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) {
150 pr_warn("printk.devkmsg: bad option string '%s'\n", str);
155 * Set sysctl string accordingly:
157 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
158 strcpy(devkmsg_log_str, "on");
159 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
160 strcpy(devkmsg_log_str, "off");
161 /* else "ratelimit" which is set by default. */
164 * Sysctl cannot change it anymore. The kernel command line setting of
165 * this parameter is to force the setting to be permanent throughout the
166 * runtime of the system. This is a precation measure against userspace
167 * trying to be a smarta** and attempting to change it up on us.
169 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
173 __setup("printk.devkmsg=", control_devkmsg);
175 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
177 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
178 void __user *buffer, size_t *lenp, loff_t *ppos)
180 char old_str[DEVKMSG_STR_MAX_SIZE];
185 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
189 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
192 err = proc_dostring(table, write, buffer, lenp, ppos);
197 err = __control_devkmsg(devkmsg_log_str);
200 * Do not accept an unknown string OR a known string with
203 if (err < 0 || (err + 1 != *lenp)) {
205 /* ... and restore old setting. */
207 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
216 /* Number of registered extended console drivers. */
217 static int nr_ext_console_drivers;
220 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
221 * macros instead of functions so that _RET_IP_ contains useful information.
223 #define down_console_sem() do { \
225 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
228 static int __down_trylock_console_sem(unsigned long ip)
234 * Here and in __up_console_sem() we need to be in safe mode,
235 * because spindump/WARN/etc from under console ->lock will
236 * deadlock in printk()->down_trylock_console_sem() otherwise.
238 printk_safe_enter_irqsave(flags);
239 lock_failed = down_trylock(&console_sem);
240 printk_safe_exit_irqrestore(flags);
244 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
247 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
249 static void __up_console_sem(unsigned long ip)
253 mutex_release(&console_lock_dep_map, 1, ip);
255 printk_safe_enter_irqsave(flags);
257 printk_safe_exit_irqrestore(flags);
259 #define up_console_sem() __up_console_sem(_RET_IP_)
262 * This is used for debugging the mess that is the VT code by
263 * keeping track if we have the console semaphore held. It's
264 * definitely not the perfect debug tool (we don't know if _WE_
265 * hold it and are racing, but it helps tracking those weird code
266 * paths in the console code where we end up in places I want
267 * locked without the console sempahore held).
269 static int console_locked, console_suspended;
272 * If exclusive_console is non-NULL then only this console is to be printed to.
274 static struct console *exclusive_console;
277 * Array of consoles built from command line options (console=)
280 #define MAX_CMDLINECONSOLES 8
282 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
284 static int preferred_console = -1;
285 int console_set_on_cmdline;
286 EXPORT_SYMBOL(console_set_on_cmdline);
288 /* Flag: console code may call schedule() */
289 static int console_may_schedule;
291 enum con_msg_format_flags {
292 MSG_FORMAT_DEFAULT = 0,
293 MSG_FORMAT_SYSLOG = (1 << 0),
296 static int console_msg_format = MSG_FORMAT_DEFAULT;
299 * The printk log buffer consists of a chain of concatenated variable
300 * length records. Every record starts with a record header, containing
301 * the overall length of the record.
303 * The heads to the first and last entry in the buffer, as well as the
304 * sequence numbers of these entries are maintained when messages are
307 * If the heads indicate available messages, the length in the header
308 * tells the start next message. A length == 0 for the next message
309 * indicates a wrap-around to the beginning of the buffer.
311 * Every record carries the monotonic timestamp in microseconds, as well as
312 * the standard userspace syslog level and syslog facility. The usual
313 * kernel messages use LOG_KERN; userspace-injected messages always carry
314 * a matching syslog facility, by default LOG_USER. The origin of every
315 * message can be reliably determined that way.
317 * The human readable log message directly follows the message header. The
318 * length of the message text is stored in the header, the stored message
321 * Optionally, a message can carry a dictionary of properties (key/value pairs),
322 * to provide userspace with a machine-readable message context.
324 * Examples for well-defined, commonly used property names are:
325 * DEVICE=b12:8 device identifier
329 * +sound:card0 subsystem:devname
330 * SUBSYSTEM=pci driver-core subsystem name
332 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
333 * follows directly after a '=' character. Every property is terminated by
334 * a '\0' character. The last property is not terminated.
336 * Example of a message structure:
337 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
338 * 0008 34 00 record is 52 bytes long
339 * 000a 0b 00 text is 11 bytes long
340 * 000c 1f 00 dictionary is 23 bytes long
341 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
342 * 0010 69 74 27 73 20 61 20 6c "it's a l"
344 * 001b 44 45 56 49 43 "DEVIC"
345 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
346 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
348 * 0032 00 00 00 padding to next message header
350 * The 'struct printk_log' buffer header must never be directly exported to
351 * userspace, it is a kernel-private implementation detail that might
352 * need to be changed in the future, when the requirements change.
354 * /dev/kmsg exports the structured data in the following line format:
355 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
357 * Users of the export format should ignore possible additional values
358 * separated by ',', and find the message after the ';' character.
360 * The optional key/value pairs are attached as continuation lines starting
361 * with a space character and terminated by a newline. All possible
362 * non-prinatable characters are escaped in the "\xff" notation.
366 LOG_NEWLINE = 2, /* text ended with a newline */
367 LOG_CONT = 8, /* text is a fragment of a continuation line */
371 u64 ts_nsec; /* timestamp in nanoseconds */
372 u16 len; /* length of entire record */
373 u16 text_len; /* length of text buffer */
374 u16 dict_len; /* length of dictionary buffer */
375 u8 facility; /* syslog facility */
376 u8 flags:5; /* internal record flags */
377 u8 level:3; /* syslog level */
378 #ifdef CONFIG_PRINTK_CALLER
379 u32 caller_id; /* thread id or processor id */
382 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
383 __packed __aligned(4)
388 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
389 * within the scheduler's rq lock. It must be released before calling
390 * console_unlock() or anything else that might wake up a process.
392 DEFINE_RAW_SPINLOCK(logbuf_lock);
395 * Helper macros to lock/unlock logbuf_lock and switch between
396 * printk-safe/unsafe modes.
398 #define logbuf_lock_irq() \
400 printk_safe_enter_irq(); \
401 raw_spin_lock(&logbuf_lock); \
404 #define logbuf_unlock_irq() \
406 raw_spin_unlock(&logbuf_lock); \
407 printk_safe_exit_irq(); \
410 #define logbuf_lock_irqsave(flags) \
412 printk_safe_enter_irqsave(flags); \
413 raw_spin_lock(&logbuf_lock); \
416 #define logbuf_unlock_irqrestore(flags) \
418 raw_spin_unlock(&logbuf_lock); \
419 printk_safe_exit_irqrestore(flags); \
423 DECLARE_WAIT_QUEUE_HEAD(log_wait);
424 /* the next printk record to read by syslog(READ) or /proc/kmsg */
425 static u64 syslog_seq;
426 static u32 syslog_idx;
427 static size_t syslog_partial;
428 static bool syslog_time;
430 /* index and sequence number of the first record stored in the buffer */
431 static u64 log_first_seq;
432 static u32 log_first_idx;
434 /* index and sequence number of the next record to store in the buffer */
435 static u64 log_next_seq;
436 static u32 log_next_idx;
438 /* the next printk record to write to the console */
439 static u64 console_seq;
440 static u32 console_idx;
441 static u64 exclusive_console_stop_seq;
443 /* the next printk record to read after the last 'clear' command */
444 static u64 clear_seq;
445 static u32 clear_idx;
447 #ifdef CONFIG_PRINTK_CALLER
448 #define PREFIX_MAX 48
450 #define PREFIX_MAX 32
452 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
454 #define LOG_LEVEL(v) ((v) & 0x07)
455 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
458 #define LOG_ALIGN __alignof__(struct printk_log)
459 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
460 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
461 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
462 static char *log_buf = __log_buf;
463 static u32 log_buf_len = __LOG_BUF_LEN;
466 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
467 * per_cpu_areas are initialised. This variable is set to true when
468 * it's safe to access per-CPU data.
470 static bool __printk_percpu_data_ready __read_mostly;
472 bool printk_percpu_data_ready(void)
474 return __printk_percpu_data_ready;
477 /* Return log buffer address */
478 char *log_buf_addr_get(void)
483 /* Return log buffer size */
484 u32 log_buf_len_get(void)
489 /* human readable text of the record */
490 static char *log_text(const struct printk_log *msg)
492 return (char *)msg + sizeof(struct printk_log);
495 /* optional key/value pair dictionary attached to the record */
496 static char *log_dict(const struct printk_log *msg)
498 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
501 /* get record by index; idx must point to valid msg */
502 static struct printk_log *log_from_idx(u32 idx)
504 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
507 * A length == 0 record is the end of buffer marker. Wrap around and
508 * read the message at the start of the buffer.
511 return (struct printk_log *)log_buf;
515 /* get next record; idx must point to valid msg */
516 static u32 log_next(u32 idx)
518 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
520 /* length == 0 indicates the end of the buffer; wrap */
522 * A length == 0 record is the end of buffer marker. Wrap around and
523 * read the message at the start of the buffer as *this* one, and
524 * return the one after that.
527 msg = (struct printk_log *)log_buf;
530 return idx + msg->len;
534 * Check whether there is enough free space for the given message.
