1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/kernel/printk.c
5 * Copyright (C) 1991, 1992 Linus Torvalds
7 * Modified to make sys_syslog() more flexible: added commands to
8 * return the last 4k of kernel messages, regardless of whether
9 * they've been read or not. Added option to suppress kernel printk's
10 * to the console. Added hook for sending the console messages
11 * elsewhere, in preparation for a serial line console (someday).
13 * Modified for sysctl support, 1/8/97, Chris Horn.
14 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
15 * manfred@colorfullife.com
16 * Rewrote bits to get rid of console_lock
17 * 01Mar01 Andrew Morton
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22 #include <linux/kernel.h>
24 #include <linux/tty.h>
25 #include <linux/tty_driver.h>
26 #include <linux/console.h>
27 #include <linux/init.h>
28 #include <linux/jiffies.h>
29 #include <linux/nmi.h>
30 #include <linux/module.h>
31 #include <linux/moduleparam.h>
32 #include <linux/delay.h>
33 #include <linux/smp.h>
34 #include <linux/security.h>
35 #include <linux/memblock.h>
36 #include <linux/syscalls.h>
37 #include <linux/crash_core.h>
38 #include <linux/kdb.h>
39 #include <linux/ratelimit.h>
40 #include <linux/kmsg_dump.h>
41 #include <linux/syslog.h>
42 #include <linux/cpu.h>
43 #include <linux/rculist.h>
44 #include <linux/poll.h>
45 #include <linux/irq_work.h>
46 #include <linux/ctype.h>
47 #include <linux/uio.h>
48 #include <linux/sched/clock.h>
49 #include <linux/sched/debug.h>
50 #include <linux/sched/task_stack.h>
52 #include <linux/uaccess.h>
53 #include <asm/sections.h>
55 #include <trace/events/initcall.h>
56 #define CREATE_TRACE_POINTS
57 #include <trace/events/printk.h>
59 #include "console_cmdline.h"
63 int console_printk[4] = {
64 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
65 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
66 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
67 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
69 EXPORT_SYMBOL_GPL(console_printk);
71 atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
72 EXPORT_SYMBOL(ignore_console_lock_warning);
75 * Low level drivers may need that to know if they can schedule in
76 * their unblank() callback or not. So let's export it.
79 EXPORT_SYMBOL(oops_in_progress);
82 * console_sem protects the console_drivers list, and also
83 * provides serialisation for access to the entire console
86 static DEFINE_SEMAPHORE(console_sem);
87 struct console *console_drivers;
88 EXPORT_SYMBOL_GPL(console_drivers);
91 * System may need to suppress printk message under certain
92 * circumstances, like after kernel panic happens.
94 int __read_mostly suppress_printk;
97 static struct lockdep_map console_lock_dep_map = {
98 .name = "console_lock"
102 enum devkmsg_log_bits {
103 __DEVKMSG_LOG_BIT_ON = 0,
104 __DEVKMSG_LOG_BIT_OFF,
105 __DEVKMSG_LOG_BIT_LOCK,
108 enum devkmsg_log_masks {
109 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
110 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
111 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
114 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
115 #define DEVKMSG_LOG_MASK_DEFAULT 0
117 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
119 static int __control_devkmsg(char *str)
126 len = str_has_prefix(str, "on");
128 devkmsg_log = DEVKMSG_LOG_MASK_ON;
132 len = str_has_prefix(str, "off");
134 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
138 len = str_has_prefix(str, "ratelimit");
140 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
147 static int __init control_devkmsg(char *str)
149 if (__control_devkmsg(str) < 0)
153 * Set sysctl string accordingly:
155 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
156 strcpy(devkmsg_log_str, "on");
157 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
158 strcpy(devkmsg_log_str, "off");
159 /* else "ratelimit" which is set by default. */
162 * Sysctl cannot change it anymore. The kernel command line setting of
163 * this parameter is to force the setting to be permanent throughout the
164 * runtime of the system. This is a precation measure against userspace
165 * trying to be a smarta** and attempting to change it up on us.
167 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
171 __setup("printk.devkmsg=", control_devkmsg);
173 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
175 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
176 void __user *buffer, size_t *lenp, loff_t *ppos)
178 char old_str[DEVKMSG_STR_MAX_SIZE];
183 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
187 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
190 err = proc_dostring(table, write, buffer, lenp, ppos);
195 err = __control_devkmsg(devkmsg_log_str);
198 * Do not accept an unknown string OR a known string with
201 if (err < 0 || (err + 1 != *lenp)) {
203 /* ... and restore old setting. */
205 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
214 /* Number of registered extended console drivers. */
215 static int nr_ext_console_drivers;
218 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
219 * macros instead of functions so that _RET_IP_ contains useful information.
221 #define down_console_sem() do { \
223 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
226 static int __down_trylock_console_sem(unsigned long ip)
232 * Here and in __up_console_sem() we need to be in safe mode,
233 * because spindump/WARN/etc from under console ->lock will
234 * deadlock in printk()->down_trylock_console_sem() otherwise.
236 printk_safe_enter_irqsave(flags);
237 lock_failed = down_trylock(&console_sem);
238 printk_safe_exit_irqrestore(flags);
242 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
245 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
247 static void __up_console_sem(unsigned long ip)
251 mutex_release(&console_lock_dep_map, ip);
253 printk_safe_enter_irqsave(flags);
255 printk_safe_exit_irqrestore(flags);
257 #define up_console_sem() __up_console_sem(_RET_IP_)
260 * This is used for debugging the mess that is the VT code by
261 * keeping track if we have the console semaphore held. It's
262 * definitely not the perfect debug tool (we don't know if _WE_
263 * hold it and are racing, but it helps tracking those weird code
264 * paths in the console code where we end up in places I want
265 * locked without the console sempahore held).
267 static int console_locked, console_suspended;
270 * If exclusive_console is non-NULL then only this console is to be printed to.
272 static struct console *exclusive_console;
275 * Array of consoles built from command line options (console=)
278 #define MAX_CMDLINECONSOLES 8
280 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
282 static int preferred_console = -1;
283 static bool has_preferred_console;
284 int console_set_on_cmdline;
285 EXPORT_SYMBOL(console_set_on_cmdline);
287 /* Flag: console code may call schedule() */
288 static int console_may_schedule;
290 enum con_msg_format_flags {
291 MSG_FORMAT_DEFAULT = 0,
292 MSG_FORMAT_SYSLOG = (1 << 0),
295 static int console_msg_format = MSG_FORMAT_DEFAULT;
298 * The printk log buffer consists of a chain of concatenated variable
299 * length records. Every record starts with a record header, containing
300 * the overall length of the record.
302 * The heads to the first and last entry in the buffer, as well as the
303 * sequence numbers of these entries are maintained when messages are
306 * If the heads indicate available messages, the length in the header
307 * tells the start next message. A length == 0 for the next message
308 * indicates a wrap-around to the beginning of the buffer.
310 * Every record carries the monotonic timestamp in microseconds, as well as
311 * the standard userspace syslog level and syslog facility. The usual
312 * kernel messages use LOG_KERN; userspace-injected messages always carry
313 * a matching syslog facility, by default LOG_USER. The origin of every
314 * message can be reliably determined that way.
316 * The human readable log message directly follows the message header. The
317 * length of the message text is stored in the header, the stored message
320 * Optionally, a message can carry a dictionary of properties (key/value pairs),
321 * to provide userspace with a machine-readable message context.
323 * Examples for well-defined, commonly used property names are:
324 * DEVICE=b12:8 device identifier
328 * +sound:card0 subsystem:devname
329 * SUBSYSTEM=pci driver-core subsystem name
331 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
332 * follows directly after a '=' character. Every property is terminated by
333 * a '\0' character. The last property is not terminated.
335 * Example of a message structure:
336 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
337 * 0008 34 00 record is 52 bytes long
338 * 000a 0b 00 text is 11 bytes long
339 * 000c 1f 00 dictionary is 23 bytes long
340 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
341 * 0010 69 74 27 73 20 61 20 6c "it's a l"
343 * 001b 44 45 56 49 43 "DEVIC"
344 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
345 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
347 * 0032 00 00 00 padding to next message header
349 * The 'struct printk_log' buffer header must never be directly exported to
350 * userspace, it is a kernel-private implementation detail that might
351 * need to be changed in the future, when the requirements change.
353 * /dev/kmsg exports the structured data in the following line format:
354 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
356 * Users of the export format should ignore possible additional values
357 * separated by ',', and find the message after the ';' character.
359 * The optional key/value pairs are attached as continuation lines starting
360 * with a space character and terminated by a newline. All possible
361 * non-prinatable characters are escaped in the "\xff" notation.
365 LOG_NEWLINE = 2, /* text ended with a newline */
366 LOG_CONT = 8, /* text is a fragment of a continuation line */
370 u64 ts_nsec; /* timestamp in nanoseconds */
371 u16 len; /* length of entire record */
372 u16 text_len; /* length of text buffer */
373 u16 dict_len; /* length of dictionary buffer */
374 u8 facility; /* syslog facility */
375 u8 flags:5; /* internal record flags */
376 u8 level:3; /* syslog level */
377 #ifdef CONFIG_PRINTK_CALLER
378 u32 caller_id; /* thread id or processor id */
381 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
382 __packed __aligned(4)
387 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
388 * within the scheduler's rq lock. It must be released before calling
389 * console_unlock() or anything else that might wake up a process.
391 DEFINE_RAW_SPINLOCK(logbuf_lock);
394 * Helper macros to lock/unlock logbuf_lock and switch between
395 * printk-safe/unsafe modes.
