Loading...
Note: File does not exist in v3.1.
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/printk.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * Modified to make sys_syslog() more flexible: added commands to
8 * return the last 4k of kernel messages, regardless of whether
9 * they've been read or not. Added option to suppress kernel printk's
10 * to the console. Added hook for sending the console messages
11 * elsewhere, in preparation for a serial line console (someday).
12 * Ted Ts'o, 2/11/93.
13 * Modified for sysctl support, 1/8/97, Chris Horn.
14 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
15 * manfred@colorfullife.com
16 * Rewrote bits to get rid of console_lock
17 * 01Mar01 Andrew Morton
18 */
19
20#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21
22#include <linux/kernel.h>
23#include <linux/mm.h>
24#include <linux/tty.h>
25#include <linux/tty_driver.h>
26#include <linux/console.h>
27#include <linux/init.h>
28#include <linux/jiffies.h>
29#include <linux/nmi.h>
30#include <linux/module.h>
31#include <linux/moduleparam.h>
32#include <linux/delay.h>
33#include <linux/smp.h>
34#include <linux/security.h>
35#include <linux/memblock.h>
36#include <linux/syscalls.h>
37#include <linux/crash_core.h>
38#include <linux/ratelimit.h>
39#include <linux/kmsg_dump.h>
40#include <linux/syslog.h>
41#include <linux/cpu.h>
42#include <linux/rculist.h>
43#include <linux/poll.h>
44#include <linux/irq_work.h>
45#include <linux/ctype.h>
46#include <linux/uio.h>
47#include <linux/sched/clock.h>
48#include <linux/sched/debug.h>
49#include <linux/sched/task_stack.h>
50
51#include <linux/uaccess.h>
52#include <asm/sections.h>
53
54#include <trace/events/initcall.h>
55#define CREATE_TRACE_POINTS
56#include <trace/events/printk.h>
57
58#include "printk_ringbuffer.h"
59#include "console_cmdline.h"
60#include "braille.h"
61#include "internal.h"
62
63int console_printk[4] = {
64 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
65 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
66 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
67 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
68};
69EXPORT_SYMBOL_GPL(console_printk);
70
71atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
72EXPORT_SYMBOL(ignore_console_lock_warning);
73
74/*
75 * Low level drivers may need that to know if they can schedule in
76 * their unblank() callback or not. So let's export it.
77 */
78int oops_in_progress;
79EXPORT_SYMBOL(oops_in_progress);
80
81/*
82 * console_sem protects the console_drivers list, and also
83 * provides serialisation for access to the entire console
84 * driver system.
85 */
86static DEFINE_SEMAPHORE(console_sem);
87struct console *console_drivers;
88EXPORT_SYMBOL_GPL(console_drivers);
89
90/*
91 * System may need to suppress printk message under certain
92 * circumstances, like after kernel panic happens.
93 */
94int __read_mostly suppress_printk;
95
96#ifdef CONFIG_LOCKDEP
97static struct lockdep_map console_lock_dep_map = {
98 .name = "console_lock"
99};
100#endif
101
102enum devkmsg_log_bits {
103 __DEVKMSG_LOG_BIT_ON = 0,
104 __DEVKMSG_LOG_BIT_OFF,
105 __DEVKMSG_LOG_BIT_LOCK,
106};
107
108enum devkmsg_log_masks {
109 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
110 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
111 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
112};
113
114/* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
115#define DEVKMSG_LOG_MASK_DEFAULT 0
116
117static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
118
119static int __control_devkmsg(char *str)
120{
121 size_t len;
122
123 if (!str)
124 return -EINVAL;
125
126 len = str_has_prefix(str, "on");
127 if (len) {
128 devkmsg_log = DEVKMSG_LOG_MASK_ON;
129 return len;
130 }
131
132 len = str_has_prefix(str, "off");
133 if (len) {
134 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
135 return len;
136 }
137
138 len = str_has_prefix(str, "ratelimit");
139 if (len) {
140 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
141 return len;
142 }
143
144 return -EINVAL;
145}
146
147static int __init control_devkmsg(char *str)
148{
149 if (__control_devkmsg(str) < 0)
150 return 1;
151
152 /*
153 * Set sysctl string accordingly:
154 */
155 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
156 strcpy(devkmsg_log_str, "on");
157 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
158 strcpy(devkmsg_log_str, "off");
159 /* else "ratelimit" which is set by default. */
160
161 /*
162 * Sysctl cannot change it anymore. The kernel command line setting of
163 * this parameter is to force the setting to be permanent throughout the
164 * runtime of the system. This is a precation measure against userspace
165 * trying to be a smarta** and attempting to change it up on us.
166 */
167 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
168
169 return 0;
170}
171__setup("printk.devkmsg=", control_devkmsg);
172
173char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
174
175int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
176 void *buffer, size_t *lenp, loff_t *ppos)
177{
178 char old_str[DEVKMSG_STR_MAX_SIZE];
179 unsigned int old;
180 int err;
181
182 if (write) {
183 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
184 return -EINVAL;
185
186 old = devkmsg_log;
187 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
188 }
189
190 err = proc_dostring(table, write, buffer, lenp, ppos);
191 if (err)
192 return err;
193
194 if (write) {
195 err = __control_devkmsg(devkmsg_log_str);
196
197 /*
198 * Do not accept an unknown string OR a known string with
199 * trailing crap...
200 */
201 if (err < 0 || (err + 1 != *lenp)) {
202
203 /* ... and restore old setting. */
204 devkmsg_log = old;
205 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
206
207 return -EINVAL;
208 }
209 }
210
211 return 0;
212}
213
214/* Number of registered extended console drivers. */
215static int nr_ext_console_drivers;
216
217/*
218 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
219 * macros instead of functions so that _RET_IP_ contains useful information.
220 */
221#define down_console_sem() do { \
222 down(&console_sem);\
223 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
224} while (0)
225
226static int __down_trylock_console_sem(unsigned long ip)
227{
228 int lock_failed;
229 unsigned long flags;
230
231 /*
232 * Here and in __up_console_sem() we need to be in safe mode,
233 * because spindump/WARN/etc from under console ->lock will
234 * deadlock in printk()->down_trylock_console_sem() otherwise.
235 */
236 printk_safe_enter_irqsave(flags);
237 lock_failed = down_trylock(&console_sem);
238 printk_safe_exit_irqrestore(flags);
239
240 if (lock_failed)
241 return 1;
242 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
243 return 0;
244}
245#define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
246
247static void __up_console_sem(unsigned long ip)
248{
249 unsigned long flags;
250
251 mutex_release(&console_lock_dep_map, ip);
252
253 printk_safe_enter_irqsave(flags);
254 up(&console_sem);
255 printk_safe_exit_irqrestore(flags);
256}
257#define up_console_sem() __up_console_sem(_RET_IP_)
258
259/*
260 * This is used for debugging the mess that is the VT code by
261 * keeping track if we have the console semaphore held. It's
262 * definitely not the perfect debug tool (we don't know if _WE_
263 * hold it and are racing, but it helps tracking those weird code
264 * paths in the console code where we end up in places I want
265 * locked without the console semaphore held).
266 */
267static int console_locked, console_suspended;
268
269/*
270 * If exclusive_console is non-NULL then only this console is to be printed to.
271 */
272static struct console *exclusive_console;
273
274/*
275 * Array of consoles built from command line options (console=)
276 */
277
278#define MAX_CMDLINECONSOLES 8
279
280static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
281
282static int preferred_console = -1;
283static bool has_preferred_console;
284int console_set_on_cmdline;
285EXPORT_SYMBOL(console_set_on_cmdline);
286
287/* Flag: console code may call schedule() */
288static int console_may_schedule;
289
290enum con_msg_format_flags {
291 MSG_FORMAT_DEFAULT = 0,
292 MSG_FORMAT_SYSLOG = (1 << 0),
293};
294
295static int console_msg_format = MSG_FORMAT_DEFAULT;
296
297/*
298 * The printk log buffer consists of a sequenced collection of records, each
299 * containing variable length message text. Every record also contains its
300 * own meta-data (@info).
301 *
302 * Every record meta-data carries the timestamp in microseconds, as well as
303 * the standard userspace syslog level and syslog facility. The usual kernel
304 * messages use LOG_KERN; userspace-injected messages always carry a matching
305 * syslog facility, by default LOG_USER. The origin of every message can be
306 * reliably determined that way.
307 *
308 * The human readable log message of a record is available in @text, the
309 * length of the message text in @text_len. The stored message is not
310 * terminated.
311 *
312 * Optionally, a record can carry a dictionary of properties (key/value
313 * pairs), to provide userspace with a machine-readable message context.
314 *
315 * Examples for well-defined, commonly used property names are:
316 * DEVICE=b12:8 device identifier
317 * b12:8 block dev_t
318 * c127:3 char dev_t
319 * n8 netdev ifindex
320 * +sound:card0 subsystem:devname
321 * SUBSYSTEM=pci driver-core subsystem name
322 *
323 * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
324 * and values are terminated by a '\0' character.
325 *
326 * Example of record values:
327 * record.text_buf = "it's a line" (unterminated)
328 * record.info.seq = 56
329 * record.info.ts_nsec = 36863
330 * record.info.text_len = 11
331 * record.info.facility = 0 (LOG_KERN)
332 * record.info.flags = 0
333 * record.info.level = 3 (LOG_ERR)
334 * record.info.caller_id = 299 (task 299)
335 * record.info.dev_info.subsystem = "pci" (terminated)
336 * record.info.dev_info.device = "+pci:0000:00:01.0" (terminated)
337 *
338 * The 'struct printk_info' buffer must never be directly exported to
339 * userspace, it is a kernel-private implementation detail that might
340 * need to be changed in the future, when the requirements change.
341 *
342 * /dev/kmsg exports the structured data in the following line format:
343 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
344 *
345 * Users of the export format should ignore possible additional values
346 * separated by ',', and find the message after the ';' character.
347 *
348 * The optional key/value pairs are attached as continuation lines starting
349 * with a space character and terminated by a newline. All possible
350 * non-prinatable characters are escaped in the "\xff" notation.
351 */
352
353enum log_flags {
354 LOG_NEWLINE = 2, /* text ended with a newline */
355 LOG_CONT = 8, /* text is a fragment of a continuation line */
356};
357
358/* syslog_lock protects syslog_* variables and write access to clear_seq. */
359static DEFINE_RAW_SPINLOCK(syslog_lock);
360
361#ifdef CONFIG_PRINTK
362DECLARE_WAIT_QUEUE_HEAD(log_wait);
363/* All 3 protected by @syslog_lock. */
364/* the next printk record to read by syslog(READ) or /proc/kmsg */
365static u64 syslog_seq;
366static size_t syslog_partial;
367static bool syslog_time;
368
369/* All 3 protected by @console_sem. */
370/* the next printk record to write to the console */
371static u64 console_seq;
372static u64 exclusive_console_stop_seq;
373static unsigned long console_dropped;
374
375struct latched_seq {
376 seqcount_latch_t latch;
377 u64 val[2];
378};
379
380/*
381 * The next printk record to read after the last 'clear' command. There are
382 * two copies (updated with seqcount_latch) so that reads can locklessly
383 * access a valid value. Writers are synchronized by @syslog_lock.
384 */
385static struct latched_seq clear_seq = {
386 .latch = SEQCNT_LATCH_ZERO(clear_seq.latch),
387 .val[0] = 0,
388 .val[1] = 0,
389};
390
391#ifdef CONFIG_PRINTK_CALLER
392#define PREFIX_MAX 48
393#else
394#define PREFIX_MAX 32
395#endif
396
397/* the maximum size of a formatted record (i.e. with prefix added per line) */
398#define CONSOLE_LOG_MAX 1024
399
400/* the maximum size allowed to be reserved for a record */
401#define LOG_LINE_MAX (CONSOLE_LOG_MAX - PREFIX_MAX)
402
403#define LOG_LEVEL(v) ((v) & 0x07)
404#define LOG_FACILITY(v) ((v) >> 3 & 0xff)
405
406/* record buffer */
407#define LOG_ALIGN __alignof__(unsigned long)
408#define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
409#define LOG_BUF_LEN_MAX (u32)(1 << 31)
410static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
411static char *log_buf = __log_buf;
412static u32 log_buf_len = __LOG_BUF_LEN;
413
414/*
415 * Define the average message size. This only affects the number of
416 * descriptors that will be available. Underestimating is better than
417 * overestimating (too many available descriptors is better than not enough).
418 */
419#define PRB_AVGBITS 5 /* 32 character average length */
420
421#if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
422#error CONFIG_LOG_BUF_SHIFT value too small.
423#endif
424_DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
425 PRB_AVGBITS, &__log_buf[0]);
426
427static struct printk_ringbuffer printk_rb_dynamic;
428
429static struct printk_ringbuffer *prb = &printk_rb_static;
430
431/*
432 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
433 * per_cpu_areas are initialised. This variable is set to true when
434 * it's safe to access per-CPU data.
435 */
436static bool __printk_percpu_data_ready __read_mostly;
437
438bool printk_percpu_data_ready(void)
439{
440 return __printk_percpu_data_ready;
441}
442
443/* Must be called under syslog_lock. */
444static void latched_seq_write(struct latched_seq *ls, u64 val)
445{
446 raw_write_seqcount_latch(&ls->latch);
447 ls->val[0] = val;
448 raw_write_seqcount_latch(&ls->latch);
449 ls->val[1] = val;
450}
451
452/* Can be called from any context. */
453static u64 latched_seq_read_nolock(struct latched_seq *ls)
454{
455 unsigned int seq;
456 unsigned int idx;
457 u64 val;
458
459 do {
460 seq = raw_read_seqcount_latch(&ls->latch);
461 idx = seq & 0x1;
462 val = ls->val[idx];
463 } while (read_seqcount_latch_retry(&ls->latch, seq));
464
465 return val;
466}
467
468/* Return log buffer address */
469char *log_buf_addr_get(void)
470{
471 return log_buf;
472}
473
474/* Return log buffer size */
475u32 log_buf_len_get(void)
476{
477 return log_buf_len;
478}
479
480/*
481 * Define how much of the log buffer we could take at maximum. The value
482 * must be greater than two. Note that only half of the buffer is available
483 * when the index points to the middle.
