2 * linux/kernel/printk.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * Modified to make sys_syslog() more flexible: added commands to
7 * return the last 4k of kernel messages, regardless of whether
8 * they've been read or not. Added option to suppress kernel printk's
9 * to the console. Added hook for sending the console messages
10 * elsewhere, in preparation for a serial line console (someday).
12 * Modified for sysctl support, 1/8/97, Chris Horn.
13 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
14 * manfred@colorfullife.com
15 * Rewrote bits to get rid of console_lock
16 * 01Mar01 Andrew Morton
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/kernel.h>
23 #include <linux/tty.h>
24 #include <linux/tty_driver.h>
25 #include <linux/console.h>
26 #include <linux/init.h>
27 #include <linux/jiffies.h>
28 #include <linux/nmi.h>
29 #include <linux/module.h>
30 #include <linux/moduleparam.h>
31 #include <linux/delay.h>
32 #include <linux/smp.h>
33 #include <linux/security.h>
34 #include <linux/memblock.h>
35 #include <linux/syscalls.h>
36 #include <linux/crash_core.h>
37 #include <linux/kdb.h>
38 #include <linux/ratelimit.h>
39 #include <linux/kmsg_dump.h>
40 #include <linux/syslog.h>
41 #include <linux/cpu.h>
42 #include <linux/rculist.h>
43 #include <linux/poll.h>
44 #include <linux/irq_work.h>
45 #include <linux/ctype.h>
46 #include <linux/uio.h>
47 #include <linux/sched/clock.h>
48 #include <linux/sched/debug.h>
49 #include <linux/sched/task_stack.h>
51 #include <linux/uaccess.h>
52 #include <asm/sections.h>
54 #include <trace/events/initcall.h>
55 #define CREATE_TRACE_POINTS
56 #include <trace/events/printk.h>
58 #include "console_cmdline.h"
62 int console_printk[4] = {
63 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
64 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
65 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
66 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
68 EXPORT_SYMBOL_GPL(console_printk);
70 atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
71 EXPORT_SYMBOL(ignore_console_lock_warning);
74 * Low level drivers may need that to know if they can schedule in
75 * their unblank() callback or not. So let's export it.
78 EXPORT_SYMBOL(oops_in_progress);
81 * console_sem protects the console_drivers list, and also
82 * provides serialisation for access to the entire console
85 static DEFINE_SEMAPHORE(console_sem);
86 struct console *console_drivers;
87 EXPORT_SYMBOL_GPL(console_drivers);
90 * System may need to suppress printk message under certain
91 * circumstances, like after kernel panic happens.
93 int __read_mostly suppress_printk;
96 static struct lockdep_map console_lock_dep_map = {
97 .name = "console_lock"
101 enum devkmsg_log_bits {
102 __DEVKMSG_LOG_BIT_ON = 0,
103 __DEVKMSG_LOG_BIT_OFF,
104 __DEVKMSG_LOG_BIT_LOCK,
107 enum devkmsg_log_masks {
108 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
109 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
110 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
113 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
114 #define DEVKMSG_LOG_MASK_DEFAULT 0
116 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
118 static int __control_devkmsg(char *str)
123 if (!strncmp(str, "on", 2)) {
124 devkmsg_log = DEVKMSG_LOG_MASK_ON;
126 } else if (!strncmp(str, "off", 3)) {
127 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
129 } else if (!strncmp(str, "ratelimit", 9)) {
130 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
136 static int __init control_devkmsg(char *str)
138 if (__control_devkmsg(str) < 0)
142 * Set sysctl string accordingly:
144 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
145 strcpy(devkmsg_log_str, "on");
146 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
147 strcpy(devkmsg_log_str, "off");
148 /* else "ratelimit" which is set by default. */
151 * Sysctl cannot change it anymore. The kernel command line setting of
152 * this parameter is to force the setting to be permanent throughout the
153 * runtime of the system. This is a precation measure against userspace
154 * trying to be a smarta** and attempting to change it up on us.
156 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
160 __setup("printk.devkmsg=", control_devkmsg);
162 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
164 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
165 void __user *buffer, size_t *lenp, loff_t *ppos)
167 char old_str[DEVKMSG_STR_MAX_SIZE];
172 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
176 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
179 err = proc_dostring(table, write, buffer, lenp, ppos);
184 err = __control_devkmsg(devkmsg_log_str);
187 * Do not accept an unknown string OR a known string with
190 if (err < 0 || (err + 1 != *lenp)) {
192 /* ... and restore old setting. */
194 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
203 /* Number of registered extended console drivers. */
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_NEWLINE = 2, /* text ended with a newline */
354 LOG_CONT = 8, /* text is a fragment of a continuation line */
358 u64 ts_nsec; /* timestamp in nanoseconds */
359 u16 len; /* length of entire record */
360 u16 text_len; /* length of text buffer */
361 u16 dict_len; /* length of dictionary buffer */
362 u8 facility; /* syslog facility */
363 u8 flags:5; /* internal record flags */
364 u8 level:3; /* syslog level */
365 #ifdef CONFIG_PRINTK_CALLER
366 u32 caller_id; /* thread id or processor id */
369 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
370 __packed __aligned(4)
375 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
376 * within the scheduler's rq lock. It must be released before calling
377 * console_unlock() or anything else that might wake up a process.
379 DEFINE_RAW_SPINLOCK(logbuf_lock);
382 * Helper macros to lock/unlock logbuf_lock and switch between
383 * printk-safe/unsafe modes.
385 #define logbuf_lock_irq() \
387 printk_safe_enter_irq(); \
388 raw_spin_lock(&logbuf_lock); \
391 #define logbuf_unlock_irq() \
393 raw_spin_unlock(&logbuf_lock); \
394 printk_safe_exit_irq(); \
397 #define logbuf_lock_irqsave(flags) \
399 printk_safe_enter_irqsave(flags); \
400 raw_spin_lock(&logbuf_lock); \
403 #define logbuf_unlock_irqrestore(flags) \
405 raw_spin_unlock(&logbuf_lock); \
406 printk_safe_exit_irqrestore(flags); \
410 DECLARE_WAIT_QUEUE_HEAD(log_wait);
411 /* the next printk record to read by syslog(READ) or /proc/kmsg */
412 static u64 syslog_seq;
413 static u32 syslog_idx;
414 static size_t syslog_partial;
415 static bool syslog_time;
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 #ifdef CONFIG_PRINTK_CALLER
435 #define PREFIX_MAX 48
437 #define PREFIX_MAX 32
439 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
441 #define LOG_LEVEL(v) ((v) & 0x07)
442 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
445 #define LOG_ALIGN __alignof__(struct printk_log)
446 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
447 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
448 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
449 static char *log_buf = __log_buf;
450 static u32 log_buf_len = __LOG_BUF_LEN;
452 /* Return log buffer address */
453 char *log_buf_addr_get(void)
458 /* Return log buffer size */
459 u32 log_buf_len_get(void)
464 /* human readable text of the record */
465 static char *log_text(const struct printk_log *msg)
467 return (char *)msg + sizeof(struct printk_log);
470 /* optional key/value pair dictionary attached to the record */
471 static char *log_dict(const struct printk_log *msg)
473 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
476 /* get record by index; idx must point to valid msg */
477 static struct printk_log *log_from_idx(u32 idx)
479 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
482 * A length == 0 record is the end of buffer marker. Wrap around and
483 * read the message at the start of the buffer.
486 return (struct printk_log *)log_buf;
490 /* get next record; idx must point to valid msg */
491 static u32 log_next(u32 idx)
493 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
495 /* length == 0 indicates the end of the buffer; wrap */
497 * A length == 0 record is the end of buffer marker. Wrap around and
498 * read the message at the start of the buffer as *this* one, and
499 * return the one after that.
502 msg = (struct printk_log *)log_buf;
505 return idx + msg->len;
509 * Check whether there is enough free space for the given message.