397 #define logbuf_lock_irq() \
399 printk_safe_enter_irq(); \
400 raw_spin_lock(&logbuf_lock); \
403 #define logbuf_unlock_irq() \
405 raw_spin_unlock(&logbuf_lock); \
406 printk_safe_exit_irq(); \
409 #define logbuf_lock_irqsave(flags) \
411 printk_safe_enter_irqsave(flags); \
412 raw_spin_lock(&logbuf_lock); \
415 #define logbuf_unlock_irqrestore(flags) \
417 raw_spin_unlock(&logbuf_lock); \
418 printk_safe_exit_irqrestore(flags); \
422 DECLARE_WAIT_QUEUE_HEAD(log_wait);
423 /* the next printk record to read by syslog(READ) or /proc/kmsg */
424 static u64 syslog_seq;
425 static u32 syslog_idx;
426 static size_t syslog_partial;
427 static bool syslog_time;
429 /* index and sequence number of the first record stored in the buffer */
430 static u64 log_first_seq;
431 static u32 log_first_idx;
433 /* index and sequence number of the next record to store in the buffer */
434 static u64 log_next_seq;
435 static u32 log_next_idx;
437 /* the next printk record to write to the console */
438 static u64 console_seq;
439 static u32 console_idx;
440 static u64 exclusive_console_stop_seq;
442 /* the next printk record to read after the last 'clear' command */
443 static u64 clear_seq;
444 static u32 clear_idx;
446 #ifdef CONFIG_PRINTK_CALLER
447 #define PREFIX_MAX 48
449 #define PREFIX_MAX 32
451 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
453 #define LOG_LEVEL(v) ((v) & 0x07)
454 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
457 #define LOG_ALIGN __alignof__(struct printk_log)
458 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
459 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
460 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
461 static char *log_buf = __log_buf;
462 static u32 log_buf_len = __LOG_BUF_LEN;
465 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
466 * per_cpu_areas are initialised. This variable is set to true when
467 * it's safe to access per-CPU data.
469 static bool __printk_percpu_data_ready __read_mostly;
471 bool printk_percpu_data_ready(void)
473 return __printk_percpu_data_ready;
476 /* Return log buffer address */
477 char *log_buf_addr_get(void)
482 /* Return log buffer size */
483 u32 log_buf_len_get(void)
488 /* human readable text of the record */
489 static char *log_text(const struct printk_log *msg)
491 return (char *)msg + sizeof(struct printk_log);
494 /* optional key/value pair dictionary attached to the record */
495 static char *log_dict(const struct printk_log *msg)
497 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
500 /* get record by index; idx must point to valid msg */
501 static struct printk_log *log_from_idx(u32 idx)
503 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
506 * A length == 0 record is the end of buffer marker. Wrap around and
507 * read the message at the start of the buffer.
510 return (struct printk_log *)log_buf;
514 /* get next record; idx must point to valid msg */
515 static u32 log_next(u32 idx)
517 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
519 /* length == 0 indicates the end of the buffer; wrap */
521 * A length == 0 record is the end of buffer marker. Wrap around and
522 * read the message at the start of the buffer as *this* one, and
523 * return the one after that.
526 msg = (struct printk_log *)log_buf;
529 return idx + msg->len;
533 * Check whether there is enough free space for the given message.
535 * The same values of first_idx and next_idx mean that the buffer
536 * is either empty or full.
538 * If the buffer is empty, we must respect the position of the indexes.
539 * They cannot be reset to the beginning of the buffer.
541 static int logbuf_has_space(u32 msg_size, bool empty)
545 if (log_next_idx > log_first_idx || empty)
546 free = max(log_buf_len - log_next_idx, log_first_idx);
548 free = log_first_idx - log_next_idx;
551 * We need space also for an empty header that signalizes wrapping
554 return free >= msg_size + sizeof(struct printk_log);
557 static int log_make_free_space(u32 msg_size)
559 while (log_first_seq < log_next_seq &&
560 !logbuf_has_space(msg_size, false)) {
561 /* drop old messages until we have enough contiguous space */
562 log_first_idx = log_next(log_first_idx);
566 if (clear_seq < log_first_seq) {
567 clear_seq = log_first_seq;
568 clear_idx = log_first_idx;
571 /* sequence numbers are equal, so the log buffer is empty */
572 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
578 /* compute the message size including the padding bytes */
579 static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
583 size = sizeof(struct printk_log) + text_len + dict_len;
584 *pad_len = (-size) & (LOG_ALIGN - 1);
591 * Define how much of the log buffer we could take at maximum. The value
592 * must be greater than two. Note that only half of the buffer is available
593 * when the index points to the middle.
595 #define MAX_LOG_TAKE_PART 4
596 static const char trunc_msg[] = "<truncated>";
598 static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
599 u16 *dict_len, u32 *pad_len)
602 * The message should not take the whole buffer. Otherwise, it might
603 * get removed too soon.
605 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
606 if (*text_len > max_text_len)
607 *text_len = max_text_len;
608 /* enable the warning message */
609 *trunc_msg_len = strlen(trunc_msg);
610 /* disable the "dict" completely */
612 /* compute the size again, count also the warning message */
613 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
616 /* insert record into the buffer, discard old ones, update heads */
617 static int log_store(u32 caller_id, int facility, int level,
618 enum log_flags flags, u64 ts_nsec,
619 const char *dict, u16 dict_len,
620 const char *text, u16 text_len)
622 struct printk_log *msg;
624 u16 trunc_msg_len = 0;
626 /* number of '\0' padding bytes to next message */
627 size = msg_used_size(text_len, dict_len, &pad_len);
629 if (log_make_free_space(size)) {
630 /* truncate the message if it is too long for empty buffer */
631 size = truncate_msg(&text_len, &trunc_msg_len,
632 &dict_len, &pad_len);
633 /* survive when the log buffer is too small for trunc_msg */
634 if (log_make_free_space(size))
638 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
640 * This message + an additional empty header does not fit
641 * at the end of the buffer. Add an empty header with len == 0
642 * to signify a wrap around.
644 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
649 msg = (struct printk_log *)(log_buf + log_next_idx);
650 memcpy(log_text(msg), text, text_len);
651 msg->text_len = text_len;
653 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
654 msg->text_len += trunc_msg_len;
656 memcpy(log_dict(msg), dict, dict_len);
657 msg->dict_len = dict_len;
658 msg->facility = facility;
659 msg->level = level & 7;
660 msg->flags = flags & 0x1f;
662 msg->ts_nsec = ts_nsec;
664 msg->ts_nsec = local_clock();
665 #ifdef CONFIG_PRINTK_CALLER
666 msg->caller_id = caller_id;
668 memset(log_dict(msg) + dict_len, 0, pad_len);
672 log_next_idx += msg->len;
675 return msg->text_len;
678 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
680 static int syslog_action_restricted(int type)
685 * Unless restricted, we allow "read all" and "get buffer size"
688 return type != SYSLOG_ACTION_READ_ALL &&
689 type != SYSLOG_ACTION_SIZE_BUFFER;
692 static int check_syslog_permissions(int type, int source)
695 * If this is from /proc/kmsg and we've already opened it, then we've
696 * already done the capabilities checks at open time.
698 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
701 if (syslog_action_restricted(type)) {
702 if (capable(CAP_SYSLOG))
705 * For historical reasons, accept CAP_SYS_ADMIN too, with
708 if (capable(CAP_SYS_ADMIN)) {
709 pr_warn_once("%s (%d): Attempt to access syslog with "
710 "CAP_SYS_ADMIN but no CAP_SYSLOG "
712 current->comm, task_pid_nr(current));
718 return security_syslog(type);
721 static void append_char(char **pp, char *e, char c)
727 static ssize_t msg_print_ext_header(char *buf, size_t size,
728 struct printk_log *msg, u64 seq)
730 u64 ts_usec = msg->ts_nsec;
732 #ifdef CONFIG_PRINTK_CALLER
733 u32 id = msg->caller_id;
735 snprintf(caller, sizeof(caller), ",caller=%c%u",
736 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
741 do_div(ts_usec, 1000);
743 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
744 (msg->facility << 3) | msg->level, seq, ts_usec,
745 msg->flags & LOG_CONT ? 'c' : '-', caller);
748 static ssize_t msg_print_ext_body(char *buf, size_t size,
749 char *dict, size_t dict_len,
750 char *text, size_t text_len)
752 char *p = buf, *e = buf + size;
755 /* escape non-printable characters */
756 for (i = 0; i < text_len; i++) {
757 unsigned char c = text[i];
759 if (c < ' ' || c >= 127 || c == '\\')
760 p += scnprintf(p, e - p, "\\x%02x", c);
762 append_char(&p, e, c);
764 append_char(&p, e, '\n');
769 for (i = 0; i < dict_len; i++) {
770 unsigned char c = dict[i];
773 append_char(&p, e, ' ');
778 append_char(&p, e, '\n');
783 if (c < ' ' || c >= 127 || c == '\\') {
784 p += scnprintf(p, e - p, "\\x%02x", c);
788 append_char(&p, e, c);
790 append_char(&p, e, '\n');
796 /* /dev/kmsg - userspace message inject/listen interface */
797 struct devkmsg_user {
800 struct ratelimit_state rs;
802 char buf[CONSOLE_EXT_LOG_MAX];
805 static __printf(3, 4) __cold
806 int devkmsg_emit(int facility, int level, const char *fmt, ...)
812 r = vprintk_emit(facility, level, NULL, 0, fmt, args);
818 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
821 int level = default_message_loglevel;
822 int facility = 1; /* LOG_USER */
823 struct file *file = iocb->ki_filp;
824 struct devkmsg_user *user = file->private_data;
825 size_t len = iov_iter_count(from);
828 if (!user || len > LOG_LINE_MAX)
831 /* Ignore when user logging is disabled. */
832 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
835 /* Ratelimit when not explicitly enabled. */
836 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
837 if (!___ratelimit(&user->rs, current->comm))
841 buf = kmalloc(len+1, GFP_KERNEL);
846 if (!copy_from_iter_full(buf, len, from)) {
852 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
853 * the decimal value represents 32bit, the lower 3 bit are the log
854 * level, the rest are the log facility.
856 * If no prefix or no userspace facility is specified, we
857 * enforce LOG_USER, to be able to reliably distinguish
858 * kernel-generated messages from userspace-injected ones.