484 */
485#define MAX_LOG_TAKE_PART 4
486static const char trunc_msg[] = "<truncated>";
487
488static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
489{
490 /*
491 * The message should not take the whole buffer. Otherwise, it might
492 * get removed too soon.
493 */
494 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
495
496 if (*text_len > max_text_len)
497 *text_len = max_text_len;
498
499 /* enable the warning message (if there is room) */
500 *trunc_msg_len = strlen(trunc_msg);
501 if (*text_len >= *trunc_msg_len)
502 *text_len -= *trunc_msg_len;
503 else
504 *trunc_msg_len = 0;
505}
506
507int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
508
509static int syslog_action_restricted(int type)
510{
511 if (dmesg_restrict)
512 return 1;
513 /*
514 * Unless restricted, we allow "read all" and "get buffer size"
515 * for everybody.
516 */
517 return type != SYSLOG_ACTION_READ_ALL &&
518 type != SYSLOG_ACTION_SIZE_BUFFER;
519}
520
521static int check_syslog_permissions(int type, int source)
522{
523 /*
524 * If this is from /proc/kmsg and we've already opened it, then we've
525 * already done the capabilities checks at open time.
526 */
527 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
528 goto ok;
529
530 if (syslog_action_restricted(type)) {
531 if (capable(CAP_SYSLOG))
532 goto ok;
533 /*
534 * For historical reasons, accept CAP_SYS_ADMIN too, with
535 * a warning.
536 */
537 if (capable(CAP_SYS_ADMIN)) {
538 pr_warn_once("%s (%d): Attempt to access syslog with "
539 "CAP_SYS_ADMIN but no CAP_SYSLOG "
540 "(deprecated).\n",
541 current->comm, task_pid_nr(current));
542 goto ok;
543 }
544 return -EPERM;
545 }
546ok:
547 return security_syslog(type);
548}
549
550static void append_char(char **pp, char *e, char c)
551{
552 if (*pp < e)
553 *(*pp)++ = c;
554}
555
556static ssize_t info_print_ext_header(char *buf, size_t size,
557 struct printk_info *info)
558{
559 u64 ts_usec = info->ts_nsec;
560 char caller[20];
561#ifdef CONFIG_PRINTK_CALLER
562 u32 id = info->caller_id;
563
564 snprintf(caller, sizeof(caller), ",caller=%c%u",
565 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
566#else
567 caller[0] = '\0';
568#endif
569
570 do_div(ts_usec, 1000);
571
572 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
573 (info->facility << 3) | info->level, info->seq,
574 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
575}
576
577static ssize_t msg_add_ext_text(char *buf, size_t size,
578 const char *text, size_t text_len,
579 unsigned char endc)
580{
581 char *p = buf, *e = buf + size;
582 size_t i;
583
584 /* escape non-printable characters */
585 for (i = 0; i < text_len; i++) {
586 unsigned char c = text[i];
587
588 if (c < ' ' || c >= 127 || c == '\\')
589 p += scnprintf(p, e - p, "\\x%02x", c);
590 else
591 append_char(&p, e, c);
592 }
593 append_char(&p, e, endc);
594
595 return p - buf;
596}
597
598static ssize_t msg_add_dict_text(char *buf, size_t size,
599 const char *key, const char *val)
600{
601 size_t val_len = strlen(val);
602 ssize_t len;
603
604 if (!val_len)
605 return 0;
606
607 len = msg_add_ext_text(buf, size, "", 0, ' '); /* dict prefix */
608 len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
609 len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
610
611 return len;
612}
613
614static ssize_t msg_print_ext_body(char *buf, size_t size,
615 char *text, size_t text_len,
616 struct dev_printk_info *dev_info)
617{
618 ssize_t len;
619
620 len = msg_add_ext_text(buf, size, text, text_len, '\n');
621
622 if (!dev_info)
623 goto out;
624
625 len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
626 dev_info->subsystem);
627 len += msg_add_dict_text(buf + len, size - len, "DEVICE",
628 dev_info->device);
629out:
630 return len;
631}
632
633/* /dev/kmsg - userspace message inject/listen interface */
634struct devkmsg_user {
635 atomic64_t seq;
636 struct ratelimit_state rs;
637 struct mutex lock;
638 char buf[CONSOLE_EXT_LOG_MAX];
639
640 struct printk_info info;
641 char text_buf[CONSOLE_EXT_LOG_MAX];
642 struct printk_record record;
643};
644
645static __printf(3, 4) __cold
646int devkmsg_emit(int facility, int level, const char *fmt, ...)
647{
648 va_list args;
649 int r;
650
651 va_start(args, fmt);
652 r = vprintk_emit(facility, level, NULL, fmt, args);
653 va_end(args);
654
655 return r;
656}
657
658static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
659{
660 char *buf, *line;
661 int level = default_message_loglevel;
662 int facility = 1; /* LOG_USER */
663 struct file *file = iocb->ki_filp;
664 struct devkmsg_user *user = file->private_data;
665 size_t len = iov_iter_count(from);
666 ssize_t ret = len;
667
668 if (!user || len > LOG_LINE_MAX)
669 return -EINVAL;
670
671 /* Ignore when user logging is disabled. */
672 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
673 return len;
674
675 /* Ratelimit when not explicitly enabled. */
676 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
677 if (!___ratelimit(&user->rs, current->comm))
678 return ret;
679 }
680
681 buf = kmalloc(len+1, GFP_KERNEL);
682 if (buf == NULL)
683 return -ENOMEM;
684
685 buf[len] = '\0';
686 if (!copy_from_iter_full(buf, len, from)) {
687 kfree(buf);
688 return -EFAULT;
689 }
690
691 /*
692 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
693 * the decimal value represents 32bit, the lower 3 bit are the log
694 * level, the rest are the log facility.
695 *
696 * If no prefix or no userspace facility is specified, we
697 * enforce LOG_USER, to be able to reliably distinguish
698 * kernel-generated messages from userspace-injected ones.
699 */
700 line = buf;
701 if (line[0] == '<') {
702 char *endp = NULL;
703 unsigned int u;
704
705 u = simple_strtoul(line + 1, &endp, 10);
706 if (endp && endp[0] == '>') {
707 level = LOG_LEVEL(u);
708 if (LOG_FACILITY(u) != 0)
709 facility = LOG_FACILITY(u);
710 endp++;
711 line = endp;
712 }
713 }
714
715 devkmsg_emit(facility, level, "%s", line);
716 kfree(buf);
717 return ret;
718}
719
720static ssize_t devkmsg_read(struct file *file, char __user *buf,
721 size_t count, loff_t *ppos)
722{
723 struct devkmsg_user *user = file->private_data;
724 struct printk_record *r = &user->record;
725 size_t len;
726 ssize_t ret;
727
728 if (!user)
729 return -EBADF;
730
731 ret = mutex_lock_interruptible(&user->lock);
732 if (ret)
733 return ret;
734
735 printk_safe_enter_irq();
736 if (!prb_read_valid(prb, atomic64_read(&user->seq), r)) {
737 if (file->f_flags & O_NONBLOCK) {
738 ret = -EAGAIN;
739 printk_safe_exit_irq();
740 goto out;
741 }
742
743 printk_safe_exit_irq();
744 ret = wait_event_interruptible(log_wait,
745 prb_read_valid(prb, atomic64_read(&user->seq), r));
746 if (ret)
747 goto out;
748 printk_safe_enter_irq();
749 }
750
751 if (r->info->seq != atomic64_read(&user->seq)) {
752 /* our last seen message is gone, return error and reset */
753 atomic64_set(&user->seq, r->info->seq);
754 ret = -EPIPE;
755 printk_safe_exit_irq();
756 goto out;
757 }
758
759 len = info_print_ext_header(user->buf, sizeof(user->buf), r->info);
760 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
761 &r->text_buf[0], r->info->text_len,
762 &r->info->dev_info);
763
764 atomic64_set(&user->seq, r->info->seq + 1);
765 printk_safe_exit_irq();
766
767 if (len > count) {
768 ret = -EINVAL;
769 goto out;
770 }
771
772 if (copy_to_user(buf, user->buf, len)) {
773 ret = -EFAULT;
774 goto out;
775 }
776 ret = len;
777out:
778 mutex_unlock(&user->lock);
779 return ret;
780}
781
782/*
783 * Be careful when modifying this function!!!
784 *
785 * Only few operations are supported because the device works only with the
786 * entire variable length messages (records). Non-standard values are
787 * returned in the other cases and has been this way for quite some time.
788 * User space applications might depend on this behavior.
789 */
790static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
791{
792 struct devkmsg_user *user = file->private_data;
793 loff_t ret = 0;
794
795 if (!user)
796 return -EBADF;
797 if (offset)
798 return -ESPIPE;
799
800 printk_safe_enter_irq();
801 switch (whence) {
802 case SEEK_SET:
803 /* the first record */
804 atomic64_set(&user->seq, prb_first_valid_seq(prb));
805 break;
806 case SEEK_DATA:
807 /*
808 * The first record after the last SYSLOG_ACTION_CLEAR,
809 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
810 * changes no global state, and does not clear anything.
811 */
812 atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
813 break;
814 case SEEK_END:
815 /* after the last record */
816 atomic64_set(&user->seq, prb_next_seq(prb));
817 break;
818 default:
819 ret = -EINVAL;
820 }
821 printk_safe_exit_irq();
822 return ret;
823}
824
825static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
826{
827 struct devkmsg_user *user = file->private_data;
828 struct printk_info info;
829 __poll_t ret = 0;
830
831 if (!user)
832 return EPOLLERR|EPOLLNVAL;
833
834 poll_wait(file, &log_wait, wait);
835
836 printk_safe_enter_irq();
837 if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
838 /* return error when data has vanished underneath us */
839 if (info.seq != atomic64_read(&user->seq))
840 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
841 else
842 ret = EPOLLIN|EPOLLRDNORM;
843 }
844 printk_safe_exit_irq();
845
846 return ret;
847}
848
849static int devkmsg_open(struct inode *inode, struct file *file)
850{
851 struct devkmsg_user *user;
852 int err;
853
854 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
855 return -EPERM;
856
857 /* write-only does not need any file context */
858 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
859 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
860 SYSLOG_FROM_READER);
861 if (err)
862 return err;
863 }
864
865 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
866 if (!user)
867 return -ENOMEM;
868
869 ratelimit_default_init(&user->rs);
870 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
871
872 mutex_init(&user->lock);
873
874 prb_rec_init_rd(&user->record, &user->info,
875 &user->text_buf[0], sizeof(user->text_buf));
876
877 printk_safe_enter_irq();
878 atomic64_set(&user->seq, prb_first_valid_seq(prb));
879 printk_safe_exit_irq();
880
881 file->private_data = user;
882 return 0;
883}
884
885static int devkmsg_release(struct inode *inode, struct file *file)
886{
887 struct devkmsg_user *user = file->private_data;
888
889 if (!user)
890 return 0;
891
892 ratelimit_state_exit(&user->rs);
893
894 mutex_destroy(&user->lock);
895 kfree(user);
896 return 0;
897}
898
899const struct file_operations kmsg_fops = {
900 .open = devkmsg_open,
901 .read = devkmsg_read,
902 .write_iter = devkmsg_write,
903 .llseek = devkmsg_llseek,
904 .poll = devkmsg_poll,
905 .release = devkmsg_release,
906};
907
908#ifdef CONFIG_CRASH_CORE
909/*
910 * This appends the listed symbols to /proc/vmcore
911 *
912 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
913 * obtain access to symbols that are otherwise very difficult to locate. These
914 * symbols are specifically used so that utilities can access and extract the
915 * dmesg log from a vmcore file after a crash.
916 */
917void log_buf_vmcoreinfo_setup(void)
918{
919 struct dev_printk_info *dev_info = NULL;
920
921 VMCOREINFO_SYMBOL(prb);
922 VMCOREINFO_SYMBOL(printk_rb_static);
923 VMCOREINFO_SYMBOL(clear_seq);
924
925 /*
926 * Export struct size and field offsets. User space tools can
927 * parse it and detect any changes to structure down the line.
