2 * linux/kernel/printk.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * Modified to make sys_syslog() more flexible: added commands to
7 * return the last 4k of kernel messages, regardless of whether
8 * they've been read or not. Added option to suppress kernel printk's
9 * to the console. Added hook for sending the console messages
10 * elsewhere, in preparation for a serial line console (someday).
12 * Modified for sysctl support, 1/8/97, Chris Horn.
13 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
14 * manfred@colorfullife.com
15 * Rewrote bits to get rid of console_lock
16 * 01Mar01 Andrew Morton
19 #include <linux/kernel.h>
21 #include <linux/tty.h>
22 #include <linux/tty_driver.h>
23 #include <linux/console.h>
24 #include <linux/init.h>
25 #include <linux/jiffies.h>
26 #include <linux/nmi.h>
27 #include <linux/module.h>
28 #include <linux/moduleparam.h>
29 #include <linux/delay.h>
30 #include <linux/smp.h>
31 #include <linux/security.h>
32 #include <linux/bootmem.h>
33 #include <linux/memblock.h>
34 #include <linux/syscalls.h>
35 #include <linux/crash_core.h>
36 #include <linux/kdb.h>
37 #include <linux/ratelimit.h>
38 #include <linux/kmsg_dump.h>
39 #include <linux/syslog.h>
40 #include <linux/cpu.h>
41 #include <linux/rculist.h>
42 #include <linux/poll.h>
43 #include <linux/irq_work.h>
44 #include <linux/ctype.h>
45 #include <linux/uio.h>
46 #include <linux/sched/clock.h>
47 #include <linux/sched/debug.h>
48 #include <linux/sched/task_stack.h>
50 #include <linux/uaccess.h>
51 #include <asm/sections.h>
53 #include <trace/events/initcall.h>
54 #define CREATE_TRACE_POINTS
55 #include <trace/events/printk.h>
57 #include "console_cmdline.h"
61 int console_printk[4] = {
62 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
63 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
64 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
65 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
68 atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
69 EXPORT_SYMBOL(ignore_console_lock_warning);
72 * Low level drivers may need that to know if they can schedule in
73 * their unblank() callback or not. So let's export it.
76 EXPORT_SYMBOL(oops_in_progress);
79 * console_sem protects the console_drivers list, and also
80 * provides serialisation for access to the entire console
83 static DEFINE_SEMAPHORE(console_sem);
84 struct console *console_drivers;
85 EXPORT_SYMBOL_GPL(console_drivers);
88 static struct lockdep_map console_lock_dep_map = {
89 .name = "console_lock"
93 enum devkmsg_log_bits {
94 __DEVKMSG_LOG_BIT_ON = 0,
95 __DEVKMSG_LOG_BIT_OFF,
96 __DEVKMSG_LOG_BIT_LOCK,
99 enum devkmsg_log_masks {
100 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
101 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
102 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
105 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
106 #define DEVKMSG_LOG_MASK_DEFAULT 0
108 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
110 static int __control_devkmsg(char *str)
115 if (!strncmp(str, "on", 2)) {
116 devkmsg_log = DEVKMSG_LOG_MASK_ON;
118 } else if (!strncmp(str, "off", 3)) {
119 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
121 } else if (!strncmp(str, "ratelimit", 9)) {
122 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
128 static int __init control_devkmsg(char *str)
130 if (__control_devkmsg(str) < 0)
134 * Set sysctl string accordingly:
136 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
137 strcpy(devkmsg_log_str, "on");
138 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
139 strcpy(devkmsg_log_str, "off");
140 /* else "ratelimit" which is set by default. */
143 * Sysctl cannot change it anymore. The kernel command line setting of
144 * this parameter is to force the setting to be permanent throughout the
145 * runtime of the system. This is a precation measure against userspace
146 * trying to be a smarta** and attempting to change it up on us.
148 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
152 __setup("printk.devkmsg=", control_devkmsg);
154 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
156 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
157 void __user *buffer, size_t *lenp, loff_t *ppos)
159 char old_str[DEVKMSG_STR_MAX_SIZE];
164 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
168 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
171 err = proc_dostring(table, write, buffer, lenp, ppos);
176 err = __control_devkmsg(devkmsg_log_str);
179 * Do not accept an unknown string OR a known string with
182 if (err < 0 || (err + 1 != *lenp)) {
184 /* ... and restore old setting. */
186 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
196 * Number of registered extended console drivers.
198 * If extended consoles are present, in-kernel cont reassembly is disabled
199 * and each fragment is stored as a separate log entry with proper
200 * continuation flag so that every emitted message has full metadata. This
201 * doesn't change the result for regular consoles or /proc/kmsg. For
202 * /dev/kmsg, as long as the reader concatenates messages according to
203 * consecutive continuation flags, the end result should be the same too.
205 static int nr_ext_console_drivers;
208 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
209 * macros instead of functions so that _RET_IP_ contains useful information.
211 #define down_console_sem() do { \
213 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
216 static int __down_trylock_console_sem(unsigned long ip)
222 * Here and in __up_console_sem() we need to be in safe mode,
223 * because spindump/WARN/etc from under console ->lock will
224 * deadlock in printk()->down_trylock_console_sem() otherwise.
226 printk_safe_enter_irqsave(flags);
227 lock_failed = down_trylock(&console_sem);
228 printk_safe_exit_irqrestore(flags);
232 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
235 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
237 static void __up_console_sem(unsigned long ip)
241 mutex_release(&console_lock_dep_map, 1, ip);
243 printk_safe_enter_irqsave(flags);
245 printk_safe_exit_irqrestore(flags);
247 #define up_console_sem() __up_console_sem(_RET_IP_)
250 * This is used for debugging the mess that is the VT code by
251 * keeping track if we have the console semaphore held. It's
252 * definitely not the perfect debug tool (we don't know if _WE_
253 * hold it and are racing, but it helps tracking those weird code
254 * paths in the console code where we end up in places I want
255 * locked without the console sempahore held).
257 static int console_locked, console_suspended;
260 * If exclusive_console is non-NULL then only this console is to be printed to.
262 static struct console *exclusive_console;
265 * Array of consoles built from command line options (console=)
268 #define MAX_CMDLINECONSOLES 8
270 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
272 static int preferred_console = -1;
273 int console_set_on_cmdline;
274 EXPORT_SYMBOL(console_set_on_cmdline);
276 /* Flag: console code may call schedule() */
277 static int console_may_schedule;
279 enum con_msg_format_flags {
280 MSG_FORMAT_DEFAULT = 0,
281 MSG_FORMAT_SYSLOG = (1 << 0),
284 static int console_msg_format = MSG_FORMAT_DEFAULT;
287 * The printk log buffer consists of a chain of concatenated variable
288 * length records. Every record starts with a record header, containing
289 * the overall length of the record.
291 * The heads to the first and last entry in the buffer, as well as the
292 * sequence numbers of these entries are maintained when messages are
295 * If the heads indicate available messages, the length in the header
296 * tells the start next message. A length == 0 for the next message
297 * indicates a wrap-around to the beginning of the buffer.
299 * Every record carries the monotonic timestamp in microseconds, as well as
300 * the standard userspace syslog level and syslog facility. The usual
301 * kernel messages use LOG_KERN; userspace-injected messages always carry
302 * a matching syslog facility, by default LOG_USER. The origin of every
303 * message can be reliably determined that way.
305 * The human readable log message directly follows the message header. The
306 * length of the message text is stored in the header, the stored message
309 * Optionally, a message can carry a dictionary of properties (key/value pairs),
310 * to provide userspace with a machine-readable message context.
312 * Examples for well-defined, commonly used property names are:
313 * DEVICE=b12:8 device identifier
317 * +sound:card0 subsystem:devname
318 * SUBSYSTEM=pci driver-core subsystem name
320 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
321 * follows directly after a '=' character. Every property is terminated by
322 * a '\0' character. The last property is not terminated.
324 * Example of a message structure:
325 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
326 * 0008 34 00 record is 52 bytes long
327 * 000a 0b 00 text is 11 bytes long
328 * 000c 1f 00 dictionary is 23 bytes long
329 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
330 * 0010 69 74 27 73 20 61 20 6c "it's a l"
332 * 001b 44 45 56 49 43 "DEVIC"
333 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
334 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
336 * 0032 00 00 00 padding to next message header
338 * The 'struct printk_log' buffer header must never be directly exported to
339 * userspace, it is a kernel-private implementation detail that might
340 * need to be changed in the future, when the requirements change.