511 * The same values of first_idx and next_idx mean that the buffer
512 * is either empty or full.
514 * If the buffer is empty, we must respect the position of the indexes.
515 * They cannot be reset to the beginning of the buffer.
517 static int logbuf_has_space(u32 msg_size, bool empty)
521 if (log_next_idx > log_first_idx || empty)
522 free = max(log_buf_len - log_next_idx, log_first_idx);
524 free = log_first_idx - log_next_idx;
527 * We need space also for an empty header that signalizes wrapping
530 return free >= msg_size + sizeof(struct printk_log);
533 static int log_make_free_space(u32 msg_size)
535 while (log_first_seq < log_next_seq &&
536 !logbuf_has_space(msg_size, false)) {
537 /* drop old messages until we have enough contiguous space */
538 log_first_idx = log_next(log_first_idx);
542 if (clear_seq < log_first_seq) {
543 clear_seq = log_first_seq;
544 clear_idx = log_first_idx;
547 /* sequence numbers are equal, so the log buffer is empty */
548 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
554 /* compute the message size including the padding bytes */
555 static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
559 size = sizeof(struct printk_log) + text_len + dict_len;
560 *pad_len = (-size) & (LOG_ALIGN - 1);
567 * Define how much of the log buffer we could take at maximum. The value
568 * must be greater than two. Note that only half of the buffer is available
569 * when the index points to the middle.
571 #define MAX_LOG_TAKE_PART 4
572 static const char trunc_msg[] = "<truncated>";
574 static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
575 u16 *dict_len, u32 *pad_len)
578 * The message should not take the whole buffer. Otherwise, it might
579 * get removed too soon.
581 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
582 if (*text_len > max_text_len)
583 *text_len = max_text_len;
584 /* enable the warning message */
585 *trunc_msg_len = strlen(trunc_msg);
586 /* disable the "dict" completely */
588 /* compute the size again, count also the warning message */
589 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
592 /* insert record into the buffer, discard old ones, update heads */
593 static int log_store(u32 caller_id, int facility, int level,
594 enum log_flags flags, u64 ts_nsec,
595 const char *dict, u16 dict_len,
596 const char *text, u16 text_len)
598 struct printk_log *msg;
600 u16 trunc_msg_len = 0;
602 /* number of '\0' padding bytes to next message */
603 size = msg_used_size(text_len, dict_len, &pad_len);
605 if (log_make_free_space(size)) {
606 /* truncate the message if it is too long for empty buffer */
607 size = truncate_msg(&text_len, &trunc_msg_len,
608 &dict_len, &pad_len);
609 /* survive when the log buffer is too small for trunc_msg */
610 if (log_make_free_space(size))
614 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
616 * This message + an additional empty header does not fit
617 * at the end of the buffer. Add an empty header with len == 0
618 * to signify a wrap around.
620 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
625 msg = (struct printk_log *)(log_buf + log_next_idx);
626 memcpy(log_text(msg), text, text_len);
627 msg->text_len = text_len;
629 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
630 msg->text_len += trunc_msg_len;
632 memcpy(log_dict(msg), dict, dict_len);
633 msg->dict_len = dict_len;
634 msg->facility = facility;
635 msg->level = level & 7;
636 msg->flags = flags & 0x1f;
638 msg->ts_nsec = ts_nsec;
640 msg->ts_nsec = local_clock();
641 #ifdef CONFIG_PRINTK_CALLER
642 msg->caller_id = caller_id;
644 memset(log_dict(msg) + dict_len, 0, pad_len);
648 log_next_idx += msg->len;
651 return msg->text_len;
654 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
656 static int syslog_action_restricted(int type)
661 * Unless restricted, we allow "read all" and "get buffer size"
664 return type != SYSLOG_ACTION_READ_ALL &&
665 type != SYSLOG_ACTION_SIZE_BUFFER;
668 static int check_syslog_permissions(int type, int source)
671 * If this is from /proc/kmsg and we've already opened it, then we've
672 * already done the capabilities checks at open time.
674 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
677 if (syslog_action_restricted(type)) {
678 if (capable(CAP_SYSLOG))
681 * For historical reasons, accept CAP_SYS_ADMIN too, with
684 if (capable(CAP_SYS_ADMIN)) {
685 pr_warn_once("%s (%d): Attempt to access syslog with "
686 "CAP_SYS_ADMIN but no CAP_SYSLOG "
688 current->comm, task_pid_nr(current));
694 return security_syslog(type);
697 static void append_char(char **pp, char *e, char c)
703 static ssize_t msg_print_ext_header(char *buf, size_t size,
704 struct printk_log *msg, u64 seq)
706 u64 ts_usec = msg->ts_nsec;
708 #ifdef CONFIG_PRINTK_CALLER
709 u32 id = msg->caller_id;
711 snprintf(caller, sizeof(caller), ",caller=%c%u",
712 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
717 do_div(ts_usec, 1000);
719 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
720 (msg->facility << 3) | msg->level, seq, ts_usec,
721 msg->flags & LOG_CONT ? 'c' : '-', caller);
724 static ssize_t msg_print_ext_body(char *buf, size_t size,
725 char *dict, size_t dict_len,
726 char *text, size_t text_len)
728 char *p = buf, *e = buf + size;
731 /* escape non-printable characters */
732 for (i = 0; i < text_len; i++) {
733 unsigned char c = text[i];
735 if (c < ' ' || c >= 127 || c == '\\')
736 p += scnprintf(p, e - p, "\\x%02x", c);
738 append_char(&p, e, c);
740 append_char(&p, e, '\n');
745 for (i = 0; i < dict_len; i++) {
746 unsigned char c = dict[i];
749 append_char(&p, e, ' ');
754 append_char(&p, e, '\n');
759 if (c < ' ' || c >= 127 || c == '\\') {
760 p += scnprintf(p, e - p, "\\x%02x", c);
764 append_char(&p, e, c);
766 append_char(&p, e, '\n');
772 /* /dev/kmsg - userspace message inject/listen interface */
773 struct devkmsg_user {
776 struct ratelimit_state rs;
778 char buf[CONSOLE_EXT_LOG_MAX];
781 static __printf(3, 4) __cold
782 int devkmsg_emit(int facility, int level, const char *fmt, ...)
788 r = vprintk_emit(facility, level, NULL, 0, fmt, args);
794 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
797 int level = default_message_loglevel;
798 int facility = 1; /* LOG_USER */
799 struct file *file = iocb->ki_filp;
800 struct devkmsg_user *user = file->private_data;
801 size_t len = iov_iter_count(from);
804 if (!user || len > LOG_LINE_MAX)
807 /* Ignore when user logging is disabled. */
808 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
811 /* Ratelimit when not explicitly enabled. */
812 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
813 if (!___ratelimit(&user->rs, current->comm))
817 buf = kmalloc(len+1, GFP_KERNEL);
822 if (!copy_from_iter_full(buf, len, from)) {
828 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
829 * the decimal value represents 32bit, the lower 3 bit are the log
830 * level, the rest are the log facility.
832 * If no prefix or no userspace facility is specified, we
833 * enforce LOG_USER, to be able to reliably distinguish
834 * kernel-generated messages from userspace-injected ones.
