2 * linux/kernel/printk.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * Modified to make sys_syslog() more flexible: added commands to
7 * return the last 4k of kernel messages, regardless of whether
8 * they've been read or not. Added option to suppress kernel printk's
9 * to the console. Added hook for sending the console messages
10 * elsewhere, in preparation for a serial line console (someday).
12 * Modified for sysctl support, 1/8/97, Chris Horn.
13 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
14 * manfred@colorfullife.com
15 * Rewrote bits to get rid of console_lock
16 * 01Mar01 Andrew Morton
19 #include <linux/kernel.h>
21 #include <linux/tty.h>
22 #include <linux/tty_driver.h>
23 #include <linux/console.h>
24 #include <linux/init.h>
25 #include <linux/jiffies.h>
26 #include <linux/nmi.h>
27 #include <linux/module.h>
28 #include <linux/moduleparam.h>
29 #include <linux/delay.h>
30 #include <linux/smp.h>
31 #include <linux/security.h>
32 #include <linux/bootmem.h>
33 #include <linux/memblock.h>
34 #include <linux/syscalls.h>
35 #include <linux/crash_core.h>
36 #include <linux/kdb.h>
37 #include <linux/ratelimit.h>
38 #include <linux/kmsg_dump.h>
39 #include <linux/syslog.h>
40 #include <linux/cpu.h>
41 #include <linux/notifier.h>
42 #include <linux/rculist.h>
43 #include <linux/poll.h>
44 #include <linux/irq_work.h>
45 #include <linux/utsname.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>
51 #include <linux/kexec.h>
53 #include <linux/uaccess.h>
54 #include <asm/sections.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 */
71 * Low level drivers may need that to know if they can schedule in
72 * their unblank() callback or not. So let's export it.
75 EXPORT_SYMBOL(oops_in_progress);
78 * console_sem protects the console_drivers list, and also
79 * provides serialisation for access to the entire console
82 static DEFINE_SEMAPHORE(console_sem);
83 struct console *console_drivers;
84 EXPORT_SYMBOL_GPL(console_drivers);
87 static struct lockdep_map console_lock_dep_map = {
88 .name = "console_lock"
92 enum devkmsg_log_bits {
93 __DEVKMSG_LOG_BIT_ON = 0,
94 __DEVKMSG_LOG_BIT_OFF,
95 __DEVKMSG_LOG_BIT_LOCK,
98 enum devkmsg_log_masks {
99 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
100 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
101 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
104 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
105 #define DEVKMSG_LOG_MASK_DEFAULT 0
107 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
109 static int __control_devkmsg(char *str)
114 if (!strncmp(str, "on", 2)) {
115 devkmsg_log = DEVKMSG_LOG_MASK_ON;
117 } else if (!strncmp(str, "off", 3)) {
118 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
120 } else if (!strncmp(str, "ratelimit", 9)) {
121 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
127 static int __init control_devkmsg(char *str)
129 if (__control_devkmsg(str) < 0)
133 * Set sysctl string accordingly:
135 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
136 strcpy(devkmsg_log_str, "on");
137 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
138 strcpy(devkmsg_log_str, "off");
139 /* else "ratelimit" which is set by default. */
142 * Sysctl cannot change it anymore. The kernel command line setting of
143 * this parameter is to force the setting to be permanent throughout the
144 * runtime of the system. This is a precation measure against userspace
145 * trying to be a smarta** and attempting to change it up on us.
147 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
151 __setup("printk.devkmsg=", control_devkmsg);
153 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
155 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
156 void __user *buffer, size_t *lenp, loff_t *ppos)
158 char old_str[DEVKMSG_STR_MAX_SIZE];
163 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
167 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
170 err = proc_dostring(table, write, buffer, lenp, ppos);
175 err = __control_devkmsg(devkmsg_log_str);
178 * Do not accept an unknown string OR a known string with
181 if (err < 0 || (err + 1 != *lenp)) {
183 /* ... and restore old setting. */
185 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
195 * Number of registered extended console drivers.
197 * If extended consoles are present, in-kernel cont reassembly is disabled
198 * and each fragment is stored as a separate log entry with proper
199 * continuation flag so that every emitted message has full metadata. This
200 * doesn't change the result for regular consoles or /proc/kmsg. For
201 * /dev/kmsg, as long as the reader concatenates messages according to
202 * consecutive continuation flags, the end result should be the same too.
204 static int nr_ext_console_drivers;
207 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
208 * macros instead of functions so that _RET_IP_ contains useful information.
210 #define down_console_sem() do { \
212 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
215 static int __down_trylock_console_sem(unsigned long ip)
221 * Here and in __up_console_sem() we need to be in safe mode,
222 * because spindump/WARN/etc from under console ->lock will
223 * deadlock in printk()->down_trylock_console_sem() otherwise.
225 printk_safe_enter_irqsave(flags);
226 lock_failed = down_trylock(&console_sem);
227 printk_safe_exit_irqrestore(flags);
231 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
234 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
236 static void __up_console_sem(unsigned long ip)
240 mutex_release(&console_lock_dep_map, 1, ip);
242 printk_safe_enter_irqsave(flags);
244 printk_safe_exit_irqrestore(flags);
246 #define up_console_sem() __up_console_sem(_RET_IP_)
249 * This is used for debugging the mess that is the VT code by
250 * keeping track if we have the console semaphore held. It's
251 * definitely not the perfect debug tool (we don't know if _WE_
252 * hold it and are racing, but it helps tracking those weird code
253 * paths in the console code where we end up in places I want
254 * locked without the console sempahore held).
256 static int console_locked, console_suspended;
259 * If exclusive_console is non-NULL then only this console is to be printed to.
261 static struct console *exclusive_console;
264 * Array of consoles built from command line options (console=)
267 #define MAX_CMDLINECONSOLES 8
269 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
271 static int preferred_console = -1;
272 int console_set_on_cmdline;
273 EXPORT_SYMBOL(console_set_on_cmdline);
275 /* Flag: console code may call schedule() */
276 static int console_may_schedule;
278 enum con_msg_format_flags {
279 MSG_FORMAT_DEFAULT = 0,
280 MSG_FORMAT_SYSLOG = (1 << 0),
283 static int console_msg_format = MSG_FORMAT_DEFAULT;
286 * The printk log buffer consists of a chain of concatenated variable
287 * length records. Every record starts with a record header, containing
288 * the overall length of the record.
290 * The heads to the first and last entry in the buffer, as well as the
291 * sequence numbers of these entries are maintained when messages are
294 * If the heads indicate available messages, the length in the header
295 * tells the start next message. A length == 0 for the next message
296 * indicates a wrap-around to the beginning of the buffer.
298 * Every record carries the monotonic timestamp in microseconds, as well as
299 * the standard userspace syslog level and syslog facility. The usual
300 * kernel messages use LOG_KERN; userspace-injected messages always carry
301 * a matching syslog facility, by default LOG_USER. The origin of every
302 * message can be reliably determined that way.
304 * The human readable log message directly follows the message header. The
305 * length of the message text is stored in the header, the stored message
308 * Optionally, a message can carry a dictionary of properties (key/value pairs),
309 * to provide userspace with a machine-readable message context.
311 * Examples for well-defined, commonly used property names are:
312 * DEVICE=b12:8 device identifier
316 * +sound:card0 subsystem:devname
317 * SUBSYSTEM=pci driver-core subsystem name
319 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
320 * follows directly after a '=' character. Every property is terminated by
321 * a '\0' character. The last property is not terminated.
323 * Example of a message structure:
324 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
325 * 0008 34 00 record is 52 bytes long
326 * 000a 0b 00 text is 11 bytes long
327 * 000c 1f 00 dictionary is 23 bytes long
328 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
329 * 0010 69 74 27 73 20 61 20 6c "it's a l"
331 * 001b 44 45 56 49 43 "DEVIC"
332 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
333 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
335 * 0032 00 00 00 padding to next message header
337 * The 'struct printk_log' buffer header must never be directly exported to
338 * userspace, it is a kernel-private implementation detail that might
339 * need to be changed in the future, when the requirements change.
341 * /dev/kmsg exports the structured data in the following line format:
342 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
344 * Users of the export format should ignore possible additional values
345 * separated by ',', and find the message after the ';' character.
347 * The optional key/value pairs are attached as continuation lines starting
348 * with a space character and terminated by a newline. All possible
349 * non-prinatable characters are escaped in the "\xff" notation.
353 LOG_NOCONS = 1, /* already flushed, do not print to console */
354 LOG_NEWLINE = 2, /* text ended with a newline */
355 LOG_PREFIX = 4, /* text started with a prefix */
356 LOG_CONT = 8, /* text is a fragment of a continuation line */
360 u64 ts_nsec; /* timestamp in nanoseconds */
361 u16 len; /* length of entire record */
362 u16 text_len; /* length of text buffer */
363 u16 dict_len; /* length of dictionary buffer */
364 u8 facility; /* syslog facility */
365 u8 flags:5; /* internal record flags */
366 u8 level:3; /* syslog level */
368 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
369 __packed __aligned(4)
374 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
375 * within the scheduler's rq lock. It must be released before calling
376 * console_unlock() or anything else that might wake up a process.
378 DEFINE_RAW_SPINLOCK(logbuf_lock);
381 * Helper macros to lock/unlock logbuf_lock and switch between
382 * printk-safe/unsafe modes.
