2 * linux/kernel/printk.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * Modified to make sys_syslog() more flexible: added commands to
7 * return the last 4k of kernel messages, regardless of whether
8 * they've been read or not. Added option to suppress kernel printk's
9 * to the console. Added hook for sending the console messages
10 * elsewhere, in preparation for a serial line console (someday).
12 * Modified for sysctl support, 1/8/97, Chris Horn.
13 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
14 * manfred@colorfullife.com
15 * Rewrote bits to get rid of console_lock
16 * 01Mar01 Andrew Morton
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/kernel.h>
23 #include <linux/tty.h>
24 #include <linux/tty_driver.h>
25 #include <linux/console.h>
26 #include <linux/init.h>
27 #include <linux/jiffies.h>
28 #include <linux/nmi.h>
29 #include <linux/module.h>
30 #include <linux/moduleparam.h>
31 #include <linux/delay.h>
32 #include <linux/smp.h>
33 #include <linux/security.h>
34 #include <linux/memblock.h>
35 #include <linux/syscalls.h>
36 #include <linux/crash_core.h>
37 #include <linux/kdb.h>
38 #include <linux/ratelimit.h>
39 #include <linux/kmsg_dump.h>
40 #include <linux/syslog.h>
41 #include <linux/cpu.h>
42 #include <linux/rculist.h>
43 #include <linux/poll.h>
44 #include <linux/irq_work.h>
45 #include <linux/ctype.h>
46 #include <linux/uio.h>
47 #include <linux/sched/clock.h>
48 #include <linux/sched/debug.h>
49 #include <linux/sched/task_stack.h>
51 #include <linux/uaccess.h>
52 #include <asm/sections.h>
54 #include <trace/events/initcall.h>
55 #define CREATE_TRACE_POINTS
56 #include <trace/events/printk.h>
58 #include "console_cmdline.h"
62 int console_printk[4] = {
63 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
64 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
65 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
66 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
69 atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
70 EXPORT_SYMBOL(ignore_console_lock_warning);
73 * Low level drivers may need that to know if they can schedule in
74 * their unblank() callback or not. So let's export it.
77 EXPORT_SYMBOL(oops_in_progress);
80 * console_sem protects the console_drivers list, and also
81 * provides serialisation for access to the entire console
84 static DEFINE_SEMAPHORE(console_sem);
85 struct console *console_drivers;
86 EXPORT_SYMBOL_GPL(console_drivers);
89 static struct lockdep_map console_lock_dep_map = {
90 .name = "console_lock"
94 enum devkmsg_log_bits {
95 __DEVKMSG_LOG_BIT_ON = 0,
96 __DEVKMSG_LOG_BIT_OFF,
97 __DEVKMSG_LOG_BIT_LOCK,
100 enum devkmsg_log_masks {
101 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
102 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
103 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
106 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
107 #define DEVKMSG_LOG_MASK_DEFAULT 0
109 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
111 static int __control_devkmsg(char *str)
116 if (!strncmp(str, "on", 2)) {
117 devkmsg_log = DEVKMSG_LOG_MASK_ON;
119 } else if (!strncmp(str, "off", 3)) {
120 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
122 } else if (!strncmp(str, "ratelimit", 9)) {
123 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
129 static int __init control_devkmsg(char *str)
131 if (__control_devkmsg(str) < 0)
135 * Set sysctl string accordingly:
137 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
138 strcpy(devkmsg_log_str, "on");
139 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
140 strcpy(devkmsg_log_str, "off");
141 /* else "ratelimit" which is set by default. */
144 * Sysctl cannot change it anymore. The kernel command line setting of
145 * this parameter is to force the setting to be permanent throughout the
146 * runtime of the system. This is a precation measure against userspace
147 * trying to be a smarta** and attempting to change it up on us.
149 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
153 __setup("printk.devkmsg=", control_devkmsg);
155 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
157 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
158 void __user *buffer, size_t *lenp, loff_t *ppos)
160 char old_str[DEVKMSG_STR_MAX_SIZE];
165 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
169 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
172 err = proc_dostring(table, write, buffer, lenp, ppos);
177 err = __control_devkmsg(devkmsg_log_str);
180 * Do not accept an unknown string OR a known string with
183 if (err < 0 || (err + 1 != *lenp)) {
185 /* ... and restore old setting. */
187 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
196 /* Number of registered extended console drivers. */
197 static int nr_ext_console_drivers;
200 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
201 * macros instead of functions so that _RET_IP_ contains useful information.
203 #define down_console_sem() do { \
205 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
208 static int __down_trylock_console_sem(unsigned long ip)
214 * Here and in __up_console_sem() we need to be in safe mode,
215 * because spindump/WARN/etc from under console ->lock will
216 * deadlock in printk()->down_trylock_console_sem() otherwise.
218 printk_safe_enter_irqsave(flags);
219 lock_failed = down_trylock(&console_sem);
220 printk_safe_exit_irqrestore(flags);
224 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
227 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
229 static void __up_console_sem(unsigned long ip)
233 mutex_release(&console_lock_dep_map, 1, ip);
235 printk_safe_enter_irqsave(flags);
237 printk_safe_exit_irqrestore(flags);
239 #define up_console_sem() __up_console_sem(_RET_IP_)
242 * This is used for debugging the mess that is the VT code by
243 * keeping track if we have the console semaphore held. It's
244 * definitely not the perfect debug tool (we don't know if _WE_
245 * hold it and are racing, but it helps tracking those weird code
246 * paths in the console code where we end up in places I want
247 * locked without the console sempahore held).
249 static int console_locked, console_suspended;
252 * If exclusive_console is non-NULL then only this console is to be printed to.
254 static struct console *exclusive_console;
257 * Array of consoles built from command line options (console=)
260 #define MAX_CMDLINECONSOLES 8
262 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
264 static int preferred_console = -1;
265 int console_set_on_cmdline;
266 EXPORT_SYMBOL(console_set_on_cmdline);
268 /* Flag: console code may call schedule() */
269 static int console_may_schedule;
271 enum con_msg_format_flags {
272 MSG_FORMAT_DEFAULT = 0,
273 MSG_FORMAT_SYSLOG = (1 << 0),
276 static int console_msg_format = MSG_FORMAT_DEFAULT;
279 * The printk log buffer consists of a chain of concatenated variable
280 * length records. Every record starts with a record header, containing
281 * the overall length of the record.
283 * The heads to the first and last entry in the buffer, as well as the
284 * sequence numbers of these entries are maintained when messages are
287 * If the heads indicate available messages, the length in the header
288 * tells the start next message. A length == 0 for the next message
289 * indicates a wrap-around to the beginning of the buffer.
291 * Every record carries the monotonic timestamp in microseconds, as well as
292 * the standard userspace syslog level and syslog facility. The usual
293 * kernel messages use LOG_KERN; userspace-injected messages always carry
294 * a matching syslog facility, by default LOG_USER. The origin of every
295 * message can be reliably determined that way.
297 * The human readable log message directly follows the message header. The
298 * length of the message text is stored in the header, the stored message
301 * Optionally, a message can carry a dictionary of properties (key/value pairs),
302 * to provide userspace with a machine-readable message context.
304 * Examples for well-defined, commonly used property names are:
305 * DEVICE=b12:8 device identifier
309 * +sound:card0 subsystem:devname
310 * SUBSYSTEM=pci driver-core subsystem name
312 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
313 * follows directly after a '=' character. Every property is terminated by
314 * a '\0' character. The last property is not terminated.
316 * Example of a message structure:
317 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
318 * 0008 34 00 record is 52 bytes long
319 * 000a 0b 00 text is 11 bytes long
320 * 000c 1f 00 dictionary is 23 bytes long
321 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
322 * 0010 69 74 27 73 20 61 20 6c "it's a l"
324 * 001b 44 45 56 49 43 "DEVIC"
325 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
326 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
328 * 0032 00 00 00 padding to next message header
330 * The 'struct printk_log' buffer header must never be directly exported to
331 * userspace, it is a kernel-private implementation detail that might
332 * need to be changed in the future, when the requirements change.
334 * /dev/kmsg exports the structured data in the following line format:
335 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
337 * Users of the export format should ignore possible additional values
338 * separated by ',', and find the message after the ';' character.
340 * The optional key/value pairs are attached as continuation lines starting
341 * with a space character and terminated by a newline. All possible
342 * non-prinatable characters are escaped in the "\xff" notation.
346 LOG_NEWLINE = 2, /* text ended with a newline */
347 LOG_CONT = 8, /* text is a fragment of a continuation line */
351 u64 ts_nsec; /* timestamp in nanoseconds */
352 u16 len; /* length of entire record */
353 u16 text_len; /* length of text buffer */
354 u16 dict_len; /* length of dictionary buffer */
355 u8 facility; /* syslog facility */
356 u8 flags:5; /* internal record flags */
357 u8 level:3; /* syslog level */
358 #ifdef CONFIG_PRINTK_CALLER
359 u32 caller_id; /* thread id or processor id */
362 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
363 __packed __aligned(4)
368 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
369 * within the scheduler's rq lock. It must be released before calling
370 * console_unlock() or anything else that might wake up a process.
372 DEFINE_RAW_SPINLOCK(logbuf_lock);
375 * Helper macros to lock/unlock logbuf_lock and switch between
376 * printk-safe/unsafe modes.
