1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/kernel/printk.c
5 * Copyright (C) 1991, 1992 Linus Torvalds
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).
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
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22 #include <linux/kernel.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/kdb.h>
39 #include <linux/ratelimit.h>
40 #include <linux/kmsg_dump.h>
41 #include <linux/syslog.h>
42 #include <linux/cpu.h>
43 #include <linux/rculist.h>
44 #include <linux/poll.h>
45 #include <linux/irq_work.h>
46 #include <linux/ctype.h>
47 #include <linux/uio.h>
48 #include <linux/sched/clock.h>
49 #include <linux/sched/debug.h>
50 #include <linux/sched/task_stack.h>
52 #include <linux/uaccess.h>
53 #include <asm/sections.h>
55 #include <trace/events/initcall.h>
56 #define CREATE_TRACE_POINTS
57 #include <trace/events/printk.h>
59 #include "console_cmdline.h"
63 int console_printk[4] = {
64 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
65 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
66 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
67 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
69 EXPORT_SYMBOL_GPL(console_printk);
71 atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
72 EXPORT_SYMBOL(ignore_console_lock_warning);
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.
79 EXPORT_SYMBOL(oops_in_progress);
82 * console_sem protects the console_drivers list, and also
83 * provides serialisation for access to the entire console
86 static DEFINE_SEMAPHORE(console_sem);
87 struct console *console_drivers;
88 EXPORT_SYMBOL_GPL(console_drivers);
91 * System may need to suppress printk message under certain
92 * circumstances, like after kernel panic happens.
94 int __read_mostly suppress_printk;
97 static struct lockdep_map console_lock_dep_map = {
98 .name = "console_lock"
102 enum devkmsg_log_bits {
103 __DEVKMSG_LOG_BIT_ON = 0,
104 __DEVKMSG_LOG_BIT_OFF,
105 __DEVKMSG_LOG_BIT_LOCK,
108 enum devkmsg_log_masks {
109 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
110 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
111 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
114 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
115 #define DEVKMSG_LOG_MASK_DEFAULT 0
117 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
119 static int __control_devkmsg(char *str)
126 len = str_has_prefix(str, "on");
128 devkmsg_log = DEVKMSG_LOG_MASK_ON;
132 len = str_has_prefix(str, "off");
134 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
138 len = str_has_prefix(str, "ratelimit");
140 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
147 static int __init control_devkmsg(char *str)
149 if (__control_devkmsg(str) < 0)
153 * Set sysctl string accordingly:
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. */
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.
167 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
171 __setup("printk.devkmsg=", control_devkmsg);
173 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
175 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
176 void __user *buffer, size_t *lenp, loff_t *ppos)
178 char old_str[DEVKMSG_STR_MAX_SIZE];
183 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
187 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
190 err = proc_dostring(table, write, buffer, lenp, ppos);
195 err = __control_devkmsg(devkmsg_log_str);
198 * Do not accept an unknown string OR a known string with
201 if (err < 0 || (err + 1 != *lenp)) {
203 /* ... and restore old setting. */
205 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
214 /* Number of registered extended console drivers. */
215 static int nr_ext_console_drivers;
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.
221 #define down_console_sem() do { \
223 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
226 static int __down_trylock_console_sem(unsigned long ip)
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.
236 printk_safe_enter_irqsave(flags);
237 lock_failed = down_trylock(&console_sem);
238 printk_safe_exit_irqrestore(flags);
242 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
245 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
247 static void __up_console_sem(unsigned long ip)
251 mutex_release(&console_lock_dep_map, ip);
253 printk_safe_enter_irqsave(flags);
255 printk_safe_exit_irqrestore(flags);
257 #define up_console_sem() __up_console_sem(_RET_IP_)
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 sempahore held).
267 static int console_locked, console_suspended;
270 * If exclusive_console is non-NULL then only this console is to be printed to.
272 static struct console *exclusive_console;
275 * Array of consoles built from command line options (console=)
278 #define MAX_CMDLINECONSOLES 8
280 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
282 static int preferred_console = -1;
283 static bool has_preferred_console;
284 int console_set_on_cmdline;
285 EXPORT_SYMBOL(console_set_on_cmdline);
287 /* Flag: console code may call schedule() */
288 static int console_may_schedule;
290 enum con_msg_format_flags {
291 MSG_FORMAT_DEFAULT = 0,
292 MSG_FORMAT_SYSLOG = (1 << 0),
295 static int console_msg_format = MSG_FORMAT_DEFAULT;
298 * The printk log buffer consists of a chain of concatenated variable
299 * length records. Every record starts with a record header, containing
300 * the overall length of the record.
302 * The heads to the first and last entry in the buffer, as well as the
303 * sequence numbers of these entries are maintained when messages are
306 * If the heads indicate available messages, the length in the header
307 * tells the start next message. A length == 0 for the next message
308 * indicates a wrap-around to the beginning of the buffer.
310 * Every record carries the monotonic timestamp in microseconds, as well as
311 * the standard userspace syslog level and syslog facility. The usual
312 * kernel messages use LOG_KERN; userspace-injected messages always carry
313 * a matching syslog facility, by default LOG_USER. The origin of every
314 * message can be reliably determined that way.
316 * The human readable log message directly follows the message header. The
317 * length of the message text is stored in the header, the stored message
320 * Optionally, a message can carry a dictionary of properties (key/value pairs),
321 * to provide userspace with a machine-readable message context.
323 * Examples for well-defined, commonly used property names are:
324 * DEVICE=b12:8 device identifier
328 * +sound:card0 subsystem:devname
329 * SUBSYSTEM=pci driver-core subsystem name
331 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
332 * follows directly after a '=' character. Every property is terminated by
333 * a '\0' character. The last property is not terminated.
335 * Example of a message structure:
336 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
337 * 0008 34 00 record is 52 bytes long
338 * 000a 0b 00 text is 11 bytes long
339 * 000c 1f 00 dictionary is 23 bytes long
340 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
341 * 0010 69 74 27 73 20 61 20 6c "it's a l"
343 * 001b 44 45 56 49 43 "DEVIC"
344 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
345 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
347 * 0032 00 00 00 padding to next message header
349 * The 'struct printk_log' buffer header must never be directly exported to
350 * userspace, it is a kernel-private implementation detail that might
351 * need to be changed in the future, when the requirements change.
353 * /dev/kmsg exports the structured data in the following line format:
354 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
356 * Users of the export format should ignore possible additional values
357 * separated by ',', and find the message after the ';' character.
359 * The optional key/value pairs are attached as continuation lines starting
360 * with a space character and terminated by a newline. All possible
361 * non-prinatable characters are escaped in the "\xff" notation.
365 LOG_NEWLINE = 2, /* text ended with a newline */
366 LOG_CONT = 8, /* text is a fragment of a continuation line */
370 u64 ts_nsec; /* timestamp in nanoseconds */
371 u16 len; /* length of entire record */
372 u16 text_len; /* length of text buffer */
373 u16 dict_len; /* length of dictionary buffer */
374 u8 facility; /* syslog facility */
375 u8 flags:5; /* internal record flags */
376 u8 level:3; /* syslog level */
377 #ifdef CONFIG_PRINTK_CALLER
378 u32 caller_id; /* thread id or processor id */
381 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
382 __packed __aligned(4)
387 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
388 * within the scheduler's rq lock. It must be released before calling
389 * console_unlock() or anything else that might wake up a process.
391 DEFINE_RAW_SPINLOCK(logbuf_lock);
394 * Helper macros to lock/unlock logbuf_lock and switch between
395 * printk-safe/unsafe modes.
397 #define logbuf_lock_irq() \
399 printk_safe_enter_irq(); \
400 raw_spin_lock(&logbuf_lock); \
403 #define logbuf_unlock_irq() \
405 raw_spin_unlock(&logbuf_lock); \
406 printk_safe_exit_irq(); \
409 #define logbuf_lock_irqsave(flags) \
411 printk_safe_enter_irqsave(flags); \
412 raw_spin_lock(&logbuf_lock); \
415 #define logbuf_unlock_irqrestore(flags) \
417 raw_spin_unlock(&logbuf_lock); \
418 printk_safe_exit_irqrestore(flags); \
422 DECLARE_WAIT_QUEUE_HEAD(log_wait);
423 /* the next printk record to read by syslog(READ) or /proc/kmsg */
424 static u64 syslog_seq;
425 static u32 syslog_idx;
426 static size_t syslog_partial;
427 static bool syslog_time;
429 /* index and sequence number of the first record stored in the buffer */
430 static u64 log_first_seq;
431 static u32 log_first_idx;
433 /* index and sequence number of the next record to store in the buffer */
434 static u64 log_next_seq;
435 static u32 log_next_idx;
437 /* the next printk record to write to the console */
438 static u64 console_seq;
439 static u32 console_idx;
440 static u64 exclusive_console_stop_seq;
442 /* the next printk record to read after the last 'clear' command */
443 static u64 clear_seq;
444 static u32 clear_idx;
446 #ifdef CONFIG_PRINTK_CALLER
447 #define PREFIX_MAX 48
449 #define PREFIX_MAX 32
451 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
453 #define LOG_LEVEL(v) ((v) & 0x07)
454 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
457 #define LOG_ALIGN __alignof__(struct printk_log)
458 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
459 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
460 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
461 static char *log_buf = __log_buf;
462 static u32 log_buf_len = __LOG_BUF_LEN;
464 /* Return log buffer address */
465 char *log_buf_addr_get(void)
470 /* Return log buffer size */
471 u32 log_buf_len_get(void)
476 /* human readable text of the record */
477 static char *log_text(const struct printk_log *msg)
479 return (char *)msg + sizeof(struct printk_log);
482 /* optional key/value pair dictionary attached to the record */
483 static char *log_dict(const struct printk_log *msg)
485 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
488 /* get record by index; idx must point to valid msg */
489 static struct printk_log *log_from_idx(u32 idx)
491 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
494 * A length == 0 record is the end of buffer marker. Wrap around and
495 * read the message at the start of the buffer.