861 if (line[0] == '<') {
865 u = simple_strtoul(line + 1, &endp, 10);
866 if (endp && endp[0] == '>') {
867 level = LOG_LEVEL(u);
868 if (LOG_FACILITY(u) != 0)
869 facility = LOG_FACILITY(u);
876 devkmsg_emit(facility, level, "%s", line);
881 static ssize_t devkmsg_read(struct file *file, char __user *buf,
882 size_t count, loff_t *ppos)
884 struct devkmsg_user *user = file->private_data;
885 struct printk_log *msg;
892 ret = mutex_lock_interruptible(&user->lock);
897 while (user->seq == log_next_seq) {
898 if (file->f_flags & O_NONBLOCK) {
905 ret = wait_event_interruptible(log_wait,
906 user->seq != log_next_seq);
912 if (user->seq < log_first_seq) {
913 /* our last seen message is gone, return error and reset */
914 user->idx = log_first_idx;
915 user->seq = log_first_seq;
921 msg = log_from_idx(user->idx);
922 len = msg_print_ext_header(user->buf, sizeof(user->buf),
924 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
925 log_dict(msg), msg->dict_len,
926 log_text(msg), msg->text_len);
928 user->idx = log_next(user->idx);
937 if (copy_to_user(buf, user->buf, len)) {
943 mutex_unlock(&user->lock);
947 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
949 struct devkmsg_user *user = file->private_data;
960 /* the first record */
961 user->idx = log_first_idx;
962 user->seq = log_first_seq;
966 * The first record after the last SYSLOG_ACTION_CLEAR,
967 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
968 * changes no global state, and does not clear anything.
970 user->idx = clear_idx;
971 user->seq = clear_seq;
974 /* after the last record */
975 user->idx = log_next_idx;
976 user->seq = log_next_seq;
980 * It isn't supported due to the record nature of this
981 * interface: _SET _DATA and _END point to very specific
982 * record positions, while _CUR would be more useful in case
983 * of a byte-based log. Because of that, return the default
984 * errno value for invalid seek operation.
995 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
997 struct devkmsg_user *user = file->private_data;
1001 return EPOLLERR|EPOLLNVAL;
1003 poll_wait(file, &log_wait, wait);
1006 if (user->seq < log_next_seq) {
1007 /* return error when data has vanished underneath us */
1008 if (user->seq < log_first_seq)
1009 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
1011 ret = EPOLLIN|EPOLLRDNORM;
1013 logbuf_unlock_irq();
1018 static int devkmsg_open(struct inode *inode, struct file *file)
1020 struct devkmsg_user *user;
1023 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
1026 /* write-only does not need any file context */
1027 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
1028 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
1029 SYSLOG_FROM_READER);
1034 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
1038 ratelimit_default_init(&user->rs);
1039 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
1041 mutex_init(&user->lock);
1044 user->idx = log_first_idx;
1045 user->seq = log_first_seq;
1046 logbuf_unlock_irq();
1048 file->private_data = user;
1052 static int devkmsg_release(struct inode *inode, struct file *file)
1054 struct devkmsg_user *user = file->private_data;
1059 ratelimit_state_exit(&user->rs);
1061 mutex_destroy(&user->lock);
1066 const struct file_operations kmsg_fops = {
1067 .open = devkmsg_open,
1068 .read = devkmsg_read,
1069 .write_iter = devkmsg_write,
1070 .llseek = devkmsg_llseek,
1071 .poll = devkmsg_poll,
1072 .release = devkmsg_release,
1075 #ifdef CONFIG_CRASH_CORE
1077 * This appends the listed symbols to /proc/vmcore
1079 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1080 * obtain access to symbols that are otherwise very difficult to locate. These
1081 * symbols are specifically used so that utilities can access and extract the
1082 * dmesg log from a vmcore file after a crash.
1084 void log_buf_vmcoreinfo_setup(void)
1086 VMCOREINFO_SYMBOL(log_buf);
1087 VMCOREINFO_SYMBOL(log_buf_len);
1088 VMCOREINFO_SYMBOL(log_first_idx);
1089 VMCOREINFO_SYMBOL(clear_idx);
1090 VMCOREINFO_SYMBOL(log_next_idx);
1092 * Export struct printk_log size and field offsets. User space tools can
1093 * parse it and detect any changes to structure down the line.
1095 VMCOREINFO_STRUCT_SIZE(printk_log);
1096 VMCOREINFO_OFFSET(printk_log, ts_nsec);
1097 VMCOREINFO_OFFSET(printk_log, len);
1098 VMCOREINFO_OFFSET(printk_log, text_len);
1099 VMCOREINFO_OFFSET(printk_log, dict_len);
1100 #ifdef CONFIG_PRINTK_CALLER
1101 VMCOREINFO_OFFSET(printk_log, caller_id);
1106 /* requested log_buf_len from kernel cmdline */
1107 static unsigned long __initdata new_log_buf_len;
1109 /* we practice scaling the ring buffer by powers of 2 */
1110 static void __init log_buf_len_update(u64 size)
1112 if (size > (u64)LOG_BUF_LEN_MAX) {
1113 size = (u64)LOG_BUF_LEN_MAX;
1114 pr_err("log_buf over 2G is not supported.\n");
1118 size = roundup_pow_of_two(size);
1119 if (size > log_buf_len)
1120 new_log_buf_len = (unsigned long)size;
1123 /* save requested log_buf_len since it's too early to process it */
1124 static int __init log_buf_len_setup(char *str)
1131 size = memparse(str, &str);
1133 log_buf_len_update(size);
1137 early_param("log_buf_len", log_buf_len_setup);
1140 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1142 static void __init log_buf_add_cpu(void)
1144 unsigned int cpu_extra;
1147 * archs should set up cpu_possible_bits properly with
1148 * set_cpu_possible() after setup_arch() but just in
1149 * case lets ensure this is valid.
1151 if (num_possible_cpus() == 1)
1154 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1156 /* by default this will only continue through for large > 64 CPUs */
1157 if (cpu_extra <= __LOG_BUF_LEN / 2)
1160 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1161 __LOG_CPU_MAX_BUF_LEN);
1162 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1164 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1166 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1168 #else /* !CONFIG_SMP */
1169 static inline void log_buf_add_cpu(void) {}
1170 #endif /* CONFIG_SMP */
1172 static void __init set_percpu_data_ready(void)
1175 /* Make sure we set this flag only after printk_safe() init is done */
1177 __printk_percpu_data_ready = true;
1180 void __init setup_log_buf(int early)
1182 unsigned long flags;
1187 * Some archs call setup_log_buf() multiple times - first is very
1188 * early, e.g. from setup_arch(), and second - when percpu_areas
1192 set_percpu_data_ready();
1194 if (log_buf != __log_buf)
1197 if (!early && !new_log_buf_len)
1200 if (!new_log_buf_len)
1203 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1204 if (unlikely(!new_log_buf)) {
1205 pr_err("log_buf_len: %lu bytes not available\n",
1210 logbuf_lock_irqsave(flags);
1211 log_buf_len = new_log_buf_len;
1212 log_buf = new_log_buf;
1213 new_log_buf_len = 0;
1214 free = __LOG_BUF_LEN - log_next_idx;
1215 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1216 logbuf_unlock_irqrestore(flags);
1218 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1219 pr_info("early log buf free: %u(%u%%)\n",
1220 free, (free * 100) / __LOG_BUF_LEN);
1223 static bool __read_mostly ignore_loglevel;
1225 static int __init ignore_loglevel_setup(char *str)
1227 ignore_loglevel = true;
1228 pr_info("debug: ignoring loglevel setting.\n");
1233 early_param("ignore_loglevel", ignore_loglevel_setup);
1234 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1235 MODULE_PARM_DESC(ignore_loglevel,
1236 "ignore loglevel setting (prints all kernel messages to the console)");
1238 static bool suppress_message_printing(int level)
1240 return (level >= console_loglevel && !ignore_loglevel);
1243 #ifdef CONFIG_BOOT_PRINTK_DELAY
1245 static int boot_delay; /* msecs delay after each printk during bootup */
1246 static unsigned long long loops_per_msec; /* based on boot_delay */
1248 static int __init boot_delay_setup(char *str)
1252 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1253 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1255 get_option(&str, &boot_delay);
1256 if (boot_delay > 10 * 1000)
1259 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1260 "HZ: %d, loops_per_msec: %llu\n",
1261 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1264 early_param("boot_delay", boot_delay_setup);
1266 static void boot_delay_msec(int level)
1268 unsigned long long k;
1269 unsigned long timeout;
1271 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1272 || suppress_message_printing(level)) {
1276 k = (unsigned long long)loops_per_msec * boot_delay;
1278 timeout = jiffies + msecs_to_jiffies(boot_delay);
1283 * use (volatile) jiffies to prevent
1284 * compiler reduction; loop termination via jiffies
1285 * is secondary and may or may not happen.
1287 if (time_after(jiffies, timeout))
1289 touch_nmi_watchdog();
1293 static inline void boot_delay_msec(int level)
1298 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1299 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1301 static size_t print_syslog(unsigned int level, char *buf)
1303 return sprintf(buf, "<%u>", level);
1306 static size_t print_time(u64 ts, char *buf)
1308 unsigned long rem_nsec = do_div(ts, 1000000000);
1310 return sprintf(buf, "[%5lu.%06lu]",
1311 (unsigned long)ts, rem_nsec / 1000);
1314 #ifdef CONFIG_PRINTK_CALLER
1315 static size_t print_caller(u32 id, char *buf)
1319 snprintf(caller, sizeof(caller), "%c%u",
1320 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1321 return sprintf(buf, "[%6s]", caller);
1324 #define print_caller(id, buf) 0
1327 static size_t print_prefix(const struct printk_log *msg, bool syslog,
1328 bool time, char *buf)
1333 len = print_syslog((msg->facility << 3) | msg->level, buf);
1336 len += print_time(msg->ts_nsec, buf + len);
1338 len += print_caller(msg->caller_id, buf + len);
1340 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1348 static size_t msg_print_text(const struct printk_log *msg, bool syslog,
1349 bool time, char *buf, size_t size)
1351 const char *text = log_text(msg);
1352 size_t text_size = msg->text_len;
1354 char prefix[PREFIX_MAX];
1355 const size_t prefix_len = print_prefix(msg, syslog, time, prefix);
1358 const char *next = memchr(text, '\n', text_size);
1362 text_len = next - text;
1364 text_size -= next - text;
1366 text_len = text_size;
1370 if (prefix_len + text_len + 1 >= size - len)
1373 memcpy(buf + len, prefix, prefix_len);
1375 memcpy(buf + len, text, text_len);
1379 /* SYSLOG_ACTION_* buffer size only calculation */
1380 len += prefix_len + text_len + 1;
1389 static int syslog_print(char __user *buf, int size)
1392 struct printk_log *msg;
1395 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1404 if (syslog_seq < log_first_seq) {
1405 /* messages are gone, move to first one */
1406 syslog_seq = log_first_seq;
1407 syslog_idx = log_first_idx;
1410 if (syslog_seq == log_next_seq) {
1411 logbuf_unlock_irq();
1416 * To keep reading/counting partial line consistent,
1417 * use printk_time value as of the beginning of a line.
