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)
124 if (!strncmp(str, "on", 2)) {
125 devkmsg_log = DEVKMSG_LOG_MASK_ON;
127 } else if (!strncmp(str, "off", 3)) {
128 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
130 } else if (!strncmp(str, "ratelimit", 9)) {
131 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
137 static int __init control_devkmsg(char *str)
139 if (__control_devkmsg(str) < 0)
143 * Set sysctl string accordingly:
145 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
146 strcpy(devkmsg_log_str, "on");
147 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
148 strcpy(devkmsg_log_str, "off");
149 /* else "ratelimit" which is set by default. */
152 * Sysctl cannot change it anymore. The kernel command line setting of
153 * this parameter is to force the setting to be permanent throughout the
154 * runtime of the system. This is a precation measure against userspace
155 * trying to be a smarta** and attempting to change it up on us.
157 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
161 __setup("printk.devkmsg=", control_devkmsg);
163 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
165 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
166 void __user *buffer, size_t *lenp, loff_t *ppos)
168 char old_str[DEVKMSG_STR_MAX_SIZE];
173 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
177 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
180 err = proc_dostring(table, write, buffer, lenp, ppos);
185 err = __control_devkmsg(devkmsg_log_str);
188 * Do not accept an unknown string OR a known string with
191 if (err < 0 || (err + 1 != *lenp)) {
193 /* ... and restore old setting. */
195 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
204 /* Number of registered extended console drivers. */
205 static int nr_ext_console_drivers;
208 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
209 * macros instead of functions so that _RET_IP_ contains useful information.
211 #define down_console_sem() do { \
213 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
216 static int __down_trylock_console_sem(unsigned long ip)
222 * Here and in __up_console_sem() we need to be in safe mode,
223 * because spindump/WARN/etc from under console ->lock will
224 * deadlock in printk()->down_trylock_console_sem() otherwise.
226 printk_safe_enter_irqsave(flags);
227 lock_failed = down_trylock(&console_sem);
228 printk_safe_exit_irqrestore(flags);
232 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
235 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
237 static void __up_console_sem(unsigned long ip)
241 mutex_release(&console_lock_dep_map, 1, ip);
243 printk_safe_enter_irqsave(flags);
245 printk_safe_exit_irqrestore(flags);
247 #define up_console_sem() __up_console_sem(_RET_IP_)
250 * This is used for debugging the mess that is the VT code by
251 * keeping track if we have the console semaphore held. It's
252 * definitely not the perfect debug tool (we don't know if _WE_
253 * hold it and are racing, but it helps tracking those weird code
254 * paths in the console code where we end up in places I want
255 * locked without the console sempahore held).
257 static int console_locked, console_suspended;
260 * If exclusive_console is non-NULL then only this console is to be printed to.
262 static struct console *exclusive_console;
265 * Array of consoles built from command line options (console=)
268 #define MAX_CMDLINECONSOLES 8
270 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
272 static int preferred_console = -1;
273 int console_set_on_cmdline;
274 EXPORT_SYMBOL(console_set_on_cmdline);
276 /* Flag: console code may call schedule() */
277 static int console_may_schedule;
279 enum con_msg_format_flags {
280 MSG_FORMAT_DEFAULT = 0,
281 MSG_FORMAT_SYSLOG = (1 << 0),
284 static int console_msg_format = MSG_FORMAT_DEFAULT;
287 * The printk log buffer consists of a chain of concatenated variable
288 * length records. Every record starts with a record header, containing
289 * the overall length of the record.
291 * The heads to the first and last entry in the buffer, as well as the
292 * sequence numbers of these entries are maintained when messages are
295 * If the heads indicate available messages, the length in the header
296 * tells the start next message. A length == 0 for the next message
297 * indicates a wrap-around to the beginning of the buffer.
299 * Every record carries the monotonic timestamp in microseconds, as well as
300 * the standard userspace syslog level and syslog facility. The usual
301 * kernel messages use LOG_KERN; userspace-injected messages always carry
302 * a matching syslog facility, by default LOG_USER. The origin of every
303 * message can be reliably determined that way.
305 * The human readable log message directly follows the message header. The
306 * length of the message text is stored in the header, the stored message
309 * Optionally, a message can carry a dictionary of properties (key/value pairs),
310 * to provide userspace with a machine-readable message context.
312 * Examples for well-defined, commonly used property names are:
313 * DEVICE=b12:8 device identifier
317 * +sound:card0 subsystem:devname
318 * SUBSYSTEM=pci driver-core subsystem name
320 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
321 * follows directly after a '=' character. Every property is terminated by
322 * a '\0' character. The last property is not terminated.
324 * Example of a message structure:
325 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
326 * 0008 34 00 record is 52 bytes long
327 * 000a 0b 00 text is 11 bytes long
328 * 000c 1f 00 dictionary is 23 bytes long
329 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
330 * 0010 69 74 27 73 20 61 20 6c "it's a l"
332 * 001b 44 45 56 49 43 "DEVIC"
333 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
334 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
336 * 0032 00 00 00 padding to next message header
338 * The 'struct printk_log' buffer header 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.
342 * /dev/kmsg exports the structured data in the following line format:
343 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
345 * Users of the export format should ignore possible additional values
346 * separated by ',', and find the message after the ';' character.
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.
354 LOG_NEWLINE = 2, /* text ended with a newline */
355 LOG_CONT = 8, /* text is a fragment of a continuation line */
359 u64 ts_nsec; /* timestamp in nanoseconds */
360 u16 len; /* length of entire record */
361 u16 text_len; /* length of text buffer */
362 u16 dict_len; /* length of dictionary buffer */
363 u8 facility; /* syslog facility */
364 u8 flags:5; /* internal record flags */
365 u8 level:3; /* syslog level */
366 #ifdef CONFIG_PRINTK_CALLER
367 u32 caller_id; /* thread id or processor id */
370 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
371 __packed __aligned(4)
376 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
377 * within the scheduler's rq lock. It must be released before calling
378 * console_unlock() or anything else that might wake up a process.
380 DEFINE_RAW_SPINLOCK(logbuf_lock);
383 * Helper macros to lock/unlock logbuf_lock and switch between
384 * printk-safe/unsafe modes.
386 #define logbuf_lock_irq() \
388 printk_safe_enter_irq(); \
389 raw_spin_lock(&logbuf_lock); \
392 #define logbuf_unlock_irq() \
394 raw_spin_unlock(&logbuf_lock); \
395 printk_safe_exit_irq(); \
398 #define logbuf_lock_irqsave(flags) \
400 printk_safe_enter_irqsave(flags); \
401 raw_spin_lock(&logbuf_lock); \
404 #define logbuf_unlock_irqrestore(flags) \
406 raw_spin_unlock(&logbuf_lock); \
407 printk_safe_exit_irqrestore(flags); \
411 DECLARE_WAIT_QUEUE_HEAD(log_wait);
412 /* the next printk record to read by syslog(READ) or /proc/kmsg */
413 static u64 syslog_seq;
414 static u32 syslog_idx;
415 static size_t syslog_partial;
416 static bool syslog_time;
418 /* index and sequence number of the first record stored in the buffer */
419 static u64 log_first_seq;
420 static u32 log_first_idx;
422 /* index and sequence number of the next record to store in the buffer */
423 static u64 log_next_seq;
424 static u32 log_next_idx;
426 /* the next printk record to write to the console */
427 static u64 console_seq;
428 static u32 console_idx;
429 static u64 exclusive_console_stop_seq;
431 /* the next printk record to read after the last 'clear' command */
432 static u64 clear_seq;
433 static u32 clear_idx;
435 #ifdef CONFIG_PRINTK_CALLER
436 #define PREFIX_MAX 48
438 #define PREFIX_MAX 32
440 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
442 #define LOG_LEVEL(v) ((v) & 0x07)
443 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
446 #define LOG_ALIGN __alignof__(struct printk_log)
447 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
448 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
449 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
450 static char *log_buf = __log_buf;
451 static u32 log_buf_len = __LOG_BUF_LEN;
453 /* Return log buffer address */
454 char *log_buf_addr_get(void)
459 /* Return log buffer size */
460 u32 log_buf_len_get(void)
465 /* human readable text of the record */
466 static char *log_text(const struct printk_log *msg)
468 return (char *)msg + sizeof(struct printk_log);
471 /* optional key/value pair dictionary attached to the record */
472 static char *log_dict(const struct printk_log *msg)
474 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
477 /* get record by index; idx must point to valid msg */
478 static struct printk_log *log_from_idx(u32 idx)
480 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
483 * A length == 0 record is the end of buffer marker. Wrap around and
484 * read the message at the start of the buffer.
487 return (struct printk_log *)log_buf;
491 /* get next record; idx must point to valid msg */
492 static u32 log_next(u32 idx)
494 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
496 /* length == 0 indicates the end of the buffer; wrap */
498 * A length == 0 record is the end of buffer marker. Wrap around and
499 * read the message at the start of the buffer as *this* one, and
500 * return the one after that.
503 msg = (struct printk_log *)log_buf;
506 return idx + msg->len;
510 * Check whether there is enough free space for the given message.
512 * The same values of first_idx and next_idx mean that the buffer
513 * is either empty or full.
515 * If the buffer is empty, we must respect the position of the indexes.
