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/kexec.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/utsname.h>
46 #include <linux/ctype.h>
47 #include <linux/uio.h>
49 #include <asm/uaccess.h>
50 #include <asm/sections.h>
52 #define CREATE_TRACE_POINTS
53 #include <trace/events/printk.h>
55 #include "console_cmdline.h"
59 int console_printk[4] = {
60 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
61 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
62 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
63 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
67 * Low level drivers may need that to know if they can schedule in
68 * their unblank() callback or not. So let's export it.
71 EXPORT_SYMBOL(oops_in_progress);
74 * console_sem protects the console_drivers list, and also
75 * provides serialisation for access to the entire console
78 static DEFINE_SEMAPHORE(console_sem);
79 struct console *console_drivers;
80 EXPORT_SYMBOL_GPL(console_drivers);
83 static struct lockdep_map console_lock_dep_map = {
84 .name = "console_lock"
88 enum devkmsg_log_bits {
89 __DEVKMSG_LOG_BIT_ON = 0,
90 __DEVKMSG_LOG_BIT_OFF,
91 __DEVKMSG_LOG_BIT_LOCK,
94 enum devkmsg_log_masks {
95 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
96 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
97 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
100 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
101 #define DEVKMSG_LOG_MASK_DEFAULT 0
103 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
105 static int __control_devkmsg(char *str)
110 if (!strncmp(str, "on", 2)) {
111 devkmsg_log = DEVKMSG_LOG_MASK_ON;
113 } else if (!strncmp(str, "off", 3)) {
114 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
116 } else if (!strncmp(str, "ratelimit", 9)) {
117 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
123 static int __init control_devkmsg(char *str)
125 if (__control_devkmsg(str) < 0)
129 * Set sysctl string accordingly:
131 if (devkmsg_log == DEVKMSG_LOG_MASK_ON) {
132 memset(devkmsg_log_str, 0, DEVKMSG_STR_MAX_SIZE);
133 strncpy(devkmsg_log_str, "on", 2);
134 } else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF) {
135 memset(devkmsg_log_str, 0, DEVKMSG_STR_MAX_SIZE);
136 strncpy(devkmsg_log_str, "off", 3);
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)
216 if (down_trylock(&console_sem))
218 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
221 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
223 #define up_console_sem() do { \
224 mutex_release(&console_lock_dep_map, 1, _RET_IP_);\
229 * This is used for debugging the mess that is the VT code by
230 * keeping track if we have the console semaphore held. It's
231 * definitely not the perfect debug tool (we don't know if _WE_
232 * hold it and are racing, but it helps tracking those weird code
233 * paths in the console code where we end up in places I want
234 * locked without the console sempahore held).
236 static int console_locked, console_suspended;
239 * If exclusive_console is non-NULL then only this console is to be printed to.
241 static struct console *exclusive_console;
244 * Array of consoles built from command line options (console=)
247 #define MAX_CMDLINECONSOLES 8
249 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
251 static int selected_console = -1;
252 static int preferred_console = -1;
253 int console_set_on_cmdline;
254 EXPORT_SYMBOL(console_set_on_cmdline);
256 /* Flag: console code may call schedule() */
257 static int console_may_schedule;
260 * The printk log buffer consists of a chain of concatenated variable
261 * length records. Every record starts with a record header, containing
262 * the overall length of the record.
264 * The heads to the first and last entry in the buffer, as well as the
265 * sequence numbers of these entries are maintained when messages are
268 * If the heads indicate available messages, the length in the header
269 * tells the start next message. A length == 0 for the next message
270 * indicates a wrap-around to the beginning of the buffer.
272 * Every record carries the monotonic timestamp in microseconds, as well as
273 * the standard userspace syslog level and syslog facility. The usual
274 * kernel messages use LOG_KERN; userspace-injected messages always carry
275 * a matching syslog facility, by default LOG_USER. The origin of every
276 * message can be reliably determined that way.
278 * The human readable log message directly follows the message header. The
279 * length of the message text is stored in the header, the stored message
282 * Optionally, a message can carry a dictionary of properties (key/value pairs),
283 * to provide userspace with a machine-readable message context.
285 * Examples for well-defined, commonly used property names are:
286 * DEVICE=b12:8 device identifier
290 * +sound:card0 subsystem:devname
291 * SUBSYSTEM=pci driver-core subsystem name
293 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
294 * follows directly after a '=' character. Every property is terminated by
295 * a '\0' character. The last property is not terminated.
297 * Example of a message structure:
298 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
299 * 0008 34 00 record is 52 bytes long
300 * 000a 0b 00 text is 11 bytes long
301 * 000c 1f 00 dictionary is 23 bytes long
302 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
303 * 0010 69 74 27 73 20 61 20 6c "it's a l"
305 * 001b 44 45 56 49 43 "DEVIC"
306 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
307 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
309 * 0032 00 00 00 padding to next message header
311 * The 'struct printk_log' buffer header must never be directly exported to
312 * userspace, it is a kernel-private implementation detail that might
313 * need to be changed in the future, when the requirements change.
315 * /dev/kmsg exports the structured data in the following line format:
316 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
318 * Users of the export format should ignore possible additional values
319 * separated by ',', and find the message after the ';' character.
321 * The optional key/value pairs are attached as continuation lines starting
322 * with a space character and terminated by a newline. All possible
323 * non-prinatable characters are escaped in the "\xff" notation.
327 LOG_NOCONS = 1, /* already flushed, do not print to console */
328 LOG_NEWLINE = 2, /* text ended with a newline */
329 LOG_PREFIX = 4, /* text started with a prefix */
330 LOG_CONT = 8, /* text is a fragment of a continuation line */
334 u64 ts_nsec; /* timestamp in nanoseconds */
335 u16 len; /* length of entire record */
336 u16 text_len; /* length of text buffer */
337 u16 dict_len; /* length of dictionary buffer */
338 u8 facility; /* syslog facility */
339 u8 flags:5; /* internal record flags */
340 u8 level:3; /* syslog level */
342 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
343 __packed __aligned(4)
348 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
349 * within the scheduler's rq lock. It must be released before calling
350 * console_unlock() or anything else that might wake up a process.
352 DEFINE_RAW_SPINLOCK(logbuf_lock);
355 DECLARE_WAIT_QUEUE_HEAD(log_wait);
356 /* the next printk record to read by syslog(READ) or /proc/kmsg */
357 static u64 syslog_seq;
358 static u32 syslog_idx;
359 static enum log_flags syslog_prev;
360 static size_t syslog_partial;
362 /* index and sequence number of the first record stored in the buffer */
363 static u64 log_first_seq;
364 static u32 log_first_idx;
366 /* index and sequence number of the next record to store in the buffer */
367 static u64 log_next_seq;
368 static u32 log_next_idx;
370 /* the next printk record to write to the console */
371 static u64 console_seq;
372 static u32 console_idx;
373 static enum log_flags console_prev;
375 /* the next printk record to read after the last 'clear' command */
376 static u64 clear_seq;
377 static u32 clear_idx;
379 #define PREFIX_MAX 32
380 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
382 #define LOG_LEVEL(v) ((v) & 0x07)
383 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
386 #define LOG_ALIGN __alignof__(struct printk_log)
387 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
388 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
389 static char *log_buf = __log_buf;
390 static u32 log_buf_len = __LOG_BUF_LEN;
392 /* Return log buffer address */
393 char *log_buf_addr_get(void)
398 /* Return log buffer size */
399 u32 log_buf_len_get(void)
404 /* human readable text of the record */
405 static char *log_text(const struct printk_log *msg)
407 return (char *)msg + sizeof(struct printk_log);
410 /* optional key/value pair dictionary attached to the record */
411 static char *log_dict(const struct printk_log *msg)
413 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
416 /* get record by index; idx must point to valid msg */
417 static struct printk_log *log_from_idx(u32 idx)
419 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
422 * A length == 0 record is the end of buffer marker. Wrap around and
423 * read the message at the start of the buffer.
426 return (struct printk_log *)log_buf;
430 /* get next record; idx must point to valid msg */
431 static u32 log_next(u32 idx)
433 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
435 /* length == 0 indicates the end of the buffer; wrap */
437 * A length == 0 record is the end of buffer marker. Wrap around and
438 * read the message at the start of the buffer as *this* one, and
439 * return the one after that.
442 msg = (struct printk_log *)log_buf;
445 return idx + msg->len;
449 * Check whether there is enough free space for the given message.
451 * The same values of first_idx and next_idx mean that the buffer
452 * is either empty or full.
454 * If the buffer is empty, we must respect the position of the indexes.
455 * They cannot be reset to the beginning of the buffer.
457 static int logbuf_has_space(u32 msg_size, bool empty)
461 if (log_next_idx > log_first_idx || empty)
462 free = max(log_buf_len - log_next_idx, log_first_idx);
464 free = log_first_idx - log_next_idx;
467 * We need space also for an empty header that signalizes wrapping
470 return free >= msg_size + sizeof(struct printk_log);
473 static int log_make_free_space(u32 msg_size)
475 while (log_first_seq < log_next_seq &&
476 !logbuf_has_space(msg_size, false)) {
477 /* drop old messages until we have enough contiguous space */
478 log_first_idx = log_next(log_first_idx);
482 if (clear_seq < log_first_seq) {
483 clear_seq = log_first_seq;
484 clear_idx = log_first_idx;
487 /* sequence numbers are equal, so the log buffer is empty */
488 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
494 /* compute the message size including the padding bytes */
495 static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
499 size = sizeof(struct printk_log) + text_len + dict_len;
500 *pad_len = (-size) & (LOG_ALIGN - 1);
507 * Define how much of the log buffer we could take at maximum. The value
508 * must be greater than two. Note that only half of the buffer is available
509 * when the index points to the middle.