378 #define logbuf_lock_irq() \
380 printk_safe_enter_irq(); \
381 raw_spin_lock(&logbuf_lock); \
384 #define logbuf_unlock_irq() \
386 raw_spin_unlock(&logbuf_lock); \
387 printk_safe_exit_irq(); \
390 #define logbuf_lock_irqsave(flags) \
392 printk_safe_enter_irqsave(flags); \
393 raw_spin_lock(&logbuf_lock); \
396 #define logbuf_unlock_irqrestore(flags) \
398 raw_spin_unlock(&logbuf_lock); \
399 printk_safe_exit_irqrestore(flags); \
403 DECLARE_WAIT_QUEUE_HEAD(log_wait);
404 /* the next printk record to read by syslog(READ) or /proc/kmsg */
405 static u64 syslog_seq;
406 static u32 syslog_idx;
407 static size_t syslog_partial;
408 static bool syslog_time;
410 /* index and sequence number of the first record stored in the buffer */
411 static u64 log_first_seq;
412 static u32 log_first_idx;
414 /* index and sequence number of the next record to store in the buffer */
415 static u64 log_next_seq;
416 static u32 log_next_idx;
418 /* the next printk record to write to the console */
419 static u64 console_seq;
420 static u32 console_idx;
421 static u64 exclusive_console_stop_seq;
423 /* the next printk record to read after the last 'clear' command */
424 static u64 clear_seq;
425 static u32 clear_idx;
427 #ifdef CONFIG_PRINTK_CALLER
428 #define PREFIX_MAX 48
430 #define PREFIX_MAX 32
432 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
434 #define LOG_LEVEL(v) ((v) & 0x07)
435 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
438 #define LOG_ALIGN __alignof__(struct printk_log)
439 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
440 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
441 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
442 static char *log_buf = __log_buf;
443 static u32 log_buf_len = __LOG_BUF_LEN;
445 /* Return log buffer address */
446 char *log_buf_addr_get(void)
451 /* Return log buffer size */
452 u32 log_buf_len_get(void)
457 /* human readable text of the record */
458 static char *log_text(const struct printk_log *msg)
460 return (char *)msg + sizeof(struct printk_log);
463 /* optional key/value pair dictionary attached to the record */
464 static char *log_dict(const struct printk_log *msg)
466 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
469 /* get record by index; idx must point to valid msg */
470 static struct printk_log *log_from_idx(u32 idx)
472 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
475 * A length == 0 record is the end of buffer marker. Wrap around and
476 * read the message at the start of the buffer.
479 return (struct printk_log *)log_buf;
483 /* get next record; idx must point to valid msg */
484 static u32 log_next(u32 idx)
486 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
488 /* length == 0 indicates the end of the buffer; wrap */
490 * A length == 0 record is the end of buffer marker. Wrap around and
491 * read the message at the start of the buffer as *this* one, and
492 * return the one after that.
495 msg = (struct printk_log *)log_buf;
498 return idx + msg->len;
502 * Check whether there is enough free space for the given message.
504 * The same values of first_idx and next_idx mean that the buffer
505 * is either empty or full.
507 * If the buffer is empty, we must respect the position of the indexes.
508 * They cannot be reset to the beginning of the buffer.
510 static int logbuf_has_space(u32 msg_size, bool empty)
514 if (log_next_idx > log_first_idx || empty)
515 free = max(log_buf_len - log_next_idx, log_first_idx);
517 free = log_first_idx - log_next_idx;
520 * We need space also for an empty header that signalizes wrapping
523 return free >= msg_size + sizeof(struct printk_log);
526 static int log_make_free_space(u32 msg_size)
528 while (log_first_seq < log_next_seq &&
529 !logbuf_has_space(msg_size, false)) {
530 /* drop old messages until we have enough contiguous space */
531 log_first_idx = log_next(log_first_idx);
535 if (clear_seq < log_first_seq) {
536 clear_seq = log_first_seq;
537 clear_idx = log_first_idx;
540 /* sequence numbers are equal, so the log buffer is empty */
541 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
547 /* compute the message size including the padding bytes */
548 static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
552 size = sizeof(struct printk_log) + text_len + dict_len;
553 *pad_len = (-size) & (LOG_ALIGN - 1);
560 * Define how much of the log buffer we could take at maximum. The value
561 * must be greater than two. Note that only half of the buffer is available
562 * when the index points to the middle.
564 #define MAX_LOG_TAKE_PART 4
565 static const char trunc_msg[] = "<truncated>";
567 static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
568 u16 *dict_len, u32 *pad_len)
571 * The message should not take the whole buffer. Otherwise, it might
572 * get removed too soon.
574 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
575 if (*text_len > max_text_len)
576 *text_len = max_text_len;
577 /* enable the warning message */
578 *trunc_msg_len = strlen(trunc_msg);
579 /* disable the "dict" completely */
581 /* compute the size again, count also the warning message */
582 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
585 /* insert record into the buffer, discard old ones, update heads */
586 static int log_store(u32 caller_id, int facility, int level,
587 enum log_flags flags, u64 ts_nsec,
588 const char *dict, u16 dict_len,
589 const char *text, u16 text_len)
591 struct printk_log *msg;
593 u16 trunc_msg_len = 0;
595 /* number of '\0' padding bytes to next message */
596 size = msg_used_size(text_len, dict_len, &pad_len);
598 if (log_make_free_space(size)) {
599 /* truncate the message if it is too long for empty buffer */
600 size = truncate_msg(&text_len, &trunc_msg_len,
601 &dict_len, &pad_len);
602 /* survive when the log buffer is too small for trunc_msg */
603 if (log_make_free_space(size))
607 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
609 * This message + an additional empty header does not fit
610 * at the end of the buffer. Add an empty header with len == 0
611 * to signify a wrap around.
613 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
618 msg = (struct printk_log *)(log_buf + log_next_idx);
619 memcpy(log_text(msg), text, text_len);
620 msg->text_len = text_len;
622 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
623 msg->text_len += trunc_msg_len;
625 memcpy(log_dict(msg), dict, dict_len);
626 msg->dict_len = dict_len;
627 msg->facility = facility;
628 msg->level = level & 7;
629 msg->flags = flags & 0x1f;
631 msg->ts_nsec = ts_nsec;
633 msg->ts_nsec = local_clock();
634 #ifdef CONFIG_PRINTK_CALLER
635 msg->caller_id = caller_id;
637 memset(log_dict(msg) + dict_len, 0, pad_len);
641 log_next_idx += msg->len;
644 return msg->text_len;
647 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
649 static int syslog_action_restricted(int type)
654 * Unless restricted, we allow "read all" and "get buffer size"
657 return type != SYSLOG_ACTION_READ_ALL &&
658 type != SYSLOG_ACTION_SIZE_BUFFER;
661 static int check_syslog_permissions(int type, int source)
664 * If this is from /proc/kmsg and we've already opened it, then we've
665 * already done the capabilities checks at open time.
667 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
670 if (syslog_action_restricted(type)) {
671 if (capable(CAP_SYSLOG))
674 * For historical reasons, accept CAP_SYS_ADMIN too, with
677 if (capable(CAP_SYS_ADMIN)) {
678 pr_warn_once("%s (%d): Attempt to access syslog with "
679 "CAP_SYS_ADMIN but no CAP_SYSLOG "
681 current->comm, task_pid_nr(current));
687 return security_syslog(type);
690 static void append_char(char **pp, char *e, char c)
696 static ssize_t msg_print_ext_header(char *buf, size_t size,
697 struct printk_log *msg, u64 seq)
699 u64 ts_usec = msg->ts_nsec;
701 #ifdef CONFIG_PRINTK_CALLER
702 u32 id = msg->caller_id;
704 snprintf(caller, sizeof(caller), ",caller=%c%u",
705 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
710 do_div(ts_usec, 1000);
712 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
713 (msg->facility << 3) | msg->level, seq, ts_usec,
714 msg->flags & LOG_CONT ? 'c' : '-', caller);
717 static ssize_t msg_print_ext_body(char *buf, size_t size,
718 char *dict, size_t dict_len,
719 char *text, size_t text_len)
721 char *p = buf, *e = buf + size;
724 /* escape non-printable characters */
725 for (i = 0; i < text_len; i++) {
726 unsigned char c = text[i];
728 if (c < ' ' || c >= 127 || c == '\\')
729 p += scnprintf(p, e - p, "\\x%02x", c);
731 append_char(&p, e, c);
733 append_char(&p, e, '\n');
738 for (i = 0; i < dict_len; i++) {
739 unsigned char c = dict[i];
742 append_char(&p, e, ' ');
747 append_char(&p, e, '\n');
752 if (c < ' ' || c >= 127 || c == '\\') {
753 p += scnprintf(p, e - p, "\\x%02x", c);
757 append_char(&p, e, c);
759 append_char(&p, e, '\n');
765 /* /dev/kmsg - userspace message inject/listen interface */
766 struct devkmsg_user {
769 struct ratelimit_state rs;
771 char buf[CONSOLE_EXT_LOG_MAX];
774 static __printf(3, 4) __cold
775 int devkmsg_emit(int facility, int level, const char *fmt, ...)
781 r = vprintk_emit(facility, level, NULL, 0, fmt, args);
787 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
790 int level = default_message_loglevel;
791 int facility = 1; /* LOG_USER */
792 struct file *file = iocb->ki_filp;
793 struct devkmsg_user *user = file->private_data;
794 size_t len = iov_iter_count(from);
797 if (!user || len > LOG_LINE_MAX)
800 /* Ignore when user logging is disabled. */
801 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
804 /* Ratelimit when not explicitly enabled. */
805 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
806 if (!___ratelimit(&user->rs, current->comm))
810 buf = kmalloc(len+1, GFP_KERNEL);
815 if (!copy_from_iter_full(buf, len, from)) {
821 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
822 * the decimal value represents 32bit, the lower 3 bit are the log
823 * level, the rest are the log facility.