498 return (struct printk_log *)log_buf;
502 /* get next record; idx must point to valid msg */
503 static u32 log_next(u32 idx)
505 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
507 /* length == 0 indicates the end of the buffer; wrap */
509 * A length == 0 record is the end of buffer marker. Wrap around and
510 * read the message at the start of the buffer as *this* one, and
511 * return the one after that.
514 msg = (struct printk_log *)log_buf;
517 return idx + msg->len;
521 * Check whether there is enough free space for the given message.
523 * The same values of first_idx and next_idx mean that the buffer
524 * is either empty or full.
526 * If the buffer is empty, we must respect the position of the indexes.
527 * They cannot be reset to the beginning of the buffer.
529 static int logbuf_has_space(u32 msg_size, bool empty)
533 if (log_next_idx > log_first_idx || empty)
534 free = max(log_buf_len - log_next_idx, log_first_idx);
536 free = log_first_idx - log_next_idx;
539 * We need space also for an empty header that signalizes wrapping
542 return free >= msg_size + sizeof(struct printk_log);
545 static int log_make_free_space(u32 msg_size)
547 while (log_first_seq < log_next_seq &&
548 !logbuf_has_space(msg_size, false)) {
549 /* drop old messages until we have enough contiguous space */
550 log_first_idx = log_next(log_first_idx);
554 if (clear_seq < log_first_seq) {
555 clear_seq = log_first_seq;
556 clear_idx = log_first_idx;
559 /* sequence numbers are equal, so the log buffer is empty */
560 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
566 /* compute the message size including the padding bytes */
567 static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
571 size = sizeof(struct printk_log) + text_len + dict_len;
572 *pad_len = (-size) & (LOG_ALIGN - 1);
579 * Define how much of the log buffer we could take at maximum. The value
580 * must be greater than two. Note that only half of the buffer is available
581 * when the index points to the middle.
583 #define MAX_LOG_TAKE_PART 4
584 static const char trunc_msg[] = "<truncated>";
586 static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
587 u16 *dict_len, u32 *pad_len)
590 * The message should not take the whole buffer. Otherwise, it might
591 * get removed too soon.
593 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
594 if (*text_len > max_text_len)
595 *text_len = max_text_len;
596 /* enable the warning message */
597 *trunc_msg_len = strlen(trunc_msg);
598 /* disable the "dict" completely */
600 /* compute the size again, count also the warning message */
601 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
604 /* insert record into the buffer, discard old ones, update heads */
605 static int log_store(u32 caller_id, int facility, int level,
606 enum log_flags flags, u64 ts_nsec,
607 const char *dict, u16 dict_len,
608 const char *text, u16 text_len)
610 struct printk_log *msg;
612 u16 trunc_msg_len = 0;
614 /* number of '\0' padding bytes to next message */
615 size = msg_used_size(text_len, dict_len, &pad_len);
617 if (log_make_free_space(size)) {
618 /* truncate the message if it is too long for empty buffer */
619 size = truncate_msg(&text_len, &trunc_msg_len,
620 &dict_len, &pad_len);
621 /* survive when the log buffer is too small for trunc_msg */
622 if (log_make_free_space(size))
626 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
628 * This message + an additional empty header does not fit
629 * at the end of the buffer. Add an empty header with len == 0
630 * to signify a wrap around.
632 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
637 msg = (struct printk_log *)(log_buf + log_next_idx);
638 memcpy(log_text(msg), text, text_len);
639 msg->text_len = text_len;
641 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
642 msg->text_len += trunc_msg_len;
644 memcpy(log_dict(msg), dict, dict_len);
645 msg->dict_len = dict_len;
646 msg->facility = facility;
647 msg->level = level & 7;
648 msg->flags = flags & 0x1f;
650 msg->ts_nsec = ts_nsec;
652 msg->ts_nsec = local_clock();
653 #ifdef CONFIG_PRINTK_CALLER
654 msg->caller_id = caller_id;
656 memset(log_dict(msg) + dict_len, 0, pad_len);
660 log_next_idx += msg->len;
663 return msg->text_len;
666 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
668 static int syslog_action_restricted(int type)
673 * Unless restricted, we allow "read all" and "get buffer size"
676 return type != SYSLOG_ACTION_READ_ALL &&
677 type != SYSLOG_ACTION_SIZE_BUFFER;
680 static int check_syslog_permissions(int type, int source)
683 * If this is from /proc/kmsg and we've already opened it, then we've
684 * already done the capabilities checks at open time.
686 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
689 if (syslog_action_restricted(type)) {
690 if (capable(CAP_SYSLOG))
693 * For historical reasons, accept CAP_SYS_ADMIN too, with
696 if (capable(CAP_SYS_ADMIN)) {
697 pr_warn_once("%s (%d): Attempt to access syslog with "
698 "CAP_SYS_ADMIN but no CAP_SYSLOG "
700 current->comm, task_pid_nr(current));
706 return security_syslog(type);
709 static void append_char(char **pp, char *e, char c)
715 static ssize_t msg_print_ext_header(char *buf, size_t size,
716 struct printk_log *msg, u64 seq)
718 u64 ts_usec = msg->ts_nsec;
720 #ifdef CONFIG_PRINTK_CALLER
721 u32 id = msg->caller_id;
723 snprintf(caller, sizeof(caller), ",caller=%c%u",
724 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
729 do_div(ts_usec, 1000);
731 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
732 (msg->facility << 3) | msg->level, seq, ts_usec,
733 msg->flags & LOG_CONT ? 'c' : '-', caller);
736 static ssize_t msg_print_ext_body(char *buf, size_t size,
737 char *dict, size_t dict_len,
738 char *text, size_t text_len)
740 char *p = buf, *e = buf + size;
743 /* escape non-printable characters */
744 for (i = 0; i < text_len; i++) {
745 unsigned char c = text[i];
747 if (c < ' ' || c >= 127 || c == '\\')
748 p += scnprintf(p, e - p, "\\x%02x", c);
750 append_char(&p, e, c);
752 append_char(&p, e, '\n');
757 for (i = 0; i < dict_len; i++) {
758 unsigned char c = dict[i];
761 append_char(&p, e, ' ');
766 append_char(&p, e, '\n');
771 if (c < ' ' || c >= 127 || c == '\\') {
772 p += scnprintf(p, e - p, "\\x%02x", c);
776 append_char(&p, e, c);
778 append_char(&p, e, '\n');
784 /* /dev/kmsg - userspace message inject/listen interface */
785 struct devkmsg_user {
788 struct ratelimit_state rs;
790 char buf[CONSOLE_EXT_LOG_MAX];
793 static __printf(3, 4) __cold
794 int devkmsg_emit(int facility, int level, const char *fmt, ...)
800 r = vprintk_emit(facility, level, NULL, 0, fmt, args);
806 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
809 int level = default_message_loglevel;
810 int facility = 1; /* LOG_USER */
811 struct file *file = iocb->ki_filp;
812 struct devkmsg_user *user = file->private_data;
813 size_t len = iov_iter_count(from);
816 if (!user || len > LOG_LINE_MAX)
819 /* Ignore when user logging is disabled. */
820 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
823 /* Ratelimit when not explicitly enabled. */
824 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
825 if (!___ratelimit(&user->rs, current->comm))
829 buf = kmalloc(len+1, GFP_KERNEL);
834 if (!copy_from_iter_full(buf, len, from)) {
840 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
841 * the decimal value represents 32bit, the lower 3 bit are the log
842 * level, the rest are the log facility.
844 * If no prefix or no userspace facility is specified, we
845 * enforce LOG_USER, to be able to reliably distinguish
846 * kernel-generated messages from userspace-injected ones.
849 if (line[0] == '<') {
853 u = simple_strtoul(line + 1, &endp, 10);
854 if (endp && endp[0] == '>') {
855 level = LOG_LEVEL(u);
856 if (LOG_FACILITY(u) != 0)
857 facility = LOG_FACILITY(u);
864 devkmsg_emit(facility, level, "%s", line);
869 static ssize_t devkmsg_read(struct file *file, char __user *buf,
870 size_t count, loff_t *ppos)
872 struct devkmsg_user *user = file->private_data;
873 struct printk_log *msg;
880 ret = mutex_lock_interruptible(&user->lock);
885 while (user->seq == log_next_seq) {
886 if (file->f_flags & O_NONBLOCK) {
893 ret = wait_event_interruptible(log_wait,
894 user->seq != log_next_seq);
900 if (user->seq < log_first_seq) {
901 /* our last seen message is gone, return error and reset */
902 user->idx = log_first_idx;
903 user->seq = log_first_seq;
909 msg = log_from_idx(user->idx);
910 len = msg_print_ext_header(user->buf, sizeof(user->buf),
912 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
913 log_dict(msg), msg->dict_len,
914 log_text(msg), msg->text_len);
916 user->idx = log_next(user->idx);
925 if (copy_to_user(buf, user->buf, len)) {
931 mutex_unlock(&user->lock);
935 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
937 struct devkmsg_user *user = file->private_data;
948 /* the first record */
949 user->idx = log_first_idx;
950 user->seq = log_first_seq;
954 * The first record after the last SYSLOG_ACTION_CLEAR,
955 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
956 * changes no global state, and does not clear anything.
