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 */
70 * Low level drivers may need that to know if they can schedule in
71 * their unblank() callback or not. So let's export it.
74 EXPORT_SYMBOL(oops_in_progress);
77 * console_sem protects the console_drivers list, and also
78 * provides serialisation for access to the entire console
81 static DEFINE_SEMAPHORE(console_sem);
82 struct console *console_drivers;
83 EXPORT_SYMBOL_GPL(console_drivers);
86 static struct lockdep_map console_lock_dep_map = {
87 .name = "console_lock"
91 enum devkmsg_log_bits {
92 __DEVKMSG_LOG_BIT_ON = 0,
93 __DEVKMSG_LOG_BIT_OFF,
94 __DEVKMSG_LOG_BIT_LOCK,
97 enum devkmsg_log_masks {
98 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
99 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
100 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
103 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
104 #define DEVKMSG_LOG_MASK_DEFAULT 0
106 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
108 static int __control_devkmsg(char *str)
113 if (!strncmp(str, "on", 2)) {
114 devkmsg_log = DEVKMSG_LOG_MASK_ON;
116 } else if (!strncmp(str, "off", 3)) {
117 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
119 } else if (!strncmp(str, "ratelimit", 9)) {
120 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
126 static int __init control_devkmsg(char *str)
128 if (__control_devkmsg(str) < 0)
132 * Set sysctl string accordingly:
134 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
135 strcpy(devkmsg_log_str, "on");
136 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
137 strcpy(devkmsg_log_str, "off");
138 /* else "ratelimit" which is set by default. */
141 * Sysctl cannot change it anymore. The kernel command line setting of
142 * this parameter is to force the setting to be permanent throughout the
143 * runtime of the system. This is a precation measure against userspace
144 * trying to be a smarta** and attempting to change it up on us.
146 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
150 __setup("printk.devkmsg=", control_devkmsg);
152 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
154 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
155 void __user *buffer, size_t *lenp, loff_t *ppos)
157 char old_str[DEVKMSG_STR_MAX_SIZE];
162 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
166 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
169 err = proc_dostring(table, write, buffer, lenp, ppos);
174 err = __control_devkmsg(devkmsg_log_str);
177 * Do not accept an unknown string OR a known string with
180 if (err < 0 || (err + 1 != *lenp)) {
182 /* ... and restore old setting. */
184 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
194 * Number of registered extended console drivers.
196 * If extended consoles are present, in-kernel cont reassembly is disabled
197 * and each fragment is stored as a separate log entry with proper
198 * continuation flag so that every emitted message has full metadata. This
199 * doesn't change the result for regular consoles or /proc/kmsg. For
200 * /dev/kmsg, as long as the reader concatenates messages according to
201 * consecutive continuation flags, the end result should be the same too.
203 static int nr_ext_console_drivers;
206 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
207 * macros instead of functions so that _RET_IP_ contains useful information.
209 #define down_console_sem() do { \
211 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
214 static int __down_trylock_console_sem(unsigned long ip)
220 * Here and in __up_console_sem() we need to be in safe mode,
221 * because spindump/WARN/etc from under console ->lock will
222 * deadlock in printk()->down_trylock_console_sem() otherwise.
224 printk_safe_enter_irqsave(flags);
225 lock_failed = down_trylock(&console_sem);
226 printk_safe_exit_irqrestore(flags);
230 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
233 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
235 static void __up_console_sem(unsigned long ip)
239 mutex_release(&console_lock_dep_map, 1, ip);
241 printk_safe_enter_irqsave(flags);
243 printk_safe_exit_irqrestore(flags);
245 #define up_console_sem() __up_console_sem(_RET_IP_)
248 * This is used for debugging the mess that is the VT code by
249 * keeping track if we have the console semaphore held. It's
250 * definitely not the perfect debug tool (we don't know if _WE_
251 * hold it and are racing, but it helps tracking those weird code
252 * paths in the console code where we end up in places I want
253 * locked without the console sempahore held).
255 static int console_locked, console_suspended;
258 * If exclusive_console is non-NULL then only this console is to be printed to.
260 static struct console *exclusive_console;
263 * Array of consoles built from command line options (console=)
266 #define MAX_CMDLINECONSOLES 8
268 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
270 static int preferred_console = -1;
271 int console_set_on_cmdline;
272 EXPORT_SYMBOL(console_set_on_cmdline);
274 /* Flag: console code may call schedule() */
275 static int console_may_schedule;
277 enum con_msg_format_flags {
278 MSG_FORMAT_DEFAULT = 0,
279 MSG_FORMAT_SYSLOG = (1 << 0),
282 static int console_msg_format = MSG_FORMAT_DEFAULT;
285 * The printk log buffer consists of a chain of concatenated variable
286 * length records. Every record starts with a record header, containing
287 * the overall length of the record.
289 * The heads to the first and last entry in the buffer, as well as the
290 * sequence numbers of these entries are maintained when messages are
293 * If the heads indicate available messages, the length in the header
294 * tells the start next message. A length == 0 for the next message
295 * indicates a wrap-around to the beginning of the buffer.
297 * Every record carries the monotonic timestamp in microseconds, as well as
298 * the standard userspace syslog level and syslog facility. The usual
299 * kernel messages use LOG_KERN; userspace-injected messages always carry
300 * a matching syslog facility, by default LOG_USER. The origin of every
301 * message can be reliably determined that way.
303 * The human readable log message directly follows the message header. The
304 * length of the message text is stored in the header, the stored message
307 * Optionally, a message can carry a dictionary of properties (key/value pairs),
308 * to provide userspace with a machine-readable message context.
310 * Examples for well-defined, commonly used property names are:
311 * DEVICE=b12:8 device identifier
315 * +sound:card0 subsystem:devname
316 * SUBSYSTEM=pci driver-core subsystem name
318 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
319 * follows directly after a '=' character. Every property is terminated by
320 * a '\0' character. The last property is not terminated.
322 * Example of a message structure:
323 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
324 * 0008 34 00 record is 52 bytes long
325 * 000a 0b 00 text is 11 bytes long
326 * 000c 1f 00 dictionary is 23 bytes long
327 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
328 * 0010 69 74 27 73 20 61 20 6c "it's a l"
330 * 001b 44 45 56 49 43 "DEVIC"
331 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
332 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
334 * 0032 00 00 00 padding to next message header
336 * The 'struct printk_log' buffer header must never be directly exported to
337 * userspace, it is a kernel-private implementation detail that might
338 * need to be changed in the future, when the requirements change.
340 * /dev/kmsg exports the structured data in the following line format:
341 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
343 * Users of the export format should ignore possible additional values
344 * separated by ',', and find the message after the ';' character.
346 * The optional key/value pairs are attached as continuation lines starting
347 * with a space character and terminated by a newline. All possible
348 * non-prinatable characters are escaped in the "\xff" notation.
352 LOG_NOCONS = 1, /* already flushed, do not print to console */
353 LOG_NEWLINE = 2, /* text ended with a newline */
354 LOG_PREFIX = 4, /* text started with a prefix */
355 LOG_CONT = 8, /* text is a fragment of a continuation line */
359 u64 ts_nsec; /* timestamp in nanoseconds */
360 u16 len; /* length of entire record */
361 u16 text_len; /* length of text buffer */
362 u16 dict_len; /* length of dictionary buffer */
363 u8 facility; /* syslog facility */
364 u8 flags:5; /* internal record flags */
365 u8 level:3; /* syslog level */
367 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
368 __packed __aligned(4)
373 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
374 * within the scheduler's rq lock. It must be released before calling
375 * console_unlock() or anything else that might wake up a process.
377 DEFINE_RAW_SPINLOCK(logbuf_lock);
380 * Helper macros to lock/unlock logbuf_lock and switch between
381 * printk-safe/unsafe modes.
383 #define logbuf_lock_irq() \
385 printk_safe_enter_irq(); \
386 raw_spin_lock(&logbuf_lock); \
389 #define logbuf_unlock_irq() \
391 raw_spin_unlock(&logbuf_lock); \
392 printk_safe_exit_irq(); \
395 #define logbuf_lock_irqsave(flags) \
397 printk_safe_enter_irqsave(flags); \
398 raw_spin_lock(&logbuf_lock); \
401 #define logbuf_unlock_irqrestore(flags) \
403 raw_spin_unlock(&logbuf_lock); \
404 printk_safe_exit_irqrestore(flags); \
408 DECLARE_WAIT_QUEUE_HEAD(log_wait);
409 /* the next printk record to read by syslog(READ) or /proc/kmsg */
410 static u64 syslog_seq;
411 static u32 syslog_idx;
412 static size_t syslog_partial;
414 /* index and sequence number of the first record stored in the buffer */
415 static u64 log_first_seq;
416 static u32 log_first_idx;
418 /* index and sequence number of the next record to store in the buffer */
419 static u64 log_next_seq;
420 static u32 log_next_idx;
422 /* the next printk record to write to the console */
423 static u64 console_seq;
424 static u32 console_idx;
426 /* the next printk record to read after the last 'clear' command */
427 static u64 clear_seq;
428 static u32 clear_idx;
430 #define PREFIX_MAX 32
431 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
433 #define LOG_LEVEL(v) ((v) & 0x07)
434 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
437 #define LOG_ALIGN __alignof__(struct printk_log)
438 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
439 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
440 static char *log_buf = __log_buf;
441 static u32 log_buf_len = __LOG_BUF_LEN;
443 /* Return log buffer address */
444 char *log_buf_addr_get(void)
449 /* Return log buffer size */
450 u32 log_buf_len_get(void)
455 /* human readable text of the record */
456 static char *log_text(const struct printk_log *msg)
458 return (char *)msg + sizeof(struct printk_log);
461 /* optional key/value pair dictionary attached to the record */
462 static char *log_dict(const struct printk_log *msg)
464 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
467 /* get record by index; idx must point to valid msg */
468 static struct printk_log *log_from_idx(u32 idx)
470 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
473 * A length == 0 record is the end of buffer marker. Wrap around and
474 * read the message at the start of the buffer.
