unsigned long ip)315 static int __down_trylock_console_sem(unsigned long ip)
316 {
317 	int lock_failed;
318 	unsigned long flags;
319 
320 	/*
321 	 * Here and in __up_console_sem() we need to be in safe mode,
322 	 * because spindump/WARN/etc from under console ->lock will
323 	 * deadlock in printk()->down_trylock_console_sem() otherwise.
324 	 */
325 	printk_safe_enter_irqsave(flags);
326 	lock_failed = down_trylock(&console_sem);
327 	printk_safe_exit_irqrestore(flags);
328 
329 	if (lock_failed)
330 		return 1;
331 	mutex_acquire(&console_lock_dep_map, 0, 1, ip);
332 	return 0;
333 }
334 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
335 
__up_console_sem(unsigned long ip)336 static void __up_console_sem(unsigned long ip)
337 {
338 	unsigned long flags;
339 
340 	mutex_release(&console_lock_dep_map, ip);
341 
342 	printk_safe_enter_irqsave(flags);
343 	up(&console_sem);
344 	printk_safe_exit_irqrestore(flags);
345 }
346 #define up_console_sem() __up_console_sem(_RET_IP_)
347 
panic_in_progress(void)348 static bool panic_in_progress(void)
349 {
350 	return unlikely(atomic_read(&panic_cpu) != PANIC_CPU_INVALID);
351 }
352 
353 /* Return true if a panic is in progress on the current CPU. */
this_cpu_in_panic(void)354 bool this_cpu_in_panic(void)
355 {
356 	/*
357 	 * We can use raw_smp_processor_id() here because it is impossible for
358 	 * the task to be migrated to the panic_cpu, or away from it. If
359 	 * panic_cpu has already been set, and we're not currently executing on
360 	 * that CPU, then we never will be.
361 	 */
362 	return unlikely(atomic_read(&panic_cpu) == raw_smp_processor_id());
363 }
364 
365 /*
366  * Return true if a panic is in progress on a remote CPU.
367  *
368  * On true, the local CPU should immediately release any printing resources
369  * that may be needed by the panic CPU.
370  */
other_cpu_in_panic(void)371 bool other_cpu_in_panic(void)
372 {
373 	return (panic_in_progress() && !this_cpu_in_panic());
374 }
375 
376 /*
377  * This is used for debugging the mess that is the VT code by
378  * keeping track if we have the console semaphore held. It's
379  * definitely not the perfect debug tool (we don't know if _WE_
380  * hold it and are racing, but it helps tracking those weird code
381  * paths in the console code where we end up in places I want
382  * locked without the console semaphore held).
383  */
384 static int console_locked;
385 
386 /*
387  *	Array of consoles built from command line options (console=)
388  */
389 
390 #define MAX_CMDLINECONSOLES 8
391 
392 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
393 
394 static int preferred_console = -1;
395 int console_set_on_cmdline;
396 EXPORT_SYMBOL(console_set_on_cmdline);
397 
398 /* Flag: console code may call schedule() */
399 static int console_may_schedule;
400 
401 enum con_msg_format_flags {
402 	MSG_FORMAT_DEFAULT	= 0,
403 	MSG_FORMAT_SYSLOG	= (1 << 0),
404 };
405 
406 static int console_msg_format = MSG_FORMAT_DEFAULT;
407 
408 /*
409  * The printk log buffer consists of a sequenced collection of records, each
410  * containing variable length message text. Every record also contains its
411  * own meta-data (@info).
412  *
413  * Every record meta-data carries the timestamp in microseconds, as well as
414  * the standard userspace syslog level and syslog facility. The usual kernel
415  * messages use LOG_KERN; userspace-injected messages always carry a matching
416  * syslog facility, by default LOG_USER. The origin of every message can be
417  * reliably determined that way.
418  *
419  * The human readable log message of a record is available in @text, the
420  * length of the message text in @text_len. The stored message is not
421  * terminated.
422  *
423  * Optionally, a record can carry a dictionary of properties (key/value
424  * pairs), to provide userspace with a machine-readable message context.
425  *
426  * Examples for well-defined, commonly used property names are:
427  *   DEVICE=b12:8               device identifier
428  *                                b12:8         block dev_t
429  *                                c127:3        char dev_t
430  *                                n8            netdev ifindex
431  *                                +sound:card0  subsystem:devname
432  *   SUBSYSTEM=pci              driver-core subsystem name
433  *
434  * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
435  * and values are terminated by a '\0' character.
436  *
437  * Example of record values:
438  *   record.text_buf                = "it's a line" (unterminated)
439  *   record.info.seq                = 56
440  *   record.info.ts_nsec            = 36863
441  *   record.info.text_len           = 11
442  *   record.info.facility           = 0 (LOG_KERN)
443  *   record.info.flags              = 0
444  *   record.info.level              = 3 (LOG_ERR)
445  *   record.info.caller_id          = 299 (task 299)
446  *   record.info.dev_info.subsystem = "pci" (terminated)
447  *   record.info.dev_info.device    = "+pci:0000:00:01.0" (terminated)
448  *
449  * The 'struct printk_info' buffer must never be directly exported to
450  * userspace, it is a kernel-private implementation detail that might
451  * need to be changed in the future, when the requirements change.
452  *
453  * /dev/kmsg exports the structured data in the following line format:
454  *   "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
455  *
456  * Users of the export format should ignore possible additional values
457  * separated by ',', and find the message after the ';' character.
458  *
459  * The optional key/value pairs are attached as continuation lines starting
460  * with a space character and terminated by a newline. All possible
461  * non-prinatable characters are escaped in the "\xff" notation.
462  */
463 
464 /* syslog_lock protects syslog_* variables and write access to clear_seq. */
465 static DEFINE_MUTEX(syslog_lock);
466 
467 /*
468  * Specifies if a legacy console is registered. If legacy consoles are
469  * present, it is necessary to perform the console lock/unlock dance
470  * whenever console flushing should occur.
471  */
472 bool have_legacy_console;
473 
474 /*
475  * Specifies if an nbcon console is registered. If nbcon consoles are present,
476  * synchronous printing of legacy consoles will not occur during panic until
477  * the backtrace has been stored to the ringbuffer.
478  */
479 bool have_nbcon_console;
480 
481 /*
482  * Specifies if a boot console is registered. If boot consoles are present,
483  * nbcon consoles cannot print simultaneously and must be synchronized by
484  * the console lock. This is because boot consoles and nbcon consoles may
485  * have mapped the same hardware.
486  */
487 bool have_boot_console;
488 
489 /* See printk_legacy_allow_panic_sync() for details. */
490 bool legacy_allow_panic_sync;
491 
492 #ifdef CONFIG_PRINTK
493 DECLARE_WAIT_QUEUE_HEAD(log_wait);
494 static DECLARE_WAIT_QUEUE_HEAD(legacy_wait);
495 /* All 3 protected by @syslog_lock. */
496 /* the next printk record to read by syslog(READ) or /proc/kmsg */
497 static u64 syslog_seq;
498 static size_t syslog_partial;
499 static bool syslog_time;
500 
501 /* True when _all_ printer threads are available for printing. */
502 bool printk_kthreads_running;
503 
504 struct latched_seq {
505 	seqcount_latch_t	latch;
506 	u64			val[2];
507 };
508 
509 /*
510  * The next printk record to read after the last 'clear' command. There are
511  * two copies (updated with seqcount_latch) so that reads can locklessly
512  * access a valid value. Writers are synchronized by @syslog_lock.
513  */
514 static struct latched_seq clear_seq = {
515 	.latch		= SEQCNT_LATCH_ZERO(clear_seq.latch),
516 	.val[0]		= 0,
517 	.val[1]		= 0,
518 };
519 
520 #define LOG_LEVEL(v)		((v) & 0x07)
521 #define LOG_FACILITY(v)		((v) >> 3 & 0xff)
522 
523 /* record buffer */
524 #define LOG_ALIGN __alignof__(unsigned long)
525 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
526 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
527 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
528 static char *log_buf = __log_buf;
529 static u32 log_buf_len = __LOG_BUF_LEN;
530 
531 /*
532  * Define the average message size. This only affects the number of
533  * descriptors that will be available. Underestimating is better than
534  * overestimating (too many available descriptors is better than not enough).
535  */
536 #define PRB_AVGBITS 5	/* 32 character average length */
537 
538 #if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
539 #error CONFIG_LOG_BUF_SHIFT value too small.
540 #endif
541 _DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
542 		 PRB_AVGBITS, &__log_buf[0]);
543 
544 static struct printk_ringbuffer printk_rb_dynamic;
545 
546 struct printk_ringbuffer *prb = &printk_rb_static;
547 
548 /*
549  * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
550  * per_cpu_areas are initialised. This variable is set to true when
551  * it's safe to access per-CPU data.
552  */
553 static bool __printk_percpu_data_ready __ro_after_init;
554 
printk_percpu_data_ready(void)555 bool printk_percpu_data_ready(void)
556 {
557 	return __printk_percpu_data_ready;
558 }
559 
560 /* Must be called under syslog_lock. */
latched_seq_write(struct latched_seq * ls,u64 val)561 static void latched_seq_write(struct latched_seq *ls, u64 val)
562 {
563 	raw_write_seqcount_latch(&ls->latch);
564 	ls->val[0] = val;
565 	raw_write_seqcount_latch(&ls->latch);
566 	ls->val[1] = val;
567 }
568 
569 /* Can be called from any context. */
latched_seq_read_nolock(struct latched_seq * ls)570 static u64 latched_seq_read_nolock(struct latched_seq *ls)
571 {
572 	unsigned int seq;
573 	unsigned int idx;
574 	u64 val;
575 
576 	do {
577 		seq = raw_read_seqcount_latch(&ls->latch);
578 		idx = seq & 0x1;
579 		val = ls->val[idx];
580 	} while (raw_read_seqcount_latch_retry(&ls->latch, seq));
581 
582 	return val;
583 }
584 
585 /* Return log buffer address */
log_buf_addr_get(void)586 char *log_buf_addr_get(void)
587 {
588 	return log_buf;
589 }
590 
591 /* Return log buffer size */
log_buf_len_get(void)592 u32 log_buf_len_get(void)
593 {
594 	return log_buf_len;
595 }
596 
597 /*
598  * Define how much of the log buffer we could take at maximum. The value
599  * must be greater than two. Note that only half of the buffer is available
600  * when the index points to the middle.
601  */
602 #define MAX_LOG_TAKE_PART 4
603 static const char trunc_msg[] = "<truncated>";
604 
truncate_msg(u16 * text_len,u16 * trunc_msg_len)605 static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
606 {
607 	/*
608 	 * The message should not take the whole buffer. Otherwise, it might
609 	 * get removed too soon.
610 	 */
611 	u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
612 
613 	if (*text_len > max_text_len)
614 		*text_len = max_text_len;
615 
616 	/* enable the warning message (if there is room) */
617 	*trunc_msg_len = strlen(trunc_msg);
618 	if (*text_len >= *trunc_msg_len)
619 		*text_len -= *trunc_msg_len;
620 	else
621 		*trunc_msg_len = 0;
622 }
623 
624 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
625 
syslog_action_restricted(int type)626 static int syslog_action_restricted(int type)
627 {
628 	if (dmesg_restrict)
629 		return 1;
630 	/*
631 	 * Unless restricted, we allow "read all" and "get buffer size"
632 	 * for everybody.
633 	 */
634 	return type != SYSLOG_ACTION_READ_ALL &&
635 	       type != SYSLOG_ACTION_SIZE_BUFFER;
636 }
637 
check_syslog_permissions(int type,int source)638 static int check_syslog_permissions(int type, int source)
639 {
640 	/*
641 	 * If this is from /proc/kmsg and we've already opened it, then we've
642 	 * already done the capabilities checks at open time.
643 	 */
644 	if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
645 		goto ok;
646 
647 	if (syslog_action_restricted(type)) {
648 		if (capable(CAP_SYSLOG))
649 			goto ok;
650 		return -EPERM;
651 	}
652 ok:
653 	return security_syslog(type);
654 }
655 
append_char(char ** pp,char * e,char c)656 static void append_char(char **pp, char *e, char c)
657 {
658 	if (*pp < e)
659 		*(*pp)++ = c;
660 }
661 
info_print_ext_header(char * buf,size_t size,struct printk_info * info)662 static ssize_t info_print_ext_header(char *buf, size_t size,
663 				     struct printk_info *info)
664 {
665 	u64 ts_usec = info->ts_nsec;
666 	char caller[20];
667 #ifdef CONFIG_PRINTK_CALLER
668 	u32 id = info->caller_id;
669 
670 	snprintf(caller, sizeof(caller), ",caller=%c%u",
671 		 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
672 #else
673 	caller[0] = '\0';
674 #endif
675 
676 	do_div(ts_usec, 1000);
677 
678 	return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
679 			 (info->facility << 3) | info->level, info->seq,
680 			 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
681 }
682 
msg_add_ext_text(char * buf,size_t size,const char * text,size_t text_len,unsigned char endc)683 static ssize_t msg_add_ext_text(char *buf, size_t size,
684 				const char *text, size_t text_len,
685 				unsigned char endc)
686 {
687 	char *p = buf, *e = buf + size;
688 	size_t i;
689 
690 	/* escape non-printable characters */
691 	for (i = 0; i < text_len; i++) {
692 		unsigned char c = text[i];
693 
694 		if (c < ' ' || c >= 127 || c == '\\')
695 			p += scnprintf(p, e - p, "\\x%02x", c);
696 		else
697 			append_char(&p, e, c);
698 	}
699 	append_char(&p, e, endc);
700 
701 	return p - buf;
702 }
703 
msg_add_dict_text(char * buf,size_t size,const char * key,const char * val)704 static ssize_t msg_add_dict_text(char *buf, size_t size,
705 				 const char *key, const char *val)
706 {
707 	size_t val_len = strlen(val);
708 	ssize_t len;
709 
710 	if (!val_len)
711 		return 0;
712 
713 	len = msg_add_ext_text(buf, size, "", 0, ' ');	/* dict prefix */
714 	len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
715 	len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
716 
717 	return len;
718 }
719 
msg_print_ext_body(char * buf,size_t size,char * text,size_t text_len,struct dev_printk_info * dev_info)720 static ssize_t msg_print_ext_body(char *buf, size_t size,
721 				  char *text, size_t text_len,
722 				  struct dev_printk_info *dev_info)
723 {
724 	ssize_t len;
725 
726 	len = msg_add_ext_text(buf, size, text, text_len, '\n');
727 
728 	if (!dev_info)
729 		goto out;
730 
731 	len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
732 				 dev_info->subsystem);
733 	len += msg_add_dict_text(buf + len, size - len, "DEVICE",
734 				 dev_info->device);
735 out:
736 	return len;
737 }
738 
739 /* /dev/kmsg - userspace message inject/listen interface */
740 struct devkmsg_user {
741 	atomic64_t seq;
742 	struct ratelimit_state rs;
743 	struct mutex lock;
744 	struct printk_buffers pbufs;
745 };
746 
747 static __printf(3, 4) __cold
devkmsg_emit(int facility,int level,const char * fmt,...)748 int devkmsg_emit(int facility, int level, const char *fmt, ...)
749 {
750 	va_list args;
751 	int r;
752 
753 	va_start(args, fmt);
754 	r = vprintk_emit(facility, level, NULL, fmt, args);
755 	va_end(args);
756 
757 	return r;
758 }
759 
devkmsg_write(struct kiocb * iocb,struct iov_iter * from)760 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
761 {
762 	char *buf, *line;
763 	int level = default_message_loglevel;
764 	int facility = 1;	/* LOG_USER */
765 	struct file *file = iocb->ki_filp;
766 	struct devkmsg_user *user = file->private_data;
767 	size_t len = iov_iter_count(from);
768 	ssize_t ret = len;
769 
770 	if (len > PRINTKRB_RECORD_MAX)
771 		return -EINVAL;
772 
773 	/* Ignore when user logging is disabled. */
774 	if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
775 		return len;
776 
777 	/* Ratelimit when not explicitly enabled. */
778 	if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
779 		if (!___ratelimit(&user->rs, current->comm))
780 			return ret;
781 	}
782 
783 	buf = kmalloc(len+1, GFP_KERNEL);
784 	if (buf == NULL)
785 		return -ENOMEM;
786 
787 	buf[len] = '\0';
788 	if (!copy_from_iter_full(buf, len, from)) {
789 		kfree(buf);
790 		return -EFAULT;
791 	}
792 
793 	/*
794 	 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
795 	 * the decimal value represents 32bit, the lower 3 bit are the log
796 	 * level, the rest are the log facility.
797 	 *
798 	 * If no prefix or no userspace facility is specified, we
799 	 * enforce LOG_USER, to be able to reliably distinguish
800 	 * kernel-generated messages from userspace-injected ones.
801 	 */
802 	line = buf;
803 	if (line[0] == '<') {
804 		char *endp = NULL;
805 		unsigned int u;
806 
807 		u = simple_strtoul(line + 1, &endp, 10);
808 		if (endp && endp[0] == '>') {
809 			level = LOG_LEVEL(u);
810 			if (LOG_FACILITY(u) != 0)
811 				facility = LOG_FACILITY(u);
812 			endp++;
813 			line = endp;
814 		}
815 	}
816 
817 	devkmsg_emit(facility, level, "%s", line);
818 	kfree(buf);
819 	return ret;
820 }
821 
devkmsg_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)822 static ssize_t devkmsg_read(struct file *file, char __user *buf,
823 			    size_t count, loff_t *ppos)
824 {
825 	struct devkmsg_user *user = file->private_data;
826 	char *outbuf = &user->pbufs.outbuf[0];
827 	struct printk_message pmsg = {
828 		.pbufs = &user->pbufs,
829 	};
830 	ssize_t ret;
831 
832 	ret = mutex_lock_interruptible(&user->lock);
833 	if (ret)
834 		return ret;
835 
836 	if (!printk_get_next_message(&pmsg, atomic64_read(&user->seq), true, false)) {
837 		if (file->f_flags & O_NONBLOCK) {
838 			ret = -EAGAIN;
839 			goto out;
840 		}
841 
842 		/*
843 		 * Guarantee this task is visible on the waitqueue before
844 		 * checking the wake condition.
845 		 *
846 		 * The full memory barrier within set_current_state() of
847 		 * prepare_to_wait_event() pairs with the full memory barrier
848 		 * within wq_has_sleeper().
849 		 *
850 		 * This pairs with __wake_up_klogd:A.
851 		 */
852 		ret = wait_event_interruptible(log_wait,
853 				printk_get_next_message(&pmsg, atomic64_read(&user->seq), true,
854 							false)); /* LMM(devkmsg_read:A) */
855 		if (ret)
856 			goto out;
857 	}
858 
859 	if (pmsg.dropped) {
860 		/* our last seen message is gone, return error and reset */
861 		atomic64_set(&user->seq, pmsg.seq);
862 		ret = -EPIPE;
863 		goto out;
864 	}
865 
866 	atomic64_set(&user->seq, pmsg.seq + 1);
867 
868 	if (pmsg.outbuf_len > count) {
869 		ret = -EINVAL;
870 		goto out;
871 	}
872 
873 	if (copy_to_user(buf, outbuf, pmsg.outbuf_len)) {
874 		ret = -EFAULT;
875 		goto out;
876 	}
877 	ret = pmsg.outbuf_len;
878 out:
879 	mutex_unlock(&user->lock);
880 	return ret;
881 }
882 
883 /*
884  * Be careful when modifying this function!!!
885  *
886  * Only few operations are supported because the device works only with the
887  * entire variable length messages (records). Non-standard values are
888  * returned in the other cases and has been this way for quite some time.
889  * User space applications might depend on this behavior.
890  */
devkmsg_llseek(struct file * file,loff_t offset,int whence)891 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
892 {
893 	struct devkmsg_user *user = file->private_data;
894 	loff_t ret = 0;
895 
896 	if (offset)
897 		return -ESPIPE;
898 
899 	switch (whence) {
900 	case SEEK_SET:
901 		/* the first record */
902 		atomic64_set(&user->seq, prb_first_valid_seq(prb));
903 		break;
904 	case SEEK_DATA:
905 		/*
906 		 * The first record after the last SYSLOG_ACTION_CLEAR,
907 		 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
908 		 * changes no global state, and does not clear anything.