536 * The same values of first_idx and next_idx mean that the buffer
537 * is either empty or full.
539 * If the buffer is empty, we must respect the position of the indexes.
540 * They cannot be reset to the beginning of the buffer.
542 static int logbuf_has_space(u32 msg_size, bool empty)
546 if (log_next_idx > log_first_idx || empty)
547 free = max(log_buf_len - log_next_idx, log_first_idx);
549 free = log_first_idx - log_next_idx;
552 * We need space also for an empty header that signalizes wrapping
555 return free >= msg_size + sizeof(struct printk_log);
558 static int log_make_free_space(u32 msg_size)
560 while (log_first_seq < log_next_seq &&
561 !logbuf_has_space(msg_size, false)) {
562 /* drop old messages until we have enough contiguous space */
563 log_first_idx = log_next(log_first_idx);
567 if (clear_seq < log_first_seq) {
568 clear_seq = log_first_seq;
569 clear_idx = log_first_idx;
572 /* sequence numbers are equal, so the log buffer is empty */
573 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
579 /* compute the message size including the padding bytes */
580 static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
584 size = sizeof(struct printk_log) + text_len + dict_len;
585 *pad_len = (-size) & (LOG_ALIGN - 1);
592 * Define how much of the log buffer we could take at maximum. The value
593 * must be greater than two. Note that only half of the buffer is available
594 * when the index points to the middle.
596 #define MAX_LOG_TAKE_PART 4
597 static const char trunc_msg[] = "<truncated>";
599 static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
600 u16 *dict_len, u32 *pad_len)
603 * The message should not take the whole buffer. Otherwise, it might
604 * get removed too soon.
606 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
607 if (*text_len > max_text_len)
608 *text_len = max_text_len;
609 /* enable the warning message */
610 *trunc_msg_len = strlen(trunc_msg);
611 /* disable the "dict" completely */
613 /* compute the size again, count also the warning message */
614 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
617 /* insert record into the buffer, discard old ones, update heads */
618 static int log_store(u32 caller_id, int facility, int level,
619 enum log_flags flags, u64 ts_nsec,
620 const char *dict, u16 dict_len,
621 const char *text, u16 text_len)
623 struct printk_log *msg;
625 u16 trunc_msg_len = 0;
627 /* number of '\0' padding bytes to next message */
628 size = msg_used_size(text_len, dict_len, &pad_len);
630 if (log_make_free_space(size)) {
631 /* truncate the message if it is too long for empty buffer */
632 size = truncate_msg(&text_len, &trunc_msg_len,
633 &dict_len, &pad_len);
634 /* survive when the log buffer is too small for trunc_msg */
635 if (log_make_free_space(size))
639 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
641 * This message + an additional empty header does not fit
642 * at the end of the buffer. Add an empty header with len == 0
643 * to signify a wrap around.
645 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
650 msg = (struct printk_log *)(log_buf + log_next_idx);
651 memcpy(log_text(msg), text, text_len);
652 msg->text_len = text_len;
654 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
655 msg->text_len += trunc_msg_len;
657 memcpy(log_dict(msg), dict, dict_len);
658 msg->dict_len = dict_len;
659 msg->facility = facility;
660 msg->level = level & 7;
661 msg->flags = flags & 0x1f;
663 msg->ts_nsec = ts_nsec;
665 msg->ts_nsec = local_clock();
666 #ifdef CONFIG_PRINTK_CALLER
667 msg->caller_id = caller_id;
669 memset(log_dict(msg) + dict_len, 0, pad_len);
673 log_next_idx += msg->len;
676 return msg->text_len;
679 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
681 static int syslog_action_restricted(int type)
686 * Unless restricted, we allow "read all" and "get buffer size"
689 return type != SYSLOG_ACTION_READ_ALL &&
690 type != SYSLOG_ACTION_SIZE_BUFFER;
693 static int check_syslog_permissions(int type, int source)
696 * If this is from /proc/kmsg and we've already opened it, then we've
697 * already done the capabilities checks at open time.
699 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
702 if (syslog_action_restricted(type)) {
703 if (capable(CAP_SYSLOG))
706 * For historical reasons, accept CAP_SYS_ADMIN too, with
709 if (capable(CAP_SYS_ADMIN)) {
710 pr_warn_once("%s (%d): Attempt to access syslog with "
711 "CAP_SYS_ADMIN but no CAP_SYSLOG "
713 current->comm, task_pid_nr(current));
719 return security_syslog(type);
722 static void append_char(char **pp, char *e, char c)
728 static ssize_t msg_print_ext_header(char *buf, size_t size,
729 struct printk_log *msg, u64 seq)
731 u64 ts_usec = msg->ts_nsec;
733 #ifdef CONFIG_PRINTK_CALLER
734 u32 id = msg->caller_id;
736 snprintf(caller, sizeof(caller), ",caller=%c%u",
737 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
742 do_div(ts_usec, 1000);
744 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
745 (msg->facility << 3) | msg->level, seq, ts_usec,
746 msg->flags & LOG_CONT ? 'c' : '-', caller);
749 static ssize_t msg_print_ext_body(char *buf, size_t size,
750 char *dict, size_t dict_len,
751 char *text, size_t text_len)
753 char *p = buf, *e = buf + size;
756 /* escape non-printable characters */
757 for (i = 0; i < text_len; i++) {
758 unsigned char c = text[i];
760 if (c < ' ' || c >= 127 || c == '\\')
761 p += scnprintf(p, e - p, "\\x%02x", c);
763 append_char(&p, e, c);
765 append_char(&p, e, '\n');
770 for (i = 0; i < dict_len; i++) {
771 unsigned char c = dict[i];
774 append_char(&p, e, ' ');
779 append_char(&p, e, '\n');
784 if (c < ' ' || c >= 127 || c == '\\') {
785 p += scnprintf(p, e - p, "\\x%02x", c);
789 append_char(&p, e, c);
791 append_char(&p, e, '\n');
797 /* /dev/kmsg - userspace message inject/listen interface */
798 struct devkmsg_user {
801 struct ratelimit_state rs;
803 char buf[CONSOLE_EXT_LOG_MAX];
806 static __printf(3, 4) __cold
807 int devkmsg_emit(int facility, int level, const char *fmt, ...)
813 r = vprintk_emit(facility, level, NULL, 0, fmt, args);
819 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
822 int level = default_message_loglevel;
823 int facility = 1; /* LOG_USER */
824 struct file *file = iocb->ki_filp;
825 struct devkmsg_user *user = file->private_data;
826 size_t len = iov_iter_count(from);
829 if (!user || len > LOG_LINE_MAX)
832 /* Ignore when user logging is disabled. */
833 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
836 /* Ratelimit when not explicitly enabled. */
837 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
838 if (!___ratelimit(&user->rs, current->comm))
842 buf = kmalloc(len+1, GFP_KERNEL);
847 if (!copy_from_iter_full(buf, len, from)) {
853 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
854 * the decimal value represents 32bit, the lower 3 bit are the log
855 * level, the rest are the log facility.
857 * If no prefix or no userspace facility is specified, we
858 * enforce LOG_USER, to be able to reliably distinguish
859 * kernel-generated messages from userspace-injected ones.
862 if (line[0] == '<') {
866 u = simple_strtoul(line + 1, &endp, 10);
867 if (endp && endp[0] == '>') {
868 level = LOG_LEVEL(u);
869 if (LOG_FACILITY(u) != 0)
870 facility = LOG_FACILITY(u);
877 devkmsg_emit(facility, level, "%s", line);
882 static ssize_t devkmsg_read(struct file *file, char __user *buf,
883 size_t count, loff_t *ppos)
885 struct devkmsg_user *user = file->private_data;
886 struct printk_log *msg;
893 ret = mutex_lock_interruptible(&user->lock);
898 while (user->seq == log_next_seq) {
899 if (file->f_flags & O_NONBLOCK) {
906 ret = wait_event_interruptible(log_wait,
907 user->seq != log_next_seq);
913 if (user->seq < log_first_seq) {
914 /* our last seen message is gone, return error and reset */
915 user->idx = log_first_idx;
916 user->seq = log_first_seq;
922 msg = log_from_idx(user->idx);
923 len = msg_print_ext_header(user->buf, sizeof(user->buf),
925 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
926 log_dict(msg), msg->dict_len,
927 log_text(msg), msg->text_len);
929 user->idx = log_next(user->idx);
938 if (copy_to_user(buf, user->buf, len)) {
944 mutex_unlock(&user->lock);
948 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
950 struct devkmsg_user *user = file->private_data;
961 /* the first record */
962 user->idx = log_first_idx;
963 user->seq = log_first_seq;
967 * The first record after the last SYSLOG_ACTION_CLEAR,
968 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
969 * changes no global state, and does not clear anything.