342 * /dev/kmsg exports the structured data in the following line format:
343 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
345 * Users of the export format should ignore possible additional values
346 * separated by ',', and find the message after the ';' character.
348 * The optional key/value pairs are attached as continuation lines starting
349 * with a space character and terminated by a newline. All possible
350 * non-prinatable characters are escaped in the "\xff" notation.
354 LOG_NOCONS = 1, /* suppress print, do not print to console */
355 LOG_NEWLINE = 2, /* text ended with a newline */
356 LOG_PREFIX = 4, /* text started with a prefix */
357 LOG_CONT = 8, /* text is a fragment of a continuation line */
361 u64 ts_nsec; /* timestamp in nanoseconds */
362 u16 len; /* length of entire record */
363 u16 text_len; /* length of text buffer */
364 u16 dict_len; /* length of dictionary buffer */
365 u8 facility; /* syslog facility */
366 u8 flags:5; /* internal record flags */
367 u8 level:3; /* syslog level */
369 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
370 __packed __aligned(4)
375 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
376 * within the scheduler's rq lock. It must be released before calling
377 * console_unlock() or anything else that might wake up a process.
379 DEFINE_RAW_SPINLOCK(logbuf_lock);
382 * Helper macros to lock/unlock logbuf_lock and switch between
383 * printk-safe/unsafe modes.
385 #define logbuf_lock_irq() \
387 printk_safe_enter_irq(); \
388 raw_spin_lock(&logbuf_lock); \
391 #define logbuf_unlock_irq() \
393 raw_spin_unlock(&logbuf_lock); \
394 printk_safe_exit_irq(); \
397 #define logbuf_lock_irqsave(flags) \
399 printk_safe_enter_irqsave(flags); \
400 raw_spin_lock(&logbuf_lock); \
403 #define logbuf_unlock_irqrestore(flags) \
405 raw_spin_unlock(&logbuf_lock); \
406 printk_safe_exit_irqrestore(flags); \
410 DECLARE_WAIT_QUEUE_HEAD(log_wait);
411 /* the next printk record to read by syslog(READ) or /proc/kmsg */
412 static u64 syslog_seq;
413 static u32 syslog_idx;
414 static size_t syslog_partial;
416 /* index and sequence number of the first record stored in the buffer */
417 static u64 log_first_seq;
418 static u32 log_first_idx;
420 /* index and sequence number of the next record to store in the buffer */
421 static u64 log_next_seq;
422 static u32 log_next_idx;
424 /* the next printk record to write to the console */
425 static u64 console_seq;
426 static u32 console_idx;
428 /* the next printk record to read after the last 'clear' command */
429 static u64 clear_seq;
430 static u32 clear_idx;
432 #define PREFIX_MAX 32
433 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
435 #define LOG_LEVEL(v) ((v) & 0x07)
436 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
439 #define LOG_ALIGN __alignof__(struct printk_log)
440 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
441 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
442 static char *log_buf = __log_buf;
443 static u32 log_buf_len = __LOG_BUF_LEN;
445 /* Return log buffer address */
446 char *log_buf_addr_get(void)
451 /* Return log buffer size */
452 u32 log_buf_len_get(void)
457 /* human readable text of the record */
458 static char *log_text(const struct printk_log *msg)
460 return (char *)msg + sizeof(struct printk_log);
463 /* optional key/value pair dictionary attached to the record */
464 static char *log_dict(const struct printk_log *msg)
466 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
469 /* get record by index; idx must point to valid msg */
470 static struct printk_log *log_from_idx(u32 idx)
472 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
475 * A length == 0 record is the end of buffer marker. Wrap around and
476 * read the message at the start of the buffer.
479 return (struct printk_log *)log_buf;
483 /* get next record; idx must point to valid msg */
484 static u32 log_next(u32 idx)
486 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
488 /* length == 0 indicates the end of the buffer; wrap */
490 * A length == 0 record is the end of buffer marker. Wrap around and
491 * read the message at the start of the buffer as *this* one, and
492 * return the one after that.
495 msg = (struct printk_log *)log_buf;
498 return idx + msg->len;
502 * Check whether there is enough free space for the given message.
504 * The same values of first_idx and next_idx mean that the buffer
505 * is either empty or full.
507 * If the buffer is empty, we must respect the position of the indexes.
508 * They cannot be reset to the beginning of the buffer.
510 static int logbuf_has_space(u32 msg_size, bool empty)
514 if (log_next_idx > log_first_idx || empty)
515 free = max(log_buf_len - log_next_idx, log_first_idx);
517 free = log_first_idx - log_next_idx;
520 * We need space also for an empty header that signalizes wrapping
523 return free >= msg_size + sizeof(struct printk_log);
526 static int log_make_free_space(u32 msg_size)
528 while (log_first_seq < log_next_seq &&
529 !logbuf_has_space(msg_size, false)) {
530 /* drop old messages until we have enough contiguous space */
531 log_first_idx = log_next(log_first_idx);
535 if (clear_seq < log_first_seq) {
536 clear_seq = log_first_seq;
537 clear_idx = log_first_idx;
540 /* sequence numbers are equal, so the log buffer is empty */
541 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
547 /* compute the message size including the padding bytes */
548 static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
552 size = sizeof(struct printk_log) + text_len + dict_len;
553 *pad_len = (-size) & (LOG_ALIGN - 1);
560 * Define how much of the log buffer we could take at maximum. The value
561 * must be greater than two. Note that only half of the buffer is available
562 * when the index points to the middle.
564 #define MAX_LOG_TAKE_PART 4
565 static const char trunc_msg[] = "<truncated>";
567 static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
568 u16 *dict_len, u32 *pad_len)
571 * The message should not take the whole buffer. Otherwise, it might
572 * get removed too soon.
574 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
575 if (*text_len > max_text_len)
576 *text_len = max_text_len;
577 /* enable the warning message */
578 *trunc_msg_len = strlen(trunc_msg);
579 /* disable the "dict" completely */
581 /* compute the size again, count also the warning message */
582 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
585 /* insert record into the buffer, discard old ones, update heads */
586 static int log_store(int facility, int level,
587 enum log_flags flags, u64 ts_nsec,
588 const char *dict, u16 dict_len,
589 const char *text, u16 text_len)
591 struct printk_log *msg;
593 u16 trunc_msg_len = 0;
595 /* number of '\0' padding bytes to next message */
596 size = msg_used_size(text_len, dict_len, &pad_len);
598 if (log_make_free_space(size)) {
599 /* truncate the message if it is too long for empty buffer */
600 size = truncate_msg(&text_len, &trunc_msg_len,
601 &dict_len, &pad_len);
602 /* survive when the log buffer is too small for trunc_msg */
603 if (log_make_free_space(size))
607 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
609 * This message + an additional empty header does not fit
610 * at the end of the buffer. Add an empty header with len == 0
611 * to signify a wrap around.
613 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
618 msg = (struct printk_log *)(log_buf + log_next_idx);
619 memcpy(log_text(msg), text, text_len);
620 msg->text_len = text_len;
622 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
623 msg->text_len += trunc_msg_len;
625 memcpy(log_dict(msg), dict, dict_len);
626 msg->dict_len = dict_len;
627 msg->facility = facility;
628 msg->level = level & 7;
629 msg->flags = flags & 0x1f;
631 msg->ts_nsec = ts_nsec;
633 msg->ts_nsec = local_clock();
634 memset(log_dict(msg) + dict_len, 0, pad_len);
638 log_next_idx += msg->len;
641 return msg->text_len;
644 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
646 static int syslog_action_restricted(int type)
651 * Unless restricted, we allow "read all" and "get buffer size"
654 return type != SYSLOG_ACTION_READ_ALL &&
655 type != SYSLOG_ACTION_SIZE_BUFFER;
658 static int check_syslog_permissions(int type, int source)
661 * If this is from /proc/kmsg and we've already opened it, then we've
662 * already done the capabilities checks at open time.