837 if (line[0] == '<') {
841 u = simple_strtoul(line + 1, &endp, 10);
842 if (endp && endp[0] == '>') {
843 level = LOG_LEVEL(u);
844 if (LOG_FACILITY(u) != 0)
845 facility = LOG_FACILITY(u);
852 devkmsg_emit(facility, level, "%s", line);
857 static ssize_t devkmsg_read(struct file *file, char __user *buf,
858 size_t count, loff_t *ppos)
860 struct devkmsg_user *user = file->private_data;
861 struct printk_log *msg;
868 ret = mutex_lock_interruptible(&user->lock);
873 while (user->seq == log_next_seq) {
874 if (file->f_flags & O_NONBLOCK) {
881 ret = wait_event_interruptible(log_wait,
882 user->seq != log_next_seq);
888 if (user->seq < log_first_seq) {
889 /* our last seen message is gone, return error and reset */
890 user->idx = log_first_idx;
891 user->seq = log_first_seq;
897 msg = log_from_idx(user->idx);
898 len = msg_print_ext_header(user->buf, sizeof(user->buf),
900 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
901 log_dict(msg), msg->dict_len,
902 log_text(msg), msg->text_len);
904 user->idx = log_next(user->idx);
913 if (copy_to_user(buf, user->buf, len)) {
919 mutex_unlock(&user->lock);
923 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
925 struct devkmsg_user *user = file->private_data;
936 /* the first record */
937 user->idx = log_first_idx;
938 user->seq = log_first_seq;
942 * The first record after the last SYSLOG_ACTION_CLEAR,
943 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
944 * changes no global state, and does not clear anything.
946 user->idx = clear_idx;
947 user->seq = clear_seq;
950 /* after the last record */
951 user->idx = log_next_idx;
952 user->seq = log_next_seq;
961 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
963 struct devkmsg_user *user = file->private_data;
967 return EPOLLERR|EPOLLNVAL;
969 poll_wait(file, &log_wait, wait);
972 if (user->seq < log_next_seq) {
973 /* return error when data has vanished underneath us */
974 if (user->seq < log_first_seq)
975 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
977 ret = EPOLLIN|EPOLLRDNORM;
984 static int devkmsg_open(struct inode *inode, struct file *file)
986 struct devkmsg_user *user;
989 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
992 /* write-only does not need any file context */
993 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
994 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
1000 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
1004 ratelimit_default_init(&user->rs);
1005 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
1007 mutex_init(&user->lock);
1010 user->idx = log_first_idx;
1011 user->seq = log_first_seq;
1012 logbuf_unlock_irq();
1014 file->private_data = user;
1018 static int devkmsg_release(struct inode *inode, struct file *file)
1020 struct devkmsg_user *user = file->private_data;
1025 ratelimit_state_exit(&user->rs);
1027 mutex_destroy(&user->lock);
1032 const struct file_operations kmsg_fops = {
1033 .open = devkmsg_open,
1034 .read = devkmsg_read,
1035 .write_iter = devkmsg_write,
1036 .llseek = devkmsg_llseek,
1037 .poll = devkmsg_poll,
1038 .release = devkmsg_release,
1041 #ifdef CONFIG_CRASH_CORE
1043 * This appends the listed symbols to /proc/vmcore
1045 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1046 * obtain access to symbols that are otherwise very difficult to locate. These
1047 * symbols are specifically used so that utilities can access and extract the
1048 * dmesg log from a vmcore file after a crash.
1050 void log_buf_vmcoreinfo_setup(void)
1052 VMCOREINFO_SYMBOL(log_buf);
1053 VMCOREINFO_SYMBOL(log_buf_len);
1054 VMCOREINFO_SYMBOL(log_first_idx);
1055 VMCOREINFO_SYMBOL(clear_idx);
1056 VMCOREINFO_SYMBOL(log_next_idx);
1058 * Export struct printk_log size and field offsets. User space tools can
1059 * parse it and detect any changes to structure down the line.
1061 VMCOREINFO_STRUCT_SIZE(printk_log);
1062 VMCOREINFO_OFFSET(printk_log, ts_nsec);
1063 VMCOREINFO_OFFSET(printk_log, len);
1064 VMCOREINFO_OFFSET(printk_log, text_len);
1065 VMCOREINFO_OFFSET(printk_log, dict_len);
1066 #ifdef CONFIG_PRINTK_CALLER
1067 VMCOREINFO_OFFSET(printk_log, caller_id);
1072 /* requested log_buf_len from kernel cmdline */
1073 static unsigned long __initdata new_log_buf_len;
1075 /* we practice scaling the ring buffer by powers of 2 */
1076 static void __init log_buf_len_update(u64 size)
1078 if (size > (u64)LOG_BUF_LEN_MAX) {
1079 size = (u64)LOG_BUF_LEN_MAX;
1080 pr_err("log_buf over 2G is not supported.\n");
1084 size = roundup_pow_of_two(size);
1085 if (size > log_buf_len)
1086 new_log_buf_len = (unsigned long)size;
1089 /* save requested log_buf_len since it's too early to process it */
1090 static int __init log_buf_len_setup(char *str)
1097 size = memparse(str, &str);
1099 log_buf_len_update(size);
1103 early_param("log_buf_len", log_buf_len_setup);
1106 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1108 static void __init log_buf_add_cpu(void)
1110 unsigned int cpu_extra;
1113 * archs should set up cpu_possible_bits properly with
1114 * set_cpu_possible() after setup_arch() but just in
1115 * case lets ensure this is valid.
1117 if (num_possible_cpus() == 1)
1120 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1122 /* by default this will only continue through for large > 64 CPUs */
1123 if (cpu_extra <= __LOG_BUF_LEN / 2)
1126 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1127 __LOG_CPU_MAX_BUF_LEN);
1128 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1130 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1132 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1134 #else /* !CONFIG_SMP */
1135 static inline void log_buf_add_cpu(void) {}
1136 #endif /* CONFIG_SMP */
1138 void __init setup_log_buf(int early)
1140 unsigned long flags;
1144 if (log_buf != __log_buf)
1147 if (!early && !new_log_buf_len)
1150 if (!new_log_buf_len)
1153 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1154 if (unlikely(!new_log_buf)) {
1155 pr_err("log_buf_len: %lu bytes not available\n",
1160 logbuf_lock_irqsave(flags);
1161 log_buf_len = new_log_buf_len;
1162 log_buf = new_log_buf;
1163 new_log_buf_len = 0;
1164 free = __LOG_BUF_LEN - log_next_idx;
1165 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1166 logbuf_unlock_irqrestore(flags);
1168 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1169 pr_info("early log buf free: %u(%u%%)\n",
1170 free, (free * 100) / __LOG_BUF_LEN);
1173 static bool __read_mostly ignore_loglevel;
1175 static int __init ignore_loglevel_setup(char *str)
1177 ignore_loglevel = true;
1178 pr_info("debug: ignoring loglevel setting.\n");
1183 early_param("ignore_loglevel", ignore_loglevel_setup);
1184 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1185 MODULE_PARM_DESC(ignore_loglevel,
1186 "ignore loglevel setting (prints all kernel messages to the console)");
1188 static bool suppress_message_printing(int level)
1190 return (level >= console_loglevel && !ignore_loglevel);
1193 #ifdef CONFIG_BOOT_PRINTK_DELAY
1195 static int boot_delay; /* msecs delay after each printk during bootup */
1196 static unsigned long long loops_per_msec; /* based on boot_delay */
1198 static int __init boot_delay_setup(char *str)
1202 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1203 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1205 get_option(&str, &boot_delay);
1206 if (boot_delay > 10 * 1000)
1209 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1210 "HZ: %d, loops_per_msec: %llu\n",
1211 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1214 early_param("boot_delay", boot_delay_setup);
1216 static void boot_delay_msec(int level)
1218 unsigned long long k;
1219 unsigned long timeout;
1221 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1222 || suppress_message_printing(level)) {
1226 k = (unsigned long long)loops_per_msec * boot_delay;
1228 timeout = jiffies + msecs_to_jiffies(boot_delay);
1233 * use (volatile) jiffies to prevent
1234 * compiler reduction; loop termination via jiffies
1235 * is secondary and may or may not happen.