971 user->idx = clear_idx;
972 user->seq = clear_seq;
975 /* after the last record */
976 user->idx = log_next_idx;
977 user->seq = log_next_seq;
986 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
988 struct devkmsg_user *user = file->private_data;
992 return EPOLLERR|EPOLLNVAL;
994 poll_wait(file, &log_wait, wait);
997 if (user->seq < log_next_seq) {
998 /* return error when data has vanished underneath us */
999 if (user->seq < log_first_seq)
1000 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
1002 ret = EPOLLIN|EPOLLRDNORM;
1004 logbuf_unlock_irq();
1009 static int devkmsg_open(struct inode *inode, struct file *file)
1011 struct devkmsg_user *user;
1014 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
1017 /* write-only does not need any file context */
1018 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
1019 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
1020 SYSLOG_FROM_READER);
1025 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
1029 ratelimit_default_init(&user->rs);
1030 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
1032 mutex_init(&user->lock);
1035 user->idx = log_first_idx;
1036 user->seq = log_first_seq;
1037 logbuf_unlock_irq();
1039 file->private_data = user;
1043 static int devkmsg_release(struct inode *inode, struct file *file)
1045 struct devkmsg_user *user = file->private_data;
1050 ratelimit_state_exit(&user->rs);
1052 mutex_destroy(&user->lock);
1057 const struct file_operations kmsg_fops = {
1058 .open = devkmsg_open,
1059 .read = devkmsg_read,
1060 .write_iter = devkmsg_write,
1061 .llseek = devkmsg_llseek,
1062 .poll = devkmsg_poll,
1063 .release = devkmsg_release,
1066 #ifdef CONFIG_CRASH_CORE
1068 * This appends the listed symbols to /proc/vmcore
1070 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1071 * obtain access to symbols that are otherwise very difficult to locate. These
1072 * symbols are specifically used so that utilities can access and extract the
1073 * dmesg log from a vmcore file after a crash.
1075 void log_buf_vmcoreinfo_setup(void)
1077 VMCOREINFO_SYMBOL(log_buf);
1078 VMCOREINFO_SYMBOL(log_buf_len);
1079 VMCOREINFO_SYMBOL(log_first_idx);
1080 VMCOREINFO_SYMBOL(clear_idx);
1081 VMCOREINFO_SYMBOL(log_next_idx);
1083 * Export struct printk_log size and field offsets. User space tools can
1084 * parse it and detect any changes to structure down the line.
1086 VMCOREINFO_STRUCT_SIZE(printk_log);
1087 VMCOREINFO_OFFSET(printk_log, ts_nsec);
1088 VMCOREINFO_OFFSET(printk_log, len);
1089 VMCOREINFO_OFFSET(printk_log, text_len);
1090 VMCOREINFO_OFFSET(printk_log, dict_len);
1091 #ifdef CONFIG_PRINTK_CALLER
1092 VMCOREINFO_OFFSET(printk_log, caller_id);
1097 /* requested log_buf_len from kernel cmdline */
1098 static unsigned long __initdata new_log_buf_len;
1100 /* we practice scaling the ring buffer by powers of 2 */
1101 static void __init log_buf_len_update(u64 size)
1103 if (size > (u64)LOG_BUF_LEN_MAX) {
1104 size = (u64)LOG_BUF_LEN_MAX;
1105 pr_err("log_buf over 2G is not supported.\n");
1109 size = roundup_pow_of_two(size);
1110 if (size > log_buf_len)
1111 new_log_buf_len = (unsigned long)size;
1114 /* save requested log_buf_len since it's too early to process it */
1115 static int __init log_buf_len_setup(char *str)
1122 size = memparse(str, &str);
1124 log_buf_len_update(size);
1128 early_param("log_buf_len", log_buf_len_setup);
1131 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1133 static void __init log_buf_add_cpu(void)
1135 unsigned int cpu_extra;
1138 * archs should set up cpu_possible_bits properly with
1139 * set_cpu_possible() after setup_arch() but just in
1140 * case lets ensure this is valid.
1142 if (num_possible_cpus() == 1)
1145 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1147 /* by default this will only continue through for large > 64 CPUs */
1148 if (cpu_extra <= __LOG_BUF_LEN / 2)
1151 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1152 __LOG_CPU_MAX_BUF_LEN);
1153 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1155 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1157 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1159 #else /* !CONFIG_SMP */
1160 static inline void log_buf_add_cpu(void) {}
1161 #endif /* CONFIG_SMP */
1163 static void __init set_percpu_data_ready(void)
1166 /* Make sure we set this flag only after printk_safe() init is done */
1168 __printk_percpu_data_ready = true;
1171 void __init setup_log_buf(int early)
1173 unsigned long flags;
1178 * Some archs call setup_log_buf() multiple times - first is very
1179 * early, e.g. from setup_arch(), and second - when percpu_areas
1183 set_percpu_data_ready();
1185 if (log_buf != __log_buf)
1188 if (!early && !new_log_buf_len)
1191 if (!new_log_buf_len)
1194 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1195 if (unlikely(!new_log_buf)) {
1196 pr_err("log_buf_len: %lu bytes not available\n",
1201 logbuf_lock_irqsave(flags);
1202 log_buf_len = new_log_buf_len;
1203 log_buf = new_log_buf;
1204 new_log_buf_len = 0;
1205 free = __LOG_BUF_LEN - log_next_idx;
1206 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1207 logbuf_unlock_irqrestore(flags);
1209 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1210 pr_info("early log buf free: %u(%u%%)\n",
1211 free, (free * 100) / __LOG_BUF_LEN);
1214 static bool __read_mostly ignore_loglevel;
1216 static int __init ignore_loglevel_setup(char *str)
1218 ignore_loglevel = true;
1219 pr_info("debug: ignoring loglevel setting.\n");
1224 early_param("ignore_loglevel", ignore_loglevel_setup);
1225 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1226 MODULE_PARM_DESC(ignore_loglevel,
1227 "ignore loglevel setting (prints all kernel messages to the console)");
1229 static bool suppress_message_printing(int level)
1231 return (level >= console_loglevel && !ignore_loglevel);
1234 #ifdef CONFIG_BOOT_PRINTK_DELAY
1236 static int boot_delay; /* msecs delay after each printk during bootup */
1237 static unsigned long long loops_per_msec; /* based on boot_delay */
1239 static int __init boot_delay_setup(char *str)
1243 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1244 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1246 get_option(&str, &boot_delay);
1247 if (boot_delay > 10 * 1000)
1250 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1251 "HZ: %d, loops_per_msec: %llu\n",
1252 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1255 early_param("boot_delay", boot_delay_setup);
1257 static void boot_delay_msec(int level)
1259 unsigned long long k;
1260 unsigned long timeout;
1262 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1263 || suppress_message_printing(level)) {
1267 k = (unsigned long long)loops_per_msec * boot_delay;
1269 timeout = jiffies + msecs_to_jiffies(boot_delay);
1274 * use (volatile) jiffies to prevent
1275 * compiler reduction; loop termination via jiffies
1276 * is secondary and may or may not happen.
1278 if (time_after(jiffies, timeout))
1280 touch_nmi_watchdog();
1284 static inline void boot_delay_msec(int level)
1289 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1290 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1292 static size_t print_syslog(unsigned int level, char *buf)
1294 return sprintf(buf, "<%u>", level);
1297 static size_t print_time(u64 ts, char *buf)
1299 unsigned long rem_nsec = do_div(ts, 1000000000);
1301 return sprintf(buf, "[%5lu.%06lu]",
1302 (unsigned long)ts, rem_nsec / 1000);
1305 #ifdef CONFIG_PRINTK_CALLER
1306 static size_t print_caller(u32 id, char *buf)
1310 snprintf(caller, sizeof(caller), "%c%u",
1311 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1312 return sprintf(buf, "[%6s]", caller);
1315 #define print_caller(id, buf) 0
1318 static size_t print_prefix(const struct printk_log *msg, bool syslog,
1319 bool time, char *buf)
1324 len = print_syslog((msg->facility << 3) | msg->level, buf);
1327 len += print_time(msg->ts_nsec, buf + len);
1329 len += print_caller(msg->caller_id, buf + len);
1331 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1339 static size_t msg_print_text(const struct printk_log *msg, bool syslog,
1340 bool time, char *buf, size_t size)
1342 const char *text = log_text(msg);
1343 size_t text_size = msg->text_len;
1345 char prefix[PREFIX_MAX];
1346 const size_t prefix_len = print_prefix(msg, syslog, time, prefix);
1349 const char *next = memchr(text, '\n', text_size);
1353 text_len = next - text;
1355 text_size -= next - text;
1357 text_len = text_size;
1361 if (prefix_len + text_len + 1 >= size - len)
1364 memcpy(buf + len, prefix, prefix_len);
1366 memcpy(buf + len, text, text_len);
1370 /* SYSLOG_ACTION_* buffer size only calculation */
1371 len += prefix_len + text_len + 1;
1380 static int syslog_print(char __user *buf, int size)
1383 struct printk_log *msg;
1386 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1395 if (syslog_seq < log_first_seq) {
1396 /* messages are gone, move to first one */
1397 syslog_seq = log_first_seq;
1398 syslog_idx = log_first_idx;
1401 if (syslog_seq == log_next_seq) {
1402 logbuf_unlock_irq();
1407 * To keep reading/counting partial line consistent,
1408 * use printk_time value as of the beginning of a line.
1410 if (!syslog_partial)
1411 syslog_time = printk_time;
1413 skip = syslog_partial;
1414 msg = log_from_idx(syslog_idx);
1415 n = msg_print_text(msg, true, syslog_time, text,
1416 LOG_LINE_MAX + PREFIX_MAX);
1417 if (n - syslog_partial <= size) {
1418 /* message fits into buffer, move forward */
1419 syslog_idx = log_next(syslog_idx);
1421 n -= syslog_partial;
1424 /* partial read(), remember position */
1426 syslog_partial += n;
1429 logbuf_unlock_irq();
1434 if (copy_to_user(buf, text + skip, n)) {
1449 static int syslog_print_all(char __user *buf, int size, bool clear)
1458 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1465 * Find first record that fits, including all following records,
1466 * into the user-provided buffer for this dump.
