2 * linux/kernel/printk.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * Modified to make sys_syslog() more flexible: added commands to
7 * return the last 4k of kernel messages, regardless of whether
8 * they've been read or not. Added option to suppress kernel printk's
9 * to the console. Added hook for sending the console messages
10 * elsewhere, in preparation for a serial line console (someday).
12 * Modified for sysctl support, 1/8/97, Chris Horn.
13 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
14 * manfred@colorfullife.com
15 * Rewrote bits to get rid of console_lock
16 * 01Mar01 Andrew Morton
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/kernel.h>
23 #include <linux/tty.h>
24 #include <linux/tty_driver.h>
25 #include <linux/console.h>
26 #include <linux/init.h>
27 #include <linux/jiffies.h>
28 #include <linux/nmi.h>
29 #include <linux/module.h>
30 #include <linux/moduleparam.h>
31 #include <linux/delay.h>
32 #include <linux/smp.h>
33 #include <linux/security.h>
34 #include <linux/bootmem.h>
35 #include <linux/memblock.h>
36 #include <linux/syscalls.h>
37 #include <linux/crash_core.h>
38 #include <linux/kdb.h>
39 #include <linux/ratelimit.h>
40 #include <linux/kmsg_dump.h>
41 #include <linux/syslog.h>
42 #include <linux/cpu.h>
43 #include <linux/rculist.h>
44 #include <linux/poll.h>
45 #include <linux/irq_work.h>
46 #include <linux/ctype.h>
47 #include <linux/uio.h>
48 #include <linux/sched/clock.h>
49 #include <linux/sched/debug.h>
50 #include <linux/sched/task_stack.h>
52 #include <linux/uaccess.h>
53 #include <asm/sections.h>
55 #include <trace/events/initcall.h>
56 #define CREATE_TRACE_POINTS
57 #include <trace/events/printk.h>
59 #include "console_cmdline.h"
63 int console_printk[4] = {
64 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
65 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
66 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
67 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
70 atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
71 EXPORT_SYMBOL(ignore_console_lock_warning);
74 * Low level drivers may need that to know if they can schedule in
75 * their unblank() callback or not. So let's export it.
78 EXPORT_SYMBOL(oops_in_progress);
81 * console_sem protects the console_drivers list, and also
82 * provides serialisation for access to the entire console
85 static DEFINE_SEMAPHORE(console_sem);
86 struct console *console_drivers;
87 EXPORT_SYMBOL_GPL(console_drivers);
90 static struct lockdep_map console_lock_dep_map = {
91 .name = "console_lock"
95 enum devkmsg_log_bits {
96 __DEVKMSG_LOG_BIT_ON = 0,
97 __DEVKMSG_LOG_BIT_OFF,
98 __DEVKMSG_LOG_BIT_LOCK,
101 enum devkmsg_log_masks {
102 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
103 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
104 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
107 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
108 #define DEVKMSG_LOG_MASK_DEFAULT 0
110 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
112 static int __control_devkmsg(char *str)
117 if (!strncmp(str, "on", 2)) {
118 devkmsg_log = DEVKMSG_LOG_MASK_ON;
120 } else if (!strncmp(str, "off", 3)) {
121 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
123 } else if (!strncmp(str, "ratelimit", 9)) {
124 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
130 static int __init control_devkmsg(char *str)
132 if (__control_devkmsg(str) < 0)
136 * Set sysctl string accordingly:
138 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
139 strcpy(devkmsg_log_str, "on");
140 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
141 strcpy(devkmsg_log_str, "off");
142 /* else "ratelimit" which is set by default. */
145 * Sysctl cannot change it anymore. The kernel command line setting of
146 * this parameter is to force the setting to be permanent throughout the
147 * runtime of the system. This is a precation measure against userspace
148 * trying to be a smarta** and attempting to change it up on us.
150 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
154 __setup("printk.devkmsg=", control_devkmsg);
156 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
158 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
159 void __user *buffer, size_t *lenp, loff_t *ppos)
161 char old_str[DEVKMSG_STR_MAX_SIZE];
166 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
170 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
173 err = proc_dostring(table, write, buffer, lenp, ppos);
178 err = __control_devkmsg(devkmsg_log_str);
181 * Do not accept an unknown string OR a known string with
184 if (err < 0 || (err + 1 != *lenp)) {
186 /* ... and restore old setting. */
188 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
197 /* Number of registered extended console drivers. */
198 static int nr_ext_console_drivers;
201 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
202 * macros instead of functions so that _RET_IP_ contains useful information.
204 #define down_console_sem() do { \
206 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
209 static int __down_trylock_console_sem(unsigned long ip)
215 * Here and in __up_console_sem() we need to be in safe mode,
216 * because spindump/WARN/etc from under console ->lock will
217 * deadlock in printk()->down_trylock_console_sem() otherwise.
219 printk_safe_enter_irqsave(flags);
220 lock_failed = down_trylock(&console_sem);
221 printk_safe_exit_irqrestore(flags);
225 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
228 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
230 static void __up_console_sem(unsigned long ip)
234 mutex_release(&console_lock_dep_map, 1, ip);
236 printk_safe_enter_irqsave(flags);
238 printk_safe_exit_irqrestore(flags);
240 #define up_console_sem() __up_console_sem(_RET_IP_)
243 * This is used for debugging the mess that is the VT code by
244 * keeping track if we have the console semaphore held. It's
245 * definitely not the perfect debug tool (we don't know if _WE_
246 * hold it and are racing, but it helps tracking those weird code
247 * paths in the console code where we end up in places I want
248 * locked without the console sempahore held).
250 static int console_locked, console_suspended;
253 * If exclusive_console is non-NULL then only this console is to be printed to.
255 static struct console *exclusive_console;
258 * Array of consoles built from command line options (console=)
261 #define MAX_CMDLINECONSOLES 8
263 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
265 static int preferred_console = -1;
266 int console_set_on_cmdline;
267 EXPORT_SYMBOL(console_set_on_cmdline);
269 /* Flag: console code may call schedule() */
270 static int console_may_schedule;
272 enum con_msg_format_flags {
273 MSG_FORMAT_DEFAULT = 0,
274 MSG_FORMAT_SYSLOG = (1 << 0),
277 static int console_msg_format = MSG_FORMAT_DEFAULT;
280 * The printk log buffer consists of a chain of concatenated variable
281 * length records. Every record starts with a record header, containing
282 * the overall length of the record.
284 * The heads to the first and last entry in the buffer, as well as the
285 * sequence numbers of these entries are maintained when messages are
288 * If the heads indicate available messages, the length in the header
289 * tells the start next message. A length == 0 for the next message
290 * indicates a wrap-around to the beginning of the buffer.
292 * Every record carries the monotonic timestamp in microseconds, as well as
293 * the standard userspace syslog level and syslog facility. The usual
294 * kernel messages use LOG_KERN; userspace-injected messages always carry
295 * a matching syslog facility, by default LOG_USER. The origin of every
296 * message can be reliably determined that way.
298 * The human readable log message directly follows the message header. The
299 * length of the message text is stored in the header, the stored message
302 * Optionally, a message can carry a dictionary of properties (key/value pairs),
303 * to provide userspace with a machine-readable message context.
305 * Examples for well-defined, commonly used property names are:
306 * DEVICE=b12:8 device identifier
310 * +sound:card0 subsystem:devname
311 * SUBSYSTEM=pci driver-core subsystem name
313 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
314 * follows directly after a '=' character. Every property is terminated by
315 * a '\0' character. The last property is not terminated.
317 * Example of a message structure:
318 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
319 * 0008 34 00 record is 52 bytes long
320 * 000a 0b 00 text is 11 bytes long
321 * 000c 1f 00 dictionary is 23 bytes long
322 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
323 * 0010 69 74 27 73 20 61 20 6c "it's a l"
325 * 001b 44 45 56 49 43 "DEVIC"
326 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
327 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
329 * 0032 00 00 00 padding to next message header
331 * The 'struct printk_log' buffer header must never be directly exported to
332 * userspace, it is a kernel-private implementation detail that might
333 * need to be changed in the future, when the requirements change.
335 * /dev/kmsg exports the structured data in the following line format:
336 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
338 * Users of the export format should ignore possible additional values
339 * separated by ',', and find the message after the ';' character.
341 * The optional key/value pairs are attached as continuation lines starting
342 * with a space character and terminated by a newline. All possible
343 * non-prinatable characters are escaped in the "\xff" notation.
347 LOG_NEWLINE = 2, /* text ended with a newline */
348 LOG_PREFIX = 4, /* text started with a prefix */
349 LOG_CONT = 8, /* text is a fragment of a continuation line */
353 u64 ts_nsec; /* timestamp in nanoseconds */
354 u16 len; /* length of entire record */
355 u16 text_len; /* length of text buffer */
356 u16 dict_len; /* length of dictionary buffer */
357 u8 facility; /* syslog facility */
358 u8 flags:5; /* internal record flags */
359 u8 level:3; /* syslog level */
361 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
362 __packed __aligned(4)
367 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
368 * within the scheduler's rq lock. It must be released before calling
369 * console_unlock() or anything else that might wake up a process.
371 DEFINE_RAW_SPINLOCK(logbuf_lock);
374 * Helper macros to lock/unlock logbuf_lock and switch between
375 * printk-safe/unsafe modes.