825 * If no prefix or no userspace facility is specified, we
826 * enforce LOG_USER, to be able to reliably distinguish
827 * kernel-generated messages from userspace-injected ones.
830 if (line[0] == '<') {
834 u = simple_strtoul(line + 1, &endp, 10);
835 if (endp && endp[0] == '>') {
836 level = LOG_LEVEL(u);
837 if (LOG_FACILITY(u) != 0)
838 facility = LOG_FACILITY(u);
845 devkmsg_emit(facility, level, "%s", line);
850 static ssize_t devkmsg_read(struct file *file, char __user *buf,
851 size_t count, loff_t *ppos)
853 struct devkmsg_user *user = file->private_data;
854 struct printk_log *msg;
861 ret = mutex_lock_interruptible(&user->lock);
866 while (user->seq == log_next_seq) {
867 if (file->f_flags & O_NONBLOCK) {
874 ret = wait_event_interruptible(log_wait,
875 user->seq != log_next_seq);
881 if (user->seq < log_first_seq) {
882 /* our last seen message is gone, return error and reset */
883 user->idx = log_first_idx;
884 user->seq = log_first_seq;
890 msg = log_from_idx(user->idx);
891 len = msg_print_ext_header(user->buf, sizeof(user->buf),
893 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
894 log_dict(msg), msg->dict_len,
895 log_text(msg), msg->text_len);
897 user->idx = log_next(user->idx);
906 if (copy_to_user(buf, user->buf, len)) {
912 mutex_unlock(&user->lock);
916 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
918 struct devkmsg_user *user = file->private_data;
929 /* the first record */
930 user->idx = log_first_idx;
931 user->seq = log_first_seq;
935 * The first record after the last SYSLOG_ACTION_CLEAR,
936 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
937 * changes no global state, and does not clear anything.
939 user->idx = clear_idx;
940 user->seq = clear_seq;
943 /* after the last record */
944 user->idx = log_next_idx;
945 user->seq = log_next_seq;
954 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
956 struct devkmsg_user *user = file->private_data;
960 return EPOLLERR|EPOLLNVAL;
962 poll_wait(file, &log_wait, wait);
965 if (user->seq < log_next_seq) {
966 /* return error when data has vanished underneath us */
967 if (user->seq < log_first_seq)
968 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
970 ret = EPOLLIN|EPOLLRDNORM;
977 static int devkmsg_open(struct inode *inode, struct file *file)
979 struct devkmsg_user *user;
982 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
985 /* write-only does not need any file context */
986 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
987 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
993 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
997 ratelimit_default_init(&user->rs);
998 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
1000 mutex_init(&user->lock);
1003 user->idx = log_first_idx;
1004 user->seq = log_first_seq;
1005 logbuf_unlock_irq();
1007 file->private_data = user;
1011 static int devkmsg_release(struct inode *inode, struct file *file)
1013 struct devkmsg_user *user = file->private_data;
1018 ratelimit_state_exit(&user->rs);
1020 mutex_destroy(&user->lock);
1025 const struct file_operations kmsg_fops = {
1026 .open = devkmsg_open,
1027 .read = devkmsg_read,
1028 .write_iter = devkmsg_write,
1029 .llseek = devkmsg_llseek,
1030 .poll = devkmsg_poll,
1031 .release = devkmsg_release,
1034 #ifdef CONFIG_CRASH_CORE
1036 * This appends the listed symbols to /proc/vmcore
1038 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1039 * obtain access to symbols that are otherwise very difficult to locate. These
1040 * symbols are specifically used so that utilities can access and extract the
1041 * dmesg log from a vmcore file after a crash.
1043 void log_buf_vmcoreinfo_setup(void)
1045 VMCOREINFO_SYMBOL(log_buf);
1046 VMCOREINFO_SYMBOL(log_buf_len);
1047 VMCOREINFO_SYMBOL(log_first_idx);
1048 VMCOREINFO_SYMBOL(clear_idx);
1049 VMCOREINFO_SYMBOL(log_next_idx);
1051 * Export struct printk_log size and field offsets. User space tools can
1052 * parse it and detect any changes to structure down the line.
1054 VMCOREINFO_STRUCT_SIZE(printk_log);
1055 VMCOREINFO_OFFSET(printk_log, ts_nsec);
1056 VMCOREINFO_OFFSET(printk_log, len);
1057 VMCOREINFO_OFFSET(printk_log, text_len);
1058 VMCOREINFO_OFFSET(printk_log, dict_len);
1059 #ifdef CONFIG_PRINTK_CALLER
1060 VMCOREINFO_OFFSET(printk_log, caller_id);
1065 /* requested log_buf_len from kernel cmdline */
1066 static unsigned long __initdata new_log_buf_len;
1068 /* we practice scaling the ring buffer by powers of 2 */
1069 static void __init log_buf_len_update(u64 size)
1071 if (size > (u64)LOG_BUF_LEN_MAX) {
1072 size = (u64)LOG_BUF_LEN_MAX;
1073 pr_err("log_buf over 2G is not supported.\n");
1077 size = roundup_pow_of_two(size);
1078 if (size > log_buf_len)
1079 new_log_buf_len = (unsigned long)size;
1082 /* save requested log_buf_len since it's too early to process it */
1083 static int __init log_buf_len_setup(char *str)
1090 size = memparse(str, &str);
1092 log_buf_len_update(size);
1096 early_param("log_buf_len", log_buf_len_setup);
1099 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1101 static void __init log_buf_add_cpu(void)
1103 unsigned int cpu_extra;
1106 * archs should set up cpu_possible_bits properly with
1107 * set_cpu_possible() after setup_arch() but just in
1108 * case lets ensure this is valid.
1110 if (num_possible_cpus() == 1)
1113 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1115 /* by default this will only continue through for large > 64 CPUs */
1116 if (cpu_extra <= __LOG_BUF_LEN / 2)
1119 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1120 __LOG_CPU_MAX_BUF_LEN);
1121 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1123 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1125 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1127 #else /* !CONFIG_SMP */
1128 static inline void log_buf_add_cpu(void) {}
1129 #endif /* CONFIG_SMP */
1131 void __init setup_log_buf(int early)
1133 unsigned long flags;
1137 if (log_buf != __log_buf)
1140 if (!early && !new_log_buf_len)
1143 if (!new_log_buf_len)
1148 memblock_alloc(new_log_buf_len, LOG_ALIGN);
1150 new_log_buf = memblock_alloc_nopanic(new_log_buf_len,
1154 if (unlikely(!new_log_buf)) {
1155 pr_err("log_buf_len: %lu bytes not available\n",
1160 logbuf_lock_irqsave(flags);
1161 log_buf_len = new_log_buf_len;
1162 log_buf = new_log_buf;
1163 new_log_buf_len = 0;
1164 free = __LOG_BUF_LEN - log_next_idx;
1165 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1166 logbuf_unlock_irqrestore(flags);
1168 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1169 pr_info("early log buf free: %u(%u%%)\n",
1170 free, (free * 100) / __LOG_BUF_LEN);
1173 static bool __read_mostly ignore_loglevel;
1175 static int __init ignore_loglevel_setup(char *str)
1177 ignore_loglevel = true;
1178 pr_info("debug: ignoring loglevel setting.\n");
1183 early_param("ignore_loglevel", ignore_loglevel_setup);
1184 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1185 MODULE_PARM_DESC(ignore_loglevel,
1186 "ignore loglevel setting (prints all kernel messages to the console)");
1188 static bool suppress_message_printing(int level)
1190 return (level >= console_loglevel && !ignore_loglevel);
1193 #ifdef CONFIG_BOOT_PRINTK_DELAY
1195 static int boot_delay; /* msecs delay after each printk during bootup */
1196 static unsigned long long loops_per_msec; /* based on boot_delay */
1198 static int __init boot_delay_setup(char *str)
1202 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1203 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1205 get_option(&str, &boot_delay);
1206 if (boot_delay > 10 * 1000)
1209 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1210 "HZ: %d, loops_per_msec: %llu\n",
1211 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1214 early_param("boot_delay", boot_delay_setup);
1216 static void boot_delay_msec(int level)
1218 unsigned long long k;
1219 unsigned long timeout;
1221 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1222 || suppress_message_printing(level)) {
1226 k = (unsigned long long)loops_per_msec * boot_delay;
1228 timeout = jiffies + msecs_to_jiffies(boot_delay);
1233 * use (volatile) jiffies to prevent
1234 * compiler reduction; loop termination via jiffies
1235 * is secondary and may or may not happen.