1470 while (seq < log_next_seq) {
1471 struct printk_log *msg = log_from_idx(idx);
1473 len += msg_print_text(msg, true, time, NULL, 0);
1474 idx = log_next(idx);
1478 /* move first record forward until length fits into the buffer */
1481 while (len > size && seq < log_next_seq) {
1482 struct printk_log *msg = log_from_idx(idx);
1484 len -= msg_print_text(msg, true, time, NULL, 0);
1485 idx = log_next(idx);
1489 /* last message fitting into this dump */
1490 next_seq = log_next_seq;
1493 while (len >= 0 && seq < next_seq) {
1494 struct printk_log *msg = log_from_idx(idx);
1495 int textlen = msg_print_text(msg, true, time, text,
1496 LOG_LINE_MAX + PREFIX_MAX);
1498 idx = log_next(idx);
1501 logbuf_unlock_irq();
1502 if (copy_to_user(buf + len, text, textlen))
1508 if (seq < log_first_seq) {
1509 /* messages are gone, move to next one */
1510 seq = log_first_seq;
1511 idx = log_first_idx;
1516 clear_seq = log_next_seq;
1517 clear_idx = log_next_idx;
1519 logbuf_unlock_irq();
1525 static void syslog_clear(void)
1528 clear_seq = log_next_seq;
1529 clear_idx = log_next_idx;
1530 logbuf_unlock_irq();
1533 int do_syslog(int type, char __user *buf, int len, int source)
1536 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1539 error = check_syslog_permissions(type, source);
1544 case SYSLOG_ACTION_CLOSE: /* Close log */
1546 case SYSLOG_ACTION_OPEN: /* Open log */
1548 case SYSLOG_ACTION_READ: /* Read from log */
1549 if (!buf || len < 0)
1553 if (!access_ok(buf, len))
1555 error = wait_event_interruptible(log_wait,
1556 syslog_seq != log_next_seq);
1559 error = syslog_print(buf, len);
1561 /* Read/clear last kernel messages */
1562 case SYSLOG_ACTION_READ_CLEAR:
1565 /* Read last kernel messages */
1566 case SYSLOG_ACTION_READ_ALL:
1567 if (!buf || len < 0)
1571 if (!access_ok(buf, len))
1573 error = syslog_print_all(buf, len, clear);
1575 /* Clear ring buffer */
1576 case SYSLOG_ACTION_CLEAR:
1579 /* Disable logging to console */
1580 case SYSLOG_ACTION_CONSOLE_OFF:
1581 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1582 saved_console_loglevel = console_loglevel;
1583 console_loglevel = minimum_console_loglevel;
1585 /* Enable logging to console */
1586 case SYSLOG_ACTION_CONSOLE_ON:
1587 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1588 console_loglevel = saved_console_loglevel;
1589 saved_console_loglevel = LOGLEVEL_DEFAULT;
1592 /* Set level of messages printed to console */
1593 case SYSLOG_ACTION_CONSOLE_LEVEL:
1594 if (len < 1 || len > 8)
1596 if (len < minimum_console_loglevel)
1597 len = minimum_console_loglevel;
1598 console_loglevel = len;
1599 /* Implicitly re-enable logging to console */
1600 saved_console_loglevel = LOGLEVEL_DEFAULT;
1602 /* Number of chars in the log buffer */
1603 case SYSLOG_ACTION_SIZE_UNREAD:
1605 if (syslog_seq < log_first_seq) {
1606 /* messages are gone, move to first one */
1607 syslog_seq = log_first_seq;
1608 syslog_idx = log_first_idx;
1611 if (source == SYSLOG_FROM_PROC) {
1613 * Short-cut for poll(/"proc/kmsg") which simply checks
1614 * for pending data, not the size; return the count of
1615 * records, not the length.
1617 error = log_next_seq - syslog_seq;
1619 u64 seq = syslog_seq;
1620 u32 idx = syslog_idx;
1621 bool time = syslog_partial ? syslog_time : printk_time;
1623 while (seq < log_next_seq) {
1624 struct printk_log *msg = log_from_idx(idx);
1626 error += msg_print_text(msg, true, time, NULL,
1629 idx = log_next(idx);
1632 error -= syslog_partial;
1634 logbuf_unlock_irq();
1636 /* Size of the log buffer */
1637 case SYSLOG_ACTION_SIZE_BUFFER:
1638 error = log_buf_len;
1648 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1650 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1654 * Special console_lock variants that help to reduce the risk of soft-lockups.
1655 * They allow to pass console_lock to another printk() call using a busy wait.
1658 #ifdef CONFIG_LOCKDEP
1659 static struct lockdep_map console_owner_dep_map = {
1660 .name = "console_owner"
1664 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1665 static struct task_struct *console_owner;
1666 static bool console_waiter;
1669 * console_lock_spinning_enable - mark beginning of code where another
1670 * thread might safely busy wait
1672 * This basically converts console_lock into a spinlock. This marks
1673 * the section where the console_lock owner can not sleep, because
1674 * there may be a waiter spinning (like a spinlock). Also it must be
1675 * ready to hand over the lock at the end of the section.
1677 static void console_lock_spinning_enable(void)
1679 raw_spin_lock(&console_owner_lock);
1680 console_owner = current;
1681 raw_spin_unlock(&console_owner_lock);
1683 /* The waiter may spin on us after setting console_owner */
1684 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1688 * console_lock_spinning_disable_and_check - mark end of code where another
1689 * thread was able to busy wait and check if there is a waiter
1691 * This is called at the end of the section where spinning is allowed.
1692 * It has two functions. First, it is a signal that it is no longer
1693 * safe to start busy waiting for the lock. Second, it checks if
1694 * there is a busy waiter and passes the lock rights to her.
1696 * Important: Callers lose the lock if there was a busy waiter.
1697 * They must not touch items synchronized by console_lock
1700 * Return: 1 if the lock rights were passed, 0 otherwise.
1702 static int console_lock_spinning_disable_and_check(void)
1706 raw_spin_lock(&console_owner_lock);
1707 waiter = READ_ONCE(console_waiter);
1708 console_owner = NULL;
1709 raw_spin_unlock(&console_owner_lock);
1712 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1716 /* The waiter is now free to continue */
1717 WRITE_ONCE(console_waiter, false);
1719 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1722 * Hand off console_lock to waiter. The waiter will perform
1723 * the up(). After this, the waiter is the console_lock owner.
1725 mutex_release(&console_lock_dep_map, 1, _THIS_IP_);
1730 * console_trylock_spinning - try to get console_lock by busy waiting
1732 * This allows to busy wait for the console_lock when the current
1733 * owner is running in specially marked sections. It means that
1734 * the current owner is running and cannot reschedule until it
1735 * is ready to lose the lock.
1737 * Return: 1 if we got the lock, 0 othrewise
1739 static int console_trylock_spinning(void)
1741 struct task_struct *owner = NULL;
1744 unsigned long flags;
1746 if (console_trylock())
1749 printk_safe_enter_irqsave(flags);
1751 raw_spin_lock(&console_owner_lock);
1752 owner = READ_ONCE(console_owner);
1753 waiter = READ_ONCE(console_waiter);
1754 if (!waiter && owner && owner != current) {
1755 WRITE_ONCE(console_waiter, true);
1758 raw_spin_unlock(&console_owner_lock);
1761 * If there is an active printk() writing to the
1762 * consoles, instead of having it write our data too,
1763 * see if we can offload that load from the active
1764 * printer, and do some printing ourselves.
1765 * Go into a spin only if there isn't already a waiter
1766 * spinning, and there is an active printer, and
1767 * that active printer isn't us (recursive printk?).
1770 printk_safe_exit_irqrestore(flags);
1774 /* We spin waiting for the owner to release us */
1775 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1776 /* Owner will clear console_waiter on hand off */
1777 while (READ_ONCE(console_waiter))
1779 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1781 printk_safe_exit_irqrestore(flags);
1783 * The owner passed the console lock to us.
1784 * Since we did not spin on console lock, annotate
1785 * this as a trylock. Otherwise lockdep will
1788 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1794 * Call the console drivers, asking them to write out
1795 * log_buf[start] to log_buf[end - 1].
1796 * The console_lock must be held.
1798 static void call_console_drivers(const char *ext_text, size_t ext_len,
1799 const char *text, size_t len)
1801 struct console *con;
1803 trace_console_rcuidle(text, len);
1805 if (!console_drivers)
1808 for_each_console(con) {
1809 if (exclusive_console && con != exclusive_console)
1811 if (!(con->flags & CON_ENABLED))
1815 if (!cpu_online(smp_processor_id()) &&
1816 !(con->flags & CON_ANYTIME))
1818 if (con->flags & CON_EXTENDED)
1819 con->write(con, ext_text, ext_len);
1821 con->write(con, text, len);
1825 int printk_delay_msec __read_mostly;
1827 static inline void printk_delay(void)
1829 if (unlikely(printk_delay_msec)) {
1830 int m = printk_delay_msec;
1834 touch_nmi_watchdog();
1839 static inline u32 printk_caller_id(void)
1841 return in_task() ? task_pid_nr(current) :
1842 0x80000000 + raw_smp_processor_id();
1846 * Continuation lines are buffered, and not committed to the record buffer
1847 * until the line is complete, or a race forces it. The line fragments
1848 * though, are printed immediately to the consoles to ensure everything has
1849 * reached the console in case of a kernel crash.
