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 void cont_flush(void);
788 static ssize_t devkmsg_read(struct file *file, char __user *buf,
789 size_t count, loff_t *ppos)
791 struct devkmsg_user *user = file->private_data;
792 struct printk_log *msg;
799 ret = mutex_lock_interruptible(&user->lock);
802 raw_spin_lock_irq(&logbuf_lock);
804 while (user->seq == log_next_seq) {
805 if (file->f_flags & O_NONBLOCK) {
807 raw_spin_unlock_irq(&logbuf_lock);
811 raw_spin_unlock_irq(&logbuf_lock);
812 ret = wait_event_interruptible(log_wait,
813 user->seq != log_next_seq);
816 raw_spin_lock_irq(&logbuf_lock);
819 if (user->seq < log_first_seq) {
820 /* our last seen message is gone, return error and reset */
821 user->idx = log_first_idx;
822 user->seq = log_first_seq;
824 raw_spin_unlock_irq(&logbuf_lock);
828 msg = log_from_idx(user->idx);
829 len = msg_print_ext_header(user->buf, sizeof(user->buf),
830 msg, user->seq, user->prev);
831 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
832 log_dict(msg), msg->dict_len,
833 log_text(msg), msg->text_len);
835 user->prev = msg->flags;
836 user->idx = log_next(user->idx);
838 raw_spin_unlock_irq(&logbuf_lock);
845 if (copy_to_user(buf, user->buf, len)) {
851 mutex_unlock(&user->lock);
855 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
857 struct devkmsg_user *user = file->private_data;
865 raw_spin_lock_irq(&logbuf_lock);
869 /* the first record */
870 user->idx = log_first_idx;
871 user->seq = log_first_seq;
875 * The first record after the last SYSLOG_ACTION_CLEAR,
876 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
877 * changes no global state, and does not clear anything.
879 user->idx = clear_idx;
880 user->seq = clear_seq;
883 /* after the last record */
884 user->idx = log_next_idx;
885 user->seq = log_next_seq;
890 raw_spin_unlock_irq(&logbuf_lock);
894 static unsigned int devkmsg_poll(struct file *file, poll_table *wait)
896 struct devkmsg_user *user = file->private_data;
900 return POLLERR|POLLNVAL;
902 poll_wait(file, &log_wait, wait);
904 raw_spin_lock_irq(&logbuf_lock);
906 if (user->seq < log_next_seq) {
907 /* return error when data has vanished underneath us */
908 if (user->seq < log_first_seq)
909 ret = POLLIN|POLLRDNORM|POLLERR|POLLPRI;
911 ret = POLLIN|POLLRDNORM;
913 raw_spin_unlock_irq(&logbuf_lock);
918 static int devkmsg_open(struct inode *inode, struct file *file)
920 struct devkmsg_user *user;
923 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
926 /* write-only does not need any file context */
927 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
928 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
934 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
938 ratelimit_default_init(&user->rs);
939 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
941 mutex_init(&user->lock);
943 raw_spin_lock_irq(&logbuf_lock);
944 user->idx = log_first_idx;
945 user->seq = log_first_seq;
946 raw_spin_unlock_irq(&logbuf_lock);
948 file->private_data = user;
952 static int devkmsg_release(struct inode *inode, struct file *file)
954 struct devkmsg_user *user = file->private_data;
959 ratelimit_state_exit(&user->rs);
961 mutex_destroy(&user->lock);
966 const struct file_operations kmsg_fops = {
967 .open = devkmsg_open,
968 .read = devkmsg_read,
969 .write_iter = devkmsg_write,
970 .llseek = devkmsg_llseek,
971 .poll = devkmsg_poll,
972 .release = devkmsg_release,
975 #ifdef CONFIG_KEXEC_CORE
977 * This appends the listed symbols to /proc/vmcore
979 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
980 * obtain access to symbols that are otherwise very difficult to locate. These
981 * symbols are specifically used so that utilities can access and extract the
982 * dmesg log from a vmcore file after a crash.
984 void log_buf_kexec_setup(void)
986 VMCOREINFO_SYMBOL(log_buf);
987 VMCOREINFO_SYMBOL(log_buf_len);
988 VMCOREINFO_SYMBOL(log_first_idx);
989 VMCOREINFO_SYMBOL(clear_idx);
990 VMCOREINFO_SYMBOL(log_next_idx);
992 * Export struct printk_log size and field offsets. User space tools can
993 * parse it and detect any changes to structure down the line.
995 VMCOREINFO_STRUCT_SIZE(printk_log);
996 VMCOREINFO_OFFSET(printk_log, ts_nsec);
997 VMCOREINFO_OFFSET(printk_log, len);
998 VMCOREINFO_OFFSET(printk_log, text_len);
999 VMCOREINFO_OFFSET(printk_log, dict_len);
1003 /* requested log_buf_len from kernel cmdline */
1004 static unsigned long __initdata new_log_buf_len;
1006 /* we practice scaling the ring buffer by powers of 2 */
1007 static void __init log_buf_len_update(unsigned size)
1010 size = roundup_pow_of_two(size);
1011 if (size > log_buf_len)
1012 new_log_buf_len = size;
1015 /* save requested log_buf_len since it's too early to process it */
1016 static int __init log_buf_len_setup(char *str)
1018 unsigned size = memparse(str, &str);
1020 log_buf_len_update(size);
1024 early_param("log_buf_len", log_buf_len_setup);
1027 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1029 static void __init log_buf_add_cpu(void)
1031 unsigned int cpu_extra;
1034 * archs should set up cpu_possible_bits properly with
1035 * set_cpu_possible() after setup_arch() but just in
1036 * case lets ensure this is valid.
1038 if (num_possible_cpus() == 1)
1041 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1043 /* by default this will only continue through for large > 64 CPUs */
1044 if (cpu_extra <= __LOG_BUF_LEN / 2)
1047 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1048 __LOG_CPU_MAX_BUF_LEN);
1049 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1051 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1053 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1055 #else /* !CONFIG_SMP */
1056 static inline void log_buf_add_cpu(void) {}
1057 #endif /* CONFIG_SMP */
1059 void __init setup_log_buf(int early)
1061 unsigned long flags;
1065 if (log_buf != __log_buf)
1068 if (!early && !new_log_buf_len)
1071 if (!new_log_buf_len)
1076 memblock_virt_alloc(new_log_buf_len, LOG_ALIGN);
1078 new_log_buf = memblock_virt_alloc_nopanic(new_log_buf_len,
1082 if (unlikely(!new_log_buf)) {
1083 pr_err("log_buf_len: %ld bytes not available\n",
1088 raw_spin_lock_irqsave(&logbuf_lock, flags);
1089 log_buf_len = new_log_buf_len;
1090 log_buf = new_log_buf;
1091 new_log_buf_len = 0;
1092 free = __LOG_BUF_LEN - log_next_idx;
1093 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1094 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
1096 pr_info("log_buf_len: %d bytes\n", log_buf_len);
1097 pr_info("early log buf free: %d(%d%%)\n",
1098 free, (free * 100) / __LOG_BUF_LEN);
1101 static bool __read_mostly ignore_loglevel;
1103 static int __init ignore_loglevel_setup(char *str)
1105 ignore_loglevel = true;
1106 pr_info("debug: ignoring loglevel setting.\n");
1111 early_param("ignore_loglevel", ignore_loglevel_setup);
1112 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1113 MODULE_PARM_DESC(ignore_loglevel,
1114 "ignore loglevel setting (prints all kernel messages to the console)");
1116 static bool suppress_message_printing(int level)
1118 return (level >= console_loglevel && !ignore_loglevel);
1121 #ifdef CONFIG_BOOT_PRINTK_DELAY
1123 static int boot_delay; /* msecs delay after each printk during bootup */
1124 static unsigned long long loops_per_msec; /* based on boot_delay */
1126 static int __init boot_delay_setup(char *str)
1130 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1131 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1133 get_option(&str, &boot_delay);
1134 if (boot_delay > 10 * 1000)
1137 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1138 "HZ: %d, loops_per_msec: %llu\n",
1139 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1142 early_param("boot_delay", boot_delay_setup);
1144 static void boot_delay_msec(int level)
1146 unsigned long long k;
1147 unsigned long timeout;
1149 if ((boot_delay == 0 || system_state != SYSTEM_BOOTING)
1150 || suppress_message_printing(level)) {
1154 k = (unsigned long long)loops_per_msec * boot_delay;
1156 timeout = jiffies + msecs_to_jiffies(boot_delay);
1161 * use (volatile) jiffies to prevent
1162 * compiler reduction; loop termination via jiffies
1163 * is secondary and may or may not happen.