1237 if (time_after(jiffies, timeout))
1239 touch_nmi_watchdog();
1243 static inline void boot_delay_msec(int level)
1248 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1249 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1251 static size_t print_syslog(unsigned int level, char *buf)
1253 return sprintf(buf, "<%u>", level);
1256 static size_t print_time(u64 ts, char *buf)
1258 unsigned long rem_nsec = do_div(ts, 1000000000);
1260 return sprintf(buf, "[%5lu.%06lu]",
1261 (unsigned long)ts, rem_nsec / 1000);
1264 #ifdef CONFIG_PRINTK_CALLER
1265 static size_t print_caller(u32 id, char *buf)
1269 snprintf(caller, sizeof(caller), "%c%u",
1270 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1271 return sprintf(buf, "[%6s]", caller);
1274 #define print_caller(id, buf) 0
1277 static size_t print_prefix(const struct printk_log *msg, bool syslog,
1278 bool time, char *buf)
1283 len = print_syslog((msg->facility << 3) | msg->level, buf);
1286 len += print_time(msg->ts_nsec, buf + len);
1288 len += print_caller(msg->caller_id, buf + len);
1290 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1298 static size_t msg_print_text(const struct printk_log *msg, bool syslog,
1299 bool time, char *buf, size_t size)
1301 const char *text = log_text(msg);
1302 size_t text_size = msg->text_len;
1304 char prefix[PREFIX_MAX];
1305 const size_t prefix_len = print_prefix(msg, syslog, time, prefix);
1308 const char *next = memchr(text, '\n', text_size);
1312 text_len = next - text;
1314 text_size -= next - text;
1316 text_len = text_size;
1320 if (prefix_len + text_len + 1 >= size - len)
1323 memcpy(buf + len, prefix, prefix_len);
1325 memcpy(buf + len, text, text_len);
1329 /* SYSLOG_ACTION_* buffer size only calculation */
1330 len += prefix_len + text_len + 1;
1339 static int syslog_print(char __user *buf, int size)
1342 struct printk_log *msg;
1345 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1354 if (syslog_seq < log_first_seq) {
1355 /* messages are gone, move to first one */
1356 syslog_seq = log_first_seq;
1357 syslog_idx = log_first_idx;
1360 if (syslog_seq == log_next_seq) {
1361 logbuf_unlock_irq();
1366 * To keep reading/counting partial line consistent,
1367 * use printk_time value as of the beginning of a line.
1369 if (!syslog_partial)
1370 syslog_time = printk_time;
1372 skip = syslog_partial;
1373 msg = log_from_idx(syslog_idx);
1374 n = msg_print_text(msg, true, syslog_time, text,
1375 LOG_LINE_MAX + PREFIX_MAX);
1376 if (n - syslog_partial <= size) {
1377 /* message fits into buffer, move forward */
1378 syslog_idx = log_next(syslog_idx);
1380 n -= syslog_partial;
1383 /* partial read(), remember position */
1385 syslog_partial += n;
1388 logbuf_unlock_irq();
1393 if (copy_to_user(buf, text + skip, n)) {
1408 static int syslog_print_all(char __user *buf, int size, bool clear)
1417 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1424 * Find first record that fits, including all following records,
1425 * into the user-provided buffer for this dump.
1429 while (seq < log_next_seq) {
1430 struct printk_log *msg = log_from_idx(idx);
1432 len += msg_print_text(msg, true, time, NULL, 0);
1433 idx = log_next(idx);
1437 /* move first record forward until length fits into the buffer */
1440 while (len > size && seq < log_next_seq) {
1441 struct printk_log *msg = log_from_idx(idx);
1443 len -= msg_print_text(msg, true, time, NULL, 0);
1444 idx = log_next(idx);
1448 /* last message fitting into this dump */
1449 next_seq = log_next_seq;
1452 while (len >= 0 && seq < next_seq) {
1453 struct printk_log *msg = log_from_idx(idx);
1454 int textlen = msg_print_text(msg, true, time, text,
1455 LOG_LINE_MAX + PREFIX_MAX);
1457 idx = log_next(idx);
1460 logbuf_unlock_irq();
1461 if (copy_to_user(buf + len, text, textlen))
1467 if (seq < log_first_seq) {
1468 /* messages are gone, move to next one */
1469 seq = log_first_seq;
1470 idx = log_first_idx;
1475 clear_seq = log_next_seq;
1476 clear_idx = log_next_idx;
1478 logbuf_unlock_irq();
1484 static void syslog_clear(void)
1487 clear_seq = log_next_seq;
1488 clear_idx = log_next_idx;
1489 logbuf_unlock_irq();
1492 int do_syslog(int type, char __user *buf, int len, int source)
1495 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1498 error = check_syslog_permissions(type, source);
1503 case SYSLOG_ACTION_CLOSE: /* Close log */
1505 case SYSLOG_ACTION_OPEN: /* Open log */
1507 case SYSLOG_ACTION_READ: /* Read from log */
1508 if (!buf || len < 0)
1512 if (!access_ok(buf, len))
1514 error = wait_event_interruptible(log_wait,
1515 syslog_seq != log_next_seq);
1518 error = syslog_print(buf, len);
1520 /* Read/clear last kernel messages */
1521 case SYSLOG_ACTION_READ_CLEAR:
1524 /* Read last kernel messages */
1525 case SYSLOG_ACTION_READ_ALL:
1526 if (!buf || len < 0)
1530 if (!access_ok(buf, len))
1532 error = syslog_print_all(buf, len, clear);
1534 /* Clear ring buffer */
1535 case SYSLOG_ACTION_CLEAR:
1538 /* Disable logging to console */
1539 case SYSLOG_ACTION_CONSOLE_OFF:
1540 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1541 saved_console_loglevel = console_loglevel;
1542 console_loglevel = minimum_console_loglevel;
1544 /* Enable logging to console */
1545 case SYSLOG_ACTION_CONSOLE_ON:
1546 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1547 console_loglevel = saved_console_loglevel;
1548 saved_console_loglevel = LOGLEVEL_DEFAULT;
1551 /* Set level of messages printed to console */
1552 case SYSLOG_ACTION_CONSOLE_LEVEL:
1553 if (len < 1 || len > 8)
1555 if (len < minimum_console_loglevel)
1556 len = minimum_console_loglevel;
1557 console_loglevel = len;
1558 /* Implicitly re-enable logging to console */
1559 saved_console_loglevel = LOGLEVEL_DEFAULT;
1561 /* Number of chars in the log buffer */
1562 case SYSLOG_ACTION_SIZE_UNREAD:
1564 if (syslog_seq < log_first_seq) {
1565 /* messages are gone, move to first one */
1566 syslog_seq = log_first_seq;
1567 syslog_idx = log_first_idx;
1570 if (source == SYSLOG_FROM_PROC) {
1572 * Short-cut for poll(/"proc/kmsg") which simply checks
1573 * for pending data, not the size; return the count of
1574 * records, not the length.
1576 error = log_next_seq - syslog_seq;
1578 u64 seq = syslog_seq;
1579 u32 idx = syslog_idx;
1580 bool time = syslog_partial ? syslog_time : printk_time;
1582 while (seq < log_next_seq) {
1583 struct printk_log *msg = log_from_idx(idx);
1585 error += msg_print_text(msg, true, time, NULL,
1588 idx = log_next(idx);
1591 error -= syslog_partial;
1593 logbuf_unlock_irq();
1595 /* Size of the log buffer */
1596 case SYSLOG_ACTION_SIZE_BUFFER:
1597 error = log_buf_len;
1607 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1609 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1613 * Special console_lock variants that help to reduce the risk of soft-lockups.
1614 * They allow to pass console_lock to another printk() call using a busy wait.
1617 #ifdef CONFIG_LOCKDEP
1618 static struct lockdep_map console_owner_dep_map = {
1619 .name = "console_owner"
1623 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1624 static struct task_struct *console_owner;
1625 static bool console_waiter;
1628 * console_lock_spinning_enable - mark beginning of code where another
1629 * thread might safely busy wait
1631 * This basically converts console_lock into a spinlock. This marks
1632 * the section where the console_lock owner can not sleep, because
1633 * there may be a waiter spinning (like a spinlock). Also it must be
1634 * ready to hand over the lock at the end of the section.
1636 static void console_lock_spinning_enable(void)
1638 raw_spin_lock(&console_owner_lock);
1639 console_owner = current;
1640 raw_spin_unlock(&console_owner_lock);
1642 /* The waiter may spin on us after setting console_owner */
1643 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1647 * console_lock_spinning_disable_and_check - mark end of code where another
1648 * thread was able to busy wait and check if there is a waiter
1650 * This is called at the end of the section where spinning is allowed.
1651 * It has two functions. First, it is a signal that it is no longer
1652 * safe to start busy waiting for the lock. Second, it checks if
1653 * there is a busy waiter and passes the lock rights to her.
1655 * Important: Callers lose the lock if there was a busy waiter.
1656 * They must not touch items synchronized by console_lock
1659 * Return: 1 if the lock rights were passed, 0 otherwise.
1661 static int console_lock_spinning_disable_and_check(void)
1665 raw_spin_lock(&console_owner_lock);
1666 waiter = READ_ONCE(console_waiter);
1667 console_owner = NULL;
1668 raw_spin_unlock(&console_owner_lock);
1671 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1675 /* The waiter is now free to continue */
1676 WRITE_ONCE(console_waiter, false);
1678 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1681 * Hand off console_lock to waiter. The waiter will perform
1682 * the up(). After this, the waiter is the console_lock owner.
1684 mutex_release(&console_lock_dep_map, 1, _THIS_IP_);
1689 * console_trylock_spinning - try to get console_lock by busy waiting
1691 * This allows to busy wait for the console_lock when the current
1692 * owner is running in specially marked sections. It means that
1693 * the current owner is running and cannot reschedule until it
1694 * is ready to lose the lock.
1696 * Return: 1 if we got the lock, 0 othrewise
1698 static int console_trylock_spinning(void)
1700 struct task_struct *owner = NULL;
1703 unsigned long flags;
1705 if (console_trylock())
1708 printk_safe_enter_irqsave(flags);
1710 raw_spin_lock(&console_owner_lock);
1711 owner = READ_ONCE(console_owner);
1712 waiter = READ_ONCE(console_waiter);
1713 if (!waiter && owner && owner != current) {
1714 WRITE_ONCE(console_waiter, true);
1717 raw_spin_unlock(&console_owner_lock);
1720 * If there is an active printk() writing to the
1721 * consoles, instead of having it write our data too,
1722 * see if we can offload that load from the active
1723 * printer, and do some printing ourselves.
1724 * Go into a spin only if there isn't already a waiter
1725 * spinning, and there is an active printer, and
1726 * that active printer isn't us (recursive printk?).