511 #define MAX_LOG_TAKE_PART 4
512 static const char trunc_msg[] = "<truncated>";
514 static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
515 u16 *dict_len, u32 *pad_len)
518 * The message should not take the whole buffer. Otherwise, it might
519 * get removed too soon.
521 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
522 if (*text_len > max_text_len)
523 *text_len = max_text_len;
524 /* enable the warning message */
525 *trunc_msg_len = strlen(trunc_msg);
526 /* disable the "dict" completely */
528 /* compute the size again, count also the warning message */
529 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
532 /* insert record into the buffer, discard old ones, update heads */
533 static int log_store(int facility, int level,
534 enum log_flags flags, u64 ts_nsec,
535 const char *dict, u16 dict_len,
536 const char *text, u16 text_len)
538 struct printk_log *msg;
540 u16 trunc_msg_len = 0;
542 /* number of '\0' padding bytes to next message */
543 size = msg_used_size(text_len, dict_len, &pad_len);
545 if (log_make_free_space(size)) {
546 /* truncate the message if it is too long for empty buffer */
547 size = truncate_msg(&text_len, &trunc_msg_len,
548 &dict_len, &pad_len);
549 /* survive when the log buffer is too small for trunc_msg */
550 if (log_make_free_space(size))
554 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
556 * This message + an additional empty header does not fit
557 * at the end of the buffer. Add an empty header with len == 0
558 * to signify a wrap around.
560 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
565 msg = (struct printk_log *)(log_buf + log_next_idx);
566 memcpy(log_text(msg), text, text_len);
567 msg->text_len = text_len;
569 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
570 msg->text_len += trunc_msg_len;
572 memcpy(log_dict(msg), dict, dict_len);
573 msg->dict_len = dict_len;
574 msg->facility = facility;
575 msg->level = level & 7;
576 msg->flags = flags & 0x1f;
578 msg->ts_nsec = ts_nsec;
580 msg->ts_nsec = local_clock();
581 memset(log_dict(msg) + dict_len, 0, pad_len);
585 log_next_idx += msg->len;
588 return msg->text_len;
591 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
593 static int syslog_action_restricted(int type)
598 * Unless restricted, we allow "read all" and "get buffer size"
601 return type != SYSLOG_ACTION_READ_ALL &&
602 type != SYSLOG_ACTION_SIZE_BUFFER;
605 int check_syslog_permissions(int type, int source)
608 * If this is from /proc/kmsg and we've already opened it, then we've
609 * already done the capabilities checks at open time.
611 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
614 if (syslog_action_restricted(type)) {
615 if (capable(CAP_SYSLOG))
618 * For historical reasons, accept CAP_SYS_ADMIN too, with
621 if (capable(CAP_SYS_ADMIN)) {
622 pr_warn_once("%s (%d): Attempt to access syslog with "
623 "CAP_SYS_ADMIN but no CAP_SYSLOG "
625 current->comm, task_pid_nr(current));
631 return security_syslog(type);
633 EXPORT_SYMBOL_GPL(check_syslog_permissions);
635 static void append_char(char **pp, char *e, char c)
641 static ssize_t msg_print_ext_header(char *buf, size_t size,
642 struct printk_log *msg, u64 seq,
643 enum log_flags prev_flags)
645 u64 ts_usec = msg->ts_nsec;
648 do_div(ts_usec, 1000);
651 * If we couldn't merge continuation line fragments during the print,
652 * export the stored flags to allow an optional external merge of the
653 * records. Merging the records isn't always neccessarily correct, like
654 * when we hit a race during printing. In most cases though, it produces
655 * better readable output. 'c' in the record flags mark the first
656 * fragment of a line, '+' the following.
658 if (msg->flags & LOG_CONT)
659 cont = (prev_flags & LOG_CONT) ? '+' : 'c';
661 return scnprintf(buf, size, "%u,%llu,%llu,%c;",
662 (msg->facility << 3) | msg->level, seq, ts_usec, cont);
665 static ssize_t msg_print_ext_body(char *buf, size_t size,
666 char *dict, size_t dict_len,
667 char *text, size_t text_len)
669 char *p = buf, *e = buf + size;
672 /* escape non-printable characters */
673 for (i = 0; i < text_len; i++) {
674 unsigned char c = text[i];
676 if (c < ' ' || c >= 127 || c == '\\')
677 p += scnprintf(p, e - p, "\\x%02x", c);
679 append_char(&p, e, c);
681 append_char(&p, e, '\n');
686 for (i = 0; i < dict_len; i++) {
687 unsigned char c = dict[i];
690 append_char(&p, e, ' ');
695 append_char(&p, e, '\n');
700 if (c < ' ' || c >= 127 || c == '\\') {
701 p += scnprintf(p, e - p, "\\x%02x", c);
705 append_char(&p, e, c);
707 append_char(&p, e, '\n');
713 /* /dev/kmsg - userspace message inject/listen interface */
714 struct devkmsg_user {
718 struct ratelimit_state rs;
720 char buf[CONSOLE_EXT_LOG_MAX];
723 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
726 int level = default_message_loglevel;
727 int facility = 1; /* LOG_USER */
728 struct file *file = iocb->ki_filp;
729 struct devkmsg_user *user = file->private_data;
730 size_t len = iov_iter_count(from);
733 if (!user || len > LOG_LINE_MAX)
736 /* Ignore when user logging is disabled. */
737 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
740 /* Ratelimit when not explicitly enabled. */
741 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
742 if (!___ratelimit(&user->rs, current->comm))
746 buf = kmalloc(len+1, GFP_KERNEL);
751 if (copy_from_iter(buf, len, from) != len) {
757 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
758 * the decimal value represents 32bit, the lower 3 bit are the log
759 * level, the rest are the log facility.
761 * If no prefix or no userspace facility is specified, we
762 * enforce LOG_USER, to be able to reliably distinguish
763 * kernel-generated messages from userspace-injected ones.
766 if (line[0] == '<') {
770 u = simple_strtoul(line + 1, &endp, 10);
771 if (endp && endp[0] == '>') {
772 level = LOG_LEVEL(u);
773 if (LOG_FACILITY(u) != 0)
774 facility = LOG_FACILITY(u);
781 printk_emit(facility, level, NULL, 0, "%s", line);
786 static ssize_t devkmsg_read(struct file *file, char __user *buf,
787 size_t count, loff_t *ppos)
789 struct devkmsg_user *user = file->private_data;
790 struct printk_log *msg;
797 ret = mutex_lock_interruptible(&user->lock);
800 raw_spin_lock_irq(&logbuf_lock);
801 while (user->seq == log_next_seq) {
802 if (file->f_flags & O_NONBLOCK) {
804 raw_spin_unlock_irq(&logbuf_lock);
808 raw_spin_unlock_irq(&logbuf_lock);
809 ret = wait_event_interruptible(log_wait,
810 user->seq != log_next_seq);
813 raw_spin_lock_irq(&logbuf_lock);
816 if (user->seq < log_first_seq) {
817 /* our last seen message is gone, return error and reset */
818 user->idx = log_first_idx;
819 user->seq = log_first_seq;
821 raw_spin_unlock_irq(&logbuf_lock);
825 msg = log_from_idx(user->idx);
826 len = msg_print_ext_header(user->buf, sizeof(user->buf),
827 msg, user->seq, user->prev);
828 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
829 log_dict(msg), msg->dict_len,
830 log_text(msg), msg->text_len);
832 user->prev = msg->flags;
833 user->idx = log_next(user->idx);
835 raw_spin_unlock_irq(&logbuf_lock);
842 if (copy_to_user(buf, user->buf, len)) {
848 mutex_unlock(&user->lock);
852 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
854 struct devkmsg_user *user = file->private_data;
862 raw_spin_lock_irq(&logbuf_lock);
865 /* the first record */
866 user->idx = log_first_idx;
867 user->seq = log_first_seq;
871 * The first record after the last SYSLOG_ACTION_CLEAR,
872 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
873 * changes no global state, and does not clear anything.