1165 if (time_after(jiffies, timeout))
1167 touch_nmi_watchdog();
1171 static inline void boot_delay_msec(int level)
1176 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1177 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1179 static size_t print_time(u64 ts, char *buf)
1181 unsigned long rem_nsec;
1186 rem_nsec = do_div(ts, 1000000000);
1189 return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
1191 return sprintf(buf, "[%5lu.%06lu] ",
1192 (unsigned long)ts, rem_nsec / 1000);
1195 static size_t print_prefix(const struct printk_log *msg, bool syslog, char *buf)
1198 unsigned int prefix = (msg->facility << 3) | msg->level;
1202 len += sprintf(buf, "<%u>", prefix);
1207 else if (prefix > 99)
1209 else if (prefix > 9)
1214 len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
1218 static size_t msg_print_text(const struct printk_log *msg, enum log_flags prev,
1219 bool syslog, char *buf, size_t size)
1221 const char *text = log_text(msg);
1222 size_t text_size = msg->text_len;
1224 bool newline = true;
1227 if ((prev & LOG_CONT) && !(msg->flags & LOG_PREFIX))
1230 if (msg->flags & LOG_CONT) {
1231 if ((prev & LOG_CONT) && !(prev & LOG_NEWLINE))
1234 if (!(msg->flags & LOG_NEWLINE))
1239 const char *next = memchr(text, '\n', text_size);
1243 text_len = next - text;
1245 text_size -= next - text;
1247 text_len = text_size;
1251 if (print_prefix(msg, syslog, NULL) +
1252 text_len + 1 >= size - len)
1256 len += print_prefix(msg, syslog, buf + len);
1257 memcpy(buf + len, text, text_len);
1259 if (next || newline)
1262 /* SYSLOG_ACTION_* buffer size only calculation */
1264 len += print_prefix(msg, syslog, NULL);
1266 if (next || newline)
1277 static int syslog_print(char __user *buf, int size)
1280 struct printk_log *msg;
1283 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1291 raw_spin_lock_irq(&logbuf_lock);
1293 if (syslog_seq < log_first_seq) {
1294 /* messages are gone, move to first one */
1295 syslog_seq = log_first_seq;
1296 syslog_idx = log_first_idx;
1300 if (syslog_seq == log_next_seq) {
1301 raw_spin_unlock_irq(&logbuf_lock);
1305 skip = syslog_partial;
1306 msg = log_from_idx(syslog_idx);
1307 n = msg_print_text(msg, syslog_prev, true, text,
1308 LOG_LINE_MAX + PREFIX_MAX);
1309 if (n - syslog_partial <= size) {
1310 /* message fits into buffer, move forward */
1311 syslog_idx = log_next(syslog_idx);
1313 syslog_prev = msg->flags;
1314 n -= syslog_partial;
1317 /* partial read(), remember position */
1319 syslog_partial += n;
1322 raw_spin_unlock_irq(&logbuf_lock);
1327 if (copy_to_user(buf, text + skip, n)) {
1342 static int syslog_print_all(char __user *buf, int size, bool clear)
1347 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1351 raw_spin_lock_irq(&logbuf_lock);
1357 enum log_flags prev;
1360 * Find first record that fits, including all following records,
1361 * into the user-provided buffer for this dump.
1366 while (seq < log_next_seq) {
1367 struct printk_log *msg = log_from_idx(idx);
1369 len += msg_print_text(msg, prev, true, NULL, 0);
1371 idx = log_next(idx);
1375 /* move first record forward until length fits into the buffer */
1379 while (len > size && seq < log_next_seq) {
1380 struct printk_log *msg = log_from_idx(idx);
1382 len -= msg_print_text(msg, prev, true, NULL, 0);
1384 idx = log_next(idx);
1388 /* last message fitting into this dump */
1389 next_seq = log_next_seq;
1392 while (len >= 0 && seq < next_seq) {
1393 struct printk_log *msg = log_from_idx(idx);
1396 textlen = msg_print_text(msg, prev, true, text,
1397 LOG_LINE_MAX + PREFIX_MAX);
1402 idx = log_next(idx);
1406 raw_spin_unlock_irq(&logbuf_lock);
1407 if (copy_to_user(buf + len, text, textlen))
1411 raw_spin_lock_irq(&logbuf_lock);
1413 if (seq < log_first_seq) {
1414 /* messages are gone, move to next one */
1415 seq = log_first_seq;
1416 idx = log_first_idx;
1423 clear_seq = log_next_seq;
1424 clear_idx = log_next_idx;
1426 raw_spin_unlock_irq(&logbuf_lock);
1432 int do_syslog(int type, char __user *buf, int len, int source)
1435 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1438 error = check_syslog_permissions(type, source);
1443 case SYSLOG_ACTION_CLOSE: /* Close log */
1445 case SYSLOG_ACTION_OPEN: /* Open log */
1447 case SYSLOG_ACTION_READ: /* Read from log */
1449 if (!buf || len < 0)
1454 if (!access_ok(VERIFY_WRITE, buf, len)) {
1458 error = wait_event_interruptible(log_wait,
1459 syslog_seq != log_next_seq);
1462 error = syslog_print(buf, len);
1464 /* Read/clear last kernel messages */
1465 case SYSLOG_ACTION_READ_CLEAR:
1468 /* Read last kernel messages */
1469 case SYSLOG_ACTION_READ_ALL:
1471 if (!buf || len < 0)
1476 if (!access_ok(VERIFY_WRITE, buf, len)) {
1480 error = syslog_print_all(buf, len, clear);
1482 /* Clear ring buffer */
1483 case SYSLOG_ACTION_CLEAR:
1484 syslog_print_all(NULL, 0, true);
1486 /* Disable logging to console */
1487 case SYSLOG_ACTION_CONSOLE_OFF:
1488 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1489 saved_console_loglevel = console_loglevel;
1490 console_loglevel = minimum_console_loglevel;
1492 /* Enable logging to console */
1493 case SYSLOG_ACTION_CONSOLE_ON:
1494 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1495 console_loglevel = saved_console_loglevel;
1496 saved_console_loglevel = LOGLEVEL_DEFAULT;
1499 /* Set level of messages printed to console */
1500 case SYSLOG_ACTION_CONSOLE_LEVEL:
1502 if (len < 1 || len > 8)
1504 if (len < minimum_console_loglevel)
1505 len = minimum_console_loglevel;
1506 console_loglevel = len;
1507 /* Implicitly re-enable logging to console */
1508 saved_console_loglevel = LOGLEVEL_DEFAULT;
1511 /* Number of chars in the log buffer */
1512 case SYSLOG_ACTION_SIZE_UNREAD:
1513 raw_spin_lock_irq(&logbuf_lock);
1515 if (syslog_seq < log_first_seq) {
1516 /* messages are gone, move to first one */
1517 syslog_seq = log_first_seq;
1518 syslog_idx = log_first_idx;
1522 if (source == SYSLOG_FROM_PROC) {
1524 * Short-cut for poll(/"proc/kmsg") which simply checks
1525 * for pending data, not the size; return the count of
1526 * records, not the length.
