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) {
131 pr_warn("printk.devkmsg: bad option string '%s'\n", str);
136 * Set sysctl string accordingly:
138 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
139 strcpy(devkmsg_log_str, "on");
140 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
141 strcpy(devkmsg_log_str, "off");
142 /* else "ratelimit" which is set by default. */
145 * Sysctl cannot change it anymore. The kernel command line setting of
146 * this parameter is to force the setting to be permanent throughout the
147 * runtime of the system. This is a precation measure against userspace
148 * trying to be a smarta** and attempting to change it up on us.
150 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
154 __setup("printk.devkmsg=", control_devkmsg);
156 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
158 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
159 void __user *buffer, size_t *lenp, loff_t *ppos)
161 char old_str[DEVKMSG_STR_MAX_SIZE];
166 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
170 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
173 err = proc_dostring(table, write, buffer, lenp, ppos);
178 err = __control_devkmsg(devkmsg_log_str);
181 * Do not accept an unknown string OR a known string with
184 if (err < 0 || (err + 1 != *lenp)) {
186 /* ... and restore old setting. */
188 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
198 * Number of registered extended console drivers.
200 * If extended consoles are present, in-kernel cont reassembly is disabled
201 * and each fragment is stored as a separate log entry with proper
202 * continuation flag so that every emitted message has full metadata. This
203 * doesn't change the result for regular consoles or /proc/kmsg. For
204 * /dev/kmsg, as long as the reader concatenates messages according to
205 * consecutive continuation flags, the end result should be the same too.
207 static int nr_ext_console_drivers;
210 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
211 * macros instead of functions so that _RET_IP_ contains useful information.
213 #define down_console_sem() do { \
215 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
218 static int __down_trylock_console_sem(unsigned long ip)
224 * Here and in __up_console_sem() we need to be in safe mode,
225 * because spindump/WARN/etc from under console ->lock will
226 * deadlock in printk()->down_trylock_console_sem() otherwise.
228 printk_safe_enter_irqsave(flags);
229 lock_failed = down_trylock(&console_sem);
230 printk_safe_exit_irqrestore(flags);
234 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
237 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
239 static void __up_console_sem(unsigned long ip)
243 mutex_release(&console_lock_dep_map, 1, ip);
245 printk_safe_enter_irqsave(flags);
247 printk_safe_exit_irqrestore(flags);
249 #define up_console_sem() __up_console_sem(_RET_IP_)
252 * This is used for debugging the mess that is the VT code by
253 * keeping track if we have the console semaphore held. It's
254 * definitely not the perfect debug tool (we don't know if _WE_
255 * hold it and are racing, but it helps tracking those weird code
256 * paths in the console code where we end up in places I want
257 * locked without the console sempahore held).
259 static int console_locked, console_suspended;
262 * If exclusive_console is non-NULL then only this console is to be printed to.
264 static struct console *exclusive_console;
267 * Array of consoles built from command line options (console=)
270 #define MAX_CMDLINECONSOLES 8
272 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
274 static int preferred_console = -1;
275 int console_set_on_cmdline;
276 EXPORT_SYMBOL(console_set_on_cmdline);
278 /* Flag: console code may call schedule() */
279 static int console_may_schedule;
281 enum con_msg_format_flags {
282 MSG_FORMAT_DEFAULT = 0,
283 MSG_FORMAT_SYSLOG = (1 << 0),
286 static int console_msg_format = MSG_FORMAT_DEFAULT;
289 * The printk log buffer consists of a chain of concatenated variable
290 * length records. Every record starts with a record header, containing
291 * the overall length of the record.
293 * The heads to the first and last entry in the buffer, as well as the
294 * sequence numbers of these entries are maintained when messages are
297 * If the heads indicate available messages, the length in the header
298 * tells the start next message. A length == 0 for the next message
299 * indicates a wrap-around to the beginning of the buffer.
301 * Every record carries the monotonic timestamp in microseconds, as well as
302 * the standard userspace syslog level and syslog facility. The usual
303 * kernel messages use LOG_KERN; userspace-injected messages always carry
304 * a matching syslog facility, by default LOG_USER. The origin of every
305 * message can be reliably determined that way.
307 * The human readable log message directly follows the message header. The
308 * length of the message text is stored in the header, the stored message
311 * Optionally, a message can carry a dictionary of properties (key/value pairs),
312 * to provide userspace with a machine-readable message context.
314 * Examples for well-defined, commonly used property names are:
315 * DEVICE=b12:8 device identifier
319 * +sound:card0 subsystem:devname
320 * SUBSYSTEM=pci driver-core subsystem name
322 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
323 * follows directly after a '=' character. Every property is terminated by
324 * a '\0' character. The last property is not terminated.
326 * Example of a message structure:
327 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
328 * 0008 34 00 record is 52 bytes long
329 * 000a 0b 00 text is 11 bytes long
330 * 000c 1f 00 dictionary is 23 bytes long
331 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
332 * 0010 69 74 27 73 20 61 20 6c "it's a l"
334 * 001b 44 45 56 49 43 "DEVIC"
335 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
336 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
338 * 0032 00 00 00 padding to next message header
340 * The 'struct printk_log' buffer header must never be directly exported to
341 * userspace, it is a kernel-private implementation detail that might
342 * need to be changed in the future, when the requirements change.
344 * /dev/kmsg exports the structured data in the following line format:
345 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
347 * Users of the export format should ignore possible additional values
348 * separated by ',', and find the message after the ';' character.
350 * The optional key/value pairs are attached as continuation lines starting
351 * with a space character and terminated by a newline. All possible
352 * non-prinatable characters are escaped in the "\xff" notation.
356 LOG_NEWLINE = 2, /* text ended with a newline */
357 LOG_PREFIX = 4, /* text started with a prefix */
358 LOG_CONT = 8, /* text is a fragment of a continuation line */
362 u64 ts_nsec; /* timestamp in nanoseconds */
363 u16 len; /* length of entire record */
364 u16 text_len; /* length of text buffer */
365 u16 dict_len; /* length of dictionary buffer */
366 u8 facility; /* syslog facility */
367 u8 flags:5; /* internal record flags */
368 u8 level:3; /* syslog level */
370 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
371 __packed __aligned(4)
376 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
377 * within the scheduler's rq lock. It must be released before calling
378 * console_unlock() or anything else that might wake up a process.
380 DEFINE_RAW_SPINLOCK(logbuf_lock);
383 * Helper macros to lock/unlock logbuf_lock and switch between
384 * printk-safe/unsafe modes.
386 #define logbuf_lock_irq() \
388 printk_safe_enter_irq(); \
389 raw_spin_lock(&logbuf_lock); \
392 #define logbuf_unlock_irq() \
394 raw_spin_unlock(&logbuf_lock); \
395 printk_safe_exit_irq(); \
398 #define logbuf_lock_irqsave(flags) \
400 printk_safe_enter_irqsave(flags); \
401 raw_spin_lock(&logbuf_lock); \
404 #define logbuf_unlock_irqrestore(flags) \
406 raw_spin_unlock(&logbuf_lock); \
407 printk_safe_exit_irqrestore(flags); \
411 DECLARE_WAIT_QUEUE_HEAD(log_wait);
412 /* the next printk record to read by syslog(READ) or /proc/kmsg */
413 static u64 syslog_seq;
414 static u32 syslog_idx;
415 static size_t syslog_partial;
417 /* index and sequence number of the first record stored in the buffer */
418 static u64 log_first_seq;
419 static u32 log_first_idx;
421 /* index and sequence number of the next record to store in the buffer */
422 static u64 log_next_seq;
423 static u32 log_next_idx;
425 /* the next printk record to write to the console */
426 static u64 console_seq;
427 static u32 console_idx;
428 static u64 exclusive_console_stop_seq;
430 /* the next printk record to read after the last 'clear' command */
431 static u64 clear_seq;
432 static u32 clear_idx;
434 #define PREFIX_MAX 32
435 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
437 #define LOG_LEVEL(v) ((v) & 0x07)
438 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
441 #define LOG_ALIGN __alignof__(struct printk_log)
442 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
443 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
444 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
445 static char *log_buf = __log_buf;
446 static u32 log_buf_len = __LOG_BUF_LEN;
449 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
450 * per_cpu_areas are initialised. This variable is set to true when
451 * it's safe to access per-CPU data.
453 static bool __printk_percpu_data_ready __read_mostly;
455 bool printk_percpu_data_ready(void)
457 return __printk_percpu_data_ready;
460 /* Return log buffer address */
461 char *log_buf_addr_get(void)
466 /* Return log buffer size */
467 u32 log_buf_len_get(void)
472 /* human readable text of the record */
473 static char *log_text(const struct printk_log *msg)
475 return (char *)msg + sizeof(struct printk_log);
478 /* optional key/value pair dictionary attached to the record */
479 static char *log_dict(const struct printk_log *msg)
481 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
484 /* get record by index; idx must point to valid msg */
485 static struct printk_log *log_from_idx(u32 idx)
487 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
490 * A length == 0 record is the end of buffer marker. Wrap around and
491 * read the message at the start of the buffer.