928 */
929
930 VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
931 VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
932 VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
933 VMCOREINFO_OFFSET(printk_ringbuffer, fail);
934
935 VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
936 VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
937 VMCOREINFO_OFFSET(prb_desc_ring, descs);
938 VMCOREINFO_OFFSET(prb_desc_ring, infos);
939 VMCOREINFO_OFFSET(prb_desc_ring, head_id);
940 VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
941
942 VMCOREINFO_STRUCT_SIZE(prb_desc);
943 VMCOREINFO_OFFSET(prb_desc, state_var);
944 VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
945
946 VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
947 VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
948 VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
949
950 VMCOREINFO_STRUCT_SIZE(printk_info);
951 VMCOREINFO_OFFSET(printk_info, seq);
952 VMCOREINFO_OFFSET(printk_info, ts_nsec);
953 VMCOREINFO_OFFSET(printk_info, text_len);
954 VMCOREINFO_OFFSET(printk_info, caller_id);
955 VMCOREINFO_OFFSET(printk_info, dev_info);
956
957 VMCOREINFO_STRUCT_SIZE(dev_printk_info);
958 VMCOREINFO_OFFSET(dev_printk_info, subsystem);
959 VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
960 VMCOREINFO_OFFSET(dev_printk_info, device);
961 VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
962
963 VMCOREINFO_STRUCT_SIZE(prb_data_ring);
964 VMCOREINFO_OFFSET(prb_data_ring, size_bits);
965 VMCOREINFO_OFFSET(prb_data_ring, data);
966 VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
967 VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
968
969 VMCOREINFO_SIZE(atomic_long_t);
970 VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
971
972 VMCOREINFO_STRUCT_SIZE(latched_seq);
973 VMCOREINFO_OFFSET(latched_seq, val);
974}
975#endif
976
977/* requested log_buf_len from kernel cmdline */
978static unsigned long __initdata new_log_buf_len;
979
980/* we practice scaling the ring buffer by powers of 2 */
981static void __init log_buf_len_update(u64 size)
982{
983 if (size > (u64)LOG_BUF_LEN_MAX) {
984 size = (u64)LOG_BUF_LEN_MAX;
985 pr_err("log_buf over 2G is not supported.\n");
986 }
987
988 if (size)
989 size = roundup_pow_of_two(size);
990 if (size > log_buf_len)
991 new_log_buf_len = (unsigned long)size;
992}
993
994/* save requested log_buf_len since it's too early to process it */
995static int __init log_buf_len_setup(char *str)
996{
997 u64 size;
998
999 if (!str)
1000 return -EINVAL;
1001
1002 size = memparse(str, &str);
1003
1004 log_buf_len_update(size);
1005
1006 return 0;
1007}
1008early_param("log_buf_len", log_buf_len_setup);
1009
1010#ifdef CONFIG_SMP
1011#define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1012
1013static void __init log_buf_add_cpu(void)
1014{
1015 unsigned int cpu_extra;
1016
1017 /*
1018 * archs should set up cpu_possible_bits properly with
1019 * set_cpu_possible() after setup_arch() but just in
1020 * case lets ensure this is valid.
1021 */
1022 if (num_possible_cpus() == 1)
1023 return;
1024
1025 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1026
1027 /* by default this will only continue through for large > 64 CPUs */
1028 if (cpu_extra <= __LOG_BUF_LEN / 2)
1029 return;
1030
1031 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1032 __LOG_CPU_MAX_BUF_LEN);
1033 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1034 cpu_extra);
1035 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1036
1037 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1038}
1039#else /* !CONFIG_SMP */
1040static inline void log_buf_add_cpu(void) {}
1041#endif /* CONFIG_SMP */
1042
1043static void __init set_percpu_data_ready(void)
1044{
1045 printk_safe_init();
1046 /* Make sure we set this flag only after printk_safe() init is done */
1047 barrier();
1048 __printk_percpu_data_ready = true;
1049}
1050
1051static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1052 struct printk_record *r)
1053{
1054 struct prb_reserved_entry e;
1055 struct printk_record dest_r;
1056
1057 prb_rec_init_wr(&dest_r, r->info->text_len);
1058
1059 if (!prb_reserve(&e, rb, &dest_r))
1060 return 0;
1061
1062 memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1063 dest_r.info->text_len = r->info->text_len;
1064 dest_r.info->facility = r->info->facility;
1065 dest_r.info->level = r->info->level;
1066 dest_r.info->flags = r->info->flags;
1067 dest_r.info->ts_nsec = r->info->ts_nsec;
1068 dest_r.info->caller_id = r->info->caller_id;
1069 memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1070
1071 prb_final_commit(&e);
1072
1073 return prb_record_text_space(&e);
1074}
1075
1076static char setup_text_buf[LOG_LINE_MAX] __initdata;
1077
1078void __init setup_log_buf(int early)
1079{
1080 struct printk_info *new_infos;
1081 unsigned int new_descs_count;
1082 struct prb_desc *new_descs;
1083 struct printk_info info;
1084 struct printk_record r;
1085 size_t new_descs_size;
1086 size_t new_infos_size;
1087 unsigned long flags;
1088 char *new_log_buf;
1089 unsigned int free;
1090 u64 seq;
1091
1092 /*
1093 * Some archs call setup_log_buf() multiple times - first is very
1094 * early, e.g. from setup_arch(), and second - when percpu_areas
1095 * are initialised.
1096 */
1097 if (!early)
1098 set_percpu_data_ready();
1099
1100 if (log_buf != __log_buf)
1101 return;
1102
1103 if (!early && !new_log_buf_len)
1104 log_buf_add_cpu();
1105
1106 if (!new_log_buf_len)
1107 return;
1108
1109 new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1110 if (new_descs_count == 0) {
1111 pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1112 return;
1113 }
1114
1115 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1116 if (unlikely(!new_log_buf)) {
1117 pr_err("log_buf_len: %lu text bytes not available\n",
1118 new_log_buf_len);
1119 return;
1120 }
1121
1122 new_descs_size = new_descs_count * sizeof(struct prb_desc);
1123 new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1124 if (unlikely(!new_descs)) {
1125 pr_err("log_buf_len: %zu desc bytes not available\n",
1126 new_descs_size);
1127 goto err_free_log_buf;
1128 }
1129
1130 new_infos_size = new_descs_count * sizeof(struct printk_info);
1131 new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1132 if (unlikely(!new_infos)) {
1133 pr_err("log_buf_len: %zu info bytes not available\n",
1134 new_infos_size);
1135 goto err_free_descs;
1136 }
1137
1138 prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1139
1140 prb_init(&printk_rb_dynamic,
1141 new_log_buf, ilog2(new_log_buf_len),
1142 new_descs, ilog2(new_descs_count),
1143 new_infos);
1144
1145 printk_safe_enter_irqsave(flags);
1146
1147 log_buf_len = new_log_buf_len;
1148 log_buf = new_log_buf;
1149 new_log_buf_len = 0;
1150
1151 free = __LOG_BUF_LEN;
1152 prb_for_each_record(0, &printk_rb_static, seq, &r)
1153 free -= add_to_rb(&printk_rb_dynamic, &r);
1154
1155 /*
1156 * This is early enough that everything is still running on the
1157 * boot CPU and interrupts are disabled. So no new messages will
1158 * appear during the transition to the dynamic buffer.
1159 */
1160 prb = &printk_rb_dynamic;
1161
1162 printk_safe_exit_irqrestore(flags);
1163
1164 if (seq != prb_next_seq(&printk_rb_static)) {
1165 pr_err("dropped %llu messages\n",
1166 prb_next_seq(&printk_rb_static) - seq);
1167 }
1168
1169 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1170 pr_info("early log buf free: %u(%u%%)\n",
1171 free, (free * 100) / __LOG_BUF_LEN);
1172 return;
1173
1174err_free_descs:
1175 memblock_free(__pa(new_descs), new_descs_size);
1176err_free_log_buf:
1177 memblock_free(__pa(new_log_buf), new_log_buf_len);
1178}
1179
1180static bool __read_mostly ignore_loglevel;
1181
1182static int __init ignore_loglevel_setup(char *str)
1183{
1184 ignore_loglevel = true;
1185 pr_info("debug: ignoring loglevel setting.\n");
1186
1187 return 0;
1188}
1189
1190early_param("ignore_loglevel", ignore_loglevel_setup);
1191module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1192MODULE_PARM_DESC(ignore_loglevel,
1193 "ignore loglevel setting (prints all kernel messages to the console)");
1194
1195static bool suppress_message_printing(int level)
1196{
1197 return (level >= console_loglevel && !ignore_loglevel);
1198}
1199
1200#ifdef CONFIG_BOOT_PRINTK_DELAY
1201
1202static int boot_delay; /* msecs delay after each printk during bootup */
1203static unsigned long long loops_per_msec; /* based on boot_delay */
1204
1205static int __init boot_delay_setup(char *str)
1206{
1207 unsigned long lpj;
1208
1209 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1210 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1211
1212 get_option(&str, &boot_delay);
1213 if (boot_delay > 10 * 1000)
1214 boot_delay = 0;
1215
1216 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1217 "HZ: %d, loops_per_msec: %llu\n",
1218 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1219 return 0;
1220}
1221early_param("boot_delay", boot_delay_setup);
1222
1223static void boot_delay_msec(int level)
1224{
1225 unsigned long long k;
1226 unsigned long timeout;
1227
1228 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1229 || suppress_message_printing(level)) {
1230 return;
1231 }
1232
1233 k = (unsigned long long)loops_per_msec * boot_delay;
1234
1235 timeout = jiffies + msecs_to_jiffies(boot_delay);
1236 while (k) {
1237 k--;
1238 cpu_relax();
1239 /*
1240 * use (volatile) jiffies to prevent
1241 * compiler reduction; loop termination via jiffies
1242 * is secondary and may or may not happen.
1243 */
1244 if (time_after(jiffies, timeout))
1245 break;
1246 touch_nmi_watchdog();
1247 }
1248}
1249#else
1250static inline void boot_delay_msec(int level)
1251{
1252}
1253#endif
1254
1255static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1256module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1257
1258static size_t print_syslog(unsigned int level, char *buf)
1259{
1260 return sprintf(buf, "<%u>", level);
1261}
1262
1263static size_t print_time(u64 ts, char *buf)
1264{
1265 unsigned long rem_nsec = do_div(ts, 1000000000);
1266
1267 return sprintf(buf, "[%5lu.%06lu]",
1268 (unsigned long)ts, rem_nsec / 1000);
1269}
1270
1271#ifdef CONFIG_PRINTK_CALLER
1272static size_t print_caller(u32 id, char *buf)
1273{
1274 char caller[12];
1275
1276 snprintf(caller, sizeof(caller), "%c%u",
1277 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1278 return sprintf(buf, "[%6s]", caller);
1279}
1280#else
1281#define print_caller(id, buf) 0
1282#endif
1283
1284static size_t info_print_prefix(const struct printk_info *info, bool syslog,
1285 bool time, char *buf)
1286{
1287 size_t len = 0;
1288
1289 if (syslog)
1290 len = print_syslog((info->facility << 3) | info->level, buf);
1291
1292 if (time)
1293 len += print_time(info->ts_nsec, buf + len);
1294
1295 len += print_caller(info->caller_id, buf + len);
1296
1297 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1298 buf[len++] = ' ';
1299 buf[len] = '\0';
1300 }
1301
1302 return len;
1303}
1304
1305/*
1306 * Prepare the record for printing. The text is shifted within the given
1307 * buffer to avoid a need for another one. The following operations are
1308 * done:
1309 *
1310 * - Add prefix for each line.
1311 * - Drop truncated lines that no longer fit into the buffer.
1312 * - Add the trailing newline that has been removed in vprintk_store().
1313 * - Add a string terminator.
1314 *
1315 * Since the produced string is always terminated, the maximum possible
1316 * return value is @r->text_buf_size - 1;
1317 *
1318 * Return: The length of the updated/prepared text, including the added
1319 * prefixes and the newline. The terminator is not counted. The dropped
1320 * line(s) are not counted.
1321 */
1322static size_t record_print_text(struct printk_record *r, bool syslog,
1323 bool time)
1324{
1325 size_t text_len = r->info->text_len;
1326 size_t buf_size = r->text_buf_size;
1327 char *text = r->text_buf;
1328 char prefix[PREFIX_MAX];
1329 bool truncated = false;
1330 size_t prefix_len;
1331 size_t line_len;
1332 size_t len = 0;
1333 char *next;
1334
1335 /*
1336 * If the message was truncated because the buffer was not large
1337 * enough, treat the available text as if it were the full text.
1338 */
1339 if (text_len > buf_size)
1340 text_len = buf_size;
1341
1342 prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1343
1344 /*
1345 * @text_len: bytes of unprocessed text
1346 * @line_len: bytes of current line _without_ newline
1347 * @text: pointer to beginning of current line
1348 * @len: number of bytes prepared in r->text_buf
1349 */
1350 for (;;) {
1351 next = memchr(text, '\n', text_len);
1352 if (next) {
1353 line_len = next - text;
1354 } else {
1355 /* Drop truncated line(s). */
1356 if (truncated)
1357 break;
1358 line_len = text_len;
1359 }
1360
1361 /*
1362 * Truncate the text if there is not enough space to add the
1363 * prefix and a trailing newline and a terminator.
1364 */
1365 if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1366 /* Drop even the current line if no space. */
1367 if (len + prefix_len + line_len + 1 + 1 > buf_size)
1368 break;
1369
1370 text_len = buf_size - len - prefix_len - 1 - 1;
1371 truncated = true;
1372 }
1373
1374 memmove(text + prefix_len, text, text_len);
1375 memcpy(text, prefix, prefix_len);
1376
1377 /*
1378 * Increment the prepared length to include the text and
1379 * prefix that were just moved+copied. Also increment for the
1380 * newline at the end of this line. If this is the last line,
1381 * there is no newline, but it will be added immediately below.
1382 */
1383 len += prefix_len + line_len + 1;
1384 if (text_len == line_len) {
1385 /*
1386 * This is the last line. Add the trailing newline
1387 * removed in vprintk_store().
1388 */
1389 text[prefix_len + line_len] = '\n';
1390 break;
1391 }
1392
1393 /*
1394 * Advance beyond the added prefix and the related line with
1395 * its newline.
1396 */
1397 text += prefix_len + line_len + 1;
1398
1399 /*
1400 * The remaining text has only decreased by the line with its
1401 * newline.
1402 *
1403 * Note that @text_len can become zero. It happens when @text
1404 * ended with a newline (either due to truncation or the
1405 * original string ending with "\n\n"). The loop is correctly
1406 * repeated and (if not truncated) an empty line with a prefix
1407 * will be prepared.
1408 */
1409 text_len -= line_len + 1;
1410 }
1411
1412 /*
1413 * If a buffer was provided, it will be terminated. Space for the
1414 * string terminator is guaranteed to be available. The terminator is
1415 * not counted in the return value.
1416 */
1417 if (buf_size > 0)
1418 r->text_buf[len] = 0;
1419
1420 return len;
1421}
1422
1423static size_t get_record_print_text_size(struct printk_info *info,
1424 unsigned int line_count,
1425 bool syslog, bool time)
1426{
1427 char prefix[PREFIX_MAX];
1428 size_t prefix_len;
1429
1430 prefix_len = info_print_prefix(info, syslog, time, prefix);
1431
1432 /*
1433 * Each line will be preceded with a prefix. The intermediate
1434 * newlines are already within the text, but a final trailing
1435 * newline will be added.