958 user->idx = clear_idx;
959 user->seq = clear_seq;
962 /* after the last record */
963 user->idx = log_next_idx;
964 user->seq = log_next_seq;
973 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
975 struct devkmsg_user *user = file->private_data;
979 return EPOLLERR|EPOLLNVAL;
981 poll_wait(file, &log_wait, wait);
984 if (user->seq < log_next_seq) {
985 /* return error when data has vanished underneath us */
986 if (user->seq < log_first_seq)
987 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
989 ret = EPOLLIN|EPOLLRDNORM;
996 static int devkmsg_open(struct inode *inode, struct file *file)
998 struct devkmsg_user *user;
1001 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
1004 /* write-only does not need any file context */
1005 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
1006 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
1007 SYSLOG_FROM_READER);
1012 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
1016 ratelimit_default_init(&user->rs);
1017 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
1019 mutex_init(&user->lock);
1022 user->idx = log_first_idx;
1023 user->seq = log_first_seq;
1024 logbuf_unlock_irq();
1026 file->private_data = user;
1030 static int devkmsg_release(struct inode *inode, struct file *file)
1032 struct devkmsg_user *user = file->private_data;
1037 ratelimit_state_exit(&user->rs);
1039 mutex_destroy(&user->lock);
1044 const struct file_operations kmsg_fops = {
1045 .open = devkmsg_open,
1046 .read = devkmsg_read,
1047 .write_iter = devkmsg_write,
1048 .llseek = devkmsg_llseek,
1049 .poll = devkmsg_poll,
1050 .release = devkmsg_release,
1053 #ifdef CONFIG_CRASH_CORE
1055 * This appends the listed symbols to /proc/vmcore
1057 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1058 * obtain access to symbols that are otherwise very difficult to locate. These
1059 * symbols are specifically used so that utilities can access and extract the
1060 * dmesg log from a vmcore file after a crash.
1062 void log_buf_vmcoreinfo_setup(void)
1064 VMCOREINFO_SYMBOL(log_buf);
1065 VMCOREINFO_SYMBOL(log_buf_len);
1066 VMCOREINFO_SYMBOL(log_first_idx);
1067 VMCOREINFO_SYMBOL(clear_idx);
1068 VMCOREINFO_SYMBOL(log_next_idx);
1070 * Export struct printk_log size and field offsets. User space tools can
1071 * parse it and detect any changes to structure down the line.
1073 VMCOREINFO_STRUCT_SIZE(printk_log);
1074 VMCOREINFO_OFFSET(printk_log, ts_nsec);
1075 VMCOREINFO_OFFSET(printk_log, len);
1076 VMCOREINFO_OFFSET(printk_log, text_len);
1077 VMCOREINFO_OFFSET(printk_log, dict_len);
1078 #ifdef CONFIG_PRINTK_CALLER
1079 VMCOREINFO_OFFSET(printk_log, caller_id);
1084 /* requested log_buf_len from kernel cmdline */
1085 static unsigned long __initdata new_log_buf_len;
1087 /* we practice scaling the ring buffer by powers of 2 */
1088 static void __init log_buf_len_update(u64 size)
1090 if (size > (u64)LOG_BUF_LEN_MAX) {
1091 size = (u64)LOG_BUF_LEN_MAX;
1092 pr_err("log_buf over 2G is not supported.\n");
1096 size = roundup_pow_of_two(size);
1097 if (size > log_buf_len)
1098 new_log_buf_len = (unsigned long)size;
1101 /* save requested log_buf_len since it's too early to process it */
1102 static int __init log_buf_len_setup(char *str)
1109 size = memparse(str, &str);
1111 log_buf_len_update(size);
1115 early_param("log_buf_len", log_buf_len_setup);
1118 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1120 static void __init log_buf_add_cpu(void)
1122 unsigned int cpu_extra;
1125 * archs should set up cpu_possible_bits properly with
1126 * set_cpu_possible() after setup_arch() but just in
1127 * case lets ensure this is valid.
1129 if (num_possible_cpus() == 1)
1132 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1134 /* by default this will only continue through for large > 64 CPUs */
1135 if (cpu_extra <= __LOG_BUF_LEN / 2)
1138 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1139 __LOG_CPU_MAX_BUF_LEN);
1140 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1142 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1144 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1146 #else /* !CONFIG_SMP */
1147 static inline void log_buf_add_cpu(void) {}
1148 #endif /* CONFIG_SMP */
1150 void __init setup_log_buf(int early)
1152 unsigned long flags;
1156 if (log_buf != __log_buf)
1159 if (!early && !new_log_buf_len)
1162 if (!new_log_buf_len)
1165 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1166 if (unlikely(!new_log_buf)) {
1167 pr_err("log_buf_len: %lu bytes not available\n",
1172 logbuf_lock_irqsave(flags);
1173 log_buf_len = new_log_buf_len;
1174 log_buf = new_log_buf;
1175 new_log_buf_len = 0;
1176 free = __LOG_BUF_LEN - log_next_idx;
1177 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1178 logbuf_unlock_irqrestore(flags);
1180 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1181 pr_info("early log buf free: %u(%u%%)\n",
1182 free, (free * 100) / __LOG_BUF_LEN);
1185 static bool __read_mostly ignore_loglevel;
1187 static int __init ignore_loglevel_setup(char *str)
1189 ignore_loglevel = true;
1190 pr_info("debug: ignoring loglevel setting.\n");
1195 early_param("ignore_loglevel", ignore_loglevel_setup);
1196 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1197 MODULE_PARM_DESC(ignore_loglevel,
1198 "ignore loglevel setting (prints all kernel messages to the console)");
1200 static bool suppress_message_printing(int level)
1202 return (level >= console_loglevel && !ignore_loglevel);
1205 #ifdef CONFIG_BOOT_PRINTK_DELAY
1207 static int boot_delay; /* msecs delay after each printk during bootup */
1208 static unsigned long long loops_per_msec; /* based on boot_delay */
1210 static int __init boot_delay_setup(char *str)
1214 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1215 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1217 get_option(&str, &boot_delay);
1218 if (boot_delay > 10 * 1000)
1221 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1222 "HZ: %d, loops_per_msec: %llu\n",
1223 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1226 early_param("boot_delay", boot_delay_setup);
1228 static void boot_delay_msec(int level)
1230 unsigned long long k;
1231 unsigned long timeout;
1233 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1234 || suppress_message_printing(level)) {
1238 k = (unsigned long long)loops_per_msec * boot_delay;
1240 timeout = jiffies + msecs_to_jiffies(boot_delay);
1245 * use (volatile) jiffies to prevent
1246 * compiler reduction; loop termination via jiffies
1247 * is secondary and may or may not happen.
1249 if (time_after(jiffies, timeout))
1251 touch_nmi_watchdog();
1255 static inline void boot_delay_msec(int level)
1260 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1261 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1263 static size_t print_syslog(unsigned int level, char *buf)
1265 return sprintf(buf, "<%u>", level);
1268 static size_t print_time(u64 ts, char *buf)
1270 unsigned long rem_nsec = do_div(ts, 1000000000);
1272 return sprintf(buf, "[%5lu.%06lu]",
1273 (unsigned long)ts, rem_nsec / 1000);
1276 #ifdef CONFIG_PRINTK_CALLER
1277 static size_t print_caller(u32 id, char *buf)
1281 snprintf(caller, sizeof(caller), "%c%u",
1282 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1283 return sprintf(buf, "[%6s]", caller);
1286 #define print_caller(id, buf) 0
1289 static size_t print_prefix(const struct printk_log *msg, bool syslog,
1290 bool time, char *buf)
1295 len = print_syslog((msg->facility << 3) | msg->level, buf);
1298 len += print_time(msg->ts_nsec, buf + len);
1300 len += print_caller(msg->caller_id, buf + len);
1302 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1310 static size_t msg_print_text(const struct printk_log *msg, bool syslog,
1311 bool time, char *buf, size_t size)
1313 const char *text = log_text(msg);
1314 size_t text_size = msg->text_len;
1316 char prefix[PREFIX_MAX];
1317 const size_t prefix_len = print_prefix(msg, syslog, time, prefix);
1320 const char *next = memchr(text, '\n', text_size);
1324 text_len = next - text;
1326 text_size -= next - text;
1328 text_len = text_size;
1332 if (prefix_len + text_len + 1 >= size - len)
1335 memcpy(buf + len, prefix, prefix_len);
1337 memcpy(buf + len, text, text_len);
1341 /* SYSLOG_ACTION_* buffer size only calculation */
1342 len += prefix_len + text_len + 1;
1351 static int syslog_print(char __user *buf, int size)
1354 struct printk_log *msg;
1357 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1366 if (syslog_seq < log_first_seq) {
1367 /* messages are gone, move to first one */
1368 syslog_seq = log_first_seq;
1369 syslog_idx = log_first_idx;
1372 if (syslog_seq == log_next_seq) {
1373 logbuf_unlock_irq();
1378 * To keep reading/counting partial line consistent,
1379 * use printk_time value as of the beginning of a line.
1381 if (!syslog_partial)
1382 syslog_time = printk_time;
1384 skip = syslog_partial;
1385 msg = log_from_idx(syslog_idx);
1386 n = msg_print_text(msg, true, syslog_time, text,
1387 LOG_LINE_MAX + PREFIX_MAX);
1388 if (n - syslog_partial <= size) {
1389 /* message fits into buffer, move forward */
1390 syslog_idx = log_next(syslog_idx);
1392 n -= syslog_partial;
1395 /* partial read(), remember position */
1397 syslog_partial += n;
1400 logbuf_unlock_irq();
1405 if (copy_to_user(buf, text + skip, n)) {
1420 static int syslog_print_all(char __user *buf, int size, bool clear)
1429 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1436 * Find first record that fits, including all following records,
1437 * into the user-provided buffer for this dump.