1419 if (!syslog_partial)
1420 syslog_time = printk_time;
1422 skip = syslog_partial;
1423 msg = log_from_idx(syslog_idx);
1424 n = msg_print_text(msg, true, syslog_time, text,
1425 LOG_LINE_MAX + PREFIX_MAX);
1426 if (n - syslog_partial <= size) {
1427 /* message fits into buffer, move forward */
1428 syslog_idx = log_next(syslog_idx);
1430 n -= syslog_partial;
1433 /* partial read(), remember position */
1435 syslog_partial += n;
1438 logbuf_unlock_irq();
1443 if (copy_to_user(buf, text + skip, n)) {
1458 static int syslog_print_all(char __user *buf, int size, bool clear)
1467 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1474 * Find first record that fits, including all following records,
1475 * into the user-provided buffer for this dump.
1479 while (seq < log_next_seq) {
1480 struct printk_log *msg = log_from_idx(idx);
1482 len += msg_print_text(msg, true, time, NULL, 0);
1483 idx = log_next(idx);
1487 /* move first record forward until length fits into the buffer */
1490 while (len > size && seq < log_next_seq) {
1491 struct printk_log *msg = log_from_idx(idx);
1493 len -= msg_print_text(msg, true, time, NULL, 0);
1494 idx = log_next(idx);
1498 /* last message fitting into this dump */
1499 next_seq = log_next_seq;
1502 while (len >= 0 && seq < next_seq) {
1503 struct printk_log *msg = log_from_idx(idx);
1504 int textlen = msg_print_text(msg, true, time, text,
1505 LOG_LINE_MAX + PREFIX_MAX);
1507 idx = log_next(idx);
1510 logbuf_unlock_irq();
1511 if (copy_to_user(buf + len, text, textlen))
1517 if (seq < log_first_seq) {
1518 /* messages are gone, move to next one */
1519 seq = log_first_seq;
1520 idx = log_first_idx;
1525 clear_seq = log_next_seq;
1526 clear_idx = log_next_idx;
1528 logbuf_unlock_irq();
1534 static void syslog_clear(void)
1537 clear_seq = log_next_seq;
1538 clear_idx = log_next_idx;
1539 logbuf_unlock_irq();
1542 int do_syslog(int type, char __user *buf, int len, int source)
1545 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1548 error = check_syslog_permissions(type, source);
1553 case SYSLOG_ACTION_CLOSE: /* Close log */
1555 case SYSLOG_ACTION_OPEN: /* Open log */
1557 case SYSLOG_ACTION_READ: /* Read from log */
1558 if (!buf || len < 0)
1562 if (!access_ok(buf, len))
1564 error = wait_event_interruptible(log_wait,
1565 syslog_seq != log_next_seq);
1568 error = syslog_print(buf, len);
1570 /* Read/clear last kernel messages */
1571 case SYSLOG_ACTION_READ_CLEAR:
1574 /* Read last kernel messages */
1575 case SYSLOG_ACTION_READ_ALL:
1576 if (!buf || len < 0)
1580 if (!access_ok(buf, len))
1582 error = syslog_print_all(buf, len, clear);
1584 /* Clear ring buffer */
1585 case SYSLOG_ACTION_CLEAR:
1588 /* Disable logging to console */
1589 case SYSLOG_ACTION_CONSOLE_OFF:
1590 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1591 saved_console_loglevel = console_loglevel;
1592 console_loglevel = minimum_console_loglevel;
1594 /* Enable logging to console */
1595 case SYSLOG_ACTION_CONSOLE_ON:
1596 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1597 console_loglevel = saved_console_loglevel;
1598 saved_console_loglevel = LOGLEVEL_DEFAULT;
1601 /* Set level of messages printed to console */
1602 case SYSLOG_ACTION_CONSOLE_LEVEL:
1603 if (len < 1 || len > 8)
1605 if (len < minimum_console_loglevel)
1606 len = minimum_console_loglevel;
1607 console_loglevel = len;
1608 /* Implicitly re-enable logging to console */
1609 saved_console_loglevel = LOGLEVEL_DEFAULT;
1611 /* Number of chars in the log buffer */
1612 case SYSLOG_ACTION_SIZE_UNREAD:
1614 if (syslog_seq < log_first_seq) {
1615 /* messages are gone, move to first one */
1616 syslog_seq = log_first_seq;
1617 syslog_idx = log_first_idx;
1620 if (source == SYSLOG_FROM_PROC) {
1622 * Short-cut for poll(/"proc/kmsg") which simply checks
1623 * for pending data, not the size; return the count of
1624 * records, not the length.
1626 error = log_next_seq - syslog_seq;
1628 u64 seq = syslog_seq;
1629 u32 idx = syslog_idx;
1630 bool time = syslog_partial ? syslog_time : printk_time;
1632 while (seq < log_next_seq) {
1633 struct printk_log *msg = log_from_idx(idx);
1635 error += msg_print_text(msg, true, time, NULL,
1638 idx = log_next(idx);
1641 error -= syslog_partial;
1643 logbuf_unlock_irq();
1645 /* Size of the log buffer */
1646 case SYSLOG_ACTION_SIZE_BUFFER:
1647 error = log_buf_len;
1657 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1659 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1663 * Special console_lock variants that help to reduce the risk of soft-lockups.
1664 * They allow to pass console_lock to another printk() call using a busy wait.
1667 #ifdef CONFIG_LOCKDEP
1668 static struct lockdep_map console_owner_dep_map = {
1669 .name = "console_owner"
1673 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1674 static struct task_struct *console_owner;
1675 static bool console_waiter;
1678 * console_lock_spinning_enable - mark beginning of code where another
1679 * thread might safely busy wait
1681 * This basically converts console_lock into a spinlock. This marks
1682 * the section where the console_lock owner can not sleep, because
1683 * there may be a waiter spinning (like a spinlock). Also it must be
1684 * ready to hand over the lock at the end of the section.
1686 static void console_lock_spinning_enable(void)
1688 raw_spin_lock(&console_owner_lock);
1689 console_owner = current;
1690 raw_spin_unlock(&console_owner_lock);
1692 /* The waiter may spin on us after setting console_owner */
1693 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1697 * console_lock_spinning_disable_and_check - mark end of code where another
1698 * thread was able to busy wait and check if there is a waiter
1700 * This is called at the end of the section where spinning is allowed.
1701 * It has two functions. First, it is a signal that it is no longer
1702 * safe to start busy waiting for the lock. Second, it checks if
1703 * there is a busy waiter and passes the lock rights to her.
1705 * Important: Callers lose the lock if there was a busy waiter.
1706 * They must not touch items synchronized by console_lock
1709 * Return: 1 if the lock rights were passed, 0 otherwise.
1711 static int console_lock_spinning_disable_and_check(void)
1715 raw_spin_lock(&console_owner_lock);
1716 waiter = READ_ONCE(console_waiter);
1717 console_owner = NULL;
1718 raw_spin_unlock(&console_owner_lock);
1721 spin_release(&console_owner_dep_map, _THIS_IP_);
1725 /* The waiter is now free to continue */
1726 WRITE_ONCE(console_waiter, false);
1728 spin_release(&console_owner_dep_map, _THIS_IP_);
1731 * Hand off console_lock to waiter. The waiter will perform
1732 * the up(). After this, the waiter is the console_lock owner.
1734 mutex_release(&console_lock_dep_map, _THIS_IP_);
1739 * console_trylock_spinning - try to get console_lock by busy waiting
1741 * This allows to busy wait for the console_lock when the current
1742 * owner is running in specially marked sections. It means that
1743 * the current owner is running and cannot reschedule until it
1744 * is ready to lose the lock.
1746 * Return: 1 if we got the lock, 0 othrewise
1748 static int console_trylock_spinning(void)
1750 struct task_struct *owner = NULL;
1753 unsigned long flags;
1755 if (console_trylock())
1758 printk_safe_enter_irqsave(flags);
1760 raw_spin_lock(&console_owner_lock);
1761 owner = READ_ONCE(console_owner);
1762 waiter = READ_ONCE(console_waiter);
1763 if (!waiter && owner && owner != current) {
1764 WRITE_ONCE(console_waiter, true);
1767 raw_spin_unlock(&console_owner_lock);
1770 * If there is an active printk() writing to the
1771 * consoles, instead of having it write our data too,
1772 * see if we can offload that load from the active
1773 * printer, and do some printing ourselves.
1774 * Go into a spin only if there isn't already a waiter
1775 * spinning, and there is an active printer, and
1776 * that active printer isn't us (recursive printk?).
1779 printk_safe_exit_irqrestore(flags);
1783 /* We spin waiting for the owner to release us */
1784 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1785 /* Owner will clear console_waiter on hand off */
1786 while (READ_ONCE(console_waiter))
1788 spin_release(&console_owner_dep_map, _THIS_IP_);
1790 printk_safe_exit_irqrestore(flags);
1792 * The owner passed the console lock to us.