875 user->idx = clear_idx;
876 user->seq = clear_seq;
879 /* after the last record */
880 user->idx = log_next_idx;
881 user->seq = log_next_seq;
886 raw_spin_unlock_irq(&logbuf_lock);
890 static unsigned int devkmsg_poll(struct file *file, poll_table *wait)
892 struct devkmsg_user *user = file->private_data;
896 return POLLERR|POLLNVAL;
898 poll_wait(file, &log_wait, wait);
900 raw_spin_lock_irq(&logbuf_lock);
901 if (user->seq < log_next_seq) {
902 /* return error when data has vanished underneath us */
903 if (user->seq < log_first_seq)
904 ret = POLLIN|POLLRDNORM|POLLERR|POLLPRI;
906 ret = POLLIN|POLLRDNORM;
908 raw_spin_unlock_irq(&logbuf_lock);
913 static int devkmsg_open(struct inode *inode, struct file *file)
915 struct devkmsg_user *user;
918 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
921 /* write-only does not need any file context */
922 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
923 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
929 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
933 ratelimit_default_init(&user->rs);
934 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
936 mutex_init(&user->lock);
938 raw_spin_lock_irq(&logbuf_lock);
939 user->idx = log_first_idx;
940 user->seq = log_first_seq;
941 raw_spin_unlock_irq(&logbuf_lock);
943 file->private_data = user;
947 static int devkmsg_release(struct inode *inode, struct file *file)
949 struct devkmsg_user *user = file->private_data;
954 ratelimit_state_exit(&user->rs);
956 mutex_destroy(&user->lock);
961 const struct file_operations kmsg_fops = {
962 .open = devkmsg_open,
963 .read = devkmsg_read,
964 .write_iter = devkmsg_write,
965 .llseek = devkmsg_llseek,
966 .poll = devkmsg_poll,
967 .release = devkmsg_release,
970 #ifdef CONFIG_KEXEC_CORE
972 * This appends the listed symbols to /proc/vmcore
974 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
975 * obtain access to symbols that are otherwise very difficult to locate. These
976 * symbols are specifically used so that utilities can access and extract the
977 * dmesg log from a vmcore file after a crash.
979 void log_buf_kexec_setup(void)
981 VMCOREINFO_SYMBOL(log_buf);
982 VMCOREINFO_SYMBOL(log_buf_len);
983 VMCOREINFO_SYMBOL(log_first_idx);
984 VMCOREINFO_SYMBOL(clear_idx);
985 VMCOREINFO_SYMBOL(log_next_idx);
987 * Export struct printk_log size and field offsets. User space tools can
988 * parse it and detect any changes to structure down the line.
990 VMCOREINFO_STRUCT_SIZE(printk_log);
991 VMCOREINFO_OFFSET(printk_log, ts_nsec);
992 VMCOREINFO_OFFSET(printk_log, len);
993 VMCOREINFO_OFFSET(printk_log, text_len);
994 VMCOREINFO_OFFSET(printk_log, dict_len);
998 /* requested log_buf_len from kernel cmdline */
999 static unsigned long __initdata new_log_buf_len;
1001 /* we practice scaling the ring buffer by powers of 2 */
1002 static void __init log_buf_len_update(unsigned size)
1005 size = roundup_pow_of_two(size);
1006 if (size > log_buf_len)
1007 new_log_buf_len = size;
1010 /* save requested log_buf_len since it's too early to process it */
1011 static int __init log_buf_len_setup(char *str)
1013 unsigned size = memparse(str, &str);
1015 log_buf_len_update(size);
1019 early_param("log_buf_len", log_buf_len_setup);
1022 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1024 static void __init log_buf_add_cpu(void)
1026 unsigned int cpu_extra;
1029 * archs should set up cpu_possible_bits properly with
1030 * set_cpu_possible() after setup_arch() but just in
1031 * case lets ensure this is valid.
1033 if (num_possible_cpus() == 1)
1036 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1038 /* by default this will only continue through for large > 64 CPUs */
1039 if (cpu_extra <= __LOG_BUF_LEN / 2)
1042 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1043 __LOG_CPU_MAX_BUF_LEN);
1044 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1046 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1048 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1050 #else /* !CONFIG_SMP */
1051 static inline void log_buf_add_cpu(void) {}
1052 #endif /* CONFIG_SMP */
1054 void __init setup_log_buf(int early)
1056 unsigned long flags;
1060 if (log_buf != __log_buf)
1063 if (!early && !new_log_buf_len)
1066 if (!new_log_buf_len)
1071 memblock_virt_alloc(new_log_buf_len, LOG_ALIGN);
1073 new_log_buf = memblock_virt_alloc_nopanic(new_log_buf_len,
1077 if (unlikely(!new_log_buf)) {
1078 pr_err("log_buf_len: %ld bytes not available\n",
1083 raw_spin_lock_irqsave(&logbuf_lock, flags);
1084 log_buf_len = new_log_buf_len;
1085 log_buf = new_log_buf;
1086 new_log_buf_len = 0;
1087 free = __LOG_BUF_LEN - log_next_idx;
1088 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1089 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
1091 pr_info("log_buf_len: %d bytes\n", log_buf_len);
1092 pr_info("early log buf free: %d(%d%%)\n",
1093 free, (free * 100) / __LOG_BUF_LEN);
1096 static bool __read_mostly ignore_loglevel;
1098 static int __init ignore_loglevel_setup(char *str)
1100 ignore_loglevel = true;
1101 pr_info("debug: ignoring loglevel setting.\n");
1106 early_param("ignore_loglevel", ignore_loglevel_setup);
1107 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1108 MODULE_PARM_DESC(ignore_loglevel,
1109 "ignore loglevel setting (prints all kernel messages to the console)");
1111 static bool suppress_message_printing(int level)
1113 return (level >= console_loglevel && !ignore_loglevel);
1116 #ifdef CONFIG_BOOT_PRINTK_DELAY
1118 static int boot_delay; /* msecs delay after each printk during bootup */
1119 static unsigned long long loops_per_msec; /* based on boot_delay */
1121 static int __init boot_delay_setup(char *str)
1125 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1126 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1128 get_option(&str, &boot_delay);
1129 if (boot_delay > 10 * 1000)
1132 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1133 "HZ: %d, loops_per_msec: %llu\n",
1134 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1137 early_param("boot_delay", boot_delay_setup);
1139 static void boot_delay_msec(int level)
1141 unsigned long long k;
1142 unsigned long timeout;
1144 if ((boot_delay == 0 || system_state != SYSTEM_BOOTING)
1145 || suppress_message_printing(level)) {
1149 k = (unsigned long long)loops_per_msec * boot_delay;
1151 timeout = jiffies + msecs_to_jiffies(boot_delay);
1156 * use (volatile) jiffies to prevent
1157 * compiler reduction; loop termination via jiffies
1158 * is secondary and may or may not happen.
1160 if (time_after(jiffies, timeout))
1162 touch_nmi_watchdog();
1166 static inline void boot_delay_msec(int level)
1171 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1172 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1174 static size_t print_time(u64 ts, char *buf)
1176 unsigned long rem_nsec;
1181 rem_nsec = do_div(ts, 1000000000);
1184 return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
1186 return sprintf(buf, "[%5lu.%06lu] ",
1187 (unsigned long)ts, rem_nsec / 1000);
1190 static size_t print_prefix(const struct printk_log *msg, bool syslog, char *buf)
1193 unsigned int prefix = (msg->facility << 3) | msg->level;
1197 len += sprintf(buf, "<%u>", prefix);
1202 else if (prefix > 99)
1204 else if (prefix > 9)
1209 len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
1213 static size_t msg_print_text(const struct printk_log *msg, enum log_flags prev,
1214 bool syslog, char *buf, size_t size)
1216 const char *text = log_text(msg);
1217 size_t text_size = msg->text_len;
1219 bool newline = true;
1222 if ((prev & LOG_CONT) && !(msg->flags & LOG_PREFIX))
1225 if (msg->flags & LOG_CONT) {
1226 if ((prev & LOG_CONT) && !(prev & LOG_NEWLINE))
1229 if (!(msg->flags & LOG_NEWLINE))
1234 const char *next = memchr(text, '\n', text_size);
1238 text_len = next - text;
1240 text_size -= next - text;
1242 text_len = text_size;
1246 if (print_prefix(msg, syslog, NULL) +
1247 text_len + 1 >= size - len)
1251 len += print_prefix(msg, syslog, buf + len);
1252 memcpy(buf + len, text, text_len);
1254 if (next || newline)
1257 /* SYSLOG_ACTION_* buffer size only calculation */
1259 len += print_prefix(msg, syslog, NULL);
1261 if (next || newline)
1272 static int syslog_print(char __user *buf, int size)
1275 struct printk_log *msg;
1278 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1286 raw_spin_lock_irq(&logbuf_lock);
1287 if (syslog_seq < log_first_seq) {
1288 /* messages are gone, move to first one */
1289 syslog_seq = log_first_seq;
1290 syslog_idx = log_first_idx;
1294 if (syslog_seq == log_next_seq) {
1295 raw_spin_unlock_irq(&logbuf_lock);
1299 skip = syslog_partial;
1300 msg = log_from_idx(syslog_idx);
1301 n = msg_print_text(msg, syslog_prev, true, text,
1302 LOG_LINE_MAX + PREFIX_MAX);
1303 if (n - syslog_partial <= size) {
1304 /* message fits into buffer, move forward */
1305 syslog_idx = log_next(syslog_idx);
1307 syslog_prev = msg->flags;
1308 n -= syslog_partial;
1311 /* partial read(), remember position */
1313 syslog_partial += n;
1316 raw_spin_unlock_irq(&logbuf_lock);
1321 if (copy_to_user(buf, text + skip, n)) {
1336 static int syslog_print_all(char __user *buf, int size, bool clear)
1341 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1345 raw_spin_lock_irq(&logbuf_lock);
1350 enum log_flags prev;
1353 * Find first record that fits, including all following records,
1354 * into the user-provided buffer for this dump.