1436 */
1437 return ((prefix_len * line_count) + info->text_len + 1);
1438}
1439
1440/*
1441 * Beginning with @start_seq, find the first record where it and all following
1442 * records up to (but not including) @max_seq fit into @size.
1443 *
1444 * @max_seq is simply an upper bound and does not need to exist. If the caller
1445 * does not require an upper bound, -1 can be used for @max_seq.
1446 */
1447static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1448 bool syslog, bool time)
1449{
1450 struct printk_info info;
1451 unsigned int line_count;
1452 size_t len = 0;
1453 u64 seq;
1454
1455 /* Determine the size of the records up to @max_seq. */
1456 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1457 if (info.seq >= max_seq)
1458 break;
1459 len += get_record_print_text_size(&info, line_count, syslog, time);
1460 }
1461
1462 /*
1463 * Adjust the upper bound for the next loop to avoid subtracting
1464 * lengths that were never added.
1465 */
1466 if (seq < max_seq)
1467 max_seq = seq;
1468
1469 /*
1470 * Move first record forward until length fits into the buffer. Ignore
1471 * newest messages that were not counted in the above cycle. Messages
1472 * might appear and get lost in the meantime. This is a best effort
1473 * that prevents an infinite loop that could occur with a retry.
1474 */
1475 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1476 if (len <= size || info.seq >= max_seq)
1477 break;
1478 len -= get_record_print_text_size(&info, line_count, syslog, time);
1479 }
1480
1481 return seq;
1482}
1483
1484static int syslog_print(char __user *buf, int size)
1485{
1486 struct printk_info info;
1487 struct printk_record r;
1488 char *text;
1489 int len = 0;
1490
1491 text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1492 if (!text)
1493 return -ENOMEM;
1494
1495 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1496
1497 while (size > 0) {
1498 size_t n;
1499 size_t skip;
1500
1501 printk_safe_enter_irq();
1502 raw_spin_lock(&syslog_lock);
1503 if (!prb_read_valid(prb, syslog_seq, &r)) {
1504 raw_spin_unlock(&syslog_lock);
1505 printk_safe_exit_irq();
1506 break;
1507 }
1508 if (r.info->seq != syslog_seq) {
1509 /* message is gone, move to next valid one */
1510 syslog_seq = r.info->seq;
1511 syslog_partial = 0;
1512 }
1513
1514 /*
1515 * To keep reading/counting partial line consistent,
1516 * use printk_time value as of the beginning of a line.
1517 */
1518 if (!syslog_partial)
1519 syslog_time = printk_time;
1520
1521 skip = syslog_partial;
1522 n = record_print_text(&r, true, syslog_time);
1523 if (n - syslog_partial <= size) {
1524 /* message fits into buffer, move forward */
1525 syslog_seq = r.info->seq + 1;
1526 n -= syslog_partial;
1527 syslog_partial = 0;
1528 } else if (!len){
1529 /* partial read(), remember position */
1530 n = size;
1531 syslog_partial += n;
1532 } else
1533 n = 0;
1534 raw_spin_unlock(&syslog_lock);
1535 printk_safe_exit_irq();
1536
1537 if (!n)
1538 break;
1539
1540 if (copy_to_user(buf, text + skip, n)) {
1541 if (!len)
1542 len = -EFAULT;
1543 break;
1544 }
1545
1546 len += n;
1547 size -= n;
1548 buf += n;
1549 }
1550
1551 kfree(text);
1552 return len;
1553}
1554
1555static int syslog_print_all(char __user *buf, int size, bool clear)
1556{
1557 struct printk_info info;
1558 struct printk_record r;
1559 char *text;
1560 int len = 0;
1561 u64 seq;
1562 bool time;
1563
1564 text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1565 if (!text)
1566 return -ENOMEM;
1567
1568 time = printk_time;
1569 printk_safe_enter_irq();
1570 /*
1571 * Find first record that fits, including all following records,
1572 * into the user-provided buffer for this dump.
1573 */
1574 seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1575 size, true, time);
1576
1577 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1578
1579 len = 0;
1580 prb_for_each_record(seq, prb, seq, &r) {
1581 int textlen;
1582
1583 textlen = record_print_text(&r, true, time);
1584
1585 if (len + textlen > size) {
1586 seq--;
1587 break;
1588 }
1589
1590 printk_safe_exit_irq();
1591 if (copy_to_user(buf + len, text, textlen))
1592 len = -EFAULT;
1593 else
1594 len += textlen;
1595 printk_safe_enter_irq();
1596
1597 if (len < 0)
1598 break;
1599 }
1600
1601 if (clear) {
1602 raw_spin_lock(&syslog_lock);
1603 latched_seq_write(&clear_seq, seq);
1604 raw_spin_unlock(&syslog_lock);
1605 }
1606 printk_safe_exit_irq();
1607
1608 kfree(text);
1609 return len;
1610}
1611
1612static void syslog_clear(void)
1613{
1614 printk_safe_enter_irq();
1615 raw_spin_lock(&syslog_lock);
1616 latched_seq_write(&clear_seq, prb_next_seq(prb));
1617 raw_spin_unlock(&syslog_lock);
1618 printk_safe_exit_irq();
1619}
1620
1621/* Return a consistent copy of @syslog_seq. */
1622static u64 read_syslog_seq_irq(void)
1623{
1624 u64 seq;
1625
1626 raw_spin_lock_irq(&syslog_lock);
1627 seq = syslog_seq;
1628 raw_spin_unlock_irq(&syslog_lock);
1629
1630 return seq;
1631}
1632
1633int do_syslog(int type, char __user *buf, int len, int source)
1634{
1635 struct printk_info info;
1636 bool clear = false;
1637 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1638 int error;
1639
1640 error = check_syslog_permissions(type, source);
1641 if (error)
1642 return error;
1643
1644 switch (type) {
1645 case SYSLOG_ACTION_CLOSE: /* Close log */
1646 break;
1647 case SYSLOG_ACTION_OPEN: /* Open log */
1648 break;
1649 case SYSLOG_ACTION_READ: /* Read from log */
1650 if (!buf || len < 0)
1651 return -EINVAL;
1652 if (!len)
1653 return 0;
1654 if (!access_ok(buf, len))
1655 return -EFAULT;
1656
1657 error = wait_event_interruptible(log_wait,
1658 prb_read_valid(prb, read_syslog_seq_irq(), NULL));
1659 if (error)
1660 return error;
1661 error = syslog_print(buf, len);
1662 break;
1663 /* Read/clear last kernel messages */
1664 case SYSLOG_ACTION_READ_CLEAR:
1665 clear = true;
1666 fallthrough;
1667 /* Read last kernel messages */
1668 case SYSLOG_ACTION_READ_ALL:
1669 if (!buf || len < 0)
1670 return -EINVAL;
1671 if (!len)
1672 return 0;
1673 if (!access_ok(buf, len))
1674 return -EFAULT;
1675 error = syslog_print_all(buf, len, clear);
1676 break;
1677 /* Clear ring buffer */
1678 case SYSLOG_ACTION_CLEAR:
1679 syslog_clear();
1680 break;
1681 /* Disable logging to console */
1682 case SYSLOG_ACTION_CONSOLE_OFF:
1683 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1684 saved_console_loglevel = console_loglevel;
1685 console_loglevel = minimum_console_loglevel;
1686 break;
1687 /* Enable logging to console */
1688 case SYSLOG_ACTION_CONSOLE_ON:
1689 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1690 console_loglevel = saved_console_loglevel;
1691 saved_console_loglevel = LOGLEVEL_DEFAULT;
1692 }
1693 break;
1694 /* Set level of messages printed to console */
1695 case SYSLOG_ACTION_CONSOLE_LEVEL:
1696 if (len < 1 || len > 8)
1697 return -EINVAL;
1698 if (len < minimum_console_loglevel)
1699 len = minimum_console_loglevel;
1700 console_loglevel = len;
1701 /* Implicitly re-enable logging to console */
1702 saved_console_loglevel = LOGLEVEL_DEFAULT;
1703 break;
1704 /* Number of chars in the log buffer */
1705 case SYSLOG_ACTION_SIZE_UNREAD:
1706 printk_safe_enter_irq();
1707 raw_spin_lock(&syslog_lock);
1708 if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1709 /* No unread messages. */
1710 raw_spin_unlock(&syslog_lock);
1711 printk_safe_exit_irq();
1712 return 0;
1713 }
1714 if (info.seq != syslog_seq) {
1715 /* messages are gone, move to first one */
1716 syslog_seq = info.seq;
1717 syslog_partial = 0;
1718 }
1719 if (source == SYSLOG_FROM_PROC) {
1720 /*
1721 * Short-cut for poll(/"proc/kmsg") which simply checks
1722 * for pending data, not the size; return the count of
1723 * records, not the length.
1724 */
1725 error = prb_next_seq(prb) - syslog_seq;
1726 } else {
1727 bool time = syslog_partial ? syslog_time : printk_time;
1728 unsigned int line_count;
1729 u64 seq;
1730
1731 prb_for_each_info(syslog_seq, prb, seq, &info,
1732 &line_count) {
1733 error += get_record_print_text_size(&info, line_count,
1734 true, time);
1735 time = printk_time;
1736 }
1737 error -= syslog_partial;
1738 }
1739 raw_spin_unlock(&syslog_lock);
1740 printk_safe_exit_irq();
1741 break;
1742 /* Size of the log buffer */
1743 case SYSLOG_ACTION_SIZE_BUFFER:
1744 error = log_buf_len;
1745 break;
1746 default:
1747 error = -EINVAL;
1748 break;
1749 }
1750
1751 return error;
1752}
1753
1754SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1755{
1756 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1757}
1758
1759/*
1760 * Special console_lock variants that help to reduce the risk of soft-lockups.
1761 * They allow to pass console_lock to another printk() call using a busy wait.
1762 */
1763
1764#ifdef CONFIG_LOCKDEP
1765static struct lockdep_map console_owner_dep_map = {
1766 .name = "console_owner"
1767};
1768#endif
1769
1770static DEFINE_RAW_SPINLOCK(console_owner_lock);
1771static struct task_struct *console_owner;
1772static bool console_waiter;
1773
1774/**
1775 * console_lock_spinning_enable - mark beginning of code where another
1776 * thread might safely busy wait
1777 *
1778 * This basically converts console_lock into a spinlock. This marks
1779 * the section where the console_lock owner can not sleep, because
1780 * there may be a waiter spinning (like a spinlock). Also it must be
1781 * ready to hand over the lock at the end of the section.
1782 */
1783static void console_lock_spinning_enable(void)
1784{
1785 raw_spin_lock(&console_owner_lock);
1786 console_owner = current;
1787 raw_spin_unlock(&console_owner_lock);
1788
1789 /* The waiter may spin on us after setting console_owner */
1790 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1791}
1792
1793/**
1794 * console_lock_spinning_disable_and_check - mark end of code where another
1795 * thread was able to busy wait and check if there is a waiter
1796 *
1797 * This is called at the end of the section where spinning is allowed.
1798 * It has two functions. First, it is a signal that it is no longer
1799 * safe to start busy waiting for the lock. Second, it checks if
1800 * there is a busy waiter and passes the lock rights to her.
1801 *
1802 * Important: Callers lose the lock if there was a busy waiter.
1803 * They must not touch items synchronized by console_lock
1804 * in this case.
1805 *
1806 * Return: 1 if the lock rights were passed, 0 otherwise.
1807 */
1808static int console_lock_spinning_disable_and_check(void)
1809{
1810 int waiter;
1811
1812 raw_spin_lock(&console_owner_lock);
1813 waiter = READ_ONCE(console_waiter);
1814 console_owner = NULL;
1815 raw_spin_unlock(&console_owner_lock);
1816
1817 if (!waiter) {
1818 spin_release(&console_owner_dep_map, _THIS_IP_);
1819 return 0;
1820 }
1821
1822 /* The waiter is now free to continue */
1823 WRITE_ONCE(console_waiter, false);
1824
1825 spin_release(&console_owner_dep_map, _THIS_IP_);
1826
1827 /*
1828 * Hand off console_lock to waiter. The waiter will perform
1829 * the up(). After this, the waiter is the console_lock owner.
1830 */
1831 mutex_release(&console_lock_dep_map, _THIS_IP_);
1832 return 1;
1833}
1834
1835/**
1836 * console_trylock_spinning - try to get console_lock by busy waiting
1837 *
1838 * This allows to busy wait for the console_lock when the current
1839 * owner is running in specially marked sections. It means that
1840 * the current owner is running and cannot reschedule until it
1841 * is ready to lose the lock.
1842 *
1843 * Return: 1 if we got the lock, 0 othrewise
1844 */
1845static int console_trylock_spinning(void)
1846{
1847 struct task_struct *owner = NULL;
1848 bool waiter;
1849 bool spin = false;
1850 unsigned long flags;
1851
1852 if (console_trylock())
1853 return 1;
1854
1855 printk_safe_enter_irqsave(flags);
1856
1857 raw_spin_lock(&console_owner_lock);
1858 owner = READ_ONCE(console_owner);
1859 waiter = READ_ONCE(console_waiter);
1860 if (!waiter && owner && owner != current) {
1861 WRITE_ONCE(console_waiter, true);
1862 spin = true;
1863 }
1864 raw_spin_unlock(&console_owner_lock);
1865
1866 /*
1867 * If there is an active printk() writing to the
1868 * consoles, instead of having it write our data too,
1869 * see if we can offload that load from the active
1870 * printer, and do some printing ourselves.
1871 * Go into a spin only if there isn't already a waiter
1872 * spinning, and there is an active printer, and
1873 * that active printer isn't us (recursive printk?).