1793 * Since we did not spin on console lock, annotate
1794 * this as a trylock. Otherwise lockdep will
1797 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1803 * Call the console drivers, asking them to write out
1804 * log_buf[start] to log_buf[end - 1].
1805 * The console_lock must be held.
1807 static void call_console_drivers(const char *ext_text, size_t ext_len,
1808 const char *text, size_t len)
1810 struct console *con;
1812 trace_console_rcuidle(text, len);
1814 for_each_console(con) {
1815 if (exclusive_console && con != exclusive_console)
1817 if (!(con->flags & CON_ENABLED))
1821 if (!cpu_online(smp_processor_id()) &&
1822 !(con->flags & CON_ANYTIME))
1824 if (con->flags & CON_EXTENDED)
1825 con->write(con, ext_text, ext_len);
1827 con->write(con, text, len);
1831 int printk_delay_msec __read_mostly;
1833 static inline void printk_delay(void)
1835 if (unlikely(printk_delay_msec)) {
1836 int m = printk_delay_msec;
1840 touch_nmi_watchdog();
1845 static inline u32 printk_caller_id(void)
1847 return in_task() ? task_pid_nr(current) :
1848 0x80000000 + raw_smp_processor_id();
1852 * Continuation lines are buffered, and not committed to the record buffer
1853 * until the line is complete, or a race forces it. The line fragments
1854 * though, are printed immediately to the consoles to ensure everything has
1855 * reached the console in case of a kernel crash.
1857 static struct cont {
1858 char buf[LOG_LINE_MAX];
1859 size_t len; /* length == 0 means unused buffer */
1860 u32 caller_id; /* printk_caller_id() of first print */
1861 u64 ts_nsec; /* time of first print */
1862 u8 level; /* log level of first message */
1863 u8 facility; /* log facility of first message */
1864 enum log_flags flags; /* prefix, newline flags */
1867 static void cont_flush(void)
1872 log_store(cont.caller_id, cont.facility, cont.level, cont.flags,
1873 cont.ts_nsec, NULL, 0, cont.buf, cont.len);
1877 static bool cont_add(u32 caller_id, int facility, int level,
1878 enum log_flags flags, const char *text, size_t len)
1880 /* If the line gets too long, split it up in separate records. */
1881 if (cont.len + len > sizeof(cont.buf)) {
1887 cont.facility = facility;
1889 cont.caller_id = caller_id;
1890 cont.ts_nsec = local_clock();
1894 memcpy(cont.buf + cont.len, text, len);
1897 // The original flags come from the first line,
1898 // but later continuations can add a newline.
1899 if (flags & LOG_NEWLINE) {
1900 cont.flags |= LOG_NEWLINE;
1907 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)
1909 const u32 caller_id = printk_caller_id();
1912 * If an earlier line was buffered, and we're a continuation
1913 * write from the same context, try to add it to the buffer.
1916 if (cont.caller_id == caller_id && (lflags & LOG_CONT)) {
1917 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1920 /* Otherwise, make sure it's flushed */
1924 /* Skip empty continuation lines that couldn't be added - they just flush */
1925 if (!text_len && (lflags & LOG_CONT))
1928 /* If it doesn't end in a newline, try to buffer the current line */
1929 if (!(lflags & LOG_NEWLINE)) {
1930 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1934 /* Store it in the record log */
1935 return log_store(caller_id, facility, level, lflags, 0,
1936 dict, dictlen, text, text_len);
1939 /* Must be called under logbuf_lock. */
1940 int vprintk_store(int facility, int level,
1941 const char *dict, size_t dictlen,
1942 const char *fmt, va_list args)
1944 static char textbuf[LOG_LINE_MAX];
1945 char *text = textbuf;
1947 enum log_flags lflags = 0;
1950 * The printf needs to come first; we need the syslog
1951 * prefix which might be passed-in as a parameter.
1953 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1955 /* mark and strip a trailing newline */
1956 if (text_len && text[text_len-1] == '\n') {
1958 lflags |= LOG_NEWLINE;
1961 /* strip kernel syslog prefix and extract log level or control flags */
1962 if (facility == 0) {
1965 while ((kern_level = printk_get_level(text)) != 0) {
1966 switch (kern_level) {
1968 if (level == LOGLEVEL_DEFAULT)
1969 level = kern_level - '0';
1971 case 'c': /* KERN_CONT */
1980 if (level == LOGLEVEL_DEFAULT)
1981 level = default_message_loglevel;
1984 lflags |= LOG_NEWLINE;
1986 return log_output(facility, level, lflags,
1987 dict, dictlen, text, text_len);
1990 asmlinkage int vprintk_emit(int facility, int level,
1991 const char *dict, size_t dictlen,
1992 const char *fmt, va_list args)
1995 bool in_sched = false, pending_output;
1996 unsigned long flags;
1999 /* Suppress unimportant messages after panic happens */
2000 if (unlikely(suppress_printk))
2003 if (level == LOGLEVEL_SCHED) {
2004 level = LOGLEVEL_DEFAULT;
2008 boot_delay_msec(level);
2011 /* This stops the holder of console_sem just where we want him */
2012 logbuf_lock_irqsave(flags);
2013 curr_log_seq = log_next_seq;
2014 printed_len = vprintk_store(facility, level, dict, dictlen, fmt, args);
2015 pending_output = (curr_log_seq != log_next_seq);
2016 logbuf_unlock_irqrestore(flags);
2018 /* If called from the scheduler, we can not call up(). */
2019 if (!in_sched && pending_output) {
2021 * Disable preemption to avoid being preempted while holding
2022 * console_sem which would prevent anyone from printing to
2027 * Try to acquire and then immediately release the console
2028 * semaphore. The release will print out buffers and wake up
2029 * /dev/kmsg and syslog() users.
2031 if (console_trylock_spinning())
2040 EXPORT_SYMBOL(vprintk_emit);
2042 asmlinkage int vprintk(const char *fmt, va_list args)
2044 return vprintk_func(fmt, args);
2046 EXPORT_SYMBOL(vprintk);
2048 int vprintk_default(const char *fmt, va_list args)
2052 #ifdef CONFIG_KGDB_KDB
2053 /* Allow to pass printk() to kdb but avoid a recursion. */
2054 if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
2055 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
2059 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
2063 EXPORT_SYMBOL_GPL(vprintk_default);
2066 * printk - print a kernel message
2067 * @fmt: format string
2069 * This is printk(). It can be called from any context. We want it to work.
2071 * We try to grab the console_lock. If we succeed, it's easy - we log the
2072 * output and call the console drivers. If we fail to get the semaphore, we
2073 * place the output into the log buffer and return. The current holder of
2074 * the console_sem will notice the new output in console_unlock(); and will
2075 * send it to the consoles before releasing the lock.
2077 * One effect of this deferred printing is that code which calls printk() and
2078 * then changes console_loglevel may break. This is because console_loglevel
2079 * is inspected when the actual printing occurs.
2084 * See the vsnprintf() documentation for format string extensions over C99.
2086 asmlinkage __visible int printk(const char *fmt, ...)
2091 va_start(args, fmt);
2092 r = vprintk_func(fmt, args);
2097 EXPORT_SYMBOL(printk);
2099 #else /* CONFIG_PRINTK */
2101 #define LOG_LINE_MAX 0
2102 #define PREFIX_MAX 0
2103 #define printk_time false
2105 static u64 syslog_seq;
2106 static u32 syslog_idx;
2107 static u64 console_seq;
2108 static u32 console_idx;
2109 static u64 exclusive_console_stop_seq;
2110 static u64 log_first_seq;
2111 static u32 log_first_idx;
2112 static u64 log_next_seq;
2113 static char *log_text(const struct printk_log *msg) { return NULL; }
2114 static char *log_dict(const struct printk_log *msg) { return NULL; }
2115 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2116 static u32 log_next(u32 idx) { return 0; }
2117 static ssize_t msg_print_ext_header(char *buf, size_t size,
2118 struct printk_log *msg,
2119 u64 seq) { return 0; }
2120 static ssize_t msg_print_ext_body(char *buf, size_t size,
2121 char *dict, size_t dict_len,
2122 char *text, size_t text_len) { return 0; }
2123 static void console_lock_spinning_enable(void) { }
2124 static int console_lock_spinning_disable_and_check(void) { return 0; }
2125 static void call_console_drivers(const char *ext_text, size_t ext_len,
2126 const char *text, size_t len) {}
2127 static size_t msg_print_text(const struct printk_log *msg, bool syslog,
2128 bool time, char *buf, size_t size) { return 0; }
2129 static bool suppress_message_printing(int level) { return false; }
2131 #endif /* CONFIG_PRINTK */
2133 #ifdef CONFIG_EARLY_PRINTK
2134 struct console *early_console;
2136 asmlinkage __visible void early_printk(const char *fmt, ...)
2146 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2149 early_console->write(early_console, buf, n);
2153 static int __add_preferred_console(char *name, int idx, char *options,
2154 char *brl_options, bool user_specified)
2156 struct console_cmdline *c;
2160 * See if this tty is not yet registered, and
2161 * if we have a slot free.
2163 for (i = 0, c = console_cmdline;
2164 i < MAX_CMDLINECONSOLES && c->name[0];
2166 if (strcmp(c->name, name) == 0 && c->index == idx) {
2168 preferred_console = i;
2170 c->user_specified = true;
2174 if (i == MAX_CMDLINECONSOLES)
2177 preferred_console = i;
2178 strlcpy(c->name, name, sizeof(c->name));
2179 c->options = options;
2180 c->user_specified = user_specified;
2181 braille_set_options(c, brl_options);
2187 static int __init console_msg_format_setup(char *str)
2189 if (!strcmp(str, "syslog"))
2190 console_msg_format = MSG_FORMAT_SYSLOG;
2191 if (!strcmp(str, "default"))
2192 console_msg_format = MSG_FORMAT_DEFAULT;
2195 __setup("console_msg_format=", console_msg_format_setup);
2198 * Set up a console. Called via do_early_param() in init/main.c
2199 * for each "console=" parameter in the boot command line.