1237 if (time_after(jiffies, timeout))
1239 touch_nmi_watchdog();
1243 static inline void boot_delay_msec(int level)
1248 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1249 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1251 static size_t print_syslog(unsigned int level, char *buf)
1253 return sprintf(buf, "<%u>", level);
1256 static size_t print_time(u64 ts, char *buf)
1258 unsigned long rem_nsec = do_div(ts, 1000000000);
1260 return sprintf(buf, "[%5lu.%06lu]",
1261 (unsigned long)ts, rem_nsec / 1000);
1264 #ifdef CONFIG_PRINTK_CALLER
1265 static size_t print_caller(u32 id, char *buf)
1269 snprintf(caller, sizeof(caller), "%c%u",
1270 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1271 return sprintf(buf, "[%6s]", caller);
1274 #define print_caller(id, buf) 0
1277 static size_t print_prefix(const struct printk_log *msg, bool syslog,
1278 bool time, char *buf)
1283 len = print_syslog((msg->facility << 3) | msg->level, buf);
1286 len += print_time(msg->ts_nsec, buf + len);
1288 len += print_caller(msg->caller_id, buf + len);
1290 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1298 static size_t msg_print_text(const struct printk_log *msg, bool syslog,
1299 bool time, char *buf, size_t size)
1301 const char *text = log_text(msg);
1302 size_t text_size = msg->text_len;
1304 char prefix[PREFIX_MAX];
1305 const size_t prefix_len = print_prefix(msg, syslog, time, prefix);
1308 const char *next = memchr(text, '\n', text_size);
1312 text_len = next - text;
1314 text_size -= next - text;
1316 text_len = text_size;
1320 if (prefix_len + text_len + 1 >= size - len)
1323 memcpy(buf + len, prefix, prefix_len);
1325 memcpy(buf + len, text, text_len);
1329 /* SYSLOG_ACTION_* buffer size only calculation */
1330 len += prefix_len + text_len + 1;
1339 static int syslog_print(char __user *buf, int size)
1342 struct printk_log *msg;
1345 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1354 if (syslog_seq < log_first_seq) {
1355 /* messages are gone, move to first one */
1356 syslog_seq = log_first_seq;
1357 syslog_idx = log_first_idx;
1360 if (syslog_seq == log_next_seq) {
1361 logbuf_unlock_irq();
1366 * To keep reading/counting partial line consistent,
1367 * use printk_time value as of the beginning of a line.
1369 if (!syslog_partial)
1370 syslog_time = printk_time;
1372 skip = syslog_partial;
1373 msg = log_from_idx(syslog_idx);
1374 n = msg_print_text(msg, true, syslog_time, text,
1375 LOG_LINE_MAX + PREFIX_MAX);
1376 if (n - syslog_partial <= size) {
1377 /* message fits into buffer, move forward */
1378 syslog_idx = log_next(syslog_idx);
1380 n -= syslog_partial;
1383 /* partial read(), remember position */
1385 syslog_partial += n;
1388 logbuf_unlock_irq();
1393 if (copy_to_user(buf, text + skip, n)) {
1408 static int syslog_print_all(char __user *buf, int size, bool clear)
1417 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1424 * Find first record that fits, including all following records,
1425 * into the user-provided buffer for this dump.
1429 while (seq < log_next_seq) {
1430 struct printk_log *msg = log_from_idx(idx);
1432 len += msg_print_text(msg, true, time, NULL, 0);
1433 idx = log_next(idx);
1437 /* move first record forward until length fits into the buffer */
1440 while (len > size && seq < log_next_seq) {
1441 struct printk_log *msg = log_from_idx(idx);
1443 len -= msg_print_text(msg, true, time, NULL, 0);
1444 idx = log_next(idx);
1448 /* last message fitting into this dump */
1449 next_seq = log_next_seq;
1452 while (len >= 0 && seq < next_seq) {
1453 struct printk_log *msg = log_from_idx(idx);
1454 int textlen = msg_print_text(msg, true, time, text,
1455 LOG_LINE_MAX + PREFIX_MAX);
1457 idx = log_next(idx);
1460 logbuf_unlock_irq();
1461 if (copy_to_user(buf + len, text, textlen))
1467 if (seq < log_first_seq) {
1468 /* messages are gone, move to next one */
1469 seq = log_first_seq;
1470 idx = log_first_idx;
1475 clear_seq = log_next_seq;
1476 clear_idx = log_next_idx;
1478 logbuf_unlock_irq();
1484 static void syslog_clear(void)
1487 clear_seq = log_next_seq;
1488 clear_idx = log_next_idx;
1489 logbuf_unlock_irq();
1492 int do_syslog(int type, char __user *buf, int len, int source)
1495 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1498 error = check_syslog_permissions(type, source);
1503 case SYSLOG_ACTION_CLOSE: /* Close log */
1505 case SYSLOG_ACTION_OPEN: /* Open log */
1507 case SYSLOG_ACTION_READ: /* Read from log */
1508 if (!buf || len < 0)
1512 if (!access_ok(buf, len))
1514 error = wait_event_interruptible(log_wait,
1515 syslog_seq != log_next_seq);
1518 error = syslog_print(buf, len);
1520 /* Read/clear last kernel messages */
1521 case SYSLOG_ACTION_READ_CLEAR:
1524 /* Read last kernel messages */
1525 case SYSLOG_ACTION_READ_ALL:
1526 if (!buf || len < 0)
1530 if (!access_ok(buf, len))
1532 error = syslog_print_all(buf, len, clear);
1534 /* Clear ring buffer */
1535 case SYSLOG_ACTION_CLEAR:
1538 /* Disable logging to console */
1539 case SYSLOG_ACTION_CONSOLE_OFF:
1540 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1541 saved_console_loglevel = console_loglevel;
1542 console_loglevel = minimum_console_loglevel;
1544 /* Enable logging to console */
1545 case SYSLOG_ACTION_CONSOLE_ON:
1546 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1547 console_loglevel = saved_console_loglevel;
1548 saved_console_loglevel = LOGLEVEL_DEFAULT;
1551 /* Set level of messages printed to console */
1552 case SYSLOG_ACTION_CONSOLE_LEVEL:
1553 if (len < 1 || len > 8)
1555 if (len < minimum_console_loglevel)
1556 len = minimum_console_loglevel;
1557 console_loglevel = len;
1558 /* Implicitly re-enable logging to console */
1559 saved_console_loglevel = LOGLEVEL_DEFAULT;
1561 /* Number of chars in the log buffer */
1562 case SYSLOG_ACTION_SIZE_UNREAD:
1564 if (syslog_seq < log_first_seq) {
1565 /* messages are gone, move to first one */
1566 syslog_seq = log_first_seq;
1567 syslog_idx = log_first_idx;
1570 if (source == SYSLOG_FROM_PROC) {
1572 * Short-cut for poll(/"proc/kmsg") which simply checks
1573 * for pending data, not the size; return the count of
1574 * records, not the length.
1576 error = log_next_seq - syslog_seq;
1578 u64 seq = syslog_seq;
1579 u32 idx = syslog_idx;
1580 bool time = syslog_partial ? syslog_time : printk_time;
1582 while (seq < log_next_seq) {
1583 struct printk_log *msg = log_from_idx(idx);
1585 error += msg_print_text(msg, true, time, NULL,
1588 idx = log_next(idx);
1591 error -= syslog_partial;
1593 logbuf_unlock_irq();
1595 /* Size of the log buffer */
1596 case SYSLOG_ACTION_SIZE_BUFFER:
1597 error = log_buf_len;
1607 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1609 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1613 * Special console_lock variants that help to reduce the risk of soft-lockups.
1614 * They allow to pass console_lock to another printk() call using a busy wait.
1617 #ifdef CONFIG_LOCKDEP
1618 static struct lockdep_map console_owner_dep_map = {
1619 .name = "console_owner"
1623 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1624 static struct task_struct *console_owner;
1625 static bool console_waiter;
1628 * console_lock_spinning_enable - mark beginning of code where another
1629 * thread might safely busy wait
1631 * This basically converts console_lock into a spinlock. This marks
1632 * the section where the console_lock owner can not sleep, because
1633 * there may be a waiter spinning (like a spinlock). Also it must be
1634 * ready to hand over the lock at the end of the section.