909 		 */
910 		atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
911 		break;
912 	case SEEK_END:
913 		/* after the last record */
914 		atomic64_set(&user->seq, prb_next_seq(prb));
915 		break;
916 	default:
917 		ret = -EINVAL;
918 	}
919 	return ret;
920 }
921 
devkmsg_poll(struct file * file,poll_table * wait)922 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
923 {
924 	struct devkmsg_user *user = file->private_data;
925 	struct printk_info info;
926 	__poll_t ret = 0;
927 
928 	poll_wait(file, &log_wait, wait);
929 
930 	if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
931 		/* return error when data has vanished underneath us */
932 		if (info.seq != atomic64_read(&user->seq))
933 			ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
934 		else
935 			ret = EPOLLIN|EPOLLRDNORM;
936 	}
937 
938 	return ret;
939 }
940 
devkmsg_open(struct inode * inode,struct file * file)941 static int devkmsg_open(struct inode *inode, struct file *file)
942 {
943 	struct devkmsg_user *user;
944 	int err;
945 
946 	if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
947 		return -EPERM;
948 
949 	/* write-only does not need any file context */
950 	if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
951 		err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
952 					       SYSLOG_FROM_READER);
953 		if (err)
954 			return err;
955 	}
956 
957 	user = kvmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
958 	if (!user)
959 		return -ENOMEM;
960 
961 	ratelimit_default_init(&user->rs);
962 	ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
963 
964 	mutex_init(&user->lock);
965 
966 	atomic64_set(&user->seq, prb_first_valid_seq(prb));
967 
968 	file->private_data = user;
969 	return 0;
970 }
971 
devkmsg_release(struct inode * inode,struct file * file)972 static int devkmsg_release(struct inode *inode, struct file *file)
973 {
974 	struct devkmsg_user *user = file->private_data;
975 
976 	ratelimit_state_exit(&user->rs);
977 
978 	mutex_destroy(&user->lock);
979 	kvfree(user);
980 	return 0;
981 }
982 
983 const struct file_operations kmsg_fops = {
984 	.open = devkmsg_open,
985 	.read = devkmsg_read,
986 	.write_iter = devkmsg_write,
987 	.llseek = devkmsg_llseek,
988 	.poll = devkmsg_poll,
989 	.release = devkmsg_release,
990 };
991 
992 #ifdef CONFIG_VMCORE_INFO
993 /*
994  * This appends the listed symbols to /proc/vmcore
995  *
996  * /proc/vmcore is used by various utilities, like crash and makedumpfile to
997  * obtain access to symbols that are otherwise very difficult to locate.  These
998  * symbols are specifically used so that utilities can access and extract the
999  * dmesg log from a vmcore file after a crash.
1000  */
log_buf_vmcoreinfo_setup(void)1001 void log_buf_vmcoreinfo_setup(void)
1002 {
1003 	struct dev_printk_info *dev_info = NULL;
1004 
1005 	VMCOREINFO_SYMBOL(prb);
1006 	VMCOREINFO_SYMBOL(printk_rb_static);
1007 	VMCOREINFO_SYMBOL(clear_seq);
1008 
1009 	/*
1010 	 * Export struct size and field offsets. User space tools can
1011 	 * parse it and detect any changes to structure down the line.
1012 	 */
1013 
1014 	VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
1015 	VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
1016 	VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
1017 	VMCOREINFO_OFFSET(printk_ringbuffer, fail);
1018 
1019 	VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
1020 	VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
1021 	VMCOREINFO_OFFSET(prb_desc_ring, descs);
1022 	VMCOREINFO_OFFSET(prb_desc_ring, infos);
1023 	VMCOREINFO_OFFSET(prb_desc_ring, head_id);
1024 	VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
1025 
1026 	VMCOREINFO_STRUCT_SIZE(prb_desc);
1027 	VMCOREINFO_OFFSET(prb_desc, state_var);
1028 	VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
1029 
1030 	VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
1031 	VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
1032 	VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
1033 
1034 	VMCOREINFO_STRUCT_SIZE(printk_info);
1035 	VMCOREINFO_OFFSET(printk_info, seq);
1036 	VMCOREINFO_OFFSET(printk_info, ts_nsec);
1037 	VMCOREINFO_OFFSET(printk_info, text_len);
1038 	VMCOREINFO_OFFSET(printk_info, caller_id);
1039 	VMCOREINFO_OFFSET(printk_info, dev_info);
1040 
1041 	VMCOREINFO_STRUCT_SIZE(dev_printk_info);
1042 	VMCOREINFO_OFFSET(dev_printk_info, subsystem);
1043 	VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
1044 	VMCOREINFO_OFFSET(dev_printk_info, device);
1045 	VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
1046 
1047 	VMCOREINFO_STRUCT_SIZE(prb_data_ring);
1048 	VMCOREINFO_OFFSET(prb_data_ring, size_bits);
1049 	VMCOREINFO_OFFSET(prb_data_ring, data);
1050 	VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
1051 	VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
1052 
1053 	VMCOREINFO_SIZE(atomic_long_t);
1054 	VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
1055 
1056 	VMCOREINFO_STRUCT_SIZE(latched_seq);
1057 	VMCOREINFO_OFFSET(latched_seq, val);
1058 }
1059 #endif
1060 
1061 /* requested log_buf_len from kernel cmdline */
1062 static unsigned long __initdata new_log_buf_len;
1063 
1064 /* we practice scaling the ring buffer by powers of 2 */
log_buf_len_update(u64 size)1065 static void __init log_buf_len_update(u64 size)
1066 {
1067 	if (size > (u64)LOG_BUF_LEN_MAX) {
1068 		size = (u64)LOG_BUF_LEN_MAX;
1069 		pr_err("log_buf over 2G is not supported.\n");
1070 	}
1071 
1072 	if (size)
1073 		size = roundup_pow_of_two(size);
1074 	if (size > log_buf_len)
1075 		new_log_buf_len = (unsigned long)size;
1076 }
1077 
1078 /* save requested log_buf_len since it's too early to process it */
log_buf_len_setup(char * str)1079 static int __init log_buf_len_setup(char *str)
1080 {
1081 	u64 size;
1082 
1083 	if (!str)
1084 		return -EINVAL;
1085 
1086 	size = memparse(str, &str);
1087 
1088 	log_buf_len_update(size);
1089 
1090 	return 0;
1091 }
1092 early_param("log_buf_len", log_buf_len_setup);
1093 
1094 #ifdef CONFIG_SMP
1095 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1096 
log_buf_add_cpu(void)1097 static void __init log_buf_add_cpu(void)
1098 {
1099 	unsigned int cpu_extra;
1100 
1101 	/*
1102 	 * archs should set up cpu_possible_bits properly with
1103 	 * set_cpu_possible() after setup_arch() but just in
1104 	 * case lets ensure this is valid.
1105 	 */
1106 	if (num_possible_cpus() == 1)
1107 		return;
1108 
1109 	cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1110 
1111 	/* by default this will only continue through for large > 64 CPUs */
1112 	if (cpu_extra <= __LOG_BUF_LEN / 2)
1113 		return;
1114 
1115 	pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1116 		__LOG_CPU_MAX_BUF_LEN);
1117 	pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1118 		cpu_extra);
1119 	pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1120 
1121 	log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1122 }
1123 #else /* !CONFIG_SMP */
log_buf_add_cpu(void)1124 static inline void log_buf_add_cpu(void) {}
1125 #endif /* CONFIG_SMP */
1126 
set_percpu_data_ready(void)1127 static void __init set_percpu_data_ready(void)
1128 {
1129 	__printk_percpu_data_ready = true;
1130 }
1131 
add_to_rb(struct printk_ringbuffer * rb,struct printk_record * r)1132 static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1133 				     struct printk_record *r)
1134 {
1135 	struct prb_reserved_entry e;
1136 	struct printk_record dest_r;
1137 
1138 	prb_rec_init_wr(&dest_r, r->info->text_len);
1139 
1140 	if (!prb_reserve(&e, rb, &dest_r))
1141 		return 0;
1142 
1143 	memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1144 	dest_r.info->text_len = r->info->text_len;
1145 	dest_r.info->facility = r->info->facility;
1146 	dest_r.info->level = r->info->level;
1147 	dest_r.info->flags = r->info->flags;
1148 	dest_r.info->ts_nsec = r->info->ts_nsec;
1149 	dest_r.info->caller_id = r->info->caller_id;
1150 	memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1151 
1152 	prb_final_commit(&e);
1153 
1154 	return prb_record_text_space(&e);
1155 }
1156 
1157 static char setup_text_buf[PRINTKRB_RECORD_MAX] __initdata;
1158 
setup_log_buf(int early)1159 void __init setup_log_buf(int early)
1160 {
1161 	struct printk_info *new_infos;
1162 	unsigned int new_descs_count;
1163 	struct prb_desc *new_descs;
1164 	struct printk_info info;
1165 	struct printk_record r;
1166 	unsigned int text_size;
1167 	size_t new_descs_size;
1168 	size_t new_infos_size;
1169 	unsigned long flags;
1170 	char *new_log_buf;
1171 	unsigned int free;
1172 	u64 seq;
1173 
1174 	/*
1175 	 * Some archs call setup_log_buf() multiple times - first is very
1176 	 * early, e.g. from setup_arch(), and second - when percpu_areas
1177 	 * are initialised.
1178 	 */
1179 	if (!early)
1180 		set_percpu_data_ready();
1181 
1182 	if (log_buf != __log_buf)
1183 		return;
1184 
1185 	if (!early && !new_log_buf_len)
1186 		log_buf_add_cpu();
1187 
1188 	if (!new_log_buf_len)
1189 		return;
1190 
1191 	new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1192 	if (new_descs_count == 0) {
1193 		pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1194 		return;
1195 	}
1196 
1197 	new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1198 	if (unlikely(!new_log_buf)) {
1199 		pr_err("log_buf_len: %lu text bytes not available\n",
1200 		       new_log_buf_len);
1201 		return;
1202 	}
1203 
1204 	new_descs_size = new_descs_count * sizeof(struct prb_desc);
1205 	new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1206 	if (unlikely(!new_descs)) {
1207 		pr_err("log_buf_len: %zu desc bytes not available\n",
1208 		       new_descs_size);
1209 		goto err_free_log_buf;
1210 	}
1211 
1212 	new_infos_size = new_descs_count * sizeof(struct printk_info);
1213 	new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1214 	if (unlikely(!new_infos)) {
1215 		pr_err("log_buf_len: %zu info bytes not available\n",
1216 		       new_infos_size);
1217 		goto err_free_descs;
1218 	}
1219 
1220 	prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1221 
1222 	prb_init(&printk_rb_dynamic,
1223 		 new_log_buf, ilog2(new_log_buf_len),
1224 		 new_descs, ilog2(new_descs_count),
1225 		 new_infos);
1226 
1227 	local_irq_save(flags);
1228 
1229 	log_buf_len = new_log_buf_len;
1230 	log_buf = new_log_buf;
1231 	new_log_buf_len = 0;
1232 
1233 	free = __LOG_BUF_LEN;
1234 	prb_for_each_record(0, &printk_rb_static, seq, &r) {
1235 		text_size = add_to_rb(&printk_rb_dynamic, &r);
1236 		if (text_size > free)
1237 			free = 0;
1238 		else
1239 			free -= text_size;
1240 	}
1241 
1242 	prb = &printk_rb_dynamic;
1243 
1244 	local_irq_restore(flags);
1245 
1246 	/*
1247 	 * Copy any remaining messages that might have appeared from
1248 	 * NMI context after copying but before switching to the
1249 	 * dynamic buffer.
1250 	 */
1251 	prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1252 		text_size = add_to_rb(&printk_rb_dynamic, &r);
1253 		if (text_size > free)
1254 			free = 0;
1255 		else
1256 			free -= text_size;
1257 	}
1258 
1259 	if (seq != prb_next_seq(&printk_rb_static)) {
1260 		pr_err("dropped %llu messages\n",
1261 		       prb_next_seq(&printk_rb_static) - seq);
1262 	}
1263 
1264 	pr_info("log_buf_len: %u bytes\n", log_buf_len);
1265 	pr_info("early log buf free: %u(%u%%)\n",
1266 		free, (free * 100) / __LOG_BUF_LEN);
1267 	return;
1268 
1269 err_free_descs:
1270 	memblock_free(new_descs, new_descs_size);
1271 err_free_log_buf:
1272 	memblock_free(new_log_buf, new_log_buf_len);
1273 }
1274 
1275 static bool __read_mostly ignore_loglevel;
1276 
ignore_loglevel_setup(char * str)1277 static int __init ignore_loglevel_setup(char *str)
1278 {
1279 	ignore_loglevel = true;
1280 	pr_info("debug: ignoring loglevel setting.\n");
1281 
1282 	return 0;
1283 }
1284 
1285 early_param("ignore_loglevel", ignore_loglevel_setup);
1286 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1287 MODULE_PARM_DESC(ignore_loglevel,
1288 		 "ignore loglevel setting (prints all kernel messages to the console)");
1289 
suppress_message_printing(int level)1290 static bool suppress_message_printing(int level)
1291 {
1292 	return (level >= console_loglevel && !ignore_loglevel);
1293 }
1294 
1295 #ifdef CONFIG_BOOT_PRINTK_DELAY
1296 
1297 static int boot_delay; /* msecs delay after each printk during bootup */
1298 static unsigned long long loops_per_msec;	/* based on boot_delay */
1299 
boot_delay_setup(char * str)1300 static int __init boot_delay_setup(char *str)
1301 {
1302 	unsigned long lpj;
1303 
1304 	lpj = preset_lpj ? preset_lpj : 1000000;	/* some guess */
1305 	loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1306 
1307 	get_option(&str, &boot_delay);
1308 	if (boot_delay > 10 * 1000)
1309 		boot_delay = 0;
1310 
1311 	pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1312 		"HZ: %d, loops_per_msec: %llu\n",
1313 		boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1314 	return 0;
1315 }
1316 early_param("boot_delay", boot_delay_setup);
1317 
boot_delay_msec(int level)1318 static void boot_delay_msec(int level)
1319 {
1320 	unsigned long long k;
1321 	unsigned long timeout;
1322 
1323 	if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1324 		|| suppress_message_printing(level)) {
1325 		return;
1326 	}
1327 
1328 	k = (unsigned long long)loops_per_msec * boot_delay;
1329 
1330 	timeout = jiffies + msecs_to_jiffies(boot_delay);
1331 	while (k) {
1332 		k--;
1333 		cpu_relax();
1334 		/*
1335 		 * use (volatile) jiffies to prevent
1336 		 * compiler reduction; loop termination via jiffies
1337 		 * is secondary and may or may not happen.
1338 		 */
1339 		if (time_after(jiffies, timeout))
1340 			break;
1341 		touch_nmi_watchdog();
1342 	}
1343 }
1344 #else
boot_delay_msec(int level)1345 static inline void boot_delay_msec(int level)
1346 {
1347 }
1348 #endif
1349 
1350 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1351 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1352 
print_syslog(unsigned int level,char * buf)1353 static size_t print_syslog(unsigned int level, char *buf)
1354 {
1355 	return sprintf(buf, "<%u>", level);
1356 }
1357 
print_time(u64 ts,char * buf)1358 static size_t print_time(u64 ts, char *buf)
1359 {
1360 	unsigned long rem_nsec = do_div(ts, 1000000000);
1361 
1362 	return sprintf(buf, "[%5lu.%06lu]",
1363 		       (unsigned long)ts, rem_nsec / 1000);
1364 }
1365 
1366 #ifdef CONFIG_PRINTK_CALLER
print_caller(u32 id,char * buf)1367 static size_t print_caller(u32 id, char *buf)
1368 {
1369 	char caller[12];
1370 
1371 	snprintf(caller, sizeof(caller), "%c%u",
1372 		 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1373 	return sprintf(buf, "[%6s]", caller);
1374 }
1375 #else
1376 #define print_caller(id, buf) 0
1377 #endif
1378 
info_print_prefix(const struct printk_info * info,bool syslog,bool time,char * buf)1379 static size_t info_print_prefix(const struct printk_info  *info, bool syslog,
1380 				bool time, char *buf)
1381 {
1382 	size_t len = 0;
1383 
1384 	if (syslog)
1385 		len = print_syslog((info->facility << 3) | info->level, buf);
1386 
1387 	if (time)
1388 		len += print_time(info->ts_nsec, buf + len);
1389 
1390 	len += print_caller(info->caller_id, buf + len);
1391 
1392 	if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1393 		buf[len++] = ' ';
1394 		buf[len] = '\0';
1395 	}
1396 
1397 	return len;
1398 }
1399 
1400 /*
1401  * Prepare the record for printing. The text is shifted within the given
1402  * buffer to avoid a need for another one. The following operations are
1403  * done:
1404  *
1405  *   - Add prefix for each line.
1406  *   - Drop truncated lines that no longer fit into the buffer.
1407  *   - Add the trailing newline that has been removed in vprintk_store().
1408  *   - Add a string terminator.
1409  *
1410  * Since the produced string is always terminated, the maximum possible
1411  * return value is @r->text_buf_size - 1;
1412  *
1413  * Return: The length of the updated/prepared text, including the added
1414  * prefixes and the newline. The terminator is not counted. The dropped
1415  * line(s) are not counted.
1416  */
record_print_text(struct printk_record * r,bool syslog,bool time)1417 static size_t record_print_text(struct printk_record *r, bool syslog,
1418 				bool time)
1419 {
1420 	size_t text_len = r->info->text_len;
1421 	size_t buf_size = r->text_buf_size;
1422 	char *text = r->text_buf;
1423 	char prefix[PRINTK_PREFIX_MAX];
1424 	bool truncated = false;
1425 	size_t prefix_len;
1426 	size_t line_len;
1427 	size_t len = 0;
1428 	char *next;
1429 
1430 	/*
1431 	 * If the message was truncated because the buffer was not large
1432 	 * enough, treat the available text as if it were the full text.
1433 	 */
1434 	if (text_len > buf_size)
1435 		text_len = buf_size;
1436 
1437 	prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1438 
1439 	/*
1440 	 * @text_len: bytes of unprocessed text
1441 	 * @line_len: bytes of current line _without_ newline
1442 	 * @text:     pointer to beginning of current line
1443 	 * @len:      number of bytes prepared in r->text_buf
1444 	 */
1445 	for (;;) {
1446 		next = memchr(text, '\n', text_len);
1447 		if (next) {
1448 			line_len = next - text;
1449 		} else {
1450 			/* Drop truncated line(s). */
1451 			if (truncated)
1452 				break;
1453 			line_len = text_len;
1454 		}
1455 
1456 		/*
1457 		 * Truncate the text if there is not enough space to add the
1458 		 * prefix and a trailing newline and a terminator.
1459 		 */
1460 		if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1461 			/* Drop even the current line if no space. */
1462 			if (len + prefix_len + line_len + 1 + 1 > buf_size)
1463 				break;
1464 
1465 			text_len = buf_size - len - prefix_len - 1 - 1;
1466 			truncated = true;
1467 		}
1468 
1469 		memmove(text + prefix_len, text, text_len);
1470 		memcpy(text, prefix, prefix_len);
1471 
1472 		/*
1473 		 * Increment the prepared length to include the text and
1474 		 * prefix that were just moved+copied. Also increment for the
1475 		 * newline at the end of this line. If this is the last line,
1476 		 * there is no newline, but it will be added immediately below.
1477 		 */
1478 		len += prefix_len + line_len + 1;
1479 		if (text_len == line_len) {
1480 			/*
1481 			 * This is the last line. Add the trailing newline
1482 			 * removed in vprintk_store().
1483 			 */
1484 			text[prefix_len + line_len] = '\n';
1485 			break;
1486 		}
1487 
1488 		/*
1489 		 * Advance beyond the added prefix and the related line with
1490 		 * its newline.
1491 		 */
1492 		text += prefix_len + line_len + 1;
1493 
1494 		/*
1495 		 * The remaining text has only decreased by the line with its
1496 		 * newline.
1497 		 *
1498 		 * Note that @text_len can become zero. It happens when @text
1499 		 * ended with a newline (either due to truncation or the
1500 		 * original string ending with "\n\n"). The loop is correctly
1501 		 * repeated and (if not truncated) an empty line with a prefix
1502 		 * will be prepared.
1503 		 */
1504 		text_len -= line_len + 1;
1505 	}
1506 
1507 	/*
1508 	 * If a buffer was provided, it will be terminated. Space for the
1509 	 * string terminator is guaranteed to be available. The terminator is
1510 	 * not counted in the return value.
1511 	 */
1512 	if (buf_size > 0)
1513 		r->text_buf[len] = 0;
1514 
1515 	return len;
1516 }
1517 
get_record_print_text_size(struct printk_info * info,unsigned int line_count,bool syslog,bool time)1518 static size_t get_record_print_text_size(struct printk_info *info,
1519 					 unsigned int line_count,
1520 					 bool syslog, bool time)
1521 {
1522 	char prefix[PRINTK_PREFIX_MAX];
1523 	size_t prefix_len;
1524 
1525 	prefix_len = info_print_prefix(info, syslog, time, prefix);
1526 
1527 	/*
1528 	 * Each line will be preceded with a prefix. The intermediate
1529 	 * newlines are already within the text, but a final trailing
1530 	 * newline will be added.