477 return (struct printk_log *)log_buf;
481 /* get next record; idx must point to valid msg */
482 static u32 log_next(u32 idx)
484 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
486 /* length == 0 indicates the end of the buffer; wrap */
488 * A length == 0 record is the end of buffer marker. Wrap around and
489 * read the message at the start of the buffer as *this* one, and
490 * return the one after that.
493 msg = (struct printk_log *)log_buf;
496 return idx + msg->len;
500 * Check whether there is enough free space for the given message.
502 * The same values of first_idx and next_idx mean that the buffer
503 * is either empty or full.
505 * If the buffer is empty, we must respect the position of the indexes.
506 * They cannot be reset to the beginning of the buffer.
508 static int logbuf_has_space(u32 msg_size, bool empty)
512 if (log_next_idx > log_first_idx || empty)
513 free = max(log_buf_len - log_next_idx, log_first_idx);
515 free = log_first_idx - log_next_idx;
518 * We need space also for an empty header that signalizes wrapping
521 return free >= msg_size + sizeof(struct printk_log);
524 static int log_make_free_space(u32 msg_size)
526 while (log_first_seq < log_next_seq &&
527 !logbuf_has_space(msg_size, false)) {
528 /* drop old messages until we have enough contiguous space */
529 log_first_idx = log_next(log_first_idx);
533 if (clear_seq < log_first_seq) {
534 clear_seq = log_first_seq;
535 clear_idx = log_first_idx;
538 /* sequence numbers are equal, so the log buffer is empty */
539 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
545 /* compute the message size including the padding bytes */
546 static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
550 size = sizeof(struct printk_log) + text_len + dict_len;
551 *pad_len = (-size) & (LOG_ALIGN - 1);
558 * Define how much of the log buffer we could take at maximum. The value
559 * must be greater than two. Note that only half of the buffer is available
560 * when the index points to the middle.
562 #define MAX_LOG_TAKE_PART 4
563 static const char trunc_msg[] = "<truncated>";
565 static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
566 u16 *dict_len, u32 *pad_len)
569 * The message should not take the whole buffer. Otherwise, it might
570 * get removed too soon.
572 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
573 if (*text_len > max_text_len)
574 *text_len = max_text_len;
575 /* enable the warning message */
576 *trunc_msg_len = strlen(trunc_msg);
577 /* disable the "dict" completely */
579 /* compute the size again, count also the warning message */
580 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
583 /* insert record into the buffer, discard old ones, update heads */
584 static int log_store(int facility, int level,
585 enum log_flags flags, u64 ts_nsec,
586 const char *dict, u16 dict_len,
587 const char *text, u16 text_len)
589 struct printk_log *msg;
591 u16 trunc_msg_len = 0;
593 /* number of '\0' padding bytes to next message */
594 size = msg_used_size(text_len, dict_len, &pad_len);
596 if (log_make_free_space(size)) {
597 /* truncate the message if it is too long for empty buffer */
598 size = truncate_msg(&text_len, &trunc_msg_len,
599 &dict_len, &pad_len);
600 /* survive when the log buffer is too small for trunc_msg */
601 if (log_make_free_space(size))
605 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
607 * This message + an additional empty header does not fit
608 * at the end of the buffer. Add an empty header with len == 0
609 * to signify a wrap around.
611 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
616 msg = (struct printk_log *)(log_buf + log_next_idx);
617 memcpy(log_text(msg), text, text_len);
618 msg->text_len = text_len;
620 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
621 msg->text_len += trunc_msg_len;
623 memcpy(log_dict(msg), dict, dict_len);
624 msg->dict_len = dict_len;
625 msg->facility = facility;
626 msg->level = level & 7;
627 msg->flags = flags & 0x1f;
629 msg->ts_nsec = ts_nsec;
631 msg->ts_nsec = local_clock();
632 memset(log_dict(msg) + dict_len, 0, pad_len);
636 log_next_idx += msg->len;
639 return msg->text_len;
642 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
644 static int syslog_action_restricted(int type)
649 * Unless restricted, we allow "read all" and "get buffer size"
652 return type != SYSLOG_ACTION_READ_ALL &&
653 type != SYSLOG_ACTION_SIZE_BUFFER;
656 static int check_syslog_permissions(int type, int source)
659 * If this is from /proc/kmsg and we've already opened it, then we've
660 * already done the capabilities checks at open time.
662 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
665 if (syslog_action_restricted(type)) {
666 if (capable(CAP_SYSLOG))
669 * For historical reasons, accept CAP_SYS_ADMIN too, with
672 if (capable(CAP_SYS_ADMIN)) {
673 pr_warn_once("%s (%d): Attempt to access syslog with "
674 "CAP_SYS_ADMIN but no CAP_SYSLOG "
676 current->comm, task_pid_nr(current));
682 return security_syslog(type);
685 static void append_char(char **pp, char *e, char c)
691 static ssize_t msg_print_ext_header(char *buf, size_t size,
692 struct printk_log *msg, u64 seq)
694 u64 ts_usec = msg->ts_nsec;
696 do_div(ts_usec, 1000);
698 return scnprintf(buf, size, "%u,%llu,%llu,%c;",
699 (msg->facility << 3) | msg->level, seq, ts_usec,
700 msg->flags & LOG_CONT ? 'c' : '-');
703 static ssize_t msg_print_ext_body(char *buf, size_t size,
704 char *dict, size_t dict_len,
705 char *text, size_t text_len)
707 char *p = buf, *e = buf + size;
710 /* escape non-printable characters */
711 for (i = 0; i < text_len; i++) {
712 unsigned char c = text[i];
714 if (c < ' ' || c >= 127 || c == '\\')
715 p += scnprintf(p, e - p, "\\x%02x", c);
717 append_char(&p, e, c);
719 append_char(&p, e, '\n');
724 for (i = 0; i < dict_len; i++) {
725 unsigned char c = dict[i];
728 append_char(&p, e, ' ');
733 append_char(&p, e, '\n');
738 if (c < ' ' || c >= 127 || c == '\\') {
739 p += scnprintf(p, e - p, "\\x%02x", c);
743 append_char(&p, e, c);
745 append_char(&p, e, '\n');
751 /* /dev/kmsg - userspace message inject/listen interface */
752 struct devkmsg_user {
755 struct ratelimit_state rs;
757 char buf[CONSOLE_EXT_LOG_MAX];
760 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
763 int level = default_message_loglevel;
764 int facility = 1; /* LOG_USER */
765 struct file *file = iocb->ki_filp;
766 struct devkmsg_user *user = file->private_data;
767 size_t len = iov_iter_count(from);
770 if (!user || len > LOG_LINE_MAX)
773 /* Ignore when user logging is disabled. */
774 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
777 /* Ratelimit when not explicitly enabled. */
778 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
779 if (!___ratelimit(&user->rs, current->comm))
783 buf = kmalloc(len+1, GFP_KERNEL);
788 if (!copy_from_iter_full(buf, len, from)) {
794 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
795 * the decimal value represents 32bit, the lower 3 bit are the log
796 * level, the rest are the log facility.
798 * If no prefix or no userspace facility is specified, we
799 * enforce LOG_USER, to be able to reliably distinguish
800 * kernel-generated messages from userspace-injected ones.
803 if (line[0] == '<') {
807 u = simple_strtoul(line + 1, &endp, 10);
808 if (endp && endp[0] == '>') {
809 level = LOG_LEVEL(u);
810 if (LOG_FACILITY(u) != 0)
811 facility = LOG_FACILITY(u);
818 printk_emit(facility, level, NULL, 0, "%s", line);
823 static ssize_t devkmsg_read(struct file *file, char __user *buf,
824 size_t count, loff_t *ppos)
826 struct devkmsg_user *user = file->private_data;
827 struct printk_log *msg;
834 ret = mutex_lock_interruptible(&user->lock);
839 while (user->seq == log_next_seq) {
840 if (file->f_flags & O_NONBLOCK) {
847 ret = wait_event_interruptible(log_wait,
848 user->seq != log_next_seq);
854 if (user->seq < log_first_seq) {
855 /* our last seen message is gone, return error and reset */
856 user->idx = log_first_idx;
857 user->seq = log_first_seq;
863 msg = log_from_idx(user->idx);
864 len = msg_print_ext_header(user->buf, sizeof(user->buf),
866 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
867 log_dict(msg), msg->dict_len,
868 log_text(msg), msg->text_len);
870 user->idx = log_next(user->idx);
879 if (copy_to_user(buf, user->buf, len)) {
885 mutex_unlock(&user->lock);
889 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
891 struct devkmsg_user *user = file->private_data;
902 /* the first record */
903 user->idx = log_first_idx;
904 user->seq = log_first_seq;
908 * The first record after the last SYSLOG_ACTION_CLEAR,
909 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
910 * changes no global state, and does not clear anything.