1874 */
1875 if (!spin) {
1876 printk_safe_exit_irqrestore(flags);
1877 return 0;
1878 }
1879
1880 /* We spin waiting for the owner to release us */
1881 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1882 /* Owner will clear console_waiter on hand off */
1883 while (READ_ONCE(console_waiter))
1884 cpu_relax();
1885 spin_release(&console_owner_dep_map, _THIS_IP_);
1886
1887 printk_safe_exit_irqrestore(flags);
1888 /*
1889 * The owner passed the console lock to us.
1890 * Since we did not spin on console lock, annotate
1891 * this as a trylock. Otherwise lockdep will
1892 * complain.
1893 */
1894 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1895
1896 return 1;
1897}
1898
1899/*
1900 * Call the console drivers, asking them to write out
1901 * log_buf[start] to log_buf[end - 1].
1902 * The console_lock must be held.
1903 */
1904static void call_console_drivers(const char *ext_text, size_t ext_len,
1905 const char *text, size_t len)
1906{
1907 static char dropped_text[64];
1908 size_t dropped_len = 0;
1909 struct console *con;
1910
1911 trace_console_rcuidle(text, len);
1912
1913 if (!console_drivers)
1914 return;
1915
1916 if (console_dropped) {
1917 dropped_len = snprintf(dropped_text, sizeof(dropped_text),
1918 "** %lu printk messages dropped **\n",
1919 console_dropped);
1920 console_dropped = 0;
1921 }
1922
1923 for_each_console(con) {
1924 if (exclusive_console && con != exclusive_console)
1925 continue;
1926 if (!(con->flags & CON_ENABLED))
1927 continue;
1928 if (!con->write)
1929 continue;
1930 if (!cpu_online(smp_processor_id()) &&
1931 !(con->flags & CON_ANYTIME))
1932 continue;
1933 if (con->flags & CON_EXTENDED)
1934 con->write(con, ext_text, ext_len);
1935 else {
1936 if (dropped_len)
1937 con->write(con, dropped_text, dropped_len);
1938 con->write(con, text, len);
1939 }
1940 }
1941}
1942
1943int printk_delay_msec __read_mostly;
1944
1945static inline void printk_delay(void)
1946{
1947 if (unlikely(printk_delay_msec)) {
1948 int m = printk_delay_msec;
1949
1950 while (m--) {
1951 mdelay(1);
1952 touch_nmi_watchdog();
1953 }
1954 }
1955}
1956
1957static inline u32 printk_caller_id(void)
1958{
1959 return in_task() ? task_pid_nr(current) :
1960 0x80000000 + raw_smp_processor_id();
1961}
1962
1963/**
1964 * parse_prefix - Parse level and control flags.
1965 *
1966 * @text: The terminated text message.
1967 * @level: A pointer to the current level value, will be updated.
1968 * @lflags: A pointer to the current log flags, will be updated.
1969 *
1970 * @level may be NULL if the caller is not interested in the parsed value.
1971 * Otherwise the variable pointed to by @level must be set to
1972 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
1973 *
1974 * @lflags may be NULL if the caller is not interested in the parsed value.
1975 * Otherwise the variable pointed to by @lflags will be OR'd with the parsed
1976 * value.
1977 *
1978 * Return: The length of the parsed level and control flags.
1979 */
1980static u16 parse_prefix(char *text, int *level, enum log_flags *lflags)
1981{
1982 u16 prefix_len = 0;
1983 int kern_level;
1984
1985 while (*text) {
1986 kern_level = printk_get_level(text);
1987 if (!kern_level)
1988 break;
1989
1990 switch (kern_level) {
1991 case '0' ... '7':
1992 if (level && *level == LOGLEVEL_DEFAULT)
1993 *level = kern_level - '0';
1994 break;
1995 case 'c': /* KERN_CONT */
1996 if (lflags)
1997 *lflags |= LOG_CONT;
1998 }
1999
2000 prefix_len += 2;
2001 text += 2;
2002 }
2003
2004 return prefix_len;
2005}
2006
2007static u16 printk_sprint(char *text, u16 size, int facility, enum log_flags *lflags,
2008 const char *fmt, va_list args)
2009{
2010 u16 text_len;
2011
2012 text_len = vscnprintf(text, size, fmt, args);
2013
2014 /* Mark and strip a trailing newline. */
2015 if (text_len && text[text_len - 1] == '\n') {
2016 text_len--;
2017 *lflags |= LOG_NEWLINE;
2018 }
2019
2020 /* Strip log level and control flags. */
2021 if (facility == 0) {
2022 u16 prefix_len;
2023
2024 prefix_len = parse_prefix(text, NULL, NULL);
2025 if (prefix_len) {
2026 text_len -= prefix_len;
2027 memmove(text, text + prefix_len, text_len);
2028 }
2029 }
2030
2031 return text_len;
2032}
2033
2034__printf(4, 0)
2035int vprintk_store(int facility, int level,
2036 const struct dev_printk_info *dev_info,
2037 const char *fmt, va_list args)
2038{
2039 const u32 caller_id = printk_caller_id();
2040 struct prb_reserved_entry e;
2041 enum log_flags lflags = 0;
2042 struct printk_record r;
2043 u16 trunc_msg_len = 0;
2044 char prefix_buf[8];
2045 u16 reserve_size;
2046 va_list args2;
2047 u16 text_len;
2048 u64 ts_nsec;
2049
2050 /*
2051 * Since the duration of printk() can vary depending on the message
2052 * and state of the ringbuffer, grab the timestamp now so that it is
2053 * close to the call of printk(). This provides a more deterministic
2054 * timestamp with respect to the caller.
2055 */
2056 ts_nsec = local_clock();
2057
2058 /*
2059 * The sprintf needs to come first since the syslog prefix might be
2060 * passed in as a parameter. An extra byte must be reserved so that
2061 * later the vscnprintf() into the reserved buffer has room for the
2062 * terminating '\0', which is not counted by vsnprintf().
2063 */
2064 va_copy(args2, args);
2065 reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2066 va_end(args2);
2067
2068 if (reserve_size > LOG_LINE_MAX)
2069 reserve_size = LOG_LINE_MAX;
2070
2071 /* Extract log level or control flags. */
2072 if (facility == 0)
2073 parse_prefix(&prefix_buf[0], &level, &lflags);
2074
2075 if (level == LOGLEVEL_DEFAULT)
2076 level = default_message_loglevel;
2077
2078 if (dev_info)
2079 lflags |= LOG_NEWLINE;
2080
2081 if (lflags & LOG_CONT) {
2082 prb_rec_init_wr(&r, reserve_size);
2083 if (prb_reserve_in_last(&e, prb, &r, caller_id, LOG_LINE_MAX)) {
2084 text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2085 facility, &lflags, fmt, args);
2086 r.info->text_len += text_len;
2087
2088 if (lflags & LOG_NEWLINE) {
2089 r.info->flags |= LOG_NEWLINE;
2090 prb_final_commit(&e);
2091 } else {
2092 prb_commit(&e);
2093 }
2094
2095 return text_len;
2096 }
2097 }
2098
2099 /*
2100 * Explicitly initialize the record before every prb_reserve() call.
2101 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2102 * structure when they fail.
2103 */
2104 prb_rec_init_wr(&r, reserve_size);
2105 if (!prb_reserve(&e, prb, &r)) {
2106 /* truncate the message if it is too long for empty buffer */
2107 truncate_msg(&reserve_size, &trunc_msg_len);
2108
2109 prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2110 if (!prb_reserve(&e, prb, &r))
2111 return 0;
2112 }
2113
2114 /* fill message */
2115 text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &lflags, fmt, args);
2116 if (trunc_msg_len)
2117 memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2118 r.info->text_len = text_len + trunc_msg_len;
2119 r.info->facility = facility;
2120 r.info->level = level & 7;
2121 r.info->flags = lflags & 0x1f;
2122 r.info->ts_nsec = ts_nsec;
2123 r.info->caller_id = caller_id;
2124 if (dev_info)
2125 memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2126
2127 /* A message without a trailing newline can be continued. */
2128 if (!(lflags & LOG_NEWLINE))
2129 prb_commit(&e);
2130 else
2131 prb_final_commit(&e);
2132
2133 return (text_len + trunc_msg_len);
2134}
2135
2136asmlinkage int vprintk_emit(int facility, int level,
2137 const struct dev_printk_info *dev_info,
2138 const char *fmt, va_list args)
2139{
2140 int printed_len;
2141 bool in_sched = false;
2142 unsigned long flags;
2143
2144 /* Suppress unimportant messages after panic happens */
2145 if (unlikely(suppress_printk))
2146 return 0;
2147
2148 if (level == LOGLEVEL_SCHED) {
2149 level = LOGLEVEL_DEFAULT;
2150 in_sched = true;
2151 }
2152
2153 boot_delay_msec(level);
2154 printk_delay();
2155
2156 printk_safe_enter_irqsave(flags);
2157 printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2158 printk_safe_exit_irqrestore(flags);
2159
2160 /* If called from the scheduler, we can not call up(). */
2161 if (!in_sched) {
2162 /*
2163 * Disable preemption to avoid being preempted while holding
2164 * console_sem which would prevent anyone from printing to
2165 * console
2166 */
2167 preempt_disable();
2168 /*
2169 * Try to acquire and then immediately release the console
2170 * semaphore. The release will print out buffers and wake up
2171 * /dev/kmsg and syslog() users.
2172 */
2173 if (console_trylock_spinning())
2174 console_unlock();
2175 preempt_enable();
2176 }
2177
2178 wake_up_klogd();
2179 return printed_len;
2180}
2181EXPORT_SYMBOL(vprintk_emit);
2182
2183int vprintk_default(const char *fmt, va_list args)
2184{
2185 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2186}
2187EXPORT_SYMBOL_GPL(vprintk_default);
2188
2189/**
2190 * printk - print a kernel message
2191 * @fmt: format string
2192 *
2193 * This is printk(). It can be called from any context. We want it to work.
2194 *
2195 * We try to grab the console_lock. If we succeed, it's easy - we log the
2196 * output and call the console drivers. If we fail to get the semaphore, we
2197 * place the output into the log buffer and return. The current holder of
2198 * the console_sem will notice the new output in console_unlock(); and will
2199 * send it to the consoles before releasing the lock.
2200 *
2201 * One effect of this deferred printing is that code which calls printk() and
2202 * then changes console_loglevel may break. This is because console_loglevel
2203 * is inspected when the actual printing occurs.
2204 *
2205 * See also:
2206 * printf(3)
2207 *
2208 * See the vsnprintf() documentation for format string extensions over C99.
2209 */
2210asmlinkage __visible int printk(const char *fmt, ...)
2211{
2212 va_list args;
2213 int r;
2214
2215 va_start(args, fmt);
2216 r = vprintk(fmt, args);
2217 va_end(args);
2218
2219 return r;
2220}
2221EXPORT_SYMBOL(printk);
2222
2223#else /* CONFIG_PRINTK */
2224
2225#define CONSOLE_LOG_MAX 0
2226#define printk_time false
2227
2228#define prb_read_valid(rb, seq, r) false
2229#define prb_first_valid_seq(rb) 0
2230
2231static u64 syslog_seq;
2232static u64 console_seq;
2233static u64 exclusive_console_stop_seq;
2234static unsigned long console_dropped;
2235
2236static size_t record_print_text(const struct printk_record *r,
2237 bool syslog, bool time)
2238{
2239 return 0;
2240}
2241static ssize_t info_print_ext_header(char *buf, size_t size,
2242 struct printk_info *info)
2243{
2244 return 0;
2245}
2246static ssize_t msg_print_ext_body(char *buf, size_t size,
2247 char *text, size_t text_len,
2248 struct dev_printk_info *dev_info) { return 0; }
2249static void console_lock_spinning_enable(void) { }
2250static int console_lock_spinning_disable_and_check(void) { return 0; }
2251static void call_console_drivers(const char *ext_text, size_t ext_len,
2252 const char *text, size_t len) {}
2253static bool suppress_message_printing(int level) { return false; }
2254
2255#endif /* CONFIG_PRINTK */
2256
2257#ifdef CONFIG_EARLY_PRINTK
2258struct console *early_console;
2259
2260asmlinkage __visible void early_printk(const char *fmt, ...)
2261{
2262 va_list ap;
2263 char buf[512];
2264 int n;
2265
2266 if (!early_console)
2267 return;
2268
2269 va_start(ap, fmt);
2270 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2271 va_end(ap);
2272
2273 early_console->write(early_console, buf, n);
2274}
2275#endif
2276
2277static int __add_preferred_console(char *name, int idx, char *options,
2278 char *brl_options, bool user_specified)
2279{
2280 struct console_cmdline *c;
2281 int i;
2282
2283 /*
2284 * See if this tty is not yet registered, and
2285 * if we have a slot free.
2286 */
2287 for (i = 0, c = console_cmdline;
2288 i < MAX_CMDLINECONSOLES && c->name[0];
2289 i++, c++) {
2290 if (strcmp(c->name, name) == 0 && c->index == idx) {
2291 if (!brl_options)
2292 preferred_console = i;
2293 if (user_specified)
2294 c->user_specified = true;
2295 return 0;
2296 }
2297 }
2298 if (i == MAX_CMDLINECONSOLES)
2299 return -E2BIG;
2300 if (!brl_options)
2301 preferred_console = i;
2302 strlcpy(c->name, name, sizeof(c->name));
2303 c->options = options;
2304 c->user_specified = user_specified;
2305 braille_set_options(c, brl_options);
2306
2307 c->index = idx;
2308 return 0;
2309}
2310
2311static int __init console_msg_format_setup(char *str)
2312{
2313 if (!strcmp(str, "syslog"))
2314 console_msg_format = MSG_FORMAT_SYSLOG;
2315 if (!strcmp(str, "default"))
2316 console_msg_format = MSG_FORMAT_DEFAULT;
2317 return 1;
2318}
2319__setup("console_msg_format=", console_msg_format_setup);
2320
2321/*
2322 * Set up a console. Called via do_early_param() in init/main.c
2323 * for each "console=" parameter in the boot command line.