1531 	 */
1532 	return ((prefix_len * line_count) + info->text_len + 1);
1533 }
1534 
1535 /*
1536  * Beginning with @start_seq, find the first record where it and all following
1537  * records up to (but not including) @max_seq fit into @size.
1538  *
1539  * @max_seq is simply an upper bound and does not need to exist. If the caller
1540  * does not require an upper bound, -1 can be used for @max_seq.
1541  */
find_first_fitting_seq(u64 start_seq,u64 max_seq,size_t size,bool syslog,bool time)1542 static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1543 				  bool syslog, bool time)
1544 {
1545 	struct printk_info info;
1546 	unsigned int line_count;
1547 	size_t len = 0;
1548 	u64 seq;
1549 
1550 	/* Determine the size of the records up to @max_seq. */
1551 	prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1552 		if (info.seq >= max_seq)
1553 			break;
1554 		len += get_record_print_text_size(&info, line_count, syslog, time);
1555 	}
1556 
1557 	/*
1558 	 * Adjust the upper bound for the next loop to avoid subtracting
1559 	 * lengths that were never added.
1560 	 */
1561 	if (seq < max_seq)
1562 		max_seq = seq;
1563 
1564 	/*
1565 	 * Move first record forward until length fits into the buffer. Ignore
1566 	 * newest messages that were not counted in the above cycle. Messages
1567 	 * might appear and get lost in the meantime. This is a best effort
1568 	 * that prevents an infinite loop that could occur with a retry.
1569 	 */
1570 	prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1571 		if (len <= size || info.seq >= max_seq)
1572 			break;
1573 		len -= get_record_print_text_size(&info, line_count, syslog, time);
1574 	}
1575 
1576 	return seq;
1577 }
1578 
1579 /* The caller is responsible for making sure @size is greater than 0. */
syslog_print(char __user * buf,int size)1580 static int syslog_print(char __user *buf, int size)
1581 {
1582 	struct printk_info info;
1583 	struct printk_record r;
1584 	char *text;
1585 	int len = 0;
1586 	u64 seq;
1587 
1588 	text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1589 	if (!text)
1590 		return -ENOMEM;
1591 
1592 	prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1593 
1594 	mutex_lock(&syslog_lock);
1595 
1596 	/*
1597 	 * Wait for the @syslog_seq record to be available. @syslog_seq may
1598 	 * change while waiting.
1599 	 */
1600 	do {
1601 		seq = syslog_seq;
1602 
1603 		mutex_unlock(&syslog_lock);
1604 		/*
1605 		 * Guarantee this task is visible on the waitqueue before
1606 		 * checking the wake condition.
1607 		 *
1608 		 * The full memory barrier within set_current_state() of
1609 		 * prepare_to_wait_event() pairs with the full memory barrier
1610 		 * within wq_has_sleeper().
1611 		 *
1612 		 * This pairs with __wake_up_klogd:A.
1613 		 */
1614 		len = wait_event_interruptible(log_wait,
1615 				prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
1616 		mutex_lock(&syslog_lock);
1617 
1618 		if (len)
1619 			goto out;
1620 	} while (syslog_seq != seq);
1621 
1622 	/*
1623 	 * Copy records that fit into the buffer. The above cycle makes sure
1624 	 * that the first record is always available.
1625 	 */
1626 	do {
1627 		size_t n;
1628 		size_t skip;
1629 		int err;
1630 
1631 		if (!prb_read_valid(prb, syslog_seq, &r))
1632 			break;
1633 
1634 		if (r.info->seq != syslog_seq) {
1635 			/* message is gone, move to next valid one */
1636 			syslog_seq = r.info->seq;
1637 			syslog_partial = 0;
1638 		}
1639 
1640 		/*
1641 		 * To keep reading/counting partial line consistent,
1642 		 * use printk_time value as of the beginning of a line.
1643 		 */
1644 		if (!syslog_partial)
1645 			syslog_time = printk_time;
1646 
1647 		skip = syslog_partial;
1648 		n = record_print_text(&r, true, syslog_time);
1649 		if (n - syslog_partial <= size) {
1650 			/* message fits into buffer, move forward */
1651 			syslog_seq = r.info->seq + 1;
1652 			n -= syslog_partial;
1653 			syslog_partial = 0;
1654 		} else if (!len){
1655 			/* partial read(), remember position */
1656 			n = size;
1657 			syslog_partial += n;
1658 		} else
1659 			n = 0;
1660 
1661 		if (!n)
1662 			break;
1663 
1664 		mutex_unlock(&syslog_lock);
1665 		err = copy_to_user(buf, text + skip, n);
1666 		mutex_lock(&syslog_lock);
1667 
1668 		if (err) {
1669 			if (!len)
1670 				len = -EFAULT;
1671 			break;
1672 		}
1673 
1674 		len += n;
1675 		size -= n;
1676 		buf += n;
1677 	} while (size);
1678 out:
1679 	mutex_unlock(&syslog_lock);
1680 	kfree(text);
1681 	return len;
1682 }
1683 
syslog_print_all(char __user * buf,int size,bool clear)1684 static int syslog_print_all(char __user *buf, int size, bool clear)
1685 {
1686 	struct printk_info info;
1687 	struct printk_record r;
1688 	char *text;
1689 	int len = 0;
1690 	u64 seq;
1691 	bool time;
1692 
1693 	text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1694 	if (!text)
1695 		return -ENOMEM;
1696 
1697 	time = printk_time;
1698 	/*
1699 	 * Find first record that fits, including all following records,
1700 	 * into the user-provided buffer for this dump.
1701 	 */
1702 	seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1703 				     size, true, time);
1704 
1705 	prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1706 
1707 	prb_for_each_record(seq, prb, seq, &r) {
1708 		int textlen;
1709 
1710 		textlen = record_print_text(&r, true, time);
1711 
1712 		if (len + textlen > size) {
1713 			seq--;
1714 			break;
1715 		}
1716 
1717 		if (copy_to_user(buf + len, text, textlen))
1718 			len = -EFAULT;
1719 		else
1720 			len += textlen;
1721 
1722 		if (len < 0)
1723 			break;
1724 	}
1725 
1726 	if (clear) {
1727 		mutex_lock(&syslog_lock);
1728 		latched_seq_write(&clear_seq, seq);
1729 		mutex_unlock(&syslog_lock);
1730 	}
1731 
1732 	kfree(text);
1733 	return len;
1734 }
1735 
syslog_clear(void)1736 static void syslog_clear(void)
1737 {
1738 	mutex_lock(&syslog_lock);
1739 	latched_seq_write(&clear_seq, prb_next_seq(prb));
1740 	mutex_unlock(&syslog_lock);
1741 }
1742 
do_syslog(int type,char __user * buf,int len,int source)1743 int do_syslog(int type, char __user *buf, int len, int source)
1744 {
1745 	struct printk_info info;
1746 	bool clear = false;
1747 	static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1748 	int error;
1749 
1750 	error = check_syslog_permissions(type, source);
1751 	if (error)
1752 		return error;
1753 
1754 	switch (type) {
1755 	case SYSLOG_ACTION_CLOSE:	/* Close log */
1756 		break;
1757 	case SYSLOG_ACTION_OPEN:	/* Open log */
1758 		break;
1759 	case SYSLOG_ACTION_READ:	/* Read from log */
1760 		if (!buf || len < 0)
1761 			return -EINVAL;
1762 		if (!len)
1763 			return 0;
1764 		if (!access_ok(buf, len))
1765 			return -EFAULT;
1766 		error = syslog_print(buf, len);
1767 		break;
1768 	/* Read/clear last kernel messages */
1769 	case SYSLOG_ACTION_READ_CLEAR:
1770 		clear = true;
1771 		fallthrough;
1772 	/* Read last kernel messages */
1773 	case SYSLOG_ACTION_READ_ALL:
1774 		if (!buf || len < 0)
1775 			return -EINVAL;
1776 		if (!len)
1777 			return 0;
1778 		if (!access_ok(buf, len))
1779 			return -EFAULT;
1780 		error = syslog_print_all(buf, len, clear);
1781 		break;
1782 	/* Clear ring buffer */
1783 	case SYSLOG_ACTION_CLEAR:
1784 		syslog_clear();
1785 		break;
1786 	/* Disable logging to console */
1787 	case SYSLOG_ACTION_CONSOLE_OFF:
1788 		if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1789 			saved_console_loglevel = console_loglevel;
1790 		console_loglevel = minimum_console_loglevel;
1791 		break;
1792 	/* Enable logging to console */
1793 	case SYSLOG_ACTION_CONSOLE_ON:
1794 		if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1795 			console_loglevel = saved_console_loglevel;
1796 			saved_console_loglevel = LOGLEVEL_DEFAULT;
1797 		}
1798 		break;
1799 	/* Set level of messages printed to console */
1800 	case SYSLOG_ACTION_CONSOLE_LEVEL:
1801 		if (len < 1 || len > 8)
1802 			return -EINVAL;
1803 		if (len < minimum_console_loglevel)
1804 			len = minimum_console_loglevel;
1805 		console_loglevel = len;
1806 		/* Implicitly re-enable logging to console */
1807 		saved_console_loglevel = LOGLEVEL_DEFAULT;
1808 		break;
1809 	/* Number of chars in the log buffer */
1810 	case SYSLOG_ACTION_SIZE_UNREAD:
1811 		mutex_lock(&syslog_lock);
1812 		if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1813 			/* No unread messages. */
1814 			mutex_unlock(&syslog_lock);
1815 			return 0;
1816 		}
1817 		if (info.seq != syslog_seq) {
1818 			/* messages are gone, move to first one */
1819 			syslog_seq = info.seq;
1820 			syslog_partial = 0;
1821 		}
1822 		if (source == SYSLOG_FROM_PROC) {
1823 			/*
1824 			 * Short-cut for poll(/"proc/kmsg") which simply checks
1825 			 * for pending data, not the size; return the count of
1826 			 * records, not the length.
1827 			 */
1828 			error = prb_next_seq(prb) - syslog_seq;
1829 		} else {
1830 			bool time = syslog_partial ? syslog_time : printk_time;
1831 			unsigned int line_count;
1832 			u64 seq;
1833 
1834 			prb_for_each_info(syslog_seq, prb, seq, &info,
1835 					  &line_count) {
1836 				error += get_record_print_text_size(&info, line_count,
1837 								    true, time);
1838 				time = printk_time;
1839 			}
1840 			error -= syslog_partial;
1841 		}
1842 		mutex_unlock(&syslog_lock);
1843 		break;
1844 	/* Size of the log buffer */
1845 	case SYSLOG_ACTION_SIZE_BUFFER:
1846 		error = log_buf_len;
1847 		break;
1848 	default:
1849 		error = -EINVAL;
1850 		break;
1851 	}
1852 
1853 	return error;
1854 }
1855 
SYSCALL_DEFINE3(syslog,int,type,char __user *,buf,int,len)1856 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1857 {
1858 	return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1859 }
1860 
1861 /*
1862  * Special console_lock variants that help to reduce the risk of soft-lockups.
1863  * They allow to pass console_lock to another printk() call using a busy wait.
1864  */
1865 
1866 #ifdef CONFIG_LOCKDEP
1867 static struct lockdep_map console_owner_dep_map = {
1868 	.name = "console_owner"
1869 };
1870 #endif
1871 
1872 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1873 static struct task_struct *console_owner;
1874 static bool console_waiter;
1875 
1876 /**
1877  * console_lock_spinning_enable - mark beginning of code where another
1878  *	thread might safely busy wait
1879  *
1880  * This basically converts console_lock into a spinlock. This marks
1881  * the section where the console_lock owner can not sleep, because
1882  * there may be a waiter spinning (like a spinlock). Also it must be
1883  * ready to hand over the lock at the end of the section.
1884  */
console_lock_spinning_enable(void)1885 void console_lock_spinning_enable(void)
1886 {
1887 	/*
1888 	 * Do not use spinning in panic(). The panic CPU wants to keep the lock.
1889 	 * Non-panic CPUs abandon the flush anyway.
1890 	 *
1891 	 * Just keep the lockdep annotation. The panic-CPU should avoid
1892 	 * taking console_owner_lock because it might cause a deadlock.
1893 	 * This looks like the easiest way how to prevent false lockdep
1894 	 * reports without handling races a lockless way.
1895 	 */
1896 	if (panic_in_progress())
1897 		goto lockdep;
1898 
1899 	raw_spin_lock(&console_owner_lock);
1900 	console_owner = current;
1901 	raw_spin_unlock(&console_owner_lock);
1902 
1903 lockdep:
1904 	/* The waiter may spin on us after setting console_owner */
1905 	spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1906 }
1907 
1908 /**
1909  * console_lock_spinning_disable_and_check - mark end of code where another
1910  *	thread was able to busy wait and check if there is a waiter
1911  * @cookie: cookie returned from console_srcu_read_lock()
1912  *
1913  * This is called at the end of the section where spinning is allowed.
1914  * It has two functions. First, it is a signal that it is no longer
1915  * safe to start busy waiting for the lock. Second, it checks if
1916  * there is a busy waiter and passes the lock rights to her.
1917  *
1918  * Important: Callers lose both the console_lock and the SRCU read lock if
1919  *	there was a busy waiter. They must not touch items synchronized by
1920  *	console_lock or SRCU read lock in this case.
1921  *
1922  * Return: 1 if the lock rights were passed, 0 otherwise.
1923  */
console_lock_spinning_disable_and_check(int cookie)1924 int console_lock_spinning_disable_and_check(int cookie)
1925 {
1926 	int waiter;
1927 
1928 	/*
1929 	 * Ignore spinning waiters during panic() because they might get stopped
1930 	 * or blocked at any time,
1931 	 *
1932 	 * It is safe because nobody is allowed to start spinning during panic
1933 	 * in the first place. If there has been a waiter then non panic CPUs
1934 	 * might stay spinning. They would get stopped anyway. The panic context
1935 	 * will never start spinning and an interrupted spin on panic CPU will
1936 	 * never continue.
1937 	 */
1938 	if (panic_in_progress()) {
1939 		/* Keep lockdep happy. */
1940 		spin_release(&console_owner_dep_map, _THIS_IP_);
1941 		return 0;
1942 	}
1943 
1944 	raw_spin_lock(&console_owner_lock);
1945 	waiter = READ_ONCE(console_waiter);
1946 	console_owner = NULL;
1947 	raw_spin_unlock(&console_owner_lock);
1948 
1949 	if (!waiter) {
1950 		spin_release(&console_owner_dep_map, _THIS_IP_);
1951 		return 0;
1952 	}
1953 
1954 	/* The waiter is now free to continue */
1955 	WRITE_ONCE(console_waiter, false);
1956 
1957 	spin_release(&console_owner_dep_map, _THIS_IP_);
1958 
1959 	/*
1960 	 * Preserve lockdep lock ordering. Release the SRCU read lock before
1961 	 * releasing the console_lock.
1962 	 */
1963 	console_srcu_read_unlock(cookie);
1964 
1965 	/*
1966 	 * Hand off console_lock to waiter. The waiter will perform
1967 	 * the up(). After this, the waiter is the console_lock owner.
1968 	 */
1969 	mutex_release(&console_lock_dep_map, _THIS_IP_);
1970 	return 1;
1971 }
1972 
1973 /**
1974  * console_trylock_spinning - try to get console_lock by busy waiting
1975  *
1976  * This allows to busy wait for the console_lock when the current
1977  * owner is running in specially marked sections. It means that
1978  * the current owner is running and cannot reschedule until it
1979  * is ready to lose the lock.
1980  *
1981  * Return: 1 if we got the lock, 0 othrewise
1982  */
console_trylock_spinning(void)1983 static int console_trylock_spinning(void)
1984 {
1985 	struct task_struct *owner = NULL;
1986 	bool waiter;
1987 	bool spin = false;
1988 	unsigned long flags;
1989 
1990 	if (console_trylock())
1991 		return 1;
1992 
1993 	/*
1994 	 * It's unsafe to spin once a panic has begun. If we are the
1995 	 * panic CPU, we may have already halted the owner of the
1996 	 * console_sem. If we are not the panic CPU, then we should
1997 	 * avoid taking console_sem, so the panic CPU has a better
1998 	 * chance of cleanly acquiring it later.
1999 	 */
2000 	if (panic_in_progress())
2001 		return 0;
2002 
2003 	printk_safe_enter_irqsave(flags);
2004 
2005 	raw_spin_lock(&console_owner_lock);
2006 	owner = READ_ONCE(console_owner);
2007 	waiter = READ_ONCE(console_waiter);
2008 	if (!waiter && owner && owner != current) {
2009 		WRITE_ONCE(console_waiter, true);
2010 		spin = true;
2011 	}
2012 	raw_spin_unlock(&console_owner_lock);
2013 
2014 	/*
2015 	 * If there is an active printk() writing to the
2016 	 * consoles, instead of having it write our data too,
2017 	 * see if we can offload that load from the active
2018 	 * printer, and do some printing ourselves.
2019 	 * Go into a spin only if there isn't already a waiter
2020 	 * spinning, and there is an active printer, and
2021 	 * that active printer isn't us (recursive printk?).
2022 	 */
2023 	if (!spin) {
2024 		printk_safe_exit_irqrestore(flags);
2025 		return 0;
2026 	}
2027 
2028 	/* We spin waiting for the owner to release us */
2029 	spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
2030 	/* Owner will clear console_waiter on hand off */
2031 	while (READ_ONCE(console_waiter))
2032 		cpu_relax();
2033 	spin_release(&console_owner_dep_map, _THIS_IP_);
2034 
2035 	printk_safe_exit_irqrestore(flags);
2036 	/*
2037 	 * The owner passed the console lock to us.
2038 	 * Since we did not spin on console lock, annotate
2039 	 * this as a trylock. Otherwise lockdep will
2040 	 * complain.
2041 	 */
2042 	mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
2043 
2044 	/*
2045 	 * Update @console_may_schedule for trylock because the previous
2046 	 * owner may have been schedulable.
2047 	 */
2048 	console_may_schedule = 0;
2049 
2050 	return 1;
2051 }
2052 
2053 /*
2054  * Recursion is tracked separately on each CPU. If NMIs are supported, an
2055  * additional NMI context per CPU is also separately tracked. Until per-CPU
2056  * is available, a separate "early tracking" is performed.
2057  */
2058 static DEFINE_PER_CPU(u8, printk_count);
2059 static u8 printk_count_early;
2060 #ifdef CONFIG_HAVE_NMI
2061 static DEFINE_PER_CPU(u8, printk_count_nmi);
2062 static u8 printk_count_nmi_early;
2063 #endif
2064 
2065 /*
2066  * Recursion is limited to keep the output sane. printk() should not require
2067  * more than 1 level of recursion (allowing, for example, printk() to trigger
2068  * a WARN), but a higher value is used in case some printk-internal errors
2069  * exist, such as the ringbuffer validation checks failing.
2070  */
2071 #define PRINTK_MAX_RECURSION 3
2072 
2073 /*
2074  * Return a pointer to the dedicated counter for the CPU+context of the
2075  * caller.
2076  */
__printk_recursion_counter(void)2077 static u8 *__printk_recursion_counter(void)
2078 {
2079 #ifdef CONFIG_HAVE_NMI
2080 	if (in_nmi()) {
2081 		if (printk_percpu_data_ready())
2082 			return this_cpu_ptr(&printk_count_nmi);
2083 		return &printk_count_nmi_early;
2084 	}
2085 #endif
2086 	if (printk_percpu_data_ready())
2087 		return this_cpu_ptr(&printk_count);
2088 	return &printk_count_early;
2089 }
2090 
2091 /*
2092  * Enter recursion tracking. Interrupts are disabled to simplify tracking.
2093  * The caller must check the boolean return value to see if the recursion is
2094  * allowed. On failure, interrupts are not disabled.
2095  *
2096  * @recursion_ptr must be a variable of type (u8 *) and is the same variable
2097  * that is passed to printk_exit_irqrestore().