912 user->idx = clear_idx;
913 user->seq = clear_seq;
916 /* after the last record */
917 user->idx = log_next_idx;
918 user->seq = log_next_seq;
927 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
929 struct devkmsg_user *user = file->private_data;
933 return EPOLLERR|EPOLLNVAL;
935 poll_wait(file, &log_wait, wait);
938 if (user->seq < log_next_seq) {
939 /* return error when data has vanished underneath us */
940 if (user->seq < log_first_seq)
941 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
943 ret = EPOLLIN|EPOLLRDNORM;
950 static int devkmsg_open(struct inode *inode, struct file *file)
952 struct devkmsg_user *user;
955 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
958 /* write-only does not need any file context */
959 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
960 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
966 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
970 ratelimit_default_init(&user->rs);
971 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
973 mutex_init(&user->lock);
976 user->idx = log_first_idx;
977 user->seq = log_first_seq;
980 file->private_data = user;
984 static int devkmsg_release(struct inode *inode, struct file *file)
986 struct devkmsg_user *user = file->private_data;
991 ratelimit_state_exit(&user->rs);
993 mutex_destroy(&user->lock);
998 const struct file_operations kmsg_fops = {
999 .open = devkmsg_open,
1000 .read = devkmsg_read,
1001 .write_iter = devkmsg_write,
1002 .llseek = devkmsg_llseek,
1003 .poll = devkmsg_poll,
1004 .release = devkmsg_release,
1007 #ifdef CONFIG_CRASH_CORE
1009 * This appends the listed symbols to /proc/vmcore
1011 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1012 * obtain access to symbols that are otherwise very difficult to locate. These
1013 * symbols are specifically used so that utilities can access and extract the
1014 * dmesg log from a vmcore file after a crash.
1016 void log_buf_vmcoreinfo_setup(void)
1018 VMCOREINFO_SYMBOL(log_buf);
1019 VMCOREINFO_SYMBOL(log_buf_len);
1020 VMCOREINFO_SYMBOL(log_first_idx);
1021 VMCOREINFO_SYMBOL(clear_idx);
1022 VMCOREINFO_SYMBOL(log_next_idx);
1024 * Export struct printk_log size and field offsets. User space tools can
1025 * parse it and detect any changes to structure down the line.
1027 VMCOREINFO_STRUCT_SIZE(printk_log);
1028 VMCOREINFO_OFFSET(printk_log, ts_nsec);
1029 VMCOREINFO_OFFSET(printk_log, len);
1030 VMCOREINFO_OFFSET(printk_log, text_len);
1031 VMCOREINFO_OFFSET(printk_log, dict_len);
1035 /* requested log_buf_len from kernel cmdline */
1036 static unsigned long __initdata new_log_buf_len;
1038 /* we practice scaling the ring buffer by powers of 2 */
1039 static void __init log_buf_len_update(unsigned size)
1042 size = roundup_pow_of_two(size);
1043 if (size > log_buf_len)
1044 new_log_buf_len = size;
1047 /* save requested log_buf_len since it's too early to process it */
1048 static int __init log_buf_len_setup(char *str)
1050 unsigned size = memparse(str, &str);
1052 log_buf_len_update(size);
1056 early_param("log_buf_len", log_buf_len_setup);
1059 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1061 static void __init log_buf_add_cpu(void)
1063 unsigned int cpu_extra;
1066 * archs should set up cpu_possible_bits properly with
1067 * set_cpu_possible() after setup_arch() but just in
1068 * case lets ensure this is valid.
1070 if (num_possible_cpus() == 1)
1073 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1075 /* by default this will only continue through for large > 64 CPUs */
1076 if (cpu_extra <= __LOG_BUF_LEN / 2)
1079 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1080 __LOG_CPU_MAX_BUF_LEN);
1081 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1083 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1085 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1087 #else /* !CONFIG_SMP */
1088 static inline void log_buf_add_cpu(void) {}
1089 #endif /* CONFIG_SMP */
1091 void __init setup_log_buf(int early)
1093 unsigned long flags;
1097 if (log_buf != __log_buf)
1100 if (!early && !new_log_buf_len)
1103 if (!new_log_buf_len)
1108 memblock_virt_alloc(new_log_buf_len, LOG_ALIGN);
1110 new_log_buf = memblock_virt_alloc_nopanic(new_log_buf_len,
1114 if (unlikely(!new_log_buf)) {
1115 pr_err("log_buf_len: %ld bytes not available\n",
1120 logbuf_lock_irqsave(flags);
1121 log_buf_len = new_log_buf_len;
1122 log_buf = new_log_buf;
1123 new_log_buf_len = 0;
1124 free = __LOG_BUF_LEN - log_next_idx;
1125 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1126 logbuf_unlock_irqrestore(flags);
1128 pr_info("log_buf_len: %d bytes\n", log_buf_len);
1129 pr_info("early log buf free: %d(%d%%)\n",
1130 free, (free * 100) / __LOG_BUF_LEN);
1133 static bool __read_mostly ignore_loglevel;
1135 static int __init ignore_loglevel_setup(char *str)
1137 ignore_loglevel = true;
1138 pr_info("debug: ignoring loglevel setting.\n");
1143 early_param("ignore_loglevel", ignore_loglevel_setup);
1144 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1145 MODULE_PARM_DESC(ignore_loglevel,
1146 "ignore loglevel setting (prints all kernel messages to the console)");
1148 static bool suppress_message_printing(int level)
1150 return (level >= console_loglevel && !ignore_loglevel);
1153 #ifdef CONFIG_BOOT_PRINTK_DELAY
1155 static int boot_delay; /* msecs delay after each printk during bootup */
1156 static unsigned long long loops_per_msec; /* based on boot_delay */
1158 static int __init boot_delay_setup(char *str)
1162 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1163 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1165 get_option(&str, &boot_delay);
1166 if (boot_delay > 10 * 1000)
1169 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1170 "HZ: %d, loops_per_msec: %llu\n",
1171 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1174 early_param("boot_delay", boot_delay_setup);
1176 static void boot_delay_msec(int level)
1178 unsigned long long k;
1179 unsigned long timeout;
1181 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1182 || suppress_message_printing(level)) {
1186 k = (unsigned long long)loops_per_msec * boot_delay;
1188 timeout = jiffies + msecs_to_jiffies(boot_delay);
1193 * use (volatile) jiffies to prevent
1194 * compiler reduction; loop termination via jiffies
1195 * is secondary and may or may not happen.
1197 if (time_after(jiffies, timeout))
1199 touch_nmi_watchdog();
1203 static inline void boot_delay_msec(int level)
1208 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1209 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1211 static size_t print_time(u64 ts, char *buf)
1213 unsigned long rem_nsec;
1218 rem_nsec = do_div(ts, 1000000000);
1221 return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
1223 return sprintf(buf, "[%5lu.%06lu] ",
1224 (unsigned long)ts, rem_nsec / 1000);
1227 static size_t print_prefix(const struct printk_log *msg, bool syslog, char *buf)
1230 unsigned int prefix = (msg->facility << 3) | msg->level;
1234 len += sprintf(buf, "<%u>", prefix);
1239 else if (prefix > 99)
1241 else if (prefix > 9)
1246 len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
1250 static size_t msg_print_text(const struct printk_log *msg, bool syslog, char *buf, size_t size)
1252 const char *text = log_text(msg);
1253 size_t text_size = msg->text_len;
1257 const char *next = memchr(text, '\n', text_size);
1261 text_len = next - text;
1263 text_size -= next - text;
1265 text_len = text_size;
1269 if (print_prefix(msg, syslog, NULL) +
1270 text_len + 1 >= size - len)
1273 len += print_prefix(msg, syslog, buf + len);
1274 memcpy(buf + len, text, text_len);
1278 /* SYSLOG_ACTION_* buffer size only calculation */
1279 len += print_prefix(msg, syslog, NULL);
1290 static int syslog_print(char __user *buf, int size)
1293 struct printk_log *msg;
1296 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1305 if (syslog_seq < log_first_seq) {
1306 /* messages are gone, move to first one */
1307 syslog_seq = log_first_seq;
1308 syslog_idx = log_first_idx;
1311 if (syslog_seq == log_next_seq) {
1312 logbuf_unlock_irq();
1316 skip = syslog_partial;
1317 msg = log_from_idx(syslog_idx);
1318 n = msg_print_text(msg, true, text, LOG_LINE_MAX + PREFIX_MAX);
1319 if (n - syslog_partial <= size) {
1320 /* message fits into buffer, move forward */
1321 syslog_idx = log_next(syslog_idx);
1323 n -= syslog_partial;
1326 /* partial read(), remember position */
1328 syslog_partial += n;
1331 logbuf_unlock_irq();
1336 if (copy_to_user(buf, text + skip, n)) {
1351 static int syslog_print_all(char __user *buf, int size, bool clear)
1356 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1367 * Find first record that fits, including all following records,
1368 * into the user-provided buffer for this dump.