377 #define logbuf_lock_irq() \
379 printk_safe_enter_irq(); \
380 raw_spin_lock(&logbuf_lock); \
383 #define logbuf_unlock_irq() \
385 raw_spin_unlock(&logbuf_lock); \
386 printk_safe_exit_irq(); \
389 #define logbuf_lock_irqsave(flags) \
391 printk_safe_enter_irqsave(flags); \
392 raw_spin_lock(&logbuf_lock); \
395 #define logbuf_unlock_irqrestore(flags) \
397 raw_spin_unlock(&logbuf_lock); \
398 printk_safe_exit_irqrestore(flags); \
402 DECLARE_WAIT_QUEUE_HEAD(log_wait);
403 /* the next printk record to read by syslog(READ) or /proc/kmsg */
404 static u64 syslog_seq;
405 static u32 syslog_idx;
406 static size_t syslog_partial;
408 /* index and sequence number of the first record stored in the buffer */
409 static u64 log_first_seq;
410 static u32 log_first_idx;
412 /* index and sequence number of the next record to store in the buffer */
413 static u64 log_next_seq;
414 static u32 log_next_idx;
416 /* the next printk record to write to the console */
417 static u64 console_seq;
418 static u32 console_idx;
419 static u64 exclusive_console_stop_seq;
421 /* the next printk record to read after the last 'clear' command */
422 static u64 clear_seq;
423 static u32 clear_idx;
425 #define PREFIX_MAX 32
426 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
428 #define LOG_LEVEL(v) ((v) & 0x07)
429 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
432 #define LOG_ALIGN __alignof__(struct printk_log)
433 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
434 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
435 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
436 static char *log_buf = __log_buf;
437 static u32 log_buf_len = __LOG_BUF_LEN;
439 /* Return log buffer address */
440 char *log_buf_addr_get(void)
445 /* Return log buffer size */
446 u32 log_buf_len_get(void)
451 /* human readable text of the record */
452 static char *log_text(const struct printk_log *msg)
454 return (char *)msg + sizeof(struct printk_log);
457 /* optional key/value pair dictionary attached to the record */
458 static char *log_dict(const struct printk_log *msg)
460 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
463 /* get record by index; idx must point to valid msg */
464 static struct printk_log *log_from_idx(u32 idx)
466 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
469 * A length == 0 record is the end of buffer marker. Wrap around and
470 * read the message at the start of the buffer.
473 return (struct printk_log *)log_buf;
477 /* get next record; idx must point to valid msg */
478 static u32 log_next(u32 idx)
480 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
482 /* length == 0 indicates the end of the buffer; wrap */
484 * A length == 0 record is the end of buffer marker. Wrap around and
485 * read the message at the start of the buffer as *this* one, and
486 * return the one after that.
489 msg = (struct printk_log *)log_buf;
492 return idx + msg->len;
496 * Check whether there is enough free space for the given message.
498 * The same values of first_idx and next_idx mean that the buffer
499 * is either empty or full.
501 * If the buffer is empty, we must respect the position of the indexes.
502 * They cannot be reset to the beginning of the buffer.
504 static int logbuf_has_space(u32 msg_size, bool empty)
508 if (log_next_idx > log_first_idx || empty)
509 free = max(log_buf_len - log_next_idx, log_first_idx);
511 free = log_first_idx - log_next_idx;
514 * We need space also for an empty header that signalizes wrapping
517 return free >= msg_size + sizeof(struct printk_log);
520 static int log_make_free_space(u32 msg_size)
522 while (log_first_seq < log_next_seq &&
523 !logbuf_has_space(msg_size, false)) {
524 /* drop old messages until we have enough contiguous space */
525 log_first_idx = log_next(log_first_idx);
529 if (clear_seq < log_first_seq) {
530 clear_seq = log_first_seq;
531 clear_idx = log_first_idx;
534 /* sequence numbers are equal, so the log buffer is empty */
535 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
541 /* compute the message size including the padding bytes */
542 static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
546 size = sizeof(struct printk_log) + text_len + dict_len;
547 *pad_len = (-size) & (LOG_ALIGN - 1);
554 * Define how much of the log buffer we could take at maximum. The value
555 * must be greater than two. Note that only half of the buffer is available
556 * when the index points to the middle.
558 #define MAX_LOG_TAKE_PART 4
559 static const char trunc_msg[] = "<truncated>";
561 static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
562 u16 *dict_len, u32 *pad_len)
565 * The message should not take the whole buffer. Otherwise, it might
566 * get removed too soon.
568 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
569 if (*text_len > max_text_len)
570 *text_len = max_text_len;
571 /* enable the warning message */
572 *trunc_msg_len = strlen(trunc_msg);
573 /* disable the "dict" completely */
575 /* compute the size again, count also the warning message */
576 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
579 /* insert record into the buffer, discard old ones, update heads */
580 static int log_store(int facility, int level,
581 enum log_flags flags, u64 ts_nsec,
582 const char *dict, u16 dict_len,
583 const char *text, u16 text_len)
585 struct printk_log *msg;
587 u16 trunc_msg_len = 0;
589 /* number of '\0' padding bytes to next message */
590 size = msg_used_size(text_len, dict_len, &pad_len);
592 if (log_make_free_space(size)) {
593 /* truncate the message if it is too long for empty buffer */
594 size = truncate_msg(&text_len, &trunc_msg_len,
595 &dict_len, &pad_len);
596 /* survive when the log buffer is too small for trunc_msg */
597 if (log_make_free_space(size))
601 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
603 * This message + an additional empty header does not fit
604 * at the end of the buffer. Add an empty header with len == 0
605 * to signify a wrap around.
607 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
612 msg = (struct printk_log *)(log_buf + log_next_idx);
613 memcpy(log_text(msg), text, text_len);
614 msg->text_len = text_len;
616 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
617 msg->text_len += trunc_msg_len;
619 memcpy(log_dict(msg), dict, dict_len);
620 msg->dict_len = dict_len;
621 msg->facility = facility;
622 msg->level = level & 7;
623 msg->flags = flags & 0x1f;
625 msg->ts_nsec = ts_nsec;
627 msg->ts_nsec = local_clock();
628 memset(log_dict(msg) + dict_len, 0, pad_len);
632 log_next_idx += msg->len;
635 return msg->text_len;
638 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
640 static int syslog_action_restricted(int type)
645 * Unless restricted, we allow "read all" and "get buffer size"
648 return type != SYSLOG_ACTION_READ_ALL &&
649 type != SYSLOG_ACTION_SIZE_BUFFER;
652 static int check_syslog_permissions(int type, int source)
655 * If this is from /proc/kmsg and we've already opened it, then we've
656 * already done the capabilities checks at open time.
658 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
661 if (syslog_action_restricted(type)) {
662 if (capable(CAP_SYSLOG))
665 * For historical reasons, accept CAP_SYS_ADMIN too, with
668 if (capable(CAP_SYS_ADMIN)) {
669 pr_warn_once("%s (%d): Attempt to access syslog with "
670 "CAP_SYS_ADMIN but no CAP_SYSLOG "
672 current->comm, task_pid_nr(current));
678 return security_syslog(type);
681 static void append_char(char **pp, char *e, char c)
687 static ssize_t msg_print_ext_header(char *buf, size_t size,
688 struct printk_log *msg, u64 seq)
690 u64 ts_usec = msg->ts_nsec;
692 do_div(ts_usec, 1000);
694 return scnprintf(buf, size, "%u,%llu,%llu,%c;",
695 (msg->facility << 3) | msg->level, seq, ts_usec,
696 msg->flags & LOG_CONT ? 'c' : '-');
699 static ssize_t msg_print_ext_body(char *buf, size_t size,
700 char *dict, size_t dict_len,
701 char *text, size_t text_len)
703 char *p = buf, *e = buf + size;
706 /* escape non-printable characters */
707 for (i = 0; i < text_len; i++) {
708 unsigned char c = text[i];
710 if (c < ' ' || c >= 127 || c == '\\')
711 p += scnprintf(p, e - p, "\\x%02x", c);
713 append_char(&p, e, c);
715 append_char(&p, e, '\n');
720 for (i = 0; i < dict_len; i++) {
721 unsigned char c = dict[i];
724 append_char(&p, e, ' ');
729 append_char(&p, e, '\n');
734 if (c < ' ' || c >= 127 || c == '\\') {
735 p += scnprintf(p, e - p, "\\x%02x", c);
739 append_char(&p, e, c);
741 append_char(&p, e, '\n');
747 /* /dev/kmsg - userspace message inject/listen interface */
748 struct devkmsg_user {
751 struct ratelimit_state rs;
753 char buf[CONSOLE_EXT_LOG_MAX];
756 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
759 int level = default_message_loglevel;
760 int facility = 1; /* LOG_USER */
761 struct file *file = iocb->ki_filp;
762 struct devkmsg_user *user = file->private_data;
763 size_t len = iov_iter_count(from);
766 if (!user || len > LOG_LINE_MAX)
769 /* Ignore when user logging is disabled. */
770 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
773 /* Ratelimit when not explicitly enabled. */
774 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
775 if (!___ratelimit(&user->rs, current->comm))
779 buf = kmalloc(len+1, GFP_KERNEL);
784 if (!copy_from_iter_full(buf, len, from)) {
790 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
791 * the decimal value represents 32bit, the lower 3 bit are the log
792 * level, the rest are the log facility.
794 * If no prefix or no userspace facility is specified, we
795 * enforce LOG_USER, to be able to reliably distinguish
796 * kernel-generated messages from userspace-injected ones.