2324 */
2325static int __init console_setup(char *str)
2326{
2327 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2328 char *s, *options, *brl_options = NULL;
2329 int idx;
2330
2331 /*
2332 * console="" or console=null have been suggested as a way to
2333 * disable console output. Use ttynull that has been created
2334 * for exactly this purpose.
2335 */
2336 if (str[0] == 0 || strcmp(str, "null") == 0) {
2337 __add_preferred_console("ttynull", 0, NULL, NULL, true);
2338 return 1;
2339 }
2340
2341 if (_braille_console_setup(&str, &brl_options))
2342 return 1;
2343
2344 /*
2345 * Decode str into name, index, options.
2346 */
2347 if (str[0] >= '0' && str[0] <= '9') {
2348 strcpy(buf, "ttyS");
2349 strncpy(buf + 4, str, sizeof(buf) - 5);
2350 } else {
2351 strncpy(buf, str, sizeof(buf) - 1);
2352 }
2353 buf[sizeof(buf) - 1] = 0;
2354 options = strchr(str, ',');
2355 if (options)
2356 *(options++) = 0;
2357#ifdef __sparc__
2358 if (!strcmp(str, "ttya"))
2359 strcpy(buf, "ttyS0");
2360 if (!strcmp(str, "ttyb"))
2361 strcpy(buf, "ttyS1");
2362#endif
2363 for (s = buf; *s; s++)
2364 if (isdigit(*s) || *s == ',')
2365 break;
2366 idx = simple_strtoul(s, NULL, 10);
2367 *s = 0;
2368
2369 __add_preferred_console(buf, idx, options, brl_options, true);
2370 console_set_on_cmdline = 1;
2371 return 1;
2372}
2373__setup("console=", console_setup);
2374
2375/**
2376 * add_preferred_console - add a device to the list of preferred consoles.
2377 * @name: device name
2378 * @idx: device index
2379 * @options: options for this console
2380 *
2381 * The last preferred console added will be used for kernel messages
2382 * and stdin/out/err for init. Normally this is used by console_setup
2383 * above to handle user-supplied console arguments; however it can also
2384 * be used by arch-specific code either to override the user or more
2385 * commonly to provide a default console (ie from PROM variables) when
2386 * the user has not supplied one.
2387 */
2388int add_preferred_console(char *name, int idx, char *options)
2389{
2390 return __add_preferred_console(name, idx, options, NULL, false);
2391}
2392
2393bool console_suspend_enabled = true;
2394EXPORT_SYMBOL(console_suspend_enabled);
2395
2396static int __init console_suspend_disable(char *str)
2397{
2398 console_suspend_enabled = false;
2399 return 1;
2400}
2401__setup("no_console_suspend", console_suspend_disable);
2402module_param_named(console_suspend, console_suspend_enabled,
2403 bool, S_IRUGO | S_IWUSR);
2404MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2405 " and hibernate operations");
2406
2407/**
2408 * suspend_console - suspend the console subsystem
2409 *
2410 * This disables printk() while we go into suspend states
2411 */
2412void suspend_console(void)
2413{
2414 if (!console_suspend_enabled)
2415 return;
2416 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2417 console_lock();
2418 console_suspended = 1;
2419 up_console_sem();
2420}
2421
2422void resume_console(void)
2423{
2424 if (!console_suspend_enabled)
2425 return;
2426 down_console_sem();
2427 console_suspended = 0;
2428 console_unlock();
2429}
2430
2431/**
2432 * console_cpu_notify - print deferred console messages after CPU hotplug
2433 * @cpu: unused
2434 *
2435 * If printk() is called from a CPU that is not online yet, the messages
2436 * will be printed on the console only if there are CON_ANYTIME consoles.
2437 * This function is called when a new CPU comes online (or fails to come
2438 * up) or goes offline.
2439 */
2440static int console_cpu_notify(unsigned int cpu)
2441{
2442 if (!cpuhp_tasks_frozen) {
2443 /* If trylock fails, someone else is doing the printing */
2444 if (console_trylock())
2445 console_unlock();
2446 }
2447 return 0;
2448}
2449
2450/**
2451 * console_lock - lock the console system for exclusive use.
2452 *
2453 * Acquires a lock which guarantees that the caller has
2454 * exclusive access to the console system and the console_drivers list.
2455 *
2456 * Can sleep, returns nothing.
2457 */
2458void console_lock(void)
2459{
2460 might_sleep();
2461
2462 down_console_sem();
2463 if (console_suspended)
2464 return;
2465 console_locked = 1;
2466 console_may_schedule = 1;
2467}
2468EXPORT_SYMBOL(console_lock);
2469
2470/**
2471 * console_trylock - try to lock the console system for exclusive use.
2472 *
2473 * Try to acquire a lock which guarantees that the caller has exclusive
2474 * access to the console system and the console_drivers list.
2475 *
2476 * returns 1 on success, and 0 on failure to acquire the lock.
2477 */
2478int console_trylock(void)
2479{
2480 if (down_trylock_console_sem())
2481 return 0;
2482 if (console_suspended) {
2483 up_console_sem();
2484 return 0;
2485 }
2486 console_locked = 1;
2487 console_may_schedule = 0;
2488 return 1;
2489}
2490EXPORT_SYMBOL(console_trylock);
2491
2492int is_console_locked(void)
2493{
2494 return console_locked;
2495}
2496EXPORT_SYMBOL(is_console_locked);
2497
2498/*
2499 * Check if we have any console that is capable of printing while cpu is
2500 * booting or shutting down. Requires console_sem.
2501 */
2502static int have_callable_console(void)
2503{
2504 struct console *con;
2505
2506 for_each_console(con)
2507 if ((con->flags & CON_ENABLED) &&
2508 (con->flags & CON_ANYTIME))
2509 return 1;
2510
2511 return 0;
2512}
2513
2514/*
2515 * Can we actually use the console at this time on this cpu?
2516 *
2517 * Console drivers may assume that per-cpu resources have been allocated. So
2518 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2519 * call them until this CPU is officially up.
2520 */
2521static inline int can_use_console(void)
2522{
2523 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2524}
2525
2526/**
2527 * console_unlock - unlock the console system
2528 *
2529 * Releases the console_lock which the caller holds on the console system
2530 * and the console driver list.
2531 *
2532 * While the console_lock was held, console output may have been buffered
2533 * by printk(). If this is the case, console_unlock(); emits
2534 * the output prior to releasing the lock.
2535 *
2536 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2537 *
2538 * console_unlock(); may be called from any context.
2539 */
2540void console_unlock(void)
2541{
2542 static char ext_text[CONSOLE_EXT_LOG_MAX];
2543 static char text[CONSOLE_LOG_MAX];
2544 unsigned long flags;
2545 bool do_cond_resched, retry;
2546 struct printk_info info;
2547 struct printk_record r;
2548 u64 __maybe_unused next_seq;
2549
2550 if (console_suspended) {
2551 up_console_sem();
2552 return;
2553 }
2554
2555 prb_rec_init_rd(&r, &info, text, sizeof(text));
2556
2557 /*
2558 * Console drivers are called with interrupts disabled, so
2559 * @console_may_schedule should be cleared before; however, we may
2560 * end up dumping a lot of lines, for example, if called from
2561 * console registration path, and should invoke cond_resched()
2562 * between lines if allowable. Not doing so can cause a very long
2563 * scheduling stall on a slow console leading to RCU stall and
2564 * softlockup warnings which exacerbate the issue with more
2565 * messages practically incapacitating the system.
2566 *
2567 * console_trylock() is not able to detect the preemptive
2568 * context reliably. Therefore the value must be stored before
2569 * and cleared after the "again" goto label.
2570 */
2571 do_cond_resched = console_may_schedule;
2572again:
2573 console_may_schedule = 0;
2574
2575 /*
2576 * We released the console_sem lock, so we need to recheck if
2577 * cpu is online and (if not) is there at least one CON_ANYTIME
2578 * console.
2579 */
2580 if (!can_use_console()) {
2581 console_locked = 0;
2582 up_console_sem();
2583 return;
2584 }
2585
2586 for (;;) {
2587 size_t ext_len = 0;
2588 size_t len;
2589
2590 printk_safe_enter_irqsave(flags);
2591skip:
2592 if (!prb_read_valid(prb, console_seq, &r))
2593 break;
2594
2595 if (console_seq != r.info->seq) {
2596 console_dropped += r.info->seq - console_seq;
2597 console_seq = r.info->seq;
2598 }
2599
2600 if (suppress_message_printing(r.info->level)) {
2601 /*
2602 * Skip record we have buffered and already printed
2603 * directly to the console when we received it, and
2604 * record that has level above the console loglevel.
2605 */
2606 console_seq++;
2607 goto skip;
2608 }
2609
2610 /* Output to all consoles once old messages replayed. */
2611 if (unlikely(exclusive_console &&
2612 console_seq >= exclusive_console_stop_seq)) {
2613 exclusive_console = NULL;
2614 }
2615
2616 /*
2617 * Handle extended console text first because later
2618 * record_print_text() will modify the record buffer in-place.
2619 */
2620 if (nr_ext_console_drivers) {
2621 ext_len = info_print_ext_header(ext_text,
2622 sizeof(ext_text),
2623 r.info);
2624 ext_len += msg_print_ext_body(ext_text + ext_len,
2625 sizeof(ext_text) - ext_len,
2626 &r.text_buf[0],
2627 r.info->text_len,
2628 &r.info->dev_info);
2629 }
2630 len = record_print_text(&r,
2631 console_msg_format & MSG_FORMAT_SYSLOG,
2632 printk_time);
2633 console_seq++;
2634
2635 /*
2636 * While actively printing out messages, if another printk()
2637 * were to occur on another CPU, it may wait for this one to
2638 * finish. This task can not be preempted if there is a
2639 * waiter waiting to take over.
2640 */
2641 console_lock_spinning_enable();
2642
2643 stop_critical_timings(); /* don't trace print latency */
2644 call_console_drivers(ext_text, ext_len, text, len);
2645 start_critical_timings();
2646
2647 if (console_lock_spinning_disable_and_check()) {
2648 printk_safe_exit_irqrestore(flags);
2649 return;
2650 }
2651
2652 printk_safe_exit_irqrestore(flags);
2653
2654 if (do_cond_resched)
2655 cond_resched();
2656 }
2657
2658 /* Get consistent value of the next-to-be-used sequence number. */
2659 next_seq = console_seq;
2660
2661 console_locked = 0;
2662 up_console_sem();
2663
2664 /*
2665 * Someone could have filled up the buffer again, so re-check if there's
2666 * something to flush. In case we cannot trylock the console_sem again,
2667 * there's a new owner and the console_unlock() from them will do the
2668 * flush, no worries.
2669 */
2670 retry = prb_read_valid(prb, next_seq, NULL);
2671 printk_safe_exit_irqrestore(flags);
2672
2673 if (retry && console_trylock())
2674 goto again;
2675}
2676EXPORT_SYMBOL(console_unlock);
2677
2678/**
2679 * console_conditional_schedule - yield the CPU if required
2680 *
2681 * If the console code is currently allowed to sleep, and
2682 * if this CPU should yield the CPU to another task, do
2683 * so here.
2684 *
2685 * Must be called within console_lock();.
2686 */
2687void __sched console_conditional_schedule(void)
2688{
2689 if (console_may_schedule)
2690 cond_resched();
2691}
2692EXPORT_SYMBOL(console_conditional_schedule);
2693
2694void console_unblank(void)
2695{
2696 struct console *c;
2697
2698 /*
2699 * console_unblank can no longer be called in interrupt context unless
2700 * oops_in_progress is set to 1..
2701 */
2702 if (oops_in_progress) {
2703 if (down_trylock_console_sem() != 0)
2704 return;
2705 } else
2706 console_lock();
2707
2708 console_locked = 1;
2709 console_may_schedule = 0;
2710 for_each_console(c)
2711 if ((c->flags & CON_ENABLED) && c->unblank)
2712 c->unblank();
2713 console_unlock();
2714}
2715
2716/**
2717 * console_flush_on_panic - flush console content on panic
2718 * @mode: flush all messages in buffer or just the pending ones
2719 *
2720 * Immediately output all pending messages no matter what.
2721 */
2722void console_flush_on_panic(enum con_flush_mode mode)
2723{
2724 /*
2725 * If someone else is holding the console lock, trylock will fail
2726 * and may_schedule may be set. Ignore and proceed to unlock so
2727 * that messages are flushed out. As this can be called from any
2728 * context and we don't want to get preempted while flushing,
2729 * ensure may_schedule is cleared.
2730 */
2731 console_trylock();
2732 console_may_schedule = 0;
2733
2734 if (mode == CONSOLE_REPLAY_ALL) {
2735 unsigned long flags;
2736
2737 printk_safe_enter_irqsave(flags);
2738 console_seq = prb_first_valid_seq(prb);
2739 printk_safe_exit_irqrestore(flags);
2740 }
2741 console_unlock();
2742}
2743
2744/*
2745 * Return the console tty driver structure and its associated index
2746 */
2747struct tty_driver *console_device(int *index)
2748{
2749 struct console *c;
2750 struct tty_driver *driver = NULL;
2751
2752 console_lock();
2753 for_each_console(c) {
2754 if (!c->device)
2755 continue;
2756 driver = c->device(c, index);
2757 if (driver)
2758 break;
2759 }
2760 console_unlock();
2761 return driver;
2762}
2763
2764/*
2765 * Prevent further output on the passed console device so that (for example)
2766 * serial drivers can disable console output before suspending a port, and can
2767 * re-enable output afterwards.