1636 static void console_lock_spinning_enable(void)
1638 raw_spin_lock(&console_owner_lock);
1639 console_owner = current;
1640 raw_spin_unlock(&console_owner_lock);
1642 /* The waiter may spin on us after setting console_owner */
1643 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1647 * console_lock_spinning_disable_and_check - mark end of code where another
1648 * thread was able to busy wait and check if there is a waiter
1650 * This is called at the end of the section where spinning is allowed.
1651 * It has two functions. First, it is a signal that it is no longer
1652 * safe to start busy waiting for the lock. Second, it checks if
1653 * there is a busy waiter and passes the lock rights to her.
1655 * Important: Callers lose the lock if there was a busy waiter.
1656 * They must not touch items synchronized by console_lock
1659 * Return: 1 if the lock rights were passed, 0 otherwise.
1661 static int console_lock_spinning_disable_and_check(void)
1665 raw_spin_lock(&console_owner_lock);
1666 waiter = READ_ONCE(console_waiter);
1667 console_owner = NULL;
1668 raw_spin_unlock(&console_owner_lock);
1671 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1675 /* The waiter is now free to continue */
1676 WRITE_ONCE(console_waiter, false);
1678 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1681 * Hand off console_lock to waiter. The waiter will perform
1682 * the up(). After this, the waiter is the console_lock owner.
1684 mutex_release(&console_lock_dep_map, 1, _THIS_IP_);
1689 * console_trylock_spinning - try to get console_lock by busy waiting
1691 * This allows to busy wait for the console_lock when the current
1692 * owner is running in specially marked sections. It means that
1693 * the current owner is running and cannot reschedule until it
1694 * is ready to lose the lock.
1696 * Return: 1 if we got the lock, 0 othrewise
1698 static int console_trylock_spinning(void)
1700 struct task_struct *owner = NULL;
1703 unsigned long flags;
1705 if (console_trylock())
1708 printk_safe_enter_irqsave(flags);
1710 raw_spin_lock(&console_owner_lock);
1711 owner = READ_ONCE(console_owner);
1712 waiter = READ_ONCE(console_waiter);
1713 if (!waiter && owner && owner != current) {
1714 WRITE_ONCE(console_waiter, true);
1717 raw_spin_unlock(&console_owner_lock);
1720 * If there is an active printk() writing to the
1721 * consoles, instead of having it write our data too,
1722 * see if we can offload that load from the active
1723 * printer, and do some printing ourselves.
1724 * Go into a spin only if there isn't already a waiter
1725 * spinning, and there is an active printer, and
1726 * that active printer isn't us (recursive printk?).
1729 printk_safe_exit_irqrestore(flags);
1733 /* We spin waiting for the owner to release us */
1734 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1735 /* Owner will clear console_waiter on hand off */
1736 while (READ_ONCE(console_waiter))
1738 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1740 printk_safe_exit_irqrestore(flags);
1742 * The owner passed the console lock to us.
1743 * Since we did not spin on console lock, annotate
1744 * this as a trylock. Otherwise lockdep will
1747 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1753 * Call the console drivers, asking them to write out
1754 * log_buf[start] to log_buf[end - 1].
1755 * The console_lock must be held.
1757 static void call_console_drivers(const char *ext_text, size_t ext_len,
1758 const char *text, size_t len)
1760 struct console *con;
1762 trace_console_rcuidle(text, len);
1764 if (!console_drivers)
1767 for_each_console(con) {
1768 if (exclusive_console && con != exclusive_console)
1770 if (!(con->flags & CON_ENABLED))
1774 if (!cpu_online(smp_processor_id()) &&
1775 !(con->flags & CON_ANYTIME))
1777 if (con->flags & CON_EXTENDED)
1778 con->write(con, ext_text, ext_len);
1780 con->write(con, text, len);
1784 int printk_delay_msec __read_mostly;
1786 static inline void printk_delay(void)
1788 if (unlikely(printk_delay_msec)) {
1789 int m = printk_delay_msec;
1793 touch_nmi_watchdog();
1798 static inline u32 printk_caller_id(void)
1800 return in_task() ? task_pid_nr(current) :
1801 0x80000000 + raw_smp_processor_id();
1805 * Continuation lines are buffered, and not committed to the record buffer
1806 * until the line is complete, or a race forces it. The line fragments
1807 * though, are printed immediately to the consoles to ensure everything has
1808 * reached the console in case of a kernel crash.
1810 static struct cont {
1811 char buf[LOG_LINE_MAX];
1812 size_t len; /* length == 0 means unused buffer */
1813 u32 caller_id; /* printk_caller_id() of first print */
1814 u64 ts_nsec; /* time of first print */
1815 u8 level; /* log level of first message */
1816 u8 facility; /* log facility of first message */
1817 enum log_flags flags; /* prefix, newline flags */
1820 static void cont_flush(void)
1825 log_store(cont.caller_id, cont.facility, cont.level, cont.flags,
1826 cont.ts_nsec, NULL, 0, cont.buf, cont.len);
1830 static bool cont_add(u32 caller_id, int facility, int level,
1831 enum log_flags flags, const char *text, size_t len)
1833 /* If the line gets too long, split it up in separate records. */
1834 if (cont.len + len > sizeof(cont.buf)) {
1840 cont.facility = facility;
1842 cont.caller_id = caller_id;
1843 cont.ts_nsec = local_clock();
1847 memcpy(cont.buf + cont.len, text, len);
1850 // The original flags come from the first line,
1851 // but later continuations can add a newline.
1852 if (flags & LOG_NEWLINE) {
1853 cont.flags |= LOG_NEWLINE;
1860 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)
1862 const u32 caller_id = printk_caller_id();
1865 * If an earlier line was buffered, and we're a continuation
1866 * write from the same context, try to add it to the buffer.
1869 if (cont.caller_id == caller_id && (lflags & LOG_CONT)) {
1870 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1873 /* Otherwise, make sure it's flushed */
1877 /* Skip empty continuation lines that couldn't be added - they just flush */
1878 if (!text_len && (lflags & LOG_CONT))
1881 /* If it doesn't end in a newline, try to buffer the current line */
1882 if (!(lflags & LOG_NEWLINE)) {
1883 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1887 /* Store it in the record log */
1888 return log_store(caller_id, facility, level, lflags, 0,
1889 dict, dictlen, text, text_len);
1892 /* Must be called under logbuf_lock. */
1893 int vprintk_store(int facility, int level,
1894 const char *dict, size_t dictlen,
1895 const char *fmt, va_list args)
1897 static char textbuf[LOG_LINE_MAX];
1898 char *text = textbuf;
1900 enum log_flags lflags = 0;
1903 * The printf needs to come first; we need the syslog
1904 * prefix which might be passed-in as a parameter.