494 return (struct printk_log *)log_buf;
498 /* get next record; idx must point to valid msg */
499 static u32 log_next(u32 idx)
501 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
503 /* length == 0 indicates the end of the buffer; wrap */
505 * A length == 0 record is the end of buffer marker. Wrap around and
506 * read the message at the start of the buffer as *this* one, and
507 * return the one after that.
510 msg = (struct printk_log *)log_buf;
513 return idx + msg->len;
517 * Check whether there is enough free space for the given message.
519 * The same values of first_idx and next_idx mean that the buffer
520 * is either empty or full.
522 * If the buffer is empty, we must respect the position of the indexes.
523 * They cannot be reset to the beginning of the buffer.
525 static int logbuf_has_space(u32 msg_size, bool empty)
529 if (log_next_idx > log_first_idx || empty)
530 free = max(log_buf_len - log_next_idx, log_first_idx);
532 free = log_first_idx - log_next_idx;
535 * We need space also for an empty header that signalizes wrapping
538 return free >= msg_size + sizeof(struct printk_log);
541 static int log_make_free_space(u32 msg_size)
543 while (log_first_seq < log_next_seq &&
544 !logbuf_has_space(msg_size, false)) {
545 /* drop old messages until we have enough contiguous space */
546 log_first_idx = log_next(log_first_idx);
550 if (clear_seq < log_first_seq) {
551 clear_seq = log_first_seq;
552 clear_idx = log_first_idx;
555 /* sequence numbers are equal, so the log buffer is empty */
556 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
562 /* compute the message size including the padding bytes */
563 static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
567 size = sizeof(struct printk_log) + text_len + dict_len;
568 *pad_len = (-size) & (LOG_ALIGN - 1);
575 * Define how much of the log buffer we could take at maximum. The value
576 * must be greater than two. Note that only half of the buffer is available
577 * when the index points to the middle.
579 #define MAX_LOG_TAKE_PART 4
580 static const char trunc_msg[] = "<truncated>";
582 static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
583 u16 *dict_len, u32 *pad_len)
586 * The message should not take the whole buffer. Otherwise, it might
587 * get removed too soon.
589 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
590 if (*text_len > max_text_len)
591 *text_len = max_text_len;
592 /* enable the warning message */
593 *trunc_msg_len = strlen(trunc_msg);
594 /* disable the "dict" completely */
596 /* compute the size again, count also the warning message */
597 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
600 /* insert record into the buffer, discard old ones, update heads */
601 static int log_store(int facility, int level,
602 enum log_flags flags, u64 ts_nsec,
603 const char *dict, u16 dict_len,
604 const char *text, u16 text_len)
606 struct printk_log *msg;
608 u16 trunc_msg_len = 0;
610 /* number of '\0' padding bytes to next message */
611 size = msg_used_size(text_len, dict_len, &pad_len);
613 if (log_make_free_space(size)) {
614 /* truncate the message if it is too long for empty buffer */
615 size = truncate_msg(&text_len, &trunc_msg_len,
616 &dict_len, &pad_len);
617 /* survive when the log buffer is too small for trunc_msg */
618 if (log_make_free_space(size))
622 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
624 * This message + an additional empty header does not fit
625 * at the end of the buffer. Add an empty header with len == 0
626 * to signify a wrap around.
628 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
633 msg = (struct printk_log *)(log_buf + log_next_idx);
634 memcpy(log_text(msg), text, text_len);
635 msg->text_len = text_len;
637 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
638 msg->text_len += trunc_msg_len;
640 memcpy(log_dict(msg), dict, dict_len);
641 msg->dict_len = dict_len;
642 msg->facility = facility;
643 msg->level = level & 7;
644 msg->flags = flags & 0x1f;
646 msg->ts_nsec = ts_nsec;
648 msg->ts_nsec = local_clock();
649 memset(log_dict(msg) + dict_len, 0, pad_len);
653 log_next_idx += msg->len;
656 return msg->text_len;
659 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
661 static int syslog_action_restricted(int type)
666 * Unless restricted, we allow "read all" and "get buffer size"
669 return type != SYSLOG_ACTION_READ_ALL &&
670 type != SYSLOG_ACTION_SIZE_BUFFER;
673 static int check_syslog_permissions(int type, int source)
676 * If this is from /proc/kmsg and we've already opened it, then we've
677 * already done the capabilities checks at open time.
679 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
682 if (syslog_action_restricted(type)) {
683 if (capable(CAP_SYSLOG))
686 * For historical reasons, accept CAP_SYS_ADMIN too, with
689 if (capable(CAP_SYS_ADMIN)) {
690 pr_warn_once("%s (%d): Attempt to access syslog with "
691 "CAP_SYS_ADMIN but no CAP_SYSLOG "
693 current->comm, task_pid_nr(current));
699 return security_syslog(type);
702 static void append_char(char **pp, char *e, char c)
708 static ssize_t msg_print_ext_header(char *buf, size_t size,
709 struct printk_log *msg, u64 seq)
711 u64 ts_usec = msg->ts_nsec;
713 do_div(ts_usec, 1000);
715 return scnprintf(buf, size, "%u,%llu,%llu,%c;",
716 (msg->facility << 3) | msg->level, seq, ts_usec,
717 msg->flags & LOG_CONT ? 'c' : '-');
720 static ssize_t msg_print_ext_body(char *buf, size_t size,
721 char *dict, size_t dict_len,
722 char *text, size_t text_len)
724 char *p = buf, *e = buf + size;
727 /* escape non-printable characters */
728 for (i = 0; i < text_len; i++) {
729 unsigned char c = text[i];
731 if (c < ' ' || c >= 127 || c == '\\')
732 p += scnprintf(p, e - p, "\\x%02x", c);
734 append_char(&p, e, c);
736 append_char(&p, e, '\n');
741 for (i = 0; i < dict_len; i++) {
742 unsigned char c = dict[i];
745 append_char(&p, e, ' ');
750 append_char(&p, e, '\n');
755 if (c < ' ' || c >= 127 || c == '\\') {
756 p += scnprintf(p, e - p, "\\x%02x", c);
760 append_char(&p, e, c);
762 append_char(&p, e, '\n');
768 /* /dev/kmsg - userspace message inject/listen interface */
769 struct devkmsg_user {
772 struct ratelimit_state rs;
774 char buf[CONSOLE_EXT_LOG_MAX];
777 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
780 int level = default_message_loglevel;
781 int facility = 1; /* LOG_USER */
782 struct file *file = iocb->ki_filp;
783 struct devkmsg_user *user = file->private_data;
784 size_t len = iov_iter_count(from);
787 if (!user || len > LOG_LINE_MAX)
790 /* Ignore when user logging is disabled. */
791 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
794 /* Ratelimit when not explicitly enabled. */
795 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
796 if (!___ratelimit(&user->rs, current->comm))
800 buf = kmalloc(len+1, GFP_KERNEL);
805 if (!copy_from_iter_full(buf, len, from)) {
811 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
812 * the decimal value represents 32bit, the lower 3 bit are the log
813 * level, the rest are the log facility.
815 * If no prefix or no userspace facility is specified, we
816 * enforce LOG_USER, to be able to reliably distinguish
817 * kernel-generated messages from userspace-injected ones.
820 if (line[0] == '<') {
824 u = simple_strtoul(line + 1, &endp, 10);
825 if (endp && endp[0] == '>') {
826 level = LOG_LEVEL(u);
827 if (LOG_FACILITY(u) != 0)
828 facility = LOG_FACILITY(u);
835 printk_emit(facility, level, NULL, 0, "%s", line);
840 static ssize_t devkmsg_read(struct file *file, char __user *buf,
841 size_t count, loff_t *ppos)
843 struct devkmsg_user *user = file->private_data;
844 struct printk_log *msg;
851 ret = mutex_lock_interruptible(&user->lock);
856 while (user->seq == log_next_seq) {
857 if (file->f_flags & O_NONBLOCK) {
864 ret = wait_event_interruptible(log_wait,
865 user->seq != log_next_seq);
871 if (user->seq < log_first_seq) {
872 /* our last seen message is gone, return error and reset */
873 user->idx = log_first_idx;
874 user->seq = log_first_seq;
880 msg = log_from_idx(user->idx);
881 len = msg_print_ext_header(user->buf, sizeof(user->buf),
883 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
884 log_dict(msg), msg->dict_len,
885 log_text(msg), msg->text_len);
887 user->idx = log_next(user->idx);
896 if (copy_to_user(buf, user->buf, len)) {
902 mutex_unlock(&user->lock);
906 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
908 struct devkmsg_user *user = file->private_data;
919 /* the first record */
920 user->idx = log_first_idx;
921 user->seq = log_first_seq;
925 * The first record after the last SYSLOG_ACTION_CLEAR,
926 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
927 * changes no global state, and does not clear anything.