1359 while (seq < log_next_seq) {
1360 struct printk_log *msg = log_from_idx(idx);
1362 len += msg_print_text(msg, prev, true, NULL, 0);
1364 idx = log_next(idx);
1368 /* move first record forward until length fits into the buffer */
1372 while (len > size && seq < log_next_seq) {
1373 struct printk_log *msg = log_from_idx(idx);
1375 len -= msg_print_text(msg, prev, true, NULL, 0);
1377 idx = log_next(idx);
1381 /* last message fitting into this dump */
1382 next_seq = log_next_seq;
1385 while (len >= 0 && seq < next_seq) {
1386 struct printk_log *msg = log_from_idx(idx);
1389 textlen = msg_print_text(msg, prev, true, text,
1390 LOG_LINE_MAX + PREFIX_MAX);
1395 idx = log_next(idx);
1399 raw_spin_unlock_irq(&logbuf_lock);
1400 if (copy_to_user(buf + len, text, textlen))
1404 raw_spin_lock_irq(&logbuf_lock);
1406 if (seq < log_first_seq) {
1407 /* messages are gone, move to next one */
1408 seq = log_first_seq;
1409 idx = log_first_idx;
1416 clear_seq = log_next_seq;
1417 clear_idx = log_next_idx;
1419 raw_spin_unlock_irq(&logbuf_lock);
1425 int do_syslog(int type, char __user *buf, int len, int source)
1428 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1431 error = check_syslog_permissions(type, source);
1436 case SYSLOG_ACTION_CLOSE: /* Close log */
1438 case SYSLOG_ACTION_OPEN: /* Open log */
1440 case SYSLOG_ACTION_READ: /* Read from log */
1442 if (!buf || len < 0)
1447 if (!access_ok(VERIFY_WRITE, buf, len)) {
1451 error = wait_event_interruptible(log_wait,
1452 syslog_seq != log_next_seq);
1455 error = syslog_print(buf, len);
1457 /* Read/clear last kernel messages */
1458 case SYSLOG_ACTION_READ_CLEAR:
1461 /* Read last kernel messages */
1462 case SYSLOG_ACTION_READ_ALL:
1464 if (!buf || len < 0)
1469 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:
1495 if (len < 1 || len > 8)
1497 if (len < minimum_console_loglevel)
1498 len = minimum_console_loglevel;
1499 console_loglevel = len;
1500 /* Implicitly re-enable logging to console */
1501 saved_console_loglevel = LOGLEVEL_DEFAULT;
1504 /* Number of chars in the log buffer */
1505 case SYSLOG_ACTION_SIZE_UNREAD:
1506 raw_spin_lock_irq(&logbuf_lock);
1507 if (syslog_seq < log_first_seq) {
1508 /* messages are gone, move to first one */
1509 syslog_seq = log_first_seq;
1510 syslog_idx = log_first_idx;
1514 if (source == SYSLOG_FROM_PROC) {
1516 * Short-cut for poll(/"proc/kmsg") which simply checks
1517 * for pending data, not the size; return the count of
1518 * records, not the length.
1520 error = log_next_seq - syslog_seq;
1522 u64 seq = syslog_seq;
1523 u32 idx = syslog_idx;
1524 enum log_flags prev = syslog_prev;
1527 while (seq < log_next_seq) {
1528 struct printk_log *msg = log_from_idx(idx);
1530 error += msg_print_text(msg, prev, true, NULL, 0);
1531 idx = log_next(idx);
1535 error -= syslog_partial;
1537 raw_spin_unlock_irq(&logbuf_lock);
1539 /* Size of the log buffer */
1540 case SYSLOG_ACTION_SIZE_BUFFER:
1541 error = log_buf_len;
1551 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1553 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1557 * Call the console drivers, asking them to write out
1558 * log_buf[start] to log_buf[end - 1].
1559 * The console_lock must be held.
1561 static void call_console_drivers(int level,
1562 const char *ext_text, size_t ext_len,
1563 const char *text, size_t len)
1565 struct console *con;
1567 trace_console(text, len);
1569 if (!console_drivers)
1572 for_each_console(con) {
1573 if (exclusive_console && con != exclusive_console)
1575 if (!(con->flags & CON_ENABLED))
1579 if (!cpu_online(smp_processor_id()) &&
1580 !(con->flags & CON_ANYTIME))
1582 if (con->flags & CON_EXTENDED)
1583 con->write(con, ext_text, ext_len);
1585 con->write(con, text, len);
1590 * Zap console related locks when oopsing.
1591 * To leave time for slow consoles to print a full oops,
1592 * only zap at most once every 30 seconds.
1594 static void zap_locks(void)
1596 static unsigned long oops_timestamp;
1598 if (time_after_eq(jiffies, oops_timestamp) &&
1599 !time_after(jiffies, oops_timestamp + 30 * HZ))
1602 oops_timestamp = jiffies;
1605 /* If a crash is occurring, make sure we can't deadlock */
1606 raw_spin_lock_init(&logbuf_lock);
1607 /* And make sure that we print immediately */
1608 sema_init(&console_sem, 1);
1611 int printk_delay_msec __read_mostly;
1613 static inline void printk_delay(void)
1615 if (unlikely(printk_delay_msec)) {
1616 int m = printk_delay_msec;
1620 touch_nmi_watchdog();
1626 * Continuation lines are buffered, and not committed to the record buffer
1627 * until the line is complete, or a race forces it. The line fragments
1628 * though, are printed immediately to the consoles to ensure everything has
1629 * reached the console in case of a kernel crash.
1631 static struct cont {
1632 char buf[LOG_LINE_MAX];
1633 size_t len; /* length == 0 means unused buffer */
1634 size_t cons; /* bytes written to console */
1635 struct task_struct *owner; /* task of first print*/
1636 u64 ts_nsec; /* time of first print */
1637 u8 level; /* log level of first message */
1638 u8 facility; /* log facility of first message */
1639 enum log_flags flags; /* prefix, newline flags */
1640 bool flushed:1; /* buffer sealed and committed */
1643 static void cont_flush(void)
1651 * If a fragment of this line was directly flushed to the
1652 * console; wait for the console to pick up the rest of the
1653 * line. LOG_NOCONS suppresses a duplicated output.
1655 log_store(cont.facility, cont.level, cont.flags | LOG_NOCONS,
1656 cont.ts_nsec, NULL, 0, cont.buf, cont.len);
1657 cont.flushed = true;
1660 * If no fragment of this line ever reached the console,
1661 * just submit it to the store and free the buffer.
1663 log_store(cont.facility, cont.level, cont.flags, 0,
1664 NULL, 0, cont.buf, cont.len);
1669 static bool cont_add(int facility, int level, enum log_flags flags, const char *text, size_t len)
1671 if (cont.len && cont.flushed)
1675 * If ext consoles are present, flush and skip in-kernel
1676 * continuation. See nr_ext_console_drivers definition. Also, if
1677 * the line gets too long, split it up in separate records.
1679 if (nr_ext_console_drivers || cont.len + len > sizeof(cont.buf)) {
1685 cont.facility = facility;
1687 cont.owner = current;
1688 cont.ts_nsec = local_clock();
1691 cont.flushed = false;
1694 memcpy(cont.buf + cont.len, text, len);
1697 // The original flags come from the first line,
1698 // but later continuations can add a newline.
1699 if (flags & LOG_NEWLINE) {
1700 cont.flags |= LOG_NEWLINE;
1704 if (cont.len > (sizeof(cont.buf) * 80) / 100)
1710 static size_t cont_print_text(char *text, size_t size)
1715 if (cont.cons == 0 && (console_prev & LOG_NEWLINE)) {
1716 textlen += print_time(cont.ts_nsec, text);
1720 len = cont.len - cont.cons;
1724 memcpy(text + textlen, cont.buf + cont.cons, len);
1726 cont.cons = cont.len;
1730 if (cont.flags & LOG_NEWLINE)
1731 text[textlen++] = '\n';
1732 /* got everything, release buffer */
1738 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)
1741 * If an earlier line was buffered, and we're a continuation
1742 * write from the same process, try to add it to the buffer.
1745 if (cont.owner == current && (lflags & LOG_CONT)) {
1746 if (cont_add(facility, level, lflags, text, text_len))
1749 /* Otherwise, make sure it's flushed */
1753 /* If it doesn't end in a newline, try to buffer the current line */
1754 if (!(lflags & LOG_NEWLINE)) {
1755 if (cont_add(facility, level, lflags, text, text_len))
1759 /* Store it in the record log */
1760 return log_store(facility, level, lflags, 0, dict, dictlen, text, text_len);
1763 asmlinkage int vprintk_emit(int facility, int level,
1764 const char *dict, size_t dictlen,
1765 const char *fmt, va_list args)
1767 static bool recursion_bug;
1768 static char textbuf[LOG_LINE_MAX];
1769 char *text = textbuf;
1770 size_t text_len = 0;
1771 enum log_flags lflags = 0;
1772 unsigned long flags;
1774 int printed_len = 0;
1775 int nmi_message_lost;
1776 bool in_sched = false;
1777 /* cpu currently holding logbuf_lock in this function */
1778 static unsigned int logbuf_cpu = UINT_MAX;
1780 if (level == LOGLEVEL_SCHED) {
1781 level = LOGLEVEL_DEFAULT;
1785 boot_delay_msec(level);
1788 local_irq_save(flags);
1789 this_cpu = smp_processor_id();
1792 * Ouch, printk recursed into itself!