384 #define logbuf_lock_irq() \
386 printk_safe_enter_irq(); \
387 raw_spin_lock(&logbuf_lock); \
390 #define logbuf_unlock_irq() \
392 raw_spin_unlock(&logbuf_lock); \
393 printk_safe_exit_irq(); \
396 #define logbuf_lock_irqsave(flags) \
398 printk_safe_enter_irqsave(flags); \
399 raw_spin_lock(&logbuf_lock); \
402 #define logbuf_unlock_irqrestore(flags) \
404 raw_spin_unlock(&logbuf_lock); \
405 printk_safe_exit_irqrestore(flags); \
409 DECLARE_WAIT_QUEUE_HEAD(log_wait);
410 /* the next printk record to read by syslog(READ) or /proc/kmsg */
411 static u64 syslog_seq;
412 static u32 syslog_idx;
413 static size_t syslog_partial;
415 /* index and sequence number of the first record stored in the buffer */
416 static u64 log_first_seq;
417 static u32 log_first_idx;
419 /* index and sequence number of the next record to store in the buffer */
420 static u64 log_next_seq;
421 static u32 log_next_idx;
423 /* the next printk record to write to the console */
424 static u64 console_seq;
425 static u32 console_idx;
427 /* the next printk record to read after the last 'clear' command */
428 static u64 clear_seq;
429 static u32 clear_idx;
431 #define PREFIX_MAX 32
432 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
434 #define LOG_LEVEL(v) ((v) & 0x07)
435 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
438 #define LOG_ALIGN __alignof__(struct printk_log)
439 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
440 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
441 static char *log_buf = __log_buf;
442 static u32 log_buf_len = __LOG_BUF_LEN;
444 /* Return log buffer address */
445 char *log_buf_addr_get(void)
450 /* Return log buffer size */
451 u32 log_buf_len_get(void)
456 /* human readable text of the record */
457 static char *log_text(const struct printk_log *msg)
459 return (char *)msg + sizeof(struct printk_log);
462 /* optional key/value pair dictionary attached to the record */
463 static char *log_dict(const struct printk_log *msg)
465 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
468 /* get record by index; idx must point to valid msg */
469 static struct printk_log *log_from_idx(u32 idx)
471 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
474 * A length == 0 record is the end of buffer marker. Wrap around and
475 * read the message at the start of the buffer.
478 return (struct printk_log *)log_buf;
482 /* get next record; idx must point to valid msg */
483 static u32 log_next(u32 idx)
485 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
487 /* length == 0 indicates the end of the buffer; wrap */
489 * A length == 0 record is the end of buffer marker. Wrap around and
490 * read the message at the start of the buffer as *this* one, and
491 * return the one after that.
494 msg = (struct printk_log *)log_buf;
497 return idx + msg->len;
501 * Check whether there is enough free space for the given message.
503 * The same values of first_idx and next_idx mean that the buffer
504 * is either empty or full.
506 * If the buffer is empty, we must respect the position of the indexes.
507 * They cannot be reset to the beginning of the buffer.
509 static int logbuf_has_space(u32 msg_size, bool empty)
513 if (log_next_idx > log_first_idx || empty)
514 free = max(log_buf_len - log_next_idx, log_first_idx);
516 free = log_first_idx - log_next_idx;
519 * We need space also for an empty header that signalizes wrapping
522 return free >= msg_size + sizeof(struct printk_log);
525 static int log_make_free_space(u32 msg_size)
527 while (log_first_seq < log_next_seq &&
528 !logbuf_has_space(msg_size, false)) {
529 /* drop old messages until we have enough contiguous space */
530 log_first_idx = log_next(log_first_idx);
534 if (clear_seq < log_first_seq) {
535 clear_seq = log_first_seq;
536 clear_idx = log_first_idx;
539 /* sequence numbers are equal, so the log buffer is empty */
540 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
546 /* compute the message size including the padding bytes */
547 static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
551 size = sizeof(struct printk_log) + text_len + dict_len;
552 *pad_len = (-size) & (LOG_ALIGN - 1);
559 * Define how much of the log buffer we could take at maximum. The value
560 * must be greater than two. Note that only half of the buffer is available
561 * when the index points to the middle.
563 #define MAX_LOG_TAKE_PART 4
564 static const char trunc_msg[] = "<truncated>";
566 static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
567 u16 *dict_len, u32 *pad_len)
570 * The message should not take the whole buffer. Otherwise, it might
571 * get removed too soon.
573 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
574 if (*text_len > max_text_len)
575 *text_len = max_text_len;
576 /* enable the warning message */
577 *trunc_msg_len = strlen(trunc_msg);
578 /* disable the "dict" completely */
580 /* compute the size again, count also the warning message */
581 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
584 /* insert record into the buffer, discard old ones, update heads */
585 static int log_store(int facility, int level,
586 enum log_flags flags, u64 ts_nsec,
587 const char *dict, u16 dict_len,
588 const char *text, u16 text_len)
590 struct printk_log *msg;
592 u16 trunc_msg_len = 0;
594 /* number of '\0' padding bytes to next message */
595 size = msg_used_size(text_len, dict_len, &pad_len);
597 if (log_make_free_space(size)) {
598 /* truncate the message if it is too long for empty buffer */
599 size = truncate_msg(&text_len, &trunc_msg_len,
600 &dict_len, &pad_len);
601 /* survive when the log buffer is too small for trunc_msg */
602 if (log_make_free_space(size))
606 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
608 * This message + an additional empty header does not fit
609 * at the end of the buffer. Add an empty header with len == 0
610 * to signify a wrap around.
612 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
617 msg = (struct printk_log *)(log_buf + log_next_idx);
618 memcpy(log_text(msg), text, text_len);
619 msg->text_len = text_len;
621 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
622 msg->text_len += trunc_msg_len;
624 memcpy(log_dict(msg), dict, dict_len);
625 msg->dict_len = dict_len;
626 msg->facility = facility;
627 msg->level = level & 7;
628 msg->flags = flags & 0x1f;
630 msg->ts_nsec = ts_nsec;
632 msg->ts_nsec = local_clock();
633 memset(log_dict(msg) + dict_len, 0, pad_len);
637 log_next_idx += msg->len;
640 return msg->text_len;
643 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
645 static int syslog_action_restricted(int type)
650 * Unless restricted, we allow "read all" and "get buffer size"
653 return type != SYSLOG_ACTION_READ_ALL &&
654 type != SYSLOG_ACTION_SIZE_BUFFER;
657 static int check_syslog_permissions(int type, int source)
660 * If this is from /proc/kmsg and we've already opened it, then we've
661 * already done the capabilities checks at open time.
663 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
666 if (syslog_action_restricted(type)) {
667 if (capable(CAP_SYSLOG))
670 * For historical reasons, accept CAP_SYS_ADMIN too, with
673 if (capable(CAP_SYS_ADMIN)) {
674 pr_warn_once("%s (%d): Attempt to access syslog with "
675 "CAP_SYS_ADMIN but no CAP_SYSLOG "
677 current->comm, task_pid_nr(current));
683 return security_syslog(type);
686 static void append_char(char **pp, char *e, char c)
692 static ssize_t msg_print_ext_header(char *buf, size_t size,
693 struct printk_log *msg, u64 seq)
695 u64 ts_usec = msg->ts_nsec;
697 do_div(ts_usec, 1000);
699 return scnprintf(buf, size, "%u,%llu,%llu,%c;",
700 (msg->facility << 3) | msg->level, seq, ts_usec,
701 msg->flags & LOG_CONT ? 'c' : '-');
704 static ssize_t msg_print_ext_body(char *buf, size_t size,
705 char *dict, size_t dict_len,
706 char *text, size_t text_len)
708 char *p = buf, *e = buf + size;
711 /* escape non-printable characters */
712 for (i = 0; i < text_len; i++) {
713 unsigned char c = text[i];
715 if (c < ' ' || c >= 127 || c == '\\')
716 p += scnprintf(p, e - p, "\\x%02x", c);
718 append_char(&p, e, c);
720 append_char(&p, e, '\n');
725 for (i = 0; i < dict_len; i++) {
726 unsigned char c = dict[i];
729 append_char(&p, e, ' ');
734 append_char(&p, e, '\n');
739 if (c < ' ' || c >= 127 || c == '\\') {
740 p += scnprintf(p, e - p, "\\x%02x", c);
744 append_char(&p, e, c);
746 append_char(&p, e, '\n');
752 /* /dev/kmsg - userspace message inject/listen interface */
753 struct devkmsg_user {
756 struct ratelimit_state rs;
758 char buf[CONSOLE_EXT_LOG_MAX];
761 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
764 int level = default_message_loglevel;
765 int facility = 1; /* LOG_USER */
766 struct file *file = iocb->ki_filp;
767 struct devkmsg_user *user = file->private_data;
768 size_t len = iov_iter_count(from);
771 if (!user || len > LOG_LINE_MAX)
774 /* Ignore when user logging is disabled. */
775 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
778 /* Ratelimit when not explicitly enabled. */
779 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
780 if (!___ratelimit(&user->rs, current->comm))
784 buf = kmalloc(len+1, GFP_KERNEL);
789 if (!copy_from_iter_full(buf, len, from)) {
795 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
796 * the decimal value represents 32bit, the lower 3 bit are the log
797 * level, the rest are the log facility.
799 * If no prefix or no userspace facility is specified, we
800 * enforce LOG_USER, to be able to reliably distinguish
801 * kernel-generated messages from userspace-injected ones.