1851 static struct cont {
1852 char buf[LOG_LINE_MAX];
1853 size_t len; /* length == 0 means unused buffer */
1854 u32 caller_id; /* printk_caller_id() of first print */
1855 u64 ts_nsec; /* time of first print */
1856 u8 level; /* log level of first message */
1857 u8 facility; /* log facility of first message */
1858 enum log_flags flags; /* prefix, newline flags */
1861 static void cont_flush(void)
1866 log_store(cont.caller_id, cont.facility, cont.level, cont.flags,
1867 cont.ts_nsec, NULL, 0, cont.buf, cont.len);
1871 static bool cont_add(u32 caller_id, int facility, int level,
1872 enum log_flags flags, const char *text, size_t len)
1874 /* If the line gets too long, split it up in separate records. */
1875 if (cont.len + len > sizeof(cont.buf)) {
1881 cont.facility = facility;
1883 cont.caller_id = caller_id;
1884 cont.ts_nsec = local_clock();
1888 memcpy(cont.buf + cont.len, text, len);
1891 // The original flags come from the first line,
1892 // but later continuations can add a newline.
1893 if (flags & LOG_NEWLINE) {
1894 cont.flags |= LOG_NEWLINE;
1901 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)
1903 const u32 caller_id = printk_caller_id();
1906 * If an earlier line was buffered, and we're a continuation
1907 * write from the same context, try to add it to the buffer.
1910 if (cont.caller_id == caller_id && (lflags & LOG_CONT)) {
1911 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1914 /* Otherwise, make sure it's flushed */
1918 /* Skip empty continuation lines that couldn't be added - they just flush */
1919 if (!text_len && (lflags & LOG_CONT))
1922 /* If it doesn't end in a newline, try to buffer the current line */
1923 if (!(lflags & LOG_NEWLINE)) {
1924 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1928 /* Store it in the record log */
1929 return log_store(caller_id, facility, level, lflags, 0,
1930 dict, dictlen, text, text_len);
1933 /* Must be called under logbuf_lock. */
1934 int vprintk_store(int facility, int level,
1935 const char *dict, size_t dictlen,
1936 const char *fmt, va_list args)
1938 static char textbuf[LOG_LINE_MAX];
1939 char *text = textbuf;
1941 enum log_flags lflags = 0;
1944 * The printf needs to come first; we need the syslog
1945 * prefix which might be passed-in as a parameter.
1947 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1949 /* mark and strip a trailing newline */
1950 if (text_len && text[text_len-1] == '\n') {
1952 lflags |= LOG_NEWLINE;
1955 /* strip kernel syslog prefix and extract log level or control flags */
1956 if (facility == 0) {
1959 while ((kern_level = printk_get_level(text)) != 0) {
1960 switch (kern_level) {
1962 if (level == LOGLEVEL_DEFAULT)
1963 level = kern_level - '0';
1965 case 'c': /* KERN_CONT */
1974 if (level == LOGLEVEL_DEFAULT)
1975 level = default_message_loglevel;
1978 lflags |= LOG_NEWLINE;
1980 return log_output(facility, level, lflags,
1981 dict, dictlen, text, text_len);
1984 asmlinkage int vprintk_emit(int facility, int level,
1985 const char *dict, size_t dictlen,
1986 const char *fmt, va_list args)
1989 bool in_sched = false, pending_output;
1990 unsigned long flags;
1993 /* Suppress unimportant messages after panic happens */
1994 if (unlikely(suppress_printk))
1997 if (level == LOGLEVEL_SCHED) {
1998 level = LOGLEVEL_DEFAULT;
2002 boot_delay_msec(level);
2005 /* This stops the holder of console_sem just where we want him */
2006 logbuf_lock_irqsave(flags);
2007 curr_log_seq = log_next_seq;
2008 printed_len = vprintk_store(facility, level, dict, dictlen, fmt, args);
2009 pending_output = (curr_log_seq != log_next_seq);
2010 logbuf_unlock_irqrestore(flags);
2012 /* If called from the scheduler, we can not call up(). */
2013 if (!in_sched && pending_output) {
2015 * Disable preemption to avoid being preempted while holding
2016 * console_sem which would prevent anyone from printing to
2021 * Try to acquire and then immediately release the console
2022 * semaphore. The release will print out buffers and wake up
2023 * /dev/kmsg and syslog() users.
2025 if (console_trylock_spinning())
2034 EXPORT_SYMBOL(vprintk_emit);
2036 asmlinkage int vprintk(const char *fmt, va_list args)
2038 return vprintk_func(fmt, args);
2040 EXPORT_SYMBOL(vprintk);
2042 int vprintk_default(const char *fmt, va_list args)
2046 #ifdef CONFIG_KGDB_KDB
2047 /* Allow to pass printk() to kdb but avoid a recursion. */
2048 if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
2049 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
2053 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
2057 EXPORT_SYMBOL_GPL(vprintk_default);
2060 * printk - print a kernel message
2061 * @fmt: format string
2063 * This is printk(). It can be called from any context. We want it to work.
2065 * We try to grab the console_lock. If we succeed, it's easy - we log the
2066 * output and call the console drivers. If we fail to get the semaphore, we
2067 * place the output into the log buffer and return. The current holder of
2068 * the console_sem will notice the new output in console_unlock(); and will
2069 * send it to the consoles before releasing the lock.
2071 * One effect of this deferred printing is that code which calls printk() and
2072 * then changes console_loglevel may break. This is because console_loglevel
2073 * is inspected when the actual printing occurs.
2078 * See the vsnprintf() documentation for format string extensions over C99.
2080 asmlinkage __visible int printk(const char *fmt, ...)
2085 va_start(args, fmt);
2086 r = vprintk_func(fmt, args);
2091 EXPORT_SYMBOL(printk);
2093 #else /* CONFIG_PRINTK */
2095 #define LOG_LINE_MAX 0
2096 #define PREFIX_MAX 0
2097 #define printk_time false
2099 static u64 syslog_seq;
2100 static u32 syslog_idx;
2101 static u64 console_seq;
2102 static u32 console_idx;
2103 static u64 exclusive_console_stop_seq;
2104 static u64 log_first_seq;
2105 static u32 log_first_idx;
2106 static u64 log_next_seq;
2107 static char *log_text(const struct printk_log *msg) { return NULL; }
2108 static char *log_dict(const struct printk_log *msg) { return NULL; }
2109 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2110 static u32 log_next(u32 idx) { return 0; }
2111 static ssize_t msg_print_ext_header(char *buf, size_t size,
2112 struct printk_log *msg,
2113 u64 seq) { return 0; }
2114 static ssize_t msg_print_ext_body(char *buf, size_t size,
2115 char *dict, size_t dict_len,
2116 char *text, size_t text_len) { return 0; }
2117 static void console_lock_spinning_enable(void) { }
2118 static int console_lock_spinning_disable_and_check(void) { return 0; }
2119 static void call_console_drivers(const char *ext_text, size_t ext_len,
2120 const char *text, size_t len) {}
2121 static size_t msg_print_text(const struct printk_log *msg, bool syslog,
2122 bool time, char *buf, size_t size) { return 0; }
2123 static bool suppress_message_printing(int level) { return false; }
2125 #endif /* CONFIG_PRINTK */
2127 #ifdef CONFIG_EARLY_PRINTK
2128 struct console *early_console;
2130 asmlinkage __visible void early_printk(const char *fmt, ...)
2140 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2143 early_console->write(early_console, buf, n);
2147 static int __add_preferred_console(char *name, int idx, char *options,
2150 struct console_cmdline *c;
2154 * See if this tty is not yet registered, and
2155 * if we have a slot free.
2157 for (i = 0, c = console_cmdline;
2158 i < MAX_CMDLINECONSOLES && c->name[0];
2160 if (strcmp(c->name, name) == 0 && c->index == idx) {
2162 preferred_console = i;
2166 if (i == MAX_CMDLINECONSOLES)
2169 preferred_console = i;
2170 strlcpy(c->name, name, sizeof(c->name));
2171 c->options = options;
2172 braille_set_options(c, brl_options);
2178 static int __init console_msg_format_setup(char *str)
2180 if (!strcmp(str, "syslog"))
2181 console_msg_format = MSG_FORMAT_SYSLOG;
2182 if (!strcmp(str, "default"))
2183 console_msg_format = MSG_FORMAT_DEFAULT;
2186 __setup("console_msg_format=", console_msg_format_setup);
2189 * Set up a console. Called via do_early_param() in init/main.c
2190 * for each "console=" parameter in the boot command line.
2192 static int __init console_setup(char *str)
2194 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2195 char *s, *options, *brl_options = NULL;
2199 * console="" or console=null have been suggested as a way to
2200 * disable console output. Use ttynull that has been created
2201 * for exacly this purpose.
2203 if (str[0] == 0 || strcmp(str, "null") == 0) {
2204 __add_preferred_console("ttynull", 0, NULL, NULL);
2208 if (_braille_console_setup(&str, &brl_options))
2212 * Decode str into name, index, options.