1906 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1908 /* mark and strip a trailing newline */
1909 if (text_len && text[text_len-1] == '\n') {
1911 lflags |= LOG_NEWLINE;
1914 /* strip kernel syslog prefix and extract log level or control flags */
1915 if (facility == 0) {
1918 while ((kern_level = printk_get_level(text)) != 0) {
1919 switch (kern_level) {
1921 if (level == LOGLEVEL_DEFAULT)
1922 level = kern_level - '0';
1924 case 'c': /* KERN_CONT */
1933 if (level == LOGLEVEL_DEFAULT)
1934 level = default_message_loglevel;
1937 lflags |= LOG_NEWLINE;
1939 return log_output(facility, level, lflags,
1940 dict, dictlen, text, text_len);
1943 asmlinkage int vprintk_emit(int facility, int level,
1944 const char *dict, size_t dictlen,
1945 const char *fmt, va_list args)
1948 bool in_sched = false, pending_output;
1949 unsigned long flags;
1952 /* Suppress unimportant messages after panic happens */
1953 if (unlikely(suppress_printk))
1956 if (level == LOGLEVEL_SCHED) {
1957 level = LOGLEVEL_DEFAULT;
1961 boot_delay_msec(level);
1964 /* This stops the holder of console_sem just where we want him */
1965 logbuf_lock_irqsave(flags);
1966 curr_log_seq = log_next_seq;
1967 printed_len = vprintk_store(facility, level, dict, dictlen, fmt, args);
1968 pending_output = (curr_log_seq != log_next_seq);
1969 logbuf_unlock_irqrestore(flags);
1971 /* If called from the scheduler, we can not call up(). */
1972 if (!in_sched && pending_output) {
1974 * Disable preemption to avoid being preempted while holding
1975 * console_sem which would prevent anyone from printing to
1980 * Try to acquire and then immediately release the console
1981 * semaphore. The release will print out buffers and wake up
1982 * /dev/kmsg and syslog() users.
1984 if (console_trylock_spinning())
1993 EXPORT_SYMBOL(vprintk_emit);
1995 asmlinkage int vprintk(const char *fmt, va_list args)
1997 return vprintk_func(fmt, args);
1999 EXPORT_SYMBOL(vprintk);
2001 int vprintk_default(const char *fmt, va_list args)
2005 #ifdef CONFIG_KGDB_KDB
2006 /* Allow to pass printk() to kdb but avoid a recursion. */
2007 if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
2008 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
2012 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
2016 EXPORT_SYMBOL_GPL(vprintk_default);
2019 * printk - print a kernel message
2020 * @fmt: format string
2022 * This is printk(). It can be called from any context. We want it to work.
2024 * We try to grab the console_lock. If we succeed, it's easy - we log the
2025 * output and call the console drivers. If we fail to get the semaphore, we
2026 * place the output into the log buffer and return. The current holder of
2027 * the console_sem will notice the new output in console_unlock(); and will
2028 * send it to the consoles before releasing the lock.
2030 * One effect of this deferred printing is that code which calls printk() and
2031 * then changes console_loglevel may break. This is because console_loglevel
2032 * is inspected when the actual printing occurs.
2037 * See the vsnprintf() documentation for format string extensions over C99.
2039 asmlinkage __visible int printk(const char *fmt, ...)
2044 va_start(args, fmt);
2045 r = vprintk_func(fmt, args);
2050 EXPORT_SYMBOL(printk);
2052 #else /* CONFIG_PRINTK */
2054 #define LOG_LINE_MAX 0
2055 #define PREFIX_MAX 0
2056 #define printk_time false
2058 static u64 syslog_seq;
2059 static u32 syslog_idx;
2060 static u64 console_seq;
2061 static u32 console_idx;
2062 static u64 exclusive_console_stop_seq;
2063 static u64 log_first_seq;
2064 static u32 log_first_idx;
2065 static u64 log_next_seq;
2066 static char *log_text(const struct printk_log *msg) { return NULL; }
2067 static char *log_dict(const struct printk_log *msg) { return NULL; }
2068 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2069 static u32 log_next(u32 idx) { return 0; }
2070 static ssize_t msg_print_ext_header(char *buf, size_t size,
2071 struct printk_log *msg,
2072 u64 seq) { return 0; }
2073 static ssize_t msg_print_ext_body(char *buf, size_t size,
2074 char *dict, size_t dict_len,
2075 char *text, size_t text_len) { return 0; }
2076 static void console_lock_spinning_enable(void) { }
2077 static int console_lock_spinning_disable_and_check(void) { return 0; }
2078 static void call_console_drivers(const char *ext_text, size_t ext_len,
2079 const char *text, size_t len) {}
2080 static size_t msg_print_text(const struct printk_log *msg, bool syslog,
2081 bool time, char *buf, size_t size) { return 0; }
2082 static bool suppress_message_printing(int level) { return false; }
2084 #endif /* CONFIG_PRINTK */
2086 #ifdef CONFIG_EARLY_PRINTK
2087 struct console *early_console;
2089 asmlinkage __visible void early_printk(const char *fmt, ...)
2099 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2102 early_console->write(early_console, buf, n);
2106 static int __add_preferred_console(char *name, int idx, char *options,
2109 struct console_cmdline *c;
2113 * See if this tty is not yet registered, and
2114 * if we have a slot free.
2116 for (i = 0, c = console_cmdline;
2117 i < MAX_CMDLINECONSOLES && c->name[0];
2119 if (strcmp(c->name, name) == 0 && c->index == idx) {
2121 preferred_console = i;
2125 if (i == MAX_CMDLINECONSOLES)
2128 preferred_console = i;
2129 strlcpy(c->name, name, sizeof(c->name));
2130 c->options = options;
2131 braille_set_options(c, brl_options);
2137 static int __init console_msg_format_setup(char *str)
2139 if (!strcmp(str, "syslog"))
2140 console_msg_format = MSG_FORMAT_SYSLOG;
2141 if (!strcmp(str, "default"))
2142 console_msg_format = MSG_FORMAT_DEFAULT;
2145 __setup("console_msg_format=", console_msg_format_setup);
2148 * Set up a console. Called via do_early_param() in init/main.c
2149 * for each "console=" parameter in the boot command line.
2151 static int __init console_setup(char *str)
2153 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2154 char *s, *options, *brl_options = NULL;
2157 if (_braille_console_setup(&str, &brl_options))
2161 * Decode str into name, index, options.
2163 if (str[0] >= '0' && str[0] <= '9') {
2164 strcpy(buf, "ttyS");
2165 strncpy(buf + 4, str, sizeof(buf) - 5);
2167 strncpy(buf, str, sizeof(buf) - 1);
2169 buf[sizeof(buf) - 1] = 0;
2170 options = strchr(str, ',');
2174 if (!strcmp(str, "ttya"))
2175 strcpy(buf, "ttyS0");
2176 if (!strcmp(str, "ttyb"))
2177 strcpy(buf, "ttyS1");
2179 for (s = buf; *s; s++)
2180 if (isdigit(*s) || *s == ',')
2182 idx = simple_strtoul(s, NULL, 10);
2185 __add_preferred_console(buf, idx, options, brl_options);
2186 console_set_on_cmdline = 1;
2189 __setup("console=", console_setup);
2192 * add_preferred_console - add a device to the list of preferred consoles.
2193 * @name: device name
2194 * @idx: device index
2195 * @options: options for this console
2197 * The last preferred console added will be used for kernel messages
2198 * and stdin/out/err for init. Normally this is used by console_setup
2199 * above to handle user-supplied console arguments; however it can also
2200 * be used by arch-specific code either to override the user or more
2201 * commonly to provide a default console (ie from PROM variables) when
2202 * the user has not supplied one.
2204 int add_preferred_console(char *name, int idx, char *options)
2206 return __add_preferred_console(name, idx, options, NULL);
2209 bool console_suspend_enabled = true;
2210 EXPORT_SYMBOL(console_suspend_enabled);
2212 static int __init console_suspend_disable(char *str)
2214 console_suspend_enabled = false;
2217 __setup("no_console_suspend", console_suspend_disable);
2218 module_param_named(console_suspend, console_suspend_enabled,
2219 bool, S_IRUGO | S_IWUSR);
2220 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2221 " and hibernate operations");
2224 * suspend_console - suspend the console subsystem
2226 * This disables printk() while we go into suspend states
2228 void suspend_console(void)
2230 if (!console_suspend_enabled)
2232 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2234 console_suspended = 1;
2238 void resume_console(void)
2240 if (!console_suspend_enabled)
2243 console_suspended = 0;
2248 * console_cpu_notify - print deferred console messages after CPU hotplug
2251 * If printk() is called from a CPU that is not online yet, the messages
2252 * will be printed on the console only if there are CON_ANYTIME consoles.