2768 */
2769void console_stop(struct console *console)
2770{
2771 console_lock();
2772 console->flags &= ~CON_ENABLED;
2773 console_unlock();
2774}
2775EXPORT_SYMBOL(console_stop);
2776
2777void console_start(struct console *console)
2778{
2779 console_lock();
2780 console->flags |= CON_ENABLED;
2781 console_unlock();
2782}
2783EXPORT_SYMBOL(console_start);
2784
2785static int __read_mostly keep_bootcon;
2786
2787static int __init keep_bootcon_setup(char *str)
2788{
2789 keep_bootcon = 1;
2790 pr_info("debug: skip boot console de-registration.\n");
2791
2792 return 0;
2793}
2794
2795early_param("keep_bootcon", keep_bootcon_setup);
2796
2797/*
2798 * This is called by register_console() to try to match
2799 * the newly registered console with any of the ones selected
2800 * by either the command line or add_preferred_console() and
2801 * setup/enable it.
2802 *
2803 * Care need to be taken with consoles that are statically
2804 * enabled such as netconsole
2805 */
2806static int try_enable_new_console(struct console *newcon, bool user_specified)
2807{
2808 struct console_cmdline *c;
2809 int i, err;
2810
2811 for (i = 0, c = console_cmdline;
2812 i < MAX_CMDLINECONSOLES && c->name[0];
2813 i++, c++) {
2814 if (c->user_specified != user_specified)
2815 continue;
2816 if (!newcon->match ||
2817 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2818 /* default matching */
2819 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2820 if (strcmp(c->name, newcon->name) != 0)
2821 continue;
2822 if (newcon->index >= 0 &&
2823 newcon->index != c->index)
2824 continue;
2825 if (newcon->index < 0)
2826 newcon->index = c->index;
2827
2828 if (_braille_register_console(newcon, c))
2829 return 0;
2830
2831 if (newcon->setup &&
2832 (err = newcon->setup(newcon, c->options)) != 0)
2833 return err;
2834 }
2835 newcon->flags |= CON_ENABLED;
2836 if (i == preferred_console) {
2837 newcon->flags |= CON_CONSDEV;
2838 has_preferred_console = true;
2839 }
2840 return 0;
2841 }
2842
2843 /*
2844 * Some consoles, such as pstore and netconsole, can be enabled even
2845 * without matching. Accept the pre-enabled consoles only when match()
2846 * and setup() had a chance to be called.
2847 */
2848 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
2849 return 0;
2850
2851 return -ENOENT;
2852}
2853
2854/*
2855 * The console driver calls this routine during kernel initialization
2856 * to register the console printing procedure with printk() and to
2857 * print any messages that were printed by the kernel before the
2858 * console driver was initialized.
2859 *
2860 * This can happen pretty early during the boot process (because of
2861 * early_printk) - sometimes before setup_arch() completes - be careful
2862 * of what kernel features are used - they may not be initialised yet.
2863 *
2864 * There are two types of consoles - bootconsoles (early_printk) and
2865 * "real" consoles (everything which is not a bootconsole) which are
2866 * handled differently.
2867 * - Any number of bootconsoles can be registered at any time.
2868 * - As soon as a "real" console is registered, all bootconsoles
2869 * will be unregistered automatically.
2870 * - Once a "real" console is registered, any attempt to register a
2871 * bootconsoles will be rejected
2872 */
2873void register_console(struct console *newcon)
2874{
2875 unsigned long flags;
2876 struct console *bcon = NULL;
2877 int err;
2878
2879 for_each_console(bcon) {
2880 if (WARN(bcon == newcon, "console '%s%d' already registered\n",
2881 bcon->name, bcon->index))
2882 return;
2883 }
2884
2885 /*
2886 * before we register a new CON_BOOT console, make sure we don't
2887 * already have a valid console
2888 */
2889 if (newcon->flags & CON_BOOT) {
2890 for_each_console(bcon) {
2891 if (!(bcon->flags & CON_BOOT)) {
2892 pr_info("Too late to register bootconsole %s%d\n",
2893 newcon->name, newcon->index);
2894 return;
2895 }
2896 }
2897 }
2898
2899 if (console_drivers && console_drivers->flags & CON_BOOT)
2900 bcon = console_drivers;
2901
2902 if (!has_preferred_console || bcon || !console_drivers)
2903 has_preferred_console = preferred_console >= 0;
2904
2905 /*
2906 * See if we want to use this console driver. If we
2907 * didn't select a console we take the first one
2908 * that registers here.
2909 */
2910 if (!has_preferred_console) {
2911 if (newcon->index < 0)
2912 newcon->index = 0;
2913 if (newcon->setup == NULL ||
2914 newcon->setup(newcon, NULL) == 0) {
2915 newcon->flags |= CON_ENABLED;
2916 if (newcon->device) {
2917 newcon->flags |= CON_CONSDEV;
2918 has_preferred_console = true;
2919 }
2920 }
2921 }
2922
2923 /* See if this console matches one we selected on the command line */
2924 err = try_enable_new_console(newcon, true);
2925
2926 /* If not, try to match against the platform default(s) */
2927 if (err == -ENOENT)
2928 err = try_enable_new_console(newcon, false);
2929
2930 /* printk() messages are not printed to the Braille console. */
2931 if (err || newcon->flags & CON_BRL)
2932 return;
2933
2934 /*
2935 * If we have a bootconsole, and are switching to a real console,
2936 * don't print everything out again, since when the boot console, and
2937 * the real console are the same physical device, it's annoying to
2938 * see the beginning boot messages twice
2939 */
2940 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2941 newcon->flags &= ~CON_PRINTBUFFER;
2942
2943 /*
2944 * Put this console in the list - keep the
2945 * preferred driver at the head of the list.
2946 */
2947 console_lock();
2948 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2949 newcon->next = console_drivers;
2950 console_drivers = newcon;
2951 if (newcon->next)
2952 newcon->next->flags &= ~CON_CONSDEV;
2953 /* Ensure this flag is always set for the head of the list */
2954 newcon->flags |= CON_CONSDEV;
2955 } else {
2956 newcon->next = console_drivers->next;
2957 console_drivers->next = newcon;
2958 }
2959
2960 if (newcon->flags & CON_EXTENDED)
2961 nr_ext_console_drivers++;
2962
2963 if (newcon->flags & CON_PRINTBUFFER) {
2964 /*
2965 * console_unlock(); will print out the buffered messages
2966 * for us.
2967 *
2968 * We're about to replay the log buffer. Only do this to the
2969 * just-registered console to avoid excessive message spam to
2970 * the already-registered consoles.
2971 *
2972 * Set exclusive_console with disabled interrupts to reduce
2973 * race window with eventual console_flush_on_panic() that
2974 * ignores console_lock.
2975 */
2976 exclusive_console = newcon;
2977 exclusive_console_stop_seq = console_seq;
2978
2979 /* Get a consistent copy of @syslog_seq. */
2980 raw_spin_lock_irqsave(&syslog_lock, flags);
2981 console_seq = syslog_seq;
2982 raw_spin_unlock_irqrestore(&syslog_lock, flags);
2983 }
2984 console_unlock();
2985 console_sysfs_notify();
2986
2987 /*
2988 * By unregistering the bootconsoles after we enable the real console
2989 * we get the "console xxx enabled" message on all the consoles -
2990 * boot consoles, real consoles, etc - this is to ensure that end
2991 * users know there might be something in the kernel's log buffer that
2992 * went to the bootconsole (that they do not see on the real console)
2993 */
2994 pr_info("%sconsole [%s%d] enabled\n",
2995 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2996 newcon->name, newcon->index);
2997 if (bcon &&
2998 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2999 !keep_bootcon) {
3000 /* We need to iterate through all boot consoles, to make
3001 * sure we print everything out, before we unregister them.
3002 */
3003 for_each_console(bcon)
3004 if (bcon->flags & CON_BOOT)
3005 unregister_console(bcon);
3006 }
3007}
3008EXPORT_SYMBOL(register_console);
3009
3010int unregister_console(struct console *console)
3011{
3012 struct console *con;
3013 int res;
3014
3015 pr_info("%sconsole [%s%d] disabled\n",
3016 (console->flags & CON_BOOT) ? "boot" : "" ,
3017 console->name, console->index);
3018
3019 res = _braille_unregister_console(console);
3020 if (res < 0)
3021 return res;
3022 if (res > 0)
3023 return 0;
3024
3025 res = -ENODEV;
3026 console_lock();
3027 if (console_drivers == console) {
3028 console_drivers=console->next;
3029 res = 0;
3030 } else {
3031 for_each_console(con) {
3032 if (con->next == console) {
3033 con->next = console->next;
3034 res = 0;
3035 break;
3036 }
3037 }
3038 }
3039
3040 if (res)
3041 goto out_disable_unlock;
3042
3043 if (console->flags & CON_EXTENDED)
3044 nr_ext_console_drivers--;
3045
3046 /*
3047 * If this isn't the last console and it has CON_CONSDEV set, we
3048 * need to set it on the next preferred console.
3049 */
3050 if (console_drivers != NULL && console->flags & CON_CONSDEV)
3051 console_drivers->flags |= CON_CONSDEV;
3052
3053 console->flags &= ~CON_ENABLED;
3054 console_unlock();
3055 console_sysfs_notify();
3056
3057 if (console->exit)
3058 res = console->exit(console);
3059
3060 return res;
3061
3062out_disable_unlock:
3063 console->flags &= ~CON_ENABLED;
3064 console_unlock();
3065
3066 return res;
3067}
3068EXPORT_SYMBOL(unregister_console);
3069
3070/*
3071 * Initialize the console device. This is called *early*, so
3072 * we can't necessarily depend on lots of kernel help here.
3073 * Just do some early initializations, and do the complex setup
3074 * later.
3075 */
3076void __init console_init(void)
3077{
3078 int ret;
3079 initcall_t call;
3080 initcall_entry_t *ce;
3081
3082 /* Setup the default TTY line discipline. */
3083 n_tty_init();
3084
3085 /*
3086 * set up the console device so that later boot sequences can
3087 * inform about problems etc..
3088 */
3089 ce = __con_initcall_start;
3090 trace_initcall_level("console");
3091 while (ce < __con_initcall_end) {
3092 call = initcall_from_entry(ce);
3093 trace_initcall_start(call);
3094 ret = call();
3095 trace_initcall_finish(call, ret);
3096 ce++;
3097 }
3098}
3099
3100/*
3101 * Some boot consoles access data that is in the init section and which will
3102 * be discarded after the initcalls have been run. To make sure that no code
3103 * will access this data, unregister the boot consoles in a late initcall.
3104 *
3105 * If for some reason, such as deferred probe or the driver being a loadable
3106 * module, the real console hasn't registered yet at this point, there will
3107 * be a brief interval in which no messages are logged to the console, which
3108 * makes it difficult to diagnose problems that occur during this time.
3109 *
3110 * To mitigate this problem somewhat, only unregister consoles whose memory
3111 * intersects with the init section. Note that all other boot consoles will
3112 * get unregistered when the real preferred console is registered.
3113 */
3114static int __init printk_late_init(void)
3115{
3116 struct console *con;
3117 int ret;
3118
3119 for_each_console(con) {
3120 if (!(con->flags & CON_BOOT))
3121 continue;
3122
3123 /* Check addresses that might be used for enabled consoles. */
3124 if (init_section_intersects(con, sizeof(*con)) ||
3125 init_section_contains(con->write, 0) ||
3126 init_section_contains(con->read, 0) ||
3127 init_section_contains(con->device, 0) ||
3128 init_section_contains(con->unblank, 0) ||
3129 init_section_contains(con->data, 0)) {
3130 /*
3131 * Please, consider moving the reported consoles out
3132 * of the init section.
3133 */
3134 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3135 con->name, con->index);
3136 unregister_console(con);
3137 }
3138 }
3139 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3140 console_cpu_notify);
3141 WARN_ON(ret < 0);
3142 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3143 console_cpu_notify, NULL);
3144 WARN_ON(ret < 0);
3145 return 0;
3146}
3147late_initcall(printk_late_init);
3148
3149#if defined CONFIG_PRINTK
3150/*
3151 * Delayed printk version, for scheduler-internal messages:
3152 */
3153#define PRINTK_PENDING_WAKEUP 0x01
3154#define PRINTK_PENDING_OUTPUT 0x02
3155
3156static DEFINE_PER_CPU(int, printk_pending);
3157
3158static void wake_up_klogd_work_func(struct irq_work *irq_work)
3159{
3160 int pending = __this_cpu_xchg(printk_pending, 0);
3161
3162 if (pending & PRINTK_PENDING_OUTPUT) {
3163 /* If trylock fails, someone else is doing the printing */
3164 if (console_trylock())
3165 console_unlock();
3166 }
3167
3168 if (pending & PRINTK_PENDING_WAKEUP)
3169 wake_up_interruptible(&log_wait);
3170}
3171
3172static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3173 IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3174
3175void wake_up_klogd(void)
3176{
3177 if (!printk_percpu_data_ready())
3178 return;
3179
3180 preempt_disable();
3181 if (waitqueue_active(&log_wait)) {
3182 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
3183 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3184 }
3185 preempt_enable();
3186}
3187
3188void defer_console_output(void)
3189{
3190 if (!printk_percpu_data_ready())
3191 return;
3192
3193 preempt_disable();
3194 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
3195 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3196 preempt_enable();
3197}
3198
3199int vprintk_deferred(const char *fmt, va_list args)
3200{
3201 int r;
3202
3203 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3204 defer_console_output();
3205
3206 return r;
3207}
3208
3209int printk_deferred(const char *fmt, ...)
3210{
3211 va_list args;
3212 int r;
3213
3214 va_start(args, fmt);
3215 r = vprintk_deferred(fmt, args);
3216 va_end(args);
3217
3218 return r;
3219}
3220
3221/*
3222 * printk rate limiting, lifted from the networking subsystem.
3223 *
3224 * This enforces a rate limit: not more than 10 kernel messages
3225 * every 5s to make a denial-of-service attack impossible.