1372 while (seq < log_next_seq) {
1373 struct printk_log *msg = log_from_idx(idx);
1375 len += msg_print_text(msg, true, NULL, 0);
1376 idx = log_next(idx);
1380 /* move first record forward until length fits into the buffer */
1383 while (len > size && seq < log_next_seq) {
1384 struct printk_log *msg = log_from_idx(idx);
1386 len -= msg_print_text(msg, true, NULL, 0);
1387 idx = log_next(idx);
1391 /* last message fitting into this dump */
1392 next_seq = log_next_seq;
1395 while (len >= 0 && seq < next_seq) {
1396 struct printk_log *msg = log_from_idx(idx);
1399 textlen = msg_print_text(msg, true, text,
1400 LOG_LINE_MAX + PREFIX_MAX);
1405 idx = log_next(idx);
1408 logbuf_unlock_irq();
1409 if (copy_to_user(buf + len, text, textlen))
1415 if (seq < log_first_seq) {
1416 /* messages are gone, move to next one */
1417 seq = log_first_seq;
1418 idx = log_first_idx;
1424 clear_seq = log_next_seq;
1425 clear_idx = log_next_idx;
1427 logbuf_unlock_irq();
1433 int do_syslog(int type, char __user *buf, int len, int source)
1436 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1439 error = check_syslog_permissions(type, source);
1444 case SYSLOG_ACTION_CLOSE: /* Close log */
1446 case SYSLOG_ACTION_OPEN: /* Open log */
1448 case SYSLOG_ACTION_READ: /* Read from log */
1449 if (!buf || len < 0)
1453 if (!access_ok(VERIFY_WRITE, buf, len))
1455 error = wait_event_interruptible(log_wait,
1456 syslog_seq != log_next_seq);
1459 error = syslog_print(buf, len);
1461 /* Read/clear last kernel messages */
1462 case SYSLOG_ACTION_READ_CLEAR:
1465 /* Read last kernel messages */
1466 case SYSLOG_ACTION_READ_ALL:
1467 if (!buf || len < 0)
1471 if (!access_ok(VERIFY_WRITE, buf, len))
1473 error = syslog_print_all(buf, len, clear);
1475 /* Clear ring buffer */
1476 case SYSLOG_ACTION_CLEAR:
1477 syslog_print_all(NULL, 0, true);
1479 /* Disable logging to console */
1480 case SYSLOG_ACTION_CONSOLE_OFF:
1481 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1482 saved_console_loglevel = console_loglevel;
1483 console_loglevel = minimum_console_loglevel;
1485 /* Enable logging to console */
1486 case SYSLOG_ACTION_CONSOLE_ON:
1487 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1488 console_loglevel = saved_console_loglevel;
1489 saved_console_loglevel = LOGLEVEL_DEFAULT;
1492 /* Set level of messages printed to console */
1493 case SYSLOG_ACTION_CONSOLE_LEVEL:
1494 if (len < 1 || len > 8)
1496 if (len < minimum_console_loglevel)
1497 len = minimum_console_loglevel;
1498 console_loglevel = len;
1499 /* Implicitly re-enable logging to console */
1500 saved_console_loglevel = LOGLEVEL_DEFAULT;
1502 /* Number of chars in the log buffer */
1503 case SYSLOG_ACTION_SIZE_UNREAD:
1505 if (syslog_seq < log_first_seq) {
1506 /* messages are gone, move to first one */
1507 syslog_seq = log_first_seq;
1508 syslog_idx = log_first_idx;
1511 if (source == SYSLOG_FROM_PROC) {
1513 * Short-cut for poll(/"proc/kmsg") which simply checks
1514 * for pending data, not the size; return the count of
1515 * records, not the length.
1517 error = log_next_seq - syslog_seq;
1519 u64 seq = syslog_seq;
1520 u32 idx = syslog_idx;
1522 while (seq < log_next_seq) {
1523 struct printk_log *msg = log_from_idx(idx);
1525 error += msg_print_text(msg, true, NULL, 0);
1526 idx = log_next(idx);
1529 error -= syslog_partial;
1531 logbuf_unlock_irq();
1533 /* Size of the log buffer */
1534 case SYSLOG_ACTION_SIZE_BUFFER:
1535 error = log_buf_len;
1545 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1547 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1551 * Special console_lock variants that help to reduce the risk of soft-lockups.
1552 * They allow to pass console_lock to another printk() call using a busy wait.
1555 #ifdef CONFIG_LOCKDEP
1556 static struct lockdep_map console_owner_dep_map = {
1557 .name = "console_owner"
1561 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1562 static struct task_struct *console_owner;
1563 static bool console_waiter;
1566 * console_lock_spinning_enable - mark beginning of code where another
1567 * thread might safely busy wait
1569 * This basically converts console_lock into a spinlock. This marks
1570 * the section where the console_lock owner can not sleep, because
1571 * there may be a waiter spinning (like a spinlock). Also it must be
1572 * ready to hand over the lock at the end of the section.
1574 static void console_lock_spinning_enable(void)
1576 raw_spin_lock(&console_owner_lock);
1577 console_owner = current;
1578 raw_spin_unlock(&console_owner_lock);
1580 /* The waiter may spin on us after setting console_owner */
1581 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1585 * console_lock_spinning_disable_and_check - mark end of code where another
1586 * thread was able to busy wait and check if there is a waiter
1588 * This is called at the end of the section where spinning is allowed.
1589 * It has two functions. First, it is a signal that it is no longer
1590 * safe to start busy waiting for the lock. Second, it checks if
1591 * there is a busy waiter and passes the lock rights to her.
1593 * Important: Callers lose the lock if there was a busy waiter.
1594 * They must not touch items synchronized by console_lock
1597 * Return: 1 if the lock rights were passed, 0 otherwise.
1599 static int console_lock_spinning_disable_and_check(void)
1603 raw_spin_lock(&console_owner_lock);
1604 waiter = READ_ONCE(console_waiter);
1605 console_owner = NULL;
1606 raw_spin_unlock(&console_owner_lock);
1609 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1613 /* The waiter is now free to continue */
1614 WRITE_ONCE(console_waiter, false);
1616 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1619 * Hand off console_lock to waiter. The waiter will perform
1620 * the up(). After this, the waiter is the console_lock owner.
1622 mutex_release(&console_lock_dep_map, 1, _THIS_IP_);
1627 * console_trylock_spinning - try to get console_lock by busy waiting
1629 * This allows to busy wait for the console_lock when the current
1630 * owner is running in specially marked sections. It means that
1631 * the current owner is running and cannot reschedule until it
1632 * is ready to lose the lock.
1634 * Return: 1 if we got the lock, 0 othrewise
1636 static int console_trylock_spinning(void)
1638 struct task_struct *owner = NULL;
1641 unsigned long flags;
1643 if (console_trylock())
1646 printk_safe_enter_irqsave(flags);
1648 raw_spin_lock(&console_owner_lock);
1649 owner = READ_ONCE(console_owner);
1650 waiter = READ_ONCE(console_waiter);
1651 if (!waiter && owner && owner != current) {
1652 WRITE_ONCE(console_waiter, true);
1655 raw_spin_unlock(&console_owner_lock);
1658 * If there is an active printk() writing to the
1659 * consoles, instead of having it write our data too,
1660 * see if we can offload that load from the active
1661 * printer, and do some printing ourselves.
1662 * Go into a spin only if there isn't already a waiter
1663 * spinning, and there is an active printer, and
1664 * that active printer isn't us (recursive printk?).
1667 printk_safe_exit_irqrestore(flags);
1671 /* We spin waiting for the owner to release us */
1672 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1673 /* Owner will clear console_waiter on hand off */
1674 while (READ_ONCE(console_waiter))
1676 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1678 printk_safe_exit_irqrestore(flags);
1680 * The owner passed the console lock to us.
1681 * Since we did not spin on console lock, annotate
1682 * this as a trylock. Otherwise lockdep will
1685 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1691 * Call the console drivers, asking them to write out
1692 * log_buf[start] to log_buf[end - 1].
1693 * The console_lock must be held.
1695 static void call_console_drivers(const char *ext_text, size_t ext_len,
1696 const char *text, size_t len)
1698 struct console *con;
1700 trace_console_rcuidle(text, len);
1702 if (!console_drivers)
1705 for_each_console(con) {
1706 if (exclusive_console && con != exclusive_console)
1708 if (!(con->flags & CON_ENABLED))
1712 if (!cpu_online(smp_processor_id()) &&
1713 !(con->flags & CON_ANYTIME))
1715 if (con->flags & CON_EXTENDED)
1716 con->write(con, ext_text, ext_len);
1718 con->write(con, text, len);
1722 int printk_delay_msec __read_mostly;
1724 static inline void printk_delay(void)
1726 if (unlikely(printk_delay_msec)) {
1727 int m = printk_delay_msec;
1731 touch_nmi_watchdog();
1737 * Continuation lines are buffered, and not committed to the record buffer
1738 * until the line is complete, or a race forces it. The line fragments
1739 * though, are printed immediately to the consoles to ensure everything has
1740 * reached the console in case of a kernel crash.