2214 if (str[0] >= '0' && str[0] <= '9') {
2215 strcpy(buf, "ttyS");
2216 strncpy(buf + 4, str, sizeof(buf) - 5);
2218 strncpy(buf, str, sizeof(buf) - 1);
2220 buf[sizeof(buf) - 1] = 0;
2221 options = strchr(str, ',');
2225 if (!strcmp(str, "ttya"))
2226 strcpy(buf, "ttyS0");
2227 if (!strcmp(str, "ttyb"))
2228 strcpy(buf, "ttyS1");
2230 for (s = buf; *s; s++)
2231 if (isdigit(*s) || *s == ',')
2233 idx = simple_strtoul(s, NULL, 10);
2236 __add_preferred_console(buf, idx, options, brl_options);
2237 console_set_on_cmdline = 1;
2240 __setup("console=", console_setup);
2243 * add_preferred_console - add a device to the list of preferred consoles.
2244 * @name: device name
2245 * @idx: device index
2246 * @options: options for this console
2248 * The last preferred console added will be used for kernel messages
2249 * and stdin/out/err for init. Normally this is used by console_setup
2250 * above to handle user-supplied console arguments; however it can also
2251 * be used by arch-specific code either to override the user or more
2252 * commonly to provide a default console (ie from PROM variables) when
2253 * the user has not supplied one.
2255 int add_preferred_console(char *name, int idx, char *options)
2257 return __add_preferred_console(name, idx, options, NULL);
2260 bool console_suspend_enabled = true;
2261 EXPORT_SYMBOL(console_suspend_enabled);
2263 static int __init console_suspend_disable(char *str)
2265 console_suspend_enabled = false;
2268 __setup("no_console_suspend", console_suspend_disable);
2269 module_param_named(console_suspend, console_suspend_enabled,
2270 bool, S_IRUGO | S_IWUSR);
2271 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2272 " and hibernate operations");
2275 * suspend_console - suspend the console subsystem
2277 * This disables printk() while we go into suspend states
2279 void suspend_console(void)
2281 if (!console_suspend_enabled)
2283 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2285 console_suspended = 1;
2289 void resume_console(void)
2291 if (!console_suspend_enabled)
2294 console_suspended = 0;
2299 * console_cpu_notify - print deferred console messages after CPU hotplug
2302 * If printk() is called from a CPU that is not online yet, the messages
2303 * will be printed on the console only if there are CON_ANYTIME consoles.
2304 * This function is called when a new CPU comes online (or fails to come
2305 * up) or goes offline.
2307 static int console_cpu_notify(unsigned int cpu)
2309 if (!cpuhp_tasks_frozen) {
2310 /* If trylock fails, someone else is doing the printing */
2311 if (console_trylock())
2318 * console_lock - lock the console system for exclusive use.
2320 * Acquires a lock which guarantees that the caller has
2321 * exclusive access to the console system and the console_drivers list.
2323 * Can sleep, returns nothing.
2325 void console_lock(void)
2330 if (console_suspended)
2333 console_may_schedule = 1;
2335 EXPORT_SYMBOL(console_lock);
2338 * console_trylock - try to lock the console system for exclusive use.
2340 * Try to acquire a lock which guarantees that the caller has exclusive
2341 * access to the console system and the console_drivers list.
2343 * returns 1 on success, and 0 on failure to acquire the lock.
2345 int console_trylock(void)
2347 if (down_trylock_console_sem())
2349 if (console_suspended) {
2354 console_may_schedule = 0;
2357 EXPORT_SYMBOL(console_trylock);
2359 int is_console_locked(void)
2361 return console_locked;
2363 EXPORT_SYMBOL(is_console_locked);
2366 * Check if we have any console that is capable of printing while cpu is
2367 * booting or shutting down. Requires console_sem.
2369 static int have_callable_console(void)
2371 struct console *con;
2373 for_each_console(con)
2374 if ((con->flags & CON_ENABLED) &&
2375 (con->flags & CON_ANYTIME))
2382 * Can we actually use the console at this time on this cpu?
2384 * Console drivers may assume that per-cpu resources have been allocated. So
2385 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2386 * call them until this CPU is officially up.
2388 static inline int can_use_console(void)
2390 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2394 * console_unlock - unlock the console system
2396 * Releases the console_lock which the caller holds on the console system
2397 * and the console driver list.
2399 * While the console_lock was held, console output may have been buffered
2400 * by printk(). If this is the case, console_unlock(); emits
2401 * the output prior to releasing the lock.
2403 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2405 * console_unlock(); may be called from any context.
2407 void console_unlock(void)
2409 static char ext_text[CONSOLE_EXT_LOG_MAX];
2410 static char text[LOG_LINE_MAX + PREFIX_MAX];
2411 unsigned long flags;
2412 bool do_cond_resched, retry;
2414 if (console_suspended) {
2420 * Console drivers are called with interrupts disabled, so
2421 * @console_may_schedule should be cleared before; however, we may
2422 * end up dumping a lot of lines, for example, if called from
2423 * console registration path, and should invoke cond_resched()
2424 * between lines if allowable. Not doing so can cause a very long
2425 * scheduling stall on a slow console leading to RCU stall and
2426 * softlockup warnings which exacerbate the issue with more
2427 * messages practically incapacitating the system.
2429 * console_trylock() is not able to detect the preemptive
2430 * context reliably. Therefore the value must be stored before
2431 * and cleared after the the "again" goto label.
2433 do_cond_resched = console_may_schedule;
2435 console_may_schedule = 0;
2438 * We released the console_sem lock, so we need to recheck if
2439 * cpu is online and (if not) is there at least one CON_ANYTIME
2442 if (!can_use_console()) {
2449 struct printk_log *msg;
2453 printk_safe_enter_irqsave(flags);
2454 raw_spin_lock(&logbuf_lock);
2455 if (console_seq < log_first_seq) {
2457 "** %llu printk messages dropped **\n",
2458 log_first_seq - console_seq);
2460 /* messages are gone, move to first one */
2461 console_seq = log_first_seq;
2462 console_idx = log_first_idx;
2467 if (console_seq == log_next_seq)
2470 msg = log_from_idx(console_idx);
2471 if (suppress_message_printing(msg->level)) {
2473 * Skip record we have buffered and already printed
2474 * directly to the console when we received it, and
2475 * record that has level above the console loglevel.
2477 console_idx = log_next(console_idx);
2482 /* Output to all consoles once old messages replayed. */
2483 if (unlikely(exclusive_console &&
2484 console_seq >= exclusive_console_stop_seq)) {
2485 exclusive_console = NULL;
2488 len += msg_print_text(msg,
2489 console_msg_format & MSG_FORMAT_SYSLOG,
2490 printk_time, text + len, sizeof(text) - len);
2491 if (nr_ext_console_drivers) {
2492 ext_len = msg_print_ext_header(ext_text,
2495 ext_len += msg_print_ext_body(ext_text + ext_len,
2496 sizeof(ext_text) - ext_len,
2497 log_dict(msg), msg->dict_len,
2498 log_text(msg), msg->text_len);
2500 console_idx = log_next(console_idx);
2502 raw_spin_unlock(&logbuf_lock);
2505 * While actively printing out messages, if another printk()
2506 * were to occur on another CPU, it may wait for this one to
2507 * finish. This task can not be preempted if there is a
2508 * waiter waiting to take over.
2510 console_lock_spinning_enable();
2512 stop_critical_timings(); /* don't trace print latency */
2513 call_console_drivers(ext_text, ext_len, text, len);
2514 start_critical_timings();
2516 if (console_lock_spinning_disable_and_check()) {
2517 printk_safe_exit_irqrestore(flags);
2521 printk_safe_exit_irqrestore(flags);
2523 if (do_cond_resched)
2529 raw_spin_unlock(&logbuf_lock);
2534 * Someone could have filled up the buffer again, so re-check if there's
2535 * something to flush. In case we cannot trylock the console_sem again,
2536 * there's a new owner and the console_unlock() from them will do the
2537 * flush, no worries.
2539 raw_spin_lock(&logbuf_lock);
2540 retry = console_seq != log_next_seq;
2541 raw_spin_unlock(&logbuf_lock);
2542 printk_safe_exit_irqrestore(flags);
2544 if (retry && console_trylock())
2547 EXPORT_SYMBOL(console_unlock);
2550 * console_conditional_schedule - yield the CPU if required
2552 * If the console code is currently allowed to sleep, and
2553 * if this CPU should yield the CPU to another task, do
2556 * Must be called within console_lock();.
2558 void __sched console_conditional_schedule(void)
2560 if (console_may_schedule)
2563 EXPORT_SYMBOL(console_conditional_schedule);
2565 void console_unblank(void)
2570 * console_unblank can no longer be called in interrupt context unless
2571 * oops_in_progress is set to 1..
2573 if (oops_in_progress) {
2574 if (down_trylock_console_sem() != 0)
2580 console_may_schedule = 0;
2582 if ((c->flags & CON_ENABLED) && c->unblank)
2588 * console_flush_on_panic - flush console content on panic
2589 * @mode: flush all messages in buffer or just the pending ones
2591 * Immediately output all pending messages no matter what.
2593 void console_flush_on_panic(enum con_flush_mode mode)
2596 * If someone else is holding the console lock, trylock will fail
2597 * and may_schedule may be set. Ignore and proceed to unlock so
2598 * that messages are flushed out. As this can be called from any
2599 * context and we don't want to get preempted while flushing,
2600 * ensure may_schedule is cleared.