799 if (line[0] == '<') {
803 u = simple_strtoul(line + 1, &endp, 10);
804 if (endp && endp[0] == '>') {
805 level = LOG_LEVEL(u);
806 if (LOG_FACILITY(u) != 0)
807 facility = LOG_FACILITY(u);
814 printk_emit(facility, level, NULL, 0, "%s", line);
819 static ssize_t devkmsg_read(struct file *file, char __user *buf,
820 size_t count, loff_t *ppos)
822 struct devkmsg_user *user = file->private_data;
823 struct printk_log *msg;
830 ret = mutex_lock_interruptible(&user->lock);
835 while (user->seq == log_next_seq) {
836 if (file->f_flags & O_NONBLOCK) {
843 ret = wait_event_interruptible(log_wait,
844 user->seq != log_next_seq);
850 if (user->seq < log_first_seq) {
851 /* our last seen message is gone, return error and reset */
852 user->idx = log_first_idx;
853 user->seq = log_first_seq;
859 msg = log_from_idx(user->idx);
860 len = msg_print_ext_header(user->buf, sizeof(user->buf),
862 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
863 log_dict(msg), msg->dict_len,
864 log_text(msg), msg->text_len);
866 user->idx = log_next(user->idx);
875 if (copy_to_user(buf, user->buf, len)) {
881 mutex_unlock(&user->lock);
885 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
887 struct devkmsg_user *user = file->private_data;
898 /* the first record */
899 user->idx = log_first_idx;
900 user->seq = log_first_seq;
904 * The first record after the last SYSLOG_ACTION_CLEAR,
905 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
906 * changes no global state, and does not clear anything.
908 user->idx = clear_idx;
909 user->seq = clear_seq;
912 /* after the last record */
913 user->idx = log_next_idx;
914 user->seq = log_next_seq;
923 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
925 struct devkmsg_user *user = file->private_data;
929 return EPOLLERR|EPOLLNVAL;
931 poll_wait(file, &log_wait, wait);
934 if (user->seq < log_next_seq) {
935 /* return error when data has vanished underneath us */
936 if (user->seq < log_first_seq)
937 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
939 ret = EPOLLIN|EPOLLRDNORM;
946 static int devkmsg_open(struct inode *inode, struct file *file)
948 struct devkmsg_user *user;
951 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
954 /* write-only does not need any file context */
955 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
956 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
962 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
966 ratelimit_default_init(&user->rs);
967 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
969 mutex_init(&user->lock);
972 user->idx = log_first_idx;
973 user->seq = log_first_seq;
976 file->private_data = user;
980 static int devkmsg_release(struct inode *inode, struct file *file)
982 struct devkmsg_user *user = file->private_data;
987 ratelimit_state_exit(&user->rs);
989 mutex_destroy(&user->lock);
994 const struct file_operations kmsg_fops = {
995 .open = devkmsg_open,
996 .read = devkmsg_read,
997 .write_iter = devkmsg_write,
998 .llseek = devkmsg_llseek,
999 .poll = devkmsg_poll,
1000 .release = devkmsg_release,
1003 #ifdef CONFIG_CRASH_CORE
1005 * This appends the listed symbols to /proc/vmcore
1007 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1008 * obtain access to symbols that are otherwise very difficult to locate. These
1009 * symbols are specifically used so that utilities can access and extract the
1010 * dmesg log from a vmcore file after a crash.
1012 void log_buf_vmcoreinfo_setup(void)
1014 VMCOREINFO_SYMBOL(log_buf);
1015 VMCOREINFO_SYMBOL(log_buf_len);
1016 VMCOREINFO_SYMBOL(log_first_idx);
1017 VMCOREINFO_SYMBOL(clear_idx);
1018 VMCOREINFO_SYMBOL(log_next_idx);
1020 * Export struct printk_log size and field offsets. User space tools can
1021 * parse it and detect any changes to structure down the line.
1023 VMCOREINFO_STRUCT_SIZE(printk_log);
1024 VMCOREINFO_OFFSET(printk_log, ts_nsec);
1025 VMCOREINFO_OFFSET(printk_log, len);
1026 VMCOREINFO_OFFSET(printk_log, text_len);
1027 VMCOREINFO_OFFSET(printk_log, dict_len);
1031 /* requested log_buf_len from kernel cmdline */
1032 static unsigned long __initdata new_log_buf_len;
1034 /* we practice scaling the ring buffer by powers of 2 */
1035 static void __init log_buf_len_update(u64 size)
1037 if (size > (u64)LOG_BUF_LEN_MAX) {
1038 size = (u64)LOG_BUF_LEN_MAX;
1039 pr_err("log_buf over 2G is not supported.\n");
1043 size = roundup_pow_of_two(size);
1044 if (size > log_buf_len)
1045 new_log_buf_len = (unsigned long)size;
1048 /* save requested log_buf_len since it's too early to process it */
1049 static int __init log_buf_len_setup(char *str)
1056 size = memparse(str, &str);
1058 log_buf_len_update(size);
1062 early_param("log_buf_len", log_buf_len_setup);
1065 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1067 static void __init log_buf_add_cpu(void)
1069 unsigned int cpu_extra;
1072 * archs should set up cpu_possible_bits properly with
1073 * set_cpu_possible() after setup_arch() but just in
1074 * case lets ensure this is valid.
1076 if (num_possible_cpus() == 1)
1079 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1081 /* by default this will only continue through for large > 64 CPUs */
1082 if (cpu_extra <= __LOG_BUF_LEN / 2)
1085 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1086 __LOG_CPU_MAX_BUF_LEN);
1087 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1089 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1091 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1093 #else /* !CONFIG_SMP */
1094 static inline void log_buf_add_cpu(void) {}
1095 #endif /* CONFIG_SMP */
1097 void __init setup_log_buf(int early)
1099 unsigned long flags;
1103 if (log_buf != __log_buf)
1106 if (!early && !new_log_buf_len)
1109 if (!new_log_buf_len)
1114 memblock_alloc(new_log_buf_len, LOG_ALIGN);
1116 new_log_buf = memblock_alloc_nopanic(new_log_buf_len,
1120 if (unlikely(!new_log_buf)) {
1121 pr_err("log_buf_len: %lu bytes not available\n",
1126 logbuf_lock_irqsave(flags);
1127 log_buf_len = new_log_buf_len;
1128 log_buf = new_log_buf;
1129 new_log_buf_len = 0;
1130 free = __LOG_BUF_LEN - log_next_idx;
1131 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1132 logbuf_unlock_irqrestore(flags);
1134 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1135 pr_info("early log buf free: %u(%u%%)\n",
1136 free, (free * 100) / __LOG_BUF_LEN);
1139 static bool __read_mostly ignore_loglevel;
1141 static int __init ignore_loglevel_setup(char *str)
1143 ignore_loglevel = true;
1144 pr_info("debug: ignoring loglevel setting.\n");
1149 early_param("ignore_loglevel", ignore_loglevel_setup);
1150 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1151 MODULE_PARM_DESC(ignore_loglevel,
1152 "ignore loglevel setting (prints all kernel messages to the console)");
1154 static bool suppress_message_printing(int level)
1156 return (level >= console_loglevel && !ignore_loglevel);
1159 #ifdef CONFIG_BOOT_PRINTK_DELAY
1161 static int boot_delay; /* msecs delay after each printk during bootup */
1162 static unsigned long long loops_per_msec; /* based on boot_delay */
1164 static int __init boot_delay_setup(char *str)
1168 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1169 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1171 get_option(&str, &boot_delay);
1172 if (boot_delay > 10 * 1000)
1175 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1176 "HZ: %d, loops_per_msec: %llu\n",
1177 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1180 early_param("boot_delay", boot_delay_setup);
1182 static void boot_delay_msec(int level)
1184 unsigned long long k;
1185 unsigned long timeout;
1187 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1188 || suppress_message_printing(level)) {
1192 k = (unsigned long long)loops_per_msec * boot_delay;
1194 timeout = jiffies + msecs_to_jiffies(boot_delay);
1199 * use (volatile) jiffies to prevent
1200 * compiler reduction; loop termination via jiffies
1201 * is secondary and may or may not happen.