2253 * This function is called when a new CPU comes online (or fails to come
2254 * up) or goes offline.
2256 static int console_cpu_notify(unsigned int cpu)
2258 if (!cpuhp_tasks_frozen) {
2259 /* If trylock fails, someone else is doing the printing */
2260 if (console_trylock())
2267 * console_lock - lock the console system for exclusive use.
2269 * Acquires a lock which guarantees that the caller has
2270 * exclusive access to the console system and the console_drivers list.
2272 * Can sleep, returns nothing.
2274 void console_lock(void)
2279 if (console_suspended)
2282 console_may_schedule = 1;
2284 EXPORT_SYMBOL(console_lock);
2287 * console_trylock - try to lock the console system for exclusive use.
2289 * Try to acquire a lock which guarantees that the caller has exclusive
2290 * access to the console system and the console_drivers list.
2292 * returns 1 on success, and 0 on failure to acquire the lock.
2294 int console_trylock(void)
2296 if (down_trylock_console_sem())
2298 if (console_suspended) {
2303 console_may_schedule = 0;
2306 EXPORT_SYMBOL(console_trylock);
2308 int is_console_locked(void)
2310 return console_locked;
2312 EXPORT_SYMBOL(is_console_locked);
2315 * Check if we have any console that is capable of printing while cpu is
2316 * booting or shutting down. Requires console_sem.
2318 static int have_callable_console(void)
2320 struct console *con;
2322 for_each_console(con)
2323 if ((con->flags & CON_ENABLED) &&
2324 (con->flags & CON_ANYTIME))
2331 * Can we actually use the console at this time on this cpu?
2333 * Console drivers may assume that per-cpu resources have been allocated. So
2334 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2335 * call them until this CPU is officially up.
2337 static inline int can_use_console(void)
2339 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2343 * console_unlock - unlock the console system
2345 * Releases the console_lock which the caller holds on the console system
2346 * and the console driver list.
2348 * While the console_lock was held, console output may have been buffered
2349 * by printk(). If this is the case, console_unlock(); emits
2350 * the output prior to releasing the lock.
2352 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2354 * console_unlock(); may be called from any context.
2356 void console_unlock(void)
2358 static char ext_text[CONSOLE_EXT_LOG_MAX];
2359 static char text[LOG_LINE_MAX + PREFIX_MAX];
2360 unsigned long flags;
2361 bool do_cond_resched, retry;
2363 if (console_suspended) {
2369 * Console drivers are called with interrupts disabled, so
2370 * @console_may_schedule should be cleared before; however, we may
2371 * end up dumping a lot of lines, for example, if called from
2372 * console registration path, and should invoke cond_resched()
2373 * between lines if allowable. Not doing so can cause a very long
2374 * scheduling stall on a slow console leading to RCU stall and
2375 * softlockup warnings which exacerbate the issue with more
2376 * messages practically incapacitating the system.
2378 * console_trylock() is not able to detect the preemptive
2379 * context reliably. Therefore the value must be stored before
2380 * and cleared after the the "again" goto label.
2382 do_cond_resched = console_may_schedule;
2384 console_may_schedule = 0;
2387 * We released the console_sem lock, so we need to recheck if
2388 * cpu is online and (if not) is there at least one CON_ANYTIME
2391 if (!can_use_console()) {
2398 struct printk_log *msg;
2402 printk_safe_enter_irqsave(flags);
2403 raw_spin_lock(&logbuf_lock);
2404 if (console_seq < log_first_seq) {
2406 "** %llu printk messages dropped **\n",
2407 log_first_seq - console_seq);
2409 /* messages are gone, move to first one */
2410 console_seq = log_first_seq;
2411 console_idx = log_first_idx;
2416 if (console_seq == log_next_seq)
2419 msg = log_from_idx(console_idx);
2420 if (suppress_message_printing(msg->level)) {
2422 * Skip record we have buffered and already printed
2423 * directly to the console when we received it, and
2424 * record that has level above the console loglevel.
2426 console_idx = log_next(console_idx);
2431 /* Output to all consoles once old messages replayed. */
2432 if (unlikely(exclusive_console &&
2433 console_seq >= exclusive_console_stop_seq)) {
2434 exclusive_console = NULL;
2437 len += msg_print_text(msg,
2438 console_msg_format & MSG_FORMAT_SYSLOG,
2439 printk_time, text + len, sizeof(text) - len);
2440 if (nr_ext_console_drivers) {
2441 ext_len = msg_print_ext_header(ext_text,
2444 ext_len += msg_print_ext_body(ext_text + ext_len,
2445 sizeof(ext_text) - ext_len,
2446 log_dict(msg), msg->dict_len,
2447 log_text(msg), msg->text_len);
2449 console_idx = log_next(console_idx);
2451 raw_spin_unlock(&logbuf_lock);
2454 * While actively printing out messages, if another printk()
2455 * were to occur on another CPU, it may wait for this one to
2456 * finish. This task can not be preempted if there is a
2457 * waiter waiting to take over.
2459 console_lock_spinning_enable();
2461 stop_critical_timings(); /* don't trace print latency */
2462 call_console_drivers(ext_text, ext_len, text, len);
2463 start_critical_timings();
2465 if (console_lock_spinning_disable_and_check()) {
2466 printk_safe_exit_irqrestore(flags);
2470 printk_safe_exit_irqrestore(flags);
2472 if (do_cond_resched)
2478 raw_spin_unlock(&logbuf_lock);
2483 * Someone could have filled up the buffer again, so re-check if there's
2484 * something to flush. In case we cannot trylock the console_sem again,
2485 * there's a new owner and the console_unlock() from them will do the
2486 * flush, no worries.
2488 raw_spin_lock(&logbuf_lock);
2489 retry = console_seq != log_next_seq;
2490 raw_spin_unlock(&logbuf_lock);
2491 printk_safe_exit_irqrestore(flags);
2493 if (retry && console_trylock())
2496 EXPORT_SYMBOL(console_unlock);
2499 * console_conditional_schedule - yield the CPU if required
2501 * If the console code is currently allowed to sleep, and
2502 * if this CPU should yield the CPU to another task, do
2505 * Must be called within console_lock();.
2507 void __sched console_conditional_schedule(void)
2509 if (console_may_schedule)
2512 EXPORT_SYMBOL(console_conditional_schedule);
2514 void console_unblank(void)
2519 * console_unblank can no longer be called in interrupt context unless
2520 * oops_in_progress is set to 1..
2522 if (oops_in_progress) {
2523 if (down_trylock_console_sem() != 0)
2529 console_may_schedule = 0;
2531 if ((c->flags & CON_ENABLED) && c->unblank)
2537 * console_flush_on_panic - flush console content on panic
2539 * Immediately output all pending messages no matter what.
2541 void console_flush_on_panic(void)
2544 * If someone else is holding the console lock, trylock will fail
2545 * and may_schedule may be set. Ignore and proceed to unlock so
2546 * that messages are flushed out. As this can be called from any
2547 * context and we don't want to get preempted while flushing,
2548 * ensure may_schedule is cleared.
2551 console_may_schedule = 0;
2556 * Return the console tty driver structure and its associated index
2558 struct tty_driver *console_device(int *index)
2561 struct tty_driver *driver = NULL;
2564 for_each_console(c) {
2567 driver = c->device(c, index);
2576 * Prevent further output on the passed console device so that (for example)
2577 * serial drivers can disable console output before suspending a port, and can
2578 * re-enable output afterwards.
2580 void console_stop(struct console *console)
2583 console->flags &= ~CON_ENABLED;
2586 EXPORT_SYMBOL(console_stop);
2588 void console_start(struct console *console)
2591 console->flags |= CON_ENABLED;
2594 EXPORT_SYMBOL(console_start);
2596 static int __read_mostly keep_bootcon;
2598 static int __init keep_bootcon_setup(char *str)
2601 pr_info("debug: skip boot console de-registration.\n");
2606 early_param("keep_bootcon", keep_bootcon_setup);
2609 * The console driver calls this routine during kernel initialization
2610 * to register the console printing procedure with printk() and to
2611 * print any messages that were printed by the kernel before the
2612 * console driver was initialized.