1794 if (unlikely(logbuf_cpu == this_cpu)) {
1796 * If a crash is occurring during printk() on this CPU,
1797 * then try to get the crash message out but make sure
1798 * we can't deadlock. Otherwise just return to avoid the
1799 * recursion and return - but flag the recursion so that
1800 * it can be printed at the next appropriate moment:
1802 if (!oops_in_progress && !lockdep_recursing(current)) {
1803 recursion_bug = true;
1804 local_irq_restore(flags);
1811 /* This stops the holder of console_sem just where we want him */
1812 raw_spin_lock(&logbuf_lock);
1813 logbuf_cpu = this_cpu;
1815 if (unlikely(recursion_bug)) {
1816 static const char recursion_msg[] =
1817 "BUG: recent printk recursion!";
1819 recursion_bug = false;
1820 /* emit KERN_CRIT message */
1821 printed_len += log_store(0, 2, LOG_PREFIX|LOG_NEWLINE, 0,
1822 NULL, 0, recursion_msg,
1823 strlen(recursion_msg));
1826 nmi_message_lost = get_nmi_message_lost();
1827 if (unlikely(nmi_message_lost)) {
1828 text_len = scnprintf(textbuf, sizeof(textbuf),
1829 "BAD LUCK: lost %d message(s) from NMI context!",
1831 printed_len += log_store(0, 2, LOG_PREFIX|LOG_NEWLINE, 0,
1832 NULL, 0, textbuf, text_len);
1836 * The printf needs to come first; we need the syslog
1837 * prefix which might be passed-in as a parameter.
1839 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1841 /* mark and strip a trailing newline */
1842 if (text_len && text[text_len-1] == '\n') {
1844 lflags |= LOG_NEWLINE;
1847 /* strip kernel syslog prefix and extract log level or control flags */
1848 if (facility == 0) {
1851 while ((kern_level = printk_get_level(text)) != 0) {
1852 switch (kern_level) {
1854 if (level == LOGLEVEL_DEFAULT)
1855 level = kern_level - '0';
1857 case 'd': /* KERN_DEFAULT */
1858 lflags |= LOG_PREFIX;
1860 case 'c': /* KERN_CONT */
1869 if (level == LOGLEVEL_DEFAULT)
1870 level = default_message_loglevel;
1873 lflags |= LOG_PREFIX|LOG_NEWLINE;
1875 printed_len += log_output(facility, level, lflags, dict, dictlen, text, text_len);
1877 logbuf_cpu = UINT_MAX;
1878 raw_spin_unlock(&logbuf_lock);
1880 local_irq_restore(flags);
1882 /* If called from the scheduler, we can not call up(). */
1886 * Try to acquire and then immediately release the console
1887 * semaphore. The release will print out buffers and wake up
1888 * /dev/kmsg and syslog() users.
1890 if (console_trylock())
1897 EXPORT_SYMBOL(vprintk_emit);
1899 asmlinkage int vprintk(const char *fmt, va_list args)
1901 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1903 EXPORT_SYMBOL(vprintk);
1905 asmlinkage int printk_emit(int facility, int level,
1906 const char *dict, size_t dictlen,
1907 const char *fmt, ...)
1912 va_start(args, fmt);
1913 r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1918 EXPORT_SYMBOL(printk_emit);
1920 int vprintk_default(const char *fmt, va_list args)
1924 #ifdef CONFIG_KGDB_KDB
1925 if (unlikely(kdb_trap_printk)) {
1926 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
1930 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1934 EXPORT_SYMBOL_GPL(vprintk_default);
1937 * printk - print a kernel message
1938 * @fmt: format string
1940 * This is printk(). It can be called from any context. We want it to work.
1942 * We try to grab the console_lock. If we succeed, it's easy - we log the
1943 * output and call the console drivers. If we fail to get the semaphore, we
1944 * place the output into the log buffer and return. The current holder of
1945 * the console_sem will notice the new output in console_unlock(); and will
1946 * send it to the consoles before releasing the lock.
1948 * One effect of this deferred printing is that code which calls printk() and
1949 * then changes console_loglevel may break. This is because console_loglevel
1950 * is inspected when the actual printing occurs.
1955 * See the vsnprintf() documentation for format string extensions over C99.
1957 asmlinkage __visible int printk(const char *fmt, ...)
1962 va_start(args, fmt);
1963 r = vprintk_func(fmt, args);
1968 EXPORT_SYMBOL(printk);
1970 #else /* CONFIG_PRINTK */
1972 #define LOG_LINE_MAX 0
1973 #define PREFIX_MAX 0
1975 static u64 syslog_seq;
1976 static u32 syslog_idx;
1977 static u64 console_seq;
1978 static u32 console_idx;
1979 static enum log_flags syslog_prev;
1980 static u64 log_first_seq;
1981 static u32 log_first_idx;
1982 static u64 log_next_seq;
1983 static enum log_flags console_prev;
1984 static struct cont {
1990 static char *log_text(const struct printk_log *msg) { return NULL; }
1991 static char *log_dict(const struct printk_log *msg) { return NULL; }
1992 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
1993 static u32 log_next(u32 idx) { return 0; }
1994 static ssize_t msg_print_ext_header(char *buf, size_t size,
1995 struct printk_log *msg, u64 seq,
1996 enum log_flags prev_flags) { return 0; }
1997 static ssize_t msg_print_ext_body(char *buf, size_t size,
1998 char *dict, size_t dict_len,
1999 char *text, size_t text_len) { return 0; }
2000 static void call_console_drivers(int level,
2001 const char *ext_text, size_t ext_len,
2002 const char *text, size_t len) {}
2003 static size_t msg_print_text(const struct printk_log *msg, enum log_flags prev,
2004 bool syslog, char *buf, size_t size) { return 0; }
2005 static size_t cont_print_text(char *text, size_t size) { return 0; }
2006 static bool suppress_message_printing(int level) { return false; }
2008 /* Still needs to be defined for users */
2009 DEFINE_PER_CPU(printk_func_t, printk_func);
2011 #endif /* CONFIG_PRINTK */
2013 #ifdef CONFIG_EARLY_PRINTK
2014 struct console *early_console;
2016 asmlinkage __visible void early_printk(const char *fmt, ...)
2026 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2029 early_console->write(early_console, buf, n);
2033 static int __add_preferred_console(char *name, int idx, char *options,
2036 struct console_cmdline *c;
2040 * See if this tty is not yet registered, and
2041 * if we have a slot free.
2043 for (i = 0, c = console_cmdline;
2044 i < MAX_CMDLINECONSOLES && c->name[0];
2046 if (strcmp(c->name, name) == 0 && c->index == idx) {
2048 selected_console = i;
2052 if (i == MAX_CMDLINECONSOLES)
2055 selected_console = i;
2056 strlcpy(c->name, name, sizeof(c->name));
2057 c->options = options;
2058 braille_set_options(c, brl_options);
2064 * Set up a console. Called via do_early_param() in init/main.c
2065 * for each "console=" parameter in the boot command line.
2067 static int __init console_setup(char *str)
2069 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2070 char *s, *options, *brl_options = NULL;
2073 if (_braille_console_setup(&str, &brl_options))
2077 * Decode str into name, index, options.
2079 if (str[0] >= '0' && str[0] <= '9') {
2080 strcpy(buf, "ttyS");
2081 strncpy(buf + 4, str, sizeof(buf) - 5);
2083 strncpy(buf, str, sizeof(buf) - 1);
2085 buf[sizeof(buf) - 1] = 0;
2086 options = strchr(str, ',');
2090 if (!strcmp(str, "ttya"))
2091 strcpy(buf, "ttyS0");
2092 if (!strcmp(str, "ttyb"))
2093 strcpy(buf, "ttyS1");
2095 for (s = buf; *s; s++)
2096 if (isdigit(*s) || *s == ',')
2098 idx = simple_strtoul(s, NULL, 10);
2101 __add_preferred_console(buf, idx, options, brl_options);
2102 console_set_on_cmdline = 1;
2105 __setup("console=", console_setup);
2108 * add_preferred_console - add a device to the list of preferred consoles.
2109 * @name: device name
2110 * @idx: device index
2111 * @options: options for this console
2113 * The last preferred console added will be used for kernel messages
2114 * and stdin/out/err for init. Normally this is used by console_setup
2115 * above to handle user-supplied console arguments; however it can also
2116 * be used by arch-specific code either to override the user or more
2117 * commonly to provide a default console (ie from PROM variables) when
2118 * the user has not supplied one.
2120 int add_preferred_console(char *name, int idx, char *options)
2122 return __add_preferred_console(name, idx, options, NULL);
2125 bool console_suspend_enabled = true;
2126 EXPORT_SYMBOL(console_suspend_enabled);
2128 static int __init console_suspend_disable(char *str)
2130 console_suspend_enabled = false;
2133 __setup("no_console_suspend", console_suspend_disable);
2134 module_param_named(console_suspend, console_suspend_enabled,
2135 bool, S_IRUGO | S_IWUSR);
2136 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2137 " and hibernate operations");
2140 * suspend_console - suspend the console subsystem
2142 * This disables printk() while we go into suspend states
2144 void suspend_console(void)
2146 if (!console_suspend_enabled)
2148 printk("Suspending console(s) (use no_console_suspend to debug)\n");
2150 console_suspended = 1;
2154 void resume_console(void)
2156 if (!console_suspend_enabled)
2159 console_suspended = 0;
2164 * console_cpu_notify - print deferred console messages after CPU hotplug
2165 * @self: notifier struct
2166 * @action: CPU hotplug event
2169 * If printk() is called from a CPU that is not online yet, the messages
2170 * will be spooled but will not show up on the console. This function is
2171 * called when a new CPU comes online (or fails to come up), and ensures
2172 * that any such output gets printed.