804 if (line[0] == '<') {
808 u = simple_strtoul(line + 1, &endp, 10);
809 if (endp && endp[0] == '>') {
810 level = LOG_LEVEL(u);
811 if (LOG_FACILITY(u) != 0)
812 facility = LOG_FACILITY(u);
819 printk_emit(facility, level, NULL, 0, "%s", line);
824 static ssize_t devkmsg_read(struct file *file, char __user *buf,
825 size_t count, loff_t *ppos)
827 struct devkmsg_user *user = file->private_data;
828 struct printk_log *msg;
835 ret = mutex_lock_interruptible(&user->lock);
840 while (user->seq == log_next_seq) {
841 if (file->f_flags & O_NONBLOCK) {
848 ret = wait_event_interruptible(log_wait,
849 user->seq != log_next_seq);
855 if (user->seq < log_first_seq) {
856 /* our last seen message is gone, return error and reset */
857 user->idx = log_first_idx;
858 user->seq = log_first_seq;
864 msg = log_from_idx(user->idx);
865 len = msg_print_ext_header(user->buf, sizeof(user->buf),
867 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
868 log_dict(msg), msg->dict_len,
869 log_text(msg), msg->text_len);
871 user->idx = log_next(user->idx);
880 if (copy_to_user(buf, user->buf, len)) {
886 mutex_unlock(&user->lock);
890 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
892 struct devkmsg_user *user = file->private_data;
903 /* the first record */
904 user->idx = log_first_idx;
905 user->seq = log_first_seq;
909 * The first record after the last SYSLOG_ACTION_CLEAR,
910 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
911 * changes no global state, and does not clear anything.
913 user->idx = clear_idx;
914 user->seq = clear_seq;
917 /* after the last record */
918 user->idx = log_next_idx;
919 user->seq = log_next_seq;
928 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
930 struct devkmsg_user *user = file->private_data;
934 return POLLERR|POLLNVAL;
936 poll_wait(file, &log_wait, wait);
939 if (user->seq < log_next_seq) {
940 /* return error when data has vanished underneath us */
941 if (user->seq < log_first_seq)
942 ret = POLLIN|POLLRDNORM|POLLERR|POLLPRI;
944 ret = POLLIN|POLLRDNORM;
951 static int devkmsg_open(struct inode *inode, struct file *file)
953 struct devkmsg_user *user;
956 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
959 /* write-only does not need any file context */
960 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
961 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
967 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
971 ratelimit_default_init(&user->rs);
972 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
974 mutex_init(&user->lock);
977 user->idx = log_first_idx;
978 user->seq = log_first_seq;
981 file->private_data = user;
985 static int devkmsg_release(struct inode *inode, struct file *file)
987 struct devkmsg_user *user = file->private_data;
992 ratelimit_state_exit(&user->rs);
994 mutex_destroy(&user->lock);
999 const struct file_operations kmsg_fops = {
1000 .open = devkmsg_open,
1001 .read = devkmsg_read,
1002 .write_iter = devkmsg_write,
1003 .llseek = devkmsg_llseek,
1004 .poll = devkmsg_poll,
1005 .release = devkmsg_release,
1008 #ifdef CONFIG_CRASH_CORE
1010 * This appends the listed symbols to /proc/vmcore
1012 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1013 * obtain access to symbols that are otherwise very difficult to locate. These
1014 * symbols are specifically used so that utilities can access and extract the
1015 * dmesg log from a vmcore file after a crash.
1017 void log_buf_vmcoreinfo_setup(void)
1019 VMCOREINFO_SYMBOL(log_buf);
1020 VMCOREINFO_SYMBOL(log_buf_len);
1021 VMCOREINFO_SYMBOL(log_first_idx);
1022 VMCOREINFO_SYMBOL(clear_idx);
1023 VMCOREINFO_SYMBOL(log_next_idx);
1025 * Export struct printk_log size and field offsets. User space tools can
1026 * parse it and detect any changes to structure down the line.
1028 VMCOREINFO_STRUCT_SIZE(printk_log);
1029 VMCOREINFO_OFFSET(printk_log, ts_nsec);
1030 VMCOREINFO_OFFSET(printk_log, len);
1031 VMCOREINFO_OFFSET(printk_log, text_len);
1032 VMCOREINFO_OFFSET(printk_log, dict_len);
1036 /* requested log_buf_len from kernel cmdline */
1037 static unsigned long __initdata new_log_buf_len;
1039 /* we practice scaling the ring buffer by powers of 2 */
1040 static void __init log_buf_len_update(unsigned size)
1043 size = roundup_pow_of_two(size);
1044 if (size > log_buf_len)
1045 new_log_buf_len = 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)
1051 unsigned size = memparse(str, &str);
1053 log_buf_len_update(size);
1057 early_param("log_buf_len", log_buf_len_setup);
1060 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1062 static void __init log_buf_add_cpu(void)
1064 unsigned int cpu_extra;
1067 * archs should set up cpu_possible_bits properly with
1068 * set_cpu_possible() after setup_arch() but just in
1069 * case lets ensure this is valid.
1071 if (num_possible_cpus() == 1)
1074 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1076 /* by default this will only continue through for large > 64 CPUs */
1077 if (cpu_extra <= __LOG_BUF_LEN / 2)
1080 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1081 __LOG_CPU_MAX_BUF_LEN);
1082 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1084 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1086 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1088 #else /* !CONFIG_SMP */
1089 static inline void log_buf_add_cpu(void) {}
1090 #endif /* CONFIG_SMP */
1092 void __init setup_log_buf(int early)
1094 unsigned long flags;
1098 if (log_buf != __log_buf)
1101 if (!early && !new_log_buf_len)
1104 if (!new_log_buf_len)
1109 memblock_virt_alloc(new_log_buf_len, LOG_ALIGN);
1111 new_log_buf = memblock_virt_alloc_nopanic(new_log_buf_len,
1115 if (unlikely(!new_log_buf)) {
1116 pr_err("log_buf_len: %ld bytes not available\n",
1121 logbuf_lock_irqsave(flags);
1122 log_buf_len = new_log_buf_len;
1123 log_buf = new_log_buf;
1124 new_log_buf_len = 0;
1125 free = __LOG_BUF_LEN - log_next_idx;
1126 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1127 logbuf_unlock_irqrestore(flags);
1129 pr_info("log_buf_len: %d bytes\n", log_buf_len);
1130 pr_info("early log buf free: %d(%d%%)\n",
1131 free, (free * 100) / __LOG_BUF_LEN);
1134 static bool __read_mostly ignore_loglevel;
1136 static int __init ignore_loglevel_setup(char *str)
1138 ignore_loglevel = true;
1139 pr_info("debug: ignoring loglevel setting.\n");
1144 early_param("ignore_loglevel", ignore_loglevel_setup);
1145 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1146 MODULE_PARM_DESC(ignore_loglevel,
1147 "ignore loglevel setting (prints all kernel messages to the console)");
1149 static bool suppress_message_printing(int level)
1151 return (level >= console_loglevel && !ignore_loglevel);
1154 #ifdef CONFIG_BOOT_PRINTK_DELAY
1156 static int boot_delay; /* msecs delay after each printk during bootup */
1157 static unsigned long long loops_per_msec; /* based on boot_delay */
1159 static int __init boot_delay_setup(char *str)
1163 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1164 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1166 get_option(&str, &boot_delay);
1167 if (boot_delay > 10 * 1000)
1170 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1171 "HZ: %d, loops_per_msec: %llu\n",
1172 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1175 early_param("boot_delay", boot_delay_setup);
1177 static void boot_delay_msec(int level)
1179 unsigned long long k;
1180 unsigned long timeout;
1182 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1183 || suppress_message_printing(level)) {
1187 k = (unsigned long long)loops_per_msec * boot_delay;
1189 timeout = jiffies + msecs_to_jiffies(boot_delay);
1194 * use (volatile) jiffies to prevent
1195 * compiler reduction; loop termination via jiffies
1196 * is secondary and may or may not happen.
1198 if (time_after(jiffies, timeout))
1200 touch_nmi_watchdog();
1204 static inline void boot_delay_msec(int level)
1209 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1210 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1212 static size_t print_time(u64 ts, char *buf)
1214 unsigned long rem_nsec;
1219 rem_nsec = do_div(ts, 1000000000);
1222 return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
1224 return sprintf(buf, "[%5lu.%06lu] ",
1225 (unsigned long)ts, rem_nsec / 1000);
1228 static size_t print_prefix(const struct printk_log *msg, bool syslog, char *buf)
1231 unsigned int prefix = (msg->facility << 3) | msg->level;
1235 len += sprintf(buf, "<%u>", prefix);
1240 else if (prefix > 99)
1242 else if (prefix > 9)
1247 len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
1251 static size_t msg_print_text(const struct printk_log *msg, bool syslog, char *buf, size_t size)
1253 const char *text = log_text(msg);
1254 size_t text_size = msg->text_len;
1258 const char *next = memchr(text, '\n', text_size);
1262 text_len = next - text;
1264 text_size -= next - text;
1266 text_len = text_size;
1270 if (print_prefix(msg, syslog, NULL) +
1271 text_len + 1 >= size - len)
1274 len += print_prefix(msg, syslog, buf + len);
1275 memcpy(buf + len, text, text_len);
1279 /* SYSLOG_ACTION_* buffer size only calculation */
1280 len += print_prefix(msg, syslog, NULL);
1291 static int syslog_print(char __user *buf, int size)
1294 struct printk_log *msg;
1297 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1306 if (syslog_seq < log_first_seq) {
1307 /* messages are gone, move to first one */
1308 syslog_seq = log_first_seq;
1309 syslog_idx = log_first_idx;
1312 if (syslog_seq == log_next_seq) {
1313 logbuf_unlock_irq();
1317 skip = syslog_partial;
1318 msg = log_from_idx(syslog_idx);
1319 n = msg_print_text(msg, true, text, LOG_LINE_MAX + PREFIX_MAX);
1320 if (n - syslog_partial <= size) {
1321 /* message fits into buffer, move forward */
1322 syslog_idx = log_next(syslog_idx);
1324 n -= syslog_partial;
1327 /* partial read(), remember position */
1329 syslog_partial += n;
1332 logbuf_unlock_irq();
1337 if (copy_to_user(buf, text + skip, n)) {
1352 static int syslog_print_all(char __user *buf, int size, bool clear)
1357 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1368 * Find first record that fits, including all following records,
1369 * into the user-provided buffer for this dump.