2201 static int __init console_setup(char *str)
2203 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2204 char *s, *options, *brl_options = NULL;
2210 if (_braille_console_setup(&str, &brl_options))
2214 * Decode str into name, index, options.
2216 if (str[0] >= '0' && str[0] <= '9') {
2217 strcpy(buf, "ttyS");
2218 strncpy(buf + 4, str, sizeof(buf) - 5);
2220 strncpy(buf, str, sizeof(buf) - 1);
2222 buf[sizeof(buf) - 1] = 0;
2223 options = strchr(str, ',');
2227 if (!strcmp(str, "ttya"))
2228 strcpy(buf, "ttyS0");
2229 if (!strcmp(str, "ttyb"))
2230 strcpy(buf, "ttyS1");
2232 for (s = buf; *s; s++)
2233 if (isdigit(*s) || *s == ',')
2235 idx = simple_strtoul(s, NULL, 10);
2238 __add_preferred_console(buf, idx, options, brl_options, true);
2239 console_set_on_cmdline = 1;
2242 __setup("console=", console_setup);
2245 * add_preferred_console - add a device to the list of preferred consoles.
2246 * @name: device name
2247 * @idx: device index
2248 * @options: options for this console
2250 * The last preferred console added will be used for kernel messages
2251 * and stdin/out/err for init. Normally this is used by console_setup
2252 * above to handle user-supplied console arguments; however it can also
2253 * be used by arch-specific code either to override the user or more
2254 * commonly to provide a default console (ie from PROM variables) when
2255 * the user has not supplied one.
2257 int add_preferred_console(char *name, int idx, char *options)
2259 return __add_preferred_console(name, idx, options, NULL, false);
2262 bool console_suspend_enabled = true;
2263 EXPORT_SYMBOL(console_suspend_enabled);
2265 static int __init console_suspend_disable(char *str)
2267 console_suspend_enabled = false;
2270 __setup("no_console_suspend", console_suspend_disable);
2271 module_param_named(console_suspend, console_suspend_enabled,
2272 bool, S_IRUGO | S_IWUSR);
2273 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2274 " and hibernate operations");
2277 * suspend_console - suspend the console subsystem
2279 * This disables printk() while we go into suspend states
2281 void suspend_console(void)
2283 if (!console_suspend_enabled)
2285 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2287 console_suspended = 1;
2291 void resume_console(void)
2293 if (!console_suspend_enabled)
2296 console_suspended = 0;
2301 * console_cpu_notify - print deferred console messages after CPU hotplug
2304 * If printk() is called from a CPU that is not online yet, the messages
2305 * will be printed on the console only if there are CON_ANYTIME consoles.
2306 * This function is called when a new CPU comes online (or fails to come
2307 * up) or goes offline.
2309 static int console_cpu_notify(unsigned int cpu)
2311 if (!cpuhp_tasks_frozen) {
2312 /* If trylock fails, someone else is doing the printing */
2313 if (console_trylock())
2320 * console_lock - lock the console system for exclusive use.
2322 * Acquires a lock which guarantees that the caller has
2323 * exclusive access to the console system and the console_drivers list.
2325 * Can sleep, returns nothing.
2327 void console_lock(void)
2332 if (console_suspended)
2335 console_may_schedule = 1;
2337 EXPORT_SYMBOL(console_lock);
2340 * console_trylock - try to lock the console system for exclusive use.
2342 * Try to acquire a lock which guarantees that the caller has exclusive
2343 * access to the console system and the console_drivers list.
2345 * returns 1 on success, and 0 on failure to acquire the lock.
2347 int console_trylock(void)
2349 if (down_trylock_console_sem())
2351 if (console_suspended) {
2356 console_may_schedule = 0;
2359 EXPORT_SYMBOL(console_trylock);
2361 int is_console_locked(void)
2363 return console_locked;
2365 EXPORT_SYMBOL(is_console_locked);
2368 * Check if we have any console that is capable of printing while cpu is
2369 * booting or shutting down. Requires console_sem.
2371 static int have_callable_console(void)
2373 struct console *con;
2375 for_each_console(con)
2376 if ((con->flags & CON_ENABLED) &&
2377 (con->flags & CON_ANYTIME))
2384 * Can we actually use the console at this time on this cpu?
2386 * Console drivers may assume that per-cpu resources have been allocated. So
2387 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2388 * call them until this CPU is officially up.
2390 static inline int can_use_console(void)
2392 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2396 * console_unlock - unlock the console system
2398 * Releases the console_lock which the caller holds on the console system
2399 * and the console driver list.
2401 * While the console_lock was held, console output may have been buffered
2402 * by printk(). If this is the case, console_unlock(); emits
2403 * the output prior to releasing the lock.
2405 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2407 * console_unlock(); may be called from any context.
2409 void console_unlock(void)
2411 static char ext_text[CONSOLE_EXT_LOG_MAX];
2412 static char text[LOG_LINE_MAX + PREFIX_MAX];
2413 unsigned long flags;
2414 bool do_cond_resched, retry;
2416 if (console_suspended) {
2422 * Console drivers are called with interrupts disabled, so
2423 * @console_may_schedule should be cleared before; however, we may
2424 * end up dumping a lot of lines, for example, if called from
2425 * console registration path, and should invoke cond_resched()
2426 * between lines if allowable. Not doing so can cause a very long
2427 * scheduling stall on a slow console leading to RCU stall and
2428 * softlockup warnings which exacerbate the issue with more
2429 * messages practically incapacitating the system.
2431 * console_trylock() is not able to detect the preemptive
2432 * context reliably. Therefore the value must be stored before
2433 * and cleared after the the "again" goto label.
2435 do_cond_resched = console_may_schedule;
2437 console_may_schedule = 0;
2440 * We released the console_sem lock, so we need to recheck if
2441 * cpu is online and (if not) is there at least one CON_ANYTIME
2444 if (!can_use_console()) {
2451 struct printk_log *msg;
2455 printk_safe_enter_irqsave(flags);
2456 raw_spin_lock(&logbuf_lock);
2457 if (console_seq < log_first_seq) {
2458 len = snprintf(text, sizeof(text),
2459 "** %llu printk messages dropped **\n",
2460 log_first_seq - console_seq);
2462 /* messages are gone, move to first one */
2463 console_seq = log_first_seq;
2464 console_idx = log_first_idx;
2469 if (console_seq == log_next_seq)
2472 msg = log_from_idx(console_idx);
2473 if (suppress_message_printing(msg->level)) {
2475 * Skip record we have buffered and already printed
2476 * directly to the console when we received it, and
2477 * record that has level above the console loglevel.
2479 console_idx = log_next(console_idx);
2484 /* Output to all consoles once old messages replayed. */
2485 if (unlikely(exclusive_console &&
2486 console_seq >= exclusive_console_stop_seq)) {
2487 exclusive_console = NULL;
2490 len += msg_print_text(msg,
2491 console_msg_format & MSG_FORMAT_SYSLOG,
2492 printk_time, text + len, sizeof(text) - len);
2493 if (nr_ext_console_drivers) {
2494 ext_len = msg_print_ext_header(ext_text,
2497 ext_len += msg_print_ext_body(ext_text + ext_len,
2498 sizeof(ext_text) - ext_len,
2499 log_dict(msg), msg->dict_len,
2500 log_text(msg), msg->text_len);
2502 console_idx = log_next(console_idx);
2504 raw_spin_unlock(&logbuf_lock);
2507 * While actively printing out messages, if another printk()
2508 * were to occur on another CPU, it may wait for this one to
2509 * finish. This task can not be preempted if there is a
2510 * waiter waiting to take over.
2512 console_lock_spinning_enable();
2514 stop_critical_timings(); /* don't trace print latency */
2515 call_console_drivers(ext_text, ext_len, text, len);
2516 start_critical_timings();
2518 if (console_lock_spinning_disable_and_check()) {
2519 printk_safe_exit_irqrestore(flags);
2523 printk_safe_exit_irqrestore(flags);
2525 if (do_cond_resched)
2531 raw_spin_unlock(&logbuf_lock);
2536 * Someone could have filled up the buffer again, so re-check if there's
2537 * something to flush. In case we cannot trylock the console_sem again,
2538 * there's a new owner and the console_unlock() from them will do the
2539 * flush, no worries.
2541 raw_spin_lock(&logbuf_lock);
2542 retry = console_seq != log_next_seq;
2543 raw_spin_unlock(&logbuf_lock);
2544 printk_safe_exit_irqrestore(flags);
2546 if (retry && console_trylock())
2549 EXPORT_SYMBOL(console_unlock);
2552 * console_conditional_schedule - yield the CPU if required
2554 * If the console code is currently allowed to sleep, and
2555 * if this CPU should yield the CPU to another task, do
2558 * Must be called within console_lock();.
2560 void __sched console_conditional_schedule(void)
2562 if (console_may_schedule)
2565 EXPORT_SYMBOL(console_conditional_schedule);
2567 void console_unblank(void)
2572 * console_unblank can no longer be called in interrupt context unless
2573 * oops_in_progress is set to 1..
2575 if (oops_in_progress) {
2576 if (down_trylock_console_sem() != 0)
2582 console_may_schedule = 0;
2584 if ((c->flags & CON_ENABLED) && c->unblank)
2590 * console_flush_on_panic - flush console content on panic
2591 * @mode: flush all messages in buffer or just the pending ones
2593 * Immediately output all pending messages no matter what.
2595 void console_flush_on_panic(enum con_flush_mode mode)
2598 * If someone else is holding the console lock, trylock will fail
2599 * and may_schedule may be set. Ignore and proceed to unlock so
2600 * that messages are flushed out. As this can be called from any
2601 * context and we don't want to get preempted while flushing,
2602 * ensure may_schedule is cleared.