664 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
667 if (syslog_action_restricted(type)) {
668 if (capable(CAP_SYSLOG))
671 * For historical reasons, accept CAP_SYS_ADMIN too, with
674 if (capable(CAP_SYS_ADMIN)) {
675 pr_warn_once("%s (%d): Attempt to access syslog with "
676 "CAP_SYS_ADMIN but no CAP_SYSLOG "
678 current->comm, task_pid_nr(current));
684 return security_syslog(type);
687 static void append_char(char **pp, char *e, char c)
693 static ssize_t msg_print_ext_header(char *buf, size_t size,
694 struct printk_log *msg, u64 seq)
696 u64 ts_usec = msg->ts_nsec;
698 do_div(ts_usec, 1000);
700 return scnprintf(buf, size, "%u,%llu,%llu,%c;",
701 (msg->facility << 3) | msg->level, seq, ts_usec,
702 msg->flags & LOG_CONT ? 'c' : '-');
705 static ssize_t msg_print_ext_body(char *buf, size_t size,
706 char *dict, size_t dict_len,
707 char *text, size_t text_len)
709 char *p = buf, *e = buf + size;
712 /* escape non-printable characters */
713 for (i = 0; i < text_len; i++) {
714 unsigned char c = text[i];
716 if (c < ' ' || c >= 127 || c == '\\')
717 p += scnprintf(p, e - p, "\\x%02x", c);
719 append_char(&p, e, c);
721 append_char(&p, e, '\n');
726 for (i = 0; i < dict_len; i++) {
727 unsigned char c = dict[i];
730 append_char(&p, e, ' ');
735 append_char(&p, e, '\n');
740 if (c < ' ' || c >= 127 || c == '\\') {
741 p += scnprintf(p, e - p, "\\x%02x", c);
745 append_char(&p, e, c);
747 append_char(&p, e, '\n');
753 /* /dev/kmsg - userspace message inject/listen interface */
754 struct devkmsg_user {
757 struct ratelimit_state rs;
759 char buf[CONSOLE_EXT_LOG_MAX];
762 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
765 int level = default_message_loglevel;
766 int facility = 1; /* LOG_USER */
767 struct file *file = iocb->ki_filp;
768 struct devkmsg_user *user = file->private_data;
769 size_t len = iov_iter_count(from);
772 if (!user || len > LOG_LINE_MAX)
775 /* Ignore when user logging is disabled. */
776 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
779 /* Ratelimit when not explicitly enabled. */
780 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
781 if (!___ratelimit(&user->rs, current->comm))
785 buf = kmalloc(len+1, GFP_KERNEL);
790 if (!copy_from_iter_full(buf, len, from)) {
796 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
797 * the decimal value represents 32bit, the lower 3 bit are the log
798 * level, the rest are the log facility.
800 * If no prefix or no userspace facility is specified, we
801 * enforce LOG_USER, to be able to reliably distinguish
802 * kernel-generated messages from userspace-injected ones.
805 if (line[0] == '<') {
809 u = simple_strtoul(line + 1, &endp, 10);
810 if (endp && endp[0] == '>') {
811 level = LOG_LEVEL(u);
812 if (LOG_FACILITY(u) != 0)
813 facility = LOG_FACILITY(u);
820 printk_emit(facility, level, NULL, 0, "%s", line);
825 static ssize_t devkmsg_read(struct file *file, char __user *buf,
826 size_t count, loff_t *ppos)
828 struct devkmsg_user *user = file->private_data;
829 struct printk_log *msg;
836 ret = mutex_lock_interruptible(&user->lock);
841 while (user->seq == log_next_seq) {
842 if (file->f_flags & O_NONBLOCK) {
849 ret = wait_event_interruptible(log_wait,
850 user->seq != log_next_seq);
856 if (user->seq < log_first_seq) {
857 /* our last seen message is gone, return error and reset */
858 user->idx = log_first_idx;
859 user->seq = log_first_seq;
865 msg = log_from_idx(user->idx);
866 len = msg_print_ext_header(user->buf, sizeof(user->buf),
868 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
869 log_dict(msg), msg->dict_len,
870 log_text(msg), msg->text_len);
872 user->idx = log_next(user->idx);
881 if (copy_to_user(buf, user->buf, len)) {
887 mutex_unlock(&user->lock);
891 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
893 struct devkmsg_user *user = file->private_data;
904 /* the first record */
905 user->idx = log_first_idx;
906 user->seq = log_first_seq;
910 * The first record after the last SYSLOG_ACTION_CLEAR,
911 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
912 * changes no global state, and does not clear anything.
914 user->idx = clear_idx;
915 user->seq = clear_seq;
918 /* after the last record */
919 user->idx = log_next_idx;
920 user->seq = log_next_seq;
929 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
931 struct devkmsg_user *user = file->private_data;
935 return EPOLLERR|EPOLLNVAL;
937 poll_wait(file, &log_wait, wait);
940 if (user->seq < log_next_seq) {
941 /* return error when data has vanished underneath us */
942 if (user->seq < log_first_seq)
943 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
945 ret = EPOLLIN|EPOLLRDNORM;
952 static int devkmsg_open(struct inode *inode, struct file *file)
954 struct devkmsg_user *user;
957 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
960 /* write-only does not need any file context */
961 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
962 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
968 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
972 ratelimit_default_init(&user->rs);
973 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
975 mutex_init(&user->lock);
978 user->idx = log_first_idx;
979 user->seq = log_first_seq;
982 file->private_data = user;
986 static int devkmsg_release(struct inode *inode, struct file *file)
988 struct devkmsg_user *user = file->private_data;
993 ratelimit_state_exit(&user->rs);
995 mutex_destroy(&user->lock);
1000 const struct file_operations kmsg_fops = {
1001 .open = devkmsg_open,
1002 .read = devkmsg_read,
1003 .write_iter = devkmsg_write,
1004 .llseek = devkmsg_llseek,
1005 .poll = devkmsg_poll,
1006 .release = devkmsg_release,
1009 #ifdef CONFIG_CRASH_CORE
1011 * This appends the listed symbols to /proc/vmcore
1013 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1014 * obtain access to symbols that are otherwise very difficult to locate. These
1015 * symbols are specifically used so that utilities can access and extract the
1016 * dmesg log from a vmcore file after a crash.
1018 void log_buf_vmcoreinfo_setup(void)
1020 VMCOREINFO_SYMBOL(log_buf);
1021 VMCOREINFO_SYMBOL(log_buf_len);
1022 VMCOREINFO_SYMBOL(log_first_idx);
1023 VMCOREINFO_SYMBOL(clear_idx);
1024 VMCOREINFO_SYMBOL(log_next_idx);
1026 * Export struct printk_log size and field offsets. User space tools can
1027 * parse it and detect any changes to structure down the line.
1029 VMCOREINFO_STRUCT_SIZE(printk_log);
1030 VMCOREINFO_OFFSET(printk_log, ts_nsec);
1031 VMCOREINFO_OFFSET(printk_log, len);
1032 VMCOREINFO_OFFSET(printk_log, text_len);
1033 VMCOREINFO_OFFSET(printk_log, dict_len);
1037 /* requested log_buf_len from kernel cmdline */
1038 static unsigned long __initdata new_log_buf_len;
1040 /* we practice scaling the ring buffer by powers of 2 */
1041 static void __init log_buf_len_update(unsigned size)
1044 size = roundup_pow_of_two(size);
1045 if (size > log_buf_len)
1046 new_log_buf_len = size;
1049 /* save requested log_buf_len since it's too early to process it */
1050 static int __init log_buf_len_setup(char *str)
1052 unsigned size = memparse(str, &str);
1054 log_buf_len_update(size);
1058 early_param("log_buf_len", log_buf_len_setup);
1061 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1063 static void __init log_buf_add_cpu(void)
1065 unsigned int cpu_extra;
1068 * archs should set up cpu_possible_bits properly with
1069 * set_cpu_possible() after setup_arch() but just in
1070 * case lets ensure this is valid.