2174 static int console_cpu_notify(struct notifier_block *self,
2175 unsigned long action, void *hcpu)
2180 case CPU_DOWN_FAILED:
2181 case CPU_UP_CANCELED:
2189 * console_lock - lock the console system for exclusive use.
2191 * Acquires a lock which guarantees that the caller has
2192 * exclusive access to the console system and the console_drivers list.
2194 * Can sleep, returns nothing.
2196 void console_lock(void)
2201 if (console_suspended)
2204 console_may_schedule = 1;
2206 EXPORT_SYMBOL(console_lock);
2209 * console_trylock - try to lock the console system for exclusive use.
2211 * Try to acquire a lock which guarantees that the caller has exclusive
2212 * access to the console system and the console_drivers list.
2214 * returns 1 on success, and 0 on failure to acquire the lock.
2216 int console_trylock(void)
2218 if (down_trylock_console_sem())
2220 if (console_suspended) {
2226 * When PREEMPT_COUNT disabled we can't reliably detect if it's
2227 * safe to schedule (e.g. calling printk while holding a spin_lock),
2228 * because preempt_disable()/preempt_enable() are just barriers there
2229 * and preempt_count() is always 0.
2231 * RCU read sections have a separate preemption counter when
2232 * PREEMPT_RCU enabled thus we must take extra care and check
2233 * rcu_preempt_depth(), otherwise RCU read sections modify
2236 console_may_schedule = !oops_in_progress &&
2238 !rcu_preempt_depth();
2241 EXPORT_SYMBOL(console_trylock);
2243 int is_console_locked(void)
2245 return console_locked;
2249 * Check if we have any console that is capable of printing while cpu is
2250 * booting or shutting down. Requires console_sem.
2252 static int have_callable_console(void)
2254 struct console *con;
2256 for_each_console(con)
2257 if ((con->flags & CON_ENABLED) &&
2258 (con->flags & CON_ANYTIME))
2265 * Can we actually use the console at this time on this cpu?
2267 * Console drivers may assume that per-cpu resources have been allocated. So
2268 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2269 * call them until this CPU is officially up.
2271 static inline int can_use_console(void)
2273 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2276 static void console_cont_flush(char *text, size_t size)
2278 unsigned long flags;
2281 raw_spin_lock_irqsave(&logbuf_lock, flags);
2286 if (suppress_message_printing(cont.level)) {
2287 cont.cons = cont.len;
2294 * We still queue earlier records, likely because the console was
2295 * busy. The earlier ones need to be printed before this one, we
2296 * did not flush any fragment so far, so just let it queue up.
2298 if (console_seq < log_next_seq && !cont.cons)
2301 len = cont_print_text(text, size);
2302 raw_spin_unlock(&logbuf_lock);
2303 stop_critical_timings();
2304 call_console_drivers(cont.level, NULL, 0, text, len);
2305 start_critical_timings();
2306 local_irq_restore(flags);
2309 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2313 * console_unlock - unlock the console system
2315 * Releases the console_lock which the caller holds on the console system
2316 * and the console driver list.
2318 * While the console_lock was held, console output may have been buffered
2319 * by printk(). If this is the case, console_unlock(); emits
2320 * the output prior to releasing the lock.
2322 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2324 * console_unlock(); may be called from any context.
2326 void console_unlock(void)
2328 static char ext_text[CONSOLE_EXT_LOG_MAX];
2329 static char text[LOG_LINE_MAX + PREFIX_MAX];
2330 static u64 seen_seq;
2331 unsigned long flags;
2332 bool wake_klogd = false;
2333 bool do_cond_resched, retry;
2335 if (console_suspended) {
2341 * Console drivers are called under logbuf_lock, so
2342 * @console_may_schedule should be cleared before; however, we may
2343 * end up dumping a lot of lines, for example, if called from
2344 * console registration path, and should invoke cond_resched()
2345 * between lines if allowable. Not doing so can cause a very long
2346 * scheduling stall on a slow console leading to RCU stall and
2347 * softlockup warnings which exacerbate the issue with more
2348 * messages practically incapacitating the system.
2350 do_cond_resched = console_may_schedule;
2351 console_may_schedule = 0;
2355 * We released the console_sem lock, so we need to recheck if
2356 * cpu is online and (if not) is there at least one CON_ANYTIME
2359 if (!can_use_console()) {
2365 /* flush buffered message fragment immediately to console */
2366 console_cont_flush(text, sizeof(text));
2369 struct printk_log *msg;
2374 raw_spin_lock_irqsave(&logbuf_lock, flags);
2375 if (seen_seq != log_next_seq) {
2377 seen_seq = log_next_seq;
2380 if (console_seq < log_first_seq) {
2381 len = sprintf(text, "** %u printk messages dropped ** ",
2382 (unsigned)(log_first_seq - console_seq));
2384 /* messages are gone, move to first one */
2385 console_seq = log_first_seq;
2386 console_idx = log_first_idx;
2392 if (console_seq == log_next_seq)
2395 msg = log_from_idx(console_idx);
2397 if ((msg->flags & LOG_NOCONS) ||
2398 suppress_message_printing(level)) {
2400 * Skip record we have buffered and already printed
2401 * directly to the console when we received it, and
2402 * record that has level above the console loglevel.
2404 console_idx = log_next(console_idx);
2407 * We will get here again when we register a new
2408 * CON_PRINTBUFFER console. Clear the flag so we
2409 * will properly dump everything later.
2411 msg->flags &= ~LOG_NOCONS;
2412 console_prev = msg->flags;
2416 len += msg_print_text(msg, console_prev, false,
2417 text + len, sizeof(text) - len);
2418 if (nr_ext_console_drivers) {
2419 ext_len = msg_print_ext_header(ext_text,
2421 msg, console_seq, console_prev);
2422 ext_len += msg_print_ext_body(ext_text + ext_len,
2423 sizeof(ext_text) - ext_len,
2424 log_dict(msg), msg->dict_len,
2425 log_text(msg), msg->text_len);
2427 console_idx = log_next(console_idx);
2429 console_prev = msg->flags;
2430 raw_spin_unlock(&logbuf_lock);
2432 stop_critical_timings(); /* don't trace print latency */
2433 call_console_drivers(level, ext_text, ext_len, text, len);
2434 start_critical_timings();
2435 local_irq_restore(flags);
2437 if (do_cond_resched)
2442 /* Release the exclusive_console once it is used */
2443 if (unlikely(exclusive_console))
2444 exclusive_console = NULL;
2446 raw_spin_unlock(&logbuf_lock);
2451 * Someone could have filled up the buffer again, so re-check if there's
2452 * something to flush. In case we cannot trylock the console_sem again,
2453 * there's a new owner and the console_unlock() from them will do the
2454 * flush, no worries.
2456 raw_spin_lock(&logbuf_lock);
2457 retry = console_seq != log_next_seq;
2458 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2460 if (retry && console_trylock())
2466 EXPORT_SYMBOL(console_unlock);
2469 * console_conditional_schedule - yield the CPU if required
2471 * If the console code is currently allowed to sleep, and
2472 * if this CPU should yield the CPU to another task, do
2475 * Must be called within console_lock();.
2477 void __sched console_conditional_schedule(void)
2479 if (console_may_schedule)
2482 EXPORT_SYMBOL(console_conditional_schedule);
2484 void console_unblank(void)
2489 * console_unblank can no longer be called in interrupt context unless
2490 * oops_in_progress is set to 1..
2492 if (oops_in_progress) {
2493 if (down_trylock_console_sem() != 0)
2499 console_may_schedule = 0;
2501 if ((c->flags & CON_ENABLED) && c->unblank)
2507 * console_flush_on_panic - flush console content on panic
2509 * Immediately output all pending messages no matter what.
2511 void console_flush_on_panic(void)
2514 * If someone else is holding the console lock, trylock will fail
2515 * and may_schedule may be set. Ignore and proceed to unlock so
2516 * that messages are flushed out. As this can be called from any
2517 * context and we don't want to get preempted while flushing,
2518 * ensure may_schedule is cleared.
2521 console_may_schedule = 0;
2526 * Return the console tty driver structure and its associated index
2528 struct tty_driver *console_device(int *index)
2531 struct tty_driver *driver = NULL;
2534 for_each_console(c) {
2537 driver = c->device(c, index);
2546 * Prevent further output on the passed console device so that (for example)
2547 * serial drivers can disable console output before suspending a port, and can
2548 * re-enable output afterwards.
2550 void console_stop(struct console *console)
2553 console->flags &= ~CON_ENABLED;
2556 EXPORT_SYMBOL(console_stop);
2558 void console_start(struct console *console)
2561 console->flags |= CON_ENABLED;
2564 EXPORT_SYMBOL(console_start);
2566 static int __read_mostly keep_bootcon;
2568 static int __init keep_bootcon_setup(char *str)
2571 pr_info("debug: skip boot console de-registration.\n");
2576 early_param("keep_bootcon", keep_bootcon_setup);
2579 * The console driver calls this routine during kernel initialization
2580 * to register the console printing procedure with printk() and to
2581 * print any messages that were printed by the kernel before the
2582 * console driver was initialized.