516 * They cannot be reset to the beginning of the buffer.
518 static int logbuf_has_space(u32 msg_size, bool empty)
522 if (log_next_idx > log_first_idx || empty)
523 free = max(log_buf_len - log_next_idx, log_first_idx);
525 free = log_first_idx - log_next_idx;
528 * We need space also for an empty header that signalizes wrapping
531 return free >= msg_size + sizeof(struct printk_log);
534 static int log_make_free_space(u32 msg_size)
536 while (log_first_seq < log_next_seq &&
537 !logbuf_has_space(msg_size, false)) {
538 /* drop old messages until we have enough contiguous space */
539 log_first_idx = log_next(log_first_idx);
543 if (clear_seq < log_first_seq) {
544 clear_seq = log_first_seq;
545 clear_idx = log_first_idx;
548 /* sequence numbers are equal, so the log buffer is empty */
549 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
555 /* compute the message size including the padding bytes */
556 static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
560 size = sizeof(struct printk_log) + text_len + dict_len;
561 *pad_len = (-size) & (LOG_ALIGN - 1);
568 * Define how much of the log buffer we could take at maximum. The value
569 * must be greater than two. Note that only half of the buffer is available
570 * when the index points to the middle.
572 #define MAX_LOG_TAKE_PART 4
573 static const char trunc_msg[] = "<truncated>";
575 static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
576 u16 *dict_len, u32 *pad_len)
579 * The message should not take the whole buffer. Otherwise, it might
580 * get removed too soon.
582 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
583 if (*text_len > max_text_len)
584 *text_len = max_text_len;
585 /* enable the warning message */
586 *trunc_msg_len = strlen(trunc_msg);
587 /* disable the "dict" completely */
589 /* compute the size again, count also the warning message */
590 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
593 /* insert record into the buffer, discard old ones, update heads */
594 static int log_store(u32 caller_id, int facility, int level,
595 enum log_flags flags, u64 ts_nsec,
596 const char *dict, u16 dict_len,
597 const char *text, u16 text_len)
599 struct printk_log *msg;
601 u16 trunc_msg_len = 0;
603 /* number of '\0' padding bytes to next message */
604 size = msg_used_size(text_len, dict_len, &pad_len);
606 if (log_make_free_space(size)) {
607 /* truncate the message if it is too long for empty buffer */
608 size = truncate_msg(&text_len, &trunc_msg_len,
609 &dict_len, &pad_len);
610 /* survive when the log buffer is too small for trunc_msg */
611 if (log_make_free_space(size))
615 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
617 * This message + an additional empty header does not fit
618 * at the end of the buffer. Add an empty header with len == 0
619 * to signify a wrap around.
621 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
626 msg = (struct printk_log *)(log_buf + log_next_idx);
627 memcpy(log_text(msg), text, text_len);
628 msg->text_len = text_len;
630 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
631 msg->text_len += trunc_msg_len;
633 memcpy(log_dict(msg), dict, dict_len);
634 msg->dict_len = dict_len;
635 msg->facility = facility;
636 msg->level = level & 7;
637 msg->flags = flags & 0x1f;
639 msg->ts_nsec = ts_nsec;
641 msg->ts_nsec = local_clock();
642 #ifdef CONFIG_PRINTK_CALLER
643 msg->caller_id = caller_id;
645 memset(log_dict(msg) + dict_len, 0, pad_len);
649 log_next_idx += msg->len;
652 return msg->text_len;
655 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
657 static int syslog_action_restricted(int type)
662 * Unless restricted, we allow "read all" and "get buffer size"
665 return type != SYSLOG_ACTION_READ_ALL &&
666 type != SYSLOG_ACTION_SIZE_BUFFER;
669 static int check_syslog_permissions(int type, int source)
672 * If this is from /proc/kmsg and we've already opened it, then we've
673 * already done the capabilities checks at open time.
675 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
678 if (syslog_action_restricted(type)) {
679 if (capable(CAP_SYSLOG))
682 * For historical reasons, accept CAP_SYS_ADMIN too, with
685 if (capable(CAP_SYS_ADMIN)) {
686 pr_warn_once("%s (%d): Attempt to access syslog with "
687 "CAP_SYS_ADMIN but no CAP_SYSLOG "
689 current->comm, task_pid_nr(current));
695 return security_syslog(type);
698 static void append_char(char **pp, char *e, char c)
704 static ssize_t msg_print_ext_header(char *buf, size_t size,
705 struct printk_log *msg, u64 seq)
707 u64 ts_usec = msg->ts_nsec;
709 #ifdef CONFIG_PRINTK_CALLER
710 u32 id = msg->caller_id;
712 snprintf(caller, sizeof(caller), ",caller=%c%u",
713 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
718 do_div(ts_usec, 1000);
720 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
721 (msg->facility << 3) | msg->level, seq, ts_usec,
722 msg->flags & LOG_CONT ? 'c' : '-', caller);
725 static ssize_t msg_print_ext_body(char *buf, size_t size,
726 char *dict, size_t dict_len,
727 char *text, size_t text_len)
729 char *p = buf, *e = buf + size;
732 /* escape non-printable characters */
733 for (i = 0; i < text_len; i++) {
734 unsigned char c = text[i];
736 if (c < ' ' || c >= 127 || c == '\\')
737 p += scnprintf(p, e - p, "\\x%02x", c);
739 append_char(&p, e, c);
741 append_char(&p, e, '\n');
746 for (i = 0; i < dict_len; i++) {
747 unsigned char c = dict[i];
750 append_char(&p, e, ' ');
755 append_char(&p, e, '\n');
760 if (c < ' ' || c >= 127 || c == '\\') {
761 p += scnprintf(p, e - p, "\\x%02x", c);
765 append_char(&p, e, c);
767 append_char(&p, e, '\n');
773 /* /dev/kmsg - userspace message inject/listen interface */
774 struct devkmsg_user {
777 struct ratelimit_state rs;
779 char buf[CONSOLE_EXT_LOG_MAX];
782 static __printf(3, 4) __cold
783 int devkmsg_emit(int facility, int level, const char *fmt, ...)
789 r = vprintk_emit(facility, level, NULL, 0, fmt, args);
795 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
798 int level = default_message_loglevel;
799 int facility = 1; /* LOG_USER */
800 struct file *file = iocb->ki_filp;
801 struct devkmsg_user *user = file->private_data;
802 size_t len = iov_iter_count(from);
805 if (!user || len > LOG_LINE_MAX)
808 /* Ignore when user logging is disabled. */
809 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
812 /* Ratelimit when not explicitly enabled. */
813 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
814 if (!___ratelimit(&user->rs, current->comm))
818 buf = kmalloc(len+1, GFP_KERNEL);
823 if (!copy_from_iter_full(buf, len, from)) {
829 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
830 * the decimal value represents 32bit, the lower 3 bit are the log
831 * level, the rest are the log facility.
833 * If no prefix or no userspace facility is specified, we
834 * enforce LOG_USER, to be able to reliably distinguish
835 * kernel-generated messages from userspace-injected ones.
838 if (line[0] == '<') {
842 u = simple_strtoul(line + 1, &endp, 10);
843 if (endp && endp[0] == '>') {
844 level = LOG_LEVEL(u);
845 if (LOG_FACILITY(u) != 0)
846 facility = LOG_FACILITY(u);
853 devkmsg_emit(facility, level, "%s", line);
858 static ssize_t devkmsg_read(struct file *file, char __user *buf,
859 size_t count, loff_t *ppos)
861 struct devkmsg_user *user = file->private_data;
862 struct printk_log *msg;
869 ret = mutex_lock_interruptible(&user->lock);
874 while (user->seq == log_next_seq) {
875 if (file->f_flags & O_NONBLOCK) {
882 ret = wait_event_interruptible(log_wait,
883 user->seq != log_next_seq);
889 if (user->seq < log_first_seq) {
890 /* our last seen message is gone, return error and reset */
891 user->idx = log_first_idx;
892 user->seq = log_first_seq;
898 msg = log_from_idx(user->idx);
899 len = msg_print_ext_header(user->buf, sizeof(user->buf),
901 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
902 log_dict(msg), msg->dict_len,
903 log_text(msg), msg->text_len);
905 user->idx = log_next(user->idx);
914 if (copy_to_user(buf, user->buf, len)) {
920 mutex_unlock(&user->lock);
924 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
926 struct devkmsg_user *user = file->private_data;
937 /* the first record */
938 user->idx = log_first_idx;
939 user->seq = log_first_seq;
943 * The first record after the last SYSLOG_ACTION_CLEAR,
944 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
945 * changes no global state, and does not clear anything.