1203 if (time_after(jiffies, timeout))
1205 touch_nmi_watchdog();
1209 static inline void boot_delay_msec(int level)
1214 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1215 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1217 static size_t print_time(u64 ts, char *buf)
1219 unsigned long rem_nsec;
1224 rem_nsec = do_div(ts, 1000000000);
1227 return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
1229 return sprintf(buf, "[%5lu.%06lu] ",
1230 (unsigned long)ts, rem_nsec / 1000);
1233 static size_t print_prefix(const struct printk_log *msg, bool syslog, char *buf)
1236 unsigned int prefix = (msg->facility << 3) | msg->level;
1240 len += sprintf(buf, "<%u>", prefix);
1245 else if (prefix > 99)
1247 else if (prefix > 9)
1252 len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
1256 static size_t msg_print_text(const struct printk_log *msg, bool syslog, char *buf, size_t size)
1258 const char *text = log_text(msg);
1259 size_t text_size = msg->text_len;
1263 const char *next = memchr(text, '\n', text_size);
1267 text_len = next - text;
1269 text_size -= next - text;
1271 text_len = text_size;
1275 if (print_prefix(msg, syslog, NULL) +
1276 text_len + 1 >= size - len)
1279 len += print_prefix(msg, syslog, buf + len);
1280 memcpy(buf + len, text, text_len);
1284 /* SYSLOG_ACTION_* buffer size only calculation */
1285 len += print_prefix(msg, syslog, NULL);
1296 static int syslog_print(char __user *buf, int size)
1299 struct printk_log *msg;
1302 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1311 if (syslog_seq < log_first_seq) {
1312 /* messages are gone, move to first one */
1313 syslog_seq = log_first_seq;
1314 syslog_idx = log_first_idx;
1317 if (syslog_seq == log_next_seq) {
1318 logbuf_unlock_irq();
1322 skip = syslog_partial;
1323 msg = log_from_idx(syslog_idx);
1324 n = msg_print_text(msg, true, text, LOG_LINE_MAX + PREFIX_MAX);
1325 if (n - syslog_partial <= size) {
1326 /* message fits into buffer, move forward */
1327 syslog_idx = log_next(syslog_idx);
1329 n -= syslog_partial;
1332 /* partial read(), remember position */
1334 syslog_partial += n;
1337 logbuf_unlock_irq();
1342 if (copy_to_user(buf, text + skip, n)) {
1357 static int syslog_print_all(char __user *buf, int size, bool clear)
1365 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1371 * Find first record that fits, including all following records,
1372 * into the user-provided buffer for this dump.
1376 while (seq < log_next_seq) {
1377 struct printk_log *msg = log_from_idx(idx);
1379 len += msg_print_text(msg, true, NULL, 0);
1380 idx = log_next(idx);
1384 /* move first record forward until length fits into the buffer */
1387 while (len > size && seq < log_next_seq) {
1388 struct printk_log *msg = log_from_idx(idx);
1390 len -= msg_print_text(msg, true, NULL, 0);
1391 idx = log_next(idx);
1395 /* last message fitting into this dump */
1396 next_seq = log_next_seq;
1399 while (len >= 0 && seq < next_seq) {
1400 struct printk_log *msg = log_from_idx(idx);
1403 textlen = msg_print_text(msg, true, text,
1404 LOG_LINE_MAX + PREFIX_MAX);
1409 idx = log_next(idx);
1412 logbuf_unlock_irq();
1413 if (copy_to_user(buf + len, text, textlen))
1419 if (seq < log_first_seq) {
1420 /* messages are gone, move to next one */
1421 seq = log_first_seq;
1422 idx = log_first_idx;
1427 clear_seq = log_next_seq;
1428 clear_idx = log_next_idx;
1430 logbuf_unlock_irq();
1436 static void syslog_clear(void)
1439 clear_seq = log_next_seq;
1440 clear_idx = log_next_idx;
1441 logbuf_unlock_irq();
1444 int do_syslog(int type, char __user *buf, int len, int source)
1447 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1450 error = check_syslog_permissions(type, source);
1455 case SYSLOG_ACTION_CLOSE: /* Close log */
1457 case SYSLOG_ACTION_OPEN: /* Open log */
1459 case SYSLOG_ACTION_READ: /* Read from log */
1460 if (!buf || len < 0)
1464 if (!access_ok(VERIFY_WRITE, buf, len))
1466 error = wait_event_interruptible(log_wait,
1467 syslog_seq != log_next_seq);
1470 error = syslog_print(buf, len);
1472 /* Read/clear last kernel messages */
1473 case SYSLOG_ACTION_READ_CLEAR:
1476 /* Read last kernel messages */
1477 case SYSLOG_ACTION_READ_ALL:
1478 if (!buf || len < 0)
1482 if (!access_ok(VERIFY_WRITE, buf, len))
1484 error = syslog_print_all(buf, len, clear);
1486 /* Clear ring buffer */
1487 case SYSLOG_ACTION_CLEAR:
1490 /* Disable logging to console */
1491 case SYSLOG_ACTION_CONSOLE_OFF:
1492 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1493 saved_console_loglevel = console_loglevel;
1494 console_loglevel = minimum_console_loglevel;
1496 /* Enable logging to console */
1497 case SYSLOG_ACTION_CONSOLE_ON:
1498 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1499 console_loglevel = saved_console_loglevel;
1500 saved_console_loglevel = LOGLEVEL_DEFAULT;
1503 /* Set level of messages printed to console */
1504 case SYSLOG_ACTION_CONSOLE_LEVEL:
1505 if (len < 1 || len > 8)
1507 if (len < minimum_console_loglevel)
1508 len = minimum_console_loglevel;
1509 console_loglevel = len;
1510 /* Implicitly re-enable logging to console */
1511 saved_console_loglevel = LOGLEVEL_DEFAULT;
1513 /* Number of chars in the log buffer */
1514 case SYSLOG_ACTION_SIZE_UNREAD:
1516 if (syslog_seq < log_first_seq) {
1517 /* messages are gone, move to first one */
1518 syslog_seq = log_first_seq;
1519 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;
1533 while (seq < log_next_seq) {
1534 struct printk_log *msg = log_from_idx(idx);
1536 error += msg_print_text(msg, true, NULL, 0);
1537 idx = log_next(idx);
1540 error -= syslog_partial;
1542 logbuf_unlock_irq();
1544 /* Size of the log buffer */
1545 case SYSLOG_ACTION_SIZE_BUFFER:
1546 error = log_buf_len;
1556 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1558 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1562 * Special console_lock variants that help to reduce the risk of soft-lockups.
1563 * They allow to pass console_lock to another printk() call using a busy wait.
1566 #ifdef CONFIG_LOCKDEP
1567 static struct lockdep_map console_owner_dep_map = {
1568 .name = "console_owner"
1572 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1573 static struct task_struct *console_owner;
1574 static bool console_waiter;
1577 * console_lock_spinning_enable - mark beginning of code where another
1578 * thread might safely busy wait
1580 * This basically converts console_lock into a spinlock. This marks
1581 * the section where the console_lock owner can not sleep, because
1582 * there may be a waiter spinning (like a spinlock). Also it must be
1583 * ready to hand over the lock at the end of the section.
1585 static void console_lock_spinning_enable(void)
1587 raw_spin_lock(&console_owner_lock);
1588 console_owner = current;
1589 raw_spin_unlock(&console_owner_lock);
1591 /* The waiter may spin on us after setting console_owner */
1592 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1596 * console_lock_spinning_disable_and_check - mark end of code where another
1597 * thread was able to busy wait and check if there is a waiter
1599 * This is called at the end of the section where spinning is allowed.
1600 * It has two functions. First, it is a signal that it is no longer
1601 * safe to start busy waiting for the lock. Second, it checks if
1602 * there is a busy waiter and passes the lock rights to her.
1604 * Important: Callers lose the lock if there was a busy waiter.
1605 * They must not touch items synchronized by console_lock
1608 * Return: 1 if the lock rights were passed, 0 otherwise.
1610 static int console_lock_spinning_disable_and_check(void)
1614 raw_spin_lock(&console_owner_lock);
1615 waiter = READ_ONCE(console_waiter);
1616 console_owner = NULL;
1617 raw_spin_unlock(&console_owner_lock);
1620 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1624 /* The waiter is now free to continue */
1625 WRITE_ONCE(console_waiter, false);
1627 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1630 * Hand off console_lock to waiter. The waiter will perform
1631 * the up(). After this, the waiter is the console_lock owner.
1633 mutex_release(&console_lock_dep_map, 1, _THIS_IP_);
1638 * console_trylock_spinning - try to get console_lock by busy waiting
1640 * This allows to busy wait for the console_lock when the current
1641 * owner is running in specially marked sections. It means that
1642 * the current owner is running and cannot reschedule until it
1643 * is ready to lose the lock.
1645 * Return: 1 if we got the lock, 0 othrewise
1647 static int console_trylock_spinning(void)
1649 struct task_struct *owner = NULL;
1652 unsigned long flags;
1654 if (console_trylock())
1657 printk_safe_enter_irqsave(flags);
1659 raw_spin_lock(&console_owner_lock);
1660 owner = READ_ONCE(console_owner);
1661 waiter = READ_ONCE(console_waiter);
1662 if (!waiter && owner && owner != current) {
1663 WRITE_ONCE(console_waiter, true);
1666 raw_spin_unlock(&console_owner_lock);
1669 * If there is an active printk() writing to the
1670 * consoles, instead of having it write our data too,
1671 * see if we can offload that load from the active
1672 * printer, and do some printing ourselves.
1673 * Go into a spin only if there isn't already a waiter
1674 * spinning, and there is an active printer, and
1675 * that active printer isn't us (recursive printk?).
1678 printk_safe_exit_irqrestore(flags);
1682 /* We spin waiting for the owner to release us */
1683 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1684 /* Owner will clear console_waiter on hand off */
1685 while (READ_ONCE(console_waiter))
1687 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1689 printk_safe_exit_irqrestore(flags);
1691 * The owner passed the console lock to us.
1692 * Since we did not spin on console lock, annotate
1693 * this as a trylock. Otherwise lockdep will
1696 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1702 * Call the console drivers, asking them to write out
1703 * log_buf[start] to log_buf[end - 1].
1704 * The console_lock must be held.