1441 while (seq < log_next_seq) {
1442 struct printk_log *msg = log_from_idx(idx);
1444 len += msg_print_text(msg, true, time, NULL, 0);
1445 idx = log_next(idx);
1449 /* move first record forward until length fits into the buffer */
1452 while (len > size && seq < log_next_seq) {
1453 struct printk_log *msg = log_from_idx(idx);
1455 len -= msg_print_text(msg, true, time, NULL, 0);
1456 idx = log_next(idx);
1460 /* last message fitting into this dump */
1461 next_seq = log_next_seq;
1464 while (len >= 0 && seq < next_seq) {
1465 struct printk_log *msg = log_from_idx(idx);
1466 int textlen = msg_print_text(msg, true, time, text,
1467 LOG_LINE_MAX + PREFIX_MAX);
1469 idx = log_next(idx);
1472 logbuf_unlock_irq();
1473 if (copy_to_user(buf + len, text, textlen))
1479 if (seq < log_first_seq) {
1480 /* messages are gone, move to next one */
1481 seq = log_first_seq;
1482 idx = log_first_idx;
1487 clear_seq = log_next_seq;
1488 clear_idx = log_next_idx;
1490 logbuf_unlock_irq();
1496 static void syslog_clear(void)
1499 clear_seq = log_next_seq;
1500 clear_idx = log_next_idx;
1501 logbuf_unlock_irq();
1504 int do_syslog(int type, char __user *buf, int len, int source)
1507 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1510 error = check_syslog_permissions(type, source);
1515 case SYSLOG_ACTION_CLOSE: /* Close log */
1517 case SYSLOG_ACTION_OPEN: /* Open log */
1519 case SYSLOG_ACTION_READ: /* Read from log */
1520 if (!buf || len < 0)
1524 if (!access_ok(buf, len))
1526 error = wait_event_interruptible(log_wait,
1527 syslog_seq != log_next_seq);
1530 error = syslog_print(buf, len);
1532 /* Read/clear last kernel messages */
1533 case SYSLOG_ACTION_READ_CLEAR:
1536 /* Read last kernel messages */
1537 case SYSLOG_ACTION_READ_ALL:
1538 if (!buf || len < 0)
1542 if (!access_ok(buf, len))
1544 error = syslog_print_all(buf, len, clear);
1546 /* Clear ring buffer */
1547 case SYSLOG_ACTION_CLEAR:
1550 /* Disable logging to console */
1551 case SYSLOG_ACTION_CONSOLE_OFF:
1552 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1553 saved_console_loglevel = console_loglevel;
1554 console_loglevel = minimum_console_loglevel;
1556 /* Enable logging to console */
1557 case SYSLOG_ACTION_CONSOLE_ON:
1558 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1559 console_loglevel = saved_console_loglevel;
1560 saved_console_loglevel = LOGLEVEL_DEFAULT;
1563 /* Set level of messages printed to console */
1564 case SYSLOG_ACTION_CONSOLE_LEVEL:
1565 if (len < 1 || len > 8)
1567 if (len < minimum_console_loglevel)
1568 len = minimum_console_loglevel;
1569 console_loglevel = len;
1570 /* Implicitly re-enable logging to console */
1571 saved_console_loglevel = LOGLEVEL_DEFAULT;
1573 /* Number of chars in the log buffer */
1574 case SYSLOG_ACTION_SIZE_UNREAD:
1576 if (syslog_seq < log_first_seq) {
1577 /* messages are gone, move to first one */
1578 syslog_seq = log_first_seq;
1579 syslog_idx = log_first_idx;
1582 if (source == SYSLOG_FROM_PROC) {
1584 * Short-cut for poll(/"proc/kmsg") which simply checks
1585 * for pending data, not the size; return the count of
1586 * records, not the length.
1588 error = log_next_seq - syslog_seq;
1590 u64 seq = syslog_seq;
1591 u32 idx = syslog_idx;
1592 bool time = syslog_partial ? syslog_time : printk_time;
1594 while (seq < log_next_seq) {
1595 struct printk_log *msg = log_from_idx(idx);
1597 error += msg_print_text(msg, true, time, NULL,
1600 idx = log_next(idx);
1603 error -= syslog_partial;
1605 logbuf_unlock_irq();
1607 /* Size of the log buffer */
1608 case SYSLOG_ACTION_SIZE_BUFFER:
1609 error = log_buf_len;
1619 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1621 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1625 * Special console_lock variants that help to reduce the risk of soft-lockups.
1626 * They allow to pass console_lock to another printk() call using a busy wait.
1629 #ifdef CONFIG_LOCKDEP
1630 static struct lockdep_map console_owner_dep_map = {
1631 .name = "console_owner"
1635 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1636 static struct task_struct *console_owner;
1637 static bool console_waiter;
1640 * console_lock_spinning_enable - mark beginning of code where another
1641 * thread might safely busy wait
1643 * This basically converts console_lock into a spinlock. This marks
1644 * the section where the console_lock owner can not sleep, because
1645 * there may be a waiter spinning (like a spinlock). Also it must be
1646 * ready to hand over the lock at the end of the section.
1648 static void console_lock_spinning_enable(void)
1650 raw_spin_lock(&console_owner_lock);
1651 console_owner = current;
1652 raw_spin_unlock(&console_owner_lock);
1654 /* The waiter may spin on us after setting console_owner */
1655 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1659 * console_lock_spinning_disable_and_check - mark end of code where another
1660 * thread was able to busy wait and check if there is a waiter
1662 * This is called at the end of the section where spinning is allowed.
1663 * It has two functions. First, it is a signal that it is no longer
1664 * safe to start busy waiting for the lock. Second, it checks if
1665 * there is a busy waiter and passes the lock rights to her.
1667 * Important: Callers lose the lock if there was a busy waiter.
1668 * They must not touch items synchronized by console_lock
1671 * Return: 1 if the lock rights were passed, 0 otherwise.
1673 static int console_lock_spinning_disable_and_check(void)
1677 raw_spin_lock(&console_owner_lock);
1678 waiter = READ_ONCE(console_waiter);
1679 console_owner = NULL;
1680 raw_spin_unlock(&console_owner_lock);
1683 spin_release(&console_owner_dep_map, _THIS_IP_);
1687 /* The waiter is now free to continue */
1688 WRITE_ONCE(console_waiter, false);
1690 spin_release(&console_owner_dep_map, _THIS_IP_);
1693 * Hand off console_lock to waiter. The waiter will perform
1694 * the up(). After this, the waiter is the console_lock owner.
1696 mutex_release(&console_lock_dep_map, _THIS_IP_);
1701 * console_trylock_spinning - try to get console_lock by busy waiting
1703 * This allows to busy wait for the console_lock when the current
1704 * owner is running in specially marked sections. It means that
1705 * the current owner is running and cannot reschedule until it
1706 * is ready to lose the lock.
1708 * Return: 1 if we got the lock, 0 othrewise
1710 static int console_trylock_spinning(void)
1712 struct task_struct *owner = NULL;
1715 unsigned long flags;
1717 if (console_trylock())
1720 printk_safe_enter_irqsave(flags);
1722 raw_spin_lock(&console_owner_lock);
1723 owner = READ_ONCE(console_owner);
1724 waiter = READ_ONCE(console_waiter);
1725 if (!waiter && owner && owner != current) {
1726 WRITE_ONCE(console_waiter, true);
1729 raw_spin_unlock(&console_owner_lock);
1732 * If there is an active printk() writing to the
1733 * consoles, instead of having it write our data too,
1734 * see if we can offload that load from the active
1735 * printer, and do some printing ourselves.
1736 * Go into a spin only if there isn't already a waiter
1737 * spinning, and there is an active printer, and
1738 * that active printer isn't us (recursive printk?).
1741 printk_safe_exit_irqrestore(flags);
1745 /* We spin waiting for the owner to release us */
1746 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1747 /* Owner will clear console_waiter on hand off */
1748 while (READ_ONCE(console_waiter))
1750 spin_release(&console_owner_dep_map, _THIS_IP_);
1752 printk_safe_exit_irqrestore(flags);
1754 * The owner passed the console lock to us.
1755 * Since we did not spin on console lock, annotate
1756 * this as a trylock. Otherwise lockdep will
1759 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1765 * Call the console drivers, asking them to write out
1766 * log_buf[start] to log_buf[end - 1].
1767 * The console_lock must be held.
1769 static void call_console_drivers(const char *ext_text, size_t ext_len,
1770 const char *text, size_t len)
1772 struct console *con;
1774 trace_console_rcuidle(text, len);
1776 if (!console_drivers)
1779 for_each_console(con) {
1780 if (exclusive_console && con != exclusive_console)
1782 if (!(con->flags & CON_ENABLED))
1786 if (!cpu_online(smp_processor_id()) &&
1787 !(con->flags & CON_ANYTIME))
1789 if (con->flags & CON_EXTENDED)
1790 con->write(con, ext_text, ext_len);
1792 con->write(con, text, len);
1796 int printk_delay_msec __read_mostly;
1798 static inline void printk_delay(void)
1800 if (unlikely(printk_delay_msec)) {
1801 int m = printk_delay_msec;
1805 touch_nmi_watchdog();
1810 static inline u32 printk_caller_id(void)
1812 return in_task() ? task_pid_nr(current) :
1813 0x80000000 + raw_smp_processor_id();
1817 * Continuation lines are buffered, and not committed to the record buffer
1818 * until the line is complete, or a race forces it. The line fragments
1819 * though, are printed immediately to the consoles to ensure everything has
1820 * reached the console in case of a kernel crash.