2098  */
2099 #define printk_enter_irqsave(recursion_ptr, flags)	\
2100 ({							\
2101 	bool success = true;				\
2102 							\
2103 	typecheck(u8 *, recursion_ptr);			\
2104 	local_irq_save(flags);				\
2105 	(recursion_ptr) = __printk_recursion_counter();	\
2106 	if (*(recursion_ptr) > PRINTK_MAX_RECURSION) {	\
2107 		local_irq_restore(flags);		\
2108 		success = false;			\
2109 	} else {					\
2110 		(*(recursion_ptr))++;			\
2111 	}						\
2112 	success;					\
2113 })
2114 
2115 /* Exit recursion tracking, restoring interrupts. */
2116 #define printk_exit_irqrestore(recursion_ptr, flags)	\
2117 	do {						\
2118 		typecheck(u8 *, recursion_ptr);		\
2119 		(*(recursion_ptr))--;			\
2120 		local_irq_restore(flags);		\
2121 	} while (0)
2122 
2123 int printk_delay_msec __read_mostly;
2124 
printk_delay(int level)2125 static inline void printk_delay(int level)
2126 {
2127 	boot_delay_msec(level);
2128 
2129 	if (unlikely(printk_delay_msec)) {
2130 		int m = printk_delay_msec;
2131 
2132 		while (m--) {
2133 			mdelay(1);
2134 			touch_nmi_watchdog();
2135 		}
2136 	}
2137 }
2138 
printk_caller_id(void)2139 static inline u32 printk_caller_id(void)
2140 {
2141 	return in_task() ? task_pid_nr(current) :
2142 		0x80000000 + smp_processor_id();
2143 }
2144 
2145 /**
2146  * printk_parse_prefix - Parse level and control flags.
2147  *
2148  * @text:     The terminated text message.
2149  * @level:    A pointer to the current level value, will be updated.
2150  * @flags:    A pointer to the current printk_info flags, will be updated.
2151  *
2152  * @level may be NULL if the caller is not interested in the parsed value.
2153  * Otherwise the variable pointed to by @level must be set to
2154  * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2155  *
2156  * @flags may be NULL if the caller is not interested in the parsed value.
2157  * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2158  * value.
2159  *
2160  * Return: The length of the parsed level and control flags.
2161  */
printk_parse_prefix(const char * text,int * level,enum printk_info_flags * flags)2162 u16 printk_parse_prefix(const char *text, int *level,
2163 			enum printk_info_flags *flags)
2164 {
2165 	u16 prefix_len = 0;
2166 	int kern_level;
2167 
2168 	while (*text) {
2169 		kern_level = printk_get_level(text);
2170 		if (!kern_level)
2171 			break;
2172 
2173 		switch (kern_level) {
2174 		case '0' ... '7':
2175 			if (level && *level == LOGLEVEL_DEFAULT)
2176 				*level = kern_level - '0';
2177 			break;
2178 		case 'c':	/* KERN_CONT */
2179 			if (flags)
2180 				*flags |= LOG_CONT;
2181 		}
2182 
2183 		prefix_len += 2;
2184 		text += 2;
2185 	}
2186 
2187 	return prefix_len;
2188 }
2189 
2190 __printf(5, 0)
printk_sprint(char * text,u16 size,int facility,enum printk_info_flags * flags,const char * fmt,va_list args)2191 static u16 printk_sprint(char *text, u16 size, int facility,
2192 			 enum printk_info_flags *flags, const char *fmt,
2193 			 va_list args)
2194 {
2195 	u16 text_len;
2196 
2197 	text_len = vscnprintf(text, size, fmt, args);
2198 
2199 	/* Mark and strip a trailing newline. */
2200 	if (text_len && text[text_len - 1] == '\n') {
2201 		text_len--;
2202 		*flags |= LOG_NEWLINE;
2203 	}
2204 
2205 	/* Strip log level and control flags. */
2206 	if (facility == 0) {
2207 		u16 prefix_len;
2208 
2209 		prefix_len = printk_parse_prefix(text, NULL, NULL);
2210 		if (prefix_len) {
2211 			text_len -= prefix_len;
2212 			memmove(text, text + prefix_len, text_len);
2213 		}
2214 	}
2215 
2216 	trace_console(text, text_len);
2217 
2218 	return text_len;
2219 }
2220 
2221 __printf(4, 0)
vprintk_store(int facility,int level,const struct dev_printk_info * dev_info,const char * fmt,va_list args)2222 int vprintk_store(int facility, int level,
2223 		  const struct dev_printk_info *dev_info,
2224 		  const char *fmt, va_list args)
2225 {
2226 	struct prb_reserved_entry e;
2227 	enum printk_info_flags flags = 0;
2228 	struct printk_record r;
2229 	unsigned long irqflags;
2230 	u16 trunc_msg_len = 0;
2231 	char prefix_buf[8];
2232 	u8 *recursion_ptr;
2233 	u16 reserve_size;
2234 	va_list args2;
2235 	u32 caller_id;
2236 	u16 text_len;
2237 	int ret = 0;
2238 	u64 ts_nsec;
2239 
2240 	if (!printk_enter_irqsave(recursion_ptr, irqflags))
2241 		return 0;
2242 
2243 	/*
2244 	 * Since the duration of printk() can vary depending on the message
2245 	 * and state of the ringbuffer, grab the timestamp now so that it is
2246 	 * close to the call of printk(). This provides a more deterministic
2247 	 * timestamp with respect to the caller.
2248 	 */
2249 	ts_nsec = local_clock();
2250 
2251 	caller_id = printk_caller_id();
2252 
2253 	/*
2254 	 * The sprintf needs to come first since the syslog prefix might be
2255 	 * passed in as a parameter. An extra byte must be reserved so that
2256 	 * later the vscnprintf() into the reserved buffer has room for the
2257 	 * terminating '\0', which is not counted by vsnprintf().
2258 	 */
2259 	va_copy(args2, args);
2260 	reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2261 	va_end(args2);
2262 
2263 	if (reserve_size > PRINTKRB_RECORD_MAX)
2264 		reserve_size = PRINTKRB_RECORD_MAX;
2265 
2266 	/* Extract log level or control flags. */
2267 	if (facility == 0)
2268 		printk_parse_prefix(&prefix_buf[0], &level, &flags);
2269 
2270 	if (level == LOGLEVEL_DEFAULT)
2271 		level = default_message_loglevel;
2272 
2273 	if (dev_info)
2274 		flags |= LOG_NEWLINE;
2275 
2276 	if (flags & LOG_CONT) {
2277 		prb_rec_init_wr(&r, reserve_size);
2278 		if (prb_reserve_in_last(&e, prb, &r, caller_id, PRINTKRB_RECORD_MAX)) {
2279 			text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2280 						 facility, &flags, fmt, args);
2281 			r.info->text_len += text_len;
2282 
2283 			if (flags & LOG_NEWLINE) {
2284 				r.info->flags |= LOG_NEWLINE;
2285 				prb_final_commit(&e);
2286 			} else {
2287 				prb_commit(&e);
2288 			}
2289 
2290 			ret = text_len;
2291 			goto out;
2292 		}
2293 	}
2294 
2295 	/*
2296 	 * Explicitly initialize the record before every prb_reserve() call.
2297 	 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2298 	 * structure when they fail.
2299 	 */
2300 	prb_rec_init_wr(&r, reserve_size);
2301 	if (!prb_reserve(&e, prb, &r)) {
2302 		/* truncate the message if it is too long for empty buffer */
2303 		truncate_msg(&reserve_size, &trunc_msg_len);
2304 
2305 		prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2306 		if (!prb_reserve(&e, prb, &r))
2307 			goto out;
2308 	}
2309 
2310 	/* fill message */
2311 	text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2312 	if (trunc_msg_len)
2313 		memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2314 	r.info->text_len = text_len + trunc_msg_len;
2315 	r.info->facility = facility;
2316 	r.info->level = level & 7;
2317 	r.info->flags = flags & 0x1f;
2318 	r.info->ts_nsec = ts_nsec;
2319 	r.info->caller_id = caller_id;
2320 	if (dev_info)
2321 		memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2322 
2323 	/* A message without a trailing newline can be continued. */
2324 	if (!(flags & LOG_NEWLINE))
2325 		prb_commit(&e);
2326 	else
2327 		prb_final_commit(&e);
2328 
2329 	ret = text_len + trunc_msg_len;
2330 out:
2331 	printk_exit_irqrestore(recursion_ptr, irqflags);
2332 	return ret;
2333 }
2334 
2335 /*
2336  * This acts as a one-way switch to allow legacy consoles to print from
2337  * the printk() caller context on a panic CPU. It also attempts to flush
2338  * the legacy consoles in this context.
2339  */
printk_legacy_allow_panic_sync(void)2340 void printk_legacy_allow_panic_sync(void)
2341 {
2342 	struct console_flush_type ft;
2343 
2344 	legacy_allow_panic_sync = true;
2345 
2346 	printk_get_console_flush_type(&ft);
2347 	if (ft.legacy_direct) {
2348 		if (console_trylock())
2349 			console_unlock();
2350 	}
2351 }
2352 
vprintk_emit(int facility,int level,const struct dev_printk_info * dev_info,const char * fmt,va_list args)2353 asmlinkage int vprintk_emit(int facility, int level,
2354 			    const struct dev_printk_info *dev_info,
2355 			    const char *fmt, va_list args)
2356 {
2357 	struct console_flush_type ft;
2358 	int printed_len;
2359 
2360 	/* Suppress unimportant messages after panic happens */
2361 	if (unlikely(suppress_printk))
2362 		return 0;
2363 
2364 	/*
2365 	 * The messages on the panic CPU are the most important. If
2366 	 * non-panic CPUs are generating any messages, they will be
2367 	 * silently dropped.
2368 	 */
2369 	if (other_cpu_in_panic() && !panic_triggering_all_cpu_backtrace)
2370 		return 0;
2371 
2372 	printk_get_console_flush_type(&ft);
2373 
2374 	/* If called from the scheduler, we can not call up(). */
2375 	if (level == LOGLEVEL_SCHED) {
2376 		level = LOGLEVEL_DEFAULT;
2377 		ft.legacy_offload |= ft.legacy_direct;
2378 		ft.legacy_direct = false;
2379 	}
2380 
2381 	printk_delay(level);
2382 
2383 	printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2384 
2385 	if (ft.nbcon_atomic)
2386 		nbcon_atomic_flush_pending();
2387 
2388 	if (ft.nbcon_offload)
2389 		nbcon_kthreads_wake();
2390 
2391 	if (ft.legacy_direct) {
2392 		/*
2393 		 * The caller may be holding system-critical or
2394 		 * timing-sensitive locks. Disable preemption during
2395 		 * printing of all remaining records to all consoles so that
2396 		 * this context can return as soon as possible. Hopefully
2397 		 * another printk() caller will take over the printing.
2398 		 */
2399 		preempt_disable();
2400 		/*
2401 		 * Try to acquire and then immediately release the console
2402 		 * semaphore. The release will print out buffers. With the
2403 		 * spinning variant, this context tries to take over the
2404 		 * printing from another printing context.
2405 		 */
2406 		if (console_trylock_spinning())
2407 			console_unlock();
2408 		preempt_enable();
2409 	}
2410 
2411 	if (ft.legacy_offload)
2412 		defer_console_output();
2413 	else
2414 		wake_up_klogd();
2415 
2416 	return printed_len;
2417 }
2418 EXPORT_SYMBOL(vprintk_emit);
2419 
vprintk_default(const char * fmt,va_list args)2420 int vprintk_default(const char *fmt, va_list args)
2421 {
2422 	return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2423 }
2424 EXPORT_SYMBOL_GPL(vprintk_default);
2425 
_printk(const char * fmt,...)2426 asmlinkage __visible int _printk(const char *fmt, ...)
2427 {
2428 	va_list args;
2429 	int r;
2430 
2431 	va_start(args, fmt);
2432 	r = vprintk(fmt, args);
2433 	va_end(args);
2434 
2435 	return r;
2436 }
2437 EXPORT_SYMBOL(_printk);
2438 
2439 static bool pr_flush(int timeout_ms, bool reset_on_progress);
2440 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress);
2441 
2442 #else /* CONFIG_PRINTK */
2443 
2444 #define printk_time		false
2445 
2446 #define prb_read_valid(rb, seq, r)	false
2447 #define prb_first_valid_seq(rb)		0
2448 #define prb_next_seq(rb)		0
2449 
2450 static u64 syslog_seq;
2451 
pr_flush(int timeout_ms,bool reset_on_progress)2452 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)2453 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress) { return true; }
2454 
2455 #endif /* CONFIG_PRINTK */
2456 
2457 #ifdef CONFIG_EARLY_PRINTK
2458 struct console *early_console;
2459 
early_printk(const char * fmt,...)2460 asmlinkage __visible void early_printk(const char *fmt, ...)
2461 {
2462 	va_list ap;
2463 	char buf[512];
2464 	int n;
2465 
2466 	if (!early_console)
2467 		return;
2468 
2469 	va_start(ap, fmt);
2470 	n = vscnprintf(buf, sizeof(buf), fmt, ap);
2471 	va_end(ap);
2472 
2473 	early_console->write(early_console, buf, n);
2474 }
2475 #endif
2476 
set_user_specified(struct console_cmdline * c,bool user_specified)2477 static void set_user_specified(struct console_cmdline *c, bool user_specified)
2478 {
2479 	if (!user_specified)
2480 		return;
2481 
2482 	/*
2483 	 * @c console was defined by the user on the command line.
2484 	 * Do not clear when added twice also by SPCR or the device tree.
2485 	 */
2486 	c->user_specified = true;
2487 	/* At least one console defined by the user on the command line. */
2488 	console_set_on_cmdline = 1;
2489 }
2490 
__add_preferred_console(const char * name,const short idx,const char * devname,char * options,char * brl_options,bool user_specified)2491 static int __add_preferred_console(const char *name, const short idx,
2492 				   const char *devname, char *options,
2493 				   char *brl_options, bool user_specified)
2494 {
2495 	struct console_cmdline *c;
2496 	int i;
2497 
2498 	if (!name && !devname)
2499 		return -EINVAL;
2500 
2501 	/*
2502 	 * We use a signed short index for struct console for device drivers to
2503 	 * indicate a not yet assigned index or port. However, a negative index
2504 	 * value is not valid when the console name and index are defined on
2505 	 * the command line.
2506 	 */
2507 	if (name && idx < 0)
2508 		return -EINVAL;
2509 
2510 	/*
2511 	 *	See if this tty is not yet registered, and
2512 	 *	if we have a slot free.
2513 	 */
2514 	for (i = 0, c = console_cmdline;
2515 	     i < MAX_CMDLINECONSOLES && (c->name[0] || c->devname[0]);
2516 	     i++, c++) {
2517 		if ((name && strcmp(c->name, name) == 0 && c->index == idx) ||
2518 		    (devname && strcmp(c->devname, devname) == 0)) {
2519 			if (!brl_options)
2520 				preferred_console = i;
2521 			set_user_specified(c, user_specified);
2522 			return 0;
2523 		}
2524 	}
2525 	if (i == MAX_CMDLINECONSOLES)
2526 		return -E2BIG;
2527 	if (!brl_options)
2528 		preferred_console = i;
2529 	if (name)
2530 		strscpy(c->name, name);
2531 	if (devname)
2532 		strscpy(c->devname, devname);
2533 	c->options = options;
2534 	set_user_specified(c, user_specified);
2535 	braille_set_options(c, brl_options);
2536 
2537 	c->index = idx;
2538 	return 0;
2539 }
2540 
console_msg_format_setup(char * str)2541 static int __init console_msg_format_setup(char *str)
2542 {
2543 	if (!strcmp(str, "syslog"))
2544 		console_msg_format = MSG_FORMAT_SYSLOG;
2545 	if (!strcmp(str, "default"))
2546 		console_msg_format = MSG_FORMAT_DEFAULT;
2547 	return 1;
2548 }
2549 __setup("console_msg_format=", console_msg_format_setup);
2550 
2551 /*
2552  * Set up a console.  Called via do_early_param() in init/main.c
2553  * for each "console=" parameter in the boot command line.
2554  */
console_setup(char * str)2555 static int __init console_setup(char *str)
2556 {
2557 	static_assert(sizeof(console_cmdline[0].devname) >= sizeof(console_cmdline[0].name) + 4);
2558 	char buf[sizeof(console_cmdline[0].devname)];
2559 	char *brl_options = NULL;
2560 	char *ttyname = NULL;
2561 	char *devname = NULL;
2562 	char *options;
2563 	char *s;
2564 	int idx;
2565 
2566 	/*
2567 	 * console="" or console=null have been suggested as a way to
2568 	 * disable console output. Use ttynull that has been created
2569 	 * for exactly this purpose.
2570 	 */
2571 	if (str[0] == 0 || strcmp(str, "null") == 0) {
2572 		__add_preferred_console("ttynull", 0, NULL, NULL, NULL, true);
2573 		return 1;
2574 	}
2575 
2576 	if (_braille_console_setup(&str, &brl_options))
2577 		return 1;
2578 
2579 	/* For a DEVNAME:0.0 style console the character device is unknown early */
2580 	if (strchr(str, ':'))
2581 		devname = buf;
2582 	else
2583 		ttyname = buf;
2584 
2585 	/*
2586 	 * Decode str into name, index, options.
2587 	 */
2588 	if (ttyname && isdigit(str[0]))
2589 		scnprintf(buf, sizeof(buf), "ttyS%s", str);
2590 	else
2591 		strscpy(buf, str);
2592 
2593 	options = strchr(str, ',');
2594 	if (options)
2595 		*(options++) = 0;
2596 
2597 #ifdef __sparc__
2598 	if (!strcmp(str, "ttya"))
2599 		strscpy(buf, "ttyS0");
2600 	if (!strcmp(str, "ttyb"))
2601 		strscpy(buf, "ttyS1");
2602 #endif
2603 
2604 	for (s = buf; *s; s++)
2605 		if ((ttyname && isdigit(*s)) || *s == ',')
2606 			break;
2607 
2608 	/* @idx will get defined when devname matches. */
2609 	if (devname)
2610 		idx = -1;
2611 	else
2612 		idx = simple_strtoul(s, NULL, 10);
2613 
2614 	*s = 0;
2615 
2616 	__add_preferred_console(ttyname, idx, devname, options, brl_options, true);
2617 	return 1;
2618 }
2619 __setup("console=", console_setup);
2620 
2621 /**
2622  * add_preferred_console - add a device to the list of preferred consoles.
2623  * @name: device name
2624  * @idx: device index
2625  * @options: options for this console
2626  *
2627  * The last preferred console added will be used for kernel messages
2628  * and stdin/out/err for init.  Normally this is used by console_setup
2629  * above to handle user-supplied console arguments; however it can also
2630  * be used by arch-specific code either to override the user or more
2631  * commonly to provide a default console (ie from PROM variables) when
2632  * the user has not supplied one.
2633  */
add_preferred_console(const char * name,const short idx,char * options)2634 int add_preferred_console(const char *name, const short idx, char *options)
2635 {
2636 	return __add_preferred_console(name, idx, NULL, options, NULL, false);
2637 }
2638 
2639 /**
2640  * match_devname_and_update_preferred_console - Update a preferred console
2641  *	when matching devname is found.
2642  * @devname: DEVNAME:0.0 style device name
2643  * @name: Name of the corresponding console driver, e.g. "ttyS"
2644  * @idx: Console index, e.g. port number.
2645  *
2646  * The function checks whether a device with the given @devname is
2647  * preferred via the console=DEVNAME:0.0 command line option.
2648  * It fills the missing console driver name and console index
2649  * so that a later register_console() call could find (match)
2650  * and enable this device.
2651  *
2652  * It might be used when a driver subsystem initializes particular
2653  * devices with already known DEVNAME:0.0 style names. And it
2654  * could predict which console driver name and index this device
2655  * would later get associated with.
2656  *
2657  * Return: 0 on success, negative error code on failure.
2658  */
match_devname_and_update_preferred_console(const char * devname,const char * name,const short idx)2659 int match_devname_and_update_preferred_console(const char *devname,
2660 					       const char *name,
2661 					       const short idx)
2662 {
2663 	struct console_cmdline *c = console_cmdline;
2664 	int i;
2665 
2666 	if (!devname || !strlen(devname) || !name || !strlen(name) || idx < 0)
2667 		return -EINVAL;
2668 
2669 	for (i = 0; i < MAX_CMDLINECONSOLES && (c->name[0] || c->devname[0]);
2670 	     i++, c++) {
2671 		if (!strcmp(devname, c->devname)) {
2672 			pr_info("associate the preferred console \"%s\" with \"%s%d\"\n",
2673 				devname, name, idx);
2674 			strscpy(c->name, name);
2675 			c->index = idx;
2676 			return 0;
2677 		}
2678 	}
2679 
2680 	return -ENOENT;
2681 }
2682 EXPORT_SYMBOL_GPL(match_devname_and_update_preferred_console);
2683 
2684 bool console_suspend_enabled = true;
2685 EXPORT_SYMBOL(console_suspend_enabled);
2686 
console_suspend_disable(char * str)2687 static int __init console_suspend_disable(char *str)
2688 {
2689 	console_suspend_enabled = false;
2690 	return 1;
2691 }
2692 __setup("no_console_suspend", console_suspend_disable);
2693 module_param_named(console_suspend, console_suspend_enabled,
2694 		bool, S_IRUGO | S_IWUSR);
2695 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2696 	" and hibernate operations");
2697 
2698 static bool printk_console_no_auto_verbose;
2699 
console_verbose(void)2700 void console_verbose(void)
2701 {
2702 	if (console_loglevel && !printk_console_no_auto_verbose)
2703 		console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2704 }
2705 EXPORT_SYMBOL_GPL(console_verbose);
2706 
2707 module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2708 MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2709 
2710 /**
2711  * suspend_console - suspend the console subsystem
2712  *
2713  * This disables printk() while we go into suspend states
2714  */
suspend_console(void)2715 void suspend_console(void)
2716 {
2717 	struct console *con;
2718 
2719 	if (!console_suspend_enabled)
2720 		return;
2721 	pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2722 	pr_flush(1000, true);
2723 
2724 	console_list_lock();
2725 	for_each_console(con)
2726 		console_srcu_write_flags(con, con->flags | CON_SUSPENDED);
2727 	console_list_unlock();
2728 
2729 	/*
2730 	 * Ensure that all SRCU list walks have completed. All printing
2731 	 * contexts must be able to see that they are suspended so that it
2732 	 * is guaranteed that all printing has stopped when this function
2733 	 * completes.