929 user->idx = clear_idx;
930 user->seq = clear_seq;
933 /* after the last record */
934 user->idx = log_next_idx;
935 user->seq = log_next_seq;
944 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
946 struct devkmsg_user *user = file->private_data;
950 return EPOLLERR|EPOLLNVAL;
952 poll_wait(file, &log_wait, wait);
955 if (user->seq < log_next_seq) {
956 /* return error when data has vanished underneath us */
957 if (user->seq < log_first_seq)
958 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
960 ret = EPOLLIN|EPOLLRDNORM;
967 static int devkmsg_open(struct inode *inode, struct file *file)
969 struct devkmsg_user *user;
972 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
975 /* write-only does not need any file context */
976 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
977 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
983 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
987 ratelimit_default_init(&user->rs);
988 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
990 mutex_init(&user->lock);
993 user->idx = log_first_idx;
994 user->seq = log_first_seq;
997 file->private_data = user;
1001 static int devkmsg_release(struct inode *inode, struct file *file)
1003 struct devkmsg_user *user = file->private_data;
1008 ratelimit_state_exit(&user->rs);
1010 mutex_destroy(&user->lock);
1015 const struct file_operations kmsg_fops = {
1016 .open = devkmsg_open,
1017 .read = devkmsg_read,
1018 .write_iter = devkmsg_write,
1019 .llseek = devkmsg_llseek,
1020 .poll = devkmsg_poll,
1021 .release = devkmsg_release,
1024 #ifdef CONFIG_CRASH_CORE
1026 * This appends the listed symbols to /proc/vmcore
1028 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1029 * obtain access to symbols that are otherwise very difficult to locate. These
1030 * symbols are specifically used so that utilities can access and extract the
1031 * dmesg log from a vmcore file after a crash.
1033 void log_buf_vmcoreinfo_setup(void)
1035 VMCOREINFO_SYMBOL(log_buf);
1036 VMCOREINFO_SYMBOL(log_buf_len);
1037 VMCOREINFO_SYMBOL(log_first_idx);
1038 VMCOREINFO_SYMBOL(clear_idx);
1039 VMCOREINFO_SYMBOL(log_next_idx);
1041 * Export struct printk_log size and field offsets. User space tools can
1042 * parse it and detect any changes to structure down the line.
1044 VMCOREINFO_STRUCT_SIZE(printk_log);
1045 VMCOREINFO_OFFSET(printk_log, ts_nsec);
1046 VMCOREINFO_OFFSET(printk_log, len);
1047 VMCOREINFO_OFFSET(printk_log, text_len);
1048 VMCOREINFO_OFFSET(printk_log, dict_len);
1052 /* requested log_buf_len from kernel cmdline */
1053 static unsigned long __initdata new_log_buf_len;
1055 /* we practice scaling the ring buffer by powers of 2 */
1056 static void __init log_buf_len_update(u64 size)
1058 if (size > (u64)LOG_BUF_LEN_MAX) {
1059 size = (u64)LOG_BUF_LEN_MAX;
1060 pr_err("log_buf over 2G is not supported.\n");
1064 size = roundup_pow_of_two(size);
1065 if (size > log_buf_len)
1066 new_log_buf_len = (unsigned long)size;
1069 /* save requested log_buf_len since it's too early to process it */
1070 static int __init log_buf_len_setup(char *str)
1077 size = memparse(str, &str);
1079 log_buf_len_update(size);
1083 early_param("log_buf_len", log_buf_len_setup);
1086 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1088 static void __init log_buf_add_cpu(void)
1090 unsigned int cpu_extra;
1093 * archs should set up cpu_possible_bits properly with
1094 * set_cpu_possible() after setup_arch() but just in
1095 * case lets ensure this is valid.
1097 if (num_possible_cpus() == 1)
1100 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1102 /* by default this will only continue through for large > 64 CPUs */
1103 if (cpu_extra <= __LOG_BUF_LEN / 2)
1106 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1107 __LOG_CPU_MAX_BUF_LEN);
1108 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1110 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1112 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1114 #else /* !CONFIG_SMP */
1115 static inline void log_buf_add_cpu(void) {}
1116 #endif /* CONFIG_SMP */
1118 static void __init set_percpu_data_ready(void)
1121 /* Make sure we set this flag only after printk_safe() init is done */
1123 __printk_percpu_data_ready = true;
1126 void __init setup_log_buf(int early)
1128 unsigned long flags;
1133 * Some archs call setup_log_buf() multiple times - first is very
1134 * early, e.g. from setup_arch(), and second - when percpu_areas
1138 set_percpu_data_ready();
1140 if (log_buf != __log_buf)
1143 if (!early && !new_log_buf_len)
1146 if (!new_log_buf_len)
1151 memblock_virt_alloc(new_log_buf_len, LOG_ALIGN);
1153 new_log_buf = memblock_virt_alloc_nopanic(new_log_buf_len,
1157 if (unlikely(!new_log_buf)) {
1158 pr_err("log_buf_len: %lu bytes not available\n",
1163 logbuf_lock_irqsave(flags);
1164 log_buf_len = new_log_buf_len;
1165 log_buf = new_log_buf;
1166 new_log_buf_len = 0;
1167 free = __LOG_BUF_LEN - log_next_idx;
1168 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1169 logbuf_unlock_irqrestore(flags);
1171 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1172 pr_info("early log buf free: %u(%u%%)\n",
1173 free, (free * 100) / __LOG_BUF_LEN);
1176 static bool __read_mostly ignore_loglevel;
1178 static int __init ignore_loglevel_setup(char *str)
1180 ignore_loglevel = true;
1181 pr_info("debug: ignoring loglevel setting.\n");
1186 early_param("ignore_loglevel", ignore_loglevel_setup);
1187 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1188 MODULE_PARM_DESC(ignore_loglevel,
1189 "ignore loglevel setting (prints all kernel messages to the console)");
1191 static bool suppress_message_printing(int level)
1193 return (level >= console_loglevel && !ignore_loglevel);
1196 #ifdef CONFIG_BOOT_PRINTK_DELAY
1198 static int boot_delay; /* msecs delay after each printk during bootup */
1199 static unsigned long long loops_per_msec; /* based on boot_delay */
1201 static int __init boot_delay_setup(char *str)
1205 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1206 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1208 get_option(&str, &boot_delay);
1209 if (boot_delay > 10 * 1000)
1212 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1213 "HZ: %d, loops_per_msec: %llu\n",
1214 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1217 early_param("boot_delay", boot_delay_setup);
1219 static void boot_delay_msec(int level)
1221 unsigned long long k;
1222 unsigned long timeout;
1224 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1225 || suppress_message_printing(level)) {
1229 k = (unsigned long long)loops_per_msec * boot_delay;
1231 timeout = jiffies + msecs_to_jiffies(boot_delay);
1236 * use (volatile) jiffies to prevent
1237 * compiler reduction; loop termination via jiffies
1238 * is secondary and may or may not happen.
1240 if (time_after(jiffies, timeout))
1242 touch_nmi_watchdog();
1246 static inline void boot_delay_msec(int level)
1251 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1252 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1254 static size_t print_time(u64 ts, char *buf)
1256 unsigned long rem_nsec;
1261 rem_nsec = do_div(ts, 1000000000);
1264 return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
1266 return sprintf(buf, "[%5lu.%06lu] ",
1267 (unsigned long)ts, rem_nsec / 1000);
1270 static size_t print_prefix(const struct printk_log *msg, bool syslog, char *buf)
1273 unsigned int prefix = (msg->facility << 3) | msg->level;
1277 len += sprintf(buf, "<%u>", prefix);
1282 else if (prefix > 99)
1284 else if (prefix > 9)
1289 len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
1293 static size_t msg_print_text(const struct printk_log *msg, bool syslog, char *buf, size_t size)
1295 const char *text = log_text(msg);
1296 size_t text_size = msg->text_len;
1300 const char *next = memchr(text, '\n', text_size);
1304 text_len = next - text;
1306 text_size -= next - text;
1308 text_len = text_size;
1312 if (print_prefix(msg, syslog, NULL) +
1313 text_len + 1 >= size - len)
1316 len += print_prefix(msg, syslog, buf + len);
1317 memcpy(buf + len, text, text_len);
1321 /* SYSLOG_ACTION_* buffer size only calculation */
1322 len += print_prefix(msg, syslog, NULL);
1333 static int syslog_print(char __user *buf, int size)
1336 struct printk_log *msg;
1339 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1348 if (syslog_seq < log_first_seq) {
1349 /* messages are gone, move to first one */
1350 syslog_seq = log_first_seq;
1351 syslog_idx = log_first_idx;
1354 if (syslog_seq == log_next_seq) {
1355 logbuf_unlock_irq();
1359 skip = syslog_partial;
1360 msg = log_from_idx(syslog_idx);
1361 n = msg_print_text(msg, true, text, LOG_LINE_MAX + PREFIX_MAX);
1362 if (n - syslog_partial <= size) {
1363 /* message fits into buffer, move forward */
1364 syslog_idx = log_next(syslog_idx);
1366 n -= syslog_partial;
1369 /* partial read(), remember position */
1371 syslog_partial += n;
1374 logbuf_unlock_irq();
1379 if (copy_to_user(buf, text + skip, n)) {
1394 static int syslog_print_all(char __user *buf, int size, bool clear)
1402 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1408 * Find first record that fits, including all following records,
1409 * into the user-provided buffer for this dump.