947 user->idx = clear_idx;
948 user->seq = clear_seq;
951 /* after the last record */
952 user->idx = log_next_idx;
953 user->seq = log_next_seq;
962 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
964 struct devkmsg_user *user = file->private_data;
968 return EPOLLERR|EPOLLNVAL;
970 poll_wait(file, &log_wait, wait);
973 if (user->seq < log_next_seq) {
974 /* return error when data has vanished underneath us */
975 if (user->seq < log_first_seq)
976 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
978 ret = EPOLLIN|EPOLLRDNORM;
985 static int devkmsg_open(struct inode *inode, struct file *file)
987 struct devkmsg_user *user;
990 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
993 /* write-only does not need any file context */
994 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
995 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
1001 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
1005 ratelimit_default_init(&user->rs);
1006 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
1008 mutex_init(&user->lock);
1011 user->idx = log_first_idx;
1012 user->seq = log_first_seq;
1013 logbuf_unlock_irq();
1015 file->private_data = user;
1019 static int devkmsg_release(struct inode *inode, struct file *file)
1021 struct devkmsg_user *user = file->private_data;
1026 ratelimit_state_exit(&user->rs);
1028 mutex_destroy(&user->lock);
1033 const struct file_operations kmsg_fops = {
1034 .open = devkmsg_open,
1035 .read = devkmsg_read,
1036 .write_iter = devkmsg_write,
1037 .llseek = devkmsg_llseek,
1038 .poll = devkmsg_poll,
1039 .release = devkmsg_release,
1042 #ifdef CONFIG_CRASH_CORE
1044 * This appends the listed symbols to /proc/vmcore
1046 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1047 * obtain access to symbols that are otherwise very difficult to locate. These
1048 * symbols are specifically used so that utilities can access and extract the
1049 * dmesg log from a vmcore file after a crash.
1051 void log_buf_vmcoreinfo_setup(void)
1053 VMCOREINFO_SYMBOL(log_buf);
1054 VMCOREINFO_SYMBOL(log_buf_len);
1055 VMCOREINFO_SYMBOL(log_first_idx);
1056 VMCOREINFO_SYMBOL(clear_idx);
1057 VMCOREINFO_SYMBOL(log_next_idx);
1059 * Export struct printk_log size and field offsets. User space tools can
1060 * parse it and detect any changes to structure down the line.
1062 VMCOREINFO_STRUCT_SIZE(printk_log);
1063 VMCOREINFO_OFFSET(printk_log, ts_nsec);
1064 VMCOREINFO_OFFSET(printk_log, len);
1065 VMCOREINFO_OFFSET(printk_log, text_len);
1066 VMCOREINFO_OFFSET(printk_log, dict_len);
1067 #ifdef CONFIG_PRINTK_CALLER
1068 VMCOREINFO_OFFSET(printk_log, caller_id);
1073 /* requested log_buf_len from kernel cmdline */
1074 static unsigned long __initdata new_log_buf_len;
1076 /* we practice scaling the ring buffer by powers of 2 */
1077 static void __init log_buf_len_update(u64 size)
1079 if (size > (u64)LOG_BUF_LEN_MAX) {
1080 size = (u64)LOG_BUF_LEN_MAX;
1081 pr_err("log_buf over 2G is not supported.\n");
1085 size = roundup_pow_of_two(size);
1086 if (size > log_buf_len)
1087 new_log_buf_len = (unsigned long)size;
1090 /* save requested log_buf_len since it's too early to process it */
1091 static int __init log_buf_len_setup(char *str)
1098 size = memparse(str, &str);
1100 log_buf_len_update(size);
1104 early_param("log_buf_len", log_buf_len_setup);
1107 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1109 static void __init log_buf_add_cpu(void)
1111 unsigned int cpu_extra;
1114 * archs should set up cpu_possible_bits properly with
1115 * set_cpu_possible() after setup_arch() but just in
1116 * case lets ensure this is valid.
1118 if (num_possible_cpus() == 1)
1121 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1123 /* by default this will only continue through for large > 64 CPUs */
1124 if (cpu_extra <= __LOG_BUF_LEN / 2)
1127 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1128 __LOG_CPU_MAX_BUF_LEN);
1129 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1131 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1133 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1135 #else /* !CONFIG_SMP */
1136 static inline void log_buf_add_cpu(void) {}
1137 #endif /* CONFIG_SMP */
1139 void __init setup_log_buf(int early)
1141 unsigned long flags;
1145 if (log_buf != __log_buf)
1148 if (!early && !new_log_buf_len)
1151 if (!new_log_buf_len)
1154 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1155 if (unlikely(!new_log_buf)) {
1156 pr_err("log_buf_len: %lu bytes not available\n",
1161 logbuf_lock_irqsave(flags);
1162 log_buf_len = new_log_buf_len;
1163 log_buf = new_log_buf;
1164 new_log_buf_len = 0;
1165 free = __LOG_BUF_LEN - log_next_idx;
1166 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1167 logbuf_unlock_irqrestore(flags);
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);
1174 static bool __read_mostly ignore_loglevel;
1176 static int __init ignore_loglevel_setup(char *str)
1178 ignore_loglevel = true;
1179 pr_info("debug: ignoring loglevel setting.\n");
1184 early_param("ignore_loglevel", ignore_loglevel_setup);
1185 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1186 MODULE_PARM_DESC(ignore_loglevel,
1187 "ignore loglevel setting (prints all kernel messages to the console)");
1189 static bool suppress_message_printing(int level)
1191 return (level >= console_loglevel && !ignore_loglevel);
1194 #ifdef CONFIG_BOOT_PRINTK_DELAY
1196 static int boot_delay; /* msecs delay after each printk during bootup */
1197 static unsigned long long loops_per_msec; /* based on boot_delay */
1199 static int __init boot_delay_setup(char *str)
1203 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1204 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1206 get_option(&str, &boot_delay);
1207 if (boot_delay > 10 * 1000)
1210 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1211 "HZ: %d, loops_per_msec: %llu\n",
1212 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1215 early_param("boot_delay", boot_delay_setup);
1217 static void boot_delay_msec(int level)
1219 unsigned long long k;
1220 unsigned long timeout;
1222 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1223 || suppress_message_printing(level)) {
1227 k = (unsigned long long)loops_per_msec * boot_delay;
1229 timeout = jiffies + msecs_to_jiffies(boot_delay);
1234 * use (volatile) jiffies to prevent
1235 * compiler reduction; loop termination via jiffies
1236 * is secondary and may or may not happen.
1238 if (time_after(jiffies, timeout))
1240 touch_nmi_watchdog();
1244 static inline void boot_delay_msec(int level)
1249 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1250 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1252 static size_t print_syslog(unsigned int level, char *buf)
1254 return sprintf(buf, "<%u>", level);
1257 static size_t print_time(u64 ts, char *buf)
1259 unsigned long rem_nsec = do_div(ts, 1000000000);
1261 return sprintf(buf, "[%5lu.%06lu]",
1262 (unsigned long)ts, rem_nsec / 1000);
1265 #ifdef CONFIG_PRINTK_CALLER
1266 static size_t print_caller(u32 id, char *buf)
1270 snprintf(caller, sizeof(caller), "%c%u",
1271 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1272 return sprintf(buf, "[%6s]", caller);
1275 #define print_caller(id, buf) 0
1278 static size_t print_prefix(const struct printk_log *msg, bool syslog,
1279 bool time, char *buf)
1284 len = print_syslog((msg->facility << 3) | msg->level, buf);
1287 len += print_time(msg->ts_nsec, buf + len);
1289 len += print_caller(msg->caller_id, buf + len);
1291 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1299 static size_t msg_print_text(const struct printk_log *msg, bool syslog,
1300 bool time, char *buf, size_t size)
1302 const char *text = log_text(msg);
1303 size_t text_size = msg->text_len;
1305 char prefix[PREFIX_MAX];
1306 const size_t prefix_len = print_prefix(msg, syslog, time, prefix);
1309 const char *next = memchr(text, '\n', text_size);
1313 text_len = next - text;
1315 text_size -= next - text;
1317 text_len = text_size;
1321 if (prefix_len + text_len + 1 >= size - len)
1324 memcpy(buf + len, prefix, prefix_len);
1326 memcpy(buf + len, text, text_len);
1330 /* SYSLOG_ACTION_* buffer size only calculation */
1331 len += prefix_len + text_len + 1;
1340 static int syslog_print(char __user *buf, int size)
1343 struct printk_log *msg;
1346 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1355 if (syslog_seq < log_first_seq) {
1356 /* messages are gone, move to first one */
1357 syslog_seq = log_first_seq;
1358 syslog_idx = log_first_idx;
1361 if (syslog_seq == log_next_seq) {
1362 logbuf_unlock_irq();
1367 * To keep reading/counting partial line consistent,
1368 * use printk_time value as of the beginning of a line.
1370 if (!syslog_partial)
1371 syslog_time = printk_time;
1373 skip = syslog_partial;
1374 msg = log_from_idx(syslog_idx);
1375 n = msg_print_text(msg, true, syslog_time, text,
1376 LOG_LINE_MAX + PREFIX_MAX);
1377 if (n - syslog_partial <= size) {
1378 /* message fits into buffer, move forward */
1379 syslog_idx = log_next(syslog_idx);
1381 n -= syslog_partial;
1384 /* partial read(), remember position */
1386 syslog_partial += n;
1389 logbuf_unlock_irq();
1394 if (copy_to_user(buf, text + skip, n)) {
1409 static int syslog_print_all(char __user *buf, int size, bool clear)
1418 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1425 * Find first record that fits, including all following records,
1426 * into the user-provided buffer for this dump.