1742 static struct cont {
1743 char buf[LOG_LINE_MAX];
1744 size_t len; /* length == 0 means unused buffer */
1745 struct task_struct *owner; /* task of first print*/
1746 u64 ts_nsec; /* time of first print */
1747 u8 level; /* log level of first message */
1748 u8 facility; /* log facility of first message */
1749 enum log_flags flags; /* prefix, newline flags */
1752 static void cont_flush(void)
1757 log_store(cont.facility, cont.level, cont.flags, cont.ts_nsec,
1758 NULL, 0, cont.buf, cont.len);
1762 static bool cont_add(int facility, int level, enum log_flags flags, const char *text, size_t len)
1765 * If ext consoles are present, flush and skip in-kernel
1766 * continuation. See nr_ext_console_drivers definition. Also, if
1767 * the line gets too long, split it up in separate records.
1769 if (nr_ext_console_drivers || cont.len + len > sizeof(cont.buf)) {
1775 cont.facility = facility;
1777 cont.owner = current;
1778 cont.ts_nsec = local_clock();
1782 memcpy(cont.buf + cont.len, text, len);
1785 // The original flags come from the first line,
1786 // but later continuations can add a newline.
1787 if (flags & LOG_NEWLINE) {
1788 cont.flags |= LOG_NEWLINE;
1792 if (cont.len > (sizeof(cont.buf) * 80) / 100)
1798 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)
1801 * If an earlier line was buffered, and we're a continuation
1802 * write from the same process, try to add it to the buffer.
1805 if (cont.owner == current && (lflags & LOG_CONT)) {
1806 if (cont_add(facility, level, lflags, text, text_len))
1809 /* Otherwise, make sure it's flushed */
1813 /* Skip empty continuation lines that couldn't be added - they just flush */
1814 if (!text_len && (lflags & LOG_CONT))
1817 /* If it doesn't end in a newline, try to buffer the current line */
1818 if (!(lflags & LOG_NEWLINE)) {
1819 if (cont_add(facility, level, lflags, text, text_len))
1823 /* Store it in the record log */
1824 return log_store(facility, level, lflags, 0, dict, dictlen, text, text_len);
1827 asmlinkage int vprintk_emit(int facility, int level,
1828 const char *dict, size_t dictlen,
1829 const char *fmt, va_list args)
1831 static char textbuf[LOG_LINE_MAX];
1832 char *text = textbuf;
1834 enum log_flags lflags = 0;
1835 unsigned long flags;
1837 bool in_sched = false;
1839 if (level == LOGLEVEL_SCHED) {
1840 level = LOGLEVEL_DEFAULT;
1844 boot_delay_msec(level);
1847 /* This stops the holder of console_sem just where we want him */
1848 logbuf_lock_irqsave(flags);
1850 * The printf needs to come first; we need the syslog
1851 * prefix which might be passed-in as a parameter.
1853 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1855 /* mark and strip a trailing newline */
1856 if (text_len && text[text_len-1] == '\n') {
1858 lflags |= LOG_NEWLINE;
1861 /* strip kernel syslog prefix and extract log level or control flags */
1862 if (facility == 0) {
1865 while ((kern_level = printk_get_level(text)) != 0) {
1866 switch (kern_level) {
1868 if (level == LOGLEVEL_DEFAULT)
1869 level = kern_level - '0';
1871 case 'd': /* KERN_DEFAULT */
1872 lflags |= LOG_PREFIX;
1874 case 'c': /* KERN_CONT */
1883 if (level == LOGLEVEL_DEFAULT)
1884 level = default_message_loglevel;
1887 lflags |= LOG_PREFIX|LOG_NEWLINE;
1889 printed_len = log_output(facility, level, lflags, dict, dictlen, text, text_len);
1891 logbuf_unlock_irqrestore(flags);
1893 /* If called from the scheduler, we can not call up(). */
1896 * Disable preemption to avoid being preempted while holding
1897 * console_sem which would prevent anyone from printing to
1902 * Try to acquire and then immediately release the console
1903 * semaphore. The release will print out buffers and wake up
1904 * /dev/kmsg and syslog() users.
1906 if (console_trylock_spinning())
1914 EXPORT_SYMBOL(vprintk_emit);
1916 asmlinkage int vprintk(const char *fmt, va_list args)
1918 return vprintk_func(fmt, args);
1920 EXPORT_SYMBOL(vprintk);
1922 asmlinkage int printk_emit(int facility, int level,
1923 const char *dict, size_t dictlen,
1924 const char *fmt, ...)
1929 va_start(args, fmt);
1930 r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1935 EXPORT_SYMBOL(printk_emit);
1937 int vprintk_default(const char *fmt, va_list args)
1941 #ifdef CONFIG_KGDB_KDB
1942 /* Allow to pass printk() to kdb but avoid a recursion. */
1943 if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
1944 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
1948 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1952 EXPORT_SYMBOL_GPL(vprintk_default);
1955 * printk - print a kernel message
1956 * @fmt: format string
1958 * This is printk(). It can be called from any context. We want it to work.
1960 * We try to grab the console_lock. If we succeed, it's easy - we log the
1961 * output and call the console drivers. If we fail to get the semaphore, we
1962 * place the output into the log buffer and return. The current holder of
1963 * the console_sem will notice the new output in console_unlock(); and will
1964 * send it to the consoles before releasing the lock.
1966 * One effect of this deferred printing is that code which calls printk() and
1967 * then changes console_loglevel may break. This is because console_loglevel
1968 * is inspected when the actual printing occurs.
1973 * See the vsnprintf() documentation for format string extensions over C99.
1975 asmlinkage __visible int printk(const char *fmt, ...)
1980 va_start(args, fmt);
1981 r = vprintk_func(fmt, args);
1986 EXPORT_SYMBOL(printk);
1988 #else /* CONFIG_PRINTK */
1990 #define LOG_LINE_MAX 0
1991 #define PREFIX_MAX 0
1993 static u64 syslog_seq;
1994 static u32 syslog_idx;
1995 static u64 console_seq;
1996 static u32 console_idx;
1997 static u64 log_first_seq;
1998 static u32 log_first_idx;
1999 static u64 log_next_seq;
2000 static char *log_text(const struct printk_log *msg) { return NULL; }
2001 static char *log_dict(const struct printk_log *msg) { return NULL; }
2002 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2003 static u32 log_next(u32 idx) { return 0; }
2004 static ssize_t msg_print_ext_header(char *buf, size_t size,
2005 struct printk_log *msg,
2006 u64 seq) { return 0; }
2007 static ssize_t msg_print_ext_body(char *buf, size_t size,
2008 char *dict, size_t dict_len,
2009 char *text, size_t text_len) { return 0; }
2010 static void console_lock_spinning_enable(void) { }
2011 static int console_lock_spinning_disable_and_check(void) { return 0; }
2012 static void call_console_drivers(const char *ext_text, size_t ext_len,
2013 const char *text, size_t len) {}
2014 static size_t msg_print_text(const struct printk_log *msg,
2015 bool syslog, char *buf, size_t size) { return 0; }
2016 static bool suppress_message_printing(int level) { return false; }
2018 #endif /* CONFIG_PRINTK */
2020 #ifdef CONFIG_EARLY_PRINTK
2021 struct console *early_console;
2023 asmlinkage __visible void early_printk(const char *fmt, ...)
2033 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2036 early_console->write(early_console, buf, n);
2040 static int __add_preferred_console(char *name, int idx, char *options,
2043 struct console_cmdline *c;
2047 * See if this tty is not yet registered, and
2048 * if we have a slot free.
2050 for (i = 0, c = console_cmdline;
2051 i < MAX_CMDLINECONSOLES && c->name[0];
2053 if (strcmp(c->name, name) == 0 && c->index == idx) {
2055 preferred_console = i;
2059 if (i == MAX_CMDLINECONSOLES)
2062 preferred_console = i;
2063 strlcpy(c->name, name, sizeof(c->name));
2064 c->options = options;
2065 braille_set_options(c, brl_options);
2071 static int __init console_msg_format_setup(char *str)
2073 if (!strcmp(str, "syslog"))
2074 console_msg_format = MSG_FORMAT_SYSLOG;
2075 if (!strcmp(str, "default"))
2076 console_msg_format = MSG_FORMAT_DEFAULT;
2079 __setup("console_msg_format=", console_msg_format_setup);
2082 * Set up a console. Called via do_early_param() in init/main.c
2083 * for each "console=" parameter in the boot command line.
2085 static int __init console_setup(char *str)
2087 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2088 char *s, *options, *brl_options = NULL;
2091 if (_braille_console_setup(&str, &brl_options))
2095 * Decode str into name, index, options.
