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 #include <linux/kernel.h>
21 #include <linux/tty.h>
22 #include <linux/tty_driver.h>
23 #include <linux/console.h>
24 #include <linux/init.h>
25 #include <linux/jiffies.h>
26 #include <linux/nmi.h>
27 #include <linux/module.h>
28 #include <linux/moduleparam.h>
29 #include <linux/delay.h>
30 #include <linux/smp.h>
31 #include <linux/security.h>
32 #include <linux/bootmem.h>
33 #include <linux/memblock.h>
34 #include <linux/syscalls.h>
35 #include <linux/crash_core.h>
36 #include <linux/kdb.h>
37 #include <linux/ratelimit.h>
38 #include <linux/kmsg_dump.h>
39 #include <linux/syslog.h>
40 #include <linux/cpu.h>
41 #include <linux/notifier.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);
197 * Number of registered extended console drivers.
199 * If extended consoles are present, in-kernel cont reassembly is disabled
200 * and each fragment is stored as a separate log entry with proper
201 * continuation flag so that every emitted message has full metadata. This
202 * doesn't change the result for regular consoles or /proc/kmsg. For
203 * /dev/kmsg, as long as the reader concatenates messages according to
204 * consecutive continuation flags, the end result should be the same too.
206 static int nr_ext_console_drivers;
209 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
210 * macros instead of functions so that _RET_IP_ contains useful information.
212 #define down_console_sem() do { \
214 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
217 static int __down_trylock_console_sem(unsigned long ip)
223 * Here and in __up_console_sem() we need to be in safe mode,
224 * because spindump/WARN/etc from under console ->lock will
225 * deadlock in printk()->down_trylock_console_sem() otherwise.
227 printk_safe_enter_irqsave(flags);
228 lock_failed = down_trylock(&console_sem);
229 printk_safe_exit_irqrestore(flags);
233 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
236 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
238 static void __up_console_sem(unsigned long ip)
242 mutex_release(&console_lock_dep_map, 1, ip);
244 printk_safe_enter_irqsave(flags);
246 printk_safe_exit_irqrestore(flags);
248 #define up_console_sem() __up_console_sem(_RET_IP_)
251 * This is used for debugging the mess that is the VT code by
252 * keeping track if we have the console semaphore held. It's
253 * definitely not the perfect debug tool (we don't know if _WE_
254 * hold it and are racing, but it helps tracking those weird code
255 * paths in the console code where we end up in places I want
256 * locked without the console sempahore held).
258 static int console_locked, console_suspended;
261 * If exclusive_console is non-NULL then only this console is to be printed to.
263 static struct console *exclusive_console;
266 * Array of consoles built from command line options (console=)
269 #define MAX_CMDLINECONSOLES 8
271 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
273 static int preferred_console = -1;
274 int console_set_on_cmdline;
275 EXPORT_SYMBOL(console_set_on_cmdline);
277 /* Flag: console code may call schedule() */
278 static int console_may_schedule;
280 enum con_msg_format_flags {
281 MSG_FORMAT_DEFAULT = 0,
282 MSG_FORMAT_SYSLOG = (1 << 0),
285 static int console_msg_format = MSG_FORMAT_DEFAULT;
288 * The printk log buffer consists of a chain of concatenated variable
289 * length records. Every record starts with a record header, containing
290 * the overall length of the record.
292 * The heads to the first and last entry in the buffer, as well as the
293 * sequence numbers of these entries are maintained when messages are
296 * If the heads indicate available messages, the length in the header
297 * tells the start next message. A length == 0 for the next message
298 * indicates a wrap-around to the beginning of the buffer.
300 * Every record carries the monotonic timestamp in microseconds, as well as
301 * the standard userspace syslog level and syslog facility. The usual
302 * kernel messages use LOG_KERN; userspace-injected messages always carry
303 * a matching syslog facility, by default LOG_USER. The origin of every
304 * message can be reliably determined that way.
306 * The human readable log message directly follows the message header. The
307 * length of the message text is stored in the header, the stored message
310 * Optionally, a message can carry a dictionary of properties (key/value pairs),
311 * to provide userspace with a machine-readable message context.
313 * Examples for well-defined, commonly used property names are:
314 * DEVICE=b12:8 device identifier
318 * +sound:card0 subsystem:devname
319 * SUBSYSTEM=pci driver-core subsystem name
321 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
322 * follows directly after a '=' character. Every property is terminated by
323 * a '\0' character. The last property is not terminated.
325 * Example of a message structure:
326 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
327 * 0008 34 00 record is 52 bytes long
328 * 000a 0b 00 text is 11 bytes long
329 * 000c 1f 00 dictionary is 23 bytes long
330 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
331 * 0010 69 74 27 73 20 61 20 6c "it's a l"
333 * 001b 44 45 56 49 43 "DEVIC"
334 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
335 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
337 * 0032 00 00 00 padding to next message header
339 * The 'struct printk_log' buffer header must never be directly exported to
340 * userspace, it is a kernel-private implementation detail that might
341 * need to be changed in the future, when the requirements change.
343 * /dev/kmsg exports the structured data in the following line format:
344 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
346 * Users of the export format should ignore possible additional values
347 * separated by ',', and find the message after the ';' character.
349 * The optional key/value pairs are attached as continuation lines starting
350 * with a space character and terminated by a newline. All possible
351 * non-prinatable characters are escaped in the "\xff" notation.
355 LOG_NOCONS = 1, /* suppress print, do not print to console */
356 LOG_NEWLINE = 2, /* text ended with a newline */
357 LOG_PREFIX = 4, /* text started with a prefix */
358 LOG_CONT = 8, /* text is a fragment of a continuation line */
362 u64 ts_nsec; /* timestamp in nanoseconds */
363 u16 len; /* length of entire record */
364 u16 text_len; /* length of text buffer */
365 u16 dict_len; /* length of dictionary buffer */
366 u8 facility; /* syslog facility */
367 u8 flags:5; /* internal record flags */
368 u8 level:3; /* syslog level */
370 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
371 __packed __aligned(4)
376 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
377 * within the scheduler's rq lock. It must be released before calling
378 * console_unlock() or anything else that might wake up a process.
380 DEFINE_RAW_SPINLOCK(logbuf_lock);
383 * Helper macros to lock/unlock logbuf_lock and switch between
384 * printk-safe/unsafe modes.
386 #define logbuf_lock_irq() \
388 printk_safe_enter_irq(); \
389 raw_spin_lock(&logbuf_lock); \
392 #define logbuf_unlock_irq() \
394 raw_spin_unlock(&logbuf_lock); \
395 printk_safe_exit_irq(); \
398 #define logbuf_lock_irqsave(flags) \
400 printk_safe_enter_irqsave(flags); \
401 raw_spin_lock(&logbuf_lock); \
404 #define logbuf_unlock_irqrestore(flags) \
406 raw_spin_unlock(&logbuf_lock); \
407 printk_safe_exit_irqrestore(flags); \
411 DECLARE_WAIT_QUEUE_HEAD(log_wait);
412 /* the next printk record to read by syslog(READ) or /proc/kmsg */
413 static u64 syslog_seq;
414 static u32 syslog_idx;
415 static size_t syslog_partial;
417 /* index and sequence number of the first record stored in the buffer */
418 static u64 log_first_seq;
419 static u32 log_first_idx;
421 /* index and sequence number of the next record to store in the buffer */
422 static u64 log_next_seq;
423 static u32 log_next_idx;
425 /* the next printk record to write to the console */
426 static u64 console_seq;
427 static u32 console_idx;
429 /* the next printk record to read after the last 'clear' command */
430 static u64 clear_seq;
431 static u32 clear_idx;
433 #define PREFIX_MAX 32
434 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
436 #define LOG_LEVEL(v) ((v) & 0x07)
437 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
440 #define LOG_ALIGN __alignof__(struct printk_log)
441 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
442 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
443 static char *log_buf = __log_buf;
444 static u32 log_buf_len = __LOG_BUF_LEN;
446 /* Return log buffer address */
447 char *log_buf_addr_get(void)
452 /* Return log buffer size */
453 u32 log_buf_len_get(void)
458 /* human readable text of the record */
459 static char *log_text(const struct printk_log *msg)
461 return (char *)msg + sizeof(struct printk_log);
464 /* optional key/value pair dictionary attached to the record */
465 static char *log_dict(const struct printk_log *msg)
467 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
470 /* get record by index; idx must point to valid msg */
471 static struct printk_log *log_from_idx(u32 idx)
473 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
476 * A length == 0 record is the end of buffer marker. Wrap around and
477 * read the message at the start of the buffer.
480 return (struct printk_log *)log_buf;
484 /* get next record; idx must point to valid msg */
485 static u32 log_next(u32 idx)
487 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
489 /* length == 0 indicates the end of the buffer; wrap */
491 * A length == 0 record is the end of buffer marker. Wrap around and
492 * read the message at the start of the buffer as *this* one, and
493 * return the one after that.
496 msg = (struct printk_log *)log_buf;
499 return idx + msg->len;
503 * Check whether there is enough free space for the given message.