1822 static struct cont {
1823 char buf[LOG_LINE_MAX];
1824 size_t len; /* length == 0 means unused buffer */
1825 u32 caller_id; /* printk_caller_id() of first print */
1826 u64 ts_nsec; /* time of first print */
1827 u8 level; /* log level of first message */
1828 u8 facility; /* log facility of first message */
1829 enum log_flags flags; /* prefix, newline flags */
1832 static void cont_flush(void)
1837 log_store(cont.caller_id, cont.facility, cont.level, cont.flags,
1838 cont.ts_nsec, NULL, 0, cont.buf, cont.len);
1842 static bool cont_add(u32 caller_id, int facility, int level,
1843 enum log_flags flags, const char *text, size_t len)
1845 /* If the line gets too long, split it up in separate records. */
1846 if (cont.len + len > sizeof(cont.buf)) {
1852 cont.facility = facility;
1854 cont.caller_id = caller_id;
1855 cont.ts_nsec = local_clock();
1859 memcpy(cont.buf + cont.len, text, len);
1862 // The original flags come from the first line,
1863 // but later continuations can add a newline.
1864 if (flags & LOG_NEWLINE) {
1865 cont.flags |= LOG_NEWLINE;
1872 static size_t log_output(int facility, int level, enum log_flags lflags, const char *dict, size_t dictlen, char *text, size_t text_len)
1874 const u32 caller_id = printk_caller_id();
1877 * If an earlier line was buffered, and we're a continuation
1878 * write from the same context, try to add it to the buffer.
1881 if (cont.caller_id == caller_id && (lflags & LOG_CONT)) {
1882 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1885 /* Otherwise, make sure it's flushed */
1889 /* Skip empty continuation lines that couldn't be added - they just flush */
1890 if (!text_len && (lflags & LOG_CONT))
1893 /* If it doesn't end in a newline, try to buffer the current line */
1894 if (!(lflags & LOG_NEWLINE)) {
1895 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1899 /* Store it in the record log */
1900 return log_store(caller_id, facility, level, lflags, 0,
1901 dict, dictlen, text, text_len);
1904 /* Must be called under logbuf_lock. */
1905 int vprintk_store(int facility, int level,
1906 const char *dict, size_t dictlen,
1907 const char *fmt, va_list args)
1909 static char textbuf[LOG_LINE_MAX];
1910 char *text = textbuf;
1912 enum log_flags lflags = 0;
1915 * The printf needs to come first; we need the syslog
1916 * prefix which might be passed-in as a parameter.
1918 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1920 /* mark and strip a trailing newline */
1921 if (text_len && text[text_len-1] == '\n') {
1923 lflags |= LOG_NEWLINE;
1926 /* strip kernel syslog prefix and extract log level or control flags */
1927 if (facility == 0) {
1930 while ((kern_level = printk_get_level(text)) != 0) {
1931 switch (kern_level) {
1933 if (level == LOGLEVEL_DEFAULT)
1934 level = kern_level - '0';
1936 case 'c': /* KERN_CONT */
1945 if (level == LOGLEVEL_DEFAULT)
1946 level = default_message_loglevel;
1949 lflags |= LOG_NEWLINE;
1951 return log_output(facility, level, lflags,
1952 dict, dictlen, text, text_len);
1955 asmlinkage int vprintk_emit(int facility, int level,
1956 const char *dict, size_t dictlen,
1957 const char *fmt, va_list args)
1960 bool in_sched = false, pending_output;
1961 unsigned long flags;
1964 /* Suppress unimportant messages after panic happens */
1965 if (unlikely(suppress_printk))
1968 if (level == LOGLEVEL_SCHED) {
1969 level = LOGLEVEL_DEFAULT;
1973 boot_delay_msec(level);
1976 /* This stops the holder of console_sem just where we want him */
1977 logbuf_lock_irqsave(flags);
1978 curr_log_seq = log_next_seq;
1979 printed_len = vprintk_store(facility, level, dict, dictlen, fmt, args);
1980 pending_output = (curr_log_seq != log_next_seq);
1981 logbuf_unlock_irqrestore(flags);
1983 /* If called from the scheduler, we can not call up(). */
1984 if (!in_sched && pending_output) {
1986 * Disable preemption to avoid being preempted while holding
1987 * console_sem which would prevent anyone from printing to
1992 * Try to acquire and then immediately release the console
1993 * semaphore. The release will print out buffers and wake up
1994 * /dev/kmsg and syslog() users.
1996 if (console_trylock_spinning())
2005 EXPORT_SYMBOL(vprintk_emit);
2007 asmlinkage int vprintk(const char *fmt, va_list args)
2009 return vprintk_func(fmt, args);
2011 EXPORT_SYMBOL(vprintk);
2013 int vprintk_default(const char *fmt, va_list args)
2017 #ifdef CONFIG_KGDB_KDB
2018 /* Allow to pass printk() to kdb but avoid a recursion. */
2019 if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
2020 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
2024 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
2028 EXPORT_SYMBOL_GPL(vprintk_default);
2031 * printk - print a kernel message
2032 * @fmt: format string
2034 * This is printk(). It can be called from any context. We want it to work.
2036 * We try to grab the console_lock. If we succeed, it's easy - we log the
2037 * output and call the console drivers. If we fail to get the semaphore, we
2038 * place the output into the log buffer and return. The current holder of
2039 * the console_sem will notice the new output in console_unlock(); and will
2040 * send it to the consoles before releasing the lock.
2042 * One effect of this deferred printing is that code which calls printk() and
2043 * then changes console_loglevel may break. This is because console_loglevel
2044 * is inspected when the actual printing occurs.
2049 * See the vsnprintf() documentation for format string extensions over C99.
2051 asmlinkage __visible int printk(const char *fmt, ...)
2056 va_start(args, fmt);
2057 r = vprintk_func(fmt, args);
2062 EXPORT_SYMBOL(printk);
2064 #else /* CONFIG_PRINTK */
2066 #define LOG_LINE_MAX 0
2067 #define PREFIX_MAX 0
2068 #define printk_time false
2070 static u64 syslog_seq;
2071 static u32 syslog_idx;
2072 static u64 console_seq;
2073 static u32 console_idx;
2074 static u64 exclusive_console_stop_seq;
2075 static u64 log_first_seq;
2076 static u32 log_first_idx;
2077 static u64 log_next_seq;
2078 static char *log_text(const struct printk_log *msg) { return NULL; }
2079 static char *log_dict(const struct printk_log *msg) { return NULL; }
2080 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2081 static u32 log_next(u32 idx) { return 0; }
2082 static ssize_t msg_print_ext_header(char *buf, size_t size,
2083 struct printk_log *msg,
2084 u64 seq) { return 0; }
2085 static ssize_t msg_print_ext_body(char *buf, size_t size,
2086 char *dict, size_t dict_len,
2087 char *text, size_t text_len) { return 0; }
2088 static void console_lock_spinning_enable(void) { }
2089 static int console_lock_spinning_disable_and_check(void) { return 0; }
2090 static void call_console_drivers(const char *ext_text, size_t ext_len,
2091 const char *text, size_t len) {}
2092 static size_t msg_print_text(const struct printk_log *msg, bool syslog,
2093 bool time, char *buf, size_t size) { return 0; }
2094 static bool suppress_message_printing(int level) { return false; }
2096 #endif /* CONFIG_PRINTK */
2098 #ifdef CONFIG_EARLY_PRINTK
2099 struct console *early_console;
2101 asmlinkage __visible void early_printk(const char *fmt, ...)
2111 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2114 early_console->write(early_console, buf, n);
2118 static int __add_preferred_console(char *name, int idx, char *options,
2119 char *brl_options, bool user_specified)
2121 struct console_cmdline *c;
2125 * See if this tty is not yet registered, and
2126 * if we have a slot free.
2128 for (i = 0, c = console_cmdline;
2129 i < MAX_CMDLINECONSOLES && c->name[0];
2131 if (strcmp(c->name, name) == 0 && c->index == idx) {
2133 preferred_console = i;
2135 c->user_specified = true;
2139 if (i == MAX_CMDLINECONSOLES)
2142 preferred_console = i;
2143 strlcpy(c->name, name, sizeof(c->name));
2144 c->options = options;
2145 c->user_specified = user_specified;
2146 braille_set_options(c, brl_options);
2152 static int __init console_msg_format_setup(char *str)
2154 if (!strcmp(str, "syslog"))
2155 console_msg_format = MSG_FORMAT_SYSLOG;
2156 if (!strcmp(str, "default"))
2157 console_msg_format = MSG_FORMAT_DEFAULT;
2160 __setup("console_msg_format=", console_msg_format_setup);
2163 * Set up a console. Called via do_early_param() in init/main.c
2164 * for each "console=" parameter in the boot command line.
2166 static int __init console_setup(char *str)
2168 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2169 char *s, *options, *brl_options = NULL;
2172 if (_braille_console_setup(&str, &brl_options))
2176 * Decode str into name, index, options.
2178 if (str[0] >= '0' && str[0] <= '9') {
2179 strcpy(buf, "ttyS");
2180 strncpy(buf + 4, str, sizeof(buf) - 5);
2182 strncpy(buf, str, sizeof(buf) - 1);
2184 buf[sizeof(buf) - 1] = 0;
2185 options = strchr(str, ',');
2189 if (!strcmp(str, "ttya"))
2190 strcpy(buf, "ttyS0");
2191 if (!strcmp(str, "ttyb"))
2192 strcpy(buf, "ttyS1");
2194 for (s = buf; *s; s++)
2195 if (isdigit(*s) || *s == ',')
2197 idx = simple_strtoul(s, NULL, 10);
2200 __add_preferred_console(buf, idx, options, brl_options, true);
2201 console_set_on_cmdline = 1;
2204 __setup("console=", console_setup);
2207 * add_preferred_console - add a device to the list of preferred consoles.