1413 while (seq < log_next_seq) {
1414 struct printk_log *msg = log_from_idx(idx);
1416 len += msg_print_text(msg, true, NULL, 0);
1417 idx = log_next(idx);
1421 /* move first record forward until length fits into the buffer */
1424 while (len > size && seq < log_next_seq) {
1425 struct printk_log *msg = log_from_idx(idx);
1427 len -= msg_print_text(msg, true, NULL, 0);
1428 idx = log_next(idx);
1432 /* last message fitting into this dump */
1433 next_seq = log_next_seq;
1436 while (len >= 0 && seq < next_seq) {
1437 struct printk_log *msg = log_from_idx(idx);
1440 textlen = msg_print_text(msg, true, text,
1441 LOG_LINE_MAX + PREFIX_MAX);
1446 idx = log_next(idx);
1449 logbuf_unlock_irq();
1450 if (copy_to_user(buf + len, text, textlen))
1456 if (seq < log_first_seq) {
1457 /* messages are gone, move to next one */
1458 seq = log_first_seq;
1459 idx = log_first_idx;
1464 clear_seq = log_next_seq;
1465 clear_idx = log_next_idx;
1467 logbuf_unlock_irq();
1473 static void syslog_clear(void)
1476 clear_seq = log_next_seq;
1477 clear_idx = log_next_idx;
1478 logbuf_unlock_irq();
1481 int do_syslog(int type, char __user *buf, int len, int source)
1484 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1487 error = check_syslog_permissions(type, source);
1492 case SYSLOG_ACTION_CLOSE: /* Close log */
1494 case SYSLOG_ACTION_OPEN: /* Open log */
1496 case SYSLOG_ACTION_READ: /* Read from log */
1497 if (!buf || len < 0)
1501 if (!access_ok(VERIFY_WRITE, buf, len))
1503 error = wait_event_interruptible(log_wait,
1504 syslog_seq != log_next_seq);
1507 error = syslog_print(buf, len);
1509 /* Read/clear last kernel messages */
1510 case SYSLOG_ACTION_READ_CLEAR:
1513 /* Read last kernel messages */
1514 case SYSLOG_ACTION_READ_ALL:
1515 if (!buf || len < 0)
1519 if (!access_ok(VERIFY_WRITE, buf, len))
1521 error = syslog_print_all(buf, len, clear);
1523 /* Clear ring buffer */
1524 case SYSLOG_ACTION_CLEAR:
1527 /* Disable logging to console */
1528 case SYSLOG_ACTION_CONSOLE_OFF:
1529 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1530 saved_console_loglevel = console_loglevel;
1531 console_loglevel = minimum_console_loglevel;
1533 /* Enable logging to console */
1534 case SYSLOG_ACTION_CONSOLE_ON:
1535 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1536 console_loglevel = saved_console_loglevel;
1537 saved_console_loglevel = LOGLEVEL_DEFAULT;
1540 /* Set level of messages printed to console */
1541 case SYSLOG_ACTION_CONSOLE_LEVEL:
1542 if (len < 1 || len > 8)
1544 if (len < minimum_console_loglevel)
1545 len = minimum_console_loglevel;
1546 console_loglevel = len;
1547 /* Implicitly re-enable logging to console */
1548 saved_console_loglevel = LOGLEVEL_DEFAULT;
1550 /* Number of chars in the log buffer */
1551 case SYSLOG_ACTION_SIZE_UNREAD:
1553 if (syslog_seq < log_first_seq) {
1554 /* messages are gone, move to first one */
1555 syslog_seq = log_first_seq;
1556 syslog_idx = log_first_idx;
1559 if (source == SYSLOG_FROM_PROC) {
1561 * Short-cut for poll(/"proc/kmsg") which simply checks
1562 * for pending data, not the size; return the count of
1563 * records, not the length.
1565 error = log_next_seq - syslog_seq;
1567 u64 seq = syslog_seq;
1568 u32 idx = syslog_idx;
1570 while (seq < log_next_seq) {
1571 struct printk_log *msg = log_from_idx(idx);
1573 error += msg_print_text(msg, true, NULL, 0);
1574 idx = log_next(idx);
1577 error -= syslog_partial;
1579 logbuf_unlock_irq();
1581 /* Size of the log buffer */
1582 case SYSLOG_ACTION_SIZE_BUFFER:
1583 error = log_buf_len;
1593 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1595 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1599 * Special console_lock variants that help to reduce the risk of soft-lockups.
1600 * They allow to pass console_lock to another printk() call using a busy wait.
1603 #ifdef CONFIG_LOCKDEP
1604 static struct lockdep_map console_owner_dep_map = {
1605 .name = "console_owner"
1609 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1610 static struct task_struct *console_owner;
1611 static bool console_waiter;
1614 * console_lock_spinning_enable - mark beginning of code where another
1615 * thread might safely busy wait
1617 * This basically converts console_lock into a spinlock. This marks
1618 * the section where the console_lock owner can not sleep, because
1619 * there may be a waiter spinning (like a spinlock). Also it must be
1620 * ready to hand over the lock at the end of the section.
1622 static void console_lock_spinning_enable(void)
1624 raw_spin_lock(&console_owner_lock);
1625 console_owner = current;
1626 raw_spin_unlock(&console_owner_lock);
1628 /* The waiter may spin on us after setting console_owner */
1629 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1633 * console_lock_spinning_disable_and_check - mark end of code where another
1634 * thread was able to busy wait and check if there is a waiter
1636 * This is called at the end of the section where spinning is allowed.
1637 * It has two functions. First, it is a signal that it is no longer
1638 * safe to start busy waiting for the lock. Second, it checks if
1639 * there is a busy waiter and passes the lock rights to her.
1641 * Important: Callers lose the lock if there was a busy waiter.
1642 * They must not touch items synchronized by console_lock
1645 * Return: 1 if the lock rights were passed, 0 otherwise.
1647 static int console_lock_spinning_disable_and_check(void)
1651 raw_spin_lock(&console_owner_lock);
1652 waiter = READ_ONCE(console_waiter);
1653 console_owner = NULL;
1654 raw_spin_unlock(&console_owner_lock);
1657 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1661 /* The waiter is now free to continue */
1662 WRITE_ONCE(console_waiter, false);
1664 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1667 * Hand off console_lock to waiter. The waiter will perform
1668 * the up(). After this, the waiter is the console_lock owner.
1670 mutex_release(&console_lock_dep_map, 1, _THIS_IP_);
1675 * console_trylock_spinning - try to get console_lock by busy waiting
1677 * This allows to busy wait for the console_lock when the current
1678 * owner is running in specially marked sections. It means that
1679 * the current owner is running and cannot reschedule until it
1680 * is ready to lose the lock.
1682 * Return: 1 if we got the lock, 0 othrewise
1684 static int console_trylock_spinning(void)
1686 struct task_struct *owner = NULL;
1689 unsigned long flags;
1691 if (console_trylock())
1694 printk_safe_enter_irqsave(flags);
1696 raw_spin_lock(&console_owner_lock);
1697 owner = READ_ONCE(console_owner);
1698 waiter = READ_ONCE(console_waiter);
1699 if (!waiter && owner && owner != current) {
1700 WRITE_ONCE(console_waiter, true);
1703 raw_spin_unlock(&console_owner_lock);
1706 * If there is an active printk() writing to the
1707 * consoles, instead of having it write our data too,
1708 * see if we can offload that load from the active
1709 * printer, and do some printing ourselves.
1710 * Go into a spin only if there isn't already a waiter
1711 * spinning, and there is an active printer, and
1712 * that active printer isn't us (recursive printk?).
1715 printk_safe_exit_irqrestore(flags);
1719 /* We spin waiting for the owner to release us */
1720 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1721 /* Owner will clear console_waiter on hand off */
1722 while (READ_ONCE(console_waiter))
1724 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1726 printk_safe_exit_irqrestore(flags);
1728 * The owner passed the console lock to us.
1729 * Since we did not spin on console lock, annotate
1730 * this as a trylock. Otherwise lockdep will
1733 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1739 * Call the console drivers, asking them to write out
1740 * log_buf[start] to log_buf[end - 1].
1741 * The console_lock must be held.
1743 static void call_console_drivers(const char *ext_text, size_t ext_len,
1744 const char *text, size_t len)
1746 struct console *con;
1748 trace_console_rcuidle(text, len);
1750 if (!console_drivers)
1753 for_each_console(con) {
1754 if (exclusive_console && con != exclusive_console)
1756 if (!(con->flags & CON_ENABLED))
1760 if (!cpu_online(smp_processor_id()) &&
1761 !(con->flags & CON_ANYTIME))
1763 if (con->flags & CON_EXTENDED)
1764 con->write(con, ext_text, ext_len);
1766 con->write(con, text, len);
1770 int printk_delay_msec __read_mostly;
1772 static inline void printk_delay(void)
1774 if (unlikely(printk_delay_msec)) {
1775 int m = printk_delay_msec;
1779 touch_nmi_watchdog();
1785 * Continuation lines are buffered, and not committed to the record buffer
1786 * until the line is complete, or a race forces it. The line fragments
1787 * though, are printed immediately to the consoles to ensure everything has
1788 * reached the console in case of a kernel crash.