1706 static void call_console_drivers(const char *ext_text, size_t ext_len,
1707 const char *text, size_t len)
1709 struct console *con;
1711 trace_console_rcuidle(text, len);
1713 if (!console_drivers)
1716 for_each_console(con) {
1717 if (exclusive_console && con != exclusive_console)
1719 if (!(con->flags & CON_ENABLED))
1723 if (!cpu_online(smp_processor_id()) &&
1724 !(con->flags & CON_ANYTIME))
1726 if (con->flags & CON_EXTENDED)
1727 con->write(con, ext_text, ext_len);
1729 con->write(con, text, len);
1733 int printk_delay_msec __read_mostly;
1735 static inline void printk_delay(void)
1737 if (unlikely(printk_delay_msec)) {
1738 int m = printk_delay_msec;
1742 touch_nmi_watchdog();
1748 * Continuation lines are buffered, and not committed to the record buffer
1749 * until the line is complete, or a race forces it. The line fragments
1750 * though, are printed immediately to the consoles to ensure everything has
1751 * reached the console in case of a kernel crash.
1753 static struct cont {
1754 char buf[LOG_LINE_MAX];
1755 size_t len; /* length == 0 means unused buffer */
1756 struct task_struct *owner; /* task of first print*/
1757 u64 ts_nsec; /* time of first print */
1758 u8 level; /* log level of first message */
1759 u8 facility; /* log facility of first message */
1760 enum log_flags flags; /* prefix, newline flags */
1763 static void cont_flush(void)
1768 log_store(cont.facility, cont.level, cont.flags, cont.ts_nsec,
1769 NULL, 0, cont.buf, cont.len);
1773 static bool cont_add(int facility, int level, enum log_flags flags, const char *text, size_t len)
1775 /* If the line gets too long, split it up in separate records. */
1776 if (cont.len + len > sizeof(cont.buf)) {
1782 cont.facility = facility;
1784 cont.owner = current;
1785 cont.ts_nsec = local_clock();
1789 memcpy(cont.buf + cont.len, text, len);
1792 // The original flags come from the first line,
1793 // but later continuations can add a newline.
1794 if (flags & LOG_NEWLINE) {
1795 cont.flags |= LOG_NEWLINE;
1802 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)
1805 * If an earlier line was buffered, and we're a continuation
1806 * write from the same process, try to add it to the buffer.
1809 if (cont.owner == current && (lflags & LOG_CONT)) {
1810 if (cont_add(facility, level, lflags, text, text_len))
1813 /* Otherwise, make sure it's flushed */
1817 /* Skip empty continuation lines that couldn't be added - they just flush */
1818 if (!text_len && (lflags & LOG_CONT))
1821 /* If it doesn't end in a newline, try to buffer the current line */
1822 if (!(lflags & LOG_NEWLINE)) {
1823 if (cont_add(facility, level, lflags, text, text_len))
1827 /* Store it in the record log */
1828 return log_store(facility, level, lflags, 0, dict, dictlen, text, text_len);
1831 /* Must be called under logbuf_lock. */
1832 int vprintk_store(int facility, int level,
1833 const char *dict, size_t dictlen,
1834 const char *fmt, va_list args)
1836 static char textbuf[LOG_LINE_MAX];
1837 char *text = textbuf;
1839 enum log_flags lflags = 0;
1842 * The printf needs to come first; we need the syslog
1843 * prefix which might be passed-in as a parameter.
1845 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1847 /* mark and strip a trailing newline */
1848 if (text_len && text[text_len-1] == '\n') {
1850 lflags |= LOG_NEWLINE;
1853 /* strip kernel syslog prefix and extract log level or control flags */
1854 if (facility == 0) {
1857 while ((kern_level = printk_get_level(text)) != 0) {
1858 switch (kern_level) {
1860 if (level == LOGLEVEL_DEFAULT)
1861 level = kern_level - '0';
1863 case 'd': /* KERN_DEFAULT */
1864 lflags |= LOG_PREFIX;
1866 case 'c': /* KERN_CONT */
1875 if (level == LOGLEVEL_DEFAULT)
1876 level = default_message_loglevel;
1879 lflags |= LOG_PREFIX|LOG_NEWLINE;
1881 return log_output(facility, level, lflags,
1882 dict, dictlen, text, text_len);
1885 asmlinkage int vprintk_emit(int facility, int level,
1886 const char *dict, size_t dictlen,
1887 const char *fmt, va_list args)
1890 bool in_sched = false, pending_output;
1891 unsigned long flags;
1894 if (level == LOGLEVEL_SCHED) {
1895 level = LOGLEVEL_DEFAULT;
1899 boot_delay_msec(level);
1902 /* This stops the holder of console_sem just where we want him */
1903 logbuf_lock_irqsave(flags);
1904 curr_log_seq = log_next_seq;
1905 printed_len = vprintk_store(facility, level, dict, dictlen, fmt, args);
1906 pending_output = (curr_log_seq != log_next_seq);
1907 logbuf_unlock_irqrestore(flags);
1909 /* If called from the scheduler, we can not call up(). */
1910 if (!in_sched && pending_output) {
1912 * Disable preemption to avoid being preempted while holding
1913 * console_sem which would prevent anyone from printing to
1918 * Try to acquire and then immediately release the console
1919 * semaphore. The release will print out buffers and wake up
1920 * /dev/kmsg and syslog() users.
1922 if (console_trylock_spinning())
1931 EXPORT_SYMBOL(vprintk_emit);
1933 asmlinkage int vprintk(const char *fmt, va_list args)
1935 return vprintk_func(fmt, args);
1937 EXPORT_SYMBOL(vprintk);
1939 asmlinkage int printk_emit(int facility, int level,
1940 const char *dict, size_t dictlen,
1941 const char *fmt, ...)
1946 va_start(args, fmt);
1947 r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1952 EXPORT_SYMBOL(printk_emit);
1954 int vprintk_default(const char *fmt, va_list args)
1958 #ifdef CONFIG_KGDB_KDB
1959 /* Allow to pass printk() to kdb but avoid a recursion. */
1960 if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
1961 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
1965 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1969 EXPORT_SYMBOL_GPL(vprintk_default);
1972 * printk - print a kernel message
1973 * @fmt: format string
1975 * This is printk(). It can be called from any context. We want it to work.
1977 * We try to grab the console_lock. If we succeed, it's easy - we log the
1978 * output and call the console drivers. If we fail to get the semaphore, we
1979 * place the output into the log buffer and return. The current holder of
1980 * the console_sem will notice the new output in console_unlock(); and will
1981 * send it to the consoles before releasing the lock.
1983 * One effect of this deferred printing is that code which calls printk() and
1984 * then changes console_loglevel may break. This is because console_loglevel
1985 * is inspected when the actual printing occurs.
1990 * See the vsnprintf() documentation for format string extensions over C99.
1992 asmlinkage __visible int printk(const char *fmt, ...)
1997 va_start(args, fmt);
1998 r = vprintk_func(fmt, args);
2003 EXPORT_SYMBOL(printk);
2005 #else /* CONFIG_PRINTK */
2007 #define LOG_LINE_MAX 0
2008 #define PREFIX_MAX 0
2010 static u64 syslog_seq;
2011 static u32 syslog_idx;
2012 static u64 console_seq;
2013 static u32 console_idx;
2014 static u64 exclusive_console_stop_seq;
2015 static u64 log_first_seq;
2016 static u32 log_first_idx;
2017 static u64 log_next_seq;
2018 static char *log_text(const struct printk_log *msg) { return NULL; }
2019 static char *log_dict(const struct printk_log *msg) { return NULL; }
2020 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2021 static u32 log_next(u32 idx) { return 0; }
2022 static ssize_t msg_print_ext_header(char *buf, size_t size,
2023 struct printk_log *msg,
2024 u64 seq) { return 0; }
2025 static ssize_t msg_print_ext_body(char *buf, size_t size,
2026 char *dict, size_t dict_len,
2027 char *text, size_t text_len) { return 0; }
2028 static void console_lock_spinning_enable(void) { }
2029 static int console_lock_spinning_disable_and_check(void) { return 0; }
2030 static void call_console_drivers(const char *ext_text, size_t ext_len,
2031 const char *text, size_t len) {}
2032 static size_t msg_print_text(const struct printk_log *msg,
2033 bool syslog, char *buf, size_t size) { return 0; }
2034 static bool suppress_message_printing(int level) { return false; }
2036 #endif /* CONFIG_PRINTK */
2038 #ifdef CONFIG_EARLY_PRINTK
2039 struct console *early_console;
2041 asmlinkage __visible void early_printk(const char *fmt, ...)
2051 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2054 early_console->write(early_console, buf, n);
2058 static int __add_preferred_console(char *name, int idx, char *options,
2061 struct console_cmdline *c;
2065 * See if this tty is not yet registered, and
2066 * if we have a slot free.
2068 for (i = 0, c = console_cmdline;
2069 i < MAX_CMDLINECONSOLES && c->name[0];
2071 if (strcmp(c->name, name) == 0 && c->index == idx) {
2073 preferred_console = i;
2077 if (i == MAX_CMDLINECONSOLES)
2080 preferred_console = i;
2081 strlcpy(c->name, name, sizeof(c->name));
2082 c->options = options;
2083 braille_set_options(c, brl_options);
2089 static int __init console_msg_format_setup(char *str)
2091 if (!strcmp(str, "syslog"))
2092 console_msg_format = MSG_FORMAT_SYSLOG;
2093 if (!strcmp(str, "default"))
2094 console_msg_format = MSG_FORMAT_DEFAULT;
2097 __setup("console_msg_format=", console_msg_format_setup);
2100 * Set up a console. Called via do_early_param() in init/main.c
2101 * for each "console=" parameter in the boot command line.