2734 	 */
2735 	synchronize_srcu(&console_srcu);
2736 }
2737 
resume_console(void)2738 void resume_console(void)
2739 {
2740 	struct console_flush_type ft;
2741 	struct console *con;
2742 
2743 	if (!console_suspend_enabled)
2744 		return;
2745 
2746 	console_list_lock();
2747 	for_each_console(con)
2748 		console_srcu_write_flags(con, con->flags & ~CON_SUSPENDED);
2749 	console_list_unlock();
2750 
2751 	/*
2752 	 * Ensure that all SRCU list walks have completed. All printing
2753 	 * contexts must be able to see they are no longer suspended so
2754 	 * that they are guaranteed to wake up and resume printing.
2755 	 */
2756 	synchronize_srcu(&console_srcu);
2757 
2758 	printk_get_console_flush_type(&ft);
2759 	if (ft.nbcon_offload)
2760 		nbcon_kthreads_wake();
2761 	if (ft.legacy_offload)
2762 		defer_console_output();
2763 
2764 	pr_flush(1000, true);
2765 }
2766 
2767 /**
2768  * console_cpu_notify - print deferred console messages after CPU hotplug
2769  * @cpu: unused
2770  *
2771  * If printk() is called from a CPU that is not online yet, the messages
2772  * will be printed on the console only if there are CON_ANYTIME consoles.
2773  * This function is called when a new CPU comes online (or fails to come
2774  * up) or goes offline.
2775  */
console_cpu_notify(unsigned int cpu)2776 static int console_cpu_notify(unsigned int cpu)
2777 {
2778 	struct console_flush_type ft;
2779 
2780 	if (!cpuhp_tasks_frozen) {
2781 		printk_get_console_flush_type(&ft);
2782 		if (ft.nbcon_atomic)
2783 			nbcon_atomic_flush_pending();
2784 		if (ft.legacy_direct) {
2785 			if (console_trylock())
2786 				console_unlock();
2787 		}
2788 	}
2789 	return 0;
2790 }
2791 
2792 /**
2793  * console_lock - block the console subsystem from printing
2794  *
2795  * Acquires a lock which guarantees that no consoles will
2796  * be in or enter their write() callback.
2797  *
2798  * Can sleep, returns nothing.
2799  */
console_lock(void)2800 void console_lock(void)
2801 {
2802 	might_sleep();
2803 
2804 	/* On panic, the console_lock must be left to the panic cpu. */
2805 	while (other_cpu_in_panic())
2806 		msleep(1000);
2807 
2808 	down_console_sem();
2809 	console_locked = 1;
2810 	console_may_schedule = 1;
2811 }
2812 EXPORT_SYMBOL(console_lock);
2813 
2814 /**
2815  * console_trylock - try to block the console subsystem from printing
2816  *
2817  * Try to acquire a lock which guarantees that no consoles will
2818  * be in or enter their write() callback.
2819  *
2820  * returns 1 on success, and 0 on failure to acquire the lock.
2821  */
console_trylock(void)2822 int console_trylock(void)
2823 {
2824 	/* On panic, the console_lock must be left to the panic cpu. */
2825 	if (other_cpu_in_panic())
2826 		return 0;
2827 	if (down_trylock_console_sem())
2828 		return 0;
2829 	console_locked = 1;
2830 	console_may_schedule = 0;
2831 	return 1;
2832 }
2833 EXPORT_SYMBOL(console_trylock);
2834 
is_console_locked(void)2835 int is_console_locked(void)
2836 {
2837 	return console_locked;
2838 }
2839 EXPORT_SYMBOL(is_console_locked);
2840 
__console_unlock(void)2841 static void __console_unlock(void)
2842 {
2843 	console_locked = 0;
2844 	up_console_sem();
2845 }
2846 
2847 #ifdef CONFIG_PRINTK
2848 
2849 /*
2850  * Prepend the message in @pmsg->pbufs->outbuf. This is achieved by shifting
2851  * the existing message over and inserting the scratchbuf message.
2852  *
2853  * @pmsg is the original printk message.
2854  * @fmt is the printf format of the message which will prepend the existing one.
2855  *
2856  * If there is not enough space in @pmsg->pbufs->outbuf, the existing
2857  * message text will be sufficiently truncated.
2858  *
2859  * If @pmsg->pbufs->outbuf is modified, @pmsg->outbuf_len is updated.
2860  */
2861 __printf(2, 3)
console_prepend_message(struct printk_message * pmsg,const char * fmt,...)2862 static void console_prepend_message(struct printk_message *pmsg, const char *fmt, ...)
2863 {
2864 	struct printk_buffers *pbufs = pmsg->pbufs;
2865 	const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2866 	const size_t outbuf_sz = sizeof(pbufs->outbuf);
2867 	char *scratchbuf = &pbufs->scratchbuf[0];
2868 	char *outbuf = &pbufs->outbuf[0];
2869 	va_list args;
2870 	size_t len;
2871 
2872 	va_start(args, fmt);
2873 	len = vscnprintf(scratchbuf, scratchbuf_sz, fmt, args);
2874 	va_end(args);
2875 
2876 	/*
2877 	 * Make sure outbuf is sufficiently large before prepending.
2878 	 * Keep at least the prefix when the message must be truncated.
2879 	 * It is a rather theoretical problem when someone tries to
2880 	 * use a minimalist buffer.
2881 	 */
2882 	if (WARN_ON_ONCE(len + PRINTK_PREFIX_MAX >= outbuf_sz))
2883 		return;
2884 
2885 	if (pmsg->outbuf_len + len >= outbuf_sz) {
2886 		/* Truncate the message, but keep it terminated. */
2887 		pmsg->outbuf_len = outbuf_sz - (len + 1);
2888 		outbuf[pmsg->outbuf_len] = 0;
2889 	}
2890 
2891 	memmove(outbuf + len, outbuf, pmsg->outbuf_len + 1);
2892 	memcpy(outbuf, scratchbuf, len);
2893 	pmsg->outbuf_len += len;
2894 }
2895 
2896 /*
2897  * Prepend the message in @pmsg->pbufs->outbuf with a "dropped message".
2898  * @pmsg->outbuf_len is updated appropriately.
2899  *
2900  * @pmsg is the printk message to prepend.
2901  *
2902  * @dropped is the dropped count to report in the dropped message.
2903  */
console_prepend_dropped(struct printk_message * pmsg,unsigned long dropped)2904 void console_prepend_dropped(struct printk_message *pmsg, unsigned long dropped)
2905 {
2906 	console_prepend_message(pmsg, "** %lu printk messages dropped **\n", dropped);
2907 }
2908 
2909 /*
2910  * Prepend the message in @pmsg->pbufs->outbuf with a "replay message".
2911  * @pmsg->outbuf_len is updated appropriately.
2912  *
2913  * @pmsg is the printk message to prepend.
2914  */
console_prepend_replay(struct printk_message * pmsg)2915 void console_prepend_replay(struct printk_message *pmsg)
2916 {
2917 	console_prepend_message(pmsg, "** replaying previous printk message **\n");
2918 }
2919 
2920 /*
2921  * Read and format the specified record (or a later record if the specified
2922  * record is not available).
2923  *
2924  * @pmsg will contain the formatted result. @pmsg->pbufs must point to a
2925  * struct printk_buffers.
2926  *
2927  * @seq is the record to read and format. If it is not available, the next
2928  * valid record is read.
2929  *
2930  * @is_extended specifies if the message should be formatted for extended
2931  * console output.
2932  *
2933  * @may_supress specifies if records may be skipped based on loglevel.
2934  *
2935  * Returns false if no record is available. Otherwise true and all fields
2936  * of @pmsg are valid. (See the documentation of struct printk_message
2937  * for information about the @pmsg fields.)
2938  */
printk_get_next_message(struct printk_message * pmsg,u64 seq,bool is_extended,bool may_suppress)2939 bool printk_get_next_message(struct printk_message *pmsg, u64 seq,
2940 			     bool is_extended, bool may_suppress)
2941 {
2942 	struct printk_buffers *pbufs = pmsg->pbufs;
2943 	const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2944 	const size_t outbuf_sz = sizeof(pbufs->outbuf);
2945 	char *scratchbuf = &pbufs->scratchbuf[0];
2946 	char *outbuf = &pbufs->outbuf[0];
2947 	struct printk_info info;
2948 	struct printk_record r;
2949 	size_t len = 0;
2950 
2951 	/*
2952 	 * Formatting extended messages requires a separate buffer, so use the
2953 	 * scratch buffer to read in the ringbuffer text.
2954 	 *
2955 	 * Formatting normal messages is done in-place, so read the ringbuffer
2956 	 * text directly into the output buffer.
2957 	 */
2958 	if (is_extended)
2959 		prb_rec_init_rd(&r, &info, scratchbuf, scratchbuf_sz);
2960 	else
2961 		prb_rec_init_rd(&r, &info, outbuf, outbuf_sz);
2962 
2963 	if (!prb_read_valid(prb, seq, &r))
2964 		return false;
2965 
2966 	pmsg->seq = r.info->seq;
2967 	pmsg->dropped = r.info->seq - seq;
2968 
2969 	/* Skip record that has level above the console loglevel. */
2970 	if (may_suppress && suppress_message_printing(r.info->level))
2971 		goto out;
2972 
2973 	if (is_extended) {
2974 		len = info_print_ext_header(outbuf, outbuf_sz, r.info);
2975 		len += msg_print_ext_body(outbuf + len, outbuf_sz - len,
2976 					  &r.text_buf[0], r.info->text_len, &r.info->dev_info);
2977 	} else {
2978 		len = record_print_text(&r, console_msg_format & MSG_FORMAT_SYSLOG, printk_time);
2979 	}
2980 out:
2981 	pmsg->outbuf_len = len;
2982 	return true;
2983 }
2984 
2985 /*
2986  * Legacy console printing from printk() caller context does not respect
2987  * raw_spinlock/spinlock nesting. For !PREEMPT_RT the lockdep warning is a
2988  * false positive. For PREEMPT_RT the false positive condition does not
2989  * occur.
2990  *
2991  * This map is used to temporarily establish LD_WAIT_SLEEP context for the
2992  * console write() callback when legacy printing to avoid false positive
2993  * lockdep complaints, thus allowing lockdep to continue to function for
2994  * real issues.
2995  */
2996 #ifdef CONFIG_PREEMPT_RT
printk_legacy_allow_spinlock_enter(void)2997 static inline void printk_legacy_allow_spinlock_enter(void) { }
printk_legacy_allow_spinlock_exit(void)2998 static inline void printk_legacy_allow_spinlock_exit(void) { }
2999 #else
3000 static DEFINE_WAIT_OVERRIDE_MAP(printk_legacy_map, LD_WAIT_SLEEP);
3001 
printk_legacy_allow_spinlock_enter(void)3002 static inline void printk_legacy_allow_spinlock_enter(void)
3003 {
3004 	lock_map_acquire_try(&printk_legacy_map);
3005 }
3006 
printk_legacy_allow_spinlock_exit(void)3007 static inline void printk_legacy_allow_spinlock_exit(void)
3008 {
3009 	lock_map_release(&printk_legacy_map);
3010 }
3011 #endif /* CONFIG_PREEMPT_RT */
3012 
3013 /*
3014  * Used as the printk buffers for non-panic, serialized console printing.
3015  * This is for legacy (!CON_NBCON) as well as all boot (CON_BOOT) consoles.
3016  * Its usage requires the console_lock held.
3017  */
3018 struct printk_buffers printk_shared_pbufs;
3019 
3020 /*
3021  * Print one record for the given console. The record printed is whatever
3022  * record is the next available record for the given console.
3023  *
3024  * @handover will be set to true if a printk waiter has taken over the
3025  * console_lock, in which case the caller is no longer holding both the
3026  * console_lock and the SRCU read lock. Otherwise it is set to false.
3027  *
3028  * @cookie is the cookie from the SRCU read lock.
3029  *
3030  * Returns false if the given console has no next record to print, otherwise
3031  * true.
3032  *
3033  * Requires the console_lock and the SRCU read lock.
3034  */
console_emit_next_record(struct console * con,bool * handover,int cookie)3035 static bool console_emit_next_record(struct console *con, bool *handover, int cookie)
3036 {
3037 	bool is_extended = console_srcu_read_flags(con) & CON_EXTENDED;
3038 	char *outbuf = &printk_shared_pbufs.outbuf[0];
3039 	struct printk_message pmsg = {
3040 		.pbufs = &printk_shared_pbufs,
3041 	};
3042 	unsigned long flags;
3043 
3044 	*handover = false;
3045 
3046 	if (!printk_get_next_message(&pmsg, con->seq, is_extended, true))
3047 		return false;
3048 
3049 	con->dropped += pmsg.dropped;
3050 
3051 	/* Skip messages of formatted length 0. */
3052 	if (pmsg.outbuf_len == 0) {
3053 		con->seq = pmsg.seq + 1;
3054 		goto skip;
3055 	}
3056 
3057 	if (con->dropped && !is_extended) {
3058 		console_prepend_dropped(&pmsg, con->dropped);
3059 		con->dropped = 0;
3060 	}
3061 
3062 	/* Write everything out to the hardware. */
3063 
3064 	if (force_legacy_kthread() && !panic_in_progress()) {
3065 		/*
3066 		 * With forced threading this function is in a task context
3067 		 * (either legacy kthread or get_init_console_seq()). There
3068 		 * is no need for concern about printk reentrance, handovers,
3069 		 * or lockdep complaints.
3070 		 */
3071 
3072 		con->write(con, outbuf, pmsg.outbuf_len);
3073 		con->seq = pmsg.seq + 1;
3074 	} else {
3075 		/*
3076 		 * While actively printing out messages, if another printk()
3077 		 * were to occur on another CPU, it may wait for this one to
3078 		 * finish. This task can not be preempted if there is a
3079 		 * waiter waiting to take over.
3080 		 *
3081 		 * Interrupts are disabled because the hand over to a waiter
3082 		 * must not be interrupted until the hand over is completed
3083 		 * (@console_waiter is cleared).
3084 		 */
3085 		printk_safe_enter_irqsave(flags);
3086 		console_lock_spinning_enable();
3087 
3088 		/* Do not trace print latency. */
3089 		stop_critical_timings();
3090 
3091 		printk_legacy_allow_spinlock_enter();
3092 		con->write(con, outbuf, pmsg.outbuf_len);
3093 		printk_legacy_allow_spinlock_exit();
3094 
3095 		start_critical_timings();
3096 
3097 		con->seq = pmsg.seq + 1;
3098 
3099 		*handover = console_lock_spinning_disable_and_check(cookie);
3100 		printk_safe_exit_irqrestore(flags);
3101 	}
3102 skip:
3103 	return true;
3104 }
3105 
3106 #else
3107 
console_emit_next_record(struct console * con,bool * handover,int cookie)3108 static bool console_emit_next_record(struct console *con, bool *handover, int cookie)
3109 {
3110 	*handover = false;
3111 	return false;
3112 }
3113 
printk_kthreads_check_locked(void)3114 static inline void printk_kthreads_check_locked(void) { }
3115 
3116 #endif /* CONFIG_PRINTK */
3117 
3118 /*
3119  * Print out all remaining records to all consoles.
3120  *
3121  * @do_cond_resched is set by the caller. It can be true only in schedulable
3122  * context.
3123  *
3124  * @next_seq is set to the sequence number after the last available record.
3125  * The value is valid only when this function returns true. It means that all
3126  * usable consoles are completely flushed.
3127  *
3128  * @handover will be set to true if a printk waiter has taken over the
3129  * console_lock, in which case the caller is no longer holding the
3130  * console_lock. Otherwise it is set to false.
3131  *
3132  * Returns true when there was at least one usable console and all messages
3133  * were flushed to all usable consoles. A returned false informs the caller
3134  * that everything was not flushed (either there were no usable consoles or
3135  * another context has taken over printing or it is a panic situation and this
3136  * is not the panic CPU). Regardless the reason, the caller should assume it
3137  * is not useful to immediately try again.
3138  *
3139  * Requires the console_lock.
3140  */
console_flush_all(bool do_cond_resched,u64 * next_seq,bool * handover)3141 static bool console_flush_all(bool do_cond_resched, u64 *next_seq, bool *handover)
3142 {
3143 	struct console_flush_type ft;
3144 	bool any_usable = false;
3145 	struct console *con;
3146 	bool any_progress;
3147 	int cookie;
3148 
3149 	*next_seq = 0;
3150 	*handover = false;
3151 
3152 	do {
3153 		any_progress = false;
3154 
3155 		printk_get_console_flush_type(&ft);
3156 
3157 		cookie = console_srcu_read_lock();
3158 		for_each_console_srcu(con) {
3159 			short flags = console_srcu_read_flags(con);
3160 			u64 printk_seq;
3161 			bool progress;
3162 
3163 			/*
3164 			 * console_flush_all() is only responsible for nbcon
3165 			 * consoles when the nbcon consoles cannot print via
3166 			 * their atomic or threaded flushing.
3167 			 */
3168 			if ((flags & CON_NBCON) && (ft.nbcon_atomic || ft.nbcon_offload))
3169 				continue;
3170 
3171 			if (!console_is_usable(con, flags, !do_cond_resched))
3172 				continue;
3173 			any_usable = true;
3174 
3175 			if (flags & CON_NBCON) {
3176 				progress = nbcon_legacy_emit_next_record(con, handover, cookie,
3177 									 !do_cond_resched);
3178 				printk_seq = nbcon_seq_read(con);
3179 			} else {
3180 				progress = console_emit_next_record(con, handover, cookie);
3181 				printk_seq = con->seq;
3182 			}
3183 
3184 			/*
3185 			 * If a handover has occurred, the SRCU read lock
3186 			 * is already released.
3187 			 */
3188 			if (*handover)
3189 				return false;
3190 
3191 			/* Track the next of the highest seq flushed. */
3192 			if (printk_seq > *next_seq)
3193 				*next_seq = printk_seq;
3194 
3195 			if (!progress)
3196 				continue;
3197 			any_progress = true;
3198 
3199 			/* Allow panic_cpu to take over the consoles safely. */
3200 			if (other_cpu_in_panic())
3201 				goto abandon;
3202 
3203 			if (do_cond_resched)
3204 				cond_resched();
3205 		}
3206 		console_srcu_read_unlock(cookie);
3207 	} while (any_progress);
3208 
3209 	return any_usable;
3210 
3211 abandon:
3212 	console_srcu_read_unlock(cookie);
3213 	return false;
3214 }
3215 
__console_flush_and_unlock(void)3216 static void __console_flush_and_unlock(void)
3217 {
3218 	bool do_cond_resched;
3219 	bool handover;
3220 	bool flushed;
3221 	u64 next_seq;
3222 
3223 	/*
3224 	 * Console drivers are called with interrupts disabled, so
3225 	 * @console_may_schedule should be cleared before; however, we may
3226 	 * end up dumping a lot of lines, for example, if called from
3227 	 * console registration path, and should invoke cond_resched()
3228 	 * between lines if allowable.  Not doing so can cause a very long
3229 	 * scheduling stall on a slow console leading to RCU stall and
3230 	 * softlockup warnings which exacerbate the issue with more
3231 	 * messages practically incapacitating the system. Therefore, create
3232 	 * a local to use for the printing loop.