1373 while (seq < log_next_seq) {
1374 struct printk_log *msg = log_from_idx(idx);
1376 len += msg_print_text(msg, true, NULL, 0);
1377 idx = log_next(idx);
1381 /* move first record forward until length fits into the buffer */
1384 while (len > size && seq < log_next_seq) {
1385 struct printk_log *msg = log_from_idx(idx);
1387 len -= msg_print_text(msg, true, NULL, 0);
1388 idx = log_next(idx);
1392 /* last message fitting into this dump */
1393 next_seq = log_next_seq;
1396 while (len >= 0 && seq < next_seq) {
1397 struct printk_log *msg = log_from_idx(idx);
1400 textlen = msg_print_text(msg, true, text,
1401 LOG_LINE_MAX + PREFIX_MAX);
1406 idx = log_next(idx);
1409 logbuf_unlock_irq();
1410 if (copy_to_user(buf + len, text, textlen))
1416 if (seq < log_first_seq) {
1417 /* messages are gone, move to next one */
1418 seq = log_first_seq;
1419 idx = log_first_idx;
1425 clear_seq = log_next_seq;
1426 clear_idx = log_next_idx;
1428 logbuf_unlock_irq();
1434 int do_syslog(int type, char __user *buf, int len, int source)
1437 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1440 error = check_syslog_permissions(type, source);
1445 case SYSLOG_ACTION_CLOSE: /* Close log */
1447 case SYSLOG_ACTION_OPEN: /* Open log */
1449 case SYSLOG_ACTION_READ: /* Read from log */
1450 if (!buf || len < 0)
1454 if (!access_ok(VERIFY_WRITE, buf, len))
1456 error = wait_event_interruptible(log_wait,
1457 syslog_seq != log_next_seq);
1460 error = syslog_print(buf, len);
1462 /* Read/clear last kernel messages */
1463 case SYSLOG_ACTION_READ_CLEAR:
1466 /* Read last kernel messages */
1467 case SYSLOG_ACTION_READ_ALL:
1468 if (!buf || len < 0)
1472 if (!access_ok(VERIFY_WRITE, buf, len))
1474 error = syslog_print_all(buf, len, clear);
1476 /* Clear ring buffer */
1477 case SYSLOG_ACTION_CLEAR:
1478 syslog_print_all(NULL, 0, true);
1480 /* Disable logging to console */
1481 case SYSLOG_ACTION_CONSOLE_OFF:
1482 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1483 saved_console_loglevel = console_loglevel;
1484 console_loglevel = minimum_console_loglevel;
1486 /* Enable logging to console */
1487 case SYSLOG_ACTION_CONSOLE_ON:
1488 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1489 console_loglevel = saved_console_loglevel;
1490 saved_console_loglevel = LOGLEVEL_DEFAULT;
1493 /* Set level of messages printed to console */
1494 case SYSLOG_ACTION_CONSOLE_LEVEL:
1495 if (len < 1 || len > 8)
1497 if (len < minimum_console_loglevel)
1498 len = minimum_console_loglevel;
1499 console_loglevel = len;
1500 /* Implicitly re-enable logging to console */
1501 saved_console_loglevel = LOGLEVEL_DEFAULT;
1503 /* Number of chars in the log buffer */
1504 case SYSLOG_ACTION_SIZE_UNREAD:
1506 if (syslog_seq < log_first_seq) {
1507 /* messages are gone, move to first one */
1508 syslog_seq = log_first_seq;
1509 syslog_idx = log_first_idx;
1512 if (source == SYSLOG_FROM_PROC) {
1514 * Short-cut for poll(/"proc/kmsg") which simply checks
1515 * for pending data, not the size; return the count of
1516 * records, not the length.
1518 error = log_next_seq - syslog_seq;
1520 u64 seq = syslog_seq;
1521 u32 idx = syslog_idx;
1523 while (seq < log_next_seq) {
1524 struct printk_log *msg = log_from_idx(idx);
1526 error += msg_print_text(msg, true, NULL, 0);
1527 idx = log_next(idx);
1530 error -= syslog_partial;
1532 logbuf_unlock_irq();
1534 /* Size of the log buffer */
1535 case SYSLOG_ACTION_SIZE_BUFFER:
1536 error = log_buf_len;
1546 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1548 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1552 * Special console_lock variants that help to reduce the risk of soft-lockups.
1553 * They allow to pass console_lock to another printk() call using a busy wait.
1556 #ifdef CONFIG_LOCKDEP
1557 static struct lockdep_map console_owner_dep_map = {
1558 .name = "console_owner"
1562 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1563 static struct task_struct *console_owner;
1564 static bool console_waiter;
1567 * console_lock_spinning_enable - mark beginning of code where another
1568 * thread might safely busy wait
1570 * This basically converts console_lock into a spinlock. This marks
1571 * the section where the console_lock owner can not sleep, because
1572 * there may be a waiter spinning (like a spinlock). Also it must be
1573 * ready to hand over the lock at the end of the section.
1575 static void console_lock_spinning_enable(void)
1577 raw_spin_lock(&console_owner_lock);
1578 console_owner = current;
1579 raw_spin_unlock(&console_owner_lock);
1581 /* The waiter may spin on us after setting console_owner */
1582 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1586 * console_lock_spinning_disable_and_check - mark end of code where another
1587 * thread was able to busy wait and check if there is a waiter
1589 * This is called at the end of the section where spinning is allowed.
1590 * It has two functions. First, it is a signal that it is no longer
1591 * safe to start busy waiting for the lock. Second, it checks if
1592 * there is a busy waiter and passes the lock rights to her.
1594 * Important: Callers lose the lock if there was a busy waiter.
1595 * They must not touch items synchronized by console_lock
1598 * Return: 1 if the lock rights were passed, 0 otherwise.
1600 static int console_lock_spinning_disable_and_check(void)
1604 raw_spin_lock(&console_owner_lock);
1605 waiter = READ_ONCE(console_waiter);
1606 console_owner = NULL;
1607 raw_spin_unlock(&console_owner_lock);
1610 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1614 /* The waiter is now free to continue */
1615 WRITE_ONCE(console_waiter, false);
1617 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1620 * Hand off console_lock to waiter. The waiter will perform
1621 * the up(). After this, the waiter is the console_lock owner.
1623 mutex_release(&console_lock_dep_map, 1, _THIS_IP_);
1628 * console_trylock_spinning - try to get console_lock by busy waiting
1630 * This allows to busy wait for the console_lock when the current
1631 * owner is running in specially marked sections. It means that
1632 * the current owner is running and cannot reschedule until it
1633 * is ready to lose the lock.
1635 * Return: 1 if we got the lock, 0 othrewise
1637 static int console_trylock_spinning(void)
1639 struct task_struct *owner = NULL;
1642 unsigned long flags;
1644 if (console_trylock())
1647 printk_safe_enter_irqsave(flags);
1649 raw_spin_lock(&console_owner_lock);
1650 owner = READ_ONCE(console_owner);
1651 waiter = READ_ONCE(console_waiter);
1652 if (!waiter && owner && owner != current) {
1653 WRITE_ONCE(console_waiter, true);
1656 raw_spin_unlock(&console_owner_lock);
1659 * If there is an active printk() writing to the
1660 * consoles, instead of having it write our data too,
1661 * see if we can offload that load from the active
1662 * printer, and do some printing ourselves.
1663 * Go into a spin only if there isn't already a waiter
1664 * spinning, and there is an active printer, and
1665 * that active printer isn't us (recursive printk?).
1668 printk_safe_exit_irqrestore(flags);
1672 /* We spin waiting for the owner to release us */
1673 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1674 /* Owner will clear console_waiter on hand off */
1675 while (READ_ONCE(console_waiter))
1677 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1679 printk_safe_exit_irqrestore(flags);
1681 * The owner passed the console lock to us.
1682 * Since we did not spin on console lock, annotate
1683 * this as a trylock. Otherwise lockdep will
1686 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1692 * Call the console drivers, asking them to write out
1693 * log_buf[start] to log_buf[end - 1].
1694 * The console_lock must be held.
1696 static void call_console_drivers(const char *ext_text, size_t ext_len,
1697 const char *text, size_t len)
1699 struct console *con;
1701 trace_console_rcuidle(text, len);
1703 if (!console_drivers)
1706 for_each_console(con) {
1707 if (exclusive_console && con != exclusive_console)
1709 if (!(con->flags & CON_ENABLED))
1713 if (!cpu_online(smp_processor_id()) &&
1714 !(con->flags & CON_ANYTIME))
1716 if (con->flags & CON_EXTENDED)
1717 con->write(con, ext_text, ext_len);
1719 con->write(con, text, len);
1723 int printk_delay_msec __read_mostly;
1725 static inline void printk_delay(void)
1727 if (unlikely(printk_delay_msec)) {
1728 int m = printk_delay_msec;
1732 touch_nmi_watchdog();
1738 * Continuation lines are buffered, and not committed to the record buffer
1739 * until the line is complete, or a race forces it. The line fragments
1740 * though, are printed immediately to the consoles to ensure everything has
1741 * reached the console in case of a kernel crash.