1430 while (seq < log_next_seq) {
1431 struct printk_log *msg = log_from_idx(idx);
1433 len += msg_print_text(msg, true, time, NULL, 0);
1434 idx = log_next(idx);
1438 /* move first record forward until length fits into the buffer */
1441 while (len > size && 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 /* last message fitting into this dump */
1450 next_seq = log_next_seq;
1453 while (len >= 0 && seq < next_seq) {
1454 struct printk_log *msg = log_from_idx(idx);
1455 int textlen = msg_print_text(msg, true, time, text,
1456 LOG_LINE_MAX + PREFIX_MAX);
1458 idx = log_next(idx);
1461 logbuf_unlock_irq();
1462 if (copy_to_user(buf + len, text, textlen))
1468 if (seq < log_first_seq) {
1469 /* messages are gone, move to next one */
1470 seq = log_first_seq;
1471 idx = log_first_idx;
1476 clear_seq = log_next_seq;
1477 clear_idx = log_next_idx;
1479 logbuf_unlock_irq();
1485 static void syslog_clear(void)
1488 clear_seq = log_next_seq;
1489 clear_idx = log_next_idx;
1490 logbuf_unlock_irq();
1493 int do_syslog(int type, char __user *buf, int len, int source)
1496 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1499 error = check_syslog_permissions(type, source);
1504 case SYSLOG_ACTION_CLOSE: /* Close log */
1506 case SYSLOG_ACTION_OPEN: /* Open log */
1508 case SYSLOG_ACTION_READ: /* Read from log */
1509 if (!buf || len < 0)
1513 if (!access_ok(buf, len))
1515 error = wait_event_interruptible(log_wait,
1516 syslog_seq != log_next_seq);
1519 error = syslog_print(buf, len);
1521 /* Read/clear last kernel messages */
1522 case SYSLOG_ACTION_READ_CLEAR:
1525 /* Read last kernel messages */
1526 case SYSLOG_ACTION_READ_ALL:
1527 if (!buf || len < 0)
1531 if (!access_ok(buf, len))
1533 error = syslog_print_all(buf, len, clear);
1535 /* Clear ring buffer */
1536 case SYSLOG_ACTION_CLEAR:
1539 /* Disable logging to console */
1540 case SYSLOG_ACTION_CONSOLE_OFF:
1541 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1542 saved_console_loglevel = console_loglevel;
1543 console_loglevel = minimum_console_loglevel;
1545 /* Enable logging to console */
1546 case SYSLOG_ACTION_CONSOLE_ON:
1547 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1548 console_loglevel = saved_console_loglevel;
1549 saved_console_loglevel = LOGLEVEL_DEFAULT;
1552 /* Set level of messages printed to console */
1553 case SYSLOG_ACTION_CONSOLE_LEVEL:
1554 if (len < 1 || len > 8)
1556 if (len < minimum_console_loglevel)
1557 len = minimum_console_loglevel;
1558 console_loglevel = len;
1559 /* Implicitly re-enable logging to console */
1560 saved_console_loglevel = LOGLEVEL_DEFAULT;
1562 /* Number of chars in the log buffer */
1563 case SYSLOG_ACTION_SIZE_UNREAD:
1565 if (syslog_seq < log_first_seq) {
1566 /* messages are gone, move to first one */
1567 syslog_seq = log_first_seq;
1568 syslog_idx = log_first_idx;
1571 if (source == SYSLOG_FROM_PROC) {
1573 * Short-cut for poll(/"proc/kmsg") which simply checks
1574 * for pending data, not the size; return the count of
1575 * records, not the length.
1577 error = log_next_seq - syslog_seq;
1579 u64 seq = syslog_seq;
1580 u32 idx = syslog_idx;
1581 bool time = syslog_partial ? syslog_time : printk_time;
1583 while (seq < log_next_seq) {
1584 struct printk_log *msg = log_from_idx(idx);
1586 error += msg_print_text(msg, true, time, NULL,
1589 idx = log_next(idx);
1592 error -= syslog_partial;
1594 logbuf_unlock_irq();
1596 /* Size of the log buffer */
1597 case SYSLOG_ACTION_SIZE_BUFFER:
1598 error = log_buf_len;
1608 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1610 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1614 * Special console_lock variants that help to reduce the risk of soft-lockups.
1615 * They allow to pass console_lock to another printk() call using a busy wait.
1618 #ifdef CONFIG_LOCKDEP
1619 static struct lockdep_map console_owner_dep_map = {
1620 .name = "console_owner"
1624 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1625 static struct task_struct *console_owner;
1626 static bool console_waiter;
1629 * console_lock_spinning_enable - mark beginning of code where another
1630 * thread might safely busy wait
1632 * This basically converts console_lock into a spinlock. This marks
1633 * the section where the console_lock owner can not sleep, because
1634 * there may be a waiter spinning (like a spinlock). Also it must be
1635 * ready to hand over the lock at the end of the section.
1637 static void console_lock_spinning_enable(void)
1639 raw_spin_lock(&console_owner_lock);
1640 console_owner = current;
1641 raw_spin_unlock(&console_owner_lock);
1643 /* The waiter may spin on us after setting console_owner */
1644 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1648 * console_lock_spinning_disable_and_check - mark end of code where another
1649 * thread was able to busy wait and check if there is a waiter
1651 * This is called at the end of the section where spinning is allowed.
1652 * It has two functions. First, it is a signal that it is no longer
1653 * safe to start busy waiting for the lock. Second, it checks if
1654 * there is a busy waiter and passes the lock rights to her.
1656 * Important: Callers lose the lock if there was a busy waiter.
1657 * They must not touch items synchronized by console_lock
1660 * Return: 1 if the lock rights were passed, 0 otherwise.
1662 static int console_lock_spinning_disable_and_check(void)
1666 raw_spin_lock(&console_owner_lock);
1667 waiter = READ_ONCE(console_waiter);
1668 console_owner = NULL;
1669 raw_spin_unlock(&console_owner_lock);
1672 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1676 /* The waiter is now free to continue */
1677 WRITE_ONCE(console_waiter, false);
1679 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1682 * Hand off console_lock to waiter. The waiter will perform
1683 * the up(). After this, the waiter is the console_lock owner.
1685 mutex_release(&console_lock_dep_map, 1, _THIS_IP_);
1690 * console_trylock_spinning - try to get console_lock by busy waiting
1692 * This allows to busy wait for the console_lock when the current
1693 * owner is running in specially marked sections. It means that
1694 * the current owner is running and cannot reschedule until it
1695 * is ready to lose the lock.
1697 * Return: 1 if we got the lock, 0 othrewise
1699 static int console_trylock_spinning(void)
1701 struct task_struct *owner = NULL;
1704 unsigned long flags;
1706 if (console_trylock())
1709 printk_safe_enter_irqsave(flags);
1711 raw_spin_lock(&console_owner_lock);
1712 owner = READ_ONCE(console_owner);
1713 waiter = READ_ONCE(console_waiter);
1714 if (!waiter && owner && owner != current) {
1715 WRITE_ONCE(console_waiter, true);
1718 raw_spin_unlock(&console_owner_lock);
1721 * If there is an active printk() writing to the
1722 * consoles, instead of having it write our data too,
1723 * see if we can offload that load from the active
1724 * printer, and do some printing ourselves.
1725 * Go into a spin only if there isn't already a waiter
1726 * spinning, and there is an active printer, and
1727 * that active printer isn't us (recursive printk?).
1730 printk_safe_exit_irqrestore(flags);
1734 /* We spin waiting for the owner to release us */
1735 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1736 /* Owner will clear console_waiter on hand off */
1737 while (READ_ONCE(console_waiter))
1739 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1741 printk_safe_exit_irqrestore(flags);
1743 * The owner passed the console lock to us.
1744 * Since we did not spin on console lock, annotate
1745 * this as a trylock. Otherwise lockdep will
1748 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1754 * Call the console drivers, asking them to write out
1755 * log_buf[start] to log_buf[end - 1].
1756 * The console_lock must be held.
1758 static void call_console_drivers(const char *ext_text, size_t ext_len,
1759 const char *text, size_t len)
1761 struct console *con;
1763 trace_console_rcuidle(text, len);
1765 if (!console_drivers)
1768 for_each_console(con) {
1769 if (exclusive_console && con != exclusive_console)
1771 if (!(con->flags & CON_ENABLED))
1775 if (!cpu_online(smp_processor_id()) &&
1776 !(con->flags & CON_ANYTIME))
1778 if (con->flags & CON_EXTENDED)
1779 con->write(con, ext_text, ext_len);
1781 con->write(con, text, len);
1785 int printk_delay_msec __read_mostly;
1787 static inline void printk_delay(void)
1789 if (unlikely(printk_delay_msec)) {
1790 int m = printk_delay_msec;
1794 touch_nmi_watchdog();
1799 static inline u32 printk_caller_id(void)
1801 return in_task() ? task_pid_nr(current) :
1802 0x80000000 + raw_smp_processor_id();
1806 * Continuation lines are buffered, and not committed to the record buffer
1807 * until the line is complete, or a race forces it. The line fragments
1808 * though, are printed immediately to the consoles to ensure everything has
1809 * reached the console in case of a kernel crash.