505 * The same values of first_idx and next_idx mean that the buffer
506 * is either empty or full.
508 * If the buffer is empty, we must respect the position of the indexes.
509 * They cannot be reset to the beginning of the buffer.
511 static int logbuf_has_space(u32 msg_size, bool empty)
515 if (log_next_idx > log_first_idx || empty)
516 free = max(log_buf_len - log_next_idx, log_first_idx);
518 free = log_first_idx - log_next_idx;
521 * We need space also for an empty header that signalizes wrapping
524 return free >= msg_size + sizeof(struct printk_log);
527 static int log_make_free_space(u32 msg_size)
529 while (log_first_seq < log_next_seq &&
530 !logbuf_has_space(msg_size, false)) {
531 /* drop old messages until we have enough contiguous space */
532 log_first_idx = log_next(log_first_idx);
536 if (clear_seq < log_first_seq) {
537 clear_seq = log_first_seq;
538 clear_idx = log_first_idx;
541 /* sequence numbers are equal, so the log buffer is empty */
542 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
548 /* compute the message size including the padding bytes */
549 static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
553 size = sizeof(struct printk_log) + text_len + dict_len;
554 *pad_len = (-size) & (LOG_ALIGN - 1);
561 * Define how much of the log buffer we could take at maximum. The value
562 * must be greater than two. Note that only half of the buffer is available
563 * when the index points to the middle.
565 #define MAX_LOG_TAKE_PART 4
566 static const char trunc_msg[] = "<truncated>";
568 static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
569 u16 *dict_len, u32 *pad_len)
572 * The message should not take the whole buffer. Otherwise, it might
573 * get removed too soon.
575 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
576 if (*text_len > max_text_len)
577 *text_len = max_text_len;
578 /* enable the warning message */
579 *trunc_msg_len = strlen(trunc_msg);
580 /* disable the "dict" completely */
582 /* compute the size again, count also the warning message */
583 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
586 /* insert record into the buffer, discard old ones, update heads */
587 static int log_store(int facility, int level,
588 enum log_flags flags, u64 ts_nsec,
589 const char *dict, u16 dict_len,
590 const char *text, u16 text_len)
592 struct printk_log *msg;
594 u16 trunc_msg_len = 0;
596 /* number of '\0' padding bytes to next message */
597 size = msg_used_size(text_len, dict_len, &pad_len);
599 if (log_make_free_space(size)) {
600 /* truncate the message if it is too long for empty buffer */
601 size = truncate_msg(&text_len, &trunc_msg_len,
602 &dict_len, &pad_len);
603 /* survive when the log buffer is too small for trunc_msg */
604 if (log_make_free_space(size))
608 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
610 * This message + an additional empty header does not fit
611 * at the end of the buffer. Add an empty header with len == 0
612 * to signify a wrap around.
614 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
619 msg = (struct printk_log *)(log_buf + log_next_idx);
620 memcpy(log_text(msg), text, text_len);
621 msg->text_len = text_len;
623 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
624 msg->text_len += trunc_msg_len;
626 memcpy(log_dict(msg), dict, dict_len);
627 msg->dict_len = dict_len;
628 msg->facility = facility;
629 msg->level = level & 7;
630 msg->flags = flags & 0x1f;
632 msg->ts_nsec = ts_nsec;
634 msg->ts_nsec = local_clock();
635 memset(log_dict(msg) + dict_len, 0, pad_len);
639 log_next_idx += msg->len;
642 return msg->text_len;
645 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
647 static int syslog_action_restricted(int type)
652 * Unless restricted, we allow "read all" and "get buffer size"
655 return type != SYSLOG_ACTION_READ_ALL &&
656 type != SYSLOG_ACTION_SIZE_BUFFER;
659 static int check_syslog_permissions(int type, int source)
662 * If this is from /proc/kmsg and we've already opened it, then we've
663 * already done the capabilities checks at open time.
665 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
668 if (syslog_action_restricted(type)) {
669 if (capable(CAP_SYSLOG))
672 * For historical reasons, accept CAP_SYS_ADMIN too, with
675 if (capable(CAP_SYS_ADMIN)) {
676 pr_warn_once("%s (%d): Attempt to access syslog with "
677 "CAP_SYS_ADMIN but no CAP_SYSLOG "
679 current->comm, task_pid_nr(current));
685 return security_syslog(type);
688 static void append_char(char **pp, char *e, char c)
694 static ssize_t msg_print_ext_header(char *buf, size_t size,
695 struct printk_log *msg, u64 seq)
697 u64 ts_usec = msg->ts_nsec;
699 do_div(ts_usec, 1000);
701 return scnprintf(buf, size, "%u,%llu,%llu,%c;",
702 (msg->facility << 3) | msg->level, seq, ts_usec,
703 msg->flags & LOG_CONT ? 'c' : '-');
706 static ssize_t msg_print_ext_body(char *buf, size_t size,
707 char *dict, size_t dict_len,
708 char *text, size_t text_len)
710 char *p = buf, *e = buf + size;
713 /* escape non-printable characters */
714 for (i = 0; i < text_len; i++) {
715 unsigned char c = text[i];
717 if (c < ' ' || c >= 127 || c == '\\')
718 p += scnprintf(p, e - p, "\\x%02x", c);
720 append_char(&p, e, c);
722 append_char(&p, e, '\n');
727 for (i = 0; i < dict_len; i++) {
728 unsigned char c = dict[i];
731 append_char(&p, e, ' ');
736 append_char(&p, e, '\n');
741 if (c < ' ' || c >= 127 || c == '\\') {
742 p += scnprintf(p, e - p, "\\x%02x", c);
746 append_char(&p, e, c);
748 append_char(&p, e, '\n');
754 /* /dev/kmsg - userspace message inject/listen interface */
755 struct devkmsg_user {
758 struct ratelimit_state rs;
760 char buf[CONSOLE_EXT_LOG_MAX];
763 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
766 int level = default_message_loglevel;
767 int facility = 1; /* LOG_USER */
768 struct file *file = iocb->ki_filp;
769 struct devkmsg_user *user = file->private_data;
770 size_t len = iov_iter_count(from);
773 if (!user || len > LOG_LINE_MAX)
776 /* Ignore when user logging is disabled. */
777 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
780 /* Ratelimit when not explicitly enabled. */
781 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
782 if (!___ratelimit(&user->rs, current->comm))
786 buf = kmalloc(len+1, GFP_KERNEL);
791 if (!copy_from_iter_full(buf, len, from)) {
797 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
798 * the decimal value represents 32bit, the lower 3 bit are the log
799 * level, the rest are the log facility.
801 * If no prefix or no userspace facility is specified, we
802 * enforce LOG_USER, to be able to reliably distinguish
803 * kernel-generated messages from userspace-injected ones.
806 if (line[0] == '<') {
810 u = simple_strtoul(line + 1, &endp, 10);
811 if (endp && endp[0] == '>') {
812 level = LOG_LEVEL(u);
813 if (LOG_FACILITY(u) != 0)
814 facility = LOG_FACILITY(u);
821 printk_emit(facility, level, NULL, 0, "%s", line);
826 static ssize_t devkmsg_read(struct file *file, char __user *buf,
827 size_t count, loff_t *ppos)
829 struct devkmsg_user *user = file->private_data;
830 struct printk_log *msg;
837 ret = mutex_lock_interruptible(&user->lock);
842 while (user->seq == log_next_seq) {
843 if (file->f_flags & O_NONBLOCK) {
850 ret = wait_event_interruptible(log_wait,
851 user->seq != log_next_seq);
857 if (user->seq < log_first_seq) {
858 /* our last seen message is gone, return error and reset */
859 user->idx = log_first_idx;
860 user->seq = log_first_seq;
866 msg = log_from_idx(user->idx);
867 len = msg_print_ext_header(user->buf, sizeof(user->buf),
869 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
870 log_dict(msg), msg->dict_len,
871 log_text(msg), msg->text_len);
873 user->idx = log_next(user->idx);
882 if (copy_to_user(buf, user->buf, len)) {
888 mutex_unlock(&user->lock);
892 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
894 struct devkmsg_user *user = file->private_data;
905 /* the first record */
906 user->idx = log_first_idx;
907 user->seq = log_first_seq;
911 * The first record after the last SYSLOG_ACTION_CLEAR,
912 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
913 * changes no global state, and does not clear anything.