2097 if (str[0] >= '0' && str[0] <= '9') {
2098 strcpy(buf, "ttyS");
2099 strncpy(buf + 4, str, sizeof(buf) - 5);
2101 strncpy(buf, str, sizeof(buf) - 1);
2103 buf[sizeof(buf) - 1] = 0;
2104 options = strchr(str, ',');
2108 if (!strcmp(str, "ttya"))
2109 strcpy(buf, "ttyS0");
2110 if (!strcmp(str, "ttyb"))
2111 strcpy(buf, "ttyS1");
2113 for (s = buf; *s; s++)
2114 if (isdigit(*s) || *s == ',')
2116 idx = simple_strtoul(s, NULL, 10);
2119 __add_preferred_console(buf, idx, options, brl_options);
2120 console_set_on_cmdline = 1;
2123 __setup("console=", console_setup);
2126 * add_preferred_console - add a device to the list of preferred consoles.
2127 * @name: device name
2128 * @idx: device index
2129 * @options: options for this console
2131 * The last preferred console added will be used for kernel messages
2132 * and stdin/out/err for init. Normally this is used by console_setup
2133 * above to handle user-supplied console arguments; however it can also
2134 * be used by arch-specific code either to override the user or more
2135 * commonly to provide a default console (ie from PROM variables) when
2136 * the user has not supplied one.
2138 int add_preferred_console(char *name, int idx, char *options)
2140 return __add_preferred_console(name, idx, options, NULL);
2143 bool console_suspend_enabled = true;
2144 EXPORT_SYMBOL(console_suspend_enabled);
2146 static int __init console_suspend_disable(char *str)
2148 console_suspend_enabled = false;
2151 __setup("no_console_suspend", console_suspend_disable);
2152 module_param_named(console_suspend, console_suspend_enabled,
2153 bool, S_IRUGO | S_IWUSR);
2154 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2155 " and hibernate operations");
2158 * suspend_console - suspend the console subsystem
2160 * This disables printk() while we go into suspend states
2162 void suspend_console(void)
2164 if (!console_suspend_enabled)
2166 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2168 console_suspended = 1;
2172 void resume_console(void)
2174 if (!console_suspend_enabled)
2177 console_suspended = 0;
2182 * console_cpu_notify - print deferred console messages after CPU hotplug
2185 * If printk() is called from a CPU that is not online yet, the messages
2186 * will be printed on the console only if there are CON_ANYTIME consoles.
2187 * This function is called when a new CPU comes online (or fails to come
2188 * up) or goes offline.
2190 static int console_cpu_notify(unsigned int cpu)
2192 if (!cpuhp_tasks_frozen) {
2193 /* If trylock fails, someone else is doing the printing */
2194 if (console_trylock())
2201 * console_lock - lock the console system for exclusive use.
2203 * Acquires a lock which guarantees that the caller has
2204 * exclusive access to the console system and the console_drivers list.
2206 * Can sleep, returns nothing.
2208 void console_lock(void)
2213 if (console_suspended)
2216 console_may_schedule = 1;
2218 EXPORT_SYMBOL(console_lock);
2221 * console_trylock - try to lock the console system for exclusive use.
2223 * Try to acquire a lock which guarantees that the caller has exclusive
2224 * access to the console system and the console_drivers list.
2226 * returns 1 on success, and 0 on failure to acquire the lock.
2228 int console_trylock(void)
2230 if (down_trylock_console_sem())
2232 if (console_suspended) {
2237 console_may_schedule = 0;
2240 EXPORT_SYMBOL(console_trylock);
2242 int is_console_locked(void)
2244 return console_locked;
2248 * Check if we have any console that is capable of printing while cpu is
2249 * booting or shutting down. Requires console_sem.
2251 static int have_callable_console(void)
2253 struct console *con;
2255 for_each_console(con)
2256 if ((con->flags & CON_ENABLED) &&
2257 (con->flags & CON_ANYTIME))
2264 * Can we actually use the console at this time on this cpu?
2266 * Console drivers may assume that per-cpu resources have been allocated. So
2267 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2268 * call them until this CPU is officially up.
2270 static inline int can_use_console(void)
2272 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2276 * console_unlock - unlock the console system
2278 * Releases the console_lock which the caller holds on the console system
2279 * and the console driver list.
2281 * While the console_lock was held, console output may have been buffered
2282 * by printk(). If this is the case, console_unlock(); emits
2283 * the output prior to releasing the lock.
2285 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2287 * console_unlock(); may be called from any context.
2289 void console_unlock(void)
2291 static char ext_text[CONSOLE_EXT_LOG_MAX];
2292 static char text[LOG_LINE_MAX + PREFIX_MAX];
2293 unsigned long flags;
2294 bool do_cond_resched, retry;
2296 if (console_suspended) {
2302 * Console drivers are called with interrupts disabled, so
2303 * @console_may_schedule should be cleared before; however, we may
2304 * end up dumping a lot of lines, for example, if called from
2305 * console registration path, and should invoke cond_resched()
2306 * between lines if allowable. Not doing so can cause a very long
2307 * scheduling stall on a slow console leading to RCU stall and
2308 * softlockup warnings which exacerbate the issue with more
2309 * messages practically incapacitating the system.
2311 * console_trylock() is not able to detect the preemptive
2312 * context reliably. Therefore the value must be stored before
2313 * and cleared after the the "again" goto label.
2315 do_cond_resched = console_may_schedule;
2317 console_may_schedule = 0;
2320 * We released the console_sem lock, so we need to recheck if
2321 * cpu is online and (if not) is there at least one CON_ANYTIME
2324 if (!can_use_console()) {
2331 struct printk_log *msg;
2335 printk_safe_enter_irqsave(flags);
2336 raw_spin_lock(&logbuf_lock);
2337 if (console_seq < log_first_seq) {
2338 len = sprintf(text, "** %u printk messages dropped **\n",
2339 (unsigned)(log_first_seq - console_seq));
2341 /* messages are gone, move to first one */
2342 console_seq = log_first_seq;
2343 console_idx = log_first_idx;
2348 if (console_seq == log_next_seq)
2351 msg = log_from_idx(console_idx);
2352 if (suppress_message_printing(msg->level)) {
2354 * Skip record we have buffered and already printed
2355 * directly to the console when we received it, and
2356 * record that has level above the console loglevel.
2358 console_idx = log_next(console_idx);
2363 len += msg_print_text(msg,
2364 console_msg_format & MSG_FORMAT_SYSLOG,
2366 sizeof(text) - len);
2367 if (nr_ext_console_drivers) {
2368 ext_len = msg_print_ext_header(ext_text,
2371 ext_len += msg_print_ext_body(ext_text + ext_len,
2372 sizeof(ext_text) - ext_len,
2373 log_dict(msg), msg->dict_len,
2374 log_text(msg), msg->text_len);
2376 console_idx = log_next(console_idx);
2378 raw_spin_unlock(&logbuf_lock);
2381 * While actively printing out messages, if another printk()
2382 * were to occur on another CPU, it may wait for this one to
2383 * finish. This task can not be preempted if there is a
2384 * waiter waiting to take over.
2386 console_lock_spinning_enable();
2388 stop_critical_timings(); /* don't trace print latency */
2389 call_console_drivers(ext_text, ext_len, text, len);
2390 start_critical_timings();
2392 if (console_lock_spinning_disable_and_check()) {
2393 printk_safe_exit_irqrestore(flags);
2397 printk_safe_exit_irqrestore(flags);
2399 if (do_cond_resched)
2405 /* Release the exclusive_console once it is used */
2406 if (unlikely(exclusive_console))
2407 exclusive_console = NULL;
2409 raw_spin_unlock(&logbuf_lock);
2414 * Someone could have filled up the buffer again, so re-check if there's
2415 * something to flush. In case we cannot trylock the console_sem again,
2416 * there's a new owner and the console_unlock() from them will do the
2417 * flush, no worries.
2419 raw_spin_lock(&logbuf_lock);
2420 retry = console_seq != log_next_seq;
2421 raw_spin_unlock(&logbuf_lock);
2422 printk_safe_exit_irqrestore(flags);
2424 if (retry && console_trylock())
2427 EXPORT_SYMBOL(console_unlock);
2430 * console_conditional_schedule - yield the CPU if required
2432 * If the console code is currently allowed to sleep, and
2433 * if this CPU should yield the CPU to another task, do
2436 * Must be called within console_lock();.
2438 void __sched console_conditional_schedule(void)
2440 if (console_may_schedule)
2443 EXPORT_SYMBOL(console_conditional_schedule);
2445 void console_unblank(void)
2450 * console_unblank can no longer be called in interrupt context unless
2451 * oops_in_progress is set to 1..
2453 if (oops_in_progress) {
2454 if (down_trylock_console_sem() != 0)
2460 console_may_schedule = 0;
2462 if ((c->flags & CON_ENABLED) && c->unblank)
2468 * console_flush_on_panic - flush console content on panic
2470 * Immediately output all pending messages no matter what.
2472 void console_flush_on_panic(void)
2475 * If someone else is holding the console lock, trylock will fail
2476 * and may_schedule may be set. Ignore and proceed to unlock so
2477 * that messages are flushed out. As this can be called from any
2478 * context and we don't want to get preempted while flushing,
2479 * ensure may_schedule is cleared.