1743 static struct cont {
1744 char buf[LOG_LINE_MAX];
1745 size_t len; /* length == 0 means unused buffer */
1746 struct task_struct *owner; /* task of first print*/
1747 u64 ts_nsec; /* time of first print */
1748 u8 level; /* log level of first message */
1749 u8 facility; /* log facility of first message */
1750 enum log_flags flags; /* prefix, newline flags */
1753 static void cont_flush(void)
1758 log_store(cont.facility, cont.level, cont.flags, cont.ts_nsec,
1759 NULL, 0, cont.buf, cont.len);
1763 static bool cont_add(int facility, int level, enum log_flags flags, const char *text, size_t len)
1766 * If ext consoles are present, flush and skip in-kernel
1767 * continuation. See nr_ext_console_drivers definition. Also, if
1768 * the line gets too long, split it up in separate records.
1770 if (nr_ext_console_drivers || cont.len + len > sizeof(cont.buf)) {
1776 cont.facility = facility;
1778 cont.owner = current;
1779 cont.ts_nsec = local_clock();
1783 memcpy(cont.buf + cont.len, text, len);
1786 // The original flags come from the first line,
1787 // but later continuations can add a newline.
1788 if (flags & LOG_NEWLINE) {
1789 cont.flags |= LOG_NEWLINE;
1793 if (cont.len > (sizeof(cont.buf) * 80) / 100)
1799 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)
1802 * If an earlier line was buffered, and we're a continuation
1803 * write from the same process, try to add it to the buffer.
1806 if (cont.owner == current && (lflags & LOG_CONT)) {
1807 if (cont_add(facility, level, lflags, text, text_len))
1810 /* Otherwise, make sure it's flushed */
1814 /* Skip empty continuation lines that couldn't be added - they just flush */
1815 if (!text_len && (lflags & LOG_CONT))
1818 /* If it doesn't end in a newline, try to buffer the current line */
1819 if (!(lflags & LOG_NEWLINE)) {
1820 if (cont_add(facility, level, lflags, text, text_len))
1824 /* Store it in the record log */
1825 return log_store(facility, level, lflags, 0, dict, dictlen, text, text_len);
1828 asmlinkage int vprintk_emit(int facility, int level,
1829 const char *dict, size_t dictlen,
1830 const char *fmt, va_list args)
1832 static char textbuf[LOG_LINE_MAX];
1833 char *text = textbuf;
1835 enum log_flags lflags = 0;
1836 unsigned long flags;
1838 bool in_sched = false;
1840 if (level == LOGLEVEL_SCHED) {
1841 level = LOGLEVEL_DEFAULT;
1845 boot_delay_msec(level);
1848 /* This stops the holder of console_sem just where we want him */
1849 logbuf_lock_irqsave(flags);
1851 * The printf needs to come first; we need the syslog
1852 * prefix which might be passed-in as a parameter.
1854 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1856 /* mark and strip a trailing newline */
1857 if (text_len && text[text_len-1] == '\n') {
1859 lflags |= LOG_NEWLINE;
1862 /* strip kernel syslog prefix and extract log level or control flags */
1863 if (facility == 0) {
1866 while ((kern_level = printk_get_level(text)) != 0) {
1867 switch (kern_level) {
1869 if (level == LOGLEVEL_DEFAULT)
1870 level = kern_level - '0';
1872 case 'd': /* KERN_DEFAULT */
1873 lflags |= LOG_PREFIX;
1875 case 'c': /* KERN_CONT */
1884 if (level == LOGLEVEL_DEFAULT)
1885 level = default_message_loglevel;
1888 lflags |= LOG_PREFIX|LOG_NEWLINE;
1890 printed_len = log_output(facility, level, lflags, dict, dictlen, text, text_len);
1892 logbuf_unlock_irqrestore(flags);
1894 /* If called from the scheduler, we can not call up(). */
1897 * Disable preemption to avoid being preempted while holding
1898 * console_sem which would prevent anyone from printing to
1903 * Try to acquire and then immediately release the console
1904 * semaphore. The release will print out buffers and wake up
1905 * /dev/kmsg and syslog() users.
1907 if (console_trylock_spinning())
1914 EXPORT_SYMBOL(vprintk_emit);
1916 asmlinkage int vprintk(const char *fmt, va_list args)
1918 return vprintk_func(fmt, args);
1920 EXPORT_SYMBOL(vprintk);
1922 asmlinkage int printk_emit(int facility, int level,
1923 const char *dict, size_t dictlen,
1924 const char *fmt, ...)
1929 va_start(args, fmt);
1930 r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1935 EXPORT_SYMBOL(printk_emit);
1937 int vprintk_default(const char *fmt, va_list args)
1941 #ifdef CONFIG_KGDB_KDB
1942 /* Allow to pass printk() to kdb but avoid a recursion. */
1943 if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
1944 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
1948 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1952 EXPORT_SYMBOL_GPL(vprintk_default);
1955 * printk - print a kernel message
1956 * @fmt: format string
1958 * This is printk(). It can be called from any context. We want it to work.
1960 * We try to grab the console_lock. If we succeed, it's easy - we log the
1961 * output and call the console drivers. If we fail to get the semaphore, we
1962 * place the output into the log buffer and return. The current holder of
1963 * the console_sem will notice the new output in console_unlock(); and will
1964 * send it to the consoles before releasing the lock.
1966 * One effect of this deferred printing is that code which calls printk() and
1967 * then changes console_loglevel may break. This is because console_loglevel
1968 * is inspected when the actual printing occurs.
1973 * See the vsnprintf() documentation for format string extensions over C99.
1975 asmlinkage __visible int printk(const char *fmt, ...)
1980 va_start(args, fmt);
1981 r = vprintk_func(fmt, args);
1986 EXPORT_SYMBOL(printk);
1988 #else /* CONFIG_PRINTK */
1990 #define LOG_LINE_MAX 0
1991 #define PREFIX_MAX 0
1993 static u64 syslog_seq;
1994 static u32 syslog_idx;
1995 static u64 console_seq;
1996 static u32 console_idx;
1997 static u64 log_first_seq;
1998 static u32 log_first_idx;
1999 static u64 log_next_seq;
2000 static char *log_text(const struct printk_log *msg) { return NULL; }
2001 static char *log_dict(const struct printk_log *msg) { return NULL; }
2002 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2003 static u32 log_next(u32 idx) { return 0; }
2004 static ssize_t msg_print_ext_header(char *buf, size_t size,
2005 struct printk_log *msg,
2006 u64 seq) { return 0; }
2007 static ssize_t msg_print_ext_body(char *buf, size_t size,
2008 char *dict, size_t dict_len,
2009 char *text, size_t text_len) { return 0; }
2010 static void console_lock_spinning_enable(void) { }
2011 static int console_lock_spinning_disable_and_check(void) { return 0; }
2012 static void call_console_drivers(const char *ext_text, size_t ext_len,
2013 const char *text, size_t len) {}
2014 static size_t msg_print_text(const struct printk_log *msg,
2015 bool syslog, char *buf, size_t size) { return 0; }
2016 static bool suppress_message_printing(int level) { return false; }
2018 #endif /* CONFIG_PRINTK */
2020 #ifdef CONFIG_EARLY_PRINTK
2021 struct console *early_console;
2023 asmlinkage __visible void early_printk(const char *fmt, ...)
2033 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2036 early_console->write(early_console, buf, n);
2040 static int __add_preferred_console(char *name, int idx, char *options,
2043 struct console_cmdline *c;
2047 * See if this tty is not yet registered, and
2048 * if we have a slot free.
2050 for (i = 0, c = console_cmdline;
2051 i < MAX_CMDLINECONSOLES && c->name[0];
2053 if (strcmp(c->name, name) == 0 && c->index == idx) {
2055 preferred_console = i;
2059 if (i == MAX_CMDLINECONSOLES)
2062 preferred_console = i;
2063 strlcpy(c->name, name, sizeof(c->name));
2064 c->options = options;
2065 braille_set_options(c, brl_options);
2071 static int __init console_msg_format_setup(char *str)
2073 if (!strcmp(str, "syslog"))
2074 console_msg_format = MSG_FORMAT_SYSLOG;
2075 if (!strcmp(str, "default"))
2076 console_msg_format = MSG_FORMAT_DEFAULT;
2079 __setup("console_msg_format=", console_msg_format_setup);
2082 * Set up a console. Called via do_early_param() in init/main.c
2083 * for each "console=" parameter in the boot command line.
2085 static int __init console_setup(char *str)
2087 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2088 char *s, *options, *brl_options = NULL;
2091 if (_braille_console_setup(&str, &brl_options))
2095 * Decode str into name, index, options.
2097 if (str[0] >= '0' && str[0] <= '9') {
2098 strcpy(buf, "ttyS");
2099 strncpy(buf + 4, str, sizeof(buf) - 5);
2101 strncpy(buf, str, sizeof(buf) - 1);
2103 buf[sizeof(buf) - 1] = 0;
2104 options = strchr(str, ',');
2108 if (!strcmp(str, "ttya"))
2109 strcpy(buf, "ttyS0");
2110 if (!strcmp(str, "ttyb"))
2111 strcpy(buf, "ttyS1");
2113 for (s = buf; *s; s++)
2114 if (isdigit(*s) || *s == ',')
2116 idx = simple_strtoul(s, NULL, 10);
2119 __add_preferred_console(buf, idx, options, brl_options);
2120 console_set_on_cmdline = 1;
2123 __setup("console=", console_setup);
2126 * add_preferred_console - add a device to the list of preferred consoles.