1729 printk_safe_exit_irqrestore(flags);
1733 /* We spin waiting for the owner to release us */
1734 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1735 /* Owner will clear console_waiter on hand off */
1736 while (READ_ONCE(console_waiter))
1738 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1740 printk_safe_exit_irqrestore(flags);
1742 * The owner passed the console lock to us.
1743 * Since we did not spin on console lock, annotate
1744 * this as a trylock. Otherwise lockdep will
1747 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1753 * Call the console drivers, asking them to write out
1754 * log_buf[start] to log_buf[end - 1].
1755 * The console_lock must be held.
1757 static void call_console_drivers(const char *ext_text, size_t ext_len,
1758 const char *text, size_t len)
1760 struct console *con;
1762 trace_console_rcuidle(text, len);
1764 if (!console_drivers)
1767 for_each_console(con) {
1768 if (exclusive_console && con != exclusive_console)
1770 if (!(con->flags & CON_ENABLED))
1774 if (!cpu_online(smp_processor_id()) &&
1775 !(con->flags & CON_ANYTIME))
1777 if (con->flags & CON_EXTENDED)
1778 con->write(con, ext_text, ext_len);
1780 con->write(con, text, len);
1784 int printk_delay_msec __read_mostly;
1786 static inline void printk_delay(void)
1788 if (unlikely(printk_delay_msec)) {
1789 int m = printk_delay_msec;
1793 touch_nmi_watchdog();
1798 static inline u32 printk_caller_id(void)
1800 return in_task() ? task_pid_nr(current) :
1801 0x80000000 + raw_smp_processor_id();
1805 * Continuation lines are buffered, and not committed to the record buffer
1806 * until the line is complete, or a race forces it. The line fragments
1807 * though, are printed immediately to the consoles to ensure everything has
1808 * reached the console in case of a kernel crash.
1810 static struct cont {
1811 char buf[LOG_LINE_MAX];
1812 size_t len; /* length == 0 means unused buffer */
1813 u32 caller_id; /* printk_caller_id() of first print */
1814 u64 ts_nsec; /* time of first print */
1815 u8 level; /* log level of first message */
1816 u8 facility; /* log facility of first message */
1817 enum log_flags flags; /* prefix, newline flags */
1820 static void cont_flush(void)
1825 log_store(cont.caller_id, cont.facility, cont.level, cont.flags,
1826 cont.ts_nsec, NULL, 0, cont.buf, cont.len);
1830 static bool cont_add(u32 caller_id, int facility, int level,
1831 enum log_flags flags, const char *text, size_t len)
1833 /* If the line gets too long, split it up in separate records. */
1834 if (cont.len + len > sizeof(cont.buf)) {
1840 cont.facility = facility;
1842 cont.caller_id = caller_id;
1843 cont.ts_nsec = local_clock();
1847 memcpy(cont.buf + cont.len, text, len);
1850 // The original flags come from the first line,
1851 // but later continuations can add a newline.
1852 if (flags & LOG_NEWLINE) {
1853 cont.flags |= LOG_NEWLINE;
1860 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)
1862 const u32 caller_id = printk_caller_id();
1865 * If an earlier line was buffered, and we're a continuation
1866 * write from the same context, try to add it to the buffer.
1869 if (cont.caller_id == caller_id && (lflags & LOG_CONT)) {
1870 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1873 /* Otherwise, make sure it's flushed */
1877 /* Skip empty continuation lines that couldn't be added - they just flush */
1878 if (!text_len && (lflags & LOG_CONT))
1881 /* If it doesn't end in a newline, try to buffer the current line */
1882 if (!(lflags & LOG_NEWLINE)) {
1883 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1887 /* Store it in the record log */
1888 return log_store(caller_id, facility, level, lflags, 0,
1889 dict, dictlen, text, text_len);
1892 /* Must be called under logbuf_lock. */
1893 int vprintk_store(int facility, int level,
1894 const char *dict, size_t dictlen,
1895 const char *fmt, va_list args)
1897 static char textbuf[LOG_LINE_MAX];
1898 char *text = textbuf;
1900 enum log_flags lflags = 0;
1903 * The printf needs to come first; we need the syslog
1904 * prefix which might be passed-in as a parameter.
1906 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1908 /* mark and strip a trailing newline */
1909 if (text_len && text[text_len-1] == '\n') {
1911 lflags |= LOG_NEWLINE;
1914 /* strip kernel syslog prefix and extract log level or control flags */
1915 if (facility == 0) {
1918 while ((kern_level = printk_get_level(text)) != 0) {
1919 switch (kern_level) {
1921 if (level == LOGLEVEL_DEFAULT)
1922 level = kern_level - '0';
1924 case 'c': /* KERN_CONT */
1933 if (level == LOGLEVEL_DEFAULT)
1934 level = default_message_loglevel;
1937 lflags |= LOG_NEWLINE;
1939 return log_output(facility, level, lflags,
1940 dict, dictlen, text, text_len);
1943 asmlinkage int vprintk_emit(int facility, int level,
1944 const char *dict, size_t dictlen,
1945 const char *fmt, va_list args)
1948 bool in_sched = false, pending_output;
1949 unsigned long flags;
1952 if (level == LOGLEVEL_SCHED) {
1953 level = LOGLEVEL_DEFAULT;
1957 boot_delay_msec(level);
1960 /* This stops the holder of console_sem just where we want him */
1961 logbuf_lock_irqsave(flags);
1962 curr_log_seq = log_next_seq;
1963 printed_len = vprintk_store(facility, level, dict, dictlen, fmt, args);
1964 pending_output = (curr_log_seq != log_next_seq);
1965 logbuf_unlock_irqrestore(flags);
1967 /* If called from the scheduler, we can not call up(). */
1968 if (!in_sched && pending_output) {
1970 * Disable preemption to avoid being preempted while holding
1971 * console_sem which would prevent anyone from printing to
1976 * Try to acquire and then immediately release the console
1977 * semaphore. The release will print out buffers and wake up
1978 * /dev/kmsg and syslog() users.
1980 if (console_trylock_spinning())
1989 EXPORT_SYMBOL(vprintk_emit);
1991 asmlinkage int vprintk(const char *fmt, va_list args)
1993 return vprintk_func(fmt, args);
1995 EXPORT_SYMBOL(vprintk);
1997 int vprintk_default(const char *fmt, va_list args)
2001 #ifdef CONFIG_KGDB_KDB
2002 /* Allow to pass printk() to kdb but avoid a recursion. */
2003 if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
2004 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
2008 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
2012 EXPORT_SYMBOL_GPL(vprintk_default);
2015 * printk - print a kernel message
2016 * @fmt: format string
2018 * This is printk(). It can be called from any context. We want it to work.
2020 * We try to grab the console_lock. If we succeed, it's easy - we log the
2021 * output and call the console drivers. If we fail to get the semaphore, we
2022 * place the output into the log buffer and return. The current holder of
2023 * the console_sem will notice the new output in console_unlock(); and will
2024 * send it to the consoles before releasing the lock.
2026 * One effect of this deferred printing is that code which calls printk() and
2027 * then changes console_loglevel may break. This is because console_loglevel
2028 * is inspected when the actual printing occurs.
2033 * See the vsnprintf() documentation for format string extensions over C99.
2035 asmlinkage __visible int printk(const char *fmt, ...)
2040 va_start(args, fmt);
2041 r = vprintk_func(fmt, args);
2046 EXPORT_SYMBOL(printk);
2048 #else /* CONFIG_PRINTK */
2050 #define LOG_LINE_MAX 0
2051 #define PREFIX_MAX 0
2052 #define printk_time false
2054 static u64 syslog_seq;
2055 static u32 syslog_idx;
2056 static u64 console_seq;
2057 static u32 console_idx;
2058 static u64 exclusive_console_stop_seq;
2059 static u64 log_first_seq;
2060 static u32 log_first_idx;
2061 static u64 log_next_seq;
2062 static char *log_text(const struct printk_log *msg) { return NULL; }
2063 static char *log_dict(const struct printk_log *msg) { return NULL; }
2064 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2065 static u32 log_next(u32 idx) { return 0; }
2066 static ssize_t msg_print_ext_header(char *buf, size_t size,
2067 struct printk_log *msg,
2068 u64 seq) { return 0; }
2069 static ssize_t msg_print_ext_body(char *buf, size_t size,
2070 char *dict, size_t dict_len,
2071 char *text, size_t text_len) { return 0; }
2072 static void console_lock_spinning_enable(void) { }
2073 static int console_lock_spinning_disable_and_check(void) { return 0; }
2074 static void call_console_drivers(const char *ext_text, size_t ext_len,
2075 const char *text, size_t len) {}
2076 static size_t msg_print_text(const struct printk_log *msg, bool syslog,
2077 bool time, char *buf, size_t size) { return 0; }
2078 static bool suppress_message_printing(int level) { return false; }
2080 #endif /* CONFIG_PRINTK */
2082 #ifdef CONFIG_EARLY_PRINTK
2083 struct console *early_console;
2085 asmlinkage __visible void early_printk(const char *fmt, ...)
2095 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2098 early_console->write(early_console, buf, n);
2102 static int __add_preferred_console(char *name, int idx, char *options,
2105 struct console_cmdline *c;
2109 * See if this tty is not yet registered, and
2110 * if we have a slot free.