1528 error = log_next_seq - syslog_seq;
1530 u64 seq = syslog_seq;
1531 u32 idx = syslog_idx;
1532 enum log_flags prev = syslog_prev;
1535 while (seq < log_next_seq) {
1536 struct printk_log *msg = log_from_idx(idx);
1538 error += msg_print_text(msg, prev, true, NULL, 0);
1539 idx = log_next(idx);
1543 error -= syslog_partial;
1545 raw_spin_unlock_irq(&logbuf_lock);
1547 /* Size of the log buffer */
1548 case SYSLOG_ACTION_SIZE_BUFFER:
1549 error = log_buf_len;
1559 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1561 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1565 * Call the console drivers, asking them to write out
1566 * log_buf[start] to log_buf[end - 1].
1567 * The console_lock must be held.
1569 static void call_console_drivers(int level,
1570 const char *ext_text, size_t ext_len,
1571 const char *text, size_t len)
1573 struct console *con;
1575 trace_console(text, len);
1577 if (!console_drivers)
1580 for_each_console(con) {
1581 if (exclusive_console && con != exclusive_console)
1583 if (!(con->flags & CON_ENABLED))
1587 if (!cpu_online(smp_processor_id()) &&
1588 !(con->flags & CON_ANYTIME))
1590 if (con->flags & CON_EXTENDED)
1591 con->write(con, ext_text, ext_len);
1593 con->write(con, text, len);
1598 * Zap console related locks when oopsing.
1599 * To leave time for slow consoles to print a full oops,
1600 * only zap at most once every 30 seconds.
1602 static void zap_locks(void)
1604 static unsigned long oops_timestamp;
1606 if (time_after_eq(jiffies, oops_timestamp) &&
1607 !time_after(jiffies, oops_timestamp + 30 * HZ))
1610 oops_timestamp = jiffies;
1613 /* If a crash is occurring, make sure we can't deadlock */
1614 raw_spin_lock_init(&logbuf_lock);
1615 /* And make sure that we print immediately */
1616 sema_init(&console_sem, 1);
1619 int printk_delay_msec __read_mostly;
1621 static inline void printk_delay(void)
1623 if (unlikely(printk_delay_msec)) {
1624 int m = printk_delay_msec;
1628 touch_nmi_watchdog();
1634 * Continuation lines are buffered, and not committed to the record buffer
1635 * until the line is complete, or a race forces it. The line fragments
1636 * though, are printed immediately to the consoles to ensure everything has
1637 * reached the console in case of a kernel crash.
1639 static struct cont {
1640 char buf[LOG_LINE_MAX];
1641 size_t len; /* length == 0 means unused buffer */
1642 size_t cons; /* bytes written to console */
1643 struct task_struct *owner; /* task of first print*/
1644 u64 ts_nsec; /* time of first print */
1645 u8 level; /* log level of first message */
1646 u8 facility; /* log facility of first message */
1647 enum log_flags flags; /* prefix, newline flags */
1648 bool flushed:1; /* buffer sealed and committed */
1651 static void cont_flush(void)
1659 * If a fragment of this line was directly flushed to the
1660 * console; wait for the console to pick up the rest of the
1661 * line. LOG_NOCONS suppresses a duplicated output.
1663 log_store(cont.facility, cont.level, cont.flags | LOG_NOCONS,
1664 cont.ts_nsec, NULL, 0, cont.buf, cont.len);
1665 cont.flushed = true;
1668 * If no fragment of this line ever reached the console,
1669 * just submit it to the store and free the buffer.
1671 log_store(cont.facility, cont.level, cont.flags, 0,
1672 NULL, 0, cont.buf, cont.len);
1677 static bool cont_add(int facility, int level, enum log_flags flags, const char *text, size_t len)
1679 if (cont.len && cont.flushed)
1683 * If ext consoles are present, flush and skip in-kernel
1684 * continuation. See nr_ext_console_drivers definition. Also, if
1685 * the line gets too long, split it up in separate records.
1687 if (nr_ext_console_drivers || cont.len + len > sizeof(cont.buf)) {
1693 cont.facility = facility;
1695 cont.owner = current;
1696 cont.ts_nsec = local_clock();
1699 cont.flushed = false;
1702 memcpy(cont.buf + cont.len, text, len);
1705 // The original flags come from the first line,
1706 // but later continuations can add a newline.
1707 if (flags & LOG_NEWLINE) {
1708 cont.flags |= LOG_NEWLINE;
1712 if (cont.len > (sizeof(cont.buf) * 80) / 100)
1718 static size_t cont_print_text(char *text, size_t size)
1723 if (cont.cons == 0 && (console_prev & LOG_NEWLINE)) {
1724 textlen += print_time(cont.ts_nsec, text);
1728 len = cont.len - cont.cons;
1732 memcpy(text + textlen, cont.buf + cont.cons, len);
1734 cont.cons = cont.len;
1738 if (cont.flags & LOG_NEWLINE)
1739 text[textlen++] = '\n';
1740 /* got everything, release buffer */
1746 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)
1749 * If an earlier line was buffered, and we're a continuation
1750 * write from the same process, try to add it to the buffer.
1753 if (cont.owner == current && (lflags & LOG_CONT)) {
1754 if (cont_add(facility, level, lflags, text, text_len))
1757 /* Otherwise, make sure it's flushed */
1761 /* Skip empty continuation lines that couldn't be added - they just flush */
1762 if (!text_len && (lflags & LOG_CONT))
1765 /* If it doesn't end in a newline, try to buffer the current line */
1766 if (!(lflags & LOG_NEWLINE)) {
1767 if (cont_add(facility, level, lflags, text, text_len))
1771 /* Store it in the record log */
1772 return log_store(facility, level, lflags, 0, dict, dictlen, text, text_len);
1775 asmlinkage int vprintk_emit(int facility, int level,
1776 const char *dict, size_t dictlen,
1777 const char *fmt, va_list args)
1779 static bool recursion_bug;
1780 static char textbuf[LOG_LINE_MAX];
1781 char *text = textbuf;
1782 size_t text_len = 0;
1783 enum log_flags lflags = 0;
1784 unsigned long flags;
1786 int printed_len = 0;
1787 int nmi_message_lost;
1788 bool in_sched = false;
1789 /* cpu currently holding logbuf_lock in this function */
1790 static unsigned int logbuf_cpu = UINT_MAX;
1792 if (level == LOGLEVEL_SCHED) {
1793 level = LOGLEVEL_DEFAULT;
1797 boot_delay_msec(level);
1800 local_irq_save(flags);
1801 this_cpu = smp_processor_id();
1804 * Ouch, printk recursed into itself!