2584 * This can happen pretty early during the boot process (because of
2585 * early_printk) - sometimes before setup_arch() completes - be careful
2586 * of what kernel features are used - they may not be initialised yet.
2588 * There are two types of consoles - bootconsoles (early_printk) and
2589 * "real" consoles (everything which is not a bootconsole) which are
2590 * handled differently.
2591 * - Any number of bootconsoles can be registered at any time.
2592 * - As soon as a "real" console is registered, all bootconsoles
2593 * will be unregistered automatically.
2594 * - Once a "real" console is registered, any attempt to register a
2595 * bootconsoles will be rejected
2597 void register_console(struct console *newcon)
2600 unsigned long flags;
2601 struct console *bcon = NULL;
2602 struct console_cmdline *c;
2604 if (console_drivers)
2605 for_each_console(bcon)
2606 if (WARN(bcon == newcon,
2607 "console '%s%d' already registered\n",
2608 bcon->name, bcon->index))
2612 * before we register a new CON_BOOT console, make sure we don't
2613 * already have a valid console
2615 if (console_drivers && newcon->flags & CON_BOOT) {
2616 /* find the last or real console */
2617 for_each_console(bcon) {
2618 if (!(bcon->flags & CON_BOOT)) {
2619 pr_info("Too late to register bootconsole %s%d\n",
2620 newcon->name, newcon->index);
2626 if (console_drivers && console_drivers->flags & CON_BOOT)
2627 bcon = console_drivers;
2629 if (preferred_console < 0 || bcon || !console_drivers)
2630 preferred_console = selected_console;
2633 * See if we want to use this console driver. If we
2634 * didn't select a console we take the first one
2635 * that registers here.
2637 if (preferred_console < 0) {
2638 if (newcon->index < 0)
2640 if (newcon->setup == NULL ||
2641 newcon->setup(newcon, NULL) == 0) {
2642 newcon->flags |= CON_ENABLED;
2643 if (newcon->device) {
2644 newcon->flags |= CON_CONSDEV;
2645 preferred_console = 0;
2651 * See if this console matches one we selected on
2654 for (i = 0, c = console_cmdline;
2655 i < MAX_CMDLINECONSOLES && c->name[0];
2657 if (!newcon->match ||
2658 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2659 /* default matching */
2660 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2661 if (strcmp(c->name, newcon->name) != 0)
2663 if (newcon->index >= 0 &&
2664 newcon->index != c->index)
2666 if (newcon->index < 0)
2667 newcon->index = c->index;
2669 if (_braille_register_console(newcon, c))
2672 if (newcon->setup &&
2673 newcon->setup(newcon, c->options) != 0)
2677 newcon->flags |= CON_ENABLED;
2678 if (i == selected_console) {
2679 newcon->flags |= CON_CONSDEV;
2680 preferred_console = selected_console;
2685 if (!(newcon->flags & CON_ENABLED))
2689 * If we have a bootconsole, and are switching to a real console,
2690 * don't print everything out again, since when the boot console, and
2691 * the real console are the same physical device, it's annoying to
2692 * see the beginning boot messages twice
2694 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2695 newcon->flags &= ~CON_PRINTBUFFER;
2698 * Put this console in the list - keep the
2699 * preferred driver at the head of the list.
2702 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2703 newcon->next = console_drivers;
2704 console_drivers = newcon;
2706 newcon->next->flags &= ~CON_CONSDEV;
2708 newcon->next = console_drivers->next;
2709 console_drivers->next = newcon;
2712 if (newcon->flags & CON_EXTENDED)
2713 if (!nr_ext_console_drivers++)
2714 pr_info("printk: continuation disabled due to ext consoles, expect more fragments in /dev/kmsg\n");
2716 if (newcon->flags & CON_PRINTBUFFER) {
2718 * console_unlock(); will print out the buffered messages
2721 raw_spin_lock_irqsave(&logbuf_lock, flags);
2722 console_seq = syslog_seq;
2723 console_idx = syslog_idx;
2724 console_prev = syslog_prev;
2725 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2727 * We're about to replay the log buffer. Only do this to the
2728 * just-registered console to avoid excessive message spam to
2729 * the already-registered consoles.
2731 exclusive_console = newcon;
2734 console_sysfs_notify();
2737 * By unregistering the bootconsoles after we enable the real console
2738 * we get the "console xxx enabled" message on all the consoles -
2739 * boot consoles, real consoles, etc - this is to ensure that end
2740 * users know there might be something in the kernel's log buffer that
2741 * went to the bootconsole (that they do not see on the real console)
2743 pr_info("%sconsole [%s%d] enabled\n",
2744 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2745 newcon->name, newcon->index);
2747 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2749 /* We need to iterate through all boot consoles, to make
2750 * sure we print everything out, before we unregister them.
2752 for_each_console(bcon)
2753 if (bcon->flags & CON_BOOT)
2754 unregister_console(bcon);
2757 EXPORT_SYMBOL(register_console);
2759 int unregister_console(struct console *console)
2761 struct console *a, *b;
2764 pr_info("%sconsole [%s%d] disabled\n",
2765 (console->flags & CON_BOOT) ? "boot" : "" ,
2766 console->name, console->index);
2768 res = _braille_unregister_console(console);
2774 if (console_drivers == console) {
2775 console_drivers=console->next;
2777 } else if (console_drivers) {
2778 for (a=console_drivers->next, b=console_drivers ;
2779 a; b=a, a=b->next) {
2788 if (!res && (console->flags & CON_EXTENDED))
2789 nr_ext_console_drivers--;
2792 * If this isn't the last console and it has CON_CONSDEV set, we
2793 * need to set it on the next preferred console.
2795 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2796 console_drivers->flags |= CON_CONSDEV;
2798 console->flags &= ~CON_ENABLED;
2800 console_sysfs_notify();
2803 EXPORT_SYMBOL(unregister_console);
2806 * Some boot consoles access data that is in the init section and which will
2807 * be discarded after the initcalls have been run. To make sure that no code
2808 * will access this data, unregister the boot consoles in a late initcall.
2810 * If for some reason, such as deferred probe or the driver being a loadable
2811 * module, the real console hasn't registered yet at this point, there will
2812 * be a brief interval in which no messages are logged to the console, which
2813 * makes it difficult to diagnose problems that occur during this time.
2815 * To mitigate this problem somewhat, only unregister consoles whose memory
2816 * intersects with the init section. Note that code exists elsewhere to get
2817 * rid of the boot console as soon as the proper console shows up, so there
2818 * won't be side-effects from postponing the removal.
2820 static int __init printk_late_init(void)
2822 struct console *con;
2824 for_each_console(con) {
2825 if (!keep_bootcon && con->flags & CON_BOOT) {
2827 * Make sure to unregister boot consoles whose data
2828 * resides in the init section before the init section
2829 * is discarded. Boot consoles whose data will stick
2830 * around will automatically be unregistered when the
2831 * proper console replaces them.
2833 if (init_section_intersects(con, sizeof(*con)))
2834 unregister_console(con);
2837 hotcpu_notifier(console_cpu_notify, 0);
2840 late_initcall(printk_late_init);
2842 #if defined CONFIG_PRINTK
2844 * Delayed printk version, for scheduler-internal messages:
2846 #define PRINTK_PENDING_WAKEUP 0x01
2847 #define PRINTK_PENDING_OUTPUT 0x02
2849 static DEFINE_PER_CPU(int, printk_pending);
2851 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2853 int pending = __this_cpu_xchg(printk_pending, 0);
2855 if (pending & PRINTK_PENDING_OUTPUT) {
2856 /* If trylock fails, someone else is doing the printing */
2857 if (console_trylock())
2861 if (pending & PRINTK_PENDING_WAKEUP)
2862 wake_up_interruptible(&log_wait);
2865 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2866 .func = wake_up_klogd_work_func,
2867 .flags = IRQ_WORK_LAZY,
2870 void wake_up_klogd(void)
2873 if (waitqueue_active(&log_wait)) {
2874 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2875 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2880 int printk_deferred(const char *fmt, ...)
2886 va_start(args, fmt);
2887 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2890 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2891 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2898 * printk rate limiting, lifted from the networking subsystem.
2900 * This enforces a rate limit: not more than 10 kernel messages
2901 * every 5s to make a denial-of-service attack impossible.
2903 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2905 int __printk_ratelimit(const char *func)
2907 return ___ratelimit(&printk_ratelimit_state, func);
2909 EXPORT_SYMBOL(__printk_ratelimit);
2912 * printk_timed_ratelimit - caller-controlled printk ratelimiting
2913 * @caller_jiffies: pointer to caller's state
2914 * @interval_msecs: minimum interval between prints
2916 * printk_timed_ratelimit() returns true if more than @interval_msecs
2917 * milliseconds have elapsed since the last time printk_timed_ratelimit()
2920 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
2921 unsigned int interval_msecs)
2923 unsigned long elapsed = jiffies - *caller_jiffies;
2925 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
2928 *caller_jiffies = jiffies;
2931 EXPORT_SYMBOL(printk_timed_ratelimit);
2933 static DEFINE_SPINLOCK(dump_list_lock);
2934 static LIST_HEAD(dump_list);
2937 * kmsg_dump_register - register a kernel log dumper.