1072 if (num_possible_cpus() == 1)
1075 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1077 /* by default this will only continue through for large > 64 CPUs */
1078 if (cpu_extra <= __LOG_BUF_LEN / 2)
1081 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1082 __LOG_CPU_MAX_BUF_LEN);
1083 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1085 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1087 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1089 #else /* !CONFIG_SMP */
1090 static inline void log_buf_add_cpu(void) {}
1091 #endif /* CONFIG_SMP */
1093 void __init setup_log_buf(int early)
1095 unsigned long flags;
1099 if (log_buf != __log_buf)
1102 if (!early && !new_log_buf_len)
1105 if (!new_log_buf_len)
1110 memblock_virt_alloc(new_log_buf_len, LOG_ALIGN);
1112 new_log_buf = memblock_virt_alloc_nopanic(new_log_buf_len,
1116 if (unlikely(!new_log_buf)) {
1117 pr_err("log_buf_len: %ld bytes not available\n",
1122 logbuf_lock_irqsave(flags);
1123 log_buf_len = new_log_buf_len;
1124 log_buf = new_log_buf;
1125 new_log_buf_len = 0;
1126 free = __LOG_BUF_LEN - log_next_idx;
1127 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1128 logbuf_unlock_irqrestore(flags);
1130 pr_info("log_buf_len: %d bytes\n", log_buf_len);
1131 pr_info("early log buf free: %d(%d%%)\n",
1132 free, (free * 100) / __LOG_BUF_LEN);
1135 static bool __read_mostly ignore_loglevel;
1137 static int __init ignore_loglevel_setup(char *str)
1139 ignore_loglevel = true;
1140 pr_info("debug: ignoring loglevel setting.\n");
1145 early_param("ignore_loglevel", ignore_loglevel_setup);
1146 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1147 MODULE_PARM_DESC(ignore_loglevel,
1148 "ignore loglevel setting (prints all kernel messages to the console)");
1150 static bool suppress_message_printing(int level)
1152 return (level >= console_loglevel && !ignore_loglevel);
1155 #ifdef CONFIG_BOOT_PRINTK_DELAY
1157 static int boot_delay; /* msecs delay after each printk during bootup */
1158 static unsigned long long loops_per_msec; /* based on boot_delay */
1160 static int __init boot_delay_setup(char *str)
1164 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1165 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1167 get_option(&str, &boot_delay);
1168 if (boot_delay > 10 * 1000)
1171 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1172 "HZ: %d, loops_per_msec: %llu\n",
1173 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1176 early_param("boot_delay", boot_delay_setup);
1178 static void boot_delay_msec(int level)
1180 unsigned long long k;
1181 unsigned long timeout;
1183 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1184 || suppress_message_printing(level)) {
1188 k = (unsigned long long)loops_per_msec * boot_delay;
1190 timeout = jiffies + msecs_to_jiffies(boot_delay);
1195 * use (volatile) jiffies to prevent
1196 * compiler reduction; loop termination via jiffies
1197 * is secondary and may or may not happen.
1199 if (time_after(jiffies, timeout))
1201 touch_nmi_watchdog();
1205 static inline void boot_delay_msec(int level)
1210 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1211 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1213 static size_t print_time(u64 ts, char *buf)
1215 unsigned long rem_nsec;
1220 rem_nsec = do_div(ts, 1000000000);
1223 return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
1225 return sprintf(buf, "[%5lu.%06lu] ",
1226 (unsigned long)ts, rem_nsec / 1000);
1229 static size_t print_prefix(const struct printk_log *msg, bool syslog, char *buf)
1232 unsigned int prefix = (msg->facility << 3) | msg->level;
1236 len += sprintf(buf, "<%u>", prefix);
1241 else if (prefix > 99)
1243 else if (prefix > 9)
1248 len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
1252 static size_t msg_print_text(const struct printk_log *msg, bool syslog, char *buf, size_t size)
1254 const char *text = log_text(msg);
1255 size_t text_size = msg->text_len;
1259 const char *next = memchr(text, '\n', text_size);
1263 text_len = next - text;
1265 text_size -= next - text;
1267 text_len = text_size;
1271 if (print_prefix(msg, syslog, NULL) +
1272 text_len + 1 >= size - len)
1275 len += print_prefix(msg, syslog, buf + len);
1276 memcpy(buf + len, text, text_len);
1280 /* SYSLOG_ACTION_* buffer size only calculation */
1281 len += print_prefix(msg, syslog, NULL);
1292 static int syslog_print(char __user *buf, int size)
1295 struct printk_log *msg;
1298 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1307 if (syslog_seq < log_first_seq) {
1308 /* messages are gone, move to first one */
1309 syslog_seq = log_first_seq;
1310 syslog_idx = log_first_idx;
1313 if (syslog_seq == log_next_seq) {
1314 logbuf_unlock_irq();
1318 skip = syslog_partial;
1319 msg = log_from_idx(syslog_idx);
1320 n = msg_print_text(msg, true, text, LOG_LINE_MAX + PREFIX_MAX);
1321 if (n - syslog_partial <= size) {
1322 /* message fits into buffer, move forward */
1323 syslog_idx = log_next(syslog_idx);
1325 n -= syslog_partial;
1328 /* partial read(), remember position */
1330 syslog_partial += n;
1333 logbuf_unlock_irq();
1338 if (copy_to_user(buf, text + skip, n)) {
1353 static int syslog_print_all(char __user *buf, int size, bool clear)
1361 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1367 * Find first record that fits, including all following records,
1368 * into the user-provided buffer for this dump.
1372 while (seq < log_next_seq) {
1373 struct printk_log *msg = log_from_idx(idx);
1375 len += msg_print_text(msg, true, NULL, 0);
1376 idx = log_next(idx);
1380 /* move first record forward until length fits into the buffer */
1383 while (len > size && seq < log_next_seq) {
1384 struct printk_log *msg = log_from_idx(idx);
1386 len -= msg_print_text(msg, true, NULL, 0);
1387 idx = log_next(idx);
1391 /* last message fitting into this dump */
1392 next_seq = log_next_seq;
1395 while (len >= 0 && seq < next_seq) {
1396 struct printk_log *msg = log_from_idx(idx);
1399 textlen = msg_print_text(msg, true, text,
1400 LOG_LINE_MAX + PREFIX_MAX);
1405 idx = log_next(idx);
1408 logbuf_unlock_irq();
1409 if (copy_to_user(buf + len, text, textlen))
1415 if (seq < log_first_seq) {
1416 /* messages are gone, move to next one */
1417 seq = log_first_seq;
1418 idx = log_first_idx;
1423 clear_seq = log_next_seq;
1424 clear_idx = log_next_idx;
1426 logbuf_unlock_irq();
1432 static void syslog_clear(void)
1435 clear_seq = log_next_seq;
1436 clear_idx = log_next_idx;
1437 logbuf_unlock_irq();
1440 int do_syslog(int type, char __user *buf, int len, int source)
1443 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1446 error = check_syslog_permissions(type, source);
1451 case SYSLOG_ACTION_CLOSE: /* Close log */
1453 case SYSLOG_ACTION_OPEN: /* Open log */
1455 case SYSLOG_ACTION_READ: /* Read from log */
1456 if (!buf || len < 0)
1460 if (!access_ok(VERIFY_WRITE, buf, len))
1462 error = wait_event_interruptible(log_wait,
1463 syslog_seq != log_next_seq);
1466 error = syslog_print(buf, len);
1468 /* Read/clear last kernel messages */
1469 case SYSLOG_ACTION_READ_CLEAR:
1472 /* Read last kernel messages */
1473 case SYSLOG_ACTION_READ_ALL:
1474 if (!buf || len < 0)
1478 if (!access_ok(VERIFY_WRITE, buf, len))
1480 error = syslog_print_all(buf, len, clear);
1482 /* Clear ring buffer */
1483 case SYSLOG_ACTION_CLEAR:
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:
1501 if (len < 1 || len > 8)
1503 if (len < minimum_console_loglevel)
1504 len = minimum_console_loglevel;
1505 console_loglevel = len;
1506 /* Implicitly re-enable logging to console */
1507 saved_console_loglevel = LOGLEVEL_DEFAULT;
1509 /* Number of chars in the log buffer */
1510 case SYSLOG_ACTION_SIZE_UNREAD:
1512 if (syslog_seq < log_first_seq) {
1513 /* messages are gone, move to first one */
1514 syslog_seq = log_first_seq;
1515 syslog_idx = log_first_idx;
1518 if (source == SYSLOG_FROM_PROC) {
1520 * Short-cut for poll(/"proc/kmsg") which simply checks
1521 * for pending data, not the size; return the count of
1522 * records, not the length.
1524 error = log_next_seq - syslog_seq;
1526 u64 seq = syslog_seq;
1527 u32 idx = syslog_idx;
1529 while (seq < log_next_seq) {
1530 struct printk_log *msg = log_from_idx(idx);
1532 error += msg_print_text(msg, true, NULL, 0);
1533 idx = log_next(idx);
1536 error -= syslog_partial;
1538 logbuf_unlock_irq();
1540 /* Size of the log buffer */
1541 case SYSLOG_ACTION_SIZE_BUFFER:
1542 error = log_buf_len;
1552 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1554 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1558 * Special console_lock variants that help to reduce the risk of soft-lockups.
1559 * They allow to pass console_lock to another printk() call using a busy wait.