1806 if (unlikely(logbuf_cpu == this_cpu)) {
1808 * If a crash is occurring during printk() on this CPU,
1809 * then try to get the crash message out but make sure
1810 * we can't deadlock. Otherwise just return to avoid the
1811 * recursion and return - but flag the recursion so that
1812 * it can be printed at the next appropriate moment:
1814 if (!oops_in_progress && !lockdep_recursing(current)) {
1815 recursion_bug = true;
1816 local_irq_restore(flags);
1823 /* This stops the holder of console_sem just where we want him */
1824 raw_spin_lock(&logbuf_lock);
1825 logbuf_cpu = this_cpu;
1827 if (unlikely(recursion_bug)) {
1828 static const char recursion_msg[] =
1829 "BUG: recent printk recursion!";
1831 recursion_bug = false;
1832 /* emit KERN_CRIT message */
1833 printed_len += log_store(0, 2, LOG_PREFIX|LOG_NEWLINE, 0,
1834 NULL, 0, recursion_msg,
1835 strlen(recursion_msg));
1838 nmi_message_lost = get_nmi_message_lost();
1839 if (unlikely(nmi_message_lost)) {
1840 text_len = scnprintf(textbuf, sizeof(textbuf),
1841 "BAD LUCK: lost %d message(s) from NMI context!",
1843 printed_len += log_store(0, 2, LOG_PREFIX|LOG_NEWLINE, 0,
1844 NULL, 0, textbuf, text_len);
1848 * The printf needs to come first; we need the syslog
1849 * prefix which might be passed-in as a parameter.
1851 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1853 /* mark and strip a trailing newline */
1854 if (text_len && text[text_len-1] == '\n') {
1856 lflags |= LOG_NEWLINE;
1859 /* strip kernel syslog prefix and extract log level or control flags */
1860 if (facility == 0) {
1863 while ((kern_level = printk_get_level(text)) != 0) {
1864 switch (kern_level) {
1866 if (level == LOGLEVEL_DEFAULT)
1867 level = kern_level - '0';
1869 case 'd': /* KERN_DEFAULT */
1870 lflags |= LOG_PREFIX;
1872 case 'c': /* KERN_CONT */
1881 if (level == LOGLEVEL_DEFAULT)
1882 level = default_message_loglevel;
1885 lflags |= LOG_PREFIX|LOG_NEWLINE;
1887 printed_len += log_output(facility, level, lflags, dict, dictlen, text, text_len);
1889 logbuf_cpu = UINT_MAX;
1890 raw_spin_unlock(&logbuf_lock);
1892 local_irq_restore(flags);
1894 /* If called from the scheduler, we can not call up(). */
1898 * Try to acquire and then immediately release the console
1899 * semaphore. The release will print out buffers and wake up
1900 * /dev/kmsg and syslog() users.
1902 if (console_trylock())
1909 EXPORT_SYMBOL(vprintk_emit);
1911 asmlinkage int vprintk(const char *fmt, va_list args)
1913 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1915 EXPORT_SYMBOL(vprintk);
1917 asmlinkage int printk_emit(int facility, int level,
1918 const char *dict, size_t dictlen,
1919 const char *fmt, ...)
1924 va_start(args, fmt);
1925 r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1930 EXPORT_SYMBOL(printk_emit);
1932 int vprintk_default(const char *fmt, va_list args)
1936 #ifdef CONFIG_KGDB_KDB
1937 if (unlikely(kdb_trap_printk)) {
1938 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
1942 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1946 EXPORT_SYMBOL_GPL(vprintk_default);
1949 * printk - print a kernel message
1950 * @fmt: format string
1952 * This is printk(). It can be called from any context. We want it to work.
1954 * We try to grab the console_lock. If we succeed, it's easy - we log the
1955 * output and call the console drivers. If we fail to get the semaphore, we
1956 * place the output into the log buffer and return. The current holder of
1957 * the console_sem will notice the new output in console_unlock(); and will
1958 * send it to the consoles before releasing the lock.
1960 * One effect of this deferred printing is that code which calls printk() and
1961 * then changes console_loglevel may break. This is because console_loglevel
1962 * is inspected when the actual printing occurs.
1967 * See the vsnprintf() documentation for format string extensions over C99.
1969 asmlinkage __visible int printk(const char *fmt, ...)
1974 va_start(args, fmt);
1975 r = vprintk_func(fmt, args);
1980 EXPORT_SYMBOL(printk);
1982 #else /* CONFIG_PRINTK */
1984 #define LOG_LINE_MAX 0
1985 #define PREFIX_MAX 0
1987 static u64 syslog_seq;
1988 static u32 syslog_idx;
1989 static u64 console_seq;
1990 static u32 console_idx;
1991 static enum log_flags syslog_prev;
1992 static u64 log_first_seq;
1993 static u32 log_first_idx;
1994 static u64 log_next_seq;
1995 static enum log_flags console_prev;
1996 static struct cont {
2002 static char *log_text(const struct printk_log *msg) { return NULL; }
2003 static char *log_dict(const struct printk_log *msg) { return NULL; }
2004 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2005 static u32 log_next(u32 idx) { return 0; }
2006 static ssize_t msg_print_ext_header(char *buf, size_t size,
2007 struct printk_log *msg, u64 seq,
2008 enum log_flags prev_flags) { return 0; }
2009 static ssize_t msg_print_ext_body(char *buf, size_t size,
2010 char *dict, size_t dict_len,
2011 char *text, size_t text_len) { return 0; }
2012 static void call_console_drivers(int level,
2013 const char *ext_text, size_t ext_len,
2014 const char *text, size_t len) {}
2015 static size_t msg_print_text(const struct printk_log *msg, enum log_flags prev,
2016 bool syslog, char *buf, size_t size) { return 0; }
2017 static size_t cont_print_text(char *text, size_t size) { return 0; }
2018 static bool suppress_message_printing(int level) { return false; }
2020 /* Still needs to be defined for users */
2021 DEFINE_PER_CPU(printk_func_t, printk_func);
2023 #endif /* CONFIG_PRINTK */
2025 #ifdef CONFIG_EARLY_PRINTK
2026 struct console *early_console;
2028 asmlinkage __visible void early_printk(const char *fmt, ...)
2038 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2041 early_console->write(early_console, buf, n);
2045 static int __add_preferred_console(char *name, int idx, char *options,
2048 struct console_cmdline *c;
2052 * See if this tty is not yet registered, and
2053 * if we have a slot free.
2055 for (i = 0, c = console_cmdline;
2056 i < MAX_CMDLINECONSOLES && c->name[0];
2058 if (strcmp(c->name, name) == 0 && c->index == idx) {
2060 selected_console = i;
2064 if (i == MAX_CMDLINECONSOLES)
2067 selected_console = i;
2068 strlcpy(c->name, name, sizeof(c->name));
2069 c->options = options;
2070 braille_set_options(c, brl_options);
2076 * Set up a console. Called via do_early_param() in init/main.c
2077 * for each "console=" parameter in the boot command line.
2079 static int __init console_setup(char *str)
2081 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2082 char *s, *options, *brl_options = NULL;
2085 if (_braille_console_setup(&str, &brl_options))
2089 * Decode str into name, index, options.