3233 	 */
3234 	do_cond_resched = console_may_schedule;
3235 
3236 	do {
3237 		console_may_schedule = 0;
3238 
3239 		flushed = console_flush_all(do_cond_resched, &next_seq, &handover);
3240 		if (!handover)
3241 			__console_unlock();
3242 
3243 		/*
3244 		 * Abort if there was a failure to flush all messages to all
3245 		 * usable consoles. Either it is not possible to flush (in
3246 		 * which case it would be an infinite loop of retrying) or
3247 		 * another context has taken over printing.
3248 		 */
3249 		if (!flushed)
3250 			break;
3251 
3252 		/*
3253 		 * Some context may have added new records after
3254 		 * console_flush_all() but before unlocking the console.
3255 		 * Re-check if there is a new record to flush. If the trylock
3256 		 * fails, another context is already handling the printing.
3257 		 */
3258 	} while (prb_read_valid(prb, next_seq, NULL) && console_trylock());
3259 }
3260 
3261 /**
3262  * console_unlock - unblock the legacy console subsystem from printing
3263  *
3264  * Releases the console_lock which the caller holds to block printing of
3265  * the legacy console subsystem.
3266  *
3267  * While the console_lock was held, console output may have been buffered
3268  * by printk(). If this is the case, console_unlock() emits the output on
3269  * legacy consoles prior to releasing the lock.
3270  *
3271  * console_unlock(); may be called from any context.
3272  */
console_unlock(void)3273 void console_unlock(void)
3274 {
3275 	struct console_flush_type ft;
3276 
3277 	printk_get_console_flush_type(&ft);
3278 	if (ft.legacy_direct)
3279 		__console_flush_and_unlock();
3280 	else
3281 		__console_unlock();
3282 }
3283 EXPORT_SYMBOL(console_unlock);
3284 
3285 /**
3286  * console_conditional_schedule - yield the CPU if required
3287  *
3288  * If the console code is currently allowed to sleep, and
3289  * if this CPU should yield the CPU to another task, do
3290  * so here.
3291  *
3292  * Must be called within console_lock();.
3293  */
console_conditional_schedule(void)3294 void __sched console_conditional_schedule(void)
3295 {
3296 	if (console_may_schedule)
3297 		cond_resched();
3298 }
3299 EXPORT_SYMBOL(console_conditional_schedule);
3300 
console_unblank(void)3301 void console_unblank(void)
3302 {
3303 	bool found_unblank = false;
3304 	struct console *c;
3305 	int cookie;
3306 
3307 	/*
3308 	 * First check if there are any consoles implementing the unblank()
3309 	 * callback. If not, there is no reason to continue and take the
3310 	 * console lock, which in particular can be dangerous if
3311 	 * @oops_in_progress is set.
3312 	 */
3313 	cookie = console_srcu_read_lock();
3314 	for_each_console_srcu(c) {
3315 		if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank) {
3316 			found_unblank = true;
3317 			break;
3318 		}
3319 	}
3320 	console_srcu_read_unlock(cookie);
3321 	if (!found_unblank)
3322 		return;
3323 
3324 	/*
3325 	 * Stop console printing because the unblank() callback may
3326 	 * assume the console is not within its write() callback.
3327 	 *
3328 	 * If @oops_in_progress is set, this may be an atomic context.
3329 	 * In that case, attempt a trylock as best-effort.
3330 	 */
3331 	if (oops_in_progress) {
3332 		/* Semaphores are not NMI-safe. */
3333 		if (in_nmi())
3334 			return;
3335 
3336 		/*
3337 		 * Attempting to trylock the console lock can deadlock
3338 		 * if another CPU was stopped while modifying the
3339 		 * semaphore. "Hope and pray" that this is not the
3340 		 * current situation.
3341 		 */
3342 		if (down_trylock_console_sem() != 0)
3343 			return;
3344 	} else
3345 		console_lock();
3346 
3347 	console_locked = 1;
3348 	console_may_schedule = 0;
3349 
3350 	cookie = console_srcu_read_lock();
3351 	for_each_console_srcu(c) {
3352 		if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank)
3353 			c->unblank();
3354 	}
3355 	console_srcu_read_unlock(cookie);
3356 
3357 	console_unlock();
3358 
3359 	if (!oops_in_progress)
3360 		pr_flush(1000, true);
3361 }
3362 
3363 /*
3364  * Rewind all consoles to the oldest available record.
3365  *
3366  * IMPORTANT: The function is safe only when called under
3367  *            console_lock(). It is not enforced because
3368  *            it is used as a best effort in panic().
3369  */
__console_rewind_all(void)3370 static void __console_rewind_all(void)
3371 {
3372 	struct console *c;
3373 	short flags;
3374 	int cookie;
3375 	u64 seq;
3376 
3377 	seq = prb_first_valid_seq(prb);
3378 
3379 	cookie = console_srcu_read_lock();
3380 	for_each_console_srcu(c) {
3381 		flags = console_srcu_read_flags(c);
3382 
3383 		if (flags & CON_NBCON) {
3384 			nbcon_seq_force(c, seq);
3385 		} else {
3386 			/*
3387 			 * This assignment is safe only when called under
3388 			 * console_lock(). On panic, legacy consoles are
3389 			 * only best effort.
3390 			 */
3391 			c->seq = seq;
3392 		}
3393 	}
3394 	console_srcu_read_unlock(cookie);
3395 }
3396 
3397 /**
3398  * console_flush_on_panic - flush console content on panic
3399  * @mode: flush all messages in buffer or just the pending ones
3400  *
3401  * Immediately output all pending messages no matter what.
3402  */
console_flush_on_panic(enum con_flush_mode mode)3403 void console_flush_on_panic(enum con_flush_mode mode)
3404 {
3405 	struct console_flush_type ft;
3406 	bool handover;
3407 	u64 next_seq;
3408 
3409 	/*
3410 	 * Ignore the console lock and flush out the messages. Attempting a
3411 	 * trylock would not be useful because:
3412 	 *
3413 	 *   - if it is contended, it must be ignored anyway
3414 	 *   - console_lock() and console_trylock() block and fail
3415 	 *     respectively in panic for non-panic CPUs
3416 	 *   - semaphores are not NMI-safe
3417 	 */
3418 
3419 	/*
3420 	 * If another context is holding the console lock,
3421 	 * @console_may_schedule might be set. Clear it so that
3422 	 * this context does not call cond_resched() while flushing.
3423 	 */
3424 	console_may_schedule = 0;
3425 
3426 	if (mode == CONSOLE_REPLAY_ALL)
3427 		__console_rewind_all();
3428 
3429 	printk_get_console_flush_type(&ft);
3430 	if (ft.nbcon_atomic)
3431 		nbcon_atomic_flush_pending();
3432 
3433 	/* Flush legacy consoles once allowed, even when dangerous. */
3434 	if (legacy_allow_panic_sync)
3435 		console_flush_all(false, &next_seq, &handover);
3436 }
3437 
3438 /*
3439  * Return the console tty driver structure and its associated index
3440  */
console_device(int * index)3441 struct tty_driver *console_device(int *index)
3442 {
3443 	struct console *c;
3444 	struct tty_driver *driver = NULL;
3445 	int cookie;
3446 
3447 	/*
3448 	 * Take console_lock to serialize device() callback with
3449 	 * other console operations. For example, fg_console is
3450 	 * modified under console_lock when switching vt.
3451 	 */
3452 	console_lock();
3453 
3454 	cookie = console_srcu_read_lock();
3455 	for_each_console_srcu(c) {
3456 		if (!c->device)
3457 			continue;
3458 		driver = c->device(c, index);
3459 		if (driver)
3460 			break;
3461 	}
3462 	console_srcu_read_unlock(cookie);
3463 
3464 	console_unlock();
3465 	return driver;
3466 }
3467 
3468 /*
3469  * Prevent further output on the passed console device so that (for example)
3470  * serial drivers can disable console output before suspending a port, and can
3471  * re-enable output afterwards.
3472  */
console_stop(struct console * console)3473 void console_stop(struct console *console)
3474 {
3475 	__pr_flush(console, 1000, true);
3476 	console_list_lock();
3477 	console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3478 	console_list_unlock();
3479 
3480 	/*
3481 	 * Ensure that all SRCU list walks have completed. All contexts must
3482 	 * be able to see that this console is disabled so that (for example)
3483 	 * the caller can suspend the port without risk of another context
3484 	 * using the port.
3485 	 */
3486 	synchronize_srcu(&console_srcu);
3487 }
3488 EXPORT_SYMBOL(console_stop);
3489 
console_start(struct console * console)3490 void console_start(struct console *console)
3491 {
3492 	struct console_flush_type ft;
3493 	bool is_nbcon;
3494 
3495 	console_list_lock();
3496 	console_srcu_write_flags(console, console->flags | CON_ENABLED);
3497 	is_nbcon = console->flags & CON_NBCON;
3498 	console_list_unlock();
3499 
3500 	/*
3501 	 * Ensure that all SRCU list walks have completed. The related
3502 	 * printing context must be able to see it is enabled so that
3503 	 * it is guaranteed to wake up and resume printing.
3504 	 */
3505 	synchronize_srcu(&console_srcu);
3506 
3507 	printk_get_console_flush_type(&ft);
3508 	if (is_nbcon && ft.nbcon_offload)
3509 		nbcon_kthread_wake(console);
3510 	else if (ft.legacy_offload)
3511 		defer_console_output();
3512 
3513 	__pr_flush(console, 1000, true);
3514 }
3515 EXPORT_SYMBOL(console_start);
3516 
3517 #ifdef CONFIG_PRINTK
3518 static int unregister_console_locked(struct console *console);
3519 
3520 /* True when system boot is far enough to create printer threads. */
3521 static bool printk_kthreads_ready __ro_after_init;
3522 
3523 static struct task_struct *printk_legacy_kthread;
3524 
legacy_kthread_should_wakeup(void)3525 static bool legacy_kthread_should_wakeup(void)
3526 {
3527 	struct console_flush_type ft;
3528 	struct console *con;
3529 	bool ret = false;
3530 	int cookie;
3531 
3532 	if (kthread_should_stop())
3533 		return true;
3534 
3535 	printk_get_console_flush_type(&ft);
3536 
3537 	cookie = console_srcu_read_lock();
3538 	for_each_console_srcu(con) {
3539 		short flags = console_srcu_read_flags(con);
3540 		u64 printk_seq;
3541 
3542 		/*
3543 		 * The legacy printer thread is only responsible for nbcon
3544 		 * consoles when the nbcon consoles cannot print via their
3545 		 * atomic or threaded flushing.
3546 		 */
3547 		if ((flags & CON_NBCON) && (ft.nbcon_atomic || ft.nbcon_offload))
3548 			continue;
3549 
3550 		if (!console_is_usable(con, flags, false))
3551 			continue;
3552 
3553 		if (flags & CON_NBCON) {
3554 			printk_seq = nbcon_seq_read(con);
3555 		} else {
3556 			/*
3557 			 * It is safe to read @seq because only this
3558 			 * thread context updates @seq.
3559 			 */
3560 			printk_seq = con->seq;
3561 		}
3562 
3563 		if (prb_read_valid(prb, printk_seq, NULL)) {
3564 			ret = true;
3565 			break;
3566 		}
3567 	}
3568 	console_srcu_read_unlock(cookie);
3569 
3570 	return ret;
3571 }
3572 
legacy_kthread_func(void * unused)3573 static int legacy_kthread_func(void *unused)
3574 {
3575 	for (;;) {
3576 		wait_event_interruptible(legacy_wait, legacy_kthread_should_wakeup());
3577 
3578 		if (kthread_should_stop())
3579 			break;
3580 
3581 		console_lock();
3582 		__console_flush_and_unlock();
3583 	}
3584 
3585 	return 0;
3586 }
3587 
legacy_kthread_create(void)3588 static bool legacy_kthread_create(void)
3589 {
3590 	struct task_struct *kt;
3591 
3592 	lockdep_assert_console_list_lock_held();
3593 
3594 	kt = kthread_run(legacy_kthread_func, NULL, "pr/legacy");
3595 	if (WARN_ON(IS_ERR(kt))) {
3596 		pr_err("failed to start legacy printing thread\n");
3597 		return false;
3598 	}
3599 
3600 	printk_legacy_kthread = kt;
3601 
3602 	/*
3603 	 * It is important that console printing threads are scheduled
3604 	 * shortly after a printk call and with generous runtime budgets.
3605 	 */
3606 	sched_set_normal(printk_legacy_kthread, -20);
3607 
3608 	return true;
3609 }
3610 
3611 /**
3612  * printk_kthreads_shutdown - shutdown all threaded printers
3613  *
3614  * On system shutdown all threaded printers are stopped. This allows printk
3615  * to transition back to atomic printing, thus providing a robust mechanism
3616  * for the final shutdown/reboot messages to be output.
3617  */
printk_kthreads_shutdown(void)3618 static void printk_kthreads_shutdown(void)
3619 {
3620 	struct console *con;
3621 
3622 	console_list_lock();
3623 	if (printk_kthreads_running) {
3624 		printk_kthreads_running = false;
3625 
3626 		for_each_console(con) {
3627 			if (con->flags & CON_NBCON)
3628 				nbcon_kthread_stop(con);
3629 		}
3630 
3631 		/*
3632 		 * The threads may have been stopped while printing a
3633 		 * backlog. Flush any records left over.
3634 		 */
3635 		nbcon_atomic_flush_pending();
3636 	}
3637 	console_list_unlock();
3638 }
3639 
3640 static struct syscore_ops printk_syscore_ops = {
3641 	.shutdown = printk_kthreads_shutdown,
3642 };
3643 
3644 /*
3645  * If appropriate, start nbcon kthreads and set @printk_kthreads_running.
3646  * If any kthreads fail to start, those consoles are unregistered.
3647  *
3648  * Must be called under console_list_lock().
3649  */
printk_kthreads_check_locked(void)3650 static void printk_kthreads_check_locked(void)
3651 {
3652 	struct hlist_node *tmp;
3653 	struct console *con;
3654 
3655 	lockdep_assert_console_list_lock_held();
3656 
3657 	if (!printk_kthreads_ready)
3658 		return;
3659 
3660 	if (have_legacy_console || have_boot_console) {
3661 		if (!printk_legacy_kthread &&
3662 		    force_legacy_kthread() &&
3663 		    !legacy_kthread_create()) {
3664 			/*
3665 			 * All legacy consoles must be unregistered. If there
3666 			 * are any nbcon consoles, they will set up their own
3667 			 * kthread.
3668 			 */
3669 			hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3670 				if (con->flags & CON_NBCON)
3671 					continue;
3672 
3673 				unregister_console_locked(con);
3674 			}
3675 		}
3676 	} else if (printk_legacy_kthread) {
3677 		kthread_stop(printk_legacy_kthread);
3678 		printk_legacy_kthread = NULL;
3679 	}
3680 
3681 	/*
3682 	 * Printer threads cannot be started as long as any boot console is
3683 	 * registered because there is no way to synchronize the hardware
3684 	 * registers between boot console code and regular console code.
3685 	 * It can only be known that there will be no new boot consoles when
3686 	 * an nbcon console is registered.
3687 	 */
3688 	if (have_boot_console || !have_nbcon_console) {
3689 		/* Clear flag in case all nbcon consoles unregistered. */
3690 		printk_kthreads_running = false;
3691 		return;
3692 	}
3693 
3694 	if (printk_kthreads_running)
3695 		return;
3696 
3697 	hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3698 		if (!(con->flags & CON_NBCON))
3699 			continue;
3700 
3701 		if (!nbcon_kthread_create(con))
3702 			unregister_console_locked(con);
3703 	}
3704 
3705 	printk_kthreads_running = true;
3706 }
3707 
printk_set_kthreads_ready(void)3708 static int __init printk_set_kthreads_ready(void)
3709 {
3710 	register_syscore_ops(&printk_syscore_ops);
3711 
3712 	console_list_lock();
3713 	printk_kthreads_ready = true;
3714 	printk_kthreads_check_locked();
3715 	console_list_unlock();
3716 
3717 	return 0;
3718 }
3719 early_initcall(printk_set_kthreads_ready);
3720 #endif /* CONFIG_PRINTK */
3721 
3722 static int __read_mostly keep_bootcon;
3723 
keep_bootcon_setup(char * str)3724 static int __init keep_bootcon_setup(char *str)
3725 {
3726 	keep_bootcon = 1;
3727 	pr_info("debug: skip boot console de-registration.\n");
3728 
3729 	return 0;
3730 }
3731 
3732 early_param("keep_bootcon", keep_bootcon_setup);
3733 
console_call_setup(struct console * newcon,char * options)3734 static int console_call_setup(struct console *newcon, char *options)
3735 {
3736 	int err;
3737 
3738 	if (!newcon->setup)
3739 		return 0;
3740 
3741 	/* Synchronize with possible boot console. */
3742 	console_lock();
3743 	err = newcon->setup(newcon, options);
3744 	console_unlock();
3745 
3746 	return err;
3747 }
3748 
3749 /*
3750  * This is called by register_console() to try to match
3751  * the newly registered console with any of the ones selected
3752  * by either the command line or add_preferred_console() and
3753  * setup/enable it.
3754  *
3755  * Care need to be taken with consoles that are statically
3756  * enabled such as netconsole
3757  */
try_enable_preferred_console(struct console * newcon,bool user_specified)3758 static int try_enable_preferred_console(struct console *newcon,
3759 					bool user_specified)
3760 {
3761 	struct console_cmdline *c;
3762 	int i, err;
3763 
3764 	for (i = 0, c = console_cmdline;
3765 	     i < MAX_CMDLINECONSOLES && (c->name[0] || c->devname[0]);
3766 	     i++, c++) {
3767 		/* Console not yet initialized? */
3768 		if (!c->name[0])
3769 			continue;
3770 		if (c->user_specified != user_specified)
3771 			continue;
3772 		if (!newcon->match ||
3773 		    newcon->match(newcon, c->name, c->index, c->options) != 0) {
3774 			/* default matching */
3775 			BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
3776 			if (strcmp(c->name, newcon->name) != 0)
3777 				continue;
3778 			if (newcon->index >= 0 &&
3779 			    newcon->index != c->index)
3780 				continue;
3781 			if (newcon->index < 0)
3782 				newcon->index = c->index;
3783 
3784 			if (_braille_register_console(newcon, c))
3785 				return 0;
3786 
3787 			err = console_call_setup(newcon, c->options);
3788 			if (err)
3789 				return err;
3790 		}
3791 		newcon->flags |= CON_ENABLED;
3792 		if (i == preferred_console)
3793 			newcon->flags |= CON_CONSDEV;
3794 		return 0;
3795 	}
3796 
3797 	/*
3798 	 * Some consoles, such as pstore and netconsole, can be enabled even
3799 	 * without matching. Accept the pre-enabled consoles only when match()
3800 	 * and setup() had a chance to be called.
3801 	 */
3802 	if (newcon->flags & CON_ENABLED && c->user_specified ==	user_specified)
3803 		return 0;
3804 
3805 	return -ENOENT;
3806 }
3807 
3808 /* Try to enable the console unconditionally */
try_enable_default_console(struct console * newcon)3809 static void try_enable_default_console(struct console *newcon)
3810 {
3811 	if (newcon->index < 0)
3812 		newcon->index = 0;
3813 
3814 	if (console_call_setup(newcon, NULL) != 0)
3815 		return;
3816 
3817 	newcon->flags |= CON_ENABLED;
3818 
3819 	if (newcon->device)
3820 		newcon->flags |= CON_CONSDEV;
3821 }
3822 
3823 /* Return the starting sequence number for a newly registered console. */
get_init_console_seq(struct console * newcon,bool bootcon_registered)3824 static u64 get_init_console_seq(struct console *newcon, bool bootcon_registered)
3825 {
3826 	struct console *con;
3827 	bool handover;
3828 	u64 init_seq;
3829 
3830 	if (newcon->flags & (CON_PRINTBUFFER | CON_BOOT)) {
3831 		/* Get a consistent copy of @syslog_seq. */
3832 		mutex_lock(&syslog_lock);
3833 		init_seq = syslog_seq;
3834 		mutex_unlock(&syslog_lock);
3835 	} else {
3836 		/* Begin with next message added to ringbuffer. */
3837 		init_seq = prb_next_seq(prb);
3838 
3839 		/*
3840 		 * If any enabled boot consoles are due to be unregistered
3841 		 * shortly, some may not be caught up and may be the same
3842 		 * device as @newcon. Since it is not known which boot console
3843 		 * is the same device, flush all consoles and, if necessary,
3844 		 * start with the message of the enabled boot console that is
3845 		 * the furthest behind.
3846 		 */
3847 		if (bootcon_registered && !keep_bootcon) {
3848 			/*
3849 			 * Hold the console_lock to stop console printing and
3850 			 * guarantee safe access to console->seq.
3851 			 */
3852 			console_lock();
3853 
3854 			/*
3855 			 * Flush all consoles and set the console to start at
3856 			 * the next unprinted sequence number.