1790 static struct cont {
1791 char buf[LOG_LINE_MAX];
1792 size_t len; /* length == 0 means unused buffer */
1793 struct task_struct *owner; /* task of first print*/
1794 u64 ts_nsec; /* time of first print */
1795 u8 level; /* log level of first message */
1796 u8 facility; /* log facility of first message */
1797 enum log_flags flags; /* prefix, newline flags */
1800 static void cont_flush(void)
1805 log_store(cont.facility, cont.level, cont.flags, cont.ts_nsec,
1806 NULL, 0, cont.buf, cont.len);
1810 static bool cont_add(int facility, int level, enum log_flags flags, const char *text, size_t len)
1813 * If ext consoles are present, flush and skip in-kernel
1814 * continuation. See nr_ext_console_drivers definition. Also, if
1815 * the line gets too long, split it up in separate records.
1817 if (nr_ext_console_drivers || cont.len + len > sizeof(cont.buf)) {
1823 cont.facility = facility;
1825 cont.owner = current;
1826 cont.ts_nsec = local_clock();
1830 memcpy(cont.buf + cont.len, text, len);
1833 // The original flags come from the first line,
1834 // but later continuations can add a newline.
1835 if (flags & LOG_NEWLINE) {
1836 cont.flags |= LOG_NEWLINE;
1840 if (cont.len > (sizeof(cont.buf) * 80) / 100)
1846 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)
1849 * If an earlier line was buffered, and we're a continuation
1850 * write from the same process, try to add it to the buffer.
1853 if (cont.owner == current && (lflags & LOG_CONT)) {
1854 if (cont_add(facility, level, lflags, text, text_len))
1857 /* Otherwise, make sure it's flushed */
1861 /* Skip empty continuation lines that couldn't be added - they just flush */
1862 if (!text_len && (lflags & LOG_CONT))
1865 /* If it doesn't end in a newline, try to buffer the current line */
1866 if (!(lflags & LOG_NEWLINE)) {
1867 if (cont_add(facility, level, lflags, text, text_len))
1871 /* Store it in the record log */
1872 return log_store(facility, level, lflags, 0, dict, dictlen, text, text_len);
1875 /* Must be called under logbuf_lock. */
1876 int vprintk_store(int facility, int level,
1877 const char *dict, size_t dictlen,
1878 const char *fmt, va_list args)
1880 static char textbuf[LOG_LINE_MAX];
1881 char *text = textbuf;
1883 enum log_flags lflags = 0;
1886 * The printf needs to come first; we need the syslog
1887 * prefix which might be passed-in as a parameter.
1889 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1891 /* mark and strip a trailing newline */
1892 if (text_len && text[text_len-1] == '\n') {
1894 lflags |= LOG_NEWLINE;
1897 /* strip kernel syslog prefix and extract log level or control flags */
1898 if (facility == 0) {
1901 while ((kern_level = printk_get_level(text)) != 0) {
1902 switch (kern_level) {
1904 if (level == LOGLEVEL_DEFAULT)
1905 level = kern_level - '0';
1907 case 'd': /* KERN_DEFAULT */
1908 lflags |= LOG_PREFIX;
1910 case 'c': /* KERN_CONT */
1919 if (level == LOGLEVEL_DEFAULT)
1920 level = default_message_loglevel;
1923 lflags |= LOG_PREFIX|LOG_NEWLINE;
1925 return log_output(facility, level, lflags,
1926 dict, dictlen, text, text_len);
1929 asmlinkage int vprintk_emit(int facility, int level,
1930 const char *dict, size_t dictlen,
1931 const char *fmt, va_list args)
1934 bool in_sched = false, pending_output;
1935 unsigned long flags;
1938 if (level == LOGLEVEL_SCHED) {
1939 level = LOGLEVEL_DEFAULT;
1943 boot_delay_msec(level);
1946 /* This stops the holder of console_sem just where we want him */
1947 logbuf_lock_irqsave(flags);
1948 curr_log_seq = log_next_seq;
1949 printed_len = vprintk_store(facility, level, dict, dictlen, fmt, args);
1950 pending_output = (curr_log_seq != log_next_seq);
1951 logbuf_unlock_irqrestore(flags);
1953 /* If called from the scheduler, we can not call up(). */
1954 if (!in_sched && pending_output) {
1956 * Disable preemption to avoid being preempted while holding
1957 * console_sem which would prevent anyone from printing to
1962 * Try to acquire and then immediately release the console
1963 * semaphore. The release will print out buffers and wake up
1964 * /dev/kmsg and syslog() users.
1966 if (console_trylock_spinning())
1975 EXPORT_SYMBOL(vprintk_emit);
1977 asmlinkage int vprintk(const char *fmt, va_list args)
1979 return vprintk_func(fmt, args);
1981 EXPORT_SYMBOL(vprintk);
1983 asmlinkage int printk_emit(int facility, int level,
1984 const char *dict, size_t dictlen,
1985 const char *fmt, ...)
1990 va_start(args, fmt);
1991 r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1996 EXPORT_SYMBOL(printk_emit);
1998 int vprintk_default(const char *fmt, va_list args)
2002 #ifdef CONFIG_KGDB_KDB
2003 /* Allow to pass printk() to kdb but avoid a recursion. */
2004 if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
2005 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
2009 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
2013 EXPORT_SYMBOL_GPL(vprintk_default);
2016 * printk - print a kernel message
2017 * @fmt: format string
2019 * This is printk(). It can be called from any context. We want it to work.
2021 * We try to grab the console_lock. If we succeed, it's easy - we log the
2022 * output and call the console drivers. If we fail to get the semaphore, we
2023 * place the output into the log buffer and return. The current holder of
2024 * the console_sem will notice the new output in console_unlock(); and will
2025 * send it to the consoles before releasing the lock.
2027 * One effect of this deferred printing is that code which calls printk() and
2028 * then changes console_loglevel may break. This is because console_loglevel
2029 * is inspected when the actual printing occurs.
2034 * See the vsnprintf() documentation for format string extensions over C99.
2036 asmlinkage __visible int printk(const char *fmt, ...)
2041 va_start(args, fmt);
2042 r = vprintk_func(fmt, args);
2047 EXPORT_SYMBOL(printk);
2049 #else /* CONFIG_PRINTK */
2051 #define LOG_LINE_MAX 0
2052 #define PREFIX_MAX 0
2054 static u64 syslog_seq;
2055 static u32 syslog_idx;
2056 static u64 console_seq;
2057 static u32 console_idx;
2058 static u64 exclusive_console_stop_seq;
2059 static u64 log_first_seq;
2060 static u32 log_first_idx;
2061 static u64 log_next_seq;
2062 static char *log_text(const struct printk_log *msg) { return NULL; }
2063 static char *log_dict(const struct printk_log *msg) { return NULL; }
2064 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2065 static u32 log_next(u32 idx) { return 0; }
2066 static ssize_t msg_print_ext_header(char *buf, size_t size,
2067 struct printk_log *msg,
2068 u64 seq) { return 0; }
2069 static ssize_t msg_print_ext_body(char *buf, size_t size,
2070 char *dict, size_t dict_len,
2071 char *text, size_t text_len) { return 0; }
2072 static void console_lock_spinning_enable(void) { }
2073 static int console_lock_spinning_disable_and_check(void) { return 0; }
2074 static void call_console_drivers(const char *ext_text, size_t ext_len,
2075 const char *text, size_t len) {}
2076 static size_t msg_print_text(const struct printk_log *msg,
2077 bool syslog, char *buf, size_t size) { return 0; }
2078 static bool suppress_message_printing(int level) { return false; }
2080 #endif /* CONFIG_PRINTK */
2082 #ifdef CONFIG_EARLY_PRINTK
2083 struct console *early_console;
2085 asmlinkage __visible void early_printk(const char *fmt, ...)
2095 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2098 early_console->write(early_console, buf, n);
2102 static int __add_preferred_console(char *name, int idx, char *options,
2105 struct console_cmdline *c;
2109 * See if this tty is not yet registered, and
2110 * if we have a slot free.
2112 for (i = 0, c = console_cmdline;
2113 i < MAX_CMDLINECONSOLES && c->name[0];
2115 if (strcmp(c->name, name) == 0 && c->index == idx) {
2117 preferred_console = i;
2121 if (i == MAX_CMDLINECONSOLES)
2124 preferred_console = i;
2125 strlcpy(c->name, name, sizeof(c->name));
2126 c->options = options;
2127 braille_set_options(c, brl_options);
2133 static int __init console_msg_format_setup(char *str)
2135 if (!strcmp(str, "syslog"))
2136 console_msg_format = MSG_FORMAT_SYSLOG;
2137 if (!strcmp(str, "default"))
2138 console_msg_format = MSG_FORMAT_DEFAULT;
2141 __setup("console_msg_format=", console_msg_format_setup);
2144 * Set up a console. Called via do_early_param() in init/main.c
2145 * for each "console=" parameter in the boot command line.
2147 static int __init console_setup(char *str)
2149 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2150 char *s, *options, *brl_options = NULL;
2154 * console="" or console=null have been suggested as a way to
2155 * disable console output. Use ttynull that has been created
2156 * for exacly this purpose.
2158 if (str[0] == 0 || strcmp(str, "null") == 0) {
2159 __add_preferred_console("ttynull", 0, NULL, NULL);
2163 if (_braille_console_setup(&str, &brl_options))
2167 * Decode str into name, index, options.