2938 * @dumper: pointer to the kmsg_dumper structure
2940 * Adds a kernel log dumper to the system. The dump callback in the
2941 * structure will be called when the kernel oopses or panics and must be
2942 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
2944 int kmsg_dump_register(struct kmsg_dumper *dumper)
2946 unsigned long flags;
2949 /* The dump callback needs to be set */
2953 spin_lock_irqsave(&dump_list_lock, flags);
2954 /* Don't allow registering multiple times */
2955 if (!dumper->registered) {
2956 dumper->registered = 1;
2957 list_add_tail_rcu(&dumper->list, &dump_list);
2960 spin_unlock_irqrestore(&dump_list_lock, flags);
2964 EXPORT_SYMBOL_GPL(kmsg_dump_register);
2967 * kmsg_dump_unregister - unregister a kmsg dumper.
2968 * @dumper: pointer to the kmsg_dumper structure
2970 * Removes a dump device from the system. Returns zero on success and
2971 * %-EINVAL otherwise.
2973 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
2975 unsigned long flags;
2978 spin_lock_irqsave(&dump_list_lock, flags);
2979 if (dumper->registered) {
2980 dumper->registered = 0;
2981 list_del_rcu(&dumper->list);
2984 spin_unlock_irqrestore(&dump_list_lock, flags);
2989 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
2991 static bool always_kmsg_dump;
2992 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
2995 * kmsg_dump - dump kernel log to kernel message dumpers.
2996 * @reason: the reason (oops, panic etc) for dumping
2998 * Call each of the registered dumper's dump() callback, which can
2999 * retrieve the kmsg records with kmsg_dump_get_line() or
3000 * kmsg_dump_get_buffer().
3002 void kmsg_dump(enum kmsg_dump_reason reason)
3004 struct kmsg_dumper *dumper;
3005 unsigned long flags;
3007 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3011 list_for_each_entry_rcu(dumper, &dump_list, list) {
3012 if (dumper->max_reason && reason > dumper->max_reason)
3015 /* initialize iterator with data about the stored records */
3016 dumper->active = true;
3018 raw_spin_lock_irqsave(&logbuf_lock, flags);
3019 dumper->cur_seq = clear_seq;
3020 dumper->cur_idx = clear_idx;
3021 dumper->next_seq = log_next_seq;
3022 dumper->next_idx = log_next_idx;
3023 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3025 /* invoke dumper which will iterate over records */
3026 dumper->dump(dumper, reason);
3028 /* reset iterator */
3029 dumper->active = false;
3035 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3036 * @dumper: registered kmsg dumper
3037 * @syslog: include the "<4>" prefixes
3038 * @line: buffer to copy the line to
3039 * @size: maximum size of the buffer
3040 * @len: length of line placed into buffer
3042 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3043 * record, and copy one record into the provided buffer.
3045 * Consecutive calls will return the next available record moving
3046 * towards the end of the buffer with the youngest messages.
3048 * A return value of FALSE indicates that there are no more records to
3051 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3053 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3054 char *line, size_t size, size_t *len)
3056 struct printk_log *msg;
3060 if (!dumper->active)
3063 if (dumper->cur_seq < log_first_seq) {
3064 /* messages are gone, move to first available one */
3065 dumper->cur_seq = log_first_seq;
3066 dumper->cur_idx = log_first_idx;
3070 if (dumper->cur_seq >= log_next_seq)
3073 msg = log_from_idx(dumper->cur_idx);
3074 l = msg_print_text(msg, 0, syslog, line, size);
3076 dumper->cur_idx = log_next(dumper->cur_idx);
3086 * kmsg_dump_get_line - retrieve one kmsg log line
3087 * @dumper: registered kmsg dumper
3088 * @syslog: include the "<4>" prefixes
3089 * @line: buffer to copy the line to
3090 * @size: maximum size of the buffer
3091 * @len: length of line placed into buffer
3093 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3094 * record, and copy one record into the provided buffer.
3096 * Consecutive calls will return the next available record moving
3097 * towards the end of the buffer with the youngest messages.
3099 * A return value of FALSE indicates that there are no more records to
3102 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3103 char *line, size_t size, size_t *len)
3105 unsigned long flags;
3108 raw_spin_lock_irqsave(&logbuf_lock, flags);
3109 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3110 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3114 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3117 * kmsg_dump_get_buffer - copy kmsg log lines
3118 * @dumper: registered kmsg dumper
3119 * @syslog: include the "<4>" prefixes
3120 * @buf: buffer to copy the line to
3121 * @size: maximum size of the buffer
3122 * @len: length of line placed into buffer
3124 * Start at the end of the kmsg buffer and fill the provided buffer
3125 * with as many of the the *youngest* kmsg records that fit into it.
3126 * If the buffer is large enough, all available kmsg records will be
3127 * copied with a single call.
3129 * Consecutive calls will fill the buffer with the next block of
3130 * available older records, not including the earlier retrieved ones.
3132 * A return value of FALSE indicates that there are no more records to
3135 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3136 char *buf, size_t size, size_t *len)
3138 unsigned long flags;
3143 enum log_flags prev;
3147 if (!dumper->active)
3150 raw_spin_lock_irqsave(&logbuf_lock, flags);
3151 if (dumper->cur_seq < log_first_seq) {
3152 /* messages are gone, move to first available one */
3153 dumper->cur_seq = log_first_seq;
3154 dumper->cur_idx = log_first_idx;
3158 if (dumper->cur_seq >= dumper->next_seq) {
3159 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3163 /* calculate length of entire buffer */
3164 seq = dumper->cur_seq;
3165 idx = dumper->cur_idx;
3167 while (seq < dumper->next_seq) {
3168 struct printk_log *msg = log_from_idx(idx);
3170 l += msg_print_text(msg, prev, true, NULL, 0);
3171 idx = log_next(idx);
3176 /* move first record forward until length fits into the buffer */
3177 seq = dumper->cur_seq;
3178 idx = dumper->cur_idx;
3180 while (l > size && seq < dumper->next_seq) {
3181 struct printk_log *msg = log_from_idx(idx);
3183 l -= msg_print_text(msg, prev, true, NULL, 0);
3184 idx = log_next(idx);
3189 /* last message in next interation */
3194 while (seq < dumper->next_seq) {
3195 struct printk_log *msg = log_from_idx(idx);
3197 l += msg_print_text(msg, prev, syslog, buf + l, size - l);
3198 idx = log_next(idx);
3203 dumper->next_seq = next_seq;
3204 dumper->next_idx = next_idx;
3206 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3212 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3215 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3216 * @dumper: registered kmsg dumper
3218 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3219 * kmsg_dump_get_buffer() can be called again and used multiple
3220 * times within the same dumper.dump() callback.
3222 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3224 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3226 dumper->cur_seq = clear_seq;
3227 dumper->cur_idx = clear_idx;
3228 dumper->next_seq = log_next_seq;
3229 dumper->next_idx = log_next_idx;
3233 * kmsg_dump_rewind - reset the interator
3234 * @dumper: registered kmsg dumper
3236 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3237 * kmsg_dump_get_buffer() can be called again and used multiple
3238 * times within the same dumper.dump() callback.
3240 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3242 unsigned long flags;
3244 raw_spin_lock_irqsave(&logbuf_lock, flags);
3245 kmsg_dump_rewind_nolock(dumper);
3246 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3248 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3250 static char dump_stack_arch_desc_str[128];
3253 * dump_stack_set_arch_desc - set arch-specific str to show with task dumps
3254 * @fmt: printf-style format string
3255 * @...: arguments for the format string
3257 * The configured string will be printed right after utsname during task
3258 * dumps. Usually used to add arch-specific system identifiers. If an
3259 * arch wants to make use of such an ID string, it should initialize this
3260 * as soon as possible during boot.
3262 void __init dump_stack_set_arch_desc(const char *fmt, ...)
3266 va_start(args, fmt);
3267 vsnprintf(dump_stack_arch_desc_str, sizeof(dump_stack_arch_desc_str),
3273 * dump_stack_print_info - print generic debug info for dump_stack()
3274 * @log_lvl: log level
3276 * Arch-specific dump_stack() implementations can use this function to
3277 * print out the same debug information as the generic dump_stack().
3279 void dump_stack_print_info(const char *log_lvl)
3281 printk("%sCPU: %d PID: %d Comm: %.20s %s %s %.*s\n",
3282 log_lvl, raw_smp_processor_id(), current->pid, current->comm,
3283 print_tainted(), init_utsname()->release,
3284 (int)strcspn(init_utsname()->version, " "),
3285 init_utsname()->version);
3287 if (dump_stack_arch_desc_str[0] != '\0')
3288 printk("%sHardware name: %s\n",
3289 log_lvl, dump_stack_arch_desc_str);
3291 print_worker_info(log_lvl, current);
3295 * show_regs_print_info - print generic debug info for show_regs()
3296 * @log_lvl: log level
3298 * show_regs() implementations can use this function to print out generic
3299 * debug information.
3301 void show_regs_print_info(const char *log_lvl)
3303 dump_stack_print_info(log_lvl);
3305 printk("%stask: %p task.stack: %p\n",
3306 log_lvl, current, task_stack_page(current));