2091 if (str[0] >= '0' && str[0] <= '9') {
2092 strcpy(buf, "ttyS");
2093 strncpy(buf + 4, str, sizeof(buf) - 5);
2095 strncpy(buf, str, sizeof(buf) - 1);
2097 buf[sizeof(buf) - 1] = 0;
2098 options = strchr(str, ',');
2102 if (!strcmp(str, "ttya"))
2103 strcpy(buf, "ttyS0");
2104 if (!strcmp(str, "ttyb"))
2105 strcpy(buf, "ttyS1");
2107 for (s = buf; *s; s++)
2108 if (isdigit(*s) || *s == ',')
2110 idx = simple_strtoul(s, NULL, 10);
2113 __add_preferred_console(buf, idx, options, brl_options);
2114 console_set_on_cmdline = 1;
2117 __setup("console=", console_setup);
2120 * add_preferred_console - add a device to the list of preferred consoles.
2121 * @name: device name
2122 * @idx: device index
2123 * @options: options for this console
2125 * The last preferred console added will be used for kernel messages
2126 * and stdin/out/err for init. Normally this is used by console_setup
2127 * above to handle user-supplied console arguments; however it can also
2128 * be used by arch-specific code either to override the user or more
2129 * commonly to provide a default console (ie from PROM variables) when
2130 * the user has not supplied one.
2132 int add_preferred_console(char *name, int idx, char *options)
2134 return __add_preferred_console(name, idx, options, NULL);
2137 bool console_suspend_enabled = true;
2138 EXPORT_SYMBOL(console_suspend_enabled);
2140 static int __init console_suspend_disable(char *str)
2142 console_suspend_enabled = false;
2145 __setup("no_console_suspend", console_suspend_disable);
2146 module_param_named(console_suspend, console_suspend_enabled,
2147 bool, S_IRUGO | S_IWUSR);
2148 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2149 " and hibernate operations");
2152 * suspend_console - suspend the console subsystem
2154 * This disables printk() while we go into suspend states
2156 void suspend_console(void)
2158 if (!console_suspend_enabled)
2160 printk("Suspending console(s) (use no_console_suspend to debug)\n");
2162 console_suspended = 1;
2166 void resume_console(void)
2168 if (!console_suspend_enabled)
2171 console_suspended = 0;
2176 * console_cpu_notify - print deferred console messages after CPU hotplug
2177 * @self: notifier struct
2178 * @action: CPU hotplug event
2181 * If printk() is called from a CPU that is not online yet, the messages
2182 * will be spooled but will not show up on the console. This function is
2183 * called when a new CPU comes online (or fails to come up), and ensures
2184 * that any such output gets printed.
2186 static int console_cpu_notify(struct notifier_block *self,
2187 unsigned long action, void *hcpu)
2192 case CPU_DOWN_FAILED:
2193 case CPU_UP_CANCELED:
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) {
2238 * When PREEMPT_COUNT disabled we can't reliably detect if it's
2239 * safe to schedule (e.g. calling printk while holding a spin_lock),
2240 * because preempt_disable()/preempt_enable() are just barriers there
2241 * and preempt_count() is always 0.
2243 * RCU read sections have a separate preemption counter when
2244 * PREEMPT_RCU enabled thus we must take extra care and check
2245 * rcu_preempt_depth(), otherwise RCU read sections modify
2248 console_may_schedule = !oops_in_progress &&
2250 !rcu_preempt_depth();
2253 EXPORT_SYMBOL(console_trylock);
2255 int is_console_locked(void)
2257 return console_locked;
2261 * Check if we have any console that is capable of printing while cpu is
2262 * booting or shutting down. Requires console_sem.
2264 static int have_callable_console(void)
2266 struct console *con;
2268 for_each_console(con)
2269 if ((con->flags & CON_ENABLED) &&
2270 (con->flags & CON_ANYTIME))
2277 * Can we actually use the console at this time on this cpu?
2279 * Console drivers may assume that per-cpu resources have been allocated. So
2280 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2281 * call them until this CPU is officially up.
2283 static inline int can_use_console(void)
2285 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2288 static void console_cont_flush(char *text, size_t size)
2290 unsigned long flags;
2293 raw_spin_lock_irqsave(&logbuf_lock, flags);
2298 if (suppress_message_printing(cont.level)) {
2299 cont.cons = cont.len;
2306 * We still queue earlier records, likely because the console was
2307 * busy. The earlier ones need to be printed before this one, we
2308 * did not flush any fragment so far, so just let it queue up.
2310 if (console_seq < log_next_seq && !cont.cons)
2313 len = cont_print_text(text, size);
2314 raw_spin_unlock(&logbuf_lock);
2315 stop_critical_timings();
2316 call_console_drivers(cont.level, NULL, 0, text, len);
2317 start_critical_timings();
2318 local_irq_restore(flags);
2321 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2325 * console_unlock - unlock the console system
2327 * Releases the console_lock which the caller holds on the console system
2328 * and the console driver list.
2330 * While the console_lock was held, console output may have been buffered
2331 * by printk(). If this is the case, console_unlock(); emits
2332 * the output prior to releasing the lock.
2334 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2336 * console_unlock(); may be called from any context.
2338 void console_unlock(void)
2340 static char ext_text[CONSOLE_EXT_LOG_MAX];
2341 static char text[LOG_LINE_MAX + PREFIX_MAX];
2342 static u64 seen_seq;
2343 unsigned long flags;
2344 bool wake_klogd = false;
2345 bool do_cond_resched, retry;
2347 if (console_suspended) {
2353 * Console drivers are called under logbuf_lock, so
2354 * @console_may_schedule should be cleared before; however, we may
2355 * end up dumping a lot of lines, for example, if called from
2356 * console registration path, and should invoke cond_resched()
2357 * between lines if allowable. Not doing so can cause a very long
2358 * scheduling stall on a slow console leading to RCU stall and
2359 * softlockup warnings which exacerbate the issue with more
2360 * messages practically incapacitating the system.
2362 do_cond_resched = console_may_schedule;
2363 console_may_schedule = 0;
2367 * We released the console_sem lock, so we need to recheck if
2368 * cpu is online and (if not) is there at least one CON_ANYTIME
2371 if (!can_use_console()) {
2377 /* flush buffered message fragment immediately to console */
2378 console_cont_flush(text, sizeof(text));
2381 struct printk_log *msg;
2386 raw_spin_lock_irqsave(&logbuf_lock, flags);
2387 if (seen_seq != log_next_seq) {
2389 seen_seq = log_next_seq;
2392 if (console_seq < log_first_seq) {
2393 len = sprintf(text, "** %u printk messages dropped ** ",
2394 (unsigned)(log_first_seq - console_seq));
2396 /* messages are gone, move to first one */
2397 console_seq = log_first_seq;
2398 console_idx = log_first_idx;
2404 if (console_seq == log_next_seq)
2407 msg = log_from_idx(console_idx);
2409 if ((msg->flags & LOG_NOCONS) ||
2410 suppress_message_printing(level)) {
2412 * Skip record we have buffered and already printed
2413 * directly to the console when we received it, and
2414 * record that has level above the console loglevel.
2416 console_idx = log_next(console_idx);
2419 * We will get here again when we register a new
2420 * CON_PRINTBUFFER console. Clear the flag so we
2421 * will properly dump everything later.