2208 * @name: device name
2209 * @idx: device index
2210 * @options: options for this console
2212 * The last preferred console added will be used for kernel messages
2213 * and stdin/out/err for init. Normally this is used by console_setup
2214 * above to handle user-supplied console arguments; however it can also
2215 * be used by arch-specific code either to override the user or more
2216 * commonly to provide a default console (ie from PROM variables) when
2217 * the user has not supplied one.
2219 int add_preferred_console(char *name, int idx, char *options)
2221 return __add_preferred_console(name, idx, options, NULL, false);
2224 bool console_suspend_enabled = true;
2225 EXPORT_SYMBOL(console_suspend_enabled);
2227 static int __init console_suspend_disable(char *str)
2229 console_suspend_enabled = false;
2232 __setup("no_console_suspend", console_suspend_disable);
2233 module_param_named(console_suspend, console_suspend_enabled,
2234 bool, S_IRUGO | S_IWUSR);
2235 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2236 " and hibernate operations");
2239 * suspend_console - suspend the console subsystem
2241 * This disables printk() while we go into suspend states
2243 void suspend_console(void)
2245 if (!console_suspend_enabled)
2247 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2249 console_suspended = 1;
2253 void resume_console(void)
2255 if (!console_suspend_enabled)
2258 console_suspended = 0;
2263 * console_cpu_notify - print deferred console messages after CPU hotplug
2266 * If printk() is called from a CPU that is not online yet, the messages
2267 * will be printed on the console only if there are CON_ANYTIME consoles.
2268 * This function is called when a new CPU comes online (or fails to come
2269 * up) or goes offline.
2271 static int console_cpu_notify(unsigned int cpu)
2273 if (!cpuhp_tasks_frozen) {
2274 /* If trylock fails, someone else is doing the printing */
2275 if (console_trylock())
2282 * console_lock - lock the console system for exclusive use.
2284 * Acquires a lock which guarantees that the caller has
2285 * exclusive access to the console system and the console_drivers list.
2287 * Can sleep, returns nothing.
2289 void console_lock(void)
2294 if (console_suspended)
2297 console_may_schedule = 1;
2299 EXPORT_SYMBOL(console_lock);
2302 * console_trylock - try to lock the console system for exclusive use.
2304 * Try to acquire a lock which guarantees that the caller has exclusive
2305 * access to the console system and the console_drivers list.
2307 * returns 1 on success, and 0 on failure to acquire the lock.
2309 int console_trylock(void)
2311 if (down_trylock_console_sem())
2313 if (console_suspended) {
2318 console_may_schedule = 0;
2321 EXPORT_SYMBOL(console_trylock);
2323 int is_console_locked(void)
2325 return console_locked;
2327 EXPORT_SYMBOL(is_console_locked);
2330 * Check if we have any console that is capable of printing while cpu is
2331 * booting or shutting down. Requires console_sem.
2333 static int have_callable_console(void)
2335 struct console *con;
2337 for_each_console(con)
2338 if ((con->flags & CON_ENABLED) &&
2339 (con->flags & CON_ANYTIME))
2346 * Can we actually use the console at this time on this cpu?
2348 * Console drivers may assume that per-cpu resources have been allocated. So
2349 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2350 * call them until this CPU is officially up.
2352 static inline int can_use_console(void)
2354 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2358 * console_unlock - unlock the console system
2360 * Releases the console_lock which the caller holds on the console system
2361 * and the console driver list.
2363 * While the console_lock was held, console output may have been buffered
2364 * by printk(). If this is the case, console_unlock(); emits
2365 * the output prior to releasing the lock.
2367 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2369 * console_unlock(); may be called from any context.
2371 void console_unlock(void)
2373 static char ext_text[CONSOLE_EXT_LOG_MAX];
2374 static char text[LOG_LINE_MAX + PREFIX_MAX];
2375 unsigned long flags;
2376 bool do_cond_resched, retry;
2378 if (console_suspended) {
2384 * Console drivers are called with interrupts disabled, so
2385 * @console_may_schedule should be cleared before; however, we may
2386 * end up dumping a lot of lines, for example, if called from
2387 * console registration path, and should invoke cond_resched()
2388 * between lines if allowable. Not doing so can cause a very long
2389 * scheduling stall on a slow console leading to RCU stall and
2390 * softlockup warnings which exacerbate the issue with more
2391 * messages practically incapacitating the system.
2393 * console_trylock() is not able to detect the preemptive
2394 * context reliably. Therefore the value must be stored before
2395 * and cleared after the the "again" goto label.
2397 do_cond_resched = console_may_schedule;
2399 console_may_schedule = 0;
2402 * We released the console_sem lock, so we need to recheck if
2403 * cpu is online and (if not) is there at least one CON_ANYTIME
2406 if (!can_use_console()) {
2413 struct printk_log *msg;
2417 printk_safe_enter_irqsave(flags);
2418 raw_spin_lock(&logbuf_lock);
2419 if (console_seq < log_first_seq) {
2420 len = snprintf(text, sizeof(text),
2421 "** %llu printk messages dropped **\n",
2422 log_first_seq - console_seq);
2424 /* messages are gone, move to first one */
2425 console_seq = log_first_seq;
2426 console_idx = log_first_idx;
2431 if (console_seq == log_next_seq)
2434 msg = log_from_idx(console_idx);
2435 if (suppress_message_printing(msg->level)) {
2437 * Skip record we have buffered and already printed
2438 * directly to the console when we received it, and
2439 * record that has level above the console loglevel.
2441 console_idx = log_next(console_idx);
2446 /* Output to all consoles once old messages replayed. */
2447 if (unlikely(exclusive_console &&
2448 console_seq >= exclusive_console_stop_seq)) {
2449 exclusive_console = NULL;
2452 len += msg_print_text(msg,
2453 console_msg_format & MSG_FORMAT_SYSLOG,
2454 printk_time, text + len, sizeof(text) - len);
2455 if (nr_ext_console_drivers) {
2456 ext_len = msg_print_ext_header(ext_text,
2459 ext_len += msg_print_ext_body(ext_text + ext_len,
2460 sizeof(ext_text) - ext_len,
2461 log_dict(msg), msg->dict_len,
2462 log_text(msg), msg->text_len);
2464 console_idx = log_next(console_idx);
2466 raw_spin_unlock(&logbuf_lock);
2469 * While actively printing out messages, if another printk()
2470 * were to occur on another CPU, it may wait for this one to
2471 * finish. This task can not be preempted if there is a
2472 * waiter waiting to take over.
2474 console_lock_spinning_enable();
2476 stop_critical_timings(); /* don't trace print latency */
2477 call_console_drivers(ext_text, ext_len, text, len);
2478 start_critical_timings();
2480 if (console_lock_spinning_disable_and_check()) {
2481 printk_safe_exit_irqrestore(flags);
2485 printk_safe_exit_irqrestore(flags);
2487 if (do_cond_resched)
2493 raw_spin_unlock(&logbuf_lock);
2498 * Someone could have filled up the buffer again, so re-check if there's
2499 * something to flush. In case we cannot trylock the console_sem again,
2500 * there's a new owner and the console_unlock() from them will do the
2501 * flush, no worries.
2503 raw_spin_lock(&logbuf_lock);
2504 retry = console_seq != log_next_seq;
2505 raw_spin_unlock(&logbuf_lock);
2506 printk_safe_exit_irqrestore(flags);
2508 if (retry && console_trylock())
2511 EXPORT_SYMBOL(console_unlock);
2514 * console_conditional_schedule - yield the CPU if required
2516 * If the console code is currently allowed to sleep, and
2517 * if this CPU should yield the CPU to another task, do
2520 * Must be called within console_lock();.
2522 void __sched console_conditional_schedule(void)
2524 if (console_may_schedule)
2527 EXPORT_SYMBOL(console_conditional_schedule);
2529 void console_unblank(void)
2534 * console_unblank can no longer be called in interrupt context unless
2535 * oops_in_progress is set to 1..
2537 if (oops_in_progress) {
2538 if (down_trylock_console_sem() != 0)
2544 console_may_schedule = 0;
2546 if ((c->flags & CON_ENABLED) && c->unblank)
2552 * console_flush_on_panic - flush console content on panic
2553 * @mode: flush all messages in buffer or just the pending ones
2555 * Immediately output all pending messages no matter what.
2557 void console_flush_on_panic(enum con_flush_mode mode)
2560 * If someone else is holding the console lock, trylock will fail
2561 * and may_schedule may be set. Ignore and proceed to unlock so
2562 * that messages are flushed out. As this can be called from any
2563 * context and we don't want to get preempted while flushing,
2564 * ensure may_schedule is cleared.
2567 console_may_schedule = 0;
2569 if (mode == CONSOLE_REPLAY_ALL) {
2570 unsigned long flags;
2572 logbuf_lock_irqsave(flags);
2573 console_seq = log_first_seq;
2574 console_idx = log_first_idx;
2575 logbuf_unlock_irqrestore(flags);
2581 * Return the console tty driver structure and its associated index
2583 struct tty_driver *console_device(int *index)
2586 struct tty_driver *driver = NULL;
2589 for_each_console(c) {
2592 driver = c->device(c, index);
2601 * Prevent further output on the passed console device so that (for example)
2602 * serial drivers can disable console output before suspending a port, and can
2603 * re-enable output afterwards.