2127 * @name: device name
2128 * @idx: device index
2129 * @options: options for this console
2131 * The last preferred console added will be used for kernel messages
2132 * and stdin/out/err for init. Normally this is used by console_setup
2133 * above to handle user-supplied console arguments; however it can also
2134 * be used by arch-specific code either to override the user or more
2135 * commonly to provide a default console (ie from PROM variables) when
2136 * the user has not supplied one.
2138 int add_preferred_console(char *name, int idx, char *options)
2140 return __add_preferred_console(name, idx, options, NULL);
2143 bool console_suspend_enabled = true;
2144 EXPORT_SYMBOL(console_suspend_enabled);
2146 static int __init console_suspend_disable(char *str)
2148 console_suspend_enabled = false;
2151 __setup("no_console_suspend", console_suspend_disable);
2152 module_param_named(console_suspend, console_suspend_enabled,
2153 bool, S_IRUGO | S_IWUSR);
2154 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2155 " and hibernate operations");
2158 * suspend_console - suspend the console subsystem
2160 * This disables printk() while we go into suspend states
2162 void suspend_console(void)
2164 if (!console_suspend_enabled)
2166 printk("Suspending console(s) (use no_console_suspend to debug)\n");
2168 console_suspended = 1;
2172 void resume_console(void)
2174 if (!console_suspend_enabled)
2177 console_suspended = 0;
2182 * console_cpu_notify - print deferred console messages after CPU hotplug
2185 * If printk() is called from a CPU that is not online yet, the messages
2186 * will be printed on the console only if there are CON_ANYTIME consoles.
2187 * This function is called when a new CPU comes online (or fails to come
2188 * up) or goes offline.
2190 static int console_cpu_notify(unsigned int cpu)
2192 if (!cpuhp_tasks_frozen) {
2193 /* If trylock fails, someone else is doing the printing */
2194 if (console_trylock())
2201 * console_lock - lock the console system for exclusive use.
2203 * Acquires a lock which guarantees that the caller has
2204 * exclusive access to the console system and the console_drivers list.
2206 * Can sleep, returns nothing.
2208 void console_lock(void)
2213 if (console_suspended)
2216 console_may_schedule = 1;
2218 EXPORT_SYMBOL(console_lock);
2221 * console_trylock - try to lock the console system for exclusive use.
2223 * Try to acquire a lock which guarantees that the caller has exclusive
2224 * access to the console system and the console_drivers list.
2226 * returns 1 on success, and 0 on failure to acquire the lock.
2228 int console_trylock(void)
2230 if (down_trylock_console_sem())
2232 if (console_suspended) {
2237 console_may_schedule = 0;
2240 EXPORT_SYMBOL(console_trylock);
2242 int is_console_locked(void)
2244 return console_locked;
2248 * Check if we have any console that is capable of printing while cpu is
2249 * booting or shutting down. Requires console_sem.
2251 static int have_callable_console(void)
2253 struct console *con;
2255 for_each_console(con)
2256 if ((con->flags & CON_ENABLED) &&
2257 (con->flags & CON_ANYTIME))
2264 * Can we actually use the console at this time on this cpu?
2266 * Console drivers may assume that per-cpu resources have been allocated. So
2267 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2268 * call them until this CPU is officially up.
2270 static inline int can_use_console(void)
2272 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2276 * console_unlock - unlock the console system
2278 * Releases the console_lock which the caller holds on the console system
2279 * and the console driver list.
2281 * While the console_lock was held, console output may have been buffered
2282 * by printk(). If this is the case, console_unlock(); emits
2283 * the output prior to releasing the lock.
2285 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2287 * console_unlock(); may be called from any context.
2289 void console_unlock(void)
2291 static char ext_text[CONSOLE_EXT_LOG_MAX];
2292 static char text[LOG_LINE_MAX + PREFIX_MAX];
2293 static u64 seen_seq;
2294 unsigned long flags;
2295 bool wake_klogd = false;
2296 bool do_cond_resched, retry;
2298 if (console_suspended) {
2304 * Console drivers are called with interrupts disabled, so
2305 * @console_may_schedule should be cleared before; however, we may
2306 * end up dumping a lot of lines, for example, if called from
2307 * console registration path, and should invoke cond_resched()
2308 * between lines if allowable. Not doing so can cause a very long
2309 * scheduling stall on a slow console leading to RCU stall and
2310 * softlockup warnings which exacerbate the issue with more
2311 * messages practically incapacitating the system.
2313 * console_trylock() is not able to detect the preemptive
2314 * context reliably. Therefore the value must be stored before
2315 * and cleared after the the "again" goto label.
2317 do_cond_resched = console_may_schedule;
2319 console_may_schedule = 0;
2322 * We released the console_sem lock, so we need to recheck if
2323 * cpu is online and (if not) is there at least one CON_ANYTIME
2326 if (!can_use_console()) {
2333 struct printk_log *msg;
2337 printk_safe_enter_irqsave(flags);
2338 raw_spin_lock(&logbuf_lock);
2339 if (seen_seq != log_next_seq) {
2341 seen_seq = log_next_seq;
2344 if (console_seq < log_first_seq) {
2345 len = sprintf(text, "** %u printk messages dropped **\n",
2346 (unsigned)(log_first_seq - console_seq));
2348 /* messages are gone, move to first one */
2349 console_seq = log_first_seq;
2350 console_idx = log_first_idx;
2355 if (console_seq == log_next_seq)
2358 msg = log_from_idx(console_idx);
2359 if (suppress_message_printing(msg->level)) {
2361 * Skip record we have buffered and already printed
2362 * directly to the console when we received it, and
2363 * record that has level above the console loglevel.
2365 console_idx = log_next(console_idx);
2370 len += msg_print_text(msg,
2371 console_msg_format & MSG_FORMAT_SYSLOG,
2373 sizeof(text) - len);
2374 if (nr_ext_console_drivers) {
2375 ext_len = msg_print_ext_header(ext_text,
2378 ext_len += msg_print_ext_body(ext_text + ext_len,
2379 sizeof(ext_text) - ext_len,
2380 log_dict(msg), msg->dict_len,
2381 log_text(msg), msg->text_len);
2383 console_idx = log_next(console_idx);
2385 raw_spin_unlock(&logbuf_lock);
2388 * While actively printing out messages, if another printk()
2389 * were to occur on another CPU, it may wait for this one to
2390 * finish. This task can not be preempted if there is a
2391 * waiter waiting to take over.
2393 console_lock_spinning_enable();
2395 stop_critical_timings(); /* don't trace print latency */
2396 call_console_drivers(ext_text, ext_len, text, len);
2397 start_critical_timings();
2399 if (console_lock_spinning_disable_and_check()) {
2400 printk_safe_exit_irqrestore(flags);
2404 printk_safe_exit_irqrestore(flags);
2406 if (do_cond_resched)
2412 /* Release the exclusive_console once it is used */
2413 if (unlikely(exclusive_console))
2414 exclusive_console = NULL;
2416 raw_spin_unlock(&logbuf_lock);
2421 * Someone could have filled up the buffer again, so re-check if there's
2422 * something to flush. In case we cannot trylock the console_sem again,
2423 * there's a new owner and the console_unlock() from them will do the
2424 * flush, no worries.
2426 raw_spin_lock(&logbuf_lock);
2427 retry = console_seq != log_next_seq;
2428 raw_spin_unlock(&logbuf_lock);
2429 printk_safe_exit_irqrestore(flags);
2431 if (retry && console_trylock())
2437 EXPORT_SYMBOL(console_unlock);
2440 * console_conditional_schedule - yield the CPU if required
2442 * If the console code is currently allowed to sleep, and
2443 * if this CPU should yield the CPU to another task, do
2446 * Must be called within console_lock();.
2448 void __sched console_conditional_schedule(void)
2450 if (console_may_schedule)
2453 EXPORT_SYMBOL(console_conditional_schedule);
2455 void console_unblank(void)
2460 * console_unblank can no longer be called in interrupt context unless
2461 * oops_in_progress is set to 1..
2463 if (oops_in_progress) {
2464 if (down_trylock_console_sem() != 0)
2470 console_may_schedule = 0;
2472 if ((c->flags & CON_ENABLED) && c->unblank)
2478 * console_flush_on_panic - flush console content on panic
2480 * Immediately output all pending messages no matter what.
2482 void console_flush_on_panic(void)
2485 * If someone else is holding the console lock, trylock will fail
2486 * and may_schedule may be set. Ignore and proceed to unlock so
2487 * that messages are flushed out. As this can be called from any
2488 * context and we don't want to get preempted while flushing,
2489 * ensure may_schedule is cleared.
2492 console_may_schedule = 0;
2497 * Return the console tty driver structure and its associated index
2499 struct tty_driver *console_device(int *index)
2502 struct tty_driver *driver = NULL;
2505 for_each_console(c) {
2508 driver = c->device(c, index);
2517 * Prevent further output on the passed console device so that (for example)
2518 * serial drivers can disable console output before suspending a port, and can
2519 * re-enable output afterwards.
2521 void console_stop(struct console *console)
2524 console->flags &= ~CON_ENABLED;
2527 EXPORT_SYMBOL(console_stop);
2529 void console_start(struct console *console)
2532 console->flags |= CON_ENABLED;
2535 EXPORT_SYMBOL(console_start);
2537 static int __read_mostly keep_bootcon;
2539 static int __init keep_bootcon_setup(char *str)
2542 pr_info("debug: skip boot console de-registration.\n");
2547 early_param("keep_bootcon", keep_bootcon_setup);
2550 * The console driver calls this routine during kernel initialization
2551 * to register the console printing procedure with printk() and to
2552 * print any messages that were printed by the kernel before the
2553 * console driver was initialized.