915 user->idx = clear_idx;
916 user->seq = clear_seq;
919 /* after the last record */
920 user->idx = log_next_idx;
921 user->seq = log_next_seq;
930 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
932 struct devkmsg_user *user = file->private_data;
936 return EPOLLERR|EPOLLNVAL;
938 poll_wait(file, &log_wait, wait);
941 if (user->seq < log_next_seq) {
942 /* return error when data has vanished underneath us */
943 if (user->seq < log_first_seq)
944 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
946 ret = EPOLLIN|EPOLLRDNORM;
953 static int devkmsg_open(struct inode *inode, struct file *file)
955 struct devkmsg_user *user;
958 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
961 /* write-only does not need any file context */
962 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
963 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
969 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
973 ratelimit_default_init(&user->rs);
974 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
976 mutex_init(&user->lock);
979 user->idx = log_first_idx;
980 user->seq = log_first_seq;
983 file->private_data = user;
987 static int devkmsg_release(struct inode *inode, struct file *file)
989 struct devkmsg_user *user = file->private_data;
994 ratelimit_state_exit(&user->rs);
996 mutex_destroy(&user->lock);
1001 const struct file_operations kmsg_fops = {
1002 .open = devkmsg_open,
1003 .read = devkmsg_read,
1004 .write_iter = devkmsg_write,
1005 .llseek = devkmsg_llseek,
1006 .poll = devkmsg_poll,
1007 .release = devkmsg_release,
1010 #ifdef CONFIG_CRASH_CORE
1012 * This appends the listed symbols to /proc/vmcore
1014 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1015 * obtain access to symbols that are otherwise very difficult to locate. These
1016 * symbols are specifically used so that utilities can access and extract the
1017 * dmesg log from a vmcore file after a crash.
1019 void log_buf_vmcoreinfo_setup(void)
1021 VMCOREINFO_SYMBOL(log_buf);
1022 VMCOREINFO_SYMBOL(log_buf_len);
1023 VMCOREINFO_SYMBOL(log_first_idx);
1024 VMCOREINFO_SYMBOL(clear_idx);
1025 VMCOREINFO_SYMBOL(log_next_idx);
1027 * Export struct printk_log size and field offsets. User space tools can
1028 * parse it and detect any changes to structure down the line.
1030 VMCOREINFO_STRUCT_SIZE(printk_log);
1031 VMCOREINFO_OFFSET(printk_log, ts_nsec);
1032 VMCOREINFO_OFFSET(printk_log, len);
1033 VMCOREINFO_OFFSET(printk_log, text_len);
1034 VMCOREINFO_OFFSET(printk_log, dict_len);
1038 /* requested log_buf_len from kernel cmdline */
1039 static unsigned long __initdata new_log_buf_len;
1041 /* we practice scaling the ring buffer by powers of 2 */
1042 static void __init log_buf_len_update(unsigned size)
1045 size = roundup_pow_of_two(size);
1046 if (size > log_buf_len)
1047 new_log_buf_len = size;
1050 /* save requested log_buf_len since it's too early to process it */
1051 static int __init log_buf_len_setup(char *str)
1053 unsigned size = memparse(str, &str);
1055 log_buf_len_update(size);
1059 early_param("log_buf_len", log_buf_len_setup);
1062 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1064 static void __init log_buf_add_cpu(void)
1066 unsigned int cpu_extra;
1069 * archs should set up cpu_possible_bits properly with
1070 * set_cpu_possible() after setup_arch() but just in
1071 * case lets ensure this is valid.
1073 if (num_possible_cpus() == 1)
1076 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1078 /* by default this will only continue through for large > 64 CPUs */
1079 if (cpu_extra <= __LOG_BUF_LEN / 2)
1082 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1083 __LOG_CPU_MAX_BUF_LEN);
1084 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1086 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1088 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1090 #else /* !CONFIG_SMP */
1091 static inline void log_buf_add_cpu(void) {}
1092 #endif /* CONFIG_SMP */
1094 void __init setup_log_buf(int early)
1096 unsigned long flags;
1100 if (log_buf != __log_buf)
1103 if (!early && !new_log_buf_len)
1106 if (!new_log_buf_len)
1111 memblock_virt_alloc(new_log_buf_len, LOG_ALIGN);
1113 new_log_buf = memblock_virt_alloc_nopanic(new_log_buf_len,
1117 if (unlikely(!new_log_buf)) {
1118 pr_err("log_buf_len: %ld bytes not available\n",
1123 logbuf_lock_irqsave(flags);
1124 log_buf_len = new_log_buf_len;
1125 log_buf = new_log_buf;
1126 new_log_buf_len = 0;
1127 free = __LOG_BUF_LEN - log_next_idx;
1128 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1129 logbuf_unlock_irqrestore(flags);
1131 pr_info("log_buf_len: %d bytes\n", log_buf_len);
1132 pr_info("early log buf free: %d(%d%%)\n",
1133 free, (free * 100) / __LOG_BUF_LEN);
1136 static bool __read_mostly ignore_loglevel;
1138 static int __init ignore_loglevel_setup(char *str)
1140 ignore_loglevel = true;
1141 pr_info("debug: ignoring loglevel setting.\n");
1146 early_param("ignore_loglevel", ignore_loglevel_setup);
1147 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1148 MODULE_PARM_DESC(ignore_loglevel,
1149 "ignore loglevel setting (prints all kernel messages to the console)");
1151 static bool suppress_message_printing(int level)
1153 return (level >= console_loglevel && !ignore_loglevel);
1156 #ifdef CONFIG_BOOT_PRINTK_DELAY
1158 static int boot_delay; /* msecs delay after each printk during bootup */
1159 static unsigned long long loops_per_msec; /* based on boot_delay */
1161 static int __init boot_delay_setup(char *str)
1165 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1166 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1168 get_option(&str, &boot_delay);
1169 if (boot_delay > 10 * 1000)
1172 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1173 "HZ: %d, loops_per_msec: %llu\n",
1174 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1177 early_param("boot_delay", boot_delay_setup);
1179 static void boot_delay_msec(int level)
1181 unsigned long long k;
1182 unsigned long timeout;
1184 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1185 || suppress_message_printing(level)) {
1189 k = (unsigned long long)loops_per_msec * boot_delay;
1191 timeout = jiffies + msecs_to_jiffies(boot_delay);
1196 * use (volatile) jiffies to prevent
1197 * compiler reduction; loop termination via jiffies
1198 * is secondary and may or may not happen.
1200 if (time_after(jiffies, timeout))
1202 touch_nmi_watchdog();
1206 static inline void boot_delay_msec(int level)
1211 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1212 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1214 static size_t print_time(u64 ts, char *buf)
1216 unsigned long rem_nsec;
1221 rem_nsec = do_div(ts, 1000000000);
1224 return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
1226 return sprintf(buf, "[%5lu.%06lu] ",
1227 (unsigned long)ts, rem_nsec / 1000);
1230 static size_t print_prefix(const struct printk_log *msg, bool syslog, char *buf)
1233 unsigned int prefix = (msg->facility << 3) | msg->level;
1237 len += sprintf(buf, "<%u>", prefix);
1242 else if (prefix > 99)
1244 else if (prefix > 9)
1249 len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
1253 static size_t msg_print_text(const struct printk_log *msg, bool syslog, char *buf, size_t size)
1255 const char *text = log_text(msg);
1256 size_t text_size = msg->text_len;
1260 const char *next = memchr(text, '\n', text_size);
1264 text_len = next - text;
1266 text_size -= next - text;
1268 text_len = text_size;
1272 if (print_prefix(msg, syslog, NULL) +
1273 text_len + 1 >= size - len)
1276 len += print_prefix(msg, syslog, buf + len);
1277 memcpy(buf + len, text, text_len);
1281 /* SYSLOG_ACTION_* buffer size only calculation */
1282 len += print_prefix(msg, syslog, NULL);
1293 static int syslog_print(char __user *buf, int size)
1296 struct printk_log *msg;
1299 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1308 if (syslog_seq < log_first_seq) {
1309 /* messages are gone, move to first one */
1310 syslog_seq = log_first_seq;
1311 syslog_idx = log_first_idx;
1314 if (syslog_seq == log_next_seq) {
1315 logbuf_unlock_irq();
1319 skip = syslog_partial;
1320 msg = log_from_idx(syslog_idx);
1321 n = msg_print_text(msg, true, text, LOG_LINE_MAX + PREFIX_MAX);
1322 if (n - syslog_partial <= size) {
1323 /* message fits into buffer, move forward */
1324 syslog_idx = log_next(syslog_idx);
1326 n -= syslog_partial;
1329 /* partial read(), remember position */
1331 syslog_partial += n;
1334 logbuf_unlock_irq();
1339 if (copy_to_user(buf, text + skip, n)) {
1354 static int syslog_print_all(char __user *buf, int size, bool clear)
1362 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1368 * Find first record that fits, including all following records,
1369 * into the user-provided buffer for this dump.