2423 msg->flags &= ~LOG_NOCONS;
2424 console_prev = msg->flags;
2428 len += msg_print_text(msg, console_prev, false,
2429 text + len, sizeof(text) - len);
2430 if (nr_ext_console_drivers) {
2431 ext_len = msg_print_ext_header(ext_text,
2433 msg, console_seq, console_prev);
2434 ext_len += msg_print_ext_body(ext_text + ext_len,
2435 sizeof(ext_text) - ext_len,
2436 log_dict(msg), msg->dict_len,
2437 log_text(msg), msg->text_len);
2439 console_idx = log_next(console_idx);
2441 console_prev = msg->flags;
2442 raw_spin_unlock(&logbuf_lock);
2444 stop_critical_timings(); /* don't trace print latency */
2445 call_console_drivers(level, ext_text, ext_len, text, len);
2446 start_critical_timings();
2447 local_irq_restore(flags);
2449 if (do_cond_resched)
2454 /* Release the exclusive_console once it is used */
2455 if (unlikely(exclusive_console))
2456 exclusive_console = NULL;
2458 raw_spin_unlock(&logbuf_lock);
2463 * Someone could have filled up the buffer again, so re-check if there's
2464 * something to flush. In case we cannot trylock the console_sem again,
2465 * there's a new owner and the console_unlock() from them will do the
2466 * flush, no worries.
2468 raw_spin_lock(&logbuf_lock);
2469 retry = console_seq != log_next_seq;
2470 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2472 if (retry && console_trylock())
2478 EXPORT_SYMBOL(console_unlock);
2481 * console_conditional_schedule - yield the CPU if required
2483 * If the console code is currently allowed to sleep, and
2484 * if this CPU should yield the CPU to another task, do
2487 * Must be called within console_lock();.
2489 void __sched console_conditional_schedule(void)
2491 if (console_may_schedule)
2494 EXPORT_SYMBOL(console_conditional_schedule);
2496 void console_unblank(void)
2501 * console_unblank can no longer be called in interrupt context unless
2502 * oops_in_progress is set to 1..
2504 if (oops_in_progress) {
2505 if (down_trylock_console_sem() != 0)
2511 console_may_schedule = 0;
2513 if ((c->flags & CON_ENABLED) && c->unblank)
2519 * console_flush_on_panic - flush console content on panic
2521 * Immediately output all pending messages no matter what.
2523 void console_flush_on_panic(void)
2526 * If someone else is holding the console lock, trylock will fail
2527 * and may_schedule may be set. Ignore and proceed to unlock so
2528 * that messages are flushed out. As this can be called from any
2529 * context and we don't want to get preempted while flushing,
2530 * ensure may_schedule is cleared.
2533 console_may_schedule = 0;
2538 * Return the console tty driver structure and its associated index
2540 struct tty_driver *console_device(int *index)
2543 struct tty_driver *driver = NULL;
2546 for_each_console(c) {
2549 driver = c->device(c, index);
2558 * Prevent further output on the passed console device so that (for example)
2559 * serial drivers can disable console output before suspending a port, and can
2560 * re-enable output afterwards.
2562 void console_stop(struct console *console)
2565 console->flags &= ~CON_ENABLED;
2568 EXPORT_SYMBOL(console_stop);
2570 void console_start(struct console *console)
2573 console->flags |= CON_ENABLED;
2576 EXPORT_SYMBOL(console_start);
2578 static int __read_mostly keep_bootcon;
2580 static int __init keep_bootcon_setup(char *str)
2583 pr_info("debug: skip boot console de-registration.\n");
2588 early_param("keep_bootcon", keep_bootcon_setup);
2591 * The console driver calls this routine during kernel initialization
2592 * to register the console printing procedure with printk() and to
2593 * print any messages that were printed by the kernel before the
2594 * console driver was initialized.
2596 * This can happen pretty early during the boot process (because of
2597 * early_printk) - sometimes before setup_arch() completes - be careful
2598 * of what kernel features are used - they may not be initialised yet.
2600 * There are two types of consoles - bootconsoles (early_printk) and
2601 * "real" consoles (everything which is not a bootconsole) which are
2602 * handled differently.
2603 * - Any number of bootconsoles can be registered at any time.
2604 * - As soon as a "real" console is registered, all bootconsoles
2605 * will be unregistered automatically.
2606 * - Once a "real" console is registered, any attempt to register a
2607 * bootconsoles will be rejected
2609 void register_console(struct console *newcon)
2612 unsigned long flags;
2613 struct console *bcon = NULL;
2614 struct console_cmdline *c;
2616 if (console_drivers)
2617 for_each_console(bcon)
2618 if (WARN(bcon == newcon,
2619 "console '%s%d' already registered\n",
2620 bcon->name, bcon->index))
2624 * before we register a new CON_BOOT console, make sure we don't
2625 * already have a valid console
2627 if (console_drivers && newcon->flags & CON_BOOT) {
2628 /* find the last or real console */
2629 for_each_console(bcon) {
2630 if (!(bcon->flags & CON_BOOT)) {
2631 pr_info("Too late to register bootconsole %s%d\n",
2632 newcon->name, newcon->index);
2638 if (console_drivers && console_drivers->flags & CON_BOOT)
2639 bcon = console_drivers;
2641 if (preferred_console < 0 || bcon || !console_drivers)
2642 preferred_console = selected_console;
2645 * See if we want to use this console driver. If we
2646 * didn't select a console we take the first one
2647 * that registers here.
2649 if (preferred_console < 0) {
2650 if (newcon->index < 0)
2652 if (newcon->setup == NULL ||
2653 newcon->setup(newcon, NULL) == 0) {
2654 newcon->flags |= CON_ENABLED;
2655 if (newcon->device) {
2656 newcon->flags |= CON_CONSDEV;
2657 preferred_console = 0;
2663 * See if this console matches one we selected on
2666 for (i = 0, c = console_cmdline;
2667 i < MAX_CMDLINECONSOLES && c->name[0];
2669 if (!newcon->match ||
2670 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2671 /* default matching */
2672 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2673 if (strcmp(c->name, newcon->name) != 0)
2675 if (newcon->index >= 0 &&
2676 newcon->index != c->index)
2678 if (newcon->index < 0)
2679 newcon->index = c->index;
2681 if (_braille_register_console(newcon, c))
2684 if (newcon->setup &&
2685 newcon->setup(newcon, c->options) != 0)
2689 newcon->flags |= CON_ENABLED;
2690 if (i == selected_console) {
2691 newcon->flags |= CON_CONSDEV;
2692 preferred_console = selected_console;
2697 if (!(newcon->flags & CON_ENABLED))
2701 * If we have a bootconsole, and are switching to a real console,
2702 * don't print everything out again, since when the boot console, and
2703 * the real console are the same physical device, it's annoying to
2704 * see the beginning boot messages twice
2706 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2707 newcon->flags &= ~CON_PRINTBUFFER;
2710 * Put this console in the list - keep the
2711 * preferred driver at the head of the list.
2714 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2715 newcon->next = console_drivers;
2716 console_drivers = newcon;
2718 newcon->next->flags &= ~CON_CONSDEV;
2720 newcon->next = console_drivers->next;
2721 console_drivers->next = newcon;
2724 if (newcon->flags & CON_EXTENDED)
2725 if (!nr_ext_console_drivers++)
2726 pr_info("printk: continuation disabled due to ext consoles, expect more fragments in /dev/kmsg\n");
2728 if (newcon->flags & CON_PRINTBUFFER) {
2730 * console_unlock(); will print out the buffered messages
2733 raw_spin_lock_irqsave(&logbuf_lock, flags);
2734 console_seq = syslog_seq;
2735 console_idx = syslog_idx;
2736 console_prev = syslog_prev;
2737 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2739 * We're about to replay the log buffer. Only do this to the
2740 * just-registered console to avoid excessive message spam to
2741 * the already-registered consoles.