2103 static int __init console_setup(char *str)
2105 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2106 char *s, *options, *brl_options = NULL;
2109 if (_braille_console_setup(&str, &brl_options))
2113 * Decode str into name, index, options.
2115 if (str[0] >= '0' && str[0] <= '9') {
2116 strcpy(buf, "ttyS");
2117 strncpy(buf + 4, str, sizeof(buf) - 5);
2119 strncpy(buf, str, sizeof(buf) - 1);
2121 buf[sizeof(buf) - 1] = 0;
2122 options = strchr(str, ',');
2126 if (!strcmp(str, "ttya"))
2127 strcpy(buf, "ttyS0");
2128 if (!strcmp(str, "ttyb"))
2129 strcpy(buf, "ttyS1");
2131 for (s = buf; *s; s++)
2132 if (isdigit(*s) || *s == ',')
2134 idx = simple_strtoul(s, NULL, 10);
2137 __add_preferred_console(buf, idx, options, brl_options);
2138 console_set_on_cmdline = 1;
2141 __setup("console=", console_setup);
2144 * add_preferred_console - add a device to the list of preferred consoles.
2145 * @name: device name
2146 * @idx: device index
2147 * @options: options for this console
2149 * The last preferred console added will be used for kernel messages
2150 * and stdin/out/err for init. Normally this is used by console_setup
2151 * above to handle user-supplied console arguments; however it can also
2152 * be used by arch-specific code either to override the user or more
2153 * commonly to provide a default console (ie from PROM variables) when
2154 * the user has not supplied one.
2156 int add_preferred_console(char *name, int idx, char *options)
2158 return __add_preferred_console(name, idx, options, NULL);
2161 bool console_suspend_enabled = true;
2162 EXPORT_SYMBOL(console_suspend_enabled);
2164 static int __init console_suspend_disable(char *str)
2166 console_suspend_enabled = false;
2169 __setup("no_console_suspend", console_suspend_disable);
2170 module_param_named(console_suspend, console_suspend_enabled,
2171 bool, S_IRUGO | S_IWUSR);
2172 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2173 " and hibernate operations");
2176 * suspend_console - suspend the console subsystem
2178 * This disables printk() while we go into suspend states
2180 void suspend_console(void)
2182 if (!console_suspend_enabled)
2184 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2186 console_suspended = 1;
2190 void resume_console(void)
2192 if (!console_suspend_enabled)
2195 console_suspended = 0;
2200 * console_cpu_notify - print deferred console messages after CPU hotplug
2203 * If printk() is called from a CPU that is not online yet, the messages
2204 * will be printed on the console only if there are CON_ANYTIME consoles.
2205 * This function is called when a new CPU comes online (or fails to come
2206 * up) or goes offline.
2208 static int console_cpu_notify(unsigned int cpu)
2210 if (!cpuhp_tasks_frozen) {
2211 /* If trylock fails, someone else is doing the printing */
2212 if (console_trylock())
2219 * console_lock - lock the console system for exclusive use.
2221 * Acquires a lock which guarantees that the caller has
2222 * exclusive access to the console system and the console_drivers list.
2224 * Can sleep, returns nothing.
2226 void console_lock(void)
2231 if (console_suspended)
2234 console_may_schedule = 1;
2236 EXPORT_SYMBOL(console_lock);
2239 * console_trylock - try to lock the console system for exclusive use.
2241 * Try to acquire a lock which guarantees that the caller has exclusive
2242 * access to the console system and the console_drivers list.
2244 * returns 1 on success, and 0 on failure to acquire the lock.
2246 int console_trylock(void)
2248 if (down_trylock_console_sem())
2250 if (console_suspended) {
2255 console_may_schedule = 0;
2258 EXPORT_SYMBOL(console_trylock);
2260 int is_console_locked(void)
2262 return console_locked;
2264 EXPORT_SYMBOL(is_console_locked);
2267 * Check if we have any console that is capable of printing while cpu is
2268 * booting or shutting down. Requires console_sem.
2270 static int have_callable_console(void)
2272 struct console *con;
2274 for_each_console(con)
2275 if ((con->flags & CON_ENABLED) &&
2276 (con->flags & CON_ANYTIME))
2283 * Can we actually use the console at this time on this cpu?
2285 * Console drivers may assume that per-cpu resources have been allocated. So
2286 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2287 * call them until this CPU is officially up.
2289 static inline int can_use_console(void)
2291 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2295 * console_unlock - unlock the console system
2297 * Releases the console_lock which the caller holds on the console system
2298 * and the console driver list.
2300 * While the console_lock was held, console output may have been buffered
2301 * by printk(). If this is the case, console_unlock(); emits
2302 * the output prior to releasing the lock.
2304 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2306 * console_unlock(); may be called from any context.
2308 void console_unlock(void)
2310 static char ext_text[CONSOLE_EXT_LOG_MAX];
2311 static char text[LOG_LINE_MAX + PREFIX_MAX];
2312 unsigned long flags;
2313 bool do_cond_resched, retry;
2315 if (console_suspended) {
2321 * Console drivers are called with interrupts disabled, so
2322 * @console_may_schedule should be cleared before; however, we may
2323 * end up dumping a lot of lines, for example, if called from
2324 * console registration path, and should invoke cond_resched()
2325 * between lines if allowable. Not doing so can cause a very long
2326 * scheduling stall on a slow console leading to RCU stall and
2327 * softlockup warnings which exacerbate the issue with more
2328 * messages practically incapacitating the system.
2330 * console_trylock() is not able to detect the preemptive
2331 * context reliably. Therefore the value must be stored before
2332 * and cleared after the the "again" goto label.
2334 do_cond_resched = console_may_schedule;
2336 console_may_schedule = 0;
2339 * We released the console_sem lock, so we need to recheck if
2340 * cpu is online and (if not) is there at least one CON_ANYTIME
2343 if (!can_use_console()) {
2350 struct printk_log *msg;
2354 printk_safe_enter_irqsave(flags);
2355 raw_spin_lock(&logbuf_lock);
2356 if (console_seq < log_first_seq) {
2358 "** %llu printk messages dropped **\n",
2359 log_first_seq - console_seq);
2361 /* messages are gone, move to first one */
2362 console_seq = log_first_seq;
2363 console_idx = log_first_idx;
2368 if (console_seq == log_next_seq)
2371 msg = log_from_idx(console_idx);
2372 if (suppress_message_printing(msg->level)) {
2374 * Skip record we have buffered and already printed
2375 * directly to the console when we received it, and
2376 * record that has level above the console loglevel.
2378 console_idx = log_next(console_idx);
2383 /* Output to all consoles once old messages replayed. */
2384 if (unlikely(exclusive_console &&
2385 console_seq >= exclusive_console_stop_seq)) {
2386 exclusive_console = NULL;
2389 len += msg_print_text(msg,
2390 console_msg_format & MSG_FORMAT_SYSLOG,
2392 sizeof(text) - len);
2393 if (nr_ext_console_drivers) {
2394 ext_len = msg_print_ext_header(ext_text,
2397 ext_len += msg_print_ext_body(ext_text + ext_len,
2398 sizeof(ext_text) - ext_len,
2399 log_dict(msg), msg->dict_len,
2400 log_text(msg), msg->text_len);
2402 console_idx = log_next(console_idx);
2404 raw_spin_unlock(&logbuf_lock);
2407 * While actively printing out messages, if another printk()
2408 * were to occur on another CPU, it may wait for this one to
2409 * finish. This task can not be preempted if there is a
2410 * waiter waiting to take over.
2412 console_lock_spinning_enable();
2414 stop_critical_timings(); /* don't trace print latency */
2415 call_console_drivers(ext_text, ext_len, text, len);
2416 start_critical_timings();
2418 if (console_lock_spinning_disable_and_check()) {
2419 printk_safe_exit_irqrestore(flags);
2423 printk_safe_exit_irqrestore(flags);
2425 if (do_cond_resched)
2431 raw_spin_unlock(&logbuf_lock);
2436 * Someone could have filled up the buffer again, so re-check if there's
2437 * something to flush. In case we cannot trylock the console_sem again,
2438 * there's a new owner and the console_unlock() from them will do the
2439 * flush, no worries.
2441 raw_spin_lock(&logbuf_lock);
2442 retry = console_seq != log_next_seq;
2443 raw_spin_unlock(&logbuf_lock);
2444 printk_safe_exit_irqrestore(flags);
2446 if (retry && console_trylock())
2449 EXPORT_SYMBOL(console_unlock);
2452 * console_conditional_schedule - yield the CPU if required
2454 * If the console code is currently allowed to sleep, and
2455 * if this CPU should yield the CPU to another task, do
2458 * Must be called within console_lock();.
2460 void __sched console_conditional_schedule(void)
2462 if (console_may_schedule)
2465 EXPORT_SYMBOL(console_conditional_schedule);
2467 void console_unblank(void)
2472 * console_unblank can no longer be called in interrupt context unless
2473 * oops_in_progress is set to 1..
2475 if (oops_in_progress) {
2476 if (down_trylock_console_sem() != 0)
2482 console_may_schedule = 0;
2484 if ((c->flags & CON_ENABLED) && c->unblank)
2490 * console_flush_on_panic - flush console content on panic
2492 * Immediately output all pending messages no matter what.
2494 void console_flush_on_panic(void)
2497 * If someone else is holding the console lock, trylock will fail
2498 * and may_schedule may be set. Ignore and proceed to unlock so
2499 * that messages are flushed out. As this can be called from any
2500 * context and we don't want to get preempted while flushing,
2501 * ensure may_schedule is cleared.