1373 while (seq < log_next_seq) {
1374 struct printk_log *msg = log_from_idx(idx);
1376 len += msg_print_text(msg, true, NULL, 0);
1377 idx = log_next(idx);
1381 /* move first record forward until length fits into the buffer */
1384 while (len > size && seq < log_next_seq) {
1385 struct printk_log *msg = log_from_idx(idx);
1387 len -= msg_print_text(msg, true, NULL, 0);
1388 idx = log_next(idx);
1392 /* last message fitting into this dump */
1393 next_seq = log_next_seq;
1396 while (len >= 0 && seq < next_seq) {
1397 struct printk_log *msg = log_from_idx(idx);
1400 textlen = msg_print_text(msg, true, text,
1401 LOG_LINE_MAX + PREFIX_MAX);
1406 idx = log_next(idx);
1409 logbuf_unlock_irq();
1410 if (copy_to_user(buf + len, text, textlen))
1416 if (seq < log_first_seq) {
1417 /* messages are gone, move to next one */
1418 seq = log_first_seq;
1419 idx = log_first_idx;
1424 clear_seq = log_next_seq;
1425 clear_idx = log_next_idx;
1427 logbuf_unlock_irq();
1433 static void syslog_clear(void)
1436 clear_seq = log_next_seq;
1437 clear_idx = log_next_idx;
1438 logbuf_unlock_irq();
1441 int do_syslog(int type, char __user *buf, int len, int source)
1444 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1447 error = check_syslog_permissions(type, source);
1452 case SYSLOG_ACTION_CLOSE: /* Close log */
1454 case SYSLOG_ACTION_OPEN: /* Open log */
1456 case SYSLOG_ACTION_READ: /* Read from log */
1457 if (!buf || len < 0)
1461 if (!access_ok(VERIFY_WRITE, buf, len))
1463 error = wait_event_interruptible(log_wait,
1464 syslog_seq != log_next_seq);
1467 error = syslog_print(buf, len);
1469 /* Read/clear last kernel messages */
1470 case SYSLOG_ACTION_READ_CLEAR:
1473 /* Read last kernel messages */
1474 case SYSLOG_ACTION_READ_ALL:
1475 if (!buf || len < 0)
1479 if (!access_ok(VERIFY_WRITE, buf, len))
1481 error = syslog_print_all(buf, len, clear);
1483 /* Clear ring buffer */
1484 case SYSLOG_ACTION_CLEAR:
1487 /* Disable logging to console */
1488 case SYSLOG_ACTION_CONSOLE_OFF:
1489 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1490 saved_console_loglevel = console_loglevel;
1491 console_loglevel = minimum_console_loglevel;
1493 /* Enable logging to console */
1494 case SYSLOG_ACTION_CONSOLE_ON:
1495 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1496 console_loglevel = saved_console_loglevel;
1497 saved_console_loglevel = LOGLEVEL_DEFAULT;
1500 /* Set level of messages printed to console */
1501 case SYSLOG_ACTION_CONSOLE_LEVEL:
1502 if (len < 1 || len > 8)
1504 if (len < minimum_console_loglevel)
1505 len = minimum_console_loglevel;
1506 console_loglevel = len;
1507 /* Implicitly re-enable logging to console */
1508 saved_console_loglevel = LOGLEVEL_DEFAULT;
1510 /* Number of chars in the log buffer */
1511 case SYSLOG_ACTION_SIZE_UNREAD:
1513 if (syslog_seq < log_first_seq) {
1514 /* messages are gone, move to first one */
1515 syslog_seq = log_first_seq;
1516 syslog_idx = log_first_idx;
1519 if (source == SYSLOG_FROM_PROC) {
1521 * Short-cut for poll(/"proc/kmsg") which simply checks
1522 * for pending data, not the size; return the count of
1523 * records, not the length.
1525 error = log_next_seq - syslog_seq;
1527 u64 seq = syslog_seq;
1528 u32 idx = syslog_idx;
1530 while (seq < log_next_seq) {
1531 struct printk_log *msg = log_from_idx(idx);
1533 error += msg_print_text(msg, true, NULL, 0);
1534 idx = log_next(idx);
1537 error -= syslog_partial;
1539 logbuf_unlock_irq();
1541 /* Size of the log buffer */
1542 case SYSLOG_ACTION_SIZE_BUFFER:
1543 error = log_buf_len;
1553 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1555 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1559 * Special console_lock variants that help to reduce the risk of soft-lockups.
1560 * They allow to pass console_lock to another printk() call using a busy wait.
1563 #ifdef CONFIG_LOCKDEP
1564 static struct lockdep_map console_owner_dep_map = {
1565 .name = "console_owner"
1569 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1570 static struct task_struct *console_owner;
1571 static bool console_waiter;
1574 * console_lock_spinning_enable - mark beginning of code where another
1575 * thread might safely busy wait
1577 * This basically converts console_lock into a spinlock. This marks
1578 * the section where the console_lock owner can not sleep, because
1579 * there may be a waiter spinning (like a spinlock). Also it must be
1580 * ready to hand over the lock at the end of the section.
1582 static void console_lock_spinning_enable(void)
1584 raw_spin_lock(&console_owner_lock);
1585 console_owner = current;
1586 raw_spin_unlock(&console_owner_lock);
1588 /* The waiter may spin on us after setting console_owner */
1589 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1593 * console_lock_spinning_disable_and_check - mark end of code where another
1594 * thread was able to busy wait and check if there is a waiter
1596 * This is called at the end of the section where spinning is allowed.
1597 * It has two functions. First, it is a signal that it is no longer
1598 * safe to start busy waiting for the lock. Second, it checks if
1599 * there is a busy waiter and passes the lock rights to her.
1601 * Important: Callers lose the lock if there was a busy waiter.
1602 * They must not touch items synchronized by console_lock
1605 * Return: 1 if the lock rights were passed, 0 otherwise.
1607 static int console_lock_spinning_disable_and_check(void)
1611 raw_spin_lock(&console_owner_lock);
1612 waiter = READ_ONCE(console_waiter);
1613 console_owner = NULL;
1614 raw_spin_unlock(&console_owner_lock);
1617 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1621 /* The waiter is now free to continue */
1622 WRITE_ONCE(console_waiter, false);
1624 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1627 * Hand off console_lock to waiter. The waiter will perform
1628 * the up(). After this, the waiter is the console_lock owner.
1630 mutex_release(&console_lock_dep_map, 1, _THIS_IP_);
1635 * console_trylock_spinning - try to get console_lock by busy waiting
1637 * This allows to busy wait for the console_lock when the current
1638 * owner is running in specially marked sections. It means that
1639 * the current owner is running and cannot reschedule until it
1640 * is ready to lose the lock.
1642 * Return: 1 if we got the lock, 0 othrewise
1644 static int console_trylock_spinning(void)
1646 struct task_struct *owner = NULL;
1649 unsigned long flags;
1651 if (console_trylock())
1654 printk_safe_enter_irqsave(flags);
1656 raw_spin_lock(&console_owner_lock);
1657 owner = READ_ONCE(console_owner);
1658 waiter = READ_ONCE(console_waiter);
1659 if (!waiter && owner && owner != current) {
1660 WRITE_ONCE(console_waiter, true);
1663 raw_spin_unlock(&console_owner_lock);
1666 * If there is an active printk() writing to the
1667 * consoles, instead of having it write our data too,
1668 * see if we can offload that load from the active
1669 * printer, and do some printing ourselves.
1670 * Go into a spin only if there isn't already a waiter
1671 * spinning, and there is an active printer, and
1672 * that active printer isn't us (recursive printk?).
1675 printk_safe_exit_irqrestore(flags);
1679 /* We spin waiting for the owner to release us */
1680 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1681 /* Owner will clear console_waiter on hand off */
1682 while (READ_ONCE(console_waiter))
1684 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1686 printk_safe_exit_irqrestore(flags);
1688 * The owner passed the console lock to us.
1689 * Since we did not spin on console lock, annotate
1690 * this as a trylock. Otherwise lockdep will
1693 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1699 * Call the console drivers, asking them to write out
1700 * log_buf[start] to log_buf[end - 1].
1701 * The console_lock must be held.
1703 static void call_console_drivers(const char *ext_text, size_t ext_len,
1704 const char *text, size_t len)
1706 struct console *con;
1708 trace_console_rcuidle(text, len);
1710 if (!console_drivers)
1713 for_each_console(con) {
1714 if (exclusive_console && con != exclusive_console)
1716 if (!(con->flags & CON_ENABLED))
1720 if (!cpu_online(smp_processor_id()) &&
1721 !(con->flags & CON_ANYTIME))
1723 if (con->flags & CON_EXTENDED)
1724 con->write(con, ext_text, ext_len);
1726 con->write(con, text, len);
1730 int printk_delay_msec __read_mostly;
1732 static inline void printk_delay(void)
1734 if (unlikely(printk_delay_msec)) {
1735 int m = printk_delay_msec;
1739 touch_nmi_watchdog();
1745 * Continuation lines are buffered, and not committed to the record buffer
1746 * until the line is complete, or a race forces it. The line fragments
1747 * though, are printed immediately to the consoles to ensure everything has
1748 * reached the console in case of a kernel crash.
1750 static struct cont {
1751 char buf[LOG_LINE_MAX];
1752 size_t len; /* length == 0 means unused buffer */
1753 struct task_struct *owner; /* task of first print*/
1754 u64 ts_nsec; /* time of first print */
1755 u8 level; /* log level of first message */
1756 u8 facility; /* log facility of first message */
1757 enum log_flags flags; /* prefix, newline flags */
1760 static void cont_flush(void)
1765 log_store(cont.facility, cont.level, cont.flags, cont.ts_nsec,
1766 NULL, 0, cont.buf, cont.len);
1770 static bool cont_add(int facility, int level, enum log_flags flags, const char *text, size_t len)
1773 * If ext consoles are present, flush and skip in-kernel
1774 * continuation. See nr_ext_console_drivers definition. Also, if
1775 * the line gets too long, split it up in separate records.