2482 console_may_schedule = 0;
2487 * Return the console tty driver structure and its associated index
2489 struct tty_driver *console_device(int *index)
2492 struct tty_driver *driver = NULL;
2495 for_each_console(c) {
2498 driver = c->device(c, index);
2507 * Prevent further output on the passed console device so that (for example)
2508 * serial drivers can disable console output before suspending a port, and can
2509 * re-enable output afterwards.
2511 void console_stop(struct console *console)
2514 console->flags &= ~CON_ENABLED;
2517 EXPORT_SYMBOL(console_stop);
2519 void console_start(struct console *console)
2522 console->flags |= CON_ENABLED;
2525 EXPORT_SYMBOL(console_start);
2527 static int __read_mostly keep_bootcon;
2529 static int __init keep_bootcon_setup(char *str)
2532 pr_info("debug: skip boot console de-registration.\n");
2537 early_param("keep_bootcon", keep_bootcon_setup);
2540 * The console driver calls this routine during kernel initialization
2541 * to register the console printing procedure with printk() and to
2542 * print any messages that were printed by the kernel before the
2543 * console driver was initialized.
2545 * This can happen pretty early during the boot process (because of
2546 * early_printk) - sometimes before setup_arch() completes - be careful
2547 * of what kernel features are used - they may not be initialised yet.
2549 * There are two types of consoles - bootconsoles (early_printk) and
2550 * "real" consoles (everything which is not a bootconsole) which are
2551 * handled differently.
2552 * - Any number of bootconsoles can be registered at any time.
2553 * - As soon as a "real" console is registered, all bootconsoles
2554 * will be unregistered automatically.
2555 * - Once a "real" console is registered, any attempt to register a
2556 * bootconsoles will be rejected
2558 void register_console(struct console *newcon)
2561 unsigned long flags;
2562 struct console *bcon = NULL;
2563 struct console_cmdline *c;
2564 static bool has_preferred;
2566 if (console_drivers)
2567 for_each_console(bcon)
2568 if (WARN(bcon == newcon,
2569 "console '%s%d' already registered\n",
2570 bcon->name, bcon->index))
2574 * before we register a new CON_BOOT console, make sure we don't
2575 * already have a valid console
2577 if (console_drivers && newcon->flags & CON_BOOT) {
2578 /* find the last or real console */
2579 for_each_console(bcon) {
2580 if (!(bcon->flags & CON_BOOT)) {
2581 pr_info("Too late to register bootconsole %s%d\n",
2582 newcon->name, newcon->index);
2588 if (console_drivers && console_drivers->flags & CON_BOOT)
2589 bcon = console_drivers;
2591 if (!has_preferred || bcon || !console_drivers)
2592 has_preferred = preferred_console >= 0;
2595 * See if we want to use this console driver. If we
2596 * didn't select a console we take the first one
2597 * that registers here.
2599 if (!has_preferred) {
2600 if (newcon->index < 0)
2602 if (newcon->setup == NULL ||
2603 newcon->setup(newcon, NULL) == 0) {
2604 newcon->flags |= CON_ENABLED;
2605 if (newcon->device) {
2606 newcon->flags |= CON_CONSDEV;
2607 has_preferred = true;
2613 * See if this console matches one we selected on
2616 for (i = 0, c = console_cmdline;
2617 i < MAX_CMDLINECONSOLES && c->name[0];
2619 if (!newcon->match ||
2620 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2621 /* default matching */
2622 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2623 if (strcmp(c->name, newcon->name) != 0)
2625 if (newcon->index >= 0 &&
2626 newcon->index != c->index)
2628 if (newcon->index < 0)
2629 newcon->index = c->index;
2631 if (_braille_register_console(newcon, c))
2634 if (newcon->setup &&
2635 newcon->setup(newcon, c->options) != 0)
2639 newcon->flags |= CON_ENABLED;
2640 if (i == preferred_console) {
2641 newcon->flags |= CON_CONSDEV;
2642 has_preferred = true;
2647 if (!(newcon->flags & CON_ENABLED))
2651 * If we have a bootconsole, and are switching to a real console,
2652 * don't print everything out again, since when the boot console, and
2653 * the real console are the same physical device, it's annoying to
2654 * see the beginning boot messages twice
2656 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2657 newcon->flags &= ~CON_PRINTBUFFER;
2660 * Put this console in the list - keep the
2661 * preferred driver at the head of the list.
2664 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2665 newcon->next = console_drivers;
2666 console_drivers = newcon;
2668 newcon->next->flags &= ~CON_CONSDEV;
2670 newcon->next = console_drivers->next;
2671 console_drivers->next = newcon;
2674 if (newcon->flags & CON_EXTENDED)
2675 if (!nr_ext_console_drivers++)
2676 pr_info("printk: continuation disabled due to ext consoles, expect more fragments in /dev/kmsg\n");
2678 if (newcon->flags & CON_PRINTBUFFER) {
2680 * console_unlock(); will print out the buffered messages
2683 logbuf_lock_irqsave(flags);
2684 console_seq = syslog_seq;
2685 console_idx = syslog_idx;
2686 logbuf_unlock_irqrestore(flags);
2688 * We're about to replay the log buffer. Only do this to the
2689 * just-registered console to avoid excessive message spam to
2690 * the already-registered consoles.
2692 exclusive_console = newcon;
2695 console_sysfs_notify();
2698 * By unregistering the bootconsoles after we enable the real console
2699 * we get the "console xxx enabled" message on all the consoles -
2700 * boot consoles, real consoles, etc - this is to ensure that end
2701 * users know there might be something in the kernel's log buffer that
2702 * went to the bootconsole (that they do not see on the real console)
2704 pr_info("%sconsole [%s%d] enabled\n",
2705 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2706 newcon->name, newcon->index);
2708 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2710 /* We need to iterate through all boot consoles, to make
2711 * sure we print everything out, before we unregister them.
2713 for_each_console(bcon)
2714 if (bcon->flags & CON_BOOT)
2715 unregister_console(bcon);
2718 EXPORT_SYMBOL(register_console);
2720 int unregister_console(struct console *console)
2722 struct console *a, *b;
2725 pr_info("%sconsole [%s%d] disabled\n",
2726 (console->flags & CON_BOOT) ? "boot" : "" ,
2727 console->name, console->index);
2729 res = _braille_unregister_console(console);
2735 if (console_drivers == console) {
2736 console_drivers=console->next;
2738 } else if (console_drivers) {
2739 for (a=console_drivers->next, b=console_drivers ;
2740 a; b=a, a=b->next) {
2749 if (!res && (console->flags & CON_EXTENDED))
2750 nr_ext_console_drivers--;
2753 * If this isn't the last console and it has CON_CONSDEV set, we
2754 * need to set it on the next preferred console.
2756 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2757 console_drivers->flags |= CON_CONSDEV;
2759 console->flags &= ~CON_ENABLED;
2761 console_sysfs_notify();
2764 EXPORT_SYMBOL(unregister_console);
2767 * Initialize the console device. This is called *early*, so
2768 * we can't necessarily depend on lots of kernel help here.
2769 * Just do some early initializations, and do the complex setup
2772 void __init console_init(void)
2777 /* Setup the default TTY line discipline. */
2781 * set up the console device so that later boot sequences can
2782 * inform about problems etc..
2784 call = __con_initcall_start;
2785 trace_initcall_level("console");
2786 while (call < __con_initcall_end) {
2787 trace_initcall_start((*call));
2789 trace_initcall_finish((*call), ret);
2795 * Some boot consoles access data that is in the init section and which will
2796 * be discarded after the initcalls have been run. To make sure that no code
2797 * will access this data, unregister the boot consoles in a late initcall.
2799 * If for some reason, such as deferred probe or the driver being a loadable
2800 * module, the real console hasn't registered yet at this point, there will
2801 * be a brief interval in which no messages are logged to the console, which
2802 * makes it difficult to diagnose problems that occur during this time.
2804 * To mitigate this problem somewhat, only unregister consoles whose memory
2805 * intersects with the init section. Note that all other boot consoles will
2806 * get unregistred when the real preferred console is registered.
2808 static int __init printk_late_init(void)
2810 struct console *con;
2813 for_each_console(con) {
2814 if (!(con->flags & CON_BOOT))
2817 /* Check addresses that might be used for enabled consoles. */
2818 if (init_section_intersects(con, sizeof(*con)) ||
2819 init_section_contains(con->write, 0) ||
2820 init_section_contains(con->read, 0) ||
2821 init_section_contains(con->device, 0) ||
2822 init_section_contains(con->unblank, 0) ||
2823 init_section_contains(con->data, 0)) {
2825 * Please, consider moving the reported consoles out
2826 * of the init section.
2828 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2829 con->name, con->index);
2830 unregister_console(con);
2833 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2834 console_cpu_notify);
2836 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2837 console_cpu_notify, NULL);
2841 late_initcall(printk_late_init);
2843 #if defined CONFIG_PRINTK
2845 * Delayed printk version, for scheduler-internal messages:
2847 #define PRINTK_PENDING_WAKEUP 0x01
2848 #define PRINTK_PENDING_OUTPUT 0x02
2850 static DEFINE_PER_CPU(int, printk_pending);
2852 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2854 int pending = __this_cpu_xchg(printk_pending, 0);
2856 if (pending & PRINTK_PENDING_OUTPUT) {
2857 /* If trylock fails, someone else is doing the printing */
2858 if (console_trylock())
2862 if (pending & PRINTK_PENDING_WAKEUP)
2863 wake_up_interruptible(&log_wait);
2866 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2867 .func = wake_up_klogd_work_func,
2868 .flags = IRQ_WORK_LAZY,
2871 void wake_up_klogd(void)
2874 if (waitqueue_active(&log_wait)) {
2875 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2876 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2881 int vprintk_deferred(const char *fmt, va_list args)
2885 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2888 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2889 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2895 int printk_deferred(const char *fmt, ...)