2605 console_may_schedule = 0;
2607 if (mode == CONSOLE_REPLAY_ALL) {
2608 unsigned long flags;
2610 logbuf_lock_irqsave(flags);
2611 console_seq = log_first_seq;
2612 console_idx = log_first_idx;
2613 logbuf_unlock_irqrestore(flags);
2619 * Return the console tty driver structure and its associated index
2621 struct tty_driver *console_device(int *index)
2624 struct tty_driver *driver = NULL;
2627 for_each_console(c) {
2630 driver = c->device(c, index);
2639 * Prevent further output on the passed console device so that (for example)
2640 * serial drivers can disable console output before suspending a port, and can
2641 * re-enable output afterwards.
2643 void console_stop(struct console *console)
2646 console->flags &= ~CON_ENABLED;
2649 EXPORT_SYMBOL(console_stop);
2651 void console_start(struct console *console)
2654 console->flags |= CON_ENABLED;
2657 EXPORT_SYMBOL(console_start);
2659 static int __read_mostly keep_bootcon;
2661 static int __init keep_bootcon_setup(char *str)
2664 pr_info("debug: skip boot console de-registration.\n");
2669 early_param("keep_bootcon", keep_bootcon_setup);
2672 * This is called by register_console() to try to match
2673 * the newly registered console with any of the ones selected
2674 * by either the command line or add_preferred_console() and
2677 * Care need to be taken with consoles that are statically
2678 * enabled such as netconsole
2680 static int try_enable_new_console(struct console *newcon, bool user_specified)
2682 struct console_cmdline *c;
2685 for (i = 0, c = console_cmdline;
2686 i < MAX_CMDLINECONSOLES && c->name[0];
2688 if (c->user_specified != user_specified)
2690 if (!newcon->match ||
2691 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2692 /* default matching */
2693 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2694 if (strcmp(c->name, newcon->name) != 0)
2696 if (newcon->index >= 0 &&
2697 newcon->index != c->index)
2699 if (newcon->index < 0)
2700 newcon->index = c->index;
2702 if (_braille_register_console(newcon, c))
2705 if (newcon->setup &&
2706 newcon->setup(newcon, c->options) != 0)
2709 newcon->flags |= CON_ENABLED;
2710 if (i == preferred_console) {
2711 newcon->flags |= CON_CONSDEV;
2712 has_preferred_console = true;
2718 * Some consoles, such as pstore and netconsole, can be enabled even
2719 * without matching. Accept the pre-enabled consoles only when match()
2720 * and setup() had a change to be called.
2722 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
2729 * The console driver calls this routine during kernel initialization
2730 * to register the console printing procedure with printk() and to
2731 * print any messages that were printed by the kernel before the
2732 * console driver was initialized.
2734 * This can happen pretty early during the boot process (because of
2735 * early_printk) - sometimes before setup_arch() completes - be careful
2736 * of what kernel features are used - they may not be initialised yet.
2738 * There are two types of consoles - bootconsoles (early_printk) and
2739 * "real" consoles (everything which is not a bootconsole) which are
2740 * handled differently.
2741 * - Any number of bootconsoles can be registered at any time.
2742 * - As soon as a "real" console is registered, all bootconsoles
2743 * will be unregistered automatically.
2744 * - Once a "real" console is registered, any attempt to register a
2745 * bootconsoles will be rejected
2747 void register_console(struct console *newcon)
2749 unsigned long flags;
2750 struct console *bcon = NULL;
2753 for_each_console(bcon) {
2754 if (WARN(bcon == newcon, "console '%s%d' already registered\n",
2755 bcon->name, bcon->index))
2760 * before we register a new CON_BOOT console, make sure we don't
2761 * already have a valid console
2763 if (newcon->flags & CON_BOOT) {
2764 for_each_console(bcon) {
2765 if (!(bcon->flags & CON_BOOT)) {
2766 pr_info("Too late to register bootconsole %s%d\n",
2767 newcon->name, newcon->index);
2773 if (console_drivers && console_drivers->flags & CON_BOOT)
2774 bcon = console_drivers;
2776 if (!has_preferred_console || bcon || !console_drivers)
2777 has_preferred_console = preferred_console >= 0;
2780 * See if we want to use this console driver. If we
2781 * didn't select a console we take the first one
2782 * that registers here.
2784 if (!has_preferred_console) {
2785 if (newcon->index < 0)
2787 if (newcon->setup == NULL ||
2788 newcon->setup(newcon, NULL) == 0) {
2789 newcon->flags |= CON_ENABLED;
2790 if (newcon->device) {
2791 newcon->flags |= CON_CONSDEV;
2792 has_preferred_console = true;
2797 /* See if this console matches one we selected on the command line */
2798 err = try_enable_new_console(newcon, true);
2800 /* If not, try to match against the platform default(s) */
2802 err = try_enable_new_console(newcon, false);
2804 /* printk() messages are not printed to the Braille console. */
2805 if (err || newcon->flags & CON_BRL)
2809 * If we have a bootconsole, and are switching to a real console,
2810 * don't print everything out again, since when the boot console, and
2811 * the real console are the same physical device, it's annoying to
2812 * see the beginning boot messages twice
2814 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2815 newcon->flags &= ~CON_PRINTBUFFER;
2818 * Put this console in the list - keep the
2819 * preferred driver at the head of the list.
2822 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2823 newcon->next = console_drivers;
2824 console_drivers = newcon;
2826 newcon->next->flags &= ~CON_CONSDEV;
2827 /* Ensure this flag is always set for the head of the list */
2828 newcon->flags |= CON_CONSDEV;
2830 newcon->next = console_drivers->next;
2831 console_drivers->next = newcon;
2834 if (newcon->flags & CON_EXTENDED)
2835 nr_ext_console_drivers++;
2837 if (newcon->flags & CON_PRINTBUFFER) {
2839 * console_unlock(); will print out the buffered messages
2842 logbuf_lock_irqsave(flags);
2844 * We're about to replay the log buffer. Only do this to the
2845 * just-registered console to avoid excessive message spam to
2846 * the already-registered consoles.
2848 * Set exclusive_console with disabled interrupts to reduce
2849 * race window with eventual console_flush_on_panic() that
2850 * ignores console_lock.
2852 exclusive_console = newcon;
2853 exclusive_console_stop_seq = console_seq;
2854 console_seq = syslog_seq;
2855 console_idx = syslog_idx;
2856 logbuf_unlock_irqrestore(flags);
2859 console_sysfs_notify();
2862 * By unregistering the bootconsoles after we enable the real console
2863 * we get the "console xxx enabled" message on all the consoles -
2864 * boot consoles, real consoles, etc - this is to ensure that end
2865 * users know there might be something in the kernel's log buffer that
2866 * went to the bootconsole (that they do not see on the real console)
2868 pr_info("%sconsole [%s%d] enabled\n",
2869 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2870 newcon->name, newcon->index);
2872 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2874 /* We need to iterate through all boot consoles, to make
2875 * sure we print everything out, before we unregister them.
2877 for_each_console(bcon)
2878 if (bcon->flags & CON_BOOT)
2879 unregister_console(bcon);
2882 EXPORT_SYMBOL(register_console);
2884 int unregister_console(struct console *console)
2886 struct console *con;
2889 pr_info("%sconsole [%s%d] disabled\n",
2890 (console->flags & CON_BOOT) ? "boot" : "" ,
2891 console->name, console->index);
2893 res = _braille_unregister_console(console);
2901 if (console_drivers == console) {
2902 console_drivers=console->next;
2905 for_each_console(con) {
2906 if (con->next == console) {
2907 con->next = console->next;
2915 goto out_disable_unlock;
2917 if (console->flags & CON_EXTENDED)
2918 nr_ext_console_drivers--;
2921 * If this isn't the last console and it has CON_CONSDEV set, we
2922 * need to set it on the next preferred console.
2924 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2925 console_drivers->flags |= CON_CONSDEV;
2927 console->flags &= ~CON_ENABLED;
2929 console_sysfs_notify();
2932 res = console->exit(console);
2937 console->flags &= ~CON_ENABLED;
2942 EXPORT_SYMBOL(unregister_console);
2945 * Initialize the console device. This is called *early*, so
2946 * we can't necessarily depend on lots of kernel help here.
2947 * Just do some early initializations, and do the complex setup
2950 void __init console_init(void)
2954 initcall_entry_t *ce;
2956 /* Setup the default TTY line discipline. */
2960 * set up the console device so that later boot sequences can
2961 * inform about problems etc..
2963 ce = __con_initcall_start;
2964 trace_initcall_level("console");
2965 while (ce < __con_initcall_end) {
2966 call = initcall_from_entry(ce);
2967 trace_initcall_start(call);
2969 trace_initcall_finish(call, ret);
2975 * Some boot consoles access data that is in the init section and which will
2976 * be discarded after the initcalls have been run. To make sure that no code
2977 * will access this data, unregister the boot consoles in a late initcall.
2979 * If for some reason, such as deferred probe or the driver being a loadable
2980 * module, the real console hasn't registered yet at this point, there will
2981 * be a brief interval in which no messages are logged to the console, which
2982 * makes it difficult to diagnose problems that occur during this time.
2984 * To mitigate this problem somewhat, only unregister consoles whose memory
2985 * intersects with the init section. Note that all other boot consoles will
2986 * get unregistred when the real preferred console is registered.
2988 static int __init printk_late_init(void)
2990 struct console *con;
2993 for_each_console(con) {
2994 if (!(con->flags & CON_BOOT))
2997 /* Check addresses that might be used for enabled consoles. */
2998 if (init_section_intersects(con, sizeof(*con)) ||
2999 init_section_contains(con->write, 0) ||
3000 init_section_contains(con->read, 0) ||
3001 init_section_contains(con->device, 0) ||
3002 init_section_contains(con->unblank, 0) ||
3003 init_section_contains(con->data, 0)) {
3005 * Please, consider moving the reported consoles out
3006 * of the init section.