1562 #ifdef CONFIG_LOCKDEP
1563 static struct lockdep_map console_owner_dep_map = {
1564 .name = "console_owner"
1568 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1569 static struct task_struct *console_owner;
1570 static bool console_waiter;
1573 * console_lock_spinning_enable - mark beginning of code where another
1574 * thread might safely busy wait
1576 * This basically converts console_lock into a spinlock. This marks
1577 * the section where the console_lock owner can not sleep, because
1578 * there may be a waiter spinning (like a spinlock). Also it must be
1579 * ready to hand over the lock at the end of the section.
1581 static void console_lock_spinning_enable(void)
1583 raw_spin_lock(&console_owner_lock);
1584 console_owner = current;
1585 raw_spin_unlock(&console_owner_lock);
1587 /* The waiter may spin on us after setting console_owner */
1588 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1592 * console_lock_spinning_disable_and_check - mark end of code where another
1593 * thread was able to busy wait and check if there is a waiter
1595 * This is called at the end of the section where spinning is allowed.
1596 * It has two functions. First, it is a signal that it is no longer
1597 * safe to start busy waiting for the lock. Second, it checks if
1598 * there is a busy waiter and passes the lock rights to her.
1600 * Important: Callers lose the lock if there was a busy waiter.
1601 * They must not touch items synchronized by console_lock
1604 * Return: 1 if the lock rights were passed, 0 otherwise.
1606 static int console_lock_spinning_disable_and_check(void)
1610 raw_spin_lock(&console_owner_lock);
1611 waiter = READ_ONCE(console_waiter);
1612 console_owner = NULL;
1613 raw_spin_unlock(&console_owner_lock);
1616 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1620 /* The waiter is now free to continue */
1621 WRITE_ONCE(console_waiter, false);
1623 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1626 * Hand off console_lock to waiter. The waiter will perform
1627 * the up(). After this, the waiter is the console_lock owner.
1629 mutex_release(&console_lock_dep_map, 1, _THIS_IP_);
1634 * console_trylock_spinning - try to get console_lock by busy waiting
1636 * This allows to busy wait for the console_lock when the current
1637 * owner is running in specially marked sections. It means that
1638 * the current owner is running and cannot reschedule until it
1639 * is ready to lose the lock.
1641 * Return: 1 if we got the lock, 0 othrewise
1643 static int console_trylock_spinning(void)
1645 struct task_struct *owner = NULL;
1648 unsigned long flags;
1650 if (console_trylock())
1653 printk_safe_enter_irqsave(flags);
1655 raw_spin_lock(&console_owner_lock);
1656 owner = READ_ONCE(console_owner);
1657 waiter = READ_ONCE(console_waiter);
1658 if (!waiter && owner && owner != current) {
1659 WRITE_ONCE(console_waiter, true);
1662 raw_spin_unlock(&console_owner_lock);
1665 * If there is an active printk() writing to the
1666 * consoles, instead of having it write our data too,
1667 * see if we can offload that load from the active
1668 * printer, and do some printing ourselves.
1669 * Go into a spin only if there isn't already a waiter
1670 * spinning, and there is an active printer, and
1671 * that active printer isn't us (recursive printk?).
1674 printk_safe_exit_irqrestore(flags);
1678 /* We spin waiting for the owner to release us */
1679 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1680 /* Owner will clear console_waiter on hand off */
1681 while (READ_ONCE(console_waiter))
1683 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1685 printk_safe_exit_irqrestore(flags);
1687 * The owner passed the console lock to us.
1688 * Since we did not spin on console lock, annotate
1689 * this as a trylock. Otherwise lockdep will
1692 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1698 * Call the console drivers, asking them to write out
1699 * log_buf[start] to log_buf[end - 1].
1700 * The console_lock must be held.
1702 static void call_console_drivers(const char *ext_text, size_t ext_len,
1703 const char *text, size_t len)
1705 struct console *con;
1707 trace_console_rcuidle(text, len);
1709 if (!console_drivers)
1712 for_each_console(con) {
1713 if (exclusive_console && con != exclusive_console)
1715 if (!(con->flags & CON_ENABLED))
1719 if (!cpu_online(smp_processor_id()) &&
1720 !(con->flags & CON_ANYTIME))
1722 if (con->flags & CON_EXTENDED)
1723 con->write(con, ext_text, ext_len);
1725 con->write(con, text, len);
1729 int printk_delay_msec __read_mostly;
1731 static inline void printk_delay(void)
1733 if (unlikely(printk_delay_msec)) {
1734 int m = printk_delay_msec;
1738 touch_nmi_watchdog();
1744 * Continuation lines are buffered, and not committed to the record buffer
1745 * until the line is complete, or a race forces it. The line fragments
1746 * though, are printed immediately to the consoles to ensure everything has
1747 * reached the console in case of a kernel crash.
1749 static struct cont {
1750 char buf[LOG_LINE_MAX];
1751 size_t len; /* length == 0 means unused buffer */
1752 struct task_struct *owner; /* task of first print*/
1753 u64 ts_nsec; /* time of first print */
1754 u8 level; /* log level of first message */
1755 u8 facility; /* log facility of first message */
1756 enum log_flags flags; /* prefix, newline flags */
1759 static void cont_flush(void)
1764 log_store(cont.facility, cont.level, cont.flags, cont.ts_nsec,
1765 NULL, 0, cont.buf, cont.len);
1769 static bool cont_add(int facility, int level, enum log_flags flags, const char *text, size_t len)
1772 * If ext consoles are present, flush and skip in-kernel
1773 * continuation. See nr_ext_console_drivers definition. Also, if
1774 * the line gets too long, split it up in separate records.
1776 if (nr_ext_console_drivers || 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;
1799 if (cont.len > (sizeof(cont.buf) * 80) / 100)
1805 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)
1808 * If an earlier line was buffered, and we're a continuation
1809 * write from the same process, try to add it to the buffer.
1812 if (cont.owner == current && (lflags & LOG_CONT)) {
1813 if (cont_add(facility, level, lflags, text, text_len))
1816 /* Otherwise, make sure it's flushed */
1820 /* Skip empty continuation lines that couldn't be added - they just flush */
1821 if (!text_len && (lflags & LOG_CONT))
1824 /* If it doesn't end in a newline, try to buffer the current line */
1825 if (!(lflags & LOG_NEWLINE)) {
1826 if (cont_add(facility, level, lflags, text, text_len))
1830 /* Store it in the record log */
1831 return log_store(facility, level, lflags, 0, dict, dictlen, text, text_len);
1834 /* Must be called under logbuf_lock. */
1835 int vprintk_store(int facility, int level,
1836 const char *dict, size_t dictlen,
1837 const char *fmt, va_list args)
1839 static char textbuf[LOG_LINE_MAX];
1840 char *text = textbuf;
1842 enum log_flags lflags = 0;
1845 * The printf needs to come first; we need the syslog
1846 * prefix which might be passed-in as a parameter.
1848 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1850 /* mark and strip a trailing newline */
1851 if (text_len && text[text_len-1] == '\n') {
1853 lflags |= LOG_NEWLINE;
1856 /* strip kernel syslog prefix and extract log level or control flags */
1857 if (facility == 0) {
1860 while ((kern_level = printk_get_level(text)) != 0) {
1861 switch (kern_level) {
1863 if (level == LOGLEVEL_DEFAULT)
1864 level = kern_level - '0';
1866 case 'd': /* KERN_DEFAULT */
1867 lflags |= LOG_PREFIX;
1869 case 'c': /* KERN_CONT */
1878 if (level == LOGLEVEL_DEFAULT)
1879 level = default_message_loglevel;
1882 lflags |= LOG_PREFIX|LOG_NEWLINE;
1884 if (suppress_message_printing(level))
1885 lflags |= LOG_NOCONS;
1887 return log_output(facility, level, lflags,
1888 dict, dictlen, text, text_len);
1891 asmlinkage int vprintk_emit(int facility, int level,
1892 const char *dict, size_t dictlen,
1893 const char *fmt, va_list args)
1896 bool in_sched = false;
1897 unsigned long flags;
1899 if (level == LOGLEVEL_SCHED) {
1900 level = LOGLEVEL_DEFAULT;
1904 boot_delay_msec(level);
1907 /* This stops the holder of console_sem just where we want him */
1908 logbuf_lock_irqsave(flags);
1909 printed_len = vprintk_store(facility, level, dict, dictlen, fmt, args);
1910 logbuf_unlock_irqrestore(flags);
1912 /* If called from the scheduler, we can not call up(). */
1915 * Disable preemption to avoid being preempted while holding
1916 * console_sem which would prevent anyone from printing to
1921 * Try to acquire and then immediately release the console
1922 * semaphore. The release will print out buffers and wake up
1923 * /dev/kmsg and syslog() users.