3857 			 */
3858 			if (!console_flush_all(true, &init_seq, &handover)) {
3859 				/*
3860 				 * Flushing failed. Just choose the lowest
3861 				 * sequence of the enabled boot consoles.
3862 				 */
3863 
3864 				/*
3865 				 * If there was a handover, this context no
3866 				 * longer holds the console_lock.
3867 				 */
3868 				if (handover)
3869 					console_lock();
3870 
3871 				init_seq = prb_next_seq(prb);
3872 				for_each_console(con) {
3873 					u64 seq;
3874 
3875 					if (!(con->flags & CON_BOOT) ||
3876 					    !(con->flags & CON_ENABLED)) {
3877 						continue;
3878 					}
3879 
3880 					if (con->flags & CON_NBCON)
3881 						seq = nbcon_seq_read(con);
3882 					else
3883 						seq = con->seq;
3884 
3885 					if (seq < init_seq)
3886 						init_seq = seq;
3887 				}
3888 			}
3889 
3890 			console_unlock();
3891 		}
3892 	}
3893 
3894 	return init_seq;
3895 }
3896 
3897 #define console_first()				\
3898 	hlist_entry(console_list.first, struct console, node)
3899 
3900 static int unregister_console_locked(struct console *console);
3901 
3902 /*
3903  * The console driver calls this routine during kernel initialization
3904  * to register the console printing procedure with printk() and to
3905  * print any messages that were printed by the kernel before the
3906  * console driver was initialized.
3907  *
3908  * This can happen pretty early during the boot process (because of
3909  * early_printk) - sometimes before setup_arch() completes - be careful
3910  * of what kernel features are used - they may not be initialised yet.
3911  *
3912  * There are two types of consoles - bootconsoles (early_printk) and
3913  * "real" consoles (everything which is not a bootconsole) which are
3914  * handled differently.
3915  *  - Any number of bootconsoles can be registered at any time.
3916  *  - As soon as a "real" console is registered, all bootconsoles
3917  *    will be unregistered automatically.
3918  *  - Once a "real" console is registered, any attempt to register a
3919  *    bootconsoles will be rejected
3920  */
register_console(struct console * newcon)3921 void register_console(struct console *newcon)
3922 {
3923 	bool use_device_lock = (newcon->flags & CON_NBCON) && newcon->write_atomic;
3924 	bool bootcon_registered = false;
3925 	bool realcon_registered = false;
3926 	struct console *con;
3927 	unsigned long flags;
3928 	u64 init_seq;
3929 	int err;
3930 
3931 	console_list_lock();
3932 
3933 	for_each_console(con) {
3934 		if (WARN(con == newcon, "console '%s%d' already registered\n",
3935 					 con->name, con->index)) {
3936 			goto unlock;
3937 		}
3938 
3939 		if (con->flags & CON_BOOT)
3940 			bootcon_registered = true;
3941 		else
3942 			realcon_registered = true;
3943 	}
3944 
3945 	/* Do not register boot consoles when there already is a real one. */
3946 	if ((newcon->flags & CON_BOOT) && realcon_registered) {
3947 		pr_info("Too late to register bootconsole %s%d\n",
3948 			newcon->name, newcon->index);
3949 		goto unlock;
3950 	}
3951 
3952 	if (newcon->flags & CON_NBCON) {
3953 		/*
3954 		 * Ensure the nbcon console buffers can be allocated
3955 		 * before modifying any global data.
3956 		 */
3957 		if (!nbcon_alloc(newcon))
3958 			goto unlock;
3959 	}
3960 
3961 	/*
3962 	 * See if we want to enable this console driver by default.
3963 	 *
3964 	 * Nope when a console is preferred by the command line, device
3965 	 * tree, or SPCR.
3966 	 *
3967 	 * The first real console with tty binding (driver) wins. More
3968 	 * consoles might get enabled before the right one is found.
3969 	 *
3970 	 * Note that a console with tty binding will have CON_CONSDEV
3971 	 * flag set and will be first in the list.
3972 	 */
3973 	if (preferred_console < 0) {
3974 		if (hlist_empty(&console_list) || !console_first()->device ||
3975 		    console_first()->flags & CON_BOOT) {
3976 			try_enable_default_console(newcon);
3977 		}
3978 	}
3979 
3980 	/* See if this console matches one we selected on the command line */
3981 	err = try_enable_preferred_console(newcon, true);
3982 
3983 	/* If not, try to match against the platform default(s) */
3984 	if (err == -ENOENT)
3985 		err = try_enable_preferred_console(newcon, false);
3986 
3987 	/* printk() messages are not printed to the Braille console. */
3988 	if (err || newcon->flags & CON_BRL) {
3989 		if (newcon->flags & CON_NBCON)
3990 			nbcon_free(newcon);
3991 		goto unlock;
3992 	}
3993 
3994 	/*
3995 	 * If we have a bootconsole, and are switching to a real console,
3996 	 * don't print everything out again, since when the boot console, and
3997 	 * the real console are the same physical device, it's annoying to
3998 	 * see the beginning boot messages twice
3999 	 */
4000 	if (bootcon_registered &&
4001 	    ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
4002 		newcon->flags &= ~CON_PRINTBUFFER;
4003 	}
4004 
4005 	newcon->dropped = 0;
4006 	init_seq = get_init_console_seq(newcon, bootcon_registered);
4007 
4008 	if (newcon->flags & CON_NBCON) {
4009 		have_nbcon_console = true;
4010 		nbcon_seq_force(newcon, init_seq);
4011 	} else {
4012 		have_legacy_console = true;
4013 		newcon->seq = init_seq;
4014 	}
4015 
4016 	if (newcon->flags & CON_BOOT)
4017 		have_boot_console = true;
4018 
4019 	/*
4020 	 * If another context is actively using the hardware of this new
4021 	 * console, it will not be aware of the nbcon synchronization. This
4022 	 * is a risk that two contexts could access the hardware
4023 	 * simultaneously if this new console is used for atomic printing
4024 	 * and the other context is still using the hardware.
4025 	 *
4026 	 * Use the driver synchronization to ensure that the hardware is not
4027 	 * in use while this new console transitions to being registered.
4028 	 */
4029 	if (use_device_lock)
4030 		newcon->device_lock(newcon, &flags);
4031 
4032 	/*
4033 	 * Put this console in the list - keep the
4034 	 * preferred driver at the head of the list.
4035 	 */
4036 	if (hlist_empty(&console_list)) {
4037 		/* Ensure CON_CONSDEV is always set for the head. */
4038 		newcon->flags |= CON_CONSDEV;
4039 		hlist_add_head_rcu(&newcon->node, &console_list);
4040 
4041 	} else if (newcon->flags & CON_CONSDEV) {
4042 		/* Only the new head can have CON_CONSDEV set. */
4043 		console_srcu_write_flags(console_first(), console_first()->flags & ~CON_CONSDEV);
4044 		hlist_add_head_rcu(&newcon->node, &console_list);
4045 
4046 	} else {
4047 		hlist_add_behind_rcu(&newcon->node, console_list.first);
4048 	}
4049 
4050 	/*
4051 	 * No need to synchronize SRCU here! The caller does not rely
4052 	 * on all contexts being able to see the new console before
4053 	 * register_console() completes.
4054 	 */
4055 
4056 	/* This new console is now registered. */
4057 	if (use_device_lock)
4058 		newcon->device_unlock(newcon, flags);
4059 
4060 	console_sysfs_notify();
4061 
4062 	/*
4063 	 * By unregistering the bootconsoles after we enable the real console
4064 	 * we get the "console xxx enabled" message on all the consoles -
4065 	 * boot consoles, real consoles, etc - this is to ensure that end
4066 	 * users know there might be something in the kernel's log buffer that
4067 	 * went to the bootconsole (that they do not see on the real console)
4068 	 */
4069 	con_printk(KERN_INFO, newcon, "enabled\n");
4070 	if (bootcon_registered &&
4071 	    ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
4072 	    !keep_bootcon) {
4073 		struct hlist_node *tmp;
4074 
4075 		hlist_for_each_entry_safe(con, tmp, &console_list, node) {
4076 			if (con->flags & CON_BOOT)
4077 				unregister_console_locked(con);
4078 		}
4079 	}
4080 
4081 	/* Changed console list, may require printer threads to start/stop. */
4082 	printk_kthreads_check_locked();
4083 unlock:
4084 	console_list_unlock();
4085 }
4086 EXPORT_SYMBOL(register_console);
4087 
4088 /* Must be called under console_list_lock(). */
unregister_console_locked(struct console * console)4089 static int unregister_console_locked(struct console *console)
4090 {
4091 	bool use_device_lock = (console->flags & CON_NBCON) && console->write_atomic;
4092 	bool found_legacy_con = false;
4093 	bool found_nbcon_con = false;
4094 	bool found_boot_con = false;
4095 	unsigned long flags;
4096 	struct console *c;
4097 	int res;
4098 
4099 	lockdep_assert_console_list_lock_held();
4100 
4101 	con_printk(KERN_INFO, console, "disabled\n");
4102 
4103 	res = _braille_unregister_console(console);
4104 	if (res < 0)
4105 		return res;
4106 	if (res > 0)
4107 		return 0;
4108 
4109 	if (!console_is_registered_locked(console))
4110 		res = -ENODEV;
4111 	else if (console_is_usable(console, console->flags, true))
4112 		__pr_flush(console, 1000, true);
4113 
4114 	/* Disable it unconditionally */
4115 	console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
4116 
4117 	if (res < 0)
4118 		return res;
4119 
4120 	/*
4121 	 * Use the driver synchronization to ensure that the hardware is not
4122 	 * in use while this console transitions to being unregistered.
4123 	 */
4124 	if (use_device_lock)
4125 		console->device_lock(console, &flags);
4126 
4127 	hlist_del_init_rcu(&console->node);
4128 
4129 	if (use_device_lock)
4130 		console->device_unlock(console, flags);
4131 
4132 	/*
4133 	 * <HISTORICAL>
4134 	 * If this isn't the last console and it has CON_CONSDEV set, we
4135 	 * need to set it on the next preferred console.
4136 	 * </HISTORICAL>
4137 	 *
4138 	 * The above makes no sense as there is no guarantee that the next
4139 	 * console has any device attached. Oh well....
4140 	 */
4141 	if (!hlist_empty(&console_list) && console->flags & CON_CONSDEV)
4142 		console_srcu_write_flags(console_first(), console_first()->flags | CON_CONSDEV);
4143 
4144 	/*
4145 	 * Ensure that all SRCU list walks have completed. All contexts
4146 	 * must not be able to see this console in the list so that any
4147 	 * exit/cleanup routines can be performed safely.
4148 	 */
4149 	synchronize_srcu(&console_srcu);
4150 
4151 	if (console->flags & CON_NBCON)
4152 		nbcon_free(console);
4153 
4154 	console_sysfs_notify();
4155 
4156 	if (console->exit)
4157 		res = console->exit(console);
4158 
4159 	/*
4160 	 * With this console gone, the global flags tracking registered
4161 	 * console types may have changed. Update them.
4162 	 */
4163 	for_each_console(c) {
4164 		if (c->flags & CON_BOOT)
4165 			found_boot_con = true;
4166 
4167 		if (c->flags & CON_NBCON)
4168 			found_nbcon_con = true;
4169 		else
4170 			found_legacy_con = true;
4171 	}
4172 	if (!found_boot_con)
4173 		have_boot_console = found_boot_con;
4174 	if (!found_legacy_con)
4175 		have_legacy_console = found_legacy_con;
4176 	if (!found_nbcon_con)
4177 		have_nbcon_console = found_nbcon_con;
4178 
4179 	/* Changed console list, may require printer threads to start/stop. */
4180 	printk_kthreads_check_locked();
4181 
4182 	return res;
4183 }
4184 
unregister_console(struct console * console)4185 int unregister_console(struct console *console)
4186 {
4187 	int res;
4188 
4189 	console_list_lock();
4190 	res = unregister_console_locked(console);
4191 	console_list_unlock();
4192 	return res;
4193 }
4194 EXPORT_SYMBOL(unregister_console);
4195 
4196 /**
4197  * console_force_preferred_locked - force a registered console preferred
4198  * @con: The registered console to force preferred.
4199  *
4200  * Must be called under console_list_lock().
4201  */
console_force_preferred_locked(struct console * con)4202 void console_force_preferred_locked(struct console *con)
4203 {
4204 	struct console *cur_pref_con;
4205 
4206 	if (!console_is_registered_locked(con))
4207 		return;
4208 
4209 	cur_pref_con = console_first();
4210 
4211 	/* Already preferred? */
4212 	if (cur_pref_con == con)
4213 		return;
4214 
4215 	/*
4216 	 * Delete, but do not re-initialize the entry. This allows the console
4217 	 * to continue to appear registered (via any hlist_unhashed_lockless()
4218 	 * checks), even though it was briefly removed from the console list.
4219 	 */
4220 	hlist_del_rcu(&con->node);
4221 
4222 	/*
4223 	 * Ensure that all SRCU list walks have completed so that the console
4224 	 * can be added to the beginning of the console list and its forward
4225 	 * list pointer can be re-initialized.
4226 	 */
4227 	synchronize_srcu(&console_srcu);
4228 
4229 	con->flags |= CON_CONSDEV;
4230 	WARN_ON(!con->device);
4231 
4232 	/* Only the new head can have CON_CONSDEV set. */
4233 	console_srcu_write_flags(cur_pref_con, cur_pref_con->flags & ~CON_CONSDEV);
4234 	hlist_add_head_rcu(&con->node, &console_list);
4235 }
4236 EXPORT_SYMBOL(console_force_preferred_locked);
4237 
4238 /*
4239  * Initialize the console device. This is called *early*, so
4240  * we can't necessarily depend on lots of kernel help here.
4241  * Just do some early initializations, and do the complex setup
4242  * later.
4243  */
console_init(void)4244 void __init console_init(void)
4245 {
4246 	int ret;
4247 	initcall_t call;
4248 	initcall_entry_t *ce;
4249 
4250 	/* Setup the default TTY line discipline. */
4251 	n_tty_init();
4252 
4253 	/*
4254 	 * set up the console device so that later boot sequences can
4255 	 * inform about problems etc..
4256 	 */
4257 	ce = __con_initcall_start;
4258 	trace_initcall_level("console");
4259 	while (ce < __con_initcall_end) {
4260 		call = initcall_from_entry(ce);
4261 		trace_initcall_start(call);
4262 		ret = call();
4263 		trace_initcall_finish(call, ret);
4264 		ce++;
4265 	}
4266 }
4267 
4268 /*
4269  * Some boot consoles access data that is in the init section and which will
4270  * be discarded after the initcalls have been run. To make sure that no code
4271  * will access this data, unregister the boot consoles in a late initcall.
4272  *
4273  * If for some reason, such as deferred probe or the driver being a loadable
4274  * module, the real console hasn't registered yet at this point, there will
4275  * be a brief interval in which no messages are logged to the console, which
4276  * makes it difficult to diagnose problems that occur during this time.
4277  *
4278  * To mitigate this problem somewhat, only unregister consoles whose memory
4279  * intersects with the init section. Note that all other boot consoles will
4280  * get unregistered when the real preferred console is registered.
4281  */
printk_late_init(void)4282 static int __init printk_late_init(void)
4283 {
4284 	struct hlist_node *tmp;
4285 	struct console *con;
4286 	int ret;
4287 
4288 	console_list_lock();
4289 	hlist_for_each_entry_safe(con, tmp, &console_list, node) {
4290 		if (!(con->flags & CON_BOOT))
4291 			continue;
4292 
4293 		/* Check addresses that might be used for enabled consoles. */
4294 		if (init_section_intersects(con, sizeof(*con)) ||
4295 		    init_section_contains(con->write, 0) ||
4296 		    init_section_contains(con->read, 0) ||
4297 		    init_section_contains(con->device, 0) ||
4298 		    init_section_contains(con->unblank, 0) ||
4299 		    init_section_contains(con->data, 0)) {
4300 			/*
4301 			 * Please, consider moving the reported consoles out
4302 			 * of the init section.
4303 			 */
4304 			pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
4305 				con->name, con->index);
4306 			unregister_console_locked(con);
4307 		}
4308 	}
4309 	console_list_unlock();
4310 
4311 	ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
4312 					console_cpu_notify);
4313 	WARN_ON(ret < 0);
4314 	ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
4315 					console_cpu_notify, NULL);
4316 	WARN_ON(ret < 0);
4317 	printk_sysctl_init();
4318 	return 0;
4319 }
4320 late_initcall(printk_late_init);
4321 
4322 #if defined CONFIG_PRINTK
4323 /* 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)4324 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
4325 {
4326 	unsigned long timeout_jiffies = msecs_to_jiffies(timeout_ms);
4327 	unsigned long remaining_jiffies = timeout_jiffies;
4328 	struct console_flush_type ft;
4329 	struct console *c;
4330 	u64 last_diff = 0;
4331 	u64 printk_seq;
4332 	short flags;
4333 	int cookie;
4334 	u64 diff;
4335 	u64 seq;
4336 
4337 	/* Sorry, pr_flush() will not work this early. */
4338 	if (system_state < SYSTEM_SCHEDULING)
4339 		return false;
4340 
4341 	might_sleep();
4342 
4343 	seq = prb_next_reserve_seq(prb);
4344 
4345 	/* Flush the consoles so that records up to @seq are printed. */
4346 	printk_get_console_flush_type(&ft);
4347 	if (ft.nbcon_atomic)
4348 		nbcon_atomic_flush_pending();
4349 	if (ft.legacy_direct) {
4350 		console_lock();
4351 		console_unlock();
4352 	}
4353 
4354 	for (;;) {
4355 		unsigned long begin_jiffies;
4356 		unsigned long slept_jiffies;
4357 
4358 		diff = 0;
4359 
4360 		/*
4361 		 * Hold the console_lock to guarantee safe access to
4362 		 * console->seq. Releasing console_lock flushes more
4363 		 * records in case @seq is still not printed on all
4364 		 * usable consoles.
4365 		 *
4366 		 * Holding the console_lock is not necessary if there
4367 		 * are no legacy or boot consoles. However, such a
4368 		 * console could register at any time. Always hold the
4369 		 * console_lock as a precaution rather than
4370 		 * synchronizing against register_console().
4371 		 */
4372 		console_lock();
4373 
4374 		cookie = console_srcu_read_lock();
4375 		for_each_console_srcu(c) {
4376 			if (con && con != c)
4377 				continue;
4378 
4379 			flags = console_srcu_read_flags(c);
4380 
4381 			/*
4382 			 * If consoles are not usable, it cannot be expected
4383 			 * that they make forward progress, so only increment
4384 			 * @diff for usable consoles.
4385 			 */
4386 			if (!console_is_usable(c, flags, true) &&
4387 			    !console_is_usable(c, flags, false)) {
4388 				continue;
4389 			}
4390 
4391 			if (flags & CON_NBCON) {
4392 				printk_seq = nbcon_seq_read(c);
4393 			} else {
4394 				printk_seq = c->seq;
4395 			}
4396 
4397 			if (printk_seq < seq)
4398 				diff += seq - printk_seq;
4399 		}
4400 		console_srcu_read_unlock(cookie);
4401 
4402 		if (diff != last_diff && reset_on_progress)
4403 			remaining_jiffies = timeout_jiffies;
4404 
4405 		console_unlock();
4406 
4407 		/* Note: @diff is 0 if there are no usable consoles. */
4408 		if (diff == 0 || remaining_jiffies == 0)
4409 			break;
4410 
4411 		/* msleep(1) might sleep much longer. Check time by jiffies. */
4412 		begin_jiffies = jiffies;
4413 		msleep(1);
4414 		slept_jiffies = jiffies - begin_jiffies;
4415 
4416 		remaining_jiffies -= min(slept_jiffies, remaining_jiffies);
4417 
4418 		last_diff = diff;
4419 	}
4420 
4421 	return (diff == 0);
4422 }
4423 
4424 /**
4425  * pr_flush() - Wait for printing threads to catch up.
4426  *
4427  * @timeout_ms:        The maximum time (in ms) to wait.
4428  * @reset_on_progress: Reset the timeout if forward progress is seen.
4429  *
4430  * A value of 0 for @timeout_ms means no waiting will occur. A value of -1
4431  * represents infinite waiting.
4432  *
4433  * If @reset_on_progress is true, the timeout will be reset whenever any
4434  * printer has been seen to make some forward progress.
4435  *
4436  * Context: Process context. May sleep while acquiring console lock.
4437  * Return: true if all usable printers are caught up.