2743 exclusive_console = newcon;
2746 console_sysfs_notify();
2749 * By unregistering the bootconsoles after we enable the real console
2750 * we get the "console xxx enabled" message on all the consoles -
2751 * boot consoles, real consoles, etc - this is to ensure that end
2752 * users know there might be something in the kernel's log buffer that
2753 * went to the bootconsole (that they do not see on the real console)
2755 pr_info("%sconsole [%s%d] enabled\n",
2756 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2757 newcon->name, newcon->index);
2759 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2761 /* We need to iterate through all boot consoles, to make
2762 * sure we print everything out, before we unregister them.
2764 for_each_console(bcon)
2765 if (bcon->flags & CON_BOOT)
2766 unregister_console(bcon);
2769 EXPORT_SYMBOL(register_console);
2771 int unregister_console(struct console *console)
2773 struct console *a, *b;
2776 pr_info("%sconsole [%s%d] disabled\n",
2777 (console->flags & CON_BOOT) ? "boot" : "" ,
2778 console->name, console->index);
2780 res = _braille_unregister_console(console);
2786 if (console_drivers == console) {
2787 console_drivers=console->next;
2789 } else if (console_drivers) {
2790 for (a=console_drivers->next, b=console_drivers ;
2791 a; b=a, a=b->next) {
2800 if (!res && (console->flags & CON_EXTENDED))
2801 nr_ext_console_drivers--;
2804 * If this isn't the last console and it has CON_CONSDEV set, we
2805 * need to set it on the next preferred console.
2807 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2808 console_drivers->flags |= CON_CONSDEV;
2810 console->flags &= ~CON_ENABLED;
2812 console_sysfs_notify();
2815 EXPORT_SYMBOL(unregister_console);
2818 * Some boot consoles access data that is in the init section and which will
2819 * be discarded after the initcalls have been run. To make sure that no code
2820 * will access this data, unregister the boot consoles in a late initcall.
2822 * If for some reason, such as deferred probe or the driver being a loadable
2823 * module, the real console hasn't registered yet at this point, there will
2824 * be a brief interval in which no messages are logged to the console, which
2825 * makes it difficult to diagnose problems that occur during this time.
2827 * To mitigate this problem somewhat, only unregister consoles whose memory
2828 * intersects with the init section. Note that code exists elsewhere to get
2829 * rid of the boot console as soon as the proper console shows up, so there
2830 * won't be side-effects from postponing the removal.
2832 static int __init printk_late_init(void)
2834 struct console *con;
2836 for_each_console(con) {
2837 if (!keep_bootcon && con->flags & CON_BOOT) {
2839 * Make sure to unregister boot consoles whose data
2840 * resides in the init section before the init section
2841 * is discarded. Boot consoles whose data will stick
2842 * around will automatically be unregistered when the
2843 * proper console replaces them.
2845 if (init_section_intersects(con, sizeof(*con)))
2846 unregister_console(con);
2849 hotcpu_notifier(console_cpu_notify, 0);
2852 late_initcall(printk_late_init);
2854 #if defined CONFIG_PRINTK
2856 * Delayed printk version, for scheduler-internal messages:
2858 #define PRINTK_PENDING_WAKEUP 0x01
2859 #define PRINTK_PENDING_OUTPUT 0x02
2861 static DEFINE_PER_CPU(int, printk_pending);
2863 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2865 int pending = __this_cpu_xchg(printk_pending, 0);
2867 if (pending & PRINTK_PENDING_OUTPUT) {
2868 /* If trylock fails, someone else is doing the printing */
2869 if (console_trylock())
2873 if (pending & PRINTK_PENDING_WAKEUP)
2874 wake_up_interruptible(&log_wait);
2877 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2878 .func = wake_up_klogd_work_func,
2879 .flags = IRQ_WORK_LAZY,
2882 void wake_up_klogd(void)
2885 if (waitqueue_active(&log_wait)) {
2886 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2887 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2892 int printk_deferred(const char *fmt, ...)
2898 va_start(args, fmt);
2899 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2902 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2903 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2910 * printk rate limiting, lifted from the networking subsystem.
2912 * This enforces a rate limit: not more than 10 kernel messages
2913 * every 5s to make a denial-of-service attack impossible.
2915 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2917 int __printk_ratelimit(const char *func)
2919 return ___ratelimit(&printk_ratelimit_state, func);
2921 EXPORT_SYMBOL(__printk_ratelimit);
2924 * printk_timed_ratelimit - caller-controlled printk ratelimiting
2925 * @caller_jiffies: pointer to caller's state
2926 * @interval_msecs: minimum interval between prints
2928 * printk_timed_ratelimit() returns true if more than @interval_msecs
2929 * milliseconds have elapsed since the last time printk_timed_ratelimit()
2932 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
2933 unsigned int interval_msecs)
2935 unsigned long elapsed = jiffies - *caller_jiffies;
2937 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
2940 *caller_jiffies = jiffies;
2943 EXPORT_SYMBOL(printk_timed_ratelimit);
2945 static DEFINE_SPINLOCK(dump_list_lock);
2946 static LIST_HEAD(dump_list);
2949 * kmsg_dump_register - register a kernel log dumper.
2950 * @dumper: pointer to the kmsg_dumper structure
2952 * Adds a kernel log dumper to the system. The dump callback in the
2953 * structure will be called when the kernel oopses or panics and must be
2954 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
2956 int kmsg_dump_register(struct kmsg_dumper *dumper)
2958 unsigned long flags;
2961 /* The dump callback needs to be set */
2965 spin_lock_irqsave(&dump_list_lock, flags);
2966 /* Don't allow registering multiple times */
2967 if (!dumper->registered) {
2968 dumper->registered = 1;
2969 list_add_tail_rcu(&dumper->list, &dump_list);
2972 spin_unlock_irqrestore(&dump_list_lock, flags);
2976 EXPORT_SYMBOL_GPL(kmsg_dump_register);
2979 * kmsg_dump_unregister - unregister a kmsg dumper.
2980 * @dumper: pointer to the kmsg_dumper structure
2982 * Removes a dump device from the system. Returns zero on success and
2983 * %-EINVAL otherwise.
2985 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
2987 unsigned long flags;
2990 spin_lock_irqsave(&dump_list_lock, flags);
2991 if (dumper->registered) {
2992 dumper->registered = 0;
2993 list_del_rcu(&dumper->list);
2996 spin_unlock_irqrestore(&dump_list_lock, flags);
3001 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3003 static bool always_kmsg_dump;
3004 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3007 * kmsg_dump - dump kernel log to kernel message dumpers.
3008 * @reason: the reason (oops, panic etc) for dumping
3010 * Call each of the registered dumper's dump() callback, which can
3011 * retrieve the kmsg records with kmsg_dump_get_line() or
3012 * kmsg_dump_get_buffer().