2605 void console_stop(struct console *console)
2608 console->flags &= ~CON_ENABLED;
2611 EXPORT_SYMBOL(console_stop);
2613 void console_start(struct console *console)
2616 console->flags |= CON_ENABLED;
2619 EXPORT_SYMBOL(console_start);
2621 static int __read_mostly keep_bootcon;
2623 static int __init keep_bootcon_setup(char *str)
2626 pr_info("debug: skip boot console de-registration.\n");
2631 early_param("keep_bootcon", keep_bootcon_setup);
2634 * This is called by register_console() to try to match
2635 * the newly registered console with any of the ones selected
2636 * by either the command line or add_preferred_console() and
2639 * Care need to be taken with consoles that are statically
2640 * enabled such as netconsole
2642 static int try_enable_new_console(struct console *newcon, bool user_specified)
2644 struct console_cmdline *c;
2647 for (i = 0, c = console_cmdline;
2648 i < MAX_CMDLINECONSOLES && c->name[0];
2650 if (c->user_specified != user_specified)
2652 if (!newcon->match ||
2653 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2654 /* default matching */
2655 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2656 if (strcmp(c->name, newcon->name) != 0)
2658 if (newcon->index >= 0 &&
2659 newcon->index != c->index)
2661 if (newcon->index < 0)
2662 newcon->index = c->index;
2664 if (_braille_register_console(newcon, c))
2667 if (newcon->setup &&
2668 newcon->setup(newcon, c->options) != 0)
2671 newcon->flags |= CON_ENABLED;
2672 if (i == preferred_console) {
2673 newcon->flags |= CON_CONSDEV;
2674 has_preferred_console = true;
2680 * Some consoles, such as pstore and netconsole, can be enabled even
2681 * without matching. Accept the pre-enabled consoles only when match()
2682 * and setup() had a change to be called.
2684 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
2691 * The console driver calls this routine during kernel initialization
2692 * to register the console printing procedure with printk() and to
2693 * print any messages that were printed by the kernel before the
2694 * console driver was initialized.
2696 * This can happen pretty early during the boot process (because of
2697 * early_printk) - sometimes before setup_arch() completes - be careful
2698 * of what kernel features are used - they may not be initialised yet.
2700 * There are two types of consoles - bootconsoles (early_printk) and
2701 * "real" consoles (everything which is not a bootconsole) which are
2702 * handled differently.
2703 * - Any number of bootconsoles can be registered at any time.
2704 * - As soon as a "real" console is registered, all bootconsoles
2705 * will be unregistered automatically.
2706 * - Once a "real" console is registered, any attempt to register a
2707 * bootconsoles will be rejected
2709 void register_console(struct console *newcon)
2711 unsigned long flags;
2712 struct console *bcon = NULL;
2715 if (console_drivers)
2716 for_each_console(bcon)
2717 if (WARN(bcon == newcon,
2718 "console '%s%d' already registered\n",
2719 bcon->name, bcon->index))
2723 * before we register a new CON_BOOT console, make sure we don't
2724 * already have a valid console
2726 if (console_drivers && newcon->flags & CON_BOOT) {
2727 /* find the last or real console */
2728 for_each_console(bcon) {
2729 if (!(bcon->flags & CON_BOOT)) {
2730 pr_info("Too late to register bootconsole %s%d\n",
2731 newcon->name, newcon->index);
2737 if (console_drivers && console_drivers->flags & CON_BOOT)
2738 bcon = console_drivers;
2740 if (!has_preferred_console || bcon || !console_drivers)
2741 has_preferred_console = preferred_console >= 0;
2744 * See if we want to use this console driver. If we
2745 * didn't select a console we take the first one
2746 * that registers here.
2748 if (!has_preferred_console) {
2749 if (newcon->index < 0)
2751 if (newcon->setup == NULL ||
2752 newcon->setup(newcon, NULL) == 0) {
2753 newcon->flags |= CON_ENABLED;
2754 if (newcon->device) {
2755 newcon->flags |= CON_CONSDEV;
2756 has_preferred_console = true;
2761 /* See if this console matches one we selected on the command line */
2762 err = try_enable_new_console(newcon, true);
2764 /* If not, try to match against the platform default(s) */
2766 err = try_enable_new_console(newcon, false);
2768 /* printk() messages are not printed to the Braille console. */
2769 if (err || newcon->flags & CON_BRL)
2773 * If we have a bootconsole, and are switching to a real console,
2774 * don't print everything out again, since when the boot console, and
2775 * the real console are the same physical device, it's annoying to
2776 * see the beginning boot messages twice
2778 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2779 newcon->flags &= ~CON_PRINTBUFFER;
2782 * Put this console in the list - keep the
2783 * preferred driver at the head of the list.
2786 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2787 newcon->next = console_drivers;
2788 console_drivers = newcon;
2790 newcon->next->flags &= ~CON_CONSDEV;
2791 /* Ensure this flag is always set for the head of the list */
2792 newcon->flags |= CON_CONSDEV;
2794 newcon->next = console_drivers->next;
2795 console_drivers->next = newcon;
2798 if (newcon->flags & CON_EXTENDED)
2799 nr_ext_console_drivers++;
2801 if (newcon->flags & CON_PRINTBUFFER) {
2803 * console_unlock(); will print out the buffered messages
2806 logbuf_lock_irqsave(flags);
2808 * We're about to replay the log buffer. Only do this to the
2809 * just-registered console to avoid excessive message spam to
2810 * the already-registered consoles.
2812 * Set exclusive_console with disabled interrupts to reduce
2813 * race window with eventual console_flush_on_panic() that
2814 * ignores console_lock.
2816 exclusive_console = newcon;
2817 exclusive_console_stop_seq = console_seq;
2818 console_seq = syslog_seq;
2819 console_idx = syslog_idx;
2820 logbuf_unlock_irqrestore(flags);
2823 console_sysfs_notify();
2826 * By unregistering the bootconsoles after we enable the real console
2827 * we get the "console xxx enabled" message on all the consoles -
2828 * boot consoles, real consoles, etc - this is to ensure that end
2829 * users know there might be something in the kernel's log buffer that
2830 * went to the bootconsole (that they do not see on the real console)
2832 pr_info("%sconsole [%s%d] enabled\n",
2833 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2834 newcon->name, newcon->index);
2836 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2838 /* We need to iterate through all boot consoles, to make
2839 * sure we print everything out, before we unregister them.
2841 for_each_console(bcon)
2842 if (bcon->flags & CON_BOOT)
2843 unregister_console(bcon);
2846 EXPORT_SYMBOL(register_console);
2848 int unregister_console(struct console *console)
2850 struct console *a, *b;
2853 pr_info("%sconsole [%s%d] disabled\n",
2854 (console->flags & CON_BOOT) ? "boot" : "" ,
2855 console->name, console->index);
2857 res = _braille_unregister_console(console);
2863 if (console_drivers == console) {
2864 console_drivers=console->next;
2866 } else if (console_drivers) {
2867 for (a=console_drivers->next, b=console_drivers ;
2868 a; b=a, a=b->next) {
2877 if (!res && (console->flags & CON_EXTENDED))
2878 nr_ext_console_drivers--;
2881 * If this isn't the last console and it has CON_CONSDEV set, we
2882 * need to set it on the next preferred console.
2884 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2885 console_drivers->flags |= CON_CONSDEV;
2887 console->flags &= ~CON_ENABLED;
2889 console_sysfs_notify();
2892 EXPORT_SYMBOL(unregister_console);
2895 * Initialize the console device. This is called *early*, so
2896 * we can't necessarily depend on lots of kernel help here.
2897 * Just do some early initializations, and do the complex setup
2900 void __init console_init(void)
2904 initcall_entry_t *ce;
2906 /* Setup the default TTY line discipline. */
2910 * set up the console device so that later boot sequences can
2911 * inform about problems etc..
2913 ce = __con_initcall_start;
2914 trace_initcall_level("console");
2915 while (ce < __con_initcall_end) {
2916 call = initcall_from_entry(ce);
2917 trace_initcall_start(call);
2919 trace_initcall_finish(call, ret);
2925 * Some boot consoles access data that is in the init section and which will
2926 * be discarded after the initcalls have been run. To make sure that no code
2927 * will access this data, unregister the boot consoles in a late initcall.
2929 * If for some reason, such as deferred probe or the driver being a loadable
2930 * module, the real console hasn't registered yet at this point, there will
2931 * be a brief interval in which no messages are logged to the console, which
2932 * makes it difficult to diagnose problems that occur during this time.
2934 * To mitigate this problem somewhat, only unregister consoles whose memory
2935 * intersects with the init section. Note that all other boot consoles will
2936 * get unregistred when the real preferred console is registered.
2938 static int __init printk_late_init(void)
2940 struct console *con;
2943 for_each_console(con) {
2944 if (!(con->flags & CON_BOOT))
2947 /* Check addresses that might be used for enabled consoles. */
2948 if (init_section_intersects(con, sizeof(*con)) ||
2949 init_section_contains(con->write, 0) ||
2950 init_section_contains(con->read, 0) ||
2951 init_section_contains(con->device, 0) ||
2952 init_section_contains(con->unblank, 0) ||
2953 init_section_contains(con->data, 0)) {
2955 * Please, consider moving the reported consoles out
2956 * of the init section.