1811 static struct cont {
1812 char buf[LOG_LINE_MAX];
1813 size_t len; /* length == 0 means unused buffer */
1814 u32 caller_id; /* printk_caller_id() of first print */
1815 u64 ts_nsec; /* time of first print */
1816 u8 level; /* log level of first message */
1817 u8 facility; /* log facility of first message */
1818 enum log_flags flags; /* prefix, newline flags */
1821 static void cont_flush(void)
1826 log_store(cont.caller_id, cont.facility, cont.level, cont.flags,
1827 cont.ts_nsec, NULL, 0, cont.buf, cont.len);
1831 static bool cont_add(u32 caller_id, int facility, int level,
1832 enum log_flags flags, const char *text, size_t len)
1834 /* If the line gets too long, split it up in separate records. */
1835 if (cont.len + len > sizeof(cont.buf)) {
1841 cont.facility = facility;
1843 cont.caller_id = caller_id;
1844 cont.ts_nsec = local_clock();
1848 memcpy(cont.buf + cont.len, text, len);
1851 // The original flags come from the first line,
1852 // but later continuations can add a newline.
1853 if (flags & LOG_NEWLINE) {
1854 cont.flags |= LOG_NEWLINE;
1861 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)
1863 const u32 caller_id = printk_caller_id();
1866 * If an earlier line was buffered, and we're a continuation
1867 * write from the same context, try to add it to the buffer.
1870 if (cont.caller_id == caller_id && (lflags & LOG_CONT)) {
1871 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1874 /* Otherwise, make sure it's flushed */
1878 /* Skip empty continuation lines that couldn't be added - they just flush */
1879 if (!text_len && (lflags & LOG_CONT))
1882 /* If it doesn't end in a newline, try to buffer the current line */
1883 if (!(lflags & LOG_NEWLINE)) {
1884 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1888 /* Store it in the record log */
1889 return log_store(caller_id, facility, level, lflags, 0,
1890 dict, dictlen, text, text_len);
1893 /* Must be called under logbuf_lock. */
1894 int vprintk_store(int facility, int level,
1895 const char *dict, size_t dictlen,
1896 const char *fmt, va_list args)
1898 static char textbuf[LOG_LINE_MAX];
1899 char *text = textbuf;
1901 enum log_flags lflags = 0;
1904 * The printf needs to come first; we need the syslog
1905 * prefix which might be passed-in as a parameter.
1907 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1909 /* mark and strip a trailing newline */
1910 if (text_len && text[text_len-1] == '\n') {
1912 lflags |= LOG_NEWLINE;
1915 /* strip kernel syslog prefix and extract log level or control flags */
1916 if (facility == 0) {
1919 while ((kern_level = printk_get_level(text)) != 0) {
1920 switch (kern_level) {
1922 if (level == LOGLEVEL_DEFAULT)
1923 level = kern_level - '0';
1925 case 'c': /* KERN_CONT */
1934 if (level == LOGLEVEL_DEFAULT)
1935 level = default_message_loglevel;
1938 lflags |= LOG_NEWLINE;
1940 return log_output(facility, level, lflags,
1941 dict, dictlen, text, text_len);
1944 asmlinkage int vprintk_emit(int facility, int level,
1945 const char *dict, size_t dictlen,
1946 const char *fmt, va_list args)
1949 bool in_sched = false, pending_output;
1950 unsigned long flags;
1953 /* Suppress unimportant messages after panic happens */
1954 if (unlikely(suppress_printk))
1957 if (level == LOGLEVEL_SCHED) {
1958 level = LOGLEVEL_DEFAULT;
1962 boot_delay_msec(level);
1965 /* This stops the holder of console_sem just where we want him */
1966 logbuf_lock_irqsave(flags);
1967 curr_log_seq = log_next_seq;
1968 printed_len = vprintk_store(facility, level, dict, dictlen, fmt, args);
1969 pending_output = (curr_log_seq != log_next_seq);
1970 logbuf_unlock_irqrestore(flags);
1972 /* If called from the scheduler, we can not call up(). */
1973 if (!in_sched && pending_output) {
1975 * Disable preemption to avoid being preempted while holding
1976 * console_sem which would prevent anyone from printing to
1981 * Try to acquire and then immediately release the console
1982 * semaphore. The release will print out buffers and wake up
1983 * /dev/kmsg and syslog() users.
1985 if (console_trylock_spinning())
1994 EXPORT_SYMBOL(vprintk_emit);
1996 asmlinkage int vprintk(const char *fmt, va_list args)
1998 return vprintk_func(fmt, args);
2000 EXPORT_SYMBOL(vprintk);
2002 int vprintk_default(const char *fmt, va_list args)
2006 #ifdef CONFIG_KGDB_KDB
2007 /* Allow to pass printk() to kdb but avoid a recursion. */
2008 if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
2009 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
2013 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
2017 EXPORT_SYMBOL_GPL(vprintk_default);
2020 * printk - print a kernel message
2021 * @fmt: format string
2023 * This is printk(). It can be called from any context. We want it to work.
2025 * We try to grab the console_lock. If we succeed, it's easy - we log the
2026 * output and call the console drivers. If we fail to get the semaphore, we
2027 * place the output into the log buffer and return. The current holder of
2028 * the console_sem will notice the new output in console_unlock(); and will
2029 * send it to the consoles before releasing the lock.
2031 * One effect of this deferred printing is that code which calls printk() and
2032 * then changes console_loglevel may break. This is because console_loglevel
2033 * is inspected when the actual printing occurs.
2038 * See the vsnprintf() documentation for format string extensions over C99.
2040 asmlinkage __visible int printk(const char *fmt, ...)
2045 va_start(args, fmt);
2046 r = vprintk_func(fmt, args);
2051 EXPORT_SYMBOL(printk);
2053 #else /* CONFIG_PRINTK */
2055 #define LOG_LINE_MAX 0
2056 #define PREFIX_MAX 0
2057 #define printk_time false
2059 static u64 syslog_seq;
2060 static u32 syslog_idx;
2061 static u64 console_seq;
2062 static u32 console_idx;
2063 static u64 exclusive_console_stop_seq;
2064 static u64 log_first_seq;
2065 static u32 log_first_idx;
2066 static u64 log_next_seq;
2067 static char *log_text(const struct printk_log *msg) { return NULL; }
2068 static char *log_dict(const struct printk_log *msg) { return NULL; }
2069 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2070 static u32 log_next(u32 idx) { return 0; }
2071 static ssize_t msg_print_ext_header(char *buf, size_t size,
2072 struct printk_log *msg,
2073 u64 seq) { return 0; }
2074 static ssize_t msg_print_ext_body(char *buf, size_t size,
2075 char *dict, size_t dict_len,
2076 char *text, size_t text_len) { return 0; }
2077 static void console_lock_spinning_enable(void) { }
2078 static int console_lock_spinning_disable_and_check(void) { return 0; }
2079 static void call_console_drivers(const char *ext_text, size_t ext_len,
2080 const char *text, size_t len) {}
2081 static size_t msg_print_text(const struct printk_log *msg, bool syslog,
2082 bool time, char *buf, size_t size) { return 0; }
2083 static bool suppress_message_printing(int level) { return false; }
2085 #endif /* CONFIG_PRINTK */
2087 #ifdef CONFIG_EARLY_PRINTK
2088 struct console *early_console;
2090 asmlinkage __visible void early_printk(const char *fmt, ...)
2100 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2103 early_console->write(early_console, buf, n);
2107 static int __add_preferred_console(char *name, int idx, char *options,
2110 struct console_cmdline *c;
2114 * See if this tty is not yet registered, and
2115 * if we have a slot free.
2117 for (i = 0, c = console_cmdline;
2118 i < MAX_CMDLINECONSOLES && c->name[0];
2120 if (strcmp(c->name, name) == 0 && c->index == idx) {
2122 preferred_console = i;
2126 if (i == MAX_CMDLINECONSOLES)
2129 preferred_console = i;
2130 strlcpy(c->name, name, sizeof(c->name));
2131 c->options = options;
2132 braille_set_options(c, brl_options);
2138 static int __init console_msg_format_setup(char *str)
2140 if (!strcmp(str, "syslog"))
2141 console_msg_format = MSG_FORMAT_SYSLOG;
2142 if (!strcmp(str, "default"))
2143 console_msg_format = MSG_FORMAT_DEFAULT;
2146 __setup("console_msg_format=", console_msg_format_setup);
2149 * Set up a console. Called via do_early_param() in init/main.c
2150 * for each "console=" parameter in the boot command line.
2152 static int __init console_setup(char *str)
2154 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2155 char *s, *options, *brl_options = NULL;
2158 if (_braille_console_setup(&str, &brl_options))
2162 * Decode str into name, index, options.
2164 if (str[0] >= '0' && str[0] <= '9') {
2165 strcpy(buf, "ttyS");
2166 strncpy(buf + 4, str, sizeof(buf) - 5);
2168 strncpy(buf, str, sizeof(buf) - 1);
2170 buf[sizeof(buf) - 1] = 0;
2171 options = strchr(str, ',');
2175 if (!strcmp(str, "ttya"))
2176 strcpy(buf, "ttyS0");
2177 if (!strcmp(str, "ttyb"))
2178 strcpy(buf, "ttyS1");
2180 for (s = buf; *s; s++)
2181 if (isdigit(*s) || *s == ',')
2183 idx = simple_strtoul(s, NULL, 10);
2186 __add_preferred_console(buf, idx, options, brl_options);
2187 console_set_on_cmdline = 1;
2190 __setup("console=", console_setup);
2193 * add_preferred_console - add a device to the list of preferred consoles.