1925 if (console_trylock_spinning())
1933 EXPORT_SYMBOL(vprintk_emit);
1935 asmlinkage int vprintk(const char *fmt, va_list args)
1937 return vprintk_func(fmt, args);
1939 EXPORT_SYMBOL(vprintk);
1941 asmlinkage int printk_emit(int facility, int level,
1942 const char *dict, size_t dictlen,
1943 const char *fmt, ...)
1948 va_start(args, fmt);
1949 r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1954 EXPORT_SYMBOL(printk_emit);
1956 int vprintk_default(const char *fmt, va_list args)
1960 #ifdef CONFIG_KGDB_KDB
1961 /* Allow to pass printk() to kdb but avoid a recursion. */
1962 if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
1963 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
1967 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1971 EXPORT_SYMBOL_GPL(vprintk_default);
1974 * printk - print a kernel message
1975 * @fmt: format string
1977 * This is printk(). It can be called from any context. We want it to work.
1979 * We try to grab the console_lock. If we succeed, it's easy - we log the
1980 * output and call the console drivers. If we fail to get the semaphore, we
1981 * place the output into the log buffer and return. The current holder of
1982 * the console_sem will notice the new output in console_unlock(); and will
1983 * send it to the consoles before releasing the lock.
1985 * One effect of this deferred printing is that code which calls printk() and
1986 * then changes console_loglevel may break. This is because console_loglevel
1987 * is inspected when the actual printing occurs.
1992 * See the vsnprintf() documentation for format string extensions over C99.
1994 asmlinkage __visible int printk(const char *fmt, ...)
1999 va_start(args, fmt);
2000 r = vprintk_func(fmt, args);
2005 EXPORT_SYMBOL(printk);
2007 #else /* CONFIG_PRINTK */
2009 #define LOG_LINE_MAX 0
2010 #define PREFIX_MAX 0
2012 static u64 syslog_seq;
2013 static u32 syslog_idx;
2014 static u64 console_seq;
2015 static u32 console_idx;
2016 static u64 log_first_seq;
2017 static u32 log_first_idx;
2018 static u64 log_next_seq;
2019 static char *log_text(const struct printk_log *msg) { return NULL; }
2020 static char *log_dict(const struct printk_log *msg) { return NULL; }
2021 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2022 static u32 log_next(u32 idx) { return 0; }
2023 static ssize_t msg_print_ext_header(char *buf, size_t size,
2024 struct printk_log *msg,
2025 u64 seq) { return 0; }
2026 static ssize_t msg_print_ext_body(char *buf, size_t size,
2027 char *dict, size_t dict_len,
2028 char *text, size_t text_len) { return 0; }
2029 static void console_lock_spinning_enable(void) { }
2030 static int console_lock_spinning_disable_and_check(void) { return 0; }
2031 static void call_console_drivers(const char *ext_text, size_t ext_len,
2032 const char *text, size_t len) {}
2033 static size_t msg_print_text(const struct printk_log *msg,
2034 bool syslog, char *buf, size_t size) { return 0; }
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) {
2357 len = sprintf(text, "** %u printk messages dropped **\n",
2358 (unsigned)(log_first_seq - console_seq));
2360 /* messages are gone, move to first one */
2361 console_seq = log_first_seq;
2362 console_idx = log_first_idx;
2367 if (console_seq == log_next_seq)
2370 msg = log_from_idx(console_idx);
2371 if (msg->flags & LOG_NOCONS) {
2373 * Skip record if !ignore_loglevel, and
2374 * record has level above the console loglevel.
2376 console_idx = log_next(console_idx);
2381 len += msg_print_text(msg,
2382 console_msg_format & MSG_FORMAT_SYSLOG,
2384 sizeof(text) - len);
2385 if (nr_ext_console_drivers) {
2386 ext_len = msg_print_ext_header(ext_text,
2389 ext_len += msg_print_ext_body(ext_text + ext_len,
2390 sizeof(ext_text) - ext_len,
2391 log_dict(msg), msg->dict_len,
2392 log_text(msg), msg->text_len);
2394 console_idx = log_next(console_idx);
2396 raw_spin_unlock(&logbuf_lock);
2399 * While actively printing out messages, if another printk()
2400 * were to occur on another CPU, it may wait for this one to
2401 * finish. This task can not be preempted if there is a
2402 * waiter waiting to take over.
2404 console_lock_spinning_enable();
2406 stop_critical_timings(); /* don't trace print latency */
2407 call_console_drivers(ext_text, ext_len, text, len);
2408 start_critical_timings();
2410 if (console_lock_spinning_disable_and_check()) {
2411 printk_safe_exit_irqrestore(flags);
2415 printk_safe_exit_irqrestore(flags);
2417 if (do_cond_resched)
2423 /* Release the exclusive_console once it is used */
2424 if (unlikely(exclusive_console))
2425 exclusive_console = NULL;
2427 raw_spin_unlock(&logbuf_lock);
2432 * Someone could have filled up the buffer again, so re-check if there's
2433 * something to flush. In case we cannot trylock the console_sem again,
2434 * there's a new owner and the console_unlock() from them will do the
2435 * flush, no worries.
2437 raw_spin_lock(&logbuf_lock);
2438 retry = console_seq != log_next_seq;
2439 raw_spin_unlock(&logbuf_lock);
2440 printk_safe_exit_irqrestore(flags);
2442 if (retry && console_trylock())
2445 EXPORT_SYMBOL(console_unlock);
2448 * console_conditional_schedule - yield the CPU if required
2450 * If the console code is currently allowed to sleep, and
2451 * if this CPU should yield the CPU to another task, do
2454 * Must be called within console_lock();.
2456 void __sched console_conditional_schedule(void)
2458 if (console_may_schedule)
2461 EXPORT_SYMBOL(console_conditional_schedule);
2463 void console_unblank(void)
2468 * console_unblank can no longer be called in interrupt context unless
2469 * oops_in_progress is set to 1..
2471 if (oops_in_progress) {
2472 if (down_trylock_console_sem() != 0)
2478 console_may_schedule = 0;
2480 if ((c->flags & CON_ENABLED) && c->unblank)
2486 * console_flush_on_panic - flush console content on panic
2488 * Immediately output all pending messages no matter what.
2490 void console_flush_on_panic(void)
2493 * If someone else is holding the console lock, trylock will fail
2494 * and may_schedule may be set. Ignore and proceed to unlock so
2495 * that messages are flushed out. As this can be called from any
2496 * context and we don't want to get preempted while flushing,
2497 * ensure may_schedule is cleared.
2500 console_may_schedule = 0;
2505 * Return the console tty driver structure and its associated index
2507 struct tty_driver *console_device(int *index)
2510 struct tty_driver *driver = NULL;
2513 for_each_console(c) {
2516 driver = c->device(c, index);
2525 * Prevent further output on the passed console device so that (for example)
2526 * serial drivers can disable console output before suspending a port, and can
2527 * re-enable output afterwards.
2529 void console_stop(struct console *console)
2532 console->flags &= ~CON_ENABLED;
2535 EXPORT_SYMBOL(console_stop);
2537 void console_start(struct console *console)
2540 console->flags |= CON_ENABLED;
2543 EXPORT_SYMBOL(console_start);
2545 static int __read_mostly keep_bootcon;
2547 static int __init keep_bootcon_setup(char *str)
2550 pr_info("debug: skip boot console de-registration.\n");
2555 early_param("keep_bootcon", keep_bootcon_setup);
2558 * The console driver calls this routine during kernel initialization
2559 * to register the console printing procedure with printk() and to
2560 * print any messages that were printed by the kernel before the
2561 * console driver was initialized.
2563 * This can happen pretty early during the boot process (because of
2564 * early_printk) - sometimes before setup_arch() completes - be careful
2565 * of what kernel features are used - they may not be initialised yet.
2567 * There are two types of consoles - bootconsoles (early_printk) and
2568 * "real" consoles (everything which is not a bootconsole) which are
2569 * handled differently.
2570 * - Any number of bootconsoles can be registered at any time.
2571 * - As soon as a "real" console is registered, all bootconsoles
2572 * will be unregistered automatically.