2112 for (i = 0, c = console_cmdline;
2113 i < MAX_CMDLINECONSOLES && c->name[0];
2115 if (strcmp(c->name, name) == 0 && c->index == idx) {
2117 preferred_console = i;
2121 if (i == MAX_CMDLINECONSOLES)
2124 preferred_console = i;
2125 strlcpy(c->name, name, sizeof(c->name));
2126 c->options = options;
2127 braille_set_options(c, brl_options);
2133 static int __init console_msg_format_setup(char *str)
2135 if (!strcmp(str, "syslog"))
2136 console_msg_format = MSG_FORMAT_SYSLOG;
2137 if (!strcmp(str, "default"))
2138 console_msg_format = MSG_FORMAT_DEFAULT;
2141 __setup("console_msg_format=", console_msg_format_setup);
2144 * Set up a console. Called via do_early_param() in init/main.c
2145 * for each "console=" parameter in the boot command line.
2147 static int __init console_setup(char *str)
2149 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2150 char *s, *options, *brl_options = NULL;
2153 if (_braille_console_setup(&str, &brl_options))
2157 * Decode str into name, index, options.
2159 if (str[0] >= '0' && str[0] <= '9') {
2160 strcpy(buf, "ttyS");
2161 strncpy(buf + 4, str, sizeof(buf) - 5);
2163 strncpy(buf, str, sizeof(buf) - 1);
2165 buf[sizeof(buf) - 1] = 0;
2166 options = strchr(str, ',');
2170 if (!strcmp(str, "ttya"))
2171 strcpy(buf, "ttyS0");
2172 if (!strcmp(str, "ttyb"))
2173 strcpy(buf, "ttyS1");
2175 for (s = buf; *s; s++)
2176 if (isdigit(*s) || *s == ',')
2178 idx = simple_strtoul(s, NULL, 10);
2181 __add_preferred_console(buf, idx, options, brl_options);
2182 console_set_on_cmdline = 1;
2185 __setup("console=", console_setup);
2188 * add_preferred_console - add a device to the list of preferred consoles.
2189 * @name: device name
2190 * @idx: device index
2191 * @options: options for this console
2193 * The last preferred console added will be used for kernel messages
2194 * and stdin/out/err for init. Normally this is used by console_setup
2195 * above to handle user-supplied console arguments; however it can also
2196 * be used by arch-specific code either to override the user or more
2197 * commonly to provide a default console (ie from PROM variables) when
2198 * the user has not supplied one.
2200 int add_preferred_console(char *name, int idx, char *options)
2202 return __add_preferred_console(name, idx, options, NULL);
2205 bool console_suspend_enabled = true;
2206 EXPORT_SYMBOL(console_suspend_enabled);
2208 static int __init console_suspend_disable(char *str)
2210 console_suspend_enabled = false;
2213 __setup("no_console_suspend", console_suspend_disable);
2214 module_param_named(console_suspend, console_suspend_enabled,
2215 bool, S_IRUGO | S_IWUSR);
2216 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2217 " and hibernate operations");
2220 * suspend_console - suspend the console subsystem
2222 * This disables printk() while we go into suspend states
2224 void suspend_console(void)
2226 if (!console_suspend_enabled)
2228 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2230 console_suspended = 1;
2234 void resume_console(void)
2236 if (!console_suspend_enabled)
2239 console_suspended = 0;
2244 * console_cpu_notify - print deferred console messages after CPU hotplug
2247 * If printk() is called from a CPU that is not online yet, the messages
2248 * will be printed on the console only if there are CON_ANYTIME consoles.
2249 * This function is called when a new CPU comes online (or fails to come
2250 * up) or goes offline.
2252 static int console_cpu_notify(unsigned int cpu)
2254 if (!cpuhp_tasks_frozen) {
2255 /* If trylock fails, someone else is doing the printing */
2256 if (console_trylock())
2263 * console_lock - lock the console system for exclusive use.
2265 * Acquires a lock which guarantees that the caller has
2266 * exclusive access to the console system and the console_drivers list.
2268 * Can sleep, returns nothing.
2270 void console_lock(void)
2275 if (console_suspended)
2278 console_may_schedule = 1;
2280 EXPORT_SYMBOL(console_lock);
2283 * console_trylock - try to lock the console system for exclusive use.
2285 * Try to acquire a lock which guarantees that the caller has exclusive
2286 * access to the console system and the console_drivers list.
2288 * returns 1 on success, and 0 on failure to acquire the lock.
2290 int console_trylock(void)
2292 if (down_trylock_console_sem())
2294 if (console_suspended) {
2299 console_may_schedule = 0;
2302 EXPORT_SYMBOL(console_trylock);
2304 int is_console_locked(void)
2306 return console_locked;
2308 EXPORT_SYMBOL(is_console_locked);
2311 * Check if we have any console that is capable of printing while cpu is
2312 * booting or shutting down. Requires console_sem.
2314 static int have_callable_console(void)
2316 struct console *con;
2318 for_each_console(con)
2319 if ((con->flags & CON_ENABLED) &&
2320 (con->flags & CON_ANYTIME))
2327 * Can we actually use the console at this time on this cpu?
2329 * Console drivers may assume that per-cpu resources have been allocated. So
2330 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2331 * call them until this CPU is officially up.
2333 static inline int can_use_console(void)
2335 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2339 * console_unlock - unlock the console system
2341 * Releases the console_lock which the caller holds on the console system
2342 * and the console driver list.
2344 * While the console_lock was held, console output may have been buffered
2345 * by printk(). If this is the case, console_unlock(); emits
2346 * the output prior to releasing the lock.
2348 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2350 * console_unlock(); may be called from any context.
2352 void console_unlock(void)
2354 static char ext_text[CONSOLE_EXT_LOG_MAX];
2355 static char text[LOG_LINE_MAX + PREFIX_MAX];
2356 unsigned long flags;
2357 bool do_cond_resched, retry;
2359 if (console_suspended) {
2365 * Console drivers are called with interrupts disabled, so
2366 * @console_may_schedule should be cleared before; however, we may
2367 * end up dumping a lot of lines, for example, if called from
2368 * console registration path, and should invoke cond_resched()
2369 * between lines if allowable. Not doing so can cause a very long
2370 * scheduling stall on a slow console leading to RCU stall and
2371 * softlockup warnings which exacerbate the issue with more
2372 * messages practically incapacitating the system.
2374 * console_trylock() is not able to detect the preemptive
2375 * context reliably. Therefore the value must be stored before
2376 * and cleared after the the "again" goto label.
2378 do_cond_resched = console_may_schedule;
2380 console_may_schedule = 0;
2383 * We released the console_sem lock, so we need to recheck if
2384 * cpu is online and (if not) is there at least one CON_ANYTIME
2387 if (!can_use_console()) {
2394 struct printk_log *msg;
2398 printk_safe_enter_irqsave(flags);
2399 raw_spin_lock(&logbuf_lock);
2400 if (console_seq < log_first_seq) {
2402 "** %llu printk messages dropped **\n",
2403 log_first_seq - console_seq);
2405 /* messages are gone, move to first one */
2406 console_seq = log_first_seq;
2407 console_idx = log_first_idx;
2412 if (console_seq == log_next_seq)
2415 msg = log_from_idx(console_idx);
2416 if (suppress_message_printing(msg->level)) {
2418 * Skip record we have buffered and already printed
2419 * directly to the console when we received it, and
2420 * record that has level above the console loglevel.
2422 console_idx = log_next(console_idx);
2427 /* Output to all consoles once old messages replayed. */
2428 if (unlikely(exclusive_console &&
2429 console_seq >= exclusive_console_stop_seq)) {
2430 exclusive_console = NULL;
2433 len += msg_print_text(msg,
2434 console_msg_format & MSG_FORMAT_SYSLOG,
2435 printk_time, text + len, sizeof(text) - len);
2436 if (nr_ext_console_drivers) {
2437 ext_len = msg_print_ext_header(ext_text,
2440 ext_len += msg_print_ext_body(ext_text + ext_len,
2441 sizeof(ext_text) - ext_len,
2442 log_dict(msg), msg->dict_len,
2443 log_text(msg), msg->text_len);
2445 console_idx = log_next(console_idx);
2447 raw_spin_unlock(&logbuf_lock);
2450 * While actively printing out messages, if another printk()
2451 * were to occur on another CPU, it may wait for this one to
2452 * finish. This task can not be preempted if there is a
2453 * waiter waiting to take over.
2455 console_lock_spinning_enable();
2457 stop_critical_timings(); /* don't trace print latency */
2458 call_console_drivers(ext_text, ext_len, text, len);
2459 start_critical_timings();
2461 if (console_lock_spinning_disable_and_check()) {
2462 printk_safe_exit_irqrestore(flags);
2466 printk_safe_exit_irqrestore(flags);
2468 if (do_cond_resched)
2474 raw_spin_unlock(&logbuf_lock);
2479 * Someone could have filled up the buffer again, so re-check if there's
2480 * something to flush. In case we cannot trylock the console_sem again,
2481 * there's a new owner and the console_unlock() from them will do the
2482 * flush, no worries.
2484 raw_spin_lock(&logbuf_lock);
2485 retry = console_seq != log_next_seq;
2486 raw_spin_unlock(&logbuf_lock);
2487 printk_safe_exit_irqrestore(flags);
2489 if (retry && console_trylock())
2492 EXPORT_SYMBOL(console_unlock);
2495 * console_conditional_schedule - yield the CPU if required
2497 * If the console code is currently allowed to sleep, and
2498 * if this CPU should yield the CPU to another task, do
2501 * Must be called within console_lock();.
2503 void __sched console_conditional_schedule(void)
2505 if (console_may_schedule)
2508 EXPORT_SYMBOL(console_conditional_schedule);
2510 void console_unblank(void)
2515 * console_unblank can no longer be called in interrupt context unless
2516 * oops_in_progress is set to 1..