2194 * @name: device name
2195 * @idx: device index
2196 * @options: options for this console
2198 * The last preferred console added will be used for kernel messages
2199 * and stdin/out/err for init. Normally this is used by console_setup
2200 * above to handle user-supplied console arguments; however it can also
2201 * be used by arch-specific code either to override the user or more
2202 * commonly to provide a default console (ie from PROM variables) when
2203 * the user has not supplied one.
2205 int add_preferred_console(char *name, int idx, char *options)
2207 return __add_preferred_console(name, idx, options, NULL);
2210 bool console_suspend_enabled = true;
2211 EXPORT_SYMBOL(console_suspend_enabled);
2213 static int __init console_suspend_disable(char *str)
2215 console_suspend_enabled = false;
2218 __setup("no_console_suspend", console_suspend_disable);
2219 module_param_named(console_suspend, console_suspend_enabled,
2220 bool, S_IRUGO | S_IWUSR);
2221 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2222 " and hibernate operations");
2225 * suspend_console - suspend the console subsystem
2227 * This disables printk() while we go into suspend states
2229 void suspend_console(void)
2231 if (!console_suspend_enabled)
2233 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2235 console_suspended = 1;
2239 void resume_console(void)
2241 if (!console_suspend_enabled)
2244 console_suspended = 0;
2249 * console_cpu_notify - print deferred console messages after CPU hotplug
2252 * If printk() is called from a CPU that is not online yet, the messages
2253 * will be printed on the console only if there are CON_ANYTIME consoles.
2254 * This function is called when a new CPU comes online (or fails to come
2255 * up) or goes offline.
2257 static int console_cpu_notify(unsigned int cpu)
2259 if (!cpuhp_tasks_frozen) {
2260 /* If trylock fails, someone else is doing the printing */
2261 if (console_trylock())
2268 * console_lock - lock the console system for exclusive use.
2270 * Acquires a lock which guarantees that the caller has
2271 * exclusive access to the console system and the console_drivers list.
2273 * Can sleep, returns nothing.
2275 void console_lock(void)
2280 if (console_suspended)
2283 console_may_schedule = 1;
2285 EXPORT_SYMBOL(console_lock);
2288 * console_trylock - try to lock the console system for exclusive use.
2290 * Try to acquire a lock which guarantees that the caller has exclusive
2291 * access to the console system and the console_drivers list.
2293 * returns 1 on success, and 0 on failure to acquire the lock.
2295 int console_trylock(void)
2297 if (down_trylock_console_sem())
2299 if (console_suspended) {
2304 console_may_schedule = 0;
2307 EXPORT_SYMBOL(console_trylock);
2309 int is_console_locked(void)
2311 return console_locked;
2313 EXPORT_SYMBOL(is_console_locked);
2316 * Check if we have any console that is capable of printing while cpu is
2317 * booting or shutting down. Requires console_sem.
2319 static int have_callable_console(void)
2321 struct console *con;
2323 for_each_console(con)
2324 if ((con->flags & CON_ENABLED) &&
2325 (con->flags & CON_ANYTIME))
2332 * Can we actually use the console at this time on this cpu?
2334 * Console drivers may assume that per-cpu resources have been allocated. So
2335 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2336 * call them until this CPU is officially up.
2338 static inline int can_use_console(void)
2340 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2344 * console_unlock - unlock the console system
2346 * Releases the console_lock which the caller holds on the console system
2347 * and the console driver list.
2349 * While the console_lock was held, console output may have been buffered
2350 * by printk(). If this is the case, console_unlock(); emits
2351 * the output prior to releasing the lock.
2353 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2355 * console_unlock(); may be called from any context.
2357 void console_unlock(void)
2359 static char ext_text[CONSOLE_EXT_LOG_MAX];
2360 static char text[LOG_LINE_MAX + PREFIX_MAX];
2361 unsigned long flags;
2362 bool do_cond_resched, retry;
2364 if (console_suspended) {
2370 * Console drivers are called with interrupts disabled, so
2371 * @console_may_schedule should be cleared before; however, we may
2372 * end up dumping a lot of lines, for example, if called from
2373 * console registration path, and should invoke cond_resched()
2374 * between lines if allowable. Not doing so can cause a very long
2375 * scheduling stall on a slow console leading to RCU stall and
2376 * softlockup warnings which exacerbate the issue with more
2377 * messages practically incapacitating the system.
2379 * console_trylock() is not able to detect the preemptive
2380 * context reliably. Therefore the value must be stored before
2381 * and cleared after the the "again" goto label.
2383 do_cond_resched = console_may_schedule;
2385 console_may_schedule = 0;
2388 * We released the console_sem lock, so we need to recheck if
2389 * cpu is online and (if not) is there at least one CON_ANYTIME
2392 if (!can_use_console()) {
2399 struct printk_log *msg;
2403 printk_safe_enter_irqsave(flags);
2404 raw_spin_lock(&logbuf_lock);
2405 if (console_seq < log_first_seq) {
2407 "** %llu printk messages dropped **\n",
2408 log_first_seq - console_seq);
2410 /* messages are gone, move to first one */
2411 console_seq = log_first_seq;
2412 console_idx = log_first_idx;
2417 if (console_seq == log_next_seq)
2420 msg = log_from_idx(console_idx);
2421 if (suppress_message_printing(msg->level)) {
2423 * Skip record we have buffered and already printed
2424 * directly to the console when we received it, and
2425 * record that has level above the console loglevel.
2427 console_idx = log_next(console_idx);
2432 /* Output to all consoles once old messages replayed. */
2433 if (unlikely(exclusive_console &&
2434 console_seq >= exclusive_console_stop_seq)) {
2435 exclusive_console = NULL;
2438 len += msg_print_text(msg,
2439 console_msg_format & MSG_FORMAT_SYSLOG,
2440 printk_time, text + len, sizeof(text) - len);
2441 if (nr_ext_console_drivers) {
2442 ext_len = msg_print_ext_header(ext_text,
2445 ext_len += msg_print_ext_body(ext_text + ext_len,
2446 sizeof(ext_text) - ext_len,
2447 log_dict(msg), msg->dict_len,
2448 log_text(msg), msg->text_len);
2450 console_idx = log_next(console_idx);
2452 raw_spin_unlock(&logbuf_lock);
2455 * While actively printing out messages, if another printk()
2456 * were to occur on another CPU, it may wait for this one to
2457 * finish. This task can not be preempted if there is a
2458 * waiter waiting to take over.
2460 console_lock_spinning_enable();
2462 stop_critical_timings(); /* don't trace print latency */
2463 call_console_drivers(ext_text, ext_len, text, len);
2464 start_critical_timings();
2466 if (console_lock_spinning_disable_and_check()) {
2467 printk_safe_exit_irqrestore(flags);
2471 printk_safe_exit_irqrestore(flags);
2473 if (do_cond_resched)
2479 raw_spin_unlock(&logbuf_lock);
2484 * Someone could have filled up the buffer again, so re-check if there's
2485 * something to flush. In case we cannot trylock the console_sem again,
2486 * there's a new owner and the console_unlock() from them will do the
2487 * flush, no worries.
2489 raw_spin_lock(&logbuf_lock);
2490 retry = console_seq != log_next_seq;
2491 raw_spin_unlock(&logbuf_lock);
2492 printk_safe_exit_irqrestore(flags);
2494 if (retry && console_trylock())
2497 EXPORT_SYMBOL(console_unlock);
2500 * console_conditional_schedule - yield the CPU if required
2502 * If the console code is currently allowed to sleep, and
2503 * if this CPU should yield the CPU to another task, do
2506 * Must be called within console_lock();.
2508 void __sched console_conditional_schedule(void)
2510 if (console_may_schedule)
2513 EXPORT_SYMBOL(console_conditional_schedule);
2515 void console_unblank(void)
2520 * console_unblank can no longer be called in interrupt context unless
2521 * oops_in_progress is set to 1..
2523 if (oops_in_progress) {
2524 if (down_trylock_console_sem() != 0)
2530 console_may_schedule = 0;
2532 if ((c->flags & CON_ENABLED) && c->unblank)
2538 * console_flush_on_panic - flush console content on panic
2539 * @mode: flush all messages in buffer or just the pending ones
2541 * Immediately output all pending messages no matter what.
2543 void console_flush_on_panic(enum con_flush_mode mode)
2546 * If someone else is holding the console lock, trylock will fail
2547 * and may_schedule may be set. Ignore and proceed to unlock so
2548 * that messages are flushed out. As this can be called from any
2549 * context and we don't want to get preempted while flushing,
2550 * ensure may_schedule is cleared.
2553 console_may_schedule = 0;
2555 if (mode == CONSOLE_REPLAY_ALL) {
2556 unsigned long flags;
2558 logbuf_lock_irqsave(flags);
2559 console_seq = log_first_seq;
2560 console_idx = log_first_idx;
2561 logbuf_unlock_irqrestore(flags);
2567 * Return the console tty driver structure and its associated index
2569 struct tty_driver *console_device(int *index)
2572 struct tty_driver *driver = NULL;
2575 for_each_console(c) {
2578 driver = c->device(c, index);
2587 * Prevent further output on the passed console device so that (for example)
2588 * serial drivers can disable console output before suspending a port, and can
2589 * re-enable output afterwards.