2504 console_may_schedule = 0;
2509 * Return the console tty driver structure and its associated index
2511 struct tty_driver *console_device(int *index)
2514 struct tty_driver *driver = NULL;
2517 for_each_console(c) {
2520 driver = c->device(c, index);
2529 * Prevent further output on the passed console device so that (for example)
2530 * serial drivers can disable console output before suspending a port, and can
2531 * re-enable output afterwards.
2533 void console_stop(struct console *console)
2536 console->flags &= ~CON_ENABLED;
2539 EXPORT_SYMBOL(console_stop);
2541 void console_start(struct console *console)
2544 console->flags |= CON_ENABLED;
2547 EXPORT_SYMBOL(console_start);
2549 static int __read_mostly keep_bootcon;
2551 static int __init keep_bootcon_setup(char *str)
2554 pr_info("debug: skip boot console de-registration.\n");
2559 early_param("keep_bootcon", keep_bootcon_setup);
2562 * The console driver calls this routine during kernel initialization
2563 * to register the console printing procedure with printk() and to
2564 * print any messages that were printed by the kernel before the
2565 * console driver was initialized.
2567 * This can happen pretty early during the boot process (because of
2568 * early_printk) - sometimes before setup_arch() completes - be careful
2569 * of what kernel features are used - they may not be initialised yet.
2571 * There are two types of consoles - bootconsoles (early_printk) and
2572 * "real" consoles (everything which is not a bootconsole) which are
2573 * handled differently.
2574 * - Any number of bootconsoles can be registered at any time.
2575 * - As soon as a "real" console is registered, all bootconsoles
2576 * will be unregistered automatically.
2577 * - Once a "real" console is registered, any attempt to register a
2578 * bootconsoles will be rejected
2580 void register_console(struct console *newcon)
2583 unsigned long flags;
2584 struct console *bcon = NULL;
2585 struct console_cmdline *c;
2586 static bool has_preferred;
2588 if (console_drivers)
2589 for_each_console(bcon)
2590 if (WARN(bcon == newcon,
2591 "console '%s%d' already registered\n",
2592 bcon->name, bcon->index))
2596 * before we register a new CON_BOOT console, make sure we don't
2597 * already have a valid console
2599 if (console_drivers && newcon->flags & CON_BOOT) {
2600 /* find the last or real console */
2601 for_each_console(bcon) {
2602 if (!(bcon->flags & CON_BOOT)) {
2603 pr_info("Too late to register bootconsole %s%d\n",
2604 newcon->name, newcon->index);
2610 if (console_drivers && console_drivers->flags & CON_BOOT)
2611 bcon = console_drivers;
2613 if (!has_preferred || bcon || !console_drivers)
2614 has_preferred = preferred_console >= 0;
2617 * See if we want to use this console driver. If we
2618 * didn't select a console we take the first one
2619 * that registers here.
2621 if (!has_preferred) {
2622 if (newcon->index < 0)
2624 if (newcon->setup == NULL ||
2625 newcon->setup(newcon, NULL) == 0) {
2626 newcon->flags |= CON_ENABLED;
2627 if (newcon->device) {
2628 newcon->flags |= CON_CONSDEV;
2629 has_preferred = true;
2635 * See if this console matches one we selected on
2638 for (i = 0, c = console_cmdline;
2639 i < MAX_CMDLINECONSOLES && c->name[0];
2641 if (!newcon->match ||
2642 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2643 /* default matching */
2644 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2645 if (strcmp(c->name, newcon->name) != 0)
2647 if (newcon->index >= 0 &&
2648 newcon->index != c->index)
2650 if (newcon->index < 0)
2651 newcon->index = c->index;
2653 if (_braille_register_console(newcon, c))
2656 if (newcon->setup &&
2657 newcon->setup(newcon, c->options) != 0)
2661 newcon->flags |= CON_ENABLED;
2662 if (i == preferred_console) {
2663 newcon->flags |= CON_CONSDEV;
2664 has_preferred = true;
2669 if (!(newcon->flags & CON_ENABLED))
2673 * If we have a bootconsole, and are switching to a real console,
2674 * don't print everything out again, since when the boot console, and
2675 * the real console are the same physical device, it's annoying to
2676 * see the beginning boot messages twice
2678 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2679 newcon->flags &= ~CON_PRINTBUFFER;
2682 * Put this console in the list - keep the
2683 * preferred driver at the head of the list.
2686 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2687 newcon->next = console_drivers;
2688 console_drivers = newcon;
2690 newcon->next->flags &= ~CON_CONSDEV;
2692 newcon->next = console_drivers->next;
2693 console_drivers->next = newcon;
2696 if (newcon->flags & CON_EXTENDED)
2697 nr_ext_console_drivers++;
2699 if (newcon->flags & CON_PRINTBUFFER) {
2701 * console_unlock(); will print out the buffered messages
2704 logbuf_lock_irqsave(flags);
2705 console_seq = syslog_seq;
2706 console_idx = syslog_idx;
2708 * We're about to replay the log buffer. Only do this to the
2709 * just-registered console to avoid excessive message spam to
2710 * the already-registered consoles.
2712 * Set exclusive_console with disabled interrupts to reduce
2713 * race window with eventual console_flush_on_panic() that
2714 * ignores console_lock.
2716 exclusive_console = newcon;
2717 exclusive_console_stop_seq = console_seq;
2718 logbuf_unlock_irqrestore(flags);
2721 console_sysfs_notify();
2724 * By unregistering the bootconsoles after we enable the real console
2725 * we get the "console xxx enabled" message on all the consoles -
2726 * boot consoles, real consoles, etc - this is to ensure that end
2727 * users know there might be something in the kernel's log buffer that
2728 * went to the bootconsole (that they do not see on the real console)
2730 pr_info("%sconsole [%s%d] enabled\n",
2731 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2732 newcon->name, newcon->index);
2734 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2736 /* We need to iterate through all boot consoles, to make
2737 * sure we print everything out, before we unregister them.
2739 for_each_console(bcon)
2740 if (bcon->flags & CON_BOOT)
2741 unregister_console(bcon);
2744 EXPORT_SYMBOL(register_console);
2746 int unregister_console(struct console *console)
2748 struct console *a, *b;
2751 pr_info("%sconsole [%s%d] disabled\n",
2752 (console->flags & CON_BOOT) ? "boot" : "" ,
2753 console->name, console->index);
2755 res = _braille_unregister_console(console);
2761 if (console_drivers == console) {
2762 console_drivers=console->next;
2764 } else if (console_drivers) {
2765 for (a=console_drivers->next, b=console_drivers ;
2766 a; b=a, a=b->next) {
2775 if (!res && (console->flags & CON_EXTENDED))
2776 nr_ext_console_drivers--;
2779 * If this isn't the last console and it has CON_CONSDEV set, we
2780 * need to set it on the next preferred console.
2782 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2783 console_drivers->flags |= CON_CONSDEV;
2785 console->flags &= ~CON_ENABLED;
2787 console_sysfs_notify();
2790 EXPORT_SYMBOL(unregister_console);
2793 * Initialize the console device. This is called *early*, so
2794 * we can't necessarily depend on lots of kernel help here.
2795 * Just do some early initializations, and do the complex setup
2798 void __init console_init(void)
2802 initcall_entry_t *ce;
2804 /* Setup the default TTY line discipline. */
2808 * set up the console device so that later boot sequences can
2809 * inform about problems etc..
2811 ce = __con_initcall_start;
2812 trace_initcall_level("console");
2813 while (ce < __con_initcall_end) {
2814 call = initcall_from_entry(ce);
2815 trace_initcall_start(call);
2817 trace_initcall_finish(call, ret);
2823 * Some boot consoles access data that is in the init section and which will
2824 * be discarded after the initcalls have been run. To make sure that no code
2825 * will access this data, unregister the boot consoles in a late initcall.
2827 * If for some reason, such as deferred probe or the driver being a loadable
2828 * module, the real console hasn't registered yet at this point, there will
2829 * be a brief interval in which no messages are logged to the console, which
2830 * makes it difficult to diagnose problems that occur during this time.
2832 * To mitigate this problem somewhat, only unregister consoles whose memory
2833 * intersects with the init section. Note that all other boot consoles will
2834 * get unregistred when the real preferred console is registered.
2836 static int __init printk_late_init(void)
2838 struct console *con;
2841 for_each_console(con) {
2842 if (!(con->flags & CON_BOOT))
2845 /* Check addresses that might be used for enabled consoles. */
2846 if (init_section_intersects(con, sizeof(*con)) ||
2847 init_section_contains(con->write, 0) ||
2848 init_section_contains(con->read, 0) ||
2849 init_section_contains(con->device, 0) ||
2850 init_section_contains(con->unblank, 0) ||
2851 init_section_contains(con->data, 0)) {
2853 * Please, consider moving the reported consoles out
2854 * of the init section.