2573 * - Once a "real" console is registered, any attempt to register a
2574 * bootconsoles will be rejected
2576 void register_console(struct console *newcon)
2579 unsigned long flags;
2580 struct console *bcon = NULL;
2581 struct console_cmdline *c;
2582 static bool has_preferred;
2584 if (console_drivers)
2585 for_each_console(bcon)
2586 if (WARN(bcon == newcon,
2587 "console '%s%d' already registered\n",
2588 bcon->name, bcon->index))
2592 * before we register a new CON_BOOT console, make sure we don't
2593 * already have a valid console
2595 if (console_drivers && newcon->flags & CON_BOOT) {
2596 /* find the last or real console */
2597 for_each_console(bcon) {
2598 if (!(bcon->flags & CON_BOOT)) {
2599 pr_info("Too late to register bootconsole %s%d\n",
2600 newcon->name, newcon->index);
2606 if (console_drivers && console_drivers->flags & CON_BOOT)
2607 bcon = console_drivers;
2609 if (!has_preferred || bcon || !console_drivers)
2610 has_preferred = preferred_console >= 0;
2613 * See if we want to use this console driver. If we
2614 * didn't select a console we take the first one
2615 * that registers here.
2617 if (!has_preferred) {
2618 if (newcon->index < 0)
2620 if (newcon->setup == NULL ||
2621 newcon->setup(newcon, NULL) == 0) {
2622 newcon->flags |= CON_ENABLED;
2623 if (newcon->device) {
2624 newcon->flags |= CON_CONSDEV;
2625 has_preferred = true;
2631 * See if this console matches one we selected on
2634 for (i = 0, c = console_cmdline;
2635 i < MAX_CMDLINECONSOLES && c->name[0];
2637 if (!newcon->match ||
2638 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2639 /* default matching */
2640 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2641 if (strcmp(c->name, newcon->name) != 0)
2643 if (newcon->index >= 0 &&
2644 newcon->index != c->index)
2646 if (newcon->index < 0)
2647 newcon->index = c->index;
2649 if (_braille_register_console(newcon, c))
2652 if (newcon->setup &&
2653 newcon->setup(newcon, c->options) != 0)
2657 newcon->flags |= CON_ENABLED;
2658 if (i == preferred_console) {
2659 newcon->flags |= CON_CONSDEV;
2660 has_preferred = true;
2665 if (!(newcon->flags & CON_ENABLED))
2669 * If we have a bootconsole, and are switching to a real console,
2670 * don't print everything out again, since when the boot console, and
2671 * the real console are the same physical device, it's annoying to
2672 * see the beginning boot messages twice
2674 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2675 newcon->flags &= ~CON_PRINTBUFFER;
2678 * Put this console in the list - keep the
2679 * preferred driver at the head of the list.
2682 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2683 newcon->next = console_drivers;
2684 console_drivers = newcon;
2686 newcon->next->flags &= ~CON_CONSDEV;
2688 newcon->next = console_drivers->next;
2689 console_drivers->next = newcon;
2692 if (newcon->flags & CON_EXTENDED)
2693 if (!nr_ext_console_drivers++)
2694 pr_info("printk: continuation disabled due to ext consoles, expect more fragments in /dev/kmsg\n");
2696 if (newcon->flags & CON_PRINTBUFFER) {
2698 * console_unlock(); will print out the buffered messages
2701 logbuf_lock_irqsave(flags);
2702 console_seq = syslog_seq;
2703 console_idx = syslog_idx;
2704 logbuf_unlock_irqrestore(flags);
2706 * We're about to replay the log buffer. Only do this to the
2707 * just-registered console to avoid excessive message spam to
2708 * the already-registered consoles.
2710 exclusive_console = newcon;
2713 console_sysfs_notify();
2716 * By unregistering the bootconsoles after we enable the real console
2717 * we get the "console xxx enabled" message on all the consoles -
2718 * boot consoles, real consoles, etc - this is to ensure that end
2719 * users know there might be something in the kernel's log buffer that
2720 * went to the bootconsole (that they do not see on the real console)
2722 pr_info("%sconsole [%s%d] enabled\n",
2723 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2724 newcon->name, newcon->index);
2726 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2728 /* We need to iterate through all boot consoles, to make
2729 * sure we print everything out, before we unregister them.
2731 for_each_console(bcon)
2732 if (bcon->flags & CON_BOOT)
2733 unregister_console(bcon);
2736 EXPORT_SYMBOL(register_console);
2738 int unregister_console(struct console *console)
2740 struct console *a, *b;
2743 pr_info("%sconsole [%s%d] disabled\n",
2744 (console->flags & CON_BOOT) ? "boot" : "" ,
2745 console->name, console->index);
2747 res = _braille_unregister_console(console);
2753 if (console_drivers == console) {
2754 console_drivers=console->next;
2756 } else if (console_drivers) {
2757 for (a=console_drivers->next, b=console_drivers ;
2758 a; b=a, a=b->next) {
2767 if (!res && (console->flags & CON_EXTENDED))
2768 nr_ext_console_drivers--;
2771 * If this isn't the last console and it has CON_CONSDEV set, we
2772 * need to set it on the next preferred console.
2774 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2775 console_drivers->flags |= CON_CONSDEV;
2777 console->flags &= ~CON_ENABLED;
2779 console_sysfs_notify();
2782 EXPORT_SYMBOL(unregister_console);
2785 * Initialize the console device. This is called *early*, so
2786 * we can't necessarily depend on lots of kernel help here.
2787 * Just do some early initializations, and do the complex setup
2790 void __init console_init(void)
2794 initcall_entry_t *ce;
2796 /* Setup the default TTY line discipline. */
2800 * set up the console device so that later boot sequences can
2801 * inform about problems etc..
2803 ce = __con_initcall_start;
2804 trace_initcall_level("console");
2805 while (ce < __con_initcall_end) {
2806 call = initcall_from_entry(ce);
2807 trace_initcall_start(call);
2809 trace_initcall_finish(call, ret);
2815 * Some boot consoles access data that is in the init section and which will
2816 * be discarded after the initcalls have been run. To make sure that no code
2817 * will access this data, unregister the boot consoles in a late initcall.
2819 * If for some reason, such as deferred probe or the driver being a loadable
2820 * module, the real console hasn't registered yet at this point, there will
2821 * be a brief interval in which no messages are logged to the console, which
2822 * makes it difficult to diagnose problems that occur during this time.
2824 * To mitigate this problem somewhat, only unregister consoles whose memory
2825 * intersects with the init section. Note that all other boot consoles will
2826 * get unregistred when the real preferred console is registered.
2828 static int __init printk_late_init(void)
2830 struct console *con;
2833 for_each_console(con) {
2834 if (!(con->flags & CON_BOOT))
2837 /* Check addresses that might be used for enabled consoles. */
2838 if (init_section_intersects(con, sizeof(*con)) ||
2839 init_section_contains(con->write, 0) ||
2840 init_section_contains(con->read, 0) ||
2841 init_section_contains(con->device, 0) ||
2842 init_section_contains(con->unblank, 0) ||
2843 init_section_contains(con->data, 0)) {
2845 * Please, consider moving the reported consoles out
2846 * of the init section.
2848 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2849 con->name, con->index);
2850 unregister_console(con);
2853 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2854 console_cpu_notify);
2856 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2857 console_cpu_notify, NULL);
2861 late_initcall(printk_late_init);
2863 #if defined CONFIG_PRINTK
2865 * Delayed printk version, for scheduler-internal messages:
2867 #define PRINTK_PENDING_WAKEUP 0x01
2868 #define PRINTK_PENDING_OUTPUT 0x02
2870 static DEFINE_PER_CPU(int, printk_pending);
2872 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2874 int pending = __this_cpu_xchg(printk_pending, 0);
2876 if (pending & PRINTK_PENDING_OUTPUT) {
2877 /* If trylock fails, someone else is doing the printing */
2878 if (console_trylock())
2882 if (pending & PRINTK_PENDING_WAKEUP)
2883 wake_up_interruptible(&log_wait);
2886 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2887 .func = wake_up_klogd_work_func,
2888 .flags = IRQ_WORK_LAZY,
2891 void wake_up_klogd(void)
2894 if (waitqueue_active(&log_wait)) {
2895 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2896 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2901 void defer_console_output(void)
2904 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2905 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2909 int vprintk_deferred(const char *fmt, va_list args)
2913 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2914 defer_console_output();
2919 int printk_deferred(const char *fmt, ...)
2924 va_start(args, fmt);
2925 r = vprintk_deferred(fmt, args);
2932 * printk rate limiting, lifted from the networking subsystem.