1777 if (nr_ext_console_drivers || cont.len + len > sizeof(cont.buf)) {
1783 cont.facility = facility;
1785 cont.owner = current;
1786 cont.ts_nsec = local_clock();
1790 memcpy(cont.buf + cont.len, text, len);
1793 // The original flags come from the first line,
1794 // but later continuations can add a newline.
1795 if (flags & LOG_NEWLINE) {
1796 cont.flags |= LOG_NEWLINE;
1800 if (cont.len > (sizeof(cont.buf) * 80) / 100)
1806 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)
1809 * If an earlier line was buffered, and we're a continuation
1810 * write from the same process, try to add it to the buffer.
1813 if (cont.owner == current && (lflags & LOG_CONT)) {
1814 if (cont_add(facility, level, lflags, text, text_len))
1817 /* Otherwise, make sure it's flushed */
1821 /* Skip empty continuation lines that couldn't be added - they just flush */
1822 if (!text_len && (lflags & LOG_CONT))
1825 /* If it doesn't end in a newline, try to buffer the current line */
1826 if (!(lflags & LOG_NEWLINE)) {
1827 if (cont_add(facility, level, lflags, text, text_len))
1831 /* Store it in the record log */
1832 return log_store(facility, level, lflags, 0, dict, dictlen, text, text_len);
1835 /* Must be called under logbuf_lock. */
1836 int vprintk_store(int facility, int level,
1837 const char *dict, size_t dictlen,
1838 const char *fmt, va_list args)
1840 static char textbuf[LOG_LINE_MAX];
1841 char *text = textbuf;
1843 enum log_flags lflags = 0;
1846 * The printf needs to come first; we need the syslog
1847 * prefix which might be passed-in as a parameter.
1849 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1851 /* mark and strip a trailing newline */
1852 if (text_len && text[text_len-1] == '\n') {
1854 lflags |= LOG_NEWLINE;
1857 /* strip kernel syslog prefix and extract log level or control flags */
1858 if (facility == 0) {
1861 while ((kern_level = printk_get_level(text)) != 0) {
1862 switch (kern_level) {
1864 if (level == LOGLEVEL_DEFAULT)
1865 level = kern_level - '0';
1867 case 'd': /* KERN_DEFAULT */
1868 lflags |= LOG_PREFIX;
1870 case 'c': /* KERN_CONT */
1879 if (level == LOGLEVEL_DEFAULT)
1880 level = default_message_loglevel;
1883 lflags |= LOG_PREFIX|LOG_NEWLINE;
1885 if (suppress_message_printing(level))
1886 lflags |= LOG_NOCONS;
1888 return log_output(facility, level, lflags,
1889 dict, dictlen, text, text_len);
1892 asmlinkage int vprintk_emit(int facility, int level,
1893 const char *dict, size_t dictlen,
1894 const char *fmt, va_list args)
1897 bool in_sched = false;
1898 unsigned long flags;
1900 if (level == LOGLEVEL_SCHED) {
1901 level = LOGLEVEL_DEFAULT;
1905 boot_delay_msec(level);
1908 /* This stops the holder of console_sem just where we want him */
1909 logbuf_lock_irqsave(flags);
1910 printed_len = vprintk_store(facility, level, dict, dictlen, fmt, args);
1911 logbuf_unlock_irqrestore(flags);
1913 /* If called from the scheduler, we can not call up(). */
1916 * Disable preemption to avoid being preempted while holding
1917 * console_sem which would prevent anyone from printing to
1922 * Try to acquire and then immediately release the console
1923 * semaphore. The release will print out buffers and wake up
1924 * /dev/kmsg and syslog() users.
1926 if (console_trylock_spinning())
1934 EXPORT_SYMBOL(vprintk_emit);
1936 asmlinkage int vprintk(const char *fmt, va_list args)
1938 return vprintk_func(fmt, args);
1940 EXPORT_SYMBOL(vprintk);
1942 asmlinkage int printk_emit(int facility, int level,
1943 const char *dict, size_t dictlen,
1944 const char *fmt, ...)
1949 va_start(args, fmt);
1950 r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1955 EXPORT_SYMBOL(printk_emit);
1957 int vprintk_default(const char *fmt, va_list args)
1961 #ifdef CONFIG_KGDB_KDB
1962 /* Allow to pass printk() to kdb but avoid a recursion. */
1963 if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
1964 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
1968 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1972 EXPORT_SYMBOL_GPL(vprintk_default);
1975 * printk - print a kernel message
1976 * @fmt: format string
1978 * This is printk(). It can be called from any context. We want it to work.
1980 * We try to grab the console_lock. If we succeed, it's easy - we log the
1981 * output and call the console drivers. If we fail to get the semaphore, we
1982 * place the output into the log buffer and return. The current holder of
1983 * the console_sem will notice the new output in console_unlock(); and will
1984 * send it to the consoles before releasing the lock.
1986 * One effect of this deferred printing is that code which calls printk() and
1987 * then changes console_loglevel may break. This is because console_loglevel
1988 * is inspected when the actual printing occurs.
1993 * See the vsnprintf() documentation for format string extensions over C99.
1995 asmlinkage __visible int printk(const char *fmt, ...)
2000 va_start(args, fmt);
2001 r = vprintk_func(fmt, args);
2006 EXPORT_SYMBOL(printk);
2008 #else /* CONFIG_PRINTK */
2010 #define LOG_LINE_MAX 0
2011 #define PREFIX_MAX 0
2013 static u64 syslog_seq;
2014 static u32 syslog_idx;
2015 static u64 console_seq;
2016 static u32 console_idx;
2017 static u64 log_first_seq;
2018 static u32 log_first_idx;
2019 static u64 log_next_seq;
2020 static char *log_text(const struct printk_log *msg) { return NULL; }
2021 static char *log_dict(const struct printk_log *msg) { return NULL; }
2022 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2023 static u32 log_next(u32 idx) { return 0; }
2024 static ssize_t msg_print_ext_header(char *buf, size_t size,
2025 struct printk_log *msg,
2026 u64 seq) { return 0; }
2027 static ssize_t msg_print_ext_body(char *buf, size_t size,
2028 char *dict, size_t dict_len,
2029 char *text, size_t text_len) { return 0; }
2030 static void console_lock_spinning_enable(void) { }
2031 static int console_lock_spinning_disable_and_check(void) { return 0; }
2032 static void call_console_drivers(const char *ext_text, size_t ext_len,
2033 const char *text, size_t len) {}
2034 static size_t msg_print_text(const struct printk_log *msg,
2035 bool syslog, char *buf, size_t size) { return 0; }
2037 #endif /* CONFIG_PRINTK */
2039 #ifdef CONFIG_EARLY_PRINTK
2040 struct console *early_console;
2042 asmlinkage __visible void early_printk(const char *fmt, ...)
2052 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2055 early_console->write(early_console, buf, n);
2059 static int __add_preferred_console(char *name, int idx, char *options,
2062 struct console_cmdline *c;
2066 * See if this tty is not yet registered, and
2067 * if we have a slot free.
2069 for (i = 0, c = console_cmdline;
2070 i < MAX_CMDLINECONSOLES && c->name[0];
2072 if (strcmp(c->name, name) == 0 && c->index == idx) {
2074 preferred_console = i;
2078 if (i == MAX_CMDLINECONSOLES)
2081 preferred_console = i;
2082 strlcpy(c->name, name, sizeof(c->name));
2083 c->options = options;
2084 braille_set_options(c, brl_options);
2090 static int __init console_msg_format_setup(char *str)
2092 if (!strcmp(str, "syslog"))
2093 console_msg_format = MSG_FORMAT_SYSLOG;
2094 if (!strcmp(str, "default"))
2095 console_msg_format = MSG_FORMAT_DEFAULT;
2098 __setup("console_msg_format=", console_msg_format_setup);
2101 * Set up a console. Called via do_early_param() in init/main.c
2102 * for each "console=" parameter in the boot command line.
2104 static int __init console_setup(char *str)
2106 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2107 char *s, *options, *brl_options = NULL;
2110 if (_braille_console_setup(&str, &brl_options))
2114 * Decode str into name, index, options.
2116 if (str[0] >= '0' && str[0] <= '9') {
2117 strcpy(buf, "ttyS");
2118 strncpy(buf + 4, str, sizeof(buf) - 5);
2120 strncpy(buf, str, sizeof(buf) - 1);
2122 buf[sizeof(buf) - 1] = 0;
2123 options = strchr(str, ',');
2127 if (!strcmp(str, "ttya"))
2128 strcpy(buf, "ttyS0");
2129 if (!strcmp(str, "ttyb"))
2130 strcpy(buf, "ttyS1");
2132 for (s = buf; *s; s++)
2133 if (isdigit(*s) || *s == ',')
2135 idx = simple_strtoul(s, NULL, 10);
2138 __add_preferred_console(buf, idx, options, brl_options);
2139 console_set_on_cmdline = 1;
2142 __setup("console=", console_setup);
2145 * add_preferred_console - add a device to the list of preferred consoles.