2555 * This can happen pretty early during the boot process (because of
2556 * early_printk) - sometimes before setup_arch() completes - be careful
2557 * of what kernel features are used - they may not be initialised yet.
2559 * There are two types of consoles - bootconsoles (early_printk) and
2560 * "real" consoles (everything which is not a bootconsole) which are
2561 * handled differently.
2562 * - Any number of bootconsoles can be registered at any time.
2563 * - As soon as a "real" console is registered, all bootconsoles
2564 * will be unregistered automatically.
2565 * - Once a "real" console is registered, any attempt to register a
2566 * bootconsoles will be rejected
2568 void register_console(struct console *newcon)
2571 unsigned long flags;
2572 struct console *bcon = NULL;
2573 struct console_cmdline *c;
2574 static bool has_preferred;
2576 if (console_drivers)
2577 for_each_console(bcon)
2578 if (WARN(bcon == newcon,
2579 "console '%s%d' already registered\n",
2580 bcon->name, bcon->index))
2584 * before we register a new CON_BOOT console, make sure we don't
2585 * already have a valid console
2587 if (console_drivers && newcon->flags & CON_BOOT) {
2588 /* find the last or real console */
2589 for_each_console(bcon) {
2590 if (!(bcon->flags & CON_BOOT)) {
2591 pr_info("Too late to register bootconsole %s%d\n",
2592 newcon->name, newcon->index);
2598 if (console_drivers && console_drivers->flags & CON_BOOT)
2599 bcon = console_drivers;
2601 if (!has_preferred || bcon || !console_drivers)
2602 has_preferred = preferred_console >= 0;
2605 * See if we want to use this console driver. If we
2606 * didn't select a console we take the first one
2607 * that registers here.
2609 if (!has_preferred) {
2610 if (newcon->index < 0)
2612 if (newcon->setup == NULL ||
2613 newcon->setup(newcon, NULL) == 0) {
2614 newcon->flags |= CON_ENABLED;
2615 if (newcon->device) {
2616 newcon->flags |= CON_CONSDEV;
2617 has_preferred = true;
2623 * See if this console matches one we selected on
2626 for (i = 0, c = console_cmdline;
2627 i < MAX_CMDLINECONSOLES && c->name[0];
2629 if (!newcon->match ||
2630 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2631 /* default matching */
2632 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2633 if (strcmp(c->name, newcon->name) != 0)
2635 if (newcon->index >= 0 &&
2636 newcon->index != c->index)
2638 if (newcon->index < 0)
2639 newcon->index = c->index;
2641 if (_braille_register_console(newcon, c))
2644 if (newcon->setup &&
2645 newcon->setup(newcon, c->options) != 0)
2649 newcon->flags |= CON_ENABLED;
2650 if (i == preferred_console) {
2651 newcon->flags |= CON_CONSDEV;
2652 has_preferred = true;
2657 if (!(newcon->flags & CON_ENABLED))
2661 * If we have a bootconsole, and are switching to a real console,
2662 * don't print everything out again, since when the boot console, and
2663 * the real console are the same physical device, it's annoying to
2664 * see the beginning boot messages twice
2666 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2667 newcon->flags &= ~CON_PRINTBUFFER;
2670 * Put this console in the list - keep the
2671 * preferred driver at the head of the list.
2674 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2675 newcon->next = console_drivers;
2676 console_drivers = newcon;
2678 newcon->next->flags &= ~CON_CONSDEV;
2680 newcon->next = console_drivers->next;
2681 console_drivers->next = newcon;
2684 if (newcon->flags & CON_EXTENDED)
2685 if (!nr_ext_console_drivers++)
2686 pr_info("printk: continuation disabled due to ext consoles, expect more fragments in /dev/kmsg\n");
2688 if (newcon->flags & CON_PRINTBUFFER) {
2690 * console_unlock(); will print out the buffered messages
2693 logbuf_lock_irqsave(flags);
2694 console_seq = syslog_seq;
2695 console_idx = syslog_idx;
2696 logbuf_unlock_irqrestore(flags);
2698 * We're about to replay the log buffer. Only do this to the
2699 * just-registered console to avoid excessive message spam to
2700 * the already-registered consoles.
2702 exclusive_console = newcon;
2705 console_sysfs_notify();
2708 * By unregistering the bootconsoles after we enable the real console
2709 * we get the "console xxx enabled" message on all the consoles -
2710 * boot consoles, real consoles, etc - this is to ensure that end
2711 * users know there might be something in the kernel's log buffer that
2712 * went to the bootconsole (that they do not see on the real console)
2714 pr_info("%sconsole [%s%d] enabled\n",
2715 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2716 newcon->name, newcon->index);
2718 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2720 /* We need to iterate through all boot consoles, to make
2721 * sure we print everything out, before we unregister them.
2723 for_each_console(bcon)
2724 if (bcon->flags & CON_BOOT)
2725 unregister_console(bcon);
2728 EXPORT_SYMBOL(register_console);
2730 int unregister_console(struct console *console)
2732 struct console *a, *b;
2735 pr_info("%sconsole [%s%d] disabled\n",
2736 (console->flags & CON_BOOT) ? "boot" : "" ,
2737 console->name, console->index);
2739 res = _braille_unregister_console(console);
2745 if (console_drivers == console) {
2746 console_drivers=console->next;
2748 } else if (console_drivers) {
2749 for (a=console_drivers->next, b=console_drivers ;
2750 a; b=a, a=b->next) {
2759 if (!res && (console->flags & CON_EXTENDED))
2760 nr_ext_console_drivers--;
2763 * If this isn't the last console and it has CON_CONSDEV set, we
2764 * need to set it on the next preferred console.
2766 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2767 console_drivers->flags |= CON_CONSDEV;
2769 console->flags &= ~CON_ENABLED;
2771 console_sysfs_notify();
2774 EXPORT_SYMBOL(unregister_console);
2777 * Initialize the console device. This is called *early*, so
2778 * we can't necessarily depend on lots of kernel help here.
2779 * Just do some early initializations, and do the complex setup
2782 void __init console_init(void)
2786 /* Setup the default TTY line discipline. */
2790 * set up the console device so that later boot sequences can
2791 * inform about problems etc..
2793 call = __con_initcall_start;
2794 while (call < __con_initcall_end) {
2801 * Some boot consoles access data that is in the init section and which will
2802 * be discarded after the initcalls have been run. To make sure that no code
2803 * will access this data, unregister the boot consoles in a late initcall.
2805 * If for some reason, such as deferred probe or the driver being a loadable
2806 * module, the real console hasn't registered yet at this point, there will
2807 * be a brief interval in which no messages are logged to the console, which
2808 * makes it difficult to diagnose problems that occur during this time.
2810 * To mitigate this problem somewhat, only unregister consoles whose memory
2811 * intersects with the init section. Note that all other boot consoles will
2812 * get unregistred when the real preferred console is registered.
2814 static int __init printk_late_init(void)
2816 struct console *con;
2819 for_each_console(con) {
2820 if (!(con->flags & CON_BOOT))
2823 /* Check addresses that might be used for enabled consoles. */
2824 if (init_section_intersects(con, sizeof(*con)) ||
2825 init_section_contains(con->write, 0) ||
2826 init_section_contains(con->read, 0) ||
2827 init_section_contains(con->device, 0) ||
2828 init_section_contains(con->unblank, 0) ||
2829 init_section_contains(con->data, 0)) {
2831 * Please, consider moving the reported consoles out
2832 * of the init section.
2834 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2835 con->name, con->index);
2836 unregister_console(con);
2839 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2840 console_cpu_notify);
2842 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2843 console_cpu_notify, NULL);
2847 late_initcall(printk_late_init);
2849 #if defined CONFIG_PRINTK
2851 * Delayed printk version, for scheduler-internal messages:
2853 #define PRINTK_PENDING_WAKEUP 0x01
2854 #define PRINTK_PENDING_OUTPUT 0x02
2856 static DEFINE_PER_CPU(int, printk_pending);
2858 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2860 int pending = __this_cpu_xchg(printk_pending, 0);
2862 if (pending & PRINTK_PENDING_OUTPUT) {
2863 /* If trylock fails, someone else is doing the printing */
2864 if (console_trylock())
2868 if (pending & PRINTK_PENDING_WAKEUP)
2869 wake_up_interruptible(&log_wait);
2872 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2873 .func = wake_up_klogd_work_func,
2874 .flags = IRQ_WORK_LAZY,
2877 void wake_up_klogd(void)
2880 if (waitqueue_active(&log_wait)) {
2881 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2882 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2887 int vprintk_deferred(const char *fmt, va_list args)
2891 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2894 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2895 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2901 int printk_deferred(const char *fmt, ...)
2906 va_start(args, fmt);
2907 r = vprintk_deferred(fmt, args);
2914 * printk rate limiting, lifted from the networking subsystem.
2916 * This enforces a rate limit: not more than 10 kernel messages
2917 * every 5s to make a denial-of-service attack impossible.