3014 void kmsg_dump(enum kmsg_dump_reason reason)
3016 struct kmsg_dumper *dumper;
3017 unsigned long flags;
3019 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3023 list_for_each_entry_rcu(dumper, &dump_list, list) {
3024 if (dumper->max_reason && reason > dumper->max_reason)
3027 /* initialize iterator with data about the stored records */
3028 dumper->active = true;
3030 raw_spin_lock_irqsave(&logbuf_lock, flags);
3032 dumper->cur_seq = clear_seq;
3033 dumper->cur_idx = clear_idx;
3034 dumper->next_seq = log_next_seq;
3035 dumper->next_idx = log_next_idx;
3036 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3038 /* invoke dumper which will iterate over records */
3039 dumper->dump(dumper, reason);
3041 /* reset iterator */
3042 dumper->active = false;
3048 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3049 * @dumper: registered kmsg dumper
3050 * @syslog: include the "<4>" prefixes
3051 * @line: buffer to copy the line to
3052 * @size: maximum size of the buffer
3053 * @len: length of line placed into buffer
3055 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3056 * record, and copy one record into the provided buffer.
3058 * Consecutive calls will return the next available record moving
3059 * towards the end of the buffer with the youngest messages.
3061 * A return value of FALSE indicates that there are no more records to
3064 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3066 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3067 char *line, size_t size, size_t *len)
3069 struct printk_log *msg;
3073 if (!dumper->active)
3076 if (dumper->cur_seq < log_first_seq) {
3077 /* messages are gone, move to first available one */
3078 dumper->cur_seq = log_first_seq;
3079 dumper->cur_idx = log_first_idx;
3083 if (dumper->cur_seq >= log_next_seq)
3086 msg = log_from_idx(dumper->cur_idx);
3087 l = msg_print_text(msg, 0, syslog, line, size);
3089 dumper->cur_idx = log_next(dumper->cur_idx);
3099 * kmsg_dump_get_line - retrieve one kmsg log line
3100 * @dumper: registered kmsg dumper
3101 * @syslog: include the "<4>" prefixes
3102 * @line: buffer to copy the line to
3103 * @size: maximum size of the buffer
3104 * @len: length of line placed into buffer
3106 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3107 * record, and copy one record into the provided buffer.
3109 * Consecutive calls will return the next available record moving
3110 * towards the end of the buffer with the youngest messages.
3112 * A return value of FALSE indicates that there are no more records to
3115 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3116 char *line, size_t size, size_t *len)
3118 unsigned long flags;
3121 raw_spin_lock_irqsave(&logbuf_lock, flags);
3123 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3124 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3128 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3131 * kmsg_dump_get_buffer - copy kmsg log lines
3132 * @dumper: registered kmsg dumper
3133 * @syslog: include the "<4>" prefixes
3134 * @buf: buffer to copy the line to
3135 * @size: maximum size of the buffer
3136 * @len: length of line placed into buffer
3138 * Start at the end of the kmsg buffer and fill the provided buffer
3139 * with as many of the the *youngest* kmsg records that fit into it.
3140 * If the buffer is large enough, all available kmsg records will be
3141 * copied with a single call.
3143 * Consecutive calls will fill the buffer with the next block of
3144 * available older records, not including the earlier retrieved ones.
3146 * A return value of FALSE indicates that there are no more records to
3149 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3150 char *buf, size_t size, size_t *len)
3152 unsigned long flags;
3157 enum log_flags prev;
3161 if (!dumper->active)
3164 raw_spin_lock_irqsave(&logbuf_lock, flags);
3166 if (dumper->cur_seq < log_first_seq) {
3167 /* messages are gone, move to first available one */
3168 dumper->cur_seq = log_first_seq;
3169 dumper->cur_idx = log_first_idx;
3173 if (dumper->cur_seq >= dumper->next_seq) {
3174 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3178 /* calculate length of entire buffer */
3179 seq = dumper->cur_seq;
3180 idx = dumper->cur_idx;
3182 while (seq < dumper->next_seq) {
3183 struct printk_log *msg = log_from_idx(idx);
3185 l += msg_print_text(msg, prev, true, NULL, 0);
3186 idx = log_next(idx);
3191 /* move first record forward until length fits into the buffer */
3192 seq = dumper->cur_seq;
3193 idx = dumper->cur_idx;
3195 while (l > size && seq < dumper->next_seq) {
3196 struct printk_log *msg = log_from_idx(idx);
3198 l -= msg_print_text(msg, prev, true, NULL, 0);
3199 idx = log_next(idx);
3204 /* last message in next interation */
3209 while (seq < dumper->next_seq) {
3210 struct printk_log *msg = log_from_idx(idx);
3212 l += msg_print_text(msg, prev, syslog, buf + l, size - l);
3213 idx = log_next(idx);
3218 dumper->next_seq = next_seq;
3219 dumper->next_idx = next_idx;
3221 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3227 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3230 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3231 * @dumper: registered kmsg dumper
3233 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3234 * kmsg_dump_get_buffer() can be called again and used multiple
3235 * times within the same dumper.dump() callback.
3237 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3239 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3241 dumper->cur_seq = clear_seq;
3242 dumper->cur_idx = clear_idx;
3243 dumper->next_seq = log_next_seq;
3244 dumper->next_idx = log_next_idx;
3248 * kmsg_dump_rewind - reset the interator
3249 * @dumper: registered kmsg dumper
3251 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3252 * kmsg_dump_get_buffer() can be called again and used multiple
3253 * times within the same dumper.dump() callback.
3255 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3257 unsigned long flags;
3259 raw_spin_lock_irqsave(&logbuf_lock, flags);
3260 kmsg_dump_rewind_nolock(dumper);
3261 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3263 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3265 static char dump_stack_arch_desc_str[128];
3268 * dump_stack_set_arch_desc - set arch-specific str to show with task dumps
3269 * @fmt: printf-style format string
3270 * @...: arguments for the format string
3272 * The configured string will be printed right after utsname during task
3273 * dumps. Usually used to add arch-specific system identifiers. If an
3274 * arch wants to make use of such an ID string, it should initialize this
3275 * as soon as possible during boot.
3277 void __init dump_stack_set_arch_desc(const char *fmt, ...)
3281 va_start(args, fmt);
3282 vsnprintf(dump_stack_arch_desc_str, sizeof(dump_stack_arch_desc_str),
3288 * dump_stack_print_info - print generic debug info for dump_stack()
3289 * @log_lvl: log level
3291 * Arch-specific dump_stack() implementations can use this function to
3292 * print out the same debug information as the generic dump_stack().
3294 void dump_stack_print_info(const char *log_lvl)
3296 printk("%sCPU: %d PID: %d Comm: %.20s %s %s %.*s\n",
3297 log_lvl, raw_smp_processor_id(), current->pid, current->comm,
3298 print_tainted(), init_utsname()->release,
3299 (int)strcspn(init_utsname()->version, " "),
3300 init_utsname()->version);
3302 if (dump_stack_arch_desc_str[0] != '\0')
3303 printk("%sHardware name: %s\n",
3304 log_lvl, dump_stack_arch_desc_str);
3306 print_worker_info(log_lvl, current);
3310 * show_regs_print_info - print generic debug info for show_regs()
3311 * @log_lvl: log level
3313 * show_regs() implementations can use this function to print out generic
3314 * debug information.
3316 void show_regs_print_info(const char *log_lvl)
3318 dump_stack_print_info(log_lvl);
3320 printk("%stask: %p task.stack: %p\n",
3321 log_lvl, current, task_stack_page(current));