3008 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3009 con->name, con->index);
3010 unregister_console(con);
3013 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3014 console_cpu_notify);
3016 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3017 console_cpu_notify, NULL);
3021 late_initcall(printk_late_init);
3023 #if defined CONFIG_PRINTK
3025 * Delayed printk version, for scheduler-internal messages:
3027 #define PRINTK_PENDING_WAKEUP 0x01
3028 #define PRINTK_PENDING_OUTPUT 0x02
3030 static DEFINE_PER_CPU(int, printk_pending);
3032 static void wake_up_klogd_work_func(struct irq_work *irq_work)
3034 int pending = __this_cpu_xchg(printk_pending, 0);
3036 if (pending & PRINTK_PENDING_OUTPUT) {
3037 /* If trylock fails, someone else is doing the printing */
3038 if (console_trylock())
3042 if (pending & PRINTK_PENDING_WAKEUP)
3043 wake_up_interruptible(&log_wait);
3046 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
3047 .func = wake_up_klogd_work_func,
3048 .flags = ATOMIC_INIT(IRQ_WORK_LAZY),
3051 void wake_up_klogd(void)
3053 if (!printk_percpu_data_ready())
3057 if (waitqueue_active(&log_wait)) {
3058 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
3059 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3064 void defer_console_output(void)
3066 if (!printk_percpu_data_ready())
3070 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
3071 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3075 int vprintk_deferred(const char *fmt, va_list args)
3079 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
3080 defer_console_output();
3085 int printk_deferred(const char *fmt, ...)
3090 va_start(args, fmt);
3091 r = vprintk_deferred(fmt, args);
3098 * printk rate limiting, lifted from the networking subsystem.
3100 * This enforces a rate limit: not more than 10 kernel messages
3101 * every 5s to make a denial-of-service attack impossible.
3103 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3105 int __printk_ratelimit(const char *func)
3107 return ___ratelimit(&printk_ratelimit_state, func);
3109 EXPORT_SYMBOL(__printk_ratelimit);
3112 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3113 * @caller_jiffies: pointer to caller's state
3114 * @interval_msecs: minimum interval between prints
3116 * printk_timed_ratelimit() returns true if more than @interval_msecs
3117 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3120 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3121 unsigned int interval_msecs)
3123 unsigned long elapsed = jiffies - *caller_jiffies;
3125 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3128 *caller_jiffies = jiffies;
3131 EXPORT_SYMBOL(printk_timed_ratelimit);
3133 static DEFINE_SPINLOCK(dump_list_lock);
3134 static LIST_HEAD(dump_list);
3137 * kmsg_dump_register - register a kernel log dumper.
3138 * @dumper: pointer to the kmsg_dumper structure
3140 * Adds a kernel log dumper to the system. The dump callback in the
3141 * structure will be called when the kernel oopses or panics and must be
3142 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3144 int kmsg_dump_register(struct kmsg_dumper *dumper)
3146 unsigned long flags;
3149 /* The dump callback needs to be set */
3153 spin_lock_irqsave(&dump_list_lock, flags);
3154 /* Don't allow registering multiple times */
3155 if (!dumper->registered) {
3156 dumper->registered = 1;
3157 list_add_tail_rcu(&dumper->list, &dump_list);
3160 spin_unlock_irqrestore(&dump_list_lock, flags);
3164 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3167 * kmsg_dump_unregister - unregister a kmsg dumper.
3168 * @dumper: pointer to the kmsg_dumper structure
3170 * Removes a dump device from the system. Returns zero on success and
3171 * %-EINVAL otherwise.
3173 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3175 unsigned long flags;
3178 spin_lock_irqsave(&dump_list_lock, flags);
3179 if (dumper->registered) {
3180 dumper->registered = 0;
3181 list_del_rcu(&dumper->list);
3184 spin_unlock_irqrestore(&dump_list_lock, flags);
3189 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3191 static bool always_kmsg_dump;
3192 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3194 const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
3197 case KMSG_DUMP_PANIC:
3199 case KMSG_DUMP_OOPS:
3201 case KMSG_DUMP_EMERG:
3203 case KMSG_DUMP_SHUTDOWN:
3209 EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
3212 * kmsg_dump - dump kernel log to kernel message dumpers.
3213 * @reason: the reason (oops, panic etc) for dumping
3215 * Call each of the registered dumper's dump() callback, which can
3216 * retrieve the kmsg records with kmsg_dump_get_line() or
3217 * kmsg_dump_get_buffer().
3219 void kmsg_dump(enum kmsg_dump_reason reason)
3221 struct kmsg_dumper *dumper;
3222 unsigned long flags;
3225 list_for_each_entry_rcu(dumper, &dump_list, list) {
3226 enum kmsg_dump_reason max_reason = dumper->max_reason;
3229 * If client has not provided a specific max_reason, default
3230 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
3232 if (max_reason == KMSG_DUMP_UNDEF) {
3233 max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
3236 if (reason > max_reason)
3239 /* initialize iterator with data about the stored records */
3240 dumper->active = true;
3242 logbuf_lock_irqsave(flags);
3243 dumper->cur_seq = clear_seq;
3244 dumper->cur_idx = clear_idx;
3245 dumper->next_seq = log_next_seq;
3246 dumper->next_idx = log_next_idx;
3247 logbuf_unlock_irqrestore(flags);
3249 /* invoke dumper which will iterate over records */
3250 dumper->dump(dumper, reason);
3252 /* reset iterator */
3253 dumper->active = false;
3259 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3260 * @dumper: registered kmsg dumper
3261 * @syslog: include the "<4>" prefixes
3262 * @line: buffer to copy the line to
3263 * @size: maximum size of the buffer
3264 * @len: length of line placed into buffer
3266 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3267 * record, and copy one record into the provided buffer.
3269 * Consecutive calls will return the next available record moving
3270 * towards the end of the buffer with the youngest messages.
3272 * A return value of FALSE indicates that there are no more records to
3275 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3277 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3278 char *line, size_t size, size_t *len)
3280 struct printk_log *msg;
3284 if (!dumper->active)
3287 if (dumper->cur_seq < log_first_seq) {
3288 /* messages are gone, move to first available one */
3289 dumper->cur_seq = log_first_seq;
3290 dumper->cur_idx = log_first_idx;
3294 if (dumper->cur_seq >= log_next_seq)
3297 msg = log_from_idx(dumper->cur_idx);
3298 l = msg_print_text(msg, syslog, printk_time, line, size);
3300 dumper->cur_idx = log_next(dumper->cur_idx);
3310 * kmsg_dump_get_line - retrieve one kmsg log line
3311 * @dumper: registered kmsg dumper
3312 * @syslog: include the "<4>" prefixes
3313 * @line: buffer to copy the line to
3314 * @size: maximum size of the buffer
3315 * @len: length of line placed into buffer
3317 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3318 * record, and copy one record into the provided buffer.
3320 * Consecutive calls will return the next available record moving
3321 * towards the end of the buffer with the youngest messages.
3323 * A return value of FALSE indicates that there are no more records to
3326 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3327 char *line, size_t size, size_t *len)
3329 unsigned long flags;
3332 logbuf_lock_irqsave(flags);
3333 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3334 logbuf_unlock_irqrestore(flags);
3338 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3341 * kmsg_dump_get_buffer - copy kmsg log lines
3342 * @dumper: registered kmsg dumper
3343 * @syslog: include the "<4>" prefixes
3344 * @buf: buffer to copy the line to
3345 * @size: maximum size of the buffer
3346 * @len: length of line placed into buffer
3348 * Start at the end of the kmsg buffer and fill the provided buffer
3349 * with as many of the the *youngest* kmsg records that fit into it.
3350 * If the buffer is large enough, all available kmsg records will be
3351 * copied with a single call.
3353 * Consecutive calls will fill the buffer with the next block of
3354 * available older records, not including the earlier retrieved ones.
3356 * A return value of FALSE indicates that there are no more records to
3359 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3360 char *buf, size_t size, size_t *len)
3362 unsigned long flags;
3369 bool time = printk_time;
3371 if (!dumper->active)
3374 logbuf_lock_irqsave(flags);
3375 if (dumper->cur_seq < log_first_seq) {
3376 /* messages are gone, move to first available one */
3377 dumper->cur_seq = log_first_seq;
3378 dumper->cur_idx = log_first_idx;
3382 if (dumper->cur_seq >= dumper->next_seq) {
3383 logbuf_unlock_irqrestore(flags);
3387 /* calculate length of entire buffer */
3388 seq = dumper->cur_seq;
3389 idx = dumper->cur_idx;
3390 while (seq < dumper->next_seq) {
3391 struct printk_log *msg = log_from_idx(idx);
3393 l += msg_print_text(msg, true, time, NULL, 0);
3394 idx = log_next(idx);
3398 /* move first record forward until length fits into the buffer */
3399 seq = dumper->cur_seq;
3400 idx = dumper->cur_idx;
3401 while (l >= size && seq < dumper->next_seq) {
3402 struct printk_log *msg = log_from_idx(idx);
3404 l -= msg_print_text(msg, true, time, NULL, 0);
3405 idx = log_next(idx);
3409 /* last message in next interation */
3414 while (seq < dumper->next_seq) {
3415 struct printk_log *msg = log_from_idx(idx);
3417 l += msg_print_text(msg, syslog, time, buf + l, size - l);
3418 idx = log_next(idx);
3422 dumper->next_seq = next_seq;
3423 dumper->next_idx = next_idx;
3425 logbuf_unlock_irqrestore(flags);
3431 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3434 * kmsg_dump_rewind_nolock - reset the iterator (unlocked version)
3435 * @dumper: registered kmsg dumper
3437 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3438 * kmsg_dump_get_buffer() can be called again and used multiple
3439 * times within the same dumper.dump() callback.
3441 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3443 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3445 dumper->cur_seq = clear_seq;
3446 dumper->cur_idx = clear_idx;
3447 dumper->next_seq = log_next_seq;
3448 dumper->next_idx = log_next_idx;
3452 * kmsg_dump_rewind - reset the iterator
3453 * @dumper: registered kmsg dumper
3455 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3456 * kmsg_dump_get_buffer() can be called again and used multiple
3457 * times within the same dumper.dump() callback.
3459 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3461 unsigned long flags;
3463 logbuf_lock_irqsave(flags);
3464 kmsg_dump_rewind_nolock(dumper);
3465 logbuf_unlock_irqrestore(flags);
3467 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);