2919 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2921 int __printk_ratelimit(const char *func)
2923 return ___ratelimit(&printk_ratelimit_state, func);
2925 EXPORT_SYMBOL(__printk_ratelimit);
2928 * printk_timed_ratelimit - caller-controlled printk ratelimiting
2929 * @caller_jiffies: pointer to caller's state
2930 * @interval_msecs: minimum interval between prints
2932 * printk_timed_ratelimit() returns true if more than @interval_msecs
2933 * milliseconds have elapsed since the last time printk_timed_ratelimit()
2936 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
2937 unsigned int interval_msecs)
2939 unsigned long elapsed = jiffies - *caller_jiffies;
2941 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
2944 *caller_jiffies = jiffies;
2947 EXPORT_SYMBOL(printk_timed_ratelimit);
2949 static DEFINE_SPINLOCK(dump_list_lock);
2950 static LIST_HEAD(dump_list);
2953 * kmsg_dump_register - register a kernel log dumper.
2954 * @dumper: pointer to the kmsg_dumper structure
2956 * Adds a kernel log dumper to the system. The dump callback in the
2957 * structure will be called when the kernel oopses or panics and must be
2958 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
2960 int kmsg_dump_register(struct kmsg_dumper *dumper)
2962 unsigned long flags;
2965 /* The dump callback needs to be set */
2969 spin_lock_irqsave(&dump_list_lock, flags);
2970 /* Don't allow registering multiple times */
2971 if (!dumper->registered) {
2972 dumper->registered = 1;
2973 list_add_tail_rcu(&dumper->list, &dump_list);
2976 spin_unlock_irqrestore(&dump_list_lock, flags);
2980 EXPORT_SYMBOL_GPL(kmsg_dump_register);
2983 * kmsg_dump_unregister - unregister a kmsg dumper.
2984 * @dumper: pointer to the kmsg_dumper structure
2986 * Removes a dump device from the system. Returns zero on success and
2987 * %-EINVAL otherwise.
2989 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
2991 unsigned long flags;
2994 spin_lock_irqsave(&dump_list_lock, flags);
2995 if (dumper->registered) {
2996 dumper->registered = 0;
2997 list_del_rcu(&dumper->list);
3000 spin_unlock_irqrestore(&dump_list_lock, flags);
3005 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3007 static bool always_kmsg_dump;
3008 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3011 * kmsg_dump - dump kernel log to kernel message dumpers.
3012 * @reason: the reason (oops, panic etc) for dumping
3014 * Call each of the registered dumper's dump() callback, which can
3015 * retrieve the kmsg records with kmsg_dump_get_line() or
3016 * kmsg_dump_get_buffer().
3018 void kmsg_dump(enum kmsg_dump_reason reason)
3020 struct kmsg_dumper *dumper;
3021 unsigned long flags;
3023 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3027 list_for_each_entry_rcu(dumper, &dump_list, list) {
3028 if (dumper->max_reason && reason > dumper->max_reason)
3031 /* initialize iterator with data about the stored records */
3032 dumper->active = true;
3034 logbuf_lock_irqsave(flags);
3035 dumper->cur_seq = clear_seq;
3036 dumper->cur_idx = clear_idx;
3037 dumper->next_seq = log_next_seq;
3038 dumper->next_idx = log_next_idx;
3039 logbuf_unlock_irqrestore(flags);
3041 /* invoke dumper which will iterate over records */
3042 dumper->dump(dumper, reason);
3044 /* reset iterator */
3045 dumper->active = false;
3051 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3052 * @dumper: registered kmsg dumper
3053 * @syslog: include the "<4>" prefixes
3054 * @line: buffer to copy the line to
3055 * @size: maximum size of the buffer
3056 * @len: length of line placed into buffer
3058 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3059 * record, and copy one record into the provided buffer.
3061 * Consecutive calls will return the next available record moving
3062 * towards the end of the buffer with the youngest messages.
3064 * A return value of FALSE indicates that there are no more records to
3067 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3069 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3070 char *line, size_t size, size_t *len)
3072 struct printk_log *msg;
3076 if (!dumper->active)
3079 if (dumper->cur_seq < log_first_seq) {
3080 /* messages are gone, move to first available one */
3081 dumper->cur_seq = log_first_seq;
3082 dumper->cur_idx = log_first_idx;
3086 if (dumper->cur_seq >= log_next_seq)
3089 msg = log_from_idx(dumper->cur_idx);
3090 l = msg_print_text(msg, syslog, line, size);
3092 dumper->cur_idx = log_next(dumper->cur_idx);
3102 * kmsg_dump_get_line - retrieve one kmsg log line
3103 * @dumper: registered kmsg dumper
3104 * @syslog: include the "<4>" prefixes
3105 * @line: buffer to copy the line to
3106 * @size: maximum size of the buffer
3107 * @len: length of line placed into buffer
3109 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3110 * record, and copy one record into the provided buffer.
3112 * Consecutive calls will return the next available record moving
3113 * towards the end of the buffer with the youngest messages.
3115 * A return value of FALSE indicates that there are no more records to
3118 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3119 char *line, size_t size, size_t *len)
3121 unsigned long flags;
3124 logbuf_lock_irqsave(flags);
3125 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3126 logbuf_unlock_irqrestore(flags);
3130 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3133 * kmsg_dump_get_buffer - copy kmsg log lines
3134 * @dumper: registered kmsg dumper
3135 * @syslog: include the "<4>" prefixes
3136 * @buf: buffer to copy the line to
3137 * @size: maximum size of the buffer
3138 * @len: length of line placed into buffer
3140 * Start at the end of the kmsg buffer and fill the provided buffer
3141 * with as many of the the *youngest* kmsg records that fit into it.
3142 * If the buffer is large enough, all available kmsg records will be
3143 * copied with a single call.
3145 * Consecutive calls will fill the buffer with the next block of
3146 * available older records, not including the earlier retrieved ones.
3148 * A return value of FALSE indicates that there are no more records to
3151 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3152 char *buf, size_t size, size_t *len)
3154 unsigned long flags;
3162 if (!dumper->active)
3165 logbuf_lock_irqsave(flags);
3166 if (dumper->cur_seq < log_first_seq) {
3167 /* messages are gone, move to first available one */
3168 dumper->cur_seq = log_first_seq;
3169 dumper->cur_idx = log_first_idx;
3173 if (dumper->cur_seq >= dumper->next_seq) {
3174 logbuf_unlock_irqrestore(flags);
3178 /* calculate length of entire buffer */
3179 seq = dumper->cur_seq;
3180 idx = dumper->cur_idx;
3181 while (seq < dumper->next_seq) {
3182 struct printk_log *msg = log_from_idx(idx);
3184 l += msg_print_text(msg, true, NULL, 0);
3185 idx = log_next(idx);
3189 /* move first record forward until length fits into the buffer */
3190 seq = dumper->cur_seq;
3191 idx = dumper->cur_idx;
3192 while (l > size && seq < dumper->next_seq) {
3193 struct printk_log *msg = log_from_idx(idx);
3195 l -= msg_print_text(msg, true, NULL, 0);
3196 idx = log_next(idx);
3200 /* last message in next interation */
3205 while (seq < dumper->next_seq) {
3206 struct printk_log *msg = log_from_idx(idx);
3208 l += msg_print_text(msg, syslog, buf + l, size - l);
3209 idx = log_next(idx);
3213 dumper->next_seq = next_seq;
3214 dumper->next_idx = next_idx;
3216 logbuf_unlock_irqrestore(flags);
3222 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3225 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3226 * @dumper: registered kmsg dumper
3228 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3229 * kmsg_dump_get_buffer() can be called again and used multiple
3230 * times within the same dumper.dump() callback.
3232 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3234 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3236 dumper->cur_seq = clear_seq;
3237 dumper->cur_idx = clear_idx;
3238 dumper->next_seq = log_next_seq;
3239 dumper->next_idx = log_next_idx;
3243 * kmsg_dump_rewind - reset the interator
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 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3252 unsigned long flags;
3254 logbuf_lock_irqsave(flags);
3255 kmsg_dump_rewind_nolock(dumper);
3256 logbuf_unlock_irqrestore(flags);
3258 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3260 static char dump_stack_arch_desc_str[128];
3263 * dump_stack_set_arch_desc - set arch-specific str to show with task dumps
3264 * @fmt: printf-style format string
3265 * @...: arguments for the format string
3267 * The configured string will be printed right after utsname during task
3268 * dumps. Usually used to add arch-specific system identifiers. If an
3269 * arch wants to make use of such an ID string, it should initialize this
3270 * as soon as possible during boot.
3272 void __init dump_stack_set_arch_desc(const char *fmt, ...)
3276 va_start(args, fmt);
3277 vsnprintf(dump_stack_arch_desc_str, sizeof(dump_stack_arch_desc_str),
3283 * dump_stack_print_info - print generic debug info for dump_stack()
3284 * @log_lvl: log level
3286 * Arch-specific dump_stack() implementations can use this function to
3287 * print out the same debug information as the generic dump_stack().
3289 void dump_stack_print_info(const char *log_lvl)
3291 printk("%sCPU: %d PID: %d Comm: %.20s %s%s %s %.*s\n",
3292 log_lvl, raw_smp_processor_id(), current->pid, current->comm,
3293 kexec_crash_loaded() ? "Kdump: loaded " : "",
3295 init_utsname()->release,
3296 (int)strcspn(init_utsname()->version, " "),
3297 init_utsname()->version);
3299 if (dump_stack_arch_desc_str[0] != '\0')
3300 printk("%sHardware name: %s\n",
3301 log_lvl, dump_stack_arch_desc_str);
3303 print_worker_info(log_lvl, current);
3307 * show_regs_print_info - print generic debug info for show_regs()
3308 * @log_lvl: log level
3310 * show_regs() implementations can use this function to print out generic
3311 * debug information.
3313 void show_regs_print_info(const char *log_lvl)
3315 dump_stack_print_info(log_lvl);