2958 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2959 con->name, con->index);
2960 unregister_console(con);
2963 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2964 console_cpu_notify);
2966 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2967 console_cpu_notify, NULL);
2971 late_initcall(printk_late_init);
2973 #if defined CONFIG_PRINTK
2975 * Delayed printk version, for scheduler-internal messages:
2977 #define PRINTK_PENDING_WAKEUP 0x01
2978 #define PRINTK_PENDING_OUTPUT 0x02
2980 static DEFINE_PER_CPU(int, printk_pending);
2982 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2984 int pending = __this_cpu_xchg(printk_pending, 0);
2986 if (pending & PRINTK_PENDING_OUTPUT) {
2987 /* If trylock fails, someone else is doing the printing */
2988 if (console_trylock())
2992 if (pending & PRINTK_PENDING_WAKEUP)
2993 wake_up_interruptible(&log_wait);
2996 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2997 .func = wake_up_klogd_work_func,
2998 .flags = ATOMIC_INIT(IRQ_WORK_LAZY),
3001 void wake_up_klogd(void)
3004 if (waitqueue_active(&log_wait)) {
3005 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
3006 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3011 void defer_console_output(void)
3014 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
3015 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3019 int vprintk_deferred(const char *fmt, va_list args)
3023 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
3024 defer_console_output();
3029 int printk_deferred(const char *fmt, ...)
3034 va_start(args, fmt);
3035 r = vprintk_deferred(fmt, args);
3042 * printk rate limiting, lifted from the networking subsystem.
3044 * This enforces a rate limit: not more than 10 kernel messages
3045 * every 5s to make a denial-of-service attack impossible.
3047 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3049 int __printk_ratelimit(const char *func)
3051 return ___ratelimit(&printk_ratelimit_state, func);
3053 EXPORT_SYMBOL(__printk_ratelimit);
3056 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3057 * @caller_jiffies: pointer to caller's state
3058 * @interval_msecs: minimum interval between prints
3060 * printk_timed_ratelimit() returns true if more than @interval_msecs
3061 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3064 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3065 unsigned int interval_msecs)
3067 unsigned long elapsed = jiffies - *caller_jiffies;
3069 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3072 *caller_jiffies = jiffies;
3075 EXPORT_SYMBOL(printk_timed_ratelimit);
3077 static DEFINE_SPINLOCK(dump_list_lock);
3078 static LIST_HEAD(dump_list);
3081 * kmsg_dump_register - register a kernel log dumper.
3082 * @dumper: pointer to the kmsg_dumper structure
3084 * Adds a kernel log dumper to the system. The dump callback in the
3085 * structure will be called when the kernel oopses or panics and must be
3086 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3088 int kmsg_dump_register(struct kmsg_dumper *dumper)
3090 unsigned long flags;
3093 /* The dump callback needs to be set */
3097 spin_lock_irqsave(&dump_list_lock, flags);
3098 /* Don't allow registering multiple times */
3099 if (!dumper->registered) {
3100 dumper->registered = 1;
3101 list_add_tail_rcu(&dumper->list, &dump_list);
3104 spin_unlock_irqrestore(&dump_list_lock, flags);
3108 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3111 * kmsg_dump_unregister - unregister a kmsg dumper.
3112 * @dumper: pointer to the kmsg_dumper structure
3114 * Removes a dump device from the system. Returns zero on success and
3115 * %-EINVAL otherwise.
3117 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3119 unsigned long flags;
3122 spin_lock_irqsave(&dump_list_lock, flags);
3123 if (dumper->registered) {
3124 dumper->registered = 0;
3125 list_del_rcu(&dumper->list);
3128 spin_unlock_irqrestore(&dump_list_lock, flags);
3133 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3135 static bool always_kmsg_dump;
3136 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3139 * kmsg_dump - dump kernel log to kernel message dumpers.
3140 * @reason: the reason (oops, panic etc) for dumping
3142 * Call each of the registered dumper's dump() callback, which can
3143 * retrieve the kmsg records with kmsg_dump_get_line() or
3144 * kmsg_dump_get_buffer().
3146 void kmsg_dump(enum kmsg_dump_reason reason)
3148 struct kmsg_dumper *dumper;
3149 unsigned long flags;
3151 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3155 list_for_each_entry_rcu(dumper, &dump_list, list) {
3156 if (dumper->max_reason && reason > dumper->max_reason)
3159 /* initialize iterator with data about the stored records */
3160 dumper->active = true;
3162 logbuf_lock_irqsave(flags);
3163 dumper->cur_seq = clear_seq;
3164 dumper->cur_idx = clear_idx;
3165 dumper->next_seq = log_next_seq;
3166 dumper->next_idx = log_next_idx;
3167 logbuf_unlock_irqrestore(flags);
3169 /* invoke dumper which will iterate over records */
3170 dumper->dump(dumper, reason);
3172 /* reset iterator */
3173 dumper->active = false;
3179 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3180 * @dumper: registered kmsg dumper
3181 * @syslog: include the "<4>" prefixes
3182 * @line: buffer to copy the line to
3183 * @size: maximum size of the buffer
3184 * @len: length of line placed into buffer
3186 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3187 * record, and copy one record into the provided buffer.
3189 * Consecutive calls will return the next available record moving
3190 * towards the end of the buffer with the youngest messages.
3192 * A return value of FALSE indicates that there are no more records to
3195 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3197 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3198 char *line, size_t size, size_t *len)
3200 struct printk_log *msg;
3204 if (!dumper->active)
3207 if (dumper->cur_seq < log_first_seq) {
3208 /* messages are gone, move to first available one */
3209 dumper->cur_seq = log_first_seq;
3210 dumper->cur_idx = log_first_idx;
3214 if (dumper->cur_seq >= log_next_seq)
3217 msg = log_from_idx(dumper->cur_idx);
3218 l = msg_print_text(msg, syslog, printk_time, line, size);
3220 dumper->cur_idx = log_next(dumper->cur_idx);
3230 * kmsg_dump_get_line - retrieve one kmsg log line
3231 * @dumper: registered kmsg dumper
3232 * @syslog: include the "<4>" prefixes
3233 * @line: buffer to copy the line to
3234 * @size: maximum size of the buffer
3235 * @len: length of line placed into buffer
3237 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3238 * record, and copy one record into the provided buffer.
3240 * Consecutive calls will return the next available record moving
3241 * towards the end of the buffer with the youngest messages.
3243 * A return value of FALSE indicates that there are no more records to
3246 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3247 char *line, size_t size, size_t *len)
3249 unsigned long flags;
3252 logbuf_lock_irqsave(flags);
3253 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3254 logbuf_unlock_irqrestore(flags);
3258 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3261 * kmsg_dump_get_buffer - copy kmsg log lines
3262 * @dumper: registered kmsg dumper
3263 * @syslog: include the "<4>" prefixes
3264 * @buf: buffer to copy the line to
3265 * @size: maximum size of the buffer
3266 * @len: length of line placed into buffer
3268 * Start at the end of the kmsg buffer and fill the provided buffer
3269 * with as many of the the *youngest* kmsg records that fit into it.
3270 * If the buffer is large enough, all available kmsg records will be
3271 * copied with a single call.
3273 * Consecutive calls will fill the buffer with the next block of
3274 * available older records, not including the earlier retrieved ones.
3276 * A return value of FALSE indicates that there are no more records to
3279 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3280 char *buf, size_t size, size_t *len)
3282 unsigned long flags;
3289 bool time = printk_time;
3291 if (!dumper->active)
3294 logbuf_lock_irqsave(flags);
3295 if (dumper->cur_seq < log_first_seq) {
3296 /* messages are gone, move to first available one */
3297 dumper->cur_seq = log_first_seq;
3298 dumper->cur_idx = log_first_idx;
3302 if (dumper->cur_seq >= dumper->next_seq) {
3303 logbuf_unlock_irqrestore(flags);
3307 /* calculate length of entire buffer */
3308 seq = dumper->cur_seq;
3309 idx = dumper->cur_idx;
3310 while (seq < dumper->next_seq) {
3311 struct printk_log *msg = log_from_idx(idx);
3313 l += msg_print_text(msg, true, time, NULL, 0);
3314 idx = log_next(idx);
3318 /* move first record forward until length fits into the buffer */
3319 seq = dumper->cur_seq;
3320 idx = dumper->cur_idx;
3321 while (l >= size && seq < dumper->next_seq) {
3322 struct printk_log *msg = log_from_idx(idx);
3324 l -= msg_print_text(msg, true, time, NULL, 0);
3325 idx = log_next(idx);
3329 /* last message in next interation */
3334 while (seq < dumper->next_seq) {
3335 struct printk_log *msg = log_from_idx(idx);
3337 l += msg_print_text(msg, syslog, time, buf + l, size - l);
3338 idx = log_next(idx);
3342 dumper->next_seq = next_seq;
3343 dumper->next_idx = next_idx;
3345 logbuf_unlock_irqrestore(flags);
3351 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3354 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3355 * @dumper: registered kmsg dumper
3357 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3358 * kmsg_dump_get_buffer() can be called again and used multiple
3359 * times within the same dumper.dump() callback.
3361 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3363 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3365 dumper->cur_seq = clear_seq;
3366 dumper->cur_idx = clear_idx;
3367 dumper->next_seq = log_next_seq;
3368 dumper->next_idx = log_next_idx;
3372 * kmsg_dump_rewind - reset the interator
3373 * @dumper: registered kmsg dumper
3375 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3376 * kmsg_dump_get_buffer() can be called again and used multiple
3377 * times within the same dumper.dump() callback.
3379 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3381 unsigned long flags;
3383 logbuf_lock_irqsave(flags);
3384 kmsg_dump_rewind_nolock(dumper);
3385 logbuf_unlock_irqrestore(flags);
3387 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);