2591 void console_stop(struct console *console)
2594 console->flags &= ~CON_ENABLED;
2597 EXPORT_SYMBOL(console_stop);
2599 void console_start(struct console *console)
2602 console->flags |= CON_ENABLED;
2605 EXPORT_SYMBOL(console_start);
2607 static int __read_mostly keep_bootcon;
2609 static int __init keep_bootcon_setup(char *str)
2612 pr_info("debug: skip boot console de-registration.\n");
2617 early_param("keep_bootcon", keep_bootcon_setup);
2620 * The console driver calls this routine during kernel initialization
2621 * to register the console printing procedure with printk() and to
2622 * print any messages that were printed by the kernel before the
2623 * console driver was initialized.
2625 * This can happen pretty early during the boot process (because of
2626 * early_printk) - sometimes before setup_arch() completes - be careful
2627 * of what kernel features are used - they may not be initialised yet.
2629 * There are two types of consoles - bootconsoles (early_printk) and
2630 * "real" consoles (everything which is not a bootconsole) which are
2631 * handled differently.
2632 * - Any number of bootconsoles can be registered at any time.
2633 * - As soon as a "real" console is registered, all bootconsoles
2634 * will be unregistered automatically.
2635 * - Once a "real" console is registered, any attempt to register a
2636 * bootconsoles will be rejected
2638 void register_console(struct console *newcon)
2641 unsigned long flags;
2642 struct console *bcon = NULL;
2643 struct console_cmdline *c;
2644 static bool has_preferred;
2646 if (console_drivers)
2647 for_each_console(bcon)
2648 if (WARN(bcon == newcon,
2649 "console '%s%d' already registered\n",
2650 bcon->name, bcon->index))
2654 * before we register a new CON_BOOT console, make sure we don't
2655 * already have a valid console
2657 if (console_drivers && newcon->flags & CON_BOOT) {
2658 /* find the last or real console */
2659 for_each_console(bcon) {
2660 if (!(bcon->flags & CON_BOOT)) {
2661 pr_info("Too late to register bootconsole %s%d\n",
2662 newcon->name, newcon->index);
2668 if (console_drivers && console_drivers->flags & CON_BOOT)
2669 bcon = console_drivers;
2671 if (!has_preferred || bcon || !console_drivers)
2672 has_preferred = preferred_console >= 0;
2675 * See if we want to use this console driver. If we
2676 * didn't select a console we take the first one
2677 * that registers here.
2679 if (!has_preferred) {
2680 if (newcon->index < 0)
2682 if (newcon->setup == NULL ||
2683 newcon->setup(newcon, NULL) == 0) {
2684 newcon->flags |= CON_ENABLED;
2685 if (newcon->device) {
2686 newcon->flags |= CON_CONSDEV;
2687 has_preferred = true;
2693 * See if this console matches one we selected on
2696 for (i = 0, c = console_cmdline;
2697 i < MAX_CMDLINECONSOLES && c->name[0];
2699 if (!newcon->match ||
2700 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2701 /* default matching */
2702 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2703 if (strcmp(c->name, newcon->name) != 0)
2705 if (newcon->index >= 0 &&
2706 newcon->index != c->index)
2708 if (newcon->index < 0)
2709 newcon->index = c->index;
2711 if (_braille_register_console(newcon, c))
2714 if (newcon->setup &&
2715 newcon->setup(newcon, c->options) != 0)
2719 newcon->flags |= CON_ENABLED;
2720 if (i == preferred_console) {
2721 newcon->flags |= CON_CONSDEV;
2722 has_preferred = true;
2727 if (!(newcon->flags & CON_ENABLED))
2731 * If we have a bootconsole, and are switching to a real console,
2732 * don't print everything out again, since when the boot console, and
2733 * the real console are the same physical device, it's annoying to
2734 * see the beginning boot messages twice
2736 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2737 newcon->flags &= ~CON_PRINTBUFFER;
2740 * Put this console in the list - keep the
2741 * preferred driver at the head of the list.
2744 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2745 newcon->next = console_drivers;
2746 console_drivers = newcon;
2748 newcon->next->flags &= ~CON_CONSDEV;
2750 newcon->next = console_drivers->next;
2751 console_drivers->next = newcon;
2754 if (newcon->flags & CON_EXTENDED)
2755 nr_ext_console_drivers++;
2757 if (newcon->flags & CON_PRINTBUFFER) {
2759 * console_unlock(); will print out the buffered messages
2762 logbuf_lock_irqsave(flags);
2763 console_seq = syslog_seq;
2764 console_idx = syslog_idx;
2766 * We're about to replay the log buffer. Only do this to the
2767 * just-registered console to avoid excessive message spam to
2768 * the already-registered consoles.
2770 * Set exclusive_console with disabled interrupts to reduce
2771 * race window with eventual console_flush_on_panic() that
2772 * ignores console_lock.
2774 exclusive_console = newcon;
2775 exclusive_console_stop_seq = console_seq;
2776 logbuf_unlock_irqrestore(flags);
2779 console_sysfs_notify();
2782 * By unregistering the bootconsoles after we enable the real console
2783 * we get the "console xxx enabled" message on all the consoles -
2784 * boot consoles, real consoles, etc - this is to ensure that end
2785 * users know there might be something in the kernel's log buffer that
2786 * went to the bootconsole (that they do not see on the real console)
2788 pr_info("%sconsole [%s%d] enabled\n",
2789 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2790 newcon->name, newcon->index);
2792 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2794 /* We need to iterate through all boot consoles, to make
2795 * sure we print everything out, before we unregister them.
2797 for_each_console(bcon)
2798 if (bcon->flags & CON_BOOT)
2799 unregister_console(bcon);
2802 EXPORT_SYMBOL(register_console);
2804 int unregister_console(struct console *console)
2806 struct console *a, *b;
2809 pr_info("%sconsole [%s%d] disabled\n",
2810 (console->flags & CON_BOOT) ? "boot" : "" ,
2811 console->name, console->index);
2813 res = _braille_unregister_console(console);
2819 if (console_drivers == console) {
2820 console_drivers=console->next;
2822 } else if (console_drivers) {
2823 for (a=console_drivers->next, b=console_drivers ;
2824 a; b=a, a=b->next) {
2833 if (!res && (console->flags & CON_EXTENDED))
2834 nr_ext_console_drivers--;
2837 * If this isn't the last console and it has CON_CONSDEV set, we
2838 * need to set it on the next preferred console.
2840 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2841 console_drivers->flags |= CON_CONSDEV;
2843 console->flags &= ~CON_ENABLED;
2845 console_sysfs_notify();
2848 EXPORT_SYMBOL(unregister_console);
2851 * Initialize the console device. This is called *early*, so
2852 * we can't necessarily depend on lots of kernel help here.
2853 * Just do some early initializations, and do the complex setup
2856 void __init console_init(void)
2860 initcall_entry_t *ce;
2862 /* Setup the default TTY line discipline. */
2866 * set up the console device so that later boot sequences can
2867 * inform about problems etc..
2869 ce = __con_initcall_start;
2870 trace_initcall_level("console");
2871 while (ce < __con_initcall_end) {
2872 call = initcall_from_entry(ce);
2873 trace_initcall_start(call);
2875 trace_initcall_finish(call, ret);
2881 * Some boot consoles access data that is in the init section and which will
2882 * be discarded after the initcalls have been run. To make sure that no code
2883 * will access this data, unregister the boot consoles in a late initcall.
2885 * If for some reason, such as deferred probe or the driver being a loadable
2886 * module, the real console hasn't registered yet at this point, there will
2887 * be a brief interval in which no messages are logged to the console, which
2888 * makes it difficult to diagnose problems that occur during this time.
2890 * To mitigate this problem somewhat, only unregister consoles whose memory
2891 * intersects with the init section. Note that all other boot consoles will
2892 * get unregistred when the real preferred console is registered.
2894 static int __init printk_late_init(void)
2896 struct console *con;
2899 for_each_console(con) {
2900 if (!(con->flags & CON_BOOT))
2903 /* Check addresses that might be used for enabled consoles. */
2904 if (init_section_intersects(con, sizeof(*con)) ||
2905 init_section_contains(con->write, 0) ||
2906 init_section_contains(con->read, 0) ||
2907 init_section_contains(con->device, 0) ||
2908 init_section_contains(con->unblank, 0) ||
2909 init_section_contains(con->data, 0)) {
2911 * Please, consider moving the reported consoles out
2912 * of the init section.