3226 */
3227DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3228
3229int __printk_ratelimit(const char *func)
3230{
3231 return ___ratelimit(&printk_ratelimit_state, func);
3232}
3233EXPORT_SYMBOL(__printk_ratelimit);
3234
3235/**
3236 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3237 * @caller_jiffies: pointer to caller's state
3238 * @interval_msecs: minimum interval between prints
3239 *
3240 * printk_timed_ratelimit() returns true if more than @interval_msecs
3241 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3242 * returned true.
3243 */
3244bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3245 unsigned int interval_msecs)
3246{
3247 unsigned long elapsed = jiffies - *caller_jiffies;
3248
3249 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3250 return false;
3251
3252 *caller_jiffies = jiffies;
3253 return true;
3254}
3255EXPORT_SYMBOL(printk_timed_ratelimit);
3256
3257static DEFINE_SPINLOCK(dump_list_lock);
3258static LIST_HEAD(dump_list);
3259
3260/**
3261 * kmsg_dump_register - register a kernel log dumper.
3262 * @dumper: pointer to the kmsg_dumper structure
3263 *
3264 * Adds a kernel log dumper to the system. The dump callback in the
3265 * structure will be called when the kernel oopses or panics and must be
3266 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3267 */
3268int kmsg_dump_register(struct kmsg_dumper *dumper)
3269{
3270 unsigned long flags;
3271 int err = -EBUSY;
3272
3273 /* The dump callback needs to be set */
3274 if (!dumper->dump)
3275 return -EINVAL;
3276
3277 spin_lock_irqsave(&dump_list_lock, flags);
3278 /* Don't allow registering multiple times */
3279 if (!dumper->registered) {
3280 dumper->registered = 1;
3281 list_add_tail_rcu(&dumper->list, &dump_list);
3282 err = 0;
3283 }
3284 spin_unlock_irqrestore(&dump_list_lock, flags);
3285
3286 return err;
3287}
3288EXPORT_SYMBOL_GPL(kmsg_dump_register);
3289
3290/**
3291 * kmsg_dump_unregister - unregister a kmsg dumper.
3292 * @dumper: pointer to the kmsg_dumper structure
3293 *
3294 * Removes a dump device from the system. Returns zero on success and
3295 * %-EINVAL otherwise.
3296 */
3297int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3298{
3299 unsigned long flags;
3300 int err = -EINVAL;
3301
3302 spin_lock_irqsave(&dump_list_lock, flags);
3303 if (dumper->registered) {
3304 dumper->registered = 0;
3305 list_del_rcu(&dumper->list);
3306 err = 0;
3307 }
3308 spin_unlock_irqrestore(&dump_list_lock, flags);
3309 synchronize_rcu();
3310
3311 return err;
3312}
3313EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3314
3315static bool always_kmsg_dump;
3316module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3317
3318const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
3319{
3320 switch (reason) {
3321 case KMSG_DUMP_PANIC:
3322 return "Panic";
3323 case KMSG_DUMP_OOPS:
3324 return "Oops";
3325 case KMSG_DUMP_EMERG:
3326 return "Emergency";
3327 case KMSG_DUMP_SHUTDOWN:
3328 return "Shutdown";
3329 default:
3330 return "Unknown";
3331 }
3332}
3333EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
3334
3335/**
3336 * kmsg_dump - dump kernel log to kernel message dumpers.
3337 * @reason: the reason (oops, panic etc) for dumping
3338 *
3339 * Call each of the registered dumper's dump() callback, which can
3340 * retrieve the kmsg records with kmsg_dump_get_line() or
3341 * kmsg_dump_get_buffer().
3342 */
3343void kmsg_dump(enum kmsg_dump_reason reason)
3344{
3345 struct kmsg_dumper *dumper;
3346
3347 rcu_read_lock();
3348 list_for_each_entry_rcu(dumper, &dump_list, list) {
3349 enum kmsg_dump_reason max_reason = dumper->max_reason;
3350
3351 /*
3352 * If client has not provided a specific max_reason, default
3353 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
3354 */
3355 if (max_reason == KMSG_DUMP_UNDEF) {
3356 max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
3357 KMSG_DUMP_OOPS;
3358 }
3359 if (reason > max_reason)
3360 continue;
3361
3362 /* invoke dumper which will iterate over records */
3363 dumper->dump(dumper, reason);
3364 }
3365 rcu_read_unlock();
3366}
3367
3368/**
3369 * kmsg_dump_get_line - retrieve one kmsg log line
3370 * @iter: kmsg dump iterator
3371 * @syslog: include the "<4>" prefixes
3372 * @line: buffer to copy the line to
3373 * @size: maximum size of the buffer
3374 * @len: length of line placed into buffer
3375 *
3376 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3377 * record, and copy one record into the provided buffer.
3378 *
3379 * Consecutive calls will return the next available record moving
3380 * towards the end of the buffer with the youngest messages.
3381 *
3382 * A return value of FALSE indicates that there are no more records to
3383 * read.
3384 */
3385bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
3386 char *line, size_t size, size_t *len)
3387{
3388 u64 min_seq = latched_seq_read_nolock(&clear_seq);
3389 struct printk_info info;
3390 unsigned int line_count;
3391 struct printk_record r;
3392 unsigned long flags;
3393 size_t l = 0;
3394 bool ret = false;
3395
3396 if (iter->cur_seq < min_seq)
3397 iter->cur_seq = min_seq;
3398
3399 printk_safe_enter_irqsave(flags);
3400 prb_rec_init_rd(&r, &info, line, size);
3401
3402 /* Read text or count text lines? */
3403 if (line) {
3404 if (!prb_read_valid(prb, iter->cur_seq, &r))
3405 goto out;
3406 l = record_print_text(&r, syslog, printk_time);
3407 } else {
3408 if (!prb_read_valid_info(prb, iter->cur_seq,
3409 &info, &line_count)) {
3410 goto out;
3411 }
3412 l = get_record_print_text_size(&info, line_count, syslog,
3413 printk_time);
3414
3415 }
3416
3417 iter->cur_seq = r.info->seq + 1;
3418 ret = true;
3419out:
3420 printk_safe_exit_irqrestore(flags);
3421 if (len)
3422 *len = l;
3423 return ret;
3424}
3425EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3426
3427/**
3428 * kmsg_dump_get_buffer - copy kmsg log lines
3429 * @iter: kmsg dump iterator
3430 * @syslog: include the "<4>" prefixes
3431 * @buf: buffer to copy the line to
3432 * @size: maximum size of the buffer
3433 * @len_out: length of line placed into buffer
3434 *
3435 * Start at the end of the kmsg buffer and fill the provided buffer
3436 * with as many of the *youngest* kmsg records that fit into it.
3437 * If the buffer is large enough, all available kmsg records will be
3438 * copied with a single call.
3439 *
3440 * Consecutive calls will fill the buffer with the next block of
3441 * available older records, not including the earlier retrieved ones.
3442 *
3443 * A return value of FALSE indicates that there are no more records to
3444 * read.
3445 */
3446bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
3447 char *buf, size_t size, size_t *len_out)
3448{
3449 u64 min_seq = latched_seq_read_nolock(&clear_seq);
3450 struct printk_info info;
3451 struct printk_record r;
3452 unsigned long flags;
3453 u64 seq;
3454 u64 next_seq;
3455 size_t len = 0;
3456 bool ret = false;
3457 bool time = printk_time;
3458
3459 if (!buf || !size)
3460 goto out;
3461
3462 if (iter->cur_seq < min_seq)
3463 iter->cur_seq = min_seq;
3464
3465 printk_safe_enter_irqsave(flags);
3466 if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
3467 if (info.seq != iter->cur_seq) {
3468 /* messages are gone, move to first available one */
3469 iter->cur_seq = info.seq;
3470 }
3471 }
3472
3473 /* last entry */
3474 if (iter->cur_seq >= iter->next_seq) {
3475 printk_safe_exit_irqrestore(flags);
3476 goto out;
3477 }
3478
3479 /*
3480 * Find first record that fits, including all following records,
3481 * into the user-provided buffer for this dump. Pass in size-1
3482 * because this function (by way of record_print_text()) will
3483 * not write more than size-1 bytes of text into @buf.
3484 */
3485 seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
3486 size - 1, syslog, time);
3487
3488 /*
3489 * Next kmsg_dump_get_buffer() invocation will dump block of
3490 * older records stored right before this one.
3491 */
3492 next_seq = seq;
3493
3494 prb_rec_init_rd(&r, &info, buf, size);
3495
3496 len = 0;
3497 prb_for_each_record(seq, prb, seq, &r) {
3498 if (r.info->seq >= iter->next_seq)
3499 break;
3500
3501 len += record_print_text(&r, syslog, time);
3502
3503 /* Adjust record to store to remaining buffer space. */
3504 prb_rec_init_rd(&r, &info, buf + len, size - len);
3505 }
3506
3507 iter->next_seq = next_seq;
3508 ret = true;
3509 printk_safe_exit_irqrestore(flags);
3510out:
3511 if (len_out)
3512 *len_out = len;
3513 return ret;
3514}
3515EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3516
3517/**
3518 * kmsg_dump_rewind - reset the iterator
3519 * @iter: kmsg dump iterator
3520 *
3521 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3522 * kmsg_dump_get_buffer() can be called again and used multiple
3523 * times within the same dumper.dump() callback.
3524 */
3525void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
3526{
3527 unsigned long flags;
3528
3529 printk_safe_enter_irqsave(flags);
3530 iter->cur_seq = latched_seq_read_nolock(&clear_seq);
3531 iter->next_seq = prb_next_seq(prb);
3532 printk_safe_exit_irqrestore(flags);
3533}
3534EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3535
3536#endif
3537
3538#ifdef CONFIG_SMP
3539static atomic_t printk_cpulock_owner = ATOMIC_INIT(-1);
3540static atomic_t printk_cpulock_nested = ATOMIC_INIT(0);
3541
3542/**
3543 * __printk_wait_on_cpu_lock() - Busy wait until the printk cpu-reentrant
3544 * spinning lock is not owned by any CPU.
3545 *
3546 * Context: Any context.
3547 */
3548void __printk_wait_on_cpu_lock(void)
3549{
3550 do {
3551 cpu_relax();
3552 } while (atomic_read(&printk_cpulock_owner) != -1);
3553}
3554EXPORT_SYMBOL(__printk_wait_on_cpu_lock);
3555
3556/**
3557 * __printk_cpu_trylock() - Try to acquire the printk cpu-reentrant
3558 * spinning lock.
3559 *
3560 * If no processor has the lock, the calling processor takes the lock and
3561 * becomes the owner. If the calling processor is already the owner of the
3562 * lock, this function succeeds immediately.
3563 *
3564 * Context: Any context. Expects interrupts to be disabled.
3565 * Return: 1 on success, otherwise 0.
3566 */
3567int __printk_cpu_trylock(void)
3568{
3569 int cpu;
3570 int old;
3571
3572 cpu = smp_processor_id();
3573
3574 /*
3575 * Guarantee loads and stores from this CPU when it is the lock owner
3576 * are _not_ visible to the previous lock owner. This pairs with
3577 * __printk_cpu_unlock:B.
3578 *
3579 * Memory barrier involvement:
3580 *
3581 * If __printk_cpu_trylock:A reads from __printk_cpu_unlock:B, then
3582 * __printk_cpu_unlock:A can never read from __printk_cpu_trylock:B.
3583 *
3584 * Relies on:
3585 *
3586 * RELEASE from __printk_cpu_unlock:A to __printk_cpu_unlock:B
3587 * of the previous CPU
3588 * matching
3589 * ACQUIRE from __printk_cpu_trylock:A to __printk_cpu_trylock:B
3590 * of this CPU
3591 */
3592 old = atomic_cmpxchg_acquire(&printk_cpulock_owner, -1,
3593 cpu); /* LMM(__printk_cpu_trylock:A) */
3594 if (old == -1) {
3595 /*
3596 * This CPU is now the owner and begins loading/storing
3597 * data: LMM(__printk_cpu_trylock:B)
3598 */
3599 return 1;
3600
3601 } else if (old == cpu) {
3602 /* This CPU is already the owner. */
3603 atomic_inc(&printk_cpulock_nested);
3604 return 1;
3605 }
3606
3607 return 0;
3608}
3609EXPORT_SYMBOL(__printk_cpu_trylock);
3610
3611/**
3612 * __printk_cpu_unlock() - Release the printk cpu-reentrant spinning lock.
3613 *
3614 * The calling processor must be the owner of the lock.
3615 *
3616 * Context: Any context. Expects interrupts to be disabled.
3617 */
3618void __printk_cpu_unlock(void)
3619{
3620 if (atomic_read(&printk_cpulock_nested)) {
3621 atomic_dec(&printk_cpulock_nested);
3622 return;
3623 }
3624
3625 /*
3626 * This CPU is finished loading/storing data:
3627 * LMM(__printk_cpu_unlock:A)
3628 */
3629
3630 /*
3631 * Guarantee loads and stores from this CPU when it was the
3632 * lock owner are visible to the next lock owner. This pairs
3633 * with __printk_cpu_trylock:A.
3634 *
3635 * Memory barrier involvement:
3636 *
3637 * If __printk_cpu_trylock:A reads from __printk_cpu_unlock:B,
3638 * then __printk_cpu_trylock:B reads from __printk_cpu_unlock:A.
3639 *
3640 * Relies on:
3641 *
3642 * RELEASE from __printk_cpu_unlock:A to __printk_cpu_unlock:B
3643 * of this CPU
3644 * matching
3645 * ACQUIRE from __printk_cpu_trylock:A to __printk_cpu_trylock:B
3646 * of the next CPU
3647 */
3648 atomic_set_release(&printk_cpulock_owner,
3649 -1); /* LMM(__printk_cpu_unlock:B) */
3650}
3651EXPORT_SYMBOL(__printk_cpu_unlock);
3652#endif /* CONFIG_SMP */