2146 * @name: device name
2147 * @idx: device index
2148 * @options: options for this console
2150 * The last preferred console added will be used for kernel messages
2151 * and stdin/out/err for init. Normally this is used by console_setup
2152 * above to handle user-supplied console arguments; however it can also
2153 * be used by arch-specific code either to override the user or more
2154 * commonly to provide a default console (ie from PROM variables) when
2155 * the user has not supplied one.
2157 int add_preferred_console(char *name, int idx, char *options)
2159 return __add_preferred_console(name, idx, options, NULL);
2162 bool console_suspend_enabled = true;
2163 EXPORT_SYMBOL(console_suspend_enabled);
2165 static int __init console_suspend_disable(char *str)
2167 console_suspend_enabled = false;
2170 __setup("no_console_suspend", console_suspend_disable);
2171 module_param_named(console_suspend, console_suspend_enabled,
2172 bool, S_IRUGO | S_IWUSR);
2173 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2174 " and hibernate operations");
2177 * suspend_console - suspend the console subsystem
2179 * This disables printk() while we go into suspend states
2181 void suspend_console(void)
2183 if (!console_suspend_enabled)
2185 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2187 console_suspended = 1;
2191 void resume_console(void)
2193 if (!console_suspend_enabled)
2196 console_suspended = 0;
2201 * console_cpu_notify - print deferred console messages after CPU hotplug
2204 * If printk() is called from a CPU that is not online yet, the messages
2205 * will be printed on the console only if there are CON_ANYTIME consoles.
2206 * This function is called when a new CPU comes online (or fails to come
2207 * up) or goes offline.
2209 static int console_cpu_notify(unsigned int cpu)
2211 if (!cpuhp_tasks_frozen) {
2212 /* If trylock fails, someone else is doing the printing */
2213 if (console_trylock())
2220 * console_lock - lock the console system for exclusive use.
2222 * Acquires a lock which guarantees that the caller has
2223 * exclusive access to the console system and the console_drivers list.
2225 * Can sleep, returns nothing.
2227 void console_lock(void)
2232 if (console_suspended)
2235 console_may_schedule = 1;
2237 EXPORT_SYMBOL(console_lock);
2240 * console_trylock - try to lock the console system for exclusive use.
2242 * Try to acquire a lock which guarantees that the caller has exclusive
2243 * access to the console system and the console_drivers list.
2245 * returns 1 on success, and 0 on failure to acquire the lock.
2247 int console_trylock(void)
2249 if (down_trylock_console_sem())
2251 if (console_suspended) {
2256 console_may_schedule = 0;
2259 EXPORT_SYMBOL(console_trylock);
2261 int is_console_locked(void)
2263 return console_locked;
2265 EXPORT_SYMBOL(is_console_locked);
2268 * Check if we have any console that is capable of printing while cpu is
2269 * booting or shutting down. Requires console_sem.
2271 static int have_callable_console(void)
2273 struct console *con;
2275 for_each_console(con)
2276 if ((con->flags & CON_ENABLED) &&
2277 (con->flags & CON_ANYTIME))
2284 * Can we actually use the console at this time on this cpu?
2286 * Console drivers may assume that per-cpu resources have been allocated. So
2287 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2288 * call them until this CPU is officially up.
2290 static inline int can_use_console(void)
2292 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2296 * console_unlock - unlock the console system
2298 * Releases the console_lock which the caller holds on the console system
2299 * and the console driver list.
2301 * While the console_lock was held, console output may have been buffered
2302 * by printk(). If this is the case, console_unlock(); emits
2303 * the output prior to releasing the lock.
2305 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2307 * console_unlock(); may be called from any context.
2309 void console_unlock(void)
2311 static char ext_text[CONSOLE_EXT_LOG_MAX];
2312 static char text[LOG_LINE_MAX + PREFIX_MAX];
2313 unsigned long flags;
2314 bool do_cond_resched, retry;
2316 if (console_suspended) {
2322 * Console drivers are called with interrupts disabled, so
2323 * @console_may_schedule should be cleared before; however, we may
2324 * end up dumping a lot of lines, for example, if called from
2325 * console registration path, and should invoke cond_resched()
2326 * between lines if allowable. Not doing so can cause a very long
2327 * scheduling stall on a slow console leading to RCU stall and
2328 * softlockup warnings which exacerbate the issue with more
2329 * messages practically incapacitating the system.
2331 * console_trylock() is not able to detect the preemptive
2332 * context reliably. Therefore the value must be stored before
2333 * and cleared after the the "again" goto label.
2335 do_cond_resched = console_may_schedule;
2337 console_may_schedule = 0;
2340 * We released the console_sem lock, so we need to recheck if
2341 * cpu is online and (if not) is there at least one CON_ANYTIME
2344 if (!can_use_console()) {
2351 struct printk_log *msg;
2355 printk_safe_enter_irqsave(flags);
2356 raw_spin_lock(&logbuf_lock);
2357 if (console_seq < log_first_seq) {
2358 len = sprintf(text, "** %u printk messages dropped **\n",
2359 (unsigned)(log_first_seq - console_seq));
2361 /* messages are gone, move to first one */
2362 console_seq = log_first_seq;
2363 console_idx = log_first_idx;
2368 if (console_seq == log_next_seq)
2371 msg = log_from_idx(console_idx);
2372 if (msg->flags & LOG_NOCONS) {
2374 * Skip record if !ignore_loglevel, and
2375 * record has level above the console loglevel.
2377 console_idx = log_next(console_idx);
2382 len += msg_print_text(msg,
2383 console_msg_format & MSG_FORMAT_SYSLOG,
2385 sizeof(text) - len);
2386 if (nr_ext_console_drivers) {
2387 ext_len = msg_print_ext_header(ext_text,
2390 ext_len += msg_print_ext_body(ext_text + ext_len,
2391 sizeof(ext_text) - ext_len,
2392 log_dict(msg), msg->dict_len,
2393 log_text(msg), msg->text_len);
2395 console_idx = log_next(console_idx);
2397 raw_spin_unlock(&logbuf_lock);
2400 * While actively printing out messages, if another printk()
2401 * were to occur on another CPU, it may wait for this one to
2402 * finish. This task can not be preempted if there is a
2403 * waiter waiting to take over.
2405 console_lock_spinning_enable();
2407 stop_critical_timings(); /* don't trace print latency */
2408 call_console_drivers(ext_text, ext_len, text, len);
2409 start_critical_timings();
2411 if (console_lock_spinning_disable_and_check()) {
2412 printk_safe_exit_irqrestore(flags);
2416 printk_safe_exit_irqrestore(flags);
2418 if (do_cond_resched)
2424 /* Release the exclusive_console once it is used */
2425 if (unlikely(exclusive_console))
2426 exclusive_console = NULL;
2428 raw_spin_unlock(&logbuf_lock);
2433 * Someone could have filled up the buffer again, so re-check if there's
2434 * something to flush. In case we cannot trylock the console_sem again,
2435 * there's a new owner and the console_unlock() from them will do the
2436 * flush, no worries.
2438 raw_spin_lock(&logbuf_lock);
2439 retry = console_seq != log_next_seq;
2440 raw_spin_unlock(&logbuf_lock);
2441 printk_safe_exit_irqrestore(flags);
2443 if (retry && console_trylock())
2446 EXPORT_SYMBOL(console_unlock);
2449 * console_conditional_schedule - yield the CPU if required
2451 * If the console code is currently allowed to sleep, and
2452 * if this CPU should yield the CPU to another task, do
2455 * Must be called within console_lock();.
2457 void __sched console_conditional_schedule(void)
2459 if (console_may_schedule)
2462 EXPORT_SYMBOL(console_conditional_schedule);
2464 void console_unblank(void)
2469 * console_unblank can no longer be called in interrupt context unless
2470 * oops_in_progress is set to 1..
2472 if (oops_in_progress) {
2473 if (down_trylock_console_sem() != 0)
2479 console_may_schedule = 0;
2481 if ((c->flags & CON_ENABLED) && c->unblank)
2487 * console_flush_on_panic - flush console content on panic
2489 * Immediately output all pending messages no matter what.
2491 void console_flush_on_panic(void)
2494 * If someone else is holding the console lock, trylock will fail
2495 * and may_schedule may be set. Ignore and proceed to unlock so
2496 * that messages are flushed out. As this can be called from any
2497 * context and we don't want to get preempted while flushing,
2498 * ensure may_schedule is cleared.
2501 console_may_schedule = 0;
2506 * Return the console tty driver structure and its associated index
2508 struct tty_driver *console_device(int *index)
2511 struct tty_driver *driver = NULL;
2514 for_each_console(c) {
2517 driver = c->device(c, index);
2526 * Prevent further output on the passed console device so that (for example)
2527 * serial drivers can disable console output before suspending a port, and can
2528 * re-enable output afterwards.