2169 if (str[0] >= '0' && str[0] <= '9') {
2170 strcpy(buf, "ttyS");
2171 strncpy(buf + 4, str, sizeof(buf) - 5);
2173 strncpy(buf, str, sizeof(buf) - 1);
2175 buf[sizeof(buf) - 1] = 0;
2176 options = strchr(str, ',');
2180 if (!strcmp(str, "ttya"))
2181 strcpy(buf, "ttyS0");
2182 if (!strcmp(str, "ttyb"))
2183 strcpy(buf, "ttyS1");
2185 for (s = buf; *s; s++)
2186 if (isdigit(*s) || *s == ',')
2188 idx = simple_strtoul(s, NULL, 10);
2191 __add_preferred_console(buf, idx, options, brl_options);
2192 console_set_on_cmdline = 1;
2195 __setup("console=", console_setup);
2198 * add_preferred_console - add a device to the list of preferred consoles.
2199 * @name: device name
2200 * @idx: device index
2201 * @options: options for this console
2203 * The last preferred console added will be used for kernel messages
2204 * and stdin/out/err for init. Normally this is used by console_setup
2205 * above to handle user-supplied console arguments; however it can also
2206 * be used by arch-specific code either to override the user or more
2207 * commonly to provide a default console (ie from PROM variables) when
2208 * the user has not supplied one.
2210 int add_preferred_console(char *name, int idx, char *options)
2212 return __add_preferred_console(name, idx, options, NULL);
2215 bool console_suspend_enabled = true;
2216 EXPORT_SYMBOL(console_suspend_enabled);
2218 static int __init console_suspend_disable(char *str)
2220 console_suspend_enabled = false;
2223 __setup("no_console_suspend", console_suspend_disable);
2224 module_param_named(console_suspend, console_suspend_enabled,
2225 bool, S_IRUGO | S_IWUSR);
2226 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2227 " and hibernate operations");
2230 * suspend_console - suspend the console subsystem
2232 * This disables printk() while we go into suspend states
2234 void suspend_console(void)
2236 if (!console_suspend_enabled)
2238 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2240 console_suspended = 1;
2244 void resume_console(void)
2246 if (!console_suspend_enabled)
2249 console_suspended = 0;
2254 * console_cpu_notify - print deferred console messages after CPU hotplug
2257 * If printk() is called from a CPU that is not online yet, the messages
2258 * will be printed on the console only if there are CON_ANYTIME consoles.
2259 * This function is called when a new CPU comes online (or fails to come
2260 * up) or goes offline.
2262 static int console_cpu_notify(unsigned int cpu)
2264 if (!cpuhp_tasks_frozen) {
2265 /* If trylock fails, someone else is doing the printing */
2266 if (console_trylock())
2273 * console_lock - lock the console system for exclusive use.
2275 * Acquires a lock which guarantees that the caller has
2276 * exclusive access to the console system and the console_drivers list.
2278 * Can sleep, returns nothing.
2280 void console_lock(void)
2285 if (console_suspended)
2288 console_may_schedule = 1;
2290 EXPORT_SYMBOL(console_lock);
2293 * console_trylock - try to lock the console system for exclusive use.
2295 * Try to acquire a lock which guarantees that the caller has exclusive
2296 * access to the console system and the console_drivers list.
2298 * returns 1 on success, and 0 on failure to acquire the lock.
2300 int console_trylock(void)
2302 if (down_trylock_console_sem())
2304 if (console_suspended) {
2309 console_may_schedule = 0;
2312 EXPORT_SYMBOL(console_trylock);
2314 int is_console_locked(void)
2316 return console_locked;
2318 EXPORT_SYMBOL(is_console_locked);
2321 * Check if we have any console that is capable of printing while cpu is
2322 * booting or shutting down. Requires console_sem.
2324 static int have_callable_console(void)
2326 struct console *con;
2328 for_each_console(con)
2329 if ((con->flags & CON_ENABLED) &&
2330 (con->flags & CON_ANYTIME))
2337 * Can we actually use the console at this time on this cpu?
2339 * Console drivers may assume that per-cpu resources have been allocated. So
2340 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2341 * call them until this CPU is officially up.
2343 static inline int can_use_console(void)
2345 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2349 * console_unlock - unlock the console system
2351 * Releases the console_lock which the caller holds on the console system
2352 * and the console driver list.
2354 * While the console_lock was held, console output may have been buffered
2355 * by printk(). If this is the case, console_unlock(); emits
2356 * the output prior to releasing the lock.
2358 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2360 * console_unlock(); may be called from any context.
2362 void console_unlock(void)
2364 static char ext_text[CONSOLE_EXT_LOG_MAX];
2365 static char text[LOG_LINE_MAX + PREFIX_MAX];
2366 unsigned long flags;
2367 bool do_cond_resched, retry;
2369 if (console_suspended) {
2375 * Console drivers are called with interrupts disabled, so
2376 * @console_may_schedule should be cleared before; however, we may
2377 * end up dumping a lot of lines, for example, if called from
2378 * console registration path, and should invoke cond_resched()
2379 * between lines if allowable. Not doing so can cause a very long
2380 * scheduling stall on a slow console leading to RCU stall and
2381 * softlockup warnings which exacerbate the issue with more
2382 * messages practically incapacitating the system.
2384 * console_trylock() is not able to detect the preemptive
2385 * context reliably. Therefore the value must be stored before
2386 * and cleared after the the "again" goto label.
2388 do_cond_resched = console_may_schedule;
2390 console_may_schedule = 0;
2393 * We released the console_sem lock, so we need to recheck if
2394 * cpu is online and (if not) is there at least one CON_ANYTIME
2397 if (!can_use_console()) {
2404 struct printk_log *msg;
2408 printk_safe_enter_irqsave(flags);
2409 raw_spin_lock(&logbuf_lock);
2410 if (console_seq < log_first_seq) {
2412 "** %llu printk messages dropped **\n",
2413 log_first_seq - console_seq);
2415 /* messages are gone, move to first one */
2416 console_seq = log_first_seq;
2417 console_idx = log_first_idx;
2422 if (console_seq == log_next_seq)
2425 msg = log_from_idx(console_idx);
2426 if (suppress_message_printing(msg->level)) {
2428 * Skip record we have buffered and already printed
2429 * directly to the console when we received it, and
2430 * record that has level above the console loglevel.
2432 console_idx = log_next(console_idx);
2437 /* Output to all consoles once old messages replayed. */
2438 if (unlikely(exclusive_console &&
2439 console_seq >= exclusive_console_stop_seq)) {
2440 exclusive_console = NULL;
2443 len += msg_print_text(msg,
2444 console_msg_format & MSG_FORMAT_SYSLOG,
2446 sizeof(text) - len);
2447 if (nr_ext_console_drivers) {
2448 ext_len = msg_print_ext_header(ext_text,
2451 ext_len += msg_print_ext_body(ext_text + ext_len,
2452 sizeof(ext_text) - ext_len,
2453 log_dict(msg), msg->dict_len,
2454 log_text(msg), msg->text_len);
2456 console_idx = log_next(console_idx);
2458 raw_spin_unlock(&logbuf_lock);
2461 * While actively printing out messages, if another printk()
2462 * were to occur on another CPU, it may wait for this one to
2463 * finish. This task can not be preempted if there is a
2464 * waiter waiting to take over.
2466 console_lock_spinning_enable();
2468 stop_critical_timings(); /* don't trace print latency */
2469 call_console_drivers(ext_text, ext_len, text, len);
2470 start_critical_timings();
2472 if (console_lock_spinning_disable_and_check()) {
2473 printk_safe_exit_irqrestore(flags);
2477 printk_safe_exit_irqrestore(flags);
2479 if (do_cond_resched)
2485 raw_spin_unlock(&logbuf_lock);
2490 * Someone could have filled up the buffer again, so re-check if there's
2491 * something to flush. In case we cannot trylock the console_sem again,
2492 * there's a new owner and the console_unlock() from them will do the
2493 * flush, no worries.
2495 raw_spin_lock(&logbuf_lock);
2496 retry = console_seq != log_next_seq;
2497 raw_spin_unlock(&logbuf_lock);
2498 printk_safe_exit_irqrestore(flags);
2500 if (retry && console_trylock())
2503 EXPORT_SYMBOL(console_unlock);
2506 * console_conditional_schedule - yield the CPU if required
2508 * If the console code is currently allowed to sleep, and
2509 * if this CPU should yield the CPU to another task, do
2512 * Must be called within console_lock();.
2514 void __sched console_conditional_schedule(void)
2516 if (console_may_schedule)
2519 EXPORT_SYMBOL(console_conditional_schedule);
2521 void console_unblank(void)
2526 * console_unblank can no longer be called in interrupt context unless
2527 * oops_in_progress is set to 1..
2529 if (oops_in_progress) {
2530 if (down_trylock_console_sem() != 0)
2536 console_may_schedule = 0;
2538 if ((c->flags & CON_ENABLED) && c->unblank)
2544 * console_flush_on_panic - flush console content on panic
2546 * Immediately output all pending messages no matter what.
2548 void console_flush_on_panic(void)
2551 * If someone else is holding the console lock, trylock will fail
2552 * and may_schedule may be set. Ignore and proceed to unlock so
2553 * that messages are flushed out. As this can be called from any
2554 * context and we don't want to get preempted while flushing,
2555 * ensure may_schedule is cleared.