2900 va_start(args, fmt);
2901 r = vprintk_deferred(fmt, args);
2908 * printk rate limiting, lifted from the networking subsystem.
2910 * This enforces a rate limit: not more than 10 kernel messages
2911 * every 5s to make a denial-of-service attack impossible.
2913 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2915 int __printk_ratelimit(const char *func)
2917 return ___ratelimit(&printk_ratelimit_state, func);
2919 EXPORT_SYMBOL(__printk_ratelimit);
2922 * printk_timed_ratelimit - caller-controlled printk ratelimiting
2923 * @caller_jiffies: pointer to caller's state
2924 * @interval_msecs: minimum interval between prints
2926 * printk_timed_ratelimit() returns true if more than @interval_msecs
2927 * milliseconds have elapsed since the last time printk_timed_ratelimit()
2930 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
2931 unsigned int interval_msecs)
2933 unsigned long elapsed = jiffies - *caller_jiffies;
2935 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
2938 *caller_jiffies = jiffies;
2941 EXPORT_SYMBOL(printk_timed_ratelimit);
2943 static DEFINE_SPINLOCK(dump_list_lock);
2944 static LIST_HEAD(dump_list);
2947 * kmsg_dump_register - register a kernel log dumper.
2948 * @dumper: pointer to the kmsg_dumper structure
2950 * Adds a kernel log dumper to the system. The dump callback in the
2951 * structure will be called when the kernel oopses or panics and must be
2952 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
2954 int kmsg_dump_register(struct kmsg_dumper *dumper)
2956 unsigned long flags;
2959 /* The dump callback needs to be set */
2963 spin_lock_irqsave(&dump_list_lock, flags);
2964 /* Don't allow registering multiple times */
2965 if (!dumper->registered) {
2966 dumper->registered = 1;
2967 list_add_tail_rcu(&dumper->list, &dump_list);
2970 spin_unlock_irqrestore(&dump_list_lock, flags);
2974 EXPORT_SYMBOL_GPL(kmsg_dump_register);
2977 * kmsg_dump_unregister - unregister a kmsg dumper.
2978 * @dumper: pointer to the kmsg_dumper structure
2980 * Removes a dump device from the system. Returns zero on success and
2981 * %-EINVAL otherwise.
2983 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
2985 unsigned long flags;
2988 spin_lock_irqsave(&dump_list_lock, flags);
2989 if (dumper->registered) {
2990 dumper->registered = 0;
2991 list_del_rcu(&dumper->list);
2994 spin_unlock_irqrestore(&dump_list_lock, flags);
2999 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3001 static bool always_kmsg_dump;
3002 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3005 * kmsg_dump - dump kernel log to kernel message dumpers.
3006 * @reason: the reason (oops, panic etc) for dumping
3008 * Call each of the registered dumper's dump() callback, which can
3009 * retrieve the kmsg records with kmsg_dump_get_line() or
3010 * kmsg_dump_get_buffer().
3012 void kmsg_dump(enum kmsg_dump_reason reason)
3014 struct kmsg_dumper *dumper;
3015 unsigned long flags;
3017 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3021 list_for_each_entry_rcu(dumper, &dump_list, list) {
3022 if (dumper->max_reason && reason > dumper->max_reason)
3025 /* initialize iterator with data about the stored records */
3026 dumper->active = true;
3028 logbuf_lock_irqsave(flags);
3029 dumper->cur_seq = clear_seq;
3030 dumper->cur_idx = clear_idx;
3031 dumper->next_seq = log_next_seq;
3032 dumper->next_idx = log_next_idx;
3033 logbuf_unlock_irqrestore(flags);
3035 /* invoke dumper which will iterate over records */
3036 dumper->dump(dumper, reason);
3038 /* reset iterator */
3039 dumper->active = false;
3045 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3046 * @dumper: registered kmsg dumper
3047 * @syslog: include the "<4>" prefixes
3048 * @line: buffer to copy the line to
3049 * @size: maximum size of the buffer
3050 * @len: length of line placed into buffer
3052 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3053 * record, and copy one record into the provided buffer.
3055 * Consecutive calls will return the next available record moving
3056 * towards the end of the buffer with the youngest messages.
3058 * A return value of FALSE indicates that there are no more records to
3061 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3063 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3064 char *line, size_t size, size_t *len)
3066 struct printk_log *msg;
3070 if (!dumper->active)
3073 if (dumper->cur_seq < log_first_seq) {
3074 /* messages are gone, move to first available one */
3075 dumper->cur_seq = log_first_seq;
3076 dumper->cur_idx = log_first_idx;
3080 if (dumper->cur_seq >= log_next_seq)
3083 msg = log_from_idx(dumper->cur_idx);
3084 l = msg_print_text(msg, syslog, line, size);
3086 dumper->cur_idx = log_next(dumper->cur_idx);
3096 * kmsg_dump_get_line - retrieve one kmsg log line
3097 * @dumper: registered kmsg dumper
3098 * @syslog: include the "<4>" prefixes
3099 * @line: buffer to copy the line to
3100 * @size: maximum size of the buffer
3101 * @len: length of line placed into buffer
3103 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3104 * record, and copy one record into the provided buffer.
3106 * Consecutive calls will return the next available record moving
3107 * towards the end of the buffer with the youngest messages.
3109 * A return value of FALSE indicates that there are no more records to
3112 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3113 char *line, size_t size, size_t *len)
3115 unsigned long flags;
3118 logbuf_lock_irqsave(flags);
3119 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3120 logbuf_unlock_irqrestore(flags);
3124 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3127 * kmsg_dump_get_buffer - copy kmsg log lines
3128 * @dumper: registered kmsg dumper
3129 * @syslog: include the "<4>" prefixes
3130 * @buf: buffer to copy the line to
3131 * @size: maximum size of the buffer
3132 * @len: length of line placed into buffer
3134 * Start at the end of the kmsg buffer and fill the provided buffer
3135 * with as many of the the *youngest* kmsg records that fit into it.
3136 * If the buffer is large enough, all available kmsg records will be
3137 * copied with a single call.
3139 * Consecutive calls will fill the buffer with the next block of
3140 * available older records, not including the earlier retrieved ones.
3142 * A return value of FALSE indicates that there are no more records to
3145 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3146 char *buf, size_t size, size_t *len)
3148 unsigned long flags;
3156 if (!dumper->active)
3159 logbuf_lock_irqsave(flags);
3160 if (dumper->cur_seq < log_first_seq) {
3161 /* messages are gone, move to first available one */
3162 dumper->cur_seq = log_first_seq;
3163 dumper->cur_idx = log_first_idx;
3167 if (dumper->cur_seq >= dumper->next_seq) {
3168 logbuf_unlock_irqrestore(flags);
3172 /* calculate length of entire buffer */
3173 seq = dumper->cur_seq;
3174 idx = dumper->cur_idx;
3175 while (seq < dumper->next_seq) {
3176 struct printk_log *msg = log_from_idx(idx);
3178 l += msg_print_text(msg, true, NULL, 0);
3179 idx = log_next(idx);
3183 /* move first record forward until length fits into the buffer */
3184 seq = dumper->cur_seq;
3185 idx = dumper->cur_idx;
3186 while (l > size && seq < dumper->next_seq) {
3187 struct printk_log *msg = log_from_idx(idx);
3189 l -= msg_print_text(msg, true, NULL, 0);
3190 idx = log_next(idx);
3194 /* last message in next interation */
3199 while (seq < dumper->next_seq) {
3200 struct printk_log *msg = log_from_idx(idx);
3202 l += msg_print_text(msg, syslog, buf + l, size - l);
3203 idx = log_next(idx);
3207 dumper->next_seq = next_seq;
3208 dumper->next_idx = next_idx;
3210 logbuf_unlock_irqrestore(flags);
3216 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3219 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3220 * @dumper: registered kmsg dumper
3222 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3223 * kmsg_dump_get_buffer() can be called again and used multiple
3224 * times within the same dumper.dump() callback.
3226 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3228 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3230 dumper->cur_seq = clear_seq;
3231 dumper->cur_idx = clear_idx;
3232 dumper->next_seq = log_next_seq;
3233 dumper->next_idx = log_next_idx;
3237 * kmsg_dump_rewind - reset the interator
3238 * @dumper: registered kmsg dumper
3240 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3241 * kmsg_dump_get_buffer() can be called again and used multiple
3242 * times within the same dumper.dump() callback.
3244 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3246 unsigned long flags;
3248 logbuf_lock_irqsave(flags);
3249 kmsg_dump_rewind_nolock(dumper);
3250 logbuf_unlock_irqrestore(flags);
3252 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);