2530 void console_stop(struct console *console)
2533 console->flags &= ~CON_ENABLED;
2536 EXPORT_SYMBOL(console_stop);
2538 void console_start(struct console *console)
2541 console->flags |= CON_ENABLED;
2544 EXPORT_SYMBOL(console_start);
2546 static int __read_mostly keep_bootcon;
2548 static int __init keep_bootcon_setup(char *str)
2551 pr_info("debug: skip boot console de-registration.\n");
2556 early_param("keep_bootcon", keep_bootcon_setup);
2559 * The console driver calls this routine during kernel initialization
2560 * to register the console printing procedure with printk() and to
2561 * print any messages that were printed by the kernel before the
2562 * console driver was initialized.
2564 * This can happen pretty early during the boot process (because of
2565 * early_printk) - sometimes before setup_arch() completes - be careful
2566 * of what kernel features are used - they may not be initialised yet.
2568 * There are two types of consoles - bootconsoles (early_printk) and
2569 * "real" consoles (everything which is not a bootconsole) which are
2570 * handled differently.
2571 * - Any number of bootconsoles can be registered at any time.
2572 * - As soon as a "real" console is registered, all bootconsoles
2573 * will be unregistered automatically.
2574 * - Once a "real" console is registered, any attempt to register a
2575 * bootconsoles will be rejected
2577 void register_console(struct console *newcon)
2580 unsigned long flags;
2581 struct console *bcon = NULL;
2582 struct console_cmdline *c;
2583 static bool has_preferred;
2585 if (console_drivers)
2586 for_each_console(bcon)
2587 if (WARN(bcon == newcon,
2588 "console '%s%d' already registered\n",
2589 bcon->name, bcon->index))
2593 * before we register a new CON_BOOT console, make sure we don't
2594 * already have a valid console
2596 if (console_drivers && newcon->flags & CON_BOOT) {
2597 /* find the last or real console */
2598 for_each_console(bcon) {
2599 if (!(bcon->flags & CON_BOOT)) {
2600 pr_info("Too late to register bootconsole %s%d\n",
2601 newcon->name, newcon->index);
2607 if (console_drivers && console_drivers->flags & CON_BOOT)
2608 bcon = console_drivers;
2610 if (!has_preferred || bcon || !console_drivers)
2611 has_preferred = preferred_console >= 0;
2614 * See if we want to use this console driver. If we
2615 * didn't select a console we take the first one
2616 * that registers here.
2618 if (!has_preferred) {
2619 if (newcon->index < 0)
2621 if (newcon->setup == NULL ||
2622 newcon->setup(newcon, NULL) == 0) {
2623 newcon->flags |= CON_ENABLED;
2624 if (newcon->device) {
2625 newcon->flags |= CON_CONSDEV;
2626 has_preferred = true;
2632 * See if this console matches one we selected on
2635 for (i = 0, c = console_cmdline;
2636 i < MAX_CMDLINECONSOLES && c->name[0];
2638 if (!newcon->match ||
2639 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2640 /* default matching */
2641 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2642 if (strcmp(c->name, newcon->name) != 0)
2644 if (newcon->index >= 0 &&
2645 newcon->index != c->index)
2647 if (newcon->index < 0)
2648 newcon->index = c->index;
2650 if (_braille_register_console(newcon, c))
2653 if (newcon->setup &&
2654 newcon->setup(newcon, c->options) != 0)
2658 newcon->flags |= CON_ENABLED;
2659 if (i == preferred_console) {
2660 newcon->flags |= CON_CONSDEV;
2661 has_preferred = true;
2666 if (!(newcon->flags & CON_ENABLED))
2670 * If we have a bootconsole, and are switching to a real console,
2671 * don't print everything out again, since when the boot console, and
2672 * the real console are the same physical device, it's annoying to
2673 * see the beginning boot messages twice
2675 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2676 newcon->flags &= ~CON_PRINTBUFFER;
2679 * Put this console in the list - keep the
2680 * preferred driver at the head of the list.
2683 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2684 newcon->next = console_drivers;
2685 console_drivers = newcon;
2687 newcon->next->flags &= ~CON_CONSDEV;
2689 newcon->next = console_drivers->next;
2690 console_drivers->next = newcon;
2693 if (newcon->flags & CON_EXTENDED)
2694 if (!nr_ext_console_drivers++)
2695 pr_info("printk: continuation disabled due to ext consoles, expect more fragments in /dev/kmsg\n");
2697 if (newcon->flags & CON_PRINTBUFFER) {
2699 * console_unlock(); will print out the buffered messages
2702 logbuf_lock_irqsave(flags);
2703 console_seq = syslog_seq;
2704 console_idx = syslog_idx;
2705 logbuf_unlock_irqrestore(flags);
2707 * We're about to replay the log buffer. Only do this to the
2708 * just-registered console to avoid excessive message spam to
2709 * the already-registered consoles.
2711 exclusive_console = newcon;
2714 console_sysfs_notify();
2717 * By unregistering the bootconsoles after we enable the real console
2718 * we get the "console xxx enabled" message on all the consoles -
2719 * boot consoles, real consoles, etc - this is to ensure that end
2720 * users know there might be something in the kernel's log buffer that
2721 * went to the bootconsole (that they do not see on the real console)
2723 pr_info("%sconsole [%s%d] enabled\n",
2724 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2725 newcon->name, newcon->index);
2727 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2729 /* We need to iterate through all boot consoles, to make
2730 * sure we print everything out, before we unregister them.
2732 for_each_console(bcon)
2733 if (bcon->flags & CON_BOOT)
2734 unregister_console(bcon);
2737 EXPORT_SYMBOL(register_console);
2739 int unregister_console(struct console *console)
2741 struct console *a, *b;
2744 pr_info("%sconsole [%s%d] disabled\n",
2745 (console->flags & CON_BOOT) ? "boot" : "" ,
2746 console->name, console->index);
2748 res = _braille_unregister_console(console);
2754 if (console_drivers == console) {
2755 console_drivers=console->next;
2757 } else if (console_drivers) {
2758 for (a=console_drivers->next, b=console_drivers ;
2759 a; b=a, a=b->next) {
2768 if (!res && (console->flags & CON_EXTENDED))
2769 nr_ext_console_drivers--;
2772 * If this isn't the last console and it has CON_CONSDEV set, we
2773 * need to set it on the next preferred console.
2775 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2776 console_drivers->flags |= CON_CONSDEV;
2778 console->flags &= ~CON_ENABLED;
2780 console_sysfs_notify();
2783 EXPORT_SYMBOL(unregister_console);
2786 * Initialize the console device. This is called *early*, so
2787 * we can't necessarily depend on lots of kernel help here.
2788 * Just do some early initializations, and do the complex setup
2791 void __init console_init(void)
2795 initcall_entry_t *ce;
2797 /* Setup the default TTY line discipline. */
2801 * set up the console device so that later boot sequences can
2802 * inform about problems etc..
2804 ce = __con_initcall_start;
2805 trace_initcall_level("console");
2806 while (ce < __con_initcall_end) {
2807 call = initcall_from_entry(ce);
2808 trace_initcall_start(call);
2810 trace_initcall_finish(call, ret);
2816 * Some boot consoles access data that is in the init section and which will
2817 * be discarded after the initcalls have been run. To make sure that no code
2818 * will access this data, unregister the boot consoles in a late initcall.
2820 * If for some reason, such as deferred probe or the driver being a loadable
2821 * module, the real console hasn't registered yet at this point, there will
2822 * be a brief interval in which no messages are logged to the console, which
2823 * makes it difficult to diagnose problems that occur during this time.
2825 * To mitigate this problem somewhat, only unregister consoles whose memory
2826 * intersects with the init section. Note that all other boot consoles will
2827 * get unregistred when the real preferred console is registered.
2829 static int __init printk_late_init(void)
2831 struct console *con;
2834 for_each_console(con) {
2835 if (!(con->flags & CON_BOOT))
2838 /* Check addresses that might be used for enabled consoles. */
2839 if (init_section_intersects(con, sizeof(*con)) ||
2840 init_section_contains(con->write, 0) ||
2841 init_section_contains(con->read, 0) ||
2842 init_section_contains(con->device, 0) ||
2843 init_section_contains(con->unblank, 0) ||
2844 init_section_contains(con->data, 0)) {
2846 * Please, consider moving the reported consoles out
2847 * of the init section.
2849 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2850 con->name, con->index);
2851 unregister_console(con);
2854 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2855 console_cpu_notify);
2857 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2858 console_cpu_notify, NULL);
2862 late_initcall(printk_late_init);
2864 #if defined CONFIG_PRINTK
2866 * Delayed printk version, for scheduler-internal messages:
2868 #define PRINTK_PENDING_WAKEUP 0x01
2869 #define PRINTK_PENDING_OUTPUT 0x02
2871 static DEFINE_PER_CPU(int, printk_pending);
2873 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2875 int pending = __this_cpu_xchg(printk_pending, 0);
2877 if (pending & PRINTK_PENDING_OUTPUT) {
2878 /* If trylock fails, someone else is doing the printing */
2879 if (console_trylock())
2883 if (pending & PRINTK_PENDING_WAKEUP)
2884 wake_up_interruptible(&log_wait);
2887 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2888 .func = wake_up_klogd_work_func,
2889 .flags = IRQ_WORK_LAZY,
2892 void wake_up_klogd(void)
2895 if (waitqueue_active(&log_wait)) {
2896 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2897 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2902 void defer_console_output(void)
2905 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2906 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2910 int vprintk_deferred(const char *fmt, va_list args)
2914 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2915 defer_console_output();
2920 int printk_deferred(const char *fmt, ...)