2518 if (oops_in_progress) {
2519 if (down_trylock_console_sem() != 0)
2525 console_may_schedule = 0;
2527 if ((c->flags & CON_ENABLED) && c->unblank)
2533 * console_flush_on_panic - flush console content on panic
2535 * Immediately output all pending messages no matter what.
2537 void console_flush_on_panic(void)
2540 * If someone else is holding the console lock, trylock will fail
2541 * and may_schedule may be set. Ignore and proceed to unlock so
2542 * that messages are flushed out. As this can be called from any
2543 * context and we don't want to get preempted while flushing,
2544 * ensure may_schedule is cleared.
2547 console_may_schedule = 0;
2552 * Return the console tty driver structure and its associated index
2554 struct tty_driver *console_device(int *index)
2557 struct tty_driver *driver = NULL;
2560 for_each_console(c) {
2563 driver = c->device(c, index);
2572 * Prevent further output on the passed console device so that (for example)
2573 * serial drivers can disable console output before suspending a port, and can
2574 * re-enable output afterwards.
2576 void console_stop(struct console *console)
2579 console->flags &= ~CON_ENABLED;
2582 EXPORT_SYMBOL(console_stop);
2584 void console_start(struct console *console)
2587 console->flags |= CON_ENABLED;
2590 EXPORT_SYMBOL(console_start);
2592 static int __read_mostly keep_bootcon;
2594 static int __init keep_bootcon_setup(char *str)
2597 pr_info("debug: skip boot console de-registration.\n");
2602 early_param("keep_bootcon", keep_bootcon_setup);
2605 * The console driver calls this routine during kernel initialization
2606 * to register the console printing procedure with printk() and to
2607 * print any messages that were printed by the kernel before the
2608 * console driver was initialized.
2610 * This can happen pretty early during the boot process (because of
2611 * early_printk) - sometimes before setup_arch() completes - be careful
2612 * of what kernel features are used - they may not be initialised yet.
2614 * There are two types of consoles - bootconsoles (early_printk) and
2615 * "real" consoles (everything which is not a bootconsole) which are
2616 * handled differently.
2617 * - Any number of bootconsoles can be registered at any time.
2618 * - As soon as a "real" console is registered, all bootconsoles
2619 * will be unregistered automatically.
2620 * - Once a "real" console is registered, any attempt to register a
2621 * bootconsoles will be rejected
2623 void register_console(struct console *newcon)
2626 unsigned long flags;
2627 struct console *bcon = NULL;
2628 struct console_cmdline *c;
2629 static bool has_preferred;
2631 if (console_drivers)
2632 for_each_console(bcon)
2633 if (WARN(bcon == newcon,
2634 "console '%s%d' already registered\n",
2635 bcon->name, bcon->index))
2639 * before we register a new CON_BOOT console, make sure we don't
2640 * already have a valid console
2642 if (console_drivers && newcon->flags & CON_BOOT) {
2643 /* find the last or real console */
2644 for_each_console(bcon) {
2645 if (!(bcon->flags & CON_BOOT)) {
2646 pr_info("Too late to register bootconsole %s%d\n",
2647 newcon->name, newcon->index);
2653 if (console_drivers && console_drivers->flags & CON_BOOT)
2654 bcon = console_drivers;
2656 if (!has_preferred || bcon || !console_drivers)
2657 has_preferred = preferred_console >= 0;
2660 * See if we want to use this console driver. If we
2661 * didn't select a console we take the first one
2662 * that registers here.
2664 if (!has_preferred) {
2665 if (newcon->index < 0)
2667 if (newcon->setup == NULL ||
2668 newcon->setup(newcon, NULL) == 0) {
2669 newcon->flags |= CON_ENABLED;
2670 if (newcon->device) {
2671 newcon->flags |= CON_CONSDEV;
2672 has_preferred = true;
2678 * See if this console matches one we selected on
2681 for (i = 0, c = console_cmdline;
2682 i < MAX_CMDLINECONSOLES && c->name[0];
2684 if (!newcon->match ||
2685 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2686 /* default matching */
2687 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2688 if (strcmp(c->name, newcon->name) != 0)
2690 if (newcon->index >= 0 &&
2691 newcon->index != c->index)
2693 if (newcon->index < 0)
2694 newcon->index = c->index;
2696 if (_braille_register_console(newcon, c))
2699 if (newcon->setup &&
2700 newcon->setup(newcon, c->options) != 0)
2704 newcon->flags |= CON_ENABLED;
2705 if (i == preferred_console) {
2706 newcon->flags |= CON_CONSDEV;
2707 has_preferred = true;
2712 if (!(newcon->flags & CON_ENABLED))
2716 * If we have a bootconsole, and are switching to a real console,
2717 * don't print everything out again, since when the boot console, and
2718 * the real console are the same physical device, it's annoying to
2719 * see the beginning boot messages twice
2721 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2722 newcon->flags &= ~CON_PRINTBUFFER;
2725 * Put this console in the list - keep the
2726 * preferred driver at the head of the list.
2729 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2730 newcon->next = console_drivers;
2731 console_drivers = newcon;
2733 newcon->next->flags &= ~CON_CONSDEV;
2735 newcon->next = console_drivers->next;
2736 console_drivers->next = newcon;
2739 if (newcon->flags & CON_EXTENDED)
2740 nr_ext_console_drivers++;
2742 if (newcon->flags & CON_PRINTBUFFER) {
2744 * console_unlock(); will print out the buffered messages
2747 logbuf_lock_irqsave(flags);
2748 console_seq = syslog_seq;
2749 console_idx = syslog_idx;
2751 * We're about to replay the log buffer. Only do this to the
2752 * just-registered console to avoid excessive message spam to
2753 * the already-registered consoles.
2755 * Set exclusive_console with disabled interrupts to reduce
2756 * race window with eventual console_flush_on_panic() that
2757 * ignores console_lock.
2759 exclusive_console = newcon;
2760 exclusive_console_stop_seq = console_seq;
2761 logbuf_unlock_irqrestore(flags);
2764 console_sysfs_notify();
2767 * By unregistering the bootconsoles after we enable the real console
2768 * we get the "console xxx enabled" message on all the consoles -
2769 * boot consoles, real consoles, etc - this is to ensure that end
2770 * users know there might be something in the kernel's log buffer that
2771 * went to the bootconsole (that they do not see on the real console)
2773 pr_info("%sconsole [%s%d] enabled\n",
2774 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2775 newcon->name, newcon->index);
2777 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2779 /* We need to iterate through all boot consoles, to make
2780 * sure we print everything out, before we unregister them.
2782 for_each_console(bcon)
2783 if (bcon->flags & CON_BOOT)
2784 unregister_console(bcon);
2787 EXPORT_SYMBOL(register_console);
2789 int unregister_console(struct console *console)
2791 struct console *a, *b;
2794 pr_info("%sconsole [%s%d] disabled\n",
2795 (console->flags & CON_BOOT) ? "boot" : "" ,
2796 console->name, console->index);
2798 res = _braille_unregister_console(console);
2804 if (console_drivers == console) {
2805 console_drivers=console->next;
2807 } else if (console_drivers) {
2808 for (a=console_drivers->next, b=console_drivers ;
2809 a; b=a, a=b->next) {
2818 if (!res && (console->flags & CON_EXTENDED))
2819 nr_ext_console_drivers--;
2822 * If this isn't the last console and it has CON_CONSDEV set, we
2823 * need to set it on the next preferred console.
2825 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2826 console_drivers->flags |= CON_CONSDEV;
2828 console->flags &= ~CON_ENABLED;
2830 console_sysfs_notify();
2833 EXPORT_SYMBOL(unregister_console);
2836 * Initialize the console device. This is called *early*, so
2837 * we can't necessarily depend on lots of kernel help here.
2838 * Just do some early initializations, and do the complex setup
2841 void __init console_init(void)
2845 initcall_entry_t *ce;
2847 /* Setup the default TTY line discipline. */
2851 * set up the console device so that later boot sequences can
2852 * inform about problems etc..
2854 ce = __con_initcall_start;
2855 trace_initcall_level("console");
2856 while (ce < __con_initcall_end) {
2857 call = initcall_from_entry(ce);
2858 trace_initcall_start(call);
2860 trace_initcall_finish(call, ret);
2866 * Some boot consoles access data that is in the init section and which will
2867 * be discarded after the initcalls have been run. To make sure that no code
2868 * will access this data, unregister the boot consoles in a late initcall.
2870 * If for some reason, such as deferred probe or the driver being a loadable
2871 * module, the real console hasn't registered yet at this point, there will
2872 * be a brief interval in which no messages are logged to the console, which
2873 * makes it difficult to diagnose problems that occur during this time.
2875 * To mitigate this problem somewhat, only unregister consoles whose memory
2876 * intersects with the init section. Note that all other boot consoles will
2877 * get unregistred when the real preferred console is registered.
2879 static int __init printk_late_init(void)
2881 struct console *con;
2884 for_each_console(con) {
2885 if (!(con->flags & CON_BOOT))
2888 /* Check addresses that might be used for enabled consoles. */
2889 if (init_section_intersects(con, sizeof(*con)) ||
2890 init_section_contains(con->write, 0) ||
2891 init_section_contains(con->read, 0) ||
2892 init_section_contains(con->device, 0) ||
2893 init_section_contains(con->unblank, 0) ||
2894 init_section_contains(con->data, 0)) {
2896 * Please, consider moving the reported consoles out
2897 * of the init section.