2856 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2857 con->name, con->index);
2858 unregister_console(con);
2861 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2862 console_cpu_notify);
2864 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2865 console_cpu_notify, NULL);
2869 late_initcall(printk_late_init);
2871 #if defined CONFIG_PRINTK
2873 * Delayed printk version, for scheduler-internal messages:
2875 #define PRINTK_PENDING_WAKEUP 0x01
2876 #define PRINTK_PENDING_OUTPUT 0x02
2878 static DEFINE_PER_CPU(int, printk_pending);
2880 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2882 int pending = __this_cpu_xchg(printk_pending, 0);
2884 if (pending & PRINTK_PENDING_OUTPUT) {
2885 /* If trylock fails, someone else is doing the printing */
2886 if (console_trylock())
2890 if (pending & PRINTK_PENDING_WAKEUP)
2891 wake_up_interruptible(&log_wait);
2894 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2895 .func = wake_up_klogd_work_func,
2896 .flags = IRQ_WORK_LAZY,
2899 void wake_up_klogd(void)
2902 if (waitqueue_active(&log_wait)) {
2903 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2904 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2909 void defer_console_output(void)
2912 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2913 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2917 int vprintk_deferred(const char *fmt, va_list args)
2921 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2922 defer_console_output();
2927 int printk_deferred(const char *fmt, ...)
2932 va_start(args, fmt);
2933 r = vprintk_deferred(fmt, args);
2940 * printk rate limiting, lifted from the networking subsystem.
2942 * This enforces a rate limit: not more than 10 kernel messages
2943 * every 5s to make a denial-of-service attack impossible.
2945 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2947 int __printk_ratelimit(const char *func)
2949 return ___ratelimit(&printk_ratelimit_state, func);
2951 EXPORT_SYMBOL(__printk_ratelimit);
2954 * printk_timed_ratelimit - caller-controlled printk ratelimiting
2955 * @caller_jiffies: pointer to caller's state
2956 * @interval_msecs: minimum interval between prints
2958 * printk_timed_ratelimit() returns true if more than @interval_msecs
2959 * milliseconds have elapsed since the last time printk_timed_ratelimit()
2962 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
2963 unsigned int interval_msecs)
2965 unsigned long elapsed = jiffies - *caller_jiffies;
2967 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
2970 *caller_jiffies = jiffies;
2973 EXPORT_SYMBOL(printk_timed_ratelimit);
2975 static DEFINE_SPINLOCK(dump_list_lock);
2976 static LIST_HEAD(dump_list);
2979 * kmsg_dump_register - register a kernel log dumper.
2980 * @dumper: pointer to the kmsg_dumper structure
2982 * Adds a kernel log dumper to the system. The dump callback in the
2983 * structure will be called when the kernel oopses or panics and must be
2984 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
2986 int kmsg_dump_register(struct kmsg_dumper *dumper)
2988 unsigned long flags;
2991 /* The dump callback needs to be set */
2995 spin_lock_irqsave(&dump_list_lock, flags);
2996 /* Don't allow registering multiple times */
2997 if (!dumper->registered) {
2998 dumper->registered = 1;
2999 list_add_tail_rcu(&dumper->list, &dump_list);
3002 spin_unlock_irqrestore(&dump_list_lock, flags);
3006 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3009 * kmsg_dump_unregister - unregister a kmsg dumper.
3010 * @dumper: pointer to the kmsg_dumper structure
3012 * Removes a dump device from the system. Returns zero on success and
3013 * %-EINVAL otherwise.
3015 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3017 unsigned long flags;
3020 spin_lock_irqsave(&dump_list_lock, flags);
3021 if (dumper->registered) {
3022 dumper->registered = 0;
3023 list_del_rcu(&dumper->list);
3026 spin_unlock_irqrestore(&dump_list_lock, flags);
3031 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3033 static bool always_kmsg_dump;
3034 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3037 * kmsg_dump - dump kernel log to kernel message dumpers.
3038 * @reason: the reason (oops, panic etc) for dumping
3040 * Call each of the registered dumper's dump() callback, which can
3041 * retrieve the kmsg records with kmsg_dump_get_line() or
3042 * kmsg_dump_get_buffer().
3044 void kmsg_dump(enum kmsg_dump_reason reason)
3046 struct kmsg_dumper *dumper;
3047 unsigned long flags;
3049 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3053 list_for_each_entry_rcu(dumper, &dump_list, list) {
3054 if (dumper->max_reason && reason > dumper->max_reason)
3057 /* initialize iterator with data about the stored records */
3058 dumper->active = true;
3060 logbuf_lock_irqsave(flags);
3061 dumper->cur_seq = clear_seq;
3062 dumper->cur_idx = clear_idx;
3063 dumper->next_seq = log_next_seq;
3064 dumper->next_idx = log_next_idx;
3065 logbuf_unlock_irqrestore(flags);
3067 /* invoke dumper which will iterate over records */
3068 dumper->dump(dumper, reason);
3070 /* reset iterator */
3071 dumper->active = false;
3077 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3078 * @dumper: registered kmsg dumper
3079 * @syslog: include the "<4>" prefixes
3080 * @line: buffer to copy the line to
3081 * @size: maximum size of the buffer
3082 * @len: length of line placed into buffer
3084 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3085 * record, and copy one record into the provided buffer.
3087 * Consecutive calls will return the next available record moving
3088 * towards the end of the buffer with the youngest messages.
3090 * A return value of FALSE indicates that there are no more records to
3093 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3095 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3096 char *line, size_t size, size_t *len)
3098 struct printk_log *msg;
3102 if (!dumper->active)
3105 if (dumper->cur_seq < log_first_seq) {
3106 /* messages are gone, move to first available one */
3107 dumper->cur_seq = log_first_seq;
3108 dumper->cur_idx = log_first_idx;
3112 if (dumper->cur_seq >= log_next_seq)
3115 msg = log_from_idx(dumper->cur_idx);
3116 l = msg_print_text(msg, syslog, line, size);
3118 dumper->cur_idx = log_next(dumper->cur_idx);
3128 * kmsg_dump_get_line - retrieve one kmsg log line
3129 * @dumper: registered kmsg dumper
3130 * @syslog: include the "<4>" prefixes
3131 * @line: buffer to copy the line to
3132 * @size: maximum size of the buffer
3133 * @len: length of line placed into buffer
3135 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3136 * record, and copy one record into the provided buffer.
3138 * Consecutive calls will return the next available record moving
3139 * towards the end of the buffer with the youngest messages.
3141 * A return value of FALSE indicates that there are no more records to
3144 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3145 char *line, size_t size, size_t *len)
3147 unsigned long flags;
3150 logbuf_lock_irqsave(flags);
3151 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3152 logbuf_unlock_irqrestore(flags);
3156 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3159 * kmsg_dump_get_buffer - copy kmsg log lines
3160 * @dumper: registered kmsg dumper
3161 * @syslog: include the "<4>" prefixes
3162 * @buf: buffer to copy the line to
3163 * @size: maximum size of the buffer
3164 * @len: length of line placed into buffer
3166 * Start at the end of the kmsg buffer and fill the provided buffer
3167 * with as many of the the *youngest* kmsg records that fit into it.
3168 * If the buffer is large enough, all available kmsg records will be
3169 * copied with a single call.
3171 * Consecutive calls will fill the buffer with the next block of
3172 * available older records, not including the earlier retrieved ones.
3174 * A return value of FALSE indicates that there are no more records to
3177 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3178 char *buf, size_t size, size_t *len)
3180 unsigned long flags;
3188 if (!dumper->active)
3191 logbuf_lock_irqsave(flags);
3192 if (dumper->cur_seq < log_first_seq) {
3193 /* messages are gone, move to first available one */
3194 dumper->cur_seq = log_first_seq;
3195 dumper->cur_idx = log_first_idx;
3199 if (dumper->cur_seq >= dumper->next_seq) {
3200 logbuf_unlock_irqrestore(flags);
3204 /* calculate length of entire buffer */
3205 seq = dumper->cur_seq;
3206 idx = dumper->cur_idx;
3207 while (seq < dumper->next_seq) {
3208 struct printk_log *msg = log_from_idx(idx);
3210 l += msg_print_text(msg, true, NULL, 0);
3211 idx = log_next(idx);
3215 /* move first record forward until length fits into the buffer */
3216 seq = dumper->cur_seq;
3217 idx = dumper->cur_idx;
3218 while (l > size && seq < dumper->next_seq) {
3219 struct printk_log *msg = log_from_idx(idx);
3221 l -= msg_print_text(msg, true, NULL, 0);
3222 idx = log_next(idx);
3226 /* last message in next interation */
3231 while (seq < dumper->next_seq) {
3232 struct printk_log *msg = log_from_idx(idx);
3234 l += msg_print_text(msg, syslog, buf + l, size - l);
3235 idx = log_next(idx);
3239 dumper->next_seq = next_seq;
3240 dumper->next_idx = next_idx;
3242 logbuf_unlock_irqrestore(flags);
3248 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3251 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3252 * @dumper: registered kmsg dumper
3254 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3255 * kmsg_dump_get_buffer() can be called again and used multiple
3256 * times within the same dumper.dump() callback.
3258 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3260 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3262 dumper->cur_seq = clear_seq;
3263 dumper->cur_idx = clear_idx;
3264 dumper->next_seq = log_next_seq;
3265 dumper->next_idx = log_next_idx;
3269 * kmsg_dump_rewind - reset the interator
3270 * @dumper: registered kmsg dumper
3272 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3273 * kmsg_dump_get_buffer() can be called again and used multiple
3274 * times within the same dumper.dump() callback.
3276 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3278 unsigned long flags;
3280 logbuf_lock_irqsave(flags);
3281 kmsg_dump_rewind_nolock(dumper);
3282 logbuf_unlock_irqrestore(flags);
3284 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);