4438  */
pr_flush(int timeout_ms,bool reset_on_progress)4439 static bool pr_flush(int timeout_ms, bool reset_on_progress)
4440 {
4441 	return __pr_flush(NULL, timeout_ms, reset_on_progress);
4442 }
4443 
4444 /*
4445  * Delayed printk version, for scheduler-internal messages:
4446  */
4447 #define PRINTK_PENDING_WAKEUP	0x01
4448 #define PRINTK_PENDING_OUTPUT	0x02
4449 
4450 static DEFINE_PER_CPU(int, printk_pending);
4451 
wake_up_klogd_work_func(struct irq_work * irq_work)4452 static void wake_up_klogd_work_func(struct irq_work *irq_work)
4453 {
4454 	int pending = this_cpu_xchg(printk_pending, 0);
4455 
4456 	if (pending & PRINTK_PENDING_OUTPUT) {
4457 		if (force_legacy_kthread()) {
4458 			if (printk_legacy_kthread)
4459 				wake_up_interruptible(&legacy_wait);
4460 		} else {
4461 			if (console_trylock())
4462 				console_unlock();
4463 		}
4464 	}
4465 
4466 	if (pending & PRINTK_PENDING_WAKEUP)
4467 		wake_up_interruptible(&log_wait);
4468 }
4469 
4470 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
4471 	IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
4472 
__wake_up_klogd(int val)4473 static void __wake_up_klogd(int val)
4474 {
4475 	if (!printk_percpu_data_ready())
4476 		return;
4477 
4478 	preempt_disable();
4479 	/*
4480 	 * Guarantee any new records can be seen by tasks preparing to wait
4481 	 * before this context checks if the wait queue is empty.
4482 	 *
4483 	 * The full memory barrier within wq_has_sleeper() pairs with the full
4484 	 * memory barrier within set_current_state() of
4485 	 * prepare_to_wait_event(), which is called after ___wait_event() adds
4486 	 * the waiter but before it has checked the wait condition.
4487 	 *
4488 	 * This pairs with devkmsg_read:A and syslog_print:A.
4489 	 */
4490 	if (wq_has_sleeper(&log_wait) || /* LMM(__wake_up_klogd:A) */
4491 	    (val & PRINTK_PENDING_OUTPUT)) {
4492 		this_cpu_or(printk_pending, val);
4493 		irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
4494 	}
4495 	preempt_enable();
4496 }
4497 
4498 /**
4499  * wake_up_klogd - Wake kernel logging daemon
4500  *
4501  * Use this function when new records have been added to the ringbuffer
4502  * and the console printing of those records has already occurred or is
4503  * known to be handled by some other context. This function will only
4504  * wake the logging daemon.
4505  *
4506  * Context: Any context.
4507  */
wake_up_klogd(void)4508 void wake_up_klogd(void)
4509 {
4510 	__wake_up_klogd(PRINTK_PENDING_WAKEUP);
4511 }
4512 
4513 /**
4514  * defer_console_output - Wake kernel logging daemon and trigger
4515  *	console printing in a deferred context
4516  *
4517  * Use this function when new records have been added to the ringbuffer,
4518  * this context is responsible for console printing those records, but
4519  * the current context is not allowed to perform the console printing.
4520  * Trigger an irq_work context to perform the console printing. This
4521  * function also wakes the logging daemon.
4522  *
4523  * Context: Any context.
4524  */
defer_console_output(void)4525 void defer_console_output(void)
4526 {
4527 	/*
4528 	 * New messages may have been added directly to the ringbuffer
4529 	 * using vprintk_store(), so wake any waiters as well.
4530 	 */
4531 	__wake_up_klogd(PRINTK_PENDING_WAKEUP | PRINTK_PENDING_OUTPUT);
4532 }
4533 
printk_trigger_flush(void)4534 void printk_trigger_flush(void)
4535 {
4536 	defer_console_output();
4537 }
4538 
vprintk_deferred(const char * fmt,va_list args)4539 int vprintk_deferred(const char *fmt, va_list args)
4540 {
4541 	return vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
4542 }
4543 
_printk_deferred(const char * fmt,...)4544 int _printk_deferred(const char *fmt, ...)
4545 {
4546 	va_list args;
4547 	int r;
4548 
4549 	va_start(args, fmt);
4550 	r = vprintk_deferred(fmt, args);
4551 	va_end(args);
4552 
4553 	return r;
4554 }
4555 
4556 /*
4557  * printk rate limiting, lifted from the networking subsystem.
4558  *
4559  * This enforces a rate limit: not more than 10 kernel messages
4560  * every 5s to make a denial-of-service attack impossible.
4561  */
4562 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
4563 
__printk_ratelimit(const char * func)4564 int __printk_ratelimit(const char *func)
4565 {
4566 	return ___ratelimit(&printk_ratelimit_state, func);
4567 }
4568 EXPORT_SYMBOL(__printk_ratelimit);
4569 
4570 /**
4571  * printk_timed_ratelimit - caller-controlled printk ratelimiting
4572  * @caller_jiffies: pointer to caller's state
4573  * @interval_msecs: minimum interval between prints
4574  *
4575  * printk_timed_ratelimit() returns true if more than @interval_msecs
4576  * milliseconds have elapsed since the last time printk_timed_ratelimit()
4577  * returned true.
4578  */
printk_timed_ratelimit(unsigned long * caller_jiffies,unsigned int interval_msecs)4579 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
4580 			unsigned int interval_msecs)
4581 {
4582 	unsigned long elapsed = jiffies - *caller_jiffies;
4583 
4584 	if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
4585 		return false;
4586 
4587 	*caller_jiffies = jiffies;
4588 	return true;
4589 }
4590 EXPORT_SYMBOL(printk_timed_ratelimit);
4591 
4592 static DEFINE_SPINLOCK(dump_list_lock);
4593 static LIST_HEAD(dump_list);
4594 
4595 /**
4596  * kmsg_dump_register - register a kernel log dumper.
4597  * @dumper: pointer to the kmsg_dumper structure
4598  *
4599  * Adds a kernel log dumper to the system. The dump callback in the
4600  * structure will be called when the kernel oopses or panics and must be
4601  * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
4602  */
kmsg_dump_register(struct kmsg_dumper * dumper)4603 int kmsg_dump_register(struct kmsg_dumper *dumper)
4604 {
4605 	unsigned long flags;
4606 	int err = -EBUSY;
4607 
4608 	/* The dump callback needs to be set */
4609 	if (!dumper->dump)
4610 		return -EINVAL;
4611 
4612 	spin_lock_irqsave(&dump_list_lock, flags);
4613 	/* Don't allow registering multiple times */
4614 	if (!dumper->registered) {
4615 		dumper->registered = 1;
4616 		list_add_tail_rcu(&dumper->list, &dump_list);
4617 		err = 0;
4618 	}
4619 	spin_unlock_irqrestore(&dump_list_lock, flags);
4620 
4621 	return err;
4622 }
4623 EXPORT_SYMBOL_GPL(kmsg_dump_register);
4624 
4625 /**
4626  * kmsg_dump_unregister - unregister a kmsg dumper.
4627  * @dumper: pointer to the kmsg_dumper structure
4628  *
4629  * Removes a dump device from the system. Returns zero on success and
4630  * %-EINVAL otherwise.
4631  */
kmsg_dump_unregister(struct kmsg_dumper * dumper)4632 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
4633 {
4634 	unsigned long flags;
4635 	int err = -EINVAL;
4636 
4637 	spin_lock_irqsave(&dump_list_lock, flags);
4638 	if (dumper->registered) {
4639 		dumper->registered = 0;
4640 		list_del_rcu(&dumper->list);
4641 		err = 0;
4642 	}
4643 	spin_unlock_irqrestore(&dump_list_lock, flags);
4644 	synchronize_rcu();
4645 
4646 	return err;
4647 }
4648 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
4649 
4650 static bool always_kmsg_dump;
4651 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
4652 
kmsg_dump_reason_str(enum kmsg_dump_reason reason)4653 const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
4654 {
4655 	switch (reason) {
4656 	case KMSG_DUMP_PANIC:
4657 		return "Panic";
4658 	case KMSG_DUMP_OOPS:
4659 		return "Oops";
4660 	case KMSG_DUMP_EMERG:
4661 		return "Emergency";
4662 	case KMSG_DUMP_SHUTDOWN:
4663 		return "Shutdown";
4664 	default:
4665 		return "Unknown";
4666 	}
4667 }
4668 EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
4669 
4670 /**
4671  * kmsg_dump_desc - dump kernel log to kernel message dumpers.
4672  * @reason: the reason (oops, panic etc) for dumping
4673  * @desc: a short string to describe what caused the panic or oops. Can be NULL
4674  * if no additional description is available.
4675  *
4676  * Call each of the registered dumper's dump() callback, which can
4677  * retrieve the kmsg records with kmsg_dump_get_line() or
4678  * kmsg_dump_get_buffer().
4679  */
kmsg_dump_desc(enum kmsg_dump_reason reason,const char * desc)4680 void kmsg_dump_desc(enum kmsg_dump_reason reason, const char *desc)
4681 {
4682 	struct kmsg_dumper *dumper;
4683 	struct kmsg_dump_detail detail = {
4684 		.reason = reason,
4685 		.description = desc};
4686 
4687 	rcu_read_lock();
4688 	list_for_each_entry_rcu(dumper, &dump_list, list) {
4689 		enum kmsg_dump_reason max_reason = dumper->max_reason;
4690 
4691 		/*
4692 		 * If client has not provided a specific max_reason, default
4693 		 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
4694 		 */
4695 		if (max_reason == KMSG_DUMP_UNDEF) {
4696 			max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
4697 							KMSG_DUMP_OOPS;
4698 		}
4699 		if (reason > max_reason)
4700 			continue;
4701 
4702 		/* invoke dumper which will iterate over records */
4703 		dumper->dump(dumper, &detail);
4704 	}
4705 	rcu_read_unlock();
4706 }
4707 
4708 /**
4709  * kmsg_dump_get_line - retrieve one kmsg log line
4710  * @iter: kmsg dump iterator
4711  * @syslog: include the "<4>" prefixes
4712  * @line: buffer to copy the line to
4713  * @size: maximum size of the buffer
4714  * @len: length of line placed into buffer
4715  *
4716  * Start at the beginning of the kmsg buffer, with the oldest kmsg
4717  * record, and copy one record into the provided buffer.
4718  *
4719  * Consecutive calls will return the next available record moving
4720  * towards the end of the buffer with the youngest messages.
4721  *
4722  * A return value of FALSE indicates that there are no more records to
4723  * read.
4724  */
kmsg_dump_get_line(struct kmsg_dump_iter * iter,bool syslog,char * line,size_t size,size_t * len)4725 bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
4726 			char *line, size_t size, size_t *len)
4727 {
4728 	u64 min_seq = latched_seq_read_nolock(&clear_seq);
4729 	struct printk_info info;
4730 	unsigned int line_count;
4731 	struct printk_record r;
4732 	size_t l = 0;
4733 	bool ret = false;
4734 
4735 	if (iter->cur_seq < min_seq)
4736 		iter->cur_seq = min_seq;
4737 
4738 	prb_rec_init_rd(&r, &info, line, size);
4739 
4740 	/* Read text or count text lines? */
4741 	if (line) {
4742 		if (!prb_read_valid(prb, iter->cur_seq, &r))
4743 			goto out;
4744 		l = record_print_text(&r, syslog, printk_time);
4745 	} else {
4746 		if (!prb_read_valid_info(prb, iter->cur_seq,
4747 					 &info, &line_count)) {
4748 			goto out;
4749 		}
4750 		l = get_record_print_text_size(&info, line_count, syslog,
4751 					       printk_time);
4752 
4753 	}
4754 
4755 	iter->cur_seq = r.info->seq + 1;
4756 	ret = true;
4757 out:
4758 	if (len)
4759 		*len = l;
4760 	return ret;
4761 }
4762 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
4763 
4764 /**
4765  * kmsg_dump_get_buffer - copy kmsg log lines
4766  * @iter: kmsg dump iterator
4767  * @syslog: include the "<4>" prefixes
4768  * @buf: buffer to copy the line to
4769  * @size: maximum size of the buffer
4770  * @len_out: length of line placed into buffer
4771  *
4772  * Start at the end of the kmsg buffer and fill the provided buffer
4773  * with as many of the *youngest* kmsg records that fit into it.
4774  * If the buffer is large enough, all available kmsg records will be
4775  * copied with a single call.
4776  *
4777  * Consecutive calls will fill the buffer with the next block of
4778  * available older records, not including the earlier retrieved ones.
4779  *
4780  * A return value of FALSE indicates that there are no more records to
4781  * read.
4782  */
kmsg_dump_get_buffer(struct kmsg_dump_iter * iter,bool syslog,char * buf,size_t size,size_t * len_out)4783 bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
4784 			  char *buf, size_t size, size_t *len_out)
4785 {
4786 	u64 min_seq = latched_seq_read_nolock(&clear_seq);
4787 	struct printk_info info;
4788 	struct printk_record r;
4789 	u64 seq;
4790 	u64 next_seq;
4791 	size_t len = 0;
4792 	bool ret = false;
4793 	bool time = printk_time;
4794 
4795 	if (!buf || !size)
4796 		goto out;
4797 
4798 	if (iter->cur_seq < min_seq)
4799 		iter->cur_seq = min_seq;
4800 
4801 	if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
4802 		if (info.seq != iter->cur_seq) {
4803 			/* messages are gone, move to first available one */
4804 			iter->cur_seq = info.seq;
4805 		}
4806 	}
4807 
4808 	/* last entry */
4809 	if (iter->cur_seq >= iter->next_seq)
4810 		goto out;
4811 
4812 	/*
4813 	 * Find first record that fits, including all following records,
4814 	 * into the user-provided buffer for this dump. Pass in size-1
4815 	 * because this function (by way of record_print_text()) will
4816 	 * not write more than size-1 bytes of text into @buf.
4817 	 */
4818 	seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
4819 				     size - 1, syslog, time);
4820 
4821 	/*
4822 	 * Next kmsg_dump_get_buffer() invocation will dump block of
4823 	 * older records stored right before this one.
4824 	 */
4825 	next_seq = seq;
4826 
4827 	prb_rec_init_rd(&r, &info, buf, size);
4828 
4829 	prb_for_each_record(seq, prb, seq, &r) {
4830 		if (r.info->seq >= iter->next_seq)
4831 			break;
4832 
4833 		len += record_print_text(&r, syslog, time);
4834 
4835 		/* Adjust record to store to remaining buffer space. */
4836 		prb_rec_init_rd(&r, &info, buf + len, size - len);
4837 	}
4838 
4839 	iter->next_seq = next_seq;
4840 	ret = true;
4841 out:
4842 	if (len_out)
4843 		*len_out = len;
4844 	return ret;
4845 }
4846 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
4847 
4848 /**
4849  * kmsg_dump_rewind - reset the iterator
4850  * @iter: kmsg dump iterator
4851  *
4852  * Reset the dumper's iterator so that kmsg_dump_get_line() and
4853  * kmsg_dump_get_buffer() can be called again and used multiple
4854  * times within the same dumper.dump() callback.
4855  */
kmsg_dump_rewind(struct kmsg_dump_iter * iter)4856 void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
4857 {
4858 	iter->cur_seq = latched_seq_read_nolock(&clear_seq);
4859 	iter->next_seq = prb_next_seq(prb);
4860 }
4861 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
4862 
4863 /**
4864  * console_try_replay_all - try to replay kernel log on consoles
4865  *
4866  * Try to obtain lock on console subsystem and replay all
4867  * available records in printk buffer on the consoles.
4868  * Does nothing if lock is not obtained.
4869  *
4870  * Context: Any, except for NMI.
4871  */
console_try_replay_all(void)4872 void console_try_replay_all(void)
4873 {
4874 	struct console_flush_type ft;
4875 
4876 	printk_get_console_flush_type(&ft);
4877 	if (console_trylock()) {
4878 		__console_rewind_all();
4879 		if (ft.nbcon_atomic)
4880 			nbcon_atomic_flush_pending();
4881 		if (ft.nbcon_offload)
4882 			nbcon_kthreads_wake();
4883 		if (ft.legacy_offload)
4884 			defer_console_output();
4885 		/* Consoles are flushed as part of console_unlock(). */
4886 		console_unlock();
4887 	}
4888 }
4889 #endif
4890 
4891 #ifdef CONFIG_SMP
4892 static atomic_t printk_cpu_sync_owner = ATOMIC_INIT(-1);
4893 static atomic_t printk_cpu_sync_nested = ATOMIC_INIT(0);
4894 
4895 /**
4896  * __printk_cpu_sync_wait() - Busy wait until the printk cpu-reentrant
4897  *                            spinning lock is not owned by any CPU.
4898  *
4899  * Context: Any context.
4900  */
__printk_cpu_sync_wait(void)4901 void __printk_cpu_sync_wait(void)
4902 {
4903 	do {
4904 		cpu_relax();
4905 	} while (atomic_read(&printk_cpu_sync_owner) != -1);
4906 }
4907 EXPORT_SYMBOL(__printk_cpu_sync_wait);
4908 
4909 /**
4910  * __printk_cpu_sync_try_get() - Try to acquire the printk cpu-reentrant
4911  *                               spinning lock.
4912  *
4913  * If no processor has the lock, the calling processor takes the lock and
4914  * becomes the owner. If the calling processor is already the owner of the
4915  * lock, this function succeeds immediately.
4916  *
4917  * Context: Any context. Expects interrupts to be disabled.
4918  * Return: 1 on success, otherwise 0.
4919  */
__printk_cpu_sync_try_get(void)4920 int __printk_cpu_sync_try_get(void)
4921 {
4922 	int cpu;
4923 	int old;
4924 
4925 	cpu = smp_processor_id();
4926 
4927 	/*
4928 	 * Guarantee loads and stores from this CPU when it is the lock owner
4929 	 * are _not_ visible to the previous lock owner. This pairs with
4930 	 * __printk_cpu_sync_put:B.
4931 	 *
4932 	 * Memory barrier involvement:
4933 	 *
4934 	 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4935 	 * then __printk_cpu_sync_put:A can never read from
4936 	 * __printk_cpu_sync_try_get:B.
4937 	 *
4938 	 * Relies on:
4939 	 *
4940 	 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4941 	 * of the previous CPU
4942 	 *    matching
4943 	 * ACQUIRE from __printk_cpu_sync_try_get:A to
4944 	 * __printk_cpu_sync_try_get:B of this CPU
4945 	 */
4946 	old = atomic_cmpxchg_acquire(&printk_cpu_sync_owner, -1,
4947 				     cpu); /* LMM(__printk_cpu_sync_try_get:A) */
4948 	if (old == -1) {
4949 		/*
4950 		 * This CPU is now the owner and begins loading/storing
4951 		 * data: LMM(__printk_cpu_sync_try_get:B)
4952 		 */
4953 		return 1;
4954 
4955 	} else if (old == cpu) {
4956 		/* This CPU is already the owner. */
4957 		atomic_inc(&printk_cpu_sync_nested);
4958 		return 1;
4959 	}
4960 
4961 	return 0;
4962 }
4963 EXPORT_SYMBOL(__printk_cpu_sync_try_get);
4964 
4965 /**
4966  * __printk_cpu_sync_put() - Release the printk cpu-reentrant spinning lock.
4967  *
4968  * The calling processor must be the owner of the lock.
4969  *
4970  * Context: Any context. Expects interrupts to be disabled.
4971  */
__printk_cpu_sync_put(void)4972 void __printk_cpu_sync_put(void)
4973 {
4974 	if (atomic_read(&printk_cpu_sync_nested)) {
4975 		atomic_dec(&printk_cpu_sync_nested);
4976 		return;
4977 	}
4978 
4979 	/*
4980 	 * This CPU is finished loading/storing data:
4981 	 * LMM(__printk_cpu_sync_put:A)
4982 	 */
4983 
4984 	/*
4985 	 * Guarantee loads and stores from this CPU when it was the
4986 	 * lock owner are visible to the next lock owner. This pairs
4987 	 * with __printk_cpu_sync_try_get:A.
4988 	 *
4989 	 * Memory barrier involvement:
4990 	 *
4991 	 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4992 	 * then __printk_cpu_sync_try_get:B reads from __printk_cpu_sync_put:A.
4993 	 *
4994 	 * Relies on:
4995 	 *
4996 	 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4997 	 * of this CPU
4998 	 *    matching
4999 	 * ACQUIRE from __printk_cpu_sync_try_get:A to
5000 	 * __printk_cpu_sync_try_get:B of the next CPU
5001 	 */
5002 	atomic_set_release(&printk_cpu_sync_owner,
5003 			   -1); /* LMM(__printk_cpu_sync_put:B) */
5004 }
5005 EXPORT_SYMBOL(__printk_cpu_sync_put);
5006 #endif /* CONFIG_SMP */
5007