2603 console_may_schedule = 0;
2605 if (mode == CONSOLE_REPLAY_ALL) {
2606 unsigned long flags;
2608 logbuf_lock_irqsave(flags);
2609 console_seq = log_first_seq;
2610 console_idx = log_first_idx;
2611 logbuf_unlock_irqrestore(flags);
2617 * Return the console tty driver structure and its associated index
2619 struct tty_driver *console_device(int *index)
2622 struct tty_driver *driver = NULL;
2625 for_each_console(c) {
2628 driver = c->device(c, index);
2637 * Prevent further output on the passed console device so that (for example)
2638 * serial drivers can disable console output before suspending a port, and can
2639 * re-enable output afterwards.
2641 void console_stop(struct console *console)
2644 console->flags &= ~CON_ENABLED;
2647 EXPORT_SYMBOL(console_stop);
2649 void console_start(struct console *console)
2652 console->flags |= CON_ENABLED;
2655 EXPORT_SYMBOL(console_start);
2657 static int __read_mostly keep_bootcon;
2659 static int __init keep_bootcon_setup(char *str)
2662 pr_info("debug: skip boot console de-registration.\n");
2667 early_param("keep_bootcon", keep_bootcon_setup);
2670 * The console driver calls this routine during kernel initialization
2671 * to register the console printing procedure with printk() and to
2672 * print any messages that were printed by the kernel before the
2673 * console driver was initialized.
2675 * This can happen pretty early during the boot process (because of
2676 * early_printk) - sometimes before setup_arch() completes - be careful
2677 * of what kernel features are used - they may not be initialised yet.
2679 * There are two types of consoles - bootconsoles (early_printk) and
2680 * "real" consoles (everything which is not a bootconsole) which are
2681 * handled differently.
2682 * - Any number of bootconsoles can be registered at any time.
2683 * - As soon as a "real" console is registered, all bootconsoles
2684 * will be unregistered automatically.
2685 * - Once a "real" console is registered, any attempt to register a
2686 * bootconsoles will be rejected
2688 void register_console(struct console *newcon)
2691 unsigned long flags;
2692 struct console *bcon = NULL;
2693 struct console_cmdline *c;
2694 static bool has_preferred;
2696 if (console_drivers)
2697 for_each_console(bcon)
2698 if (WARN(bcon == newcon,
2699 "console '%s%d' already registered\n",
2700 bcon->name, bcon->index))
2704 * before we register a new CON_BOOT console, make sure we don't
2705 * already have a valid console
2707 if (console_drivers && newcon->flags & CON_BOOT) {
2708 /* find the last or real console */
2709 for_each_console(bcon) {
2710 if (!(bcon->flags & CON_BOOT)) {
2711 pr_info("Too late to register bootconsole %s%d\n",
2712 newcon->name, newcon->index);
2718 if (console_drivers && console_drivers->flags & CON_BOOT)
2719 bcon = console_drivers;
2721 if (!has_preferred || bcon || !console_drivers)
2722 has_preferred = preferred_console >= 0;
2725 * See if we want to use this console driver. If we
2726 * didn't select a console we take the first one
2727 * that registers here.
2729 if (!has_preferred) {
2730 if (newcon->index < 0)
2732 if (newcon->setup == NULL ||
2733 newcon->setup(newcon, NULL) == 0) {
2734 newcon->flags |= CON_ENABLED;
2735 if (newcon->device) {
2736 newcon->flags |= CON_CONSDEV;
2737 has_preferred = true;
2743 * See if this console matches one we selected on
2746 for (i = 0, c = console_cmdline;
2747 i < MAX_CMDLINECONSOLES && c->name[0];
2749 if (!newcon->match ||
2750 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2751 /* default matching */
2752 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2753 if (strcmp(c->name, newcon->name) != 0)
2755 if (newcon->index >= 0 &&
2756 newcon->index != c->index)
2758 if (newcon->index < 0)
2759 newcon->index = c->index;
2761 if (_braille_register_console(newcon, c))
2764 if (newcon->setup &&
2765 newcon->setup(newcon, c->options) != 0)
2769 newcon->flags |= CON_ENABLED;
2770 if (i == preferred_console) {
2771 newcon->flags |= CON_CONSDEV;
2772 has_preferred = true;
2777 if (!(newcon->flags & CON_ENABLED))
2781 * If we have a bootconsole, and are switching to a real console,
2782 * don't print everything out again, since when the boot console, and
2783 * the real console are the same physical device, it's annoying to
2784 * see the beginning boot messages twice
2786 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2787 newcon->flags &= ~CON_PRINTBUFFER;
2790 * Put this console in the list - keep the
2791 * preferred driver at the head of the list.
2794 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2795 newcon->next = console_drivers;
2796 console_drivers = newcon;
2798 newcon->next->flags &= ~CON_CONSDEV;
2800 newcon->next = console_drivers->next;
2801 console_drivers->next = newcon;
2804 if (newcon->flags & CON_EXTENDED)
2805 nr_ext_console_drivers++;
2807 if (newcon->flags & CON_PRINTBUFFER) {
2809 * console_unlock(); will print out the buffered messages
2812 logbuf_lock_irqsave(flags);
2814 * We're about to replay the log buffer. Only do this to the
2815 * just-registered console to avoid excessive message spam to
2816 * the already-registered consoles.
2818 * Set exclusive_console with disabled interrupts to reduce
2819 * race window with eventual console_flush_on_panic() that
2820 * ignores console_lock.
2822 exclusive_console = newcon;
2823 exclusive_console_stop_seq = console_seq;
2824 console_seq = syslog_seq;
2825 console_idx = syslog_idx;
2826 logbuf_unlock_irqrestore(flags);
2829 console_sysfs_notify();
2832 * By unregistering the bootconsoles after we enable the real console
2833 * we get the "console xxx enabled" message on all the consoles -
2834 * boot consoles, real consoles, etc - this is to ensure that end
2835 * users know there might be something in the kernel's log buffer that
2836 * went to the bootconsole (that they do not see on the real console)
2838 pr_info("%sconsole [%s%d] enabled\n",
2839 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2840 newcon->name, newcon->index);
2842 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2844 /* We need to iterate through all boot consoles, to make
2845 * sure we print everything out, before we unregister them.
2847 for_each_console(bcon)
2848 if (bcon->flags & CON_BOOT)
2849 unregister_console(bcon);
2852 EXPORT_SYMBOL(register_console);
2854 int unregister_console(struct console *console)
2856 struct console *a, *b;
2859 pr_info("%sconsole [%s%d] disabled\n",
2860 (console->flags & CON_BOOT) ? "boot" : "" ,
2861 console->name, console->index);
2863 res = _braille_unregister_console(console);
2869 if (console_drivers == console) {
2870 console_drivers=console->next;
2872 } else if (console_drivers) {
2873 for (a=console_drivers->next, b=console_drivers ;
2874 a; b=a, a=b->next) {
2883 if (!res && (console->flags & CON_EXTENDED))
2884 nr_ext_console_drivers--;
2887 * If this isn't the last console and it has CON_CONSDEV set, we
2888 * need to set it on the next preferred console.
2890 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2891 console_drivers->flags |= CON_CONSDEV;
2893 console->flags &= ~CON_ENABLED;
2895 console_sysfs_notify();
2898 EXPORT_SYMBOL(unregister_console);
2901 * Initialize the console device. This is called *early*, so
2902 * we can't necessarily depend on lots of kernel help here.
2903 * Just do some early initializations, and do the complex setup
2906 void __init console_init(void)
2910 initcall_entry_t *ce;
2912 /* Setup the default TTY line discipline. */
2916 * set up the console device so that later boot sequences can
2917 * inform about problems etc..
2919 ce = __con_initcall_start;
2920 trace_initcall_level("console");
2921 while (ce < __con_initcall_end) {
2922 call = initcall_from_entry(ce);
2923 trace_initcall_start(call);
2925 trace_initcall_finish(call, ret);
2931 * Some boot consoles access data that is in the init section and which will
2932 * be discarded after the initcalls have been run. To make sure that no code
2933 * will access this data, unregister the boot consoles in a late initcall.
2935 * If for some reason, such as deferred probe or the driver being a loadable
2936 * module, the real console hasn't registered yet at this point, there will
2937 * be a brief interval in which no messages are logged to the console, which
2938 * makes it difficult to diagnose problems that occur during this time.
2940 * To mitigate this problem somewhat, only unregister consoles whose memory
2941 * intersects with the init section. Note that all other boot consoles will
2942 * get unregistred when the real preferred console is registered.
2944 static int __init printk_late_init(void)
2946 struct console *con;
2949 for_each_console(con) {
2950 if (!(con->flags & CON_BOOT))
2953 /* Check addresses that might be used for enabled consoles. */
2954 if (init_section_intersects(con, sizeof(*con)) ||
2955 init_section_contains(con->write, 0) ||
2956 init_section_contains(con->read, 0) ||
2957 init_section_contains(con->device, 0) ||
2958 init_section_contains(con->unblank, 0) ||
2959 init_section_contains(con->data, 0)) {
2961 * Please, consider moving the reported consoles out
2962 * of the init section.