2899 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2900 con->name, con->index);
2901 unregister_console(con);
2904 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2905 console_cpu_notify);
2907 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2908 console_cpu_notify, NULL);
2912 late_initcall(printk_late_init);
2914 #if defined CONFIG_PRINTK
2916 * Delayed printk version, for scheduler-internal messages:
2918 #define PRINTK_PENDING_WAKEUP 0x01
2919 #define PRINTK_PENDING_OUTPUT 0x02
2921 static DEFINE_PER_CPU(int, printk_pending);
2923 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2925 int pending = __this_cpu_xchg(printk_pending, 0);
2927 if (pending & PRINTK_PENDING_OUTPUT) {
2928 /* If trylock fails, someone else is doing the printing */
2929 if (console_trylock())
2933 if (pending & PRINTK_PENDING_WAKEUP)
2934 wake_up_interruptible(&log_wait);
2937 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2938 .func = wake_up_klogd_work_func,
2939 .flags = IRQ_WORK_LAZY,
2942 void wake_up_klogd(void)
2945 if (waitqueue_active(&log_wait)) {
2946 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2947 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2952 void defer_console_output(void)
2955 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2956 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2960 int vprintk_deferred(const char *fmt, va_list args)
2964 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2965 defer_console_output();
2970 int printk_deferred(const char *fmt, ...)
2975 va_start(args, fmt);
2976 r = vprintk_deferred(fmt, args);
2983 * printk rate limiting, lifted from the networking subsystem.
2985 * This enforces a rate limit: not more than 10 kernel messages
2986 * every 5s to make a denial-of-service attack impossible.
2988 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2990 int __printk_ratelimit(const char *func)
2992 return ___ratelimit(&printk_ratelimit_state, func);
2994 EXPORT_SYMBOL(__printk_ratelimit);
2997 * printk_timed_ratelimit - caller-controlled printk ratelimiting
2998 * @caller_jiffies: pointer to caller's state
2999 * @interval_msecs: minimum interval between prints
3001 * printk_timed_ratelimit() returns true if more than @interval_msecs
3002 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3005 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3006 unsigned int interval_msecs)
3008 unsigned long elapsed = jiffies - *caller_jiffies;
3010 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3013 *caller_jiffies = jiffies;
3016 EXPORT_SYMBOL(printk_timed_ratelimit);
3018 static DEFINE_SPINLOCK(dump_list_lock);
3019 static LIST_HEAD(dump_list);
3022 * kmsg_dump_register - register a kernel log dumper.
3023 * @dumper: pointer to the kmsg_dumper structure
3025 * Adds a kernel log dumper to the system. The dump callback in the
3026 * structure will be called when the kernel oopses or panics and must be
3027 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3029 int kmsg_dump_register(struct kmsg_dumper *dumper)
3031 unsigned long flags;
3034 /* The dump callback needs to be set */
3038 spin_lock_irqsave(&dump_list_lock, flags);
3039 /* Don't allow registering multiple times */
3040 if (!dumper->registered) {
3041 dumper->registered = 1;
3042 list_add_tail_rcu(&dumper->list, &dump_list);
3045 spin_unlock_irqrestore(&dump_list_lock, flags);
3049 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3052 * kmsg_dump_unregister - unregister a kmsg dumper.
3053 * @dumper: pointer to the kmsg_dumper structure
3055 * Removes a dump device from the system. Returns zero on success and
3056 * %-EINVAL otherwise.
3058 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3060 unsigned long flags;
3063 spin_lock_irqsave(&dump_list_lock, flags);
3064 if (dumper->registered) {
3065 dumper->registered = 0;
3066 list_del_rcu(&dumper->list);
3069 spin_unlock_irqrestore(&dump_list_lock, flags);
3074 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3076 static bool always_kmsg_dump;
3077 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3080 * kmsg_dump - dump kernel log to kernel message dumpers.
3081 * @reason: the reason (oops, panic etc) for dumping
3083 * Call each of the registered dumper's dump() callback, which can
3084 * retrieve the kmsg records with kmsg_dump_get_line() or
3085 * kmsg_dump_get_buffer().
3087 void kmsg_dump(enum kmsg_dump_reason reason)
3089 struct kmsg_dumper *dumper;
3090 unsigned long flags;
3092 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3096 list_for_each_entry_rcu(dumper, &dump_list, list) {
3097 if (dumper->max_reason && reason > dumper->max_reason)
3100 /* initialize iterator with data about the stored records */
3101 dumper->active = true;
3103 logbuf_lock_irqsave(flags);
3104 dumper->cur_seq = clear_seq;
3105 dumper->cur_idx = clear_idx;
3106 dumper->next_seq = log_next_seq;
3107 dumper->next_idx = log_next_idx;
3108 logbuf_unlock_irqrestore(flags);
3110 /* invoke dumper which will iterate over records */
3111 dumper->dump(dumper, reason);
3113 /* reset iterator */
3114 dumper->active = false;
3120 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3121 * @dumper: registered kmsg dumper
3122 * @syslog: include the "<4>" prefixes
3123 * @line: buffer to copy the line to
3124 * @size: maximum size of the buffer
3125 * @len: length of line placed into buffer
3127 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3128 * record, and copy one record into the provided buffer.
3130 * Consecutive calls will return the next available record moving
3131 * towards the end of the buffer with the youngest messages.
3133 * A return value of FALSE indicates that there are no more records to
3136 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3138 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3139 char *line, size_t size, size_t *len)
3141 struct printk_log *msg;
3145 if (!dumper->active)
3148 if (dumper->cur_seq < log_first_seq) {
3149 /* messages are gone, move to first available one */
3150 dumper->cur_seq = log_first_seq;
3151 dumper->cur_idx = log_first_idx;
3155 if (dumper->cur_seq >= log_next_seq)
3158 msg = log_from_idx(dumper->cur_idx);
3159 l = msg_print_text(msg, syslog, printk_time, line, size);
3161 dumper->cur_idx = log_next(dumper->cur_idx);
3171 * kmsg_dump_get_line - retrieve one kmsg log line
3172 * @dumper: registered kmsg dumper
3173 * @syslog: include the "<4>" prefixes
3174 * @line: buffer to copy the line to
3175 * @size: maximum size of the buffer
3176 * @len: length of line placed into buffer
3178 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3179 * record, and copy one record into the provided buffer.
3181 * Consecutive calls will return the next available record moving
3182 * towards the end of the buffer with the youngest messages.
3184 * A return value of FALSE indicates that there are no more records to
3187 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3188 char *line, size_t size, size_t *len)
3190 unsigned long flags;
3193 logbuf_lock_irqsave(flags);
3194 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3195 logbuf_unlock_irqrestore(flags);
3199 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3202 * kmsg_dump_get_buffer - copy kmsg log lines
3203 * @dumper: registered kmsg dumper
3204 * @syslog: include the "<4>" prefixes
3205 * @buf: buffer to copy the line to
3206 * @size: maximum size of the buffer
3207 * @len: length of line placed into buffer
3209 * Start at the end of the kmsg buffer and fill the provided buffer
3210 * with as many of the the *youngest* kmsg records that fit into it.
3211 * If the buffer is large enough, all available kmsg records will be
3212 * copied with a single call.
3214 * Consecutive calls will fill the buffer with the next block of
3215 * available older records, not including the earlier retrieved ones.
3217 * A return value of FALSE indicates that there are no more records to
3220 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3221 char *buf, size_t size, size_t *len)
3223 unsigned long flags;
3230 bool time = printk_time;
3232 if (!dumper->active)
3235 logbuf_lock_irqsave(flags);
3236 if (dumper->cur_seq < log_first_seq) {
3237 /* messages are gone, move to first available one */
3238 dumper->cur_seq = log_first_seq;
3239 dumper->cur_idx = log_first_idx;
3243 if (dumper->cur_seq >= dumper->next_seq) {
3244 logbuf_unlock_irqrestore(flags);
3248 /* calculate length of entire buffer */
3249 seq = dumper->cur_seq;
3250 idx = dumper->cur_idx;
3251 while (seq < dumper->next_seq) {
3252 struct printk_log *msg = log_from_idx(idx);
3254 l += msg_print_text(msg, true, time, NULL, 0);
3255 idx = log_next(idx);
3259 /* move first record forward until length fits into the buffer */
3260 seq = dumper->cur_seq;
3261 idx = dumper->cur_idx;
3262 while (l > size && seq < dumper->next_seq) {
3263 struct printk_log *msg = log_from_idx(idx);
3265 l -= msg_print_text(msg, true, time, NULL, 0);
3266 idx = log_next(idx);
3270 /* last message in next interation */
3275 while (seq < dumper->next_seq) {
3276 struct printk_log *msg = log_from_idx(idx);
3278 l += msg_print_text(msg, syslog, time, buf + l, size - l);
3279 idx = log_next(idx);
3283 dumper->next_seq = next_seq;
3284 dumper->next_idx = next_idx;
3286 logbuf_unlock_irqrestore(flags);
3292 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3295 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3296 * @dumper: registered kmsg dumper
3298 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3299 * kmsg_dump_get_buffer() can be called again and used multiple
3300 * times within the same dumper.dump() callback.
3302 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3304 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3306 dumper->cur_seq = clear_seq;
3307 dumper->cur_idx = clear_idx;
3308 dumper->next_seq = log_next_seq;
3309 dumper->next_idx = log_next_idx;
3313 * kmsg_dump_rewind - reset the interator
3314 * @dumper: registered kmsg dumper
3316 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3317 * kmsg_dump_get_buffer() can be called again and used multiple
3318 * times within the same dumper.dump() callback.
3320 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3322 unsigned long flags;
3324 logbuf_lock_irqsave(flags);
3325 kmsg_dump_rewind_nolock(dumper);
3326 logbuf_unlock_irqrestore(flags);
3328 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);