2558 console_may_schedule = 0;
2563 * Return the console tty driver structure and its associated index
2565 struct tty_driver *console_device(int *index)
2568 struct tty_driver *driver = NULL;
2571 for_each_console(c) {
2574 driver = c->device(c, index);
2583 * Prevent further output on the passed console device so that (for example)
2584 * serial drivers can disable console output before suspending a port, and can
2585 * re-enable output afterwards.
2587 void console_stop(struct console *console)
2590 console->flags &= ~CON_ENABLED;
2593 EXPORT_SYMBOL(console_stop);
2595 void console_start(struct console *console)
2598 console->flags |= CON_ENABLED;
2601 EXPORT_SYMBOL(console_start);
2603 static int __read_mostly keep_bootcon;
2605 static int __init keep_bootcon_setup(char *str)
2608 pr_info("debug: skip boot console de-registration.\n");
2613 early_param("keep_bootcon", keep_bootcon_setup);
2616 * The console driver calls this routine during kernel initialization
2617 * to register the console printing procedure with printk() and to
2618 * print any messages that were printed by the kernel before the
2619 * console driver was initialized.
2621 * This can happen pretty early during the boot process (because of
2622 * early_printk) - sometimes before setup_arch() completes - be careful
2623 * of what kernel features are used - they may not be initialised yet.
2625 * There are two types of consoles - bootconsoles (early_printk) and
2626 * "real" consoles (everything which is not a bootconsole) which are
2627 * handled differently.
2628 * - Any number of bootconsoles can be registered at any time.
2629 * - As soon as a "real" console is registered, all bootconsoles
2630 * will be unregistered automatically.
2631 * - Once a "real" console is registered, any attempt to register a
2632 * bootconsoles will be rejected
2634 void register_console(struct console *newcon)
2637 unsigned long flags;
2638 struct console *bcon = NULL;
2639 struct console_cmdline *c;
2640 static bool has_preferred;
2642 if (console_drivers)
2643 for_each_console(bcon)
2644 if (WARN(bcon == newcon,
2645 "console '%s%d' already registered\n",
2646 bcon->name, bcon->index))
2650 * before we register a new CON_BOOT console, make sure we don't
2651 * already have a valid console
2653 if (console_drivers && newcon->flags & CON_BOOT) {
2654 /* find the last or real console */
2655 for_each_console(bcon) {
2656 if (!(bcon->flags & CON_BOOT)) {
2657 pr_info("Too late to register bootconsole %s%d\n",
2658 newcon->name, newcon->index);
2664 if (console_drivers && console_drivers->flags & CON_BOOT)
2665 bcon = console_drivers;
2667 if (!has_preferred || bcon || !console_drivers)
2668 has_preferred = preferred_console >= 0;
2671 * See if we want to use this console driver. If we
2672 * didn't select a console we take the first one
2673 * that registers here.
2675 if (!has_preferred) {
2676 if (newcon->index < 0)
2678 if (newcon->setup == NULL ||
2679 newcon->setup(newcon, NULL) == 0) {
2680 newcon->flags |= CON_ENABLED;
2681 if (newcon->device) {
2682 newcon->flags |= CON_CONSDEV;
2683 has_preferred = true;
2689 * See if this console matches one we selected on
2692 for (i = 0, c = console_cmdline;
2693 i < MAX_CMDLINECONSOLES && c->name[0];
2695 if (!newcon->match ||
2696 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2697 /* default matching */
2698 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2699 if (strcmp(c->name, newcon->name) != 0)
2701 if (newcon->index >= 0 &&
2702 newcon->index != c->index)
2704 if (newcon->index < 0)
2705 newcon->index = c->index;
2707 if (_braille_register_console(newcon, c))
2710 if (newcon->setup &&
2711 newcon->setup(newcon, c->options) != 0)
2715 newcon->flags |= CON_ENABLED;
2716 if (i == preferred_console) {
2717 newcon->flags |= CON_CONSDEV;
2718 has_preferred = true;
2723 if (!(newcon->flags & CON_ENABLED))
2727 * If we have a bootconsole, and are switching to a real console,
2728 * don't print everything out again, since when the boot console, and
2729 * the real console are the same physical device, it's annoying to
2730 * see the beginning boot messages twice
2732 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2733 newcon->flags &= ~CON_PRINTBUFFER;
2736 * Put this console in the list - keep the
2737 * preferred driver at the head of the list.
2740 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2741 newcon->next = console_drivers;
2742 console_drivers = newcon;
2744 newcon->next->flags &= ~CON_CONSDEV;
2746 newcon->next = console_drivers->next;
2747 console_drivers->next = newcon;
2750 if (newcon->flags & CON_EXTENDED)
2751 if (!nr_ext_console_drivers++)
2752 pr_info("printk: continuation disabled due to ext consoles, expect more fragments in /dev/kmsg\n");
2754 if (newcon->flags & CON_PRINTBUFFER) {
2756 * console_unlock(); will print out the buffered messages
2759 logbuf_lock_irqsave(flags);
2761 * We're about to replay the log buffer. Only do this to the
2762 * just-registered console to avoid excessive message spam to
2763 * the already-registered consoles.
2765 * Set exclusive_console with disabled interrupts to reduce
2766 * race window with eventual console_flush_on_panic() that
2767 * ignores console_lock.
2769 exclusive_console = newcon;
2770 exclusive_console_stop_seq = console_seq;
2771 console_seq = syslog_seq;
2772 console_idx = syslog_idx;
2773 logbuf_unlock_irqrestore(flags);
2776 console_sysfs_notify();
2779 * By unregistering the bootconsoles after we enable the real console
2780 * we get the "console xxx enabled" message on all the consoles -
2781 * boot consoles, real consoles, etc - this is to ensure that end
2782 * users know there might be something in the kernel's log buffer that
2783 * went to the bootconsole (that they do not see on the real console)
2785 pr_info("%sconsole [%s%d] enabled\n",
2786 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2787 newcon->name, newcon->index);
2789 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2791 /* We need to iterate through all boot consoles, to make
2792 * sure we print everything out, before we unregister them.
2794 for_each_console(bcon)
2795 if (bcon->flags & CON_BOOT)
2796 unregister_console(bcon);
2799 EXPORT_SYMBOL(register_console);
2801 int unregister_console(struct console *console)
2803 struct console *a, *b;
2806 pr_info("%sconsole [%s%d] disabled\n",
2807 (console->flags & CON_BOOT) ? "boot" : "" ,
2808 console->name, console->index);
2810 res = _braille_unregister_console(console);
2816 if (console_drivers == console) {
2817 console_drivers=console->next;
2819 } else if (console_drivers) {
2820 for (a=console_drivers->next, b=console_drivers ;
2821 a; b=a, a=b->next) {
2830 if (!res && (console->flags & CON_EXTENDED))
2831 nr_ext_console_drivers--;
2834 * If this isn't the last console and it has CON_CONSDEV set, we
2835 * need to set it on the next preferred console.
2837 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2838 console_drivers->flags |= CON_CONSDEV;
2840 console->flags &= ~CON_ENABLED;
2842 console_sysfs_notify();
2845 EXPORT_SYMBOL(unregister_console);
2848 * Initialize the console device. This is called *early*, so
2849 * we can't necessarily depend on lots of kernel help here.
2850 * Just do some early initializations, and do the complex setup
2853 void __init console_init(void)
2857 initcall_entry_t *ce;
2859 /* Setup the default TTY line discipline. */
2863 * set up the console device so that later boot sequences can
2864 * inform about problems etc..
2866 ce = __con_initcall_start;
2867 trace_initcall_level("console");
2868 while (ce < __con_initcall_end) {
2869 call = initcall_from_entry(ce);
2870 trace_initcall_start(call);
2872 trace_initcall_finish(call, ret);
2878 * Some boot consoles access data that is in the init section and which will
2879 * be discarded after the initcalls have been run. To make sure that no code
2880 * will access this data, unregister the boot consoles in a late initcall.
2882 * If for some reason, such as deferred probe or the driver being a loadable
2883 * module, the real console hasn't registered yet at this point, there will
2884 * be a brief interval in which no messages are logged to the console, which
2885 * makes it difficult to diagnose problems that occur during this time.
2887 * To mitigate this problem somewhat, only unregister consoles whose memory
2888 * intersects with the init section. Note that all other boot consoles will
2889 * get unregistred when the real preferred console is registered.
2891 static int __init printk_late_init(void)
2893 struct console *con;
2896 for_each_console(con) {
2897 if (!(con->flags & CON_BOOT))
2900 /* Check addresses that might be used for enabled consoles. */
2901 if (init_section_intersects(con, sizeof(*con)) ||
2902 init_section_contains(con->write, 0) ||
2903 init_section_contains(con->read, 0) ||
2904 init_section_contains(con->device, 0) ||
2905 init_section_contains(con->unblank, 0) ||
2906 init_section_contains(con->data, 0)) {
2908 * Please, consider moving the reported consoles out
2909 * of the init section.