2934 * This enforces a rate limit: not more than 10 kernel messages
2935 * every 5s to make a denial-of-service attack impossible.
2937 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2939 int __printk_ratelimit(const char *func)
2941 return ___ratelimit(&printk_ratelimit_state, func);
2943 EXPORT_SYMBOL(__printk_ratelimit);
2946 * printk_timed_ratelimit - caller-controlled printk ratelimiting
2947 * @caller_jiffies: pointer to caller's state
2948 * @interval_msecs: minimum interval between prints
2950 * printk_timed_ratelimit() returns true if more than @interval_msecs
2951 * milliseconds have elapsed since the last time printk_timed_ratelimit()
2954 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
2955 unsigned int interval_msecs)
2957 unsigned long elapsed = jiffies - *caller_jiffies;
2959 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
2962 *caller_jiffies = jiffies;
2965 EXPORT_SYMBOL(printk_timed_ratelimit);
2967 static DEFINE_SPINLOCK(dump_list_lock);
2968 static LIST_HEAD(dump_list);
2971 * kmsg_dump_register - register a kernel log dumper.
2972 * @dumper: pointer to the kmsg_dumper structure
2974 * Adds a kernel log dumper to the system. The dump callback in the
2975 * structure will be called when the kernel oopses or panics and must be
2976 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
2978 int kmsg_dump_register(struct kmsg_dumper *dumper)
2980 unsigned long flags;
2983 /* The dump callback needs to be set */
2987 spin_lock_irqsave(&dump_list_lock, flags);
2988 /* Don't allow registering multiple times */
2989 if (!dumper->registered) {
2990 dumper->registered = 1;
2991 list_add_tail_rcu(&dumper->list, &dump_list);
2994 spin_unlock_irqrestore(&dump_list_lock, flags);
2998 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3001 * kmsg_dump_unregister - unregister a kmsg dumper.
3002 * @dumper: pointer to the kmsg_dumper structure
3004 * Removes a dump device from the system. Returns zero on success and
3005 * %-EINVAL otherwise.
3007 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3009 unsigned long flags;
3012 spin_lock_irqsave(&dump_list_lock, flags);
3013 if (dumper->registered) {
3014 dumper->registered = 0;
3015 list_del_rcu(&dumper->list);
3018 spin_unlock_irqrestore(&dump_list_lock, flags);
3023 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3025 static bool always_kmsg_dump;
3026 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3029 * kmsg_dump - dump kernel log to kernel message dumpers.
3030 * @reason: the reason (oops, panic etc) for dumping
3032 * Call each of the registered dumper's dump() callback, which can
3033 * retrieve the kmsg records with kmsg_dump_get_line() or
3034 * kmsg_dump_get_buffer().
3036 void kmsg_dump(enum kmsg_dump_reason reason)
3038 struct kmsg_dumper *dumper;
3039 unsigned long flags;
3041 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3045 list_for_each_entry_rcu(dumper, &dump_list, list) {
3046 if (dumper->max_reason && reason > dumper->max_reason)
3049 /* initialize iterator with data about the stored records */
3050 dumper->active = true;
3052 logbuf_lock_irqsave(flags);
3053 dumper->cur_seq = clear_seq;
3054 dumper->cur_idx = clear_idx;
3055 dumper->next_seq = log_next_seq;
3056 dumper->next_idx = log_next_idx;
3057 logbuf_unlock_irqrestore(flags);
3059 /* invoke dumper which will iterate over records */
3060 dumper->dump(dumper, reason);
3062 /* reset iterator */
3063 dumper->active = false;
3069 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3070 * @dumper: registered kmsg dumper
3071 * @syslog: include the "<4>" prefixes
3072 * @line: buffer to copy the line to
3073 * @size: maximum size of the buffer
3074 * @len: length of line placed into buffer
3076 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3077 * record, and copy one record into the provided buffer.
3079 * Consecutive calls will return the next available record moving
3080 * towards the end of the buffer with the youngest messages.
3082 * A return value of FALSE indicates that there are no more records to
3085 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3087 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3088 char *line, size_t size, size_t *len)
3090 struct printk_log *msg;
3094 if (!dumper->active)
3097 if (dumper->cur_seq < log_first_seq) {
3098 /* messages are gone, move to first available one */
3099 dumper->cur_seq = log_first_seq;
3100 dumper->cur_idx = log_first_idx;
3104 if (dumper->cur_seq >= log_next_seq)
3107 msg = log_from_idx(dumper->cur_idx);
3108 l = msg_print_text(msg, syslog, line, size);
3110 dumper->cur_idx = log_next(dumper->cur_idx);
3120 * kmsg_dump_get_line - retrieve one kmsg log line
3121 * @dumper: registered kmsg dumper
3122 * @syslog: include the "<4>" prefixes
3123 * @line: buffer to copy the line to
3124 * @size: maximum size of the buffer
3125 * @len: length of line placed into buffer
3127 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3128 * record, and copy one record into the provided buffer.
3130 * Consecutive calls will return the next available record moving
3131 * towards the end of the buffer with the youngest messages.
3133 * A return value of FALSE indicates that there are no more records to
3136 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3137 char *line, size_t size, size_t *len)
3139 unsigned long flags;
3142 logbuf_lock_irqsave(flags);
3143 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3144 logbuf_unlock_irqrestore(flags);
3148 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3151 * kmsg_dump_get_buffer - copy kmsg log lines
3152 * @dumper: registered kmsg dumper
3153 * @syslog: include the "<4>" prefixes
3154 * @buf: buffer to copy the line to
3155 * @size: maximum size of the buffer
3156 * @len: length of line placed into buffer
3158 * Start at the end of the kmsg buffer and fill the provided buffer
3159 * with as many of the the *youngest* kmsg records that fit into it.
3160 * If the buffer is large enough, all available kmsg records will be
3161 * copied with a single call.
3163 * Consecutive calls will fill the buffer with the next block of
3164 * available older records, not including the earlier retrieved ones.
3166 * A return value of FALSE indicates that there are no more records to
3169 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3170 char *buf, size_t size, size_t *len)
3172 unsigned long flags;
3180 if (!dumper->active)
3183 logbuf_lock_irqsave(flags);
3184 if (dumper->cur_seq < log_first_seq) {
3185 /* messages are gone, move to first available one */
3186 dumper->cur_seq = log_first_seq;
3187 dumper->cur_idx = log_first_idx;
3191 if (dumper->cur_seq >= dumper->next_seq) {
3192 logbuf_unlock_irqrestore(flags);
3196 /* calculate length of entire buffer */
3197 seq = dumper->cur_seq;
3198 idx = dumper->cur_idx;
3199 while (seq < dumper->next_seq) {
3200 struct printk_log *msg = log_from_idx(idx);
3202 l += msg_print_text(msg, true, NULL, 0);
3203 idx = log_next(idx);
3207 /* move first record forward until length fits into the buffer */
3208 seq = dumper->cur_seq;
3209 idx = dumper->cur_idx;
3210 while (l > size && seq < dumper->next_seq) {
3211 struct printk_log *msg = log_from_idx(idx);
3213 l -= msg_print_text(msg, true, NULL, 0);
3214 idx = log_next(idx);
3218 /* last message in next interation */
3223 while (seq < dumper->next_seq) {
3224 struct printk_log *msg = log_from_idx(idx);
3226 l += msg_print_text(msg, syslog, buf + l, size - l);
3227 idx = log_next(idx);
3231 dumper->next_seq = next_seq;
3232 dumper->next_idx = next_idx;
3234 logbuf_unlock_irqrestore(flags);
3240 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3243 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3244 * @dumper: registered kmsg dumper
3246 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3247 * kmsg_dump_get_buffer() can be called again and used multiple
3248 * times within the same dumper.dump() callback.
3250 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3252 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3254 dumper->cur_seq = clear_seq;
3255 dumper->cur_idx = clear_idx;
3256 dumper->next_seq = log_next_seq;
3257 dumper->next_idx = log_next_idx;
3261 * kmsg_dump_rewind - reset the interator
3262 * @dumper: registered kmsg dumper
3264 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3265 * kmsg_dump_get_buffer() can be called again and used multiple
3266 * times within the same dumper.dump() callback.
3268 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3270 unsigned long flags;
3272 logbuf_lock_irqsave(flags);
3273 kmsg_dump_rewind_nolock(dumper);
3274 logbuf_unlock_irqrestore(flags);
3276 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);