2925 va_start(args, fmt);
2926 r = vprintk_deferred(fmt, args);
2933 * printk rate limiting, lifted from the networking subsystem.
2935 * This enforces a rate limit: not more than 10 kernel messages
2936 * every 5s to make a denial-of-service attack impossible.
2938 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2940 int __printk_ratelimit(const char *func)
2942 return ___ratelimit(&printk_ratelimit_state, func);
2944 EXPORT_SYMBOL(__printk_ratelimit);
2947 * printk_timed_ratelimit - caller-controlled printk ratelimiting
2948 * @caller_jiffies: pointer to caller's state
2949 * @interval_msecs: minimum interval between prints
2951 * printk_timed_ratelimit() returns true if more than @interval_msecs
2952 * milliseconds have elapsed since the last time printk_timed_ratelimit()
2955 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
2956 unsigned int interval_msecs)
2958 unsigned long elapsed = jiffies - *caller_jiffies;
2960 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
2963 *caller_jiffies = jiffies;
2966 EXPORT_SYMBOL(printk_timed_ratelimit);
2968 static DEFINE_SPINLOCK(dump_list_lock);
2969 static LIST_HEAD(dump_list);
2972 * kmsg_dump_register - register a kernel log dumper.
2973 * @dumper: pointer to the kmsg_dumper structure
2975 * Adds a kernel log dumper to the system. The dump callback in the
2976 * structure will be called when the kernel oopses or panics and must be
2977 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
2979 int kmsg_dump_register(struct kmsg_dumper *dumper)
2981 unsigned long flags;
2984 /* The dump callback needs to be set */
2988 spin_lock_irqsave(&dump_list_lock, flags);
2989 /* Don't allow registering multiple times */
2990 if (!dumper->registered) {
2991 dumper->registered = 1;
2992 list_add_tail_rcu(&dumper->list, &dump_list);
2995 spin_unlock_irqrestore(&dump_list_lock, flags);
2999 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3002 * kmsg_dump_unregister - unregister a kmsg dumper.
3003 * @dumper: pointer to the kmsg_dumper structure
3005 * Removes a dump device from the system. Returns zero on success and
3006 * %-EINVAL otherwise.
3008 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3010 unsigned long flags;
3013 spin_lock_irqsave(&dump_list_lock, flags);
3014 if (dumper->registered) {
3015 dumper->registered = 0;
3016 list_del_rcu(&dumper->list);
3019 spin_unlock_irqrestore(&dump_list_lock, flags);
3024 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3026 static bool always_kmsg_dump;
3027 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3030 * kmsg_dump - dump kernel log to kernel message dumpers.
3031 * @reason: the reason (oops, panic etc) for dumping
3033 * Call each of the registered dumper's dump() callback, which can
3034 * retrieve the kmsg records with kmsg_dump_get_line() or
3035 * kmsg_dump_get_buffer().
3037 void kmsg_dump(enum kmsg_dump_reason reason)
3039 struct kmsg_dumper *dumper;
3040 unsigned long flags;
3042 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3046 list_for_each_entry_rcu(dumper, &dump_list, list) {
3047 if (dumper->max_reason && reason > dumper->max_reason)
3050 /* initialize iterator with data about the stored records */
3051 dumper->active = true;
3053 logbuf_lock_irqsave(flags);
3054 dumper->cur_seq = clear_seq;
3055 dumper->cur_idx = clear_idx;
3056 dumper->next_seq = log_next_seq;
3057 dumper->next_idx = log_next_idx;
3058 logbuf_unlock_irqrestore(flags);
3060 /* invoke dumper which will iterate over records */
3061 dumper->dump(dumper, reason);
3063 /* reset iterator */
3064 dumper->active = false;
3070 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3071 * @dumper: registered kmsg dumper
3072 * @syslog: include the "<4>" prefixes
3073 * @line: buffer to copy the line to
3074 * @size: maximum size of the buffer
3075 * @len: length of line placed into buffer
3077 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3078 * record, and copy one record into the provided buffer.
3080 * Consecutive calls will return the next available record moving
3081 * towards the end of the buffer with the youngest messages.
3083 * A return value of FALSE indicates that there are no more records to
3086 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3088 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3089 char *line, size_t size, size_t *len)
3091 struct printk_log *msg;
3095 if (!dumper->active)
3098 if (dumper->cur_seq < log_first_seq) {
3099 /* messages are gone, move to first available one */
3100 dumper->cur_seq = log_first_seq;
3101 dumper->cur_idx = log_first_idx;
3105 if (dumper->cur_seq >= log_next_seq)
3108 msg = log_from_idx(dumper->cur_idx);
3109 l = msg_print_text(msg, syslog, line, size);
3111 dumper->cur_idx = log_next(dumper->cur_idx);
3121 * kmsg_dump_get_line - retrieve one kmsg log line
3122 * @dumper: registered kmsg dumper
3123 * @syslog: include the "<4>" prefixes
3124 * @line: buffer to copy the line to
3125 * @size: maximum size of the buffer
3126 * @len: length of line placed into buffer
3128 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3129 * record, and copy one record into the provided buffer.
3131 * Consecutive calls will return the next available record moving
3132 * towards the end of the buffer with the youngest messages.
3134 * A return value of FALSE indicates that there are no more records to
3137 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3138 char *line, size_t size, size_t *len)
3140 unsigned long flags;
3143 logbuf_lock_irqsave(flags);
3144 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3145 logbuf_unlock_irqrestore(flags);
3149 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3152 * kmsg_dump_get_buffer - copy kmsg log lines
3153 * @dumper: registered kmsg dumper
3154 * @syslog: include the "<4>" prefixes
3155 * @buf: buffer to copy the line to
3156 * @size: maximum size of the buffer
3157 * @len: length of line placed into buffer
3159 * Start at the end of the kmsg buffer and fill the provided buffer
3160 * with as many of the the *youngest* kmsg records that fit into it.
3161 * If the buffer is large enough, all available kmsg records will be
3162 * copied with a single call.
3164 * Consecutive calls will fill the buffer with the next block of
3165 * available older records, not including the earlier retrieved ones.
3167 * A return value of FALSE indicates that there are no more records to
3170 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3171 char *buf, size_t size, size_t *len)
3173 unsigned long flags;
3181 if (!dumper->active)
3184 logbuf_lock_irqsave(flags);
3185 if (dumper->cur_seq < log_first_seq) {
3186 /* messages are gone, move to first available one */
3187 dumper->cur_seq = log_first_seq;
3188 dumper->cur_idx = log_first_idx;
3192 if (dumper->cur_seq >= dumper->next_seq) {
3193 logbuf_unlock_irqrestore(flags);
3197 /* calculate length of entire buffer */
3198 seq = dumper->cur_seq;
3199 idx = dumper->cur_idx;
3200 while (seq < dumper->next_seq) {
3201 struct printk_log *msg = log_from_idx(idx);
3203 l += msg_print_text(msg, true, NULL, 0);
3204 idx = log_next(idx);
3208 /* move first record forward until length fits into the buffer */
3209 seq = dumper->cur_seq;
3210 idx = dumper->cur_idx;
3211 while (l > size && seq < dumper->next_seq) {
3212 struct printk_log *msg = log_from_idx(idx);
3214 l -= msg_print_text(msg, true, NULL, 0);
3215 idx = log_next(idx);
3219 /* last message in next interation */
3224 while (seq < dumper->next_seq) {
3225 struct printk_log *msg = log_from_idx(idx);
3227 l += msg_print_text(msg, syslog, buf + l, size - l);
3228 idx = log_next(idx);
3232 dumper->next_seq = next_seq;
3233 dumper->next_idx = next_idx;
3235 logbuf_unlock_irqrestore(flags);
3241 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3244 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3245 * @dumper: registered kmsg dumper
3247 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3248 * kmsg_dump_get_buffer() can be called again and used multiple
3249 * times within the same dumper.dump() callback.
3251 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3253 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3255 dumper->cur_seq = clear_seq;
3256 dumper->cur_idx = clear_idx;
3257 dumper->next_seq = log_next_seq;
3258 dumper->next_idx = log_next_idx;
3262 * kmsg_dump_rewind - reset the interator
3263 * @dumper: registered kmsg dumper
3265 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3266 * kmsg_dump_get_buffer() can be called again and used multiple
3267 * times within the same dumper.dump() callback.
3269 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3271 unsigned long flags;
3273 logbuf_lock_irqsave(flags);
3274 kmsg_dump_rewind_nolock(dumper);
3275 logbuf_unlock_irqrestore(flags);
3277 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);