2964 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2965 con->name, con->index);
2966 unregister_console(con);
2969 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2970 console_cpu_notify);
2972 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2973 console_cpu_notify, NULL);
2977 late_initcall(printk_late_init);
2979 #if defined CONFIG_PRINTK
2981 * Delayed printk version, for scheduler-internal messages:
2983 #define PRINTK_PENDING_WAKEUP 0x01
2984 #define PRINTK_PENDING_OUTPUT 0x02
2986 static DEFINE_PER_CPU(int, printk_pending);
2988 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2990 int pending = __this_cpu_xchg(printk_pending, 0);
2992 if (pending & PRINTK_PENDING_OUTPUT) {
2993 /* If trylock fails, someone else is doing the printing */
2994 if (console_trylock())
2998 if (pending & PRINTK_PENDING_WAKEUP)
2999 wake_up_interruptible(&log_wait);
3002 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
3003 .func = wake_up_klogd_work_func,
3004 .flags = IRQ_WORK_LAZY,
3007 void wake_up_klogd(void)
3009 if (!printk_percpu_data_ready())
3013 if (waitqueue_active(&log_wait)) {
3014 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
3015 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3020 void defer_console_output(void)
3022 if (!printk_percpu_data_ready())
3026 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
3027 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3031 int vprintk_deferred(const char *fmt, va_list args)
3035 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
3036 defer_console_output();
3041 int printk_deferred(const char *fmt, ...)
3046 va_start(args, fmt);
3047 r = vprintk_deferred(fmt, args);
3054 * printk rate limiting, lifted from the networking subsystem.
3056 * This enforces a rate limit: not more than 10 kernel messages
3057 * every 5s to make a denial-of-service attack impossible.
3059 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3061 int __printk_ratelimit(const char *func)
3063 return ___ratelimit(&printk_ratelimit_state, func);
3065 EXPORT_SYMBOL(__printk_ratelimit);
3068 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3069 * @caller_jiffies: pointer to caller's state
3070 * @interval_msecs: minimum interval between prints
3072 * printk_timed_ratelimit() returns true if more than @interval_msecs
3073 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3076 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3077 unsigned int interval_msecs)
3079 unsigned long elapsed = jiffies - *caller_jiffies;
3081 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3084 *caller_jiffies = jiffies;
3087 EXPORT_SYMBOL(printk_timed_ratelimit);
3089 static DEFINE_SPINLOCK(dump_list_lock);
3090 static LIST_HEAD(dump_list);
3093 * kmsg_dump_register - register a kernel log dumper.
3094 * @dumper: pointer to the kmsg_dumper structure
3096 * Adds a kernel log dumper to the system. The dump callback in the
3097 * structure will be called when the kernel oopses or panics and must be
3098 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3100 int kmsg_dump_register(struct kmsg_dumper *dumper)
3102 unsigned long flags;
3105 /* The dump callback needs to be set */
3109 spin_lock_irqsave(&dump_list_lock, flags);
3110 /* Don't allow registering multiple times */
3111 if (!dumper->registered) {
3112 dumper->registered = 1;
3113 list_add_tail_rcu(&dumper->list, &dump_list);
3116 spin_unlock_irqrestore(&dump_list_lock, flags);
3120 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3123 * kmsg_dump_unregister - unregister a kmsg dumper.
3124 * @dumper: pointer to the kmsg_dumper structure
3126 * Removes a dump device from the system. Returns zero on success and
3127 * %-EINVAL otherwise.
3129 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3131 unsigned long flags;
3134 spin_lock_irqsave(&dump_list_lock, flags);
3135 if (dumper->registered) {
3136 dumper->registered = 0;
3137 list_del_rcu(&dumper->list);
3140 spin_unlock_irqrestore(&dump_list_lock, flags);
3145 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3147 static bool always_kmsg_dump;
3148 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3151 * kmsg_dump - dump kernel log to kernel message dumpers.
3152 * @reason: the reason (oops, panic etc) for dumping
3154 * Call each of the registered dumper's dump() callback, which can
3155 * retrieve the kmsg records with kmsg_dump_get_line() or
3156 * kmsg_dump_get_buffer().
3158 void kmsg_dump(enum kmsg_dump_reason reason)
3160 struct kmsg_dumper *dumper;
3161 unsigned long flags;
3163 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3167 list_for_each_entry_rcu(dumper, &dump_list, list) {
3168 if (dumper->max_reason && reason > dumper->max_reason)
3171 /* initialize iterator with data about the stored records */
3172 dumper->active = true;
3174 logbuf_lock_irqsave(flags);
3175 dumper->cur_seq = clear_seq;
3176 dumper->cur_idx = clear_idx;
3177 dumper->next_seq = log_next_seq;
3178 dumper->next_idx = log_next_idx;
3179 logbuf_unlock_irqrestore(flags);
3181 /* invoke dumper which will iterate over records */
3182 dumper->dump(dumper, reason);
3184 /* reset iterator */
3185 dumper->active = false;
3191 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3192 * @dumper: registered kmsg dumper
3193 * @syslog: include the "<4>" prefixes
3194 * @line: buffer to copy the line to
3195 * @size: maximum size of the buffer
3196 * @len: length of line placed into buffer
3198 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3199 * record, and copy one record into the provided buffer.
3201 * Consecutive calls will return the next available record moving
3202 * towards the end of the buffer with the youngest messages.
3204 * A return value of FALSE indicates that there are no more records to
3207 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3209 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3210 char *line, size_t size, size_t *len)
3212 struct printk_log *msg;
3216 if (!dumper->active)
3219 if (dumper->cur_seq < log_first_seq) {
3220 /* messages are gone, move to first available one */
3221 dumper->cur_seq = log_first_seq;
3222 dumper->cur_idx = log_first_idx;
3226 if (dumper->cur_seq >= log_next_seq)
3229 msg = log_from_idx(dumper->cur_idx);
3230 l = msg_print_text(msg, syslog, printk_time, line, size);
3232 dumper->cur_idx = log_next(dumper->cur_idx);
3242 * kmsg_dump_get_line - retrieve one kmsg log line
3243 * @dumper: registered kmsg dumper
3244 * @syslog: include the "<4>" prefixes
3245 * @line: buffer to copy the line to
3246 * @size: maximum size of the buffer
3247 * @len: length of line placed into buffer
3249 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3250 * record, and copy one record into the provided buffer.
3252 * Consecutive calls will return the next available record moving
3253 * towards the end of the buffer with the youngest messages.
3255 * A return value of FALSE indicates that there are no more records to
3258 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3259 char *line, size_t size, size_t *len)
3261 unsigned long flags;
3264 logbuf_lock_irqsave(flags);
3265 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3266 logbuf_unlock_irqrestore(flags);
3270 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3273 * kmsg_dump_get_buffer - copy kmsg log lines
3274 * @dumper: registered kmsg dumper
3275 * @syslog: include the "<4>" prefixes
3276 * @buf: buffer to copy the line to
3277 * @size: maximum size of the buffer
3278 * @len: length of line placed into buffer
3280 * Start at the end of the kmsg buffer and fill the provided buffer
3281 * with as many of the the *youngest* kmsg records that fit into it.
3282 * If the buffer is large enough, all available kmsg records will be
3283 * copied with a single call.
3285 * Consecutive calls will fill the buffer with the next block of
3286 * available older records, not including the earlier retrieved ones.
3288 * A return value of FALSE indicates that there are no more records to
3291 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3292 char *buf, size_t size, size_t *len)
3294 unsigned long flags;
3301 bool time = printk_time;
3303 if (!dumper->active)
3306 logbuf_lock_irqsave(flags);
3307 if (dumper->cur_seq < log_first_seq) {
3308 /* messages are gone, move to first available one */
3309 dumper->cur_seq = log_first_seq;
3310 dumper->cur_idx = log_first_idx;
3314 if (dumper->cur_seq >= dumper->next_seq) {
3315 logbuf_unlock_irqrestore(flags);
3319 /* calculate length of entire buffer */
3320 seq = dumper->cur_seq;
3321 idx = dumper->cur_idx;
3322 while (seq < dumper->next_seq) {
3323 struct printk_log *msg = log_from_idx(idx);
3325 l += msg_print_text(msg, true, time, NULL, 0);
3326 idx = log_next(idx);
3330 /* move first record forward until length fits into the buffer */
3331 seq = dumper->cur_seq;
3332 idx = dumper->cur_idx;
3333 while (l >= size && seq < dumper->next_seq) {
3334 struct printk_log *msg = log_from_idx(idx);
3336 l -= msg_print_text(msg, true, time, NULL, 0);
3337 idx = log_next(idx);
3341 /* last message in next interation */
3346 while (seq < dumper->next_seq) {
3347 struct printk_log *msg = log_from_idx(idx);
3349 l += msg_print_text(msg, syslog, time, buf + l, size - l);
3350 idx = log_next(idx);
3354 dumper->next_seq = next_seq;
3355 dumper->next_idx = next_idx;
3357 logbuf_unlock_irqrestore(flags);
3363 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3366 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3367 * @dumper: registered kmsg dumper
3369 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3370 * kmsg_dump_get_buffer() can be called again and used multiple
3371 * times within the same dumper.dump() callback.
3373 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3375 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3377 dumper->cur_seq = clear_seq;
3378 dumper->cur_idx = clear_idx;
3379 dumper->next_seq = log_next_seq;
3380 dumper->next_idx = log_next_idx;
3384 * kmsg_dump_rewind - reset the interator
3385 * @dumper: registered kmsg dumper
3387 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3388 * kmsg_dump_get_buffer() can be called again and used multiple
3389 * times within the same dumper.dump() callback.
3391 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3393 unsigned long flags;
3395 logbuf_lock_irqsave(flags);
3396 kmsg_dump_rewind_nolock(dumper);
3397 logbuf_unlock_irqrestore(flags);
3399 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);