2914 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2915 con->name, con->index);
2916 unregister_console(con);
2919 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2920 console_cpu_notify);
2922 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2923 console_cpu_notify, NULL);
2927 late_initcall(printk_late_init);
2929 #if defined CONFIG_PRINTK
2931 * Delayed printk version, for scheduler-internal messages:
2933 #define PRINTK_PENDING_WAKEUP 0x01
2934 #define PRINTK_PENDING_OUTPUT 0x02
2936 static DEFINE_PER_CPU(int, printk_pending);
2938 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2940 int pending = __this_cpu_xchg(printk_pending, 0);
2942 if (pending & PRINTK_PENDING_OUTPUT) {
2943 /* If trylock fails, someone else is doing the printing */
2944 if (console_trylock())
2948 if (pending & PRINTK_PENDING_WAKEUP)
2949 wake_up_interruptible(&log_wait);
2952 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2953 .func = wake_up_klogd_work_func,
2954 .flags = IRQ_WORK_LAZY,
2957 void wake_up_klogd(void)
2960 if (waitqueue_active(&log_wait)) {
2961 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2962 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2967 void defer_console_output(void)
2970 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2971 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2975 int vprintk_deferred(const char *fmt, va_list args)
2979 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2980 defer_console_output();
2985 int printk_deferred(const char *fmt, ...)
2990 va_start(args, fmt);
2991 r = vprintk_deferred(fmt, args);
2998 * printk rate limiting, lifted from the networking subsystem.
3000 * This enforces a rate limit: not more than 10 kernel messages
3001 * every 5s to make a denial-of-service attack impossible.
3003 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3005 int __printk_ratelimit(const char *func)
3007 return ___ratelimit(&printk_ratelimit_state, func);
3009 EXPORT_SYMBOL(__printk_ratelimit);
3012 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3013 * @caller_jiffies: pointer to caller's state
3014 * @interval_msecs: minimum interval between prints
3016 * printk_timed_ratelimit() returns true if more than @interval_msecs
3017 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3020 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3021 unsigned int interval_msecs)
3023 unsigned long elapsed = jiffies - *caller_jiffies;
3025 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3028 *caller_jiffies = jiffies;
3031 EXPORT_SYMBOL(printk_timed_ratelimit);
3033 static DEFINE_SPINLOCK(dump_list_lock);
3034 static LIST_HEAD(dump_list);
3037 * kmsg_dump_register - register a kernel log dumper.
3038 * @dumper: pointer to the kmsg_dumper structure
3040 * Adds a kernel log dumper to the system. The dump callback in the
3041 * structure will be called when the kernel oopses or panics and must be
3042 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3044 int kmsg_dump_register(struct kmsg_dumper *dumper)
3046 unsigned long flags;
3049 /* The dump callback needs to be set */
3053 spin_lock_irqsave(&dump_list_lock, flags);
3054 /* Don't allow registering multiple times */
3055 if (!dumper->registered) {
3056 dumper->registered = 1;
3057 list_add_tail_rcu(&dumper->list, &dump_list);
3060 spin_unlock_irqrestore(&dump_list_lock, flags);
3064 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3067 * kmsg_dump_unregister - unregister a kmsg dumper.
3068 * @dumper: pointer to the kmsg_dumper structure
3070 * Removes a dump device from the system. Returns zero on success and
3071 * %-EINVAL otherwise.
3073 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3075 unsigned long flags;
3078 spin_lock_irqsave(&dump_list_lock, flags);
3079 if (dumper->registered) {
3080 dumper->registered = 0;
3081 list_del_rcu(&dumper->list);
3084 spin_unlock_irqrestore(&dump_list_lock, flags);
3089 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3091 static bool always_kmsg_dump;
3092 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3095 * kmsg_dump - dump kernel log to kernel message dumpers.
3096 * @reason: the reason (oops, panic etc) for dumping
3098 * Call each of the registered dumper's dump() callback, which can
3099 * retrieve the kmsg records with kmsg_dump_get_line() or
3100 * kmsg_dump_get_buffer().
3102 void kmsg_dump(enum kmsg_dump_reason reason)
3104 struct kmsg_dumper *dumper;
3105 unsigned long flags;
3107 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3111 list_for_each_entry_rcu(dumper, &dump_list, list) {
3112 if (dumper->max_reason && reason > dumper->max_reason)
3115 /* initialize iterator with data about the stored records */
3116 dumper->active = true;
3118 logbuf_lock_irqsave(flags);
3119 dumper->cur_seq = clear_seq;
3120 dumper->cur_idx = clear_idx;
3121 dumper->next_seq = log_next_seq;
3122 dumper->next_idx = log_next_idx;
3123 logbuf_unlock_irqrestore(flags);
3125 /* invoke dumper which will iterate over records */
3126 dumper->dump(dumper, reason);
3128 /* reset iterator */
3129 dumper->active = false;
3135 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3136 * @dumper: registered kmsg dumper
3137 * @syslog: include the "<4>" prefixes
3138 * @line: buffer to copy the line to
3139 * @size: maximum size of the buffer
3140 * @len: length of line placed into buffer
3142 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3143 * record, and copy one record into the provided buffer.
3145 * Consecutive calls will return the next available record moving
3146 * towards the end of the buffer with the youngest messages.
3148 * A return value of FALSE indicates that there are no more records to
3151 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3153 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3154 char *line, size_t size, size_t *len)
3156 struct printk_log *msg;
3160 if (!dumper->active)
3163 if (dumper->cur_seq < log_first_seq) {
3164 /* messages are gone, move to first available one */
3165 dumper->cur_seq = log_first_seq;
3166 dumper->cur_idx = log_first_idx;
3170 if (dumper->cur_seq >= log_next_seq)
3173 msg = log_from_idx(dumper->cur_idx);
3174 l = msg_print_text(msg, syslog, printk_time, line, size);
3176 dumper->cur_idx = log_next(dumper->cur_idx);
3186 * kmsg_dump_get_line - retrieve one kmsg log line
3187 * @dumper: registered kmsg dumper
3188 * @syslog: include the "<4>" prefixes
3189 * @line: buffer to copy the line to
3190 * @size: maximum size of the buffer
3191 * @len: length of line placed into buffer
3193 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3194 * record, and copy one record into the provided buffer.
3196 * Consecutive calls will return the next available record moving
3197 * towards the end of the buffer with the youngest messages.
3199 * A return value of FALSE indicates that there are no more records to
3202 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3203 char *line, size_t size, size_t *len)
3205 unsigned long flags;
3208 logbuf_lock_irqsave(flags);
3209 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3210 logbuf_unlock_irqrestore(flags);
3214 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3217 * kmsg_dump_get_buffer - copy kmsg log lines
3218 * @dumper: registered kmsg dumper
3219 * @syslog: include the "<4>" prefixes
3220 * @buf: buffer to copy the line to
3221 * @size: maximum size of the buffer
3222 * @len: length of line placed into buffer
3224 * Start at the end of the kmsg buffer and fill the provided buffer
3225 * with as many of the the *youngest* kmsg records that fit into it.
3226 * If the buffer is large enough, all available kmsg records will be
3227 * copied with a single call.
3229 * Consecutive calls will fill the buffer with the next block of
3230 * available older records, not including the earlier retrieved ones.
3232 * A return value of FALSE indicates that there are no more records to
3235 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3236 char *buf, size_t size, size_t *len)
3238 unsigned long flags;
3245 bool time = printk_time;
3247 if (!dumper->active)
3250 logbuf_lock_irqsave(flags);
3251 if (dumper->cur_seq < log_first_seq) {
3252 /* messages are gone, move to first available one */
3253 dumper->cur_seq = log_first_seq;
3254 dumper->cur_idx = log_first_idx;
3258 if (dumper->cur_seq >= dumper->next_seq) {
3259 logbuf_unlock_irqrestore(flags);
3263 /* calculate length of entire buffer */
3264 seq = dumper->cur_seq;
3265 idx = dumper->cur_idx;
3266 while (seq < dumper->next_seq) {
3267 struct printk_log *msg = log_from_idx(idx);
3269 l += msg_print_text(msg, true, time, NULL, 0);
3270 idx = log_next(idx);
3274 /* move first record forward until length fits into the buffer */
3275 seq = dumper->cur_seq;
3276 idx = dumper->cur_idx;
3277 while (l > size && seq < dumper->next_seq) {
3278 struct printk_log *msg = log_from_idx(idx);
3280 l -= msg_print_text(msg, true, time, NULL, 0);
3281 idx = log_next(idx);
3285 /* last message in next interation */
3290 while (seq < dumper->next_seq) {
3291 struct printk_log *msg = log_from_idx(idx);
3293 l += msg_print_text(msg, syslog, time, buf + l, size - l);
3294 idx = log_next(idx);
3298 dumper->next_seq = next_seq;
3299 dumper->next_idx = next_idx;
3301 logbuf_unlock_irqrestore(flags);
3307 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3310 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3311 * @dumper: registered kmsg dumper
3313 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3314 * kmsg_dump_get_buffer() can be called again and used multiple
3315 * times within the same dumper.dump() callback.
3317 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3319 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3321 dumper->cur_seq = clear_seq;
3322 dumper->cur_idx = clear_idx;
3323 dumper->next_seq = log_next_seq;
3324 dumper->next_idx = log_next_idx;
3328 * kmsg_dump_rewind - reset the interator
3329 * @dumper: registered kmsg dumper
3331 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3332 * kmsg_dump_get_buffer() can be called again and used multiple
3333 * times within the same dumper.dump() callback.
3335 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3337 unsigned long flags;
3339 logbuf_lock_irqsave(flags);
3340 kmsg_dump_rewind_nolock(dumper);
3341 logbuf_unlock_irqrestore(flags);
3343 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);