2911 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2912 con->name, con->index);
2913 unregister_console(con);
2916 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2917 console_cpu_notify);
2919 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2920 console_cpu_notify, NULL);
2924 late_initcall(printk_late_init);
2926 #if defined CONFIG_PRINTK
2928 * Delayed printk version, for scheduler-internal messages:
2930 #define PRINTK_PENDING_WAKEUP 0x01
2931 #define PRINTK_PENDING_OUTPUT 0x02
2933 static DEFINE_PER_CPU(int, printk_pending);
2935 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2937 int pending = __this_cpu_xchg(printk_pending, 0);
2939 if (pending & PRINTK_PENDING_OUTPUT) {
2940 /* If trylock fails, someone else is doing the printing */
2941 if (console_trylock())
2945 if (pending & PRINTK_PENDING_WAKEUP)
2946 wake_up_interruptible(&log_wait);
2949 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2950 .func = wake_up_klogd_work_func,
2951 .flags = IRQ_WORK_LAZY,
2954 void wake_up_klogd(void)
2956 if (!printk_percpu_data_ready())
2960 if (waitqueue_active(&log_wait)) {
2961 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2962 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2967 void defer_console_output(void)
2969 if (!printk_percpu_data_ready())
2973 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2974 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2978 int vprintk_deferred(const char *fmt, va_list args)
2982 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2983 defer_console_output();
2988 int printk_deferred(const char *fmt, ...)
2993 va_start(args, fmt);
2994 r = vprintk_deferred(fmt, args);
3001 * printk rate limiting, lifted from the networking subsystem.
3003 * This enforces a rate limit: not more than 10 kernel messages
3004 * every 5s to make a denial-of-service attack impossible.
3006 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3008 int __printk_ratelimit(const char *func)
3010 return ___ratelimit(&printk_ratelimit_state, func);
3012 EXPORT_SYMBOL(__printk_ratelimit);
3015 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3016 * @caller_jiffies: pointer to caller's state
3017 * @interval_msecs: minimum interval between prints
3019 * printk_timed_ratelimit() returns true if more than @interval_msecs
3020 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3023 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3024 unsigned int interval_msecs)
3026 unsigned long elapsed = jiffies - *caller_jiffies;
3028 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3031 *caller_jiffies = jiffies;
3034 EXPORT_SYMBOL(printk_timed_ratelimit);
3036 static DEFINE_SPINLOCK(dump_list_lock);
3037 static LIST_HEAD(dump_list);
3040 * kmsg_dump_register - register a kernel log dumper.
3041 * @dumper: pointer to the kmsg_dumper structure
3043 * Adds a kernel log dumper to the system. The dump callback in the
3044 * structure will be called when the kernel oopses or panics and must be
3045 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3047 int kmsg_dump_register(struct kmsg_dumper *dumper)
3049 unsigned long flags;
3052 /* The dump callback needs to be set */
3056 spin_lock_irqsave(&dump_list_lock, flags);
3057 /* Don't allow registering multiple times */
3058 if (!dumper->registered) {
3059 dumper->registered = 1;
3060 list_add_tail_rcu(&dumper->list, &dump_list);
3063 spin_unlock_irqrestore(&dump_list_lock, flags);
3067 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3070 * kmsg_dump_unregister - unregister a kmsg dumper.
3071 * @dumper: pointer to the kmsg_dumper structure
3073 * Removes a dump device from the system. Returns zero on success and
3074 * %-EINVAL otherwise.
3076 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3078 unsigned long flags;
3081 spin_lock_irqsave(&dump_list_lock, flags);
3082 if (dumper->registered) {
3083 dumper->registered = 0;
3084 list_del_rcu(&dumper->list);
3087 spin_unlock_irqrestore(&dump_list_lock, flags);
3092 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3094 static bool always_kmsg_dump;
3095 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3098 * kmsg_dump - dump kernel log to kernel message dumpers.
3099 * @reason: the reason (oops, panic etc) for dumping
3101 * Call each of the registered dumper's dump() callback, which can
3102 * retrieve the kmsg records with kmsg_dump_get_line() or
3103 * kmsg_dump_get_buffer().
3105 void kmsg_dump(enum kmsg_dump_reason reason)
3107 struct kmsg_dumper *dumper;
3108 unsigned long flags;
3110 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3114 list_for_each_entry_rcu(dumper, &dump_list, list) {
3115 if (dumper->max_reason && reason > dumper->max_reason)
3118 /* initialize iterator with data about the stored records */
3119 dumper->active = true;
3121 logbuf_lock_irqsave(flags);
3122 dumper->cur_seq = clear_seq;
3123 dumper->cur_idx = clear_idx;
3124 dumper->next_seq = log_next_seq;
3125 dumper->next_idx = log_next_idx;
3126 logbuf_unlock_irqrestore(flags);
3128 /* invoke dumper which will iterate over records */
3129 dumper->dump(dumper, reason);
3131 /* reset iterator */
3132 dumper->active = false;
3138 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3139 * @dumper: registered kmsg dumper
3140 * @syslog: include the "<4>" prefixes
3141 * @line: buffer to copy the line to
3142 * @size: maximum size of the buffer
3143 * @len: length of line placed into buffer
3145 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3146 * record, and copy one record into the provided buffer.
3148 * Consecutive calls will return the next available record moving
3149 * towards the end of the buffer with the youngest messages.
3151 * A return value of FALSE indicates that there are no more records to
3154 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3156 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3157 char *line, size_t size, size_t *len)
3159 struct printk_log *msg;
3163 if (!dumper->active)
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 >= log_next_seq)
3176 msg = log_from_idx(dumper->cur_idx);
3177 l = msg_print_text(msg, syslog, line, size);
3179 dumper->cur_idx = log_next(dumper->cur_idx);
3189 * kmsg_dump_get_line - retrieve one kmsg log line
3190 * @dumper: registered kmsg dumper
3191 * @syslog: include the "<4>" prefixes
3192 * @line: buffer to copy the line to
3193 * @size: maximum size of the buffer
3194 * @len: length of line placed into buffer
3196 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3197 * record, and copy one record into the provided buffer.
3199 * Consecutive calls will return the next available record moving
3200 * towards the end of the buffer with the youngest messages.
3202 * A return value of FALSE indicates that there are no more records to
3205 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3206 char *line, size_t size, size_t *len)
3208 unsigned long flags;
3211 logbuf_lock_irqsave(flags);
3212 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3213 logbuf_unlock_irqrestore(flags);
3217 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3220 * kmsg_dump_get_buffer - copy kmsg log lines
3221 * @dumper: registered kmsg dumper
3222 * @syslog: include the "<4>" prefixes
3223 * @buf: buffer to copy the line to
3224 * @size: maximum size of the buffer
3225 * @len: length of line placed into buffer
3227 * Start at the end of the kmsg buffer and fill the provided buffer
3228 * with as many of the the *youngest* kmsg records that fit into it.
3229 * If the buffer is large enough, all available kmsg records will be
3230 * copied with a single call.
3232 * Consecutive calls will fill the buffer with the next block of
3233 * available older records, not including the earlier retrieved ones.
3235 * A return value of FALSE indicates that there are no more records to
3238 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3239 char *buf, size_t size, size_t *len)
3241 unsigned long flags;
3249 if (!dumper->active)
3252 logbuf_lock_irqsave(flags);
3253 if (dumper->cur_seq < log_first_seq) {
3254 /* messages are gone, move to first available one */
3255 dumper->cur_seq = log_first_seq;
3256 dumper->cur_idx = log_first_idx;
3260 if (dumper->cur_seq >= dumper->next_seq) {
3261 logbuf_unlock_irqrestore(flags);
3265 /* calculate length of entire buffer */
3266 seq = dumper->cur_seq;
3267 idx = dumper->cur_idx;
3268 while (seq < dumper->next_seq) {
3269 struct printk_log *msg = log_from_idx(idx);
3271 l += msg_print_text(msg, true, NULL, 0);
3272 idx = log_next(idx);
3276 /* move first record forward until length fits into the buffer */
3277 seq = dumper->cur_seq;
3278 idx = dumper->cur_idx;
3279 while (l >= size && seq < dumper->next_seq) {
3280 struct printk_log *msg = log_from_idx(idx);
3282 l -= msg_print_text(msg, true, NULL, 0);
3283 idx = log_next(idx);
3287 /* last message in next interation */
3292 while (seq < dumper->next_seq) {
3293 struct printk_log *msg = log_from_idx(idx);
3295 l += msg_print_text(msg, syslog, buf + l, size - l);
3296 idx = log_next(idx);
3300 dumper->next_seq = next_seq;
3301 dumper->next_idx = next_idx;
3303 logbuf_unlock_irqrestore(flags);
3309 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3312 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3313 * @dumper: registered kmsg dumper
3315 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3316 * kmsg_dump_get_buffer() can be called again and used multiple
3317 * times within the same dumper.dump() callback.
3319 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3321 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3323 dumper->cur_seq = clear_seq;
3324 dumper->cur_idx = clear_idx;
3325 dumper->next_seq = log_next_seq;
3326 dumper->next_idx = log_next_idx;
3330 * kmsg_dump_rewind - reset the interator
3331 * @dumper: registered kmsg dumper
3333 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3334 * kmsg_dump_get_buffer() can be called again and used multiple
3335 * times within the same dumper.dump() callback.
3337 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3339 unsigned long flags;
3341 logbuf_lock_irqsave(flags);
3342 kmsg_dump_rewind_nolock(dumper);
3343 logbuf_unlock_irqrestore(flags);
3345 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);