2614 * This can happen pretty early during the boot process (because of
2615 * early_printk) - sometimes before setup_arch() completes - be careful
2616 * of what kernel features are used - they may not be initialised yet.
2618 * There are two types of consoles - bootconsoles (early_printk) and
2619 * "real" consoles (everything which is not a bootconsole) which are
2620 * handled differently.
2621 * - Any number of bootconsoles can be registered at any time.
2622 * - As soon as a "real" console is registered, all bootconsoles
2623 * will be unregistered automatically.
2624 * - Once a "real" console is registered, any attempt to register a
2625 * bootconsoles will be rejected
2627 void register_console(struct console *newcon)
2630 unsigned long flags;
2631 struct console *bcon = NULL;
2632 struct console_cmdline *c;
2633 static bool has_preferred;
2635 if (console_drivers)
2636 for_each_console(bcon)
2637 if (WARN(bcon == newcon,
2638 "console '%s%d' already registered\n",
2639 bcon->name, bcon->index))
2643 * before we register a new CON_BOOT console, make sure we don't
2644 * already have a valid console
2646 if (console_drivers && newcon->flags & CON_BOOT) {
2647 /* find the last or real console */
2648 for_each_console(bcon) {
2649 if (!(bcon->flags & CON_BOOT)) {
2650 pr_info("Too late to register bootconsole %s%d\n",
2651 newcon->name, newcon->index);
2657 if (console_drivers && console_drivers->flags & CON_BOOT)
2658 bcon = console_drivers;
2660 if (!has_preferred || bcon || !console_drivers)
2661 has_preferred = preferred_console >= 0;
2664 * See if we want to use this console driver. If we
2665 * didn't select a console we take the first one
2666 * that registers here.
2668 if (!has_preferred) {
2669 if (newcon->index < 0)
2671 if (newcon->setup == NULL ||
2672 newcon->setup(newcon, NULL) == 0) {
2673 newcon->flags |= CON_ENABLED;
2674 if (newcon->device) {
2675 newcon->flags |= CON_CONSDEV;
2676 has_preferred = true;
2682 * See if this console matches one we selected on
2685 for (i = 0, c = console_cmdline;
2686 i < MAX_CMDLINECONSOLES && c->name[0];
2688 if (!newcon->match ||
2689 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2690 /* default matching */
2691 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2692 if (strcmp(c->name, newcon->name) != 0)
2694 if (newcon->index >= 0 &&
2695 newcon->index != c->index)
2697 if (newcon->index < 0)
2698 newcon->index = c->index;
2700 if (_braille_register_console(newcon, c))
2703 if (newcon->setup &&
2704 newcon->setup(newcon, c->options) != 0)
2708 newcon->flags |= CON_ENABLED;
2709 if (i == preferred_console) {
2710 newcon->flags |= CON_CONSDEV;
2711 has_preferred = true;
2716 if (!(newcon->flags & CON_ENABLED))
2720 * If we have a bootconsole, and are switching to a real console,
2721 * don't print everything out again, since when the boot console, and
2722 * the real console are the same physical device, it's annoying to
2723 * see the beginning boot messages twice
2725 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2726 newcon->flags &= ~CON_PRINTBUFFER;
2729 * Put this console in the list - keep the
2730 * preferred driver at the head of the list.
2733 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2734 newcon->next = console_drivers;
2735 console_drivers = newcon;
2737 newcon->next->flags &= ~CON_CONSDEV;
2739 newcon->next = console_drivers->next;
2740 console_drivers->next = newcon;
2743 if (newcon->flags & CON_EXTENDED)
2744 nr_ext_console_drivers++;
2746 if (newcon->flags & CON_PRINTBUFFER) {
2748 * console_unlock(); will print out the buffered messages
2751 logbuf_lock_irqsave(flags);
2752 console_seq = syslog_seq;
2753 console_idx = syslog_idx;
2755 * We're about to replay the log buffer. Only do this to the
2756 * just-registered console to avoid excessive message spam to
2757 * the already-registered consoles.
2759 * Set exclusive_console with disabled interrupts to reduce
2760 * race window with eventual console_flush_on_panic() that
2761 * ignores console_lock.
2763 exclusive_console = newcon;
2764 exclusive_console_stop_seq = console_seq;
2765 logbuf_unlock_irqrestore(flags);
2768 console_sysfs_notify();
2771 * By unregistering the bootconsoles after we enable the real console
2772 * we get the "console xxx enabled" message on all the consoles -
2773 * boot consoles, real consoles, etc - this is to ensure that end
2774 * users know there might be something in the kernel's log buffer that
2775 * went to the bootconsole (that they do not see on the real console)
2777 pr_info("%sconsole [%s%d] enabled\n",
2778 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2779 newcon->name, newcon->index);
2781 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2783 /* We need to iterate through all boot consoles, to make
2784 * sure we print everything out, before we unregister them.
2786 for_each_console(bcon)
2787 if (bcon->flags & CON_BOOT)
2788 unregister_console(bcon);
2791 EXPORT_SYMBOL(register_console);
2793 int unregister_console(struct console *console)
2795 struct console *a, *b;
2798 pr_info("%sconsole [%s%d] disabled\n",
2799 (console->flags & CON_BOOT) ? "boot" : "" ,
2800 console->name, console->index);
2802 res = _braille_unregister_console(console);
2808 if (console_drivers == console) {
2809 console_drivers=console->next;
2811 } else if (console_drivers) {
2812 for (a=console_drivers->next, b=console_drivers ;
2813 a; b=a, a=b->next) {
2822 if (!res && (console->flags & CON_EXTENDED))
2823 nr_ext_console_drivers--;
2826 * If this isn't the last console and it has CON_CONSDEV set, we
2827 * need to set it on the next preferred console.
2829 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2830 console_drivers->flags |= CON_CONSDEV;
2832 console->flags &= ~CON_ENABLED;
2834 console_sysfs_notify();
2837 EXPORT_SYMBOL(unregister_console);
2840 * Initialize the console device. This is called *early*, so
2841 * we can't necessarily depend on lots of kernel help here.
2842 * Just do some early initializations, and do the complex setup
2845 void __init console_init(void)
2849 initcall_entry_t *ce;
2851 /* Setup the default TTY line discipline. */
2855 * set up the console device so that later boot sequences can
2856 * inform about problems etc..
2858 ce = __con_initcall_start;
2859 trace_initcall_level("console");
2860 while (ce < __con_initcall_end) {
2861 call = initcall_from_entry(ce);
2862 trace_initcall_start(call);
2864 trace_initcall_finish(call, ret);
2870 * Some boot consoles access data that is in the init section and which will
2871 * be discarded after the initcalls have been run. To make sure that no code
2872 * will access this data, unregister the boot consoles in a late initcall.
2874 * If for some reason, such as deferred probe or the driver being a loadable
2875 * module, the real console hasn't registered yet at this point, there will
2876 * be a brief interval in which no messages are logged to the console, which
2877 * makes it difficult to diagnose problems that occur during this time.
2879 * To mitigate this problem somewhat, only unregister consoles whose memory
2880 * intersects with the init section. Note that all other boot consoles will
2881 * get unregistred when the real preferred console is registered.
2883 static int __init printk_late_init(void)
2885 struct console *con;
2888 for_each_console(con) {
2889 if (!(con->flags & CON_BOOT))
2892 /* Check addresses that might be used for enabled consoles. */
2893 if (init_section_intersects(con, sizeof(*con)) ||
2894 init_section_contains(con->write, 0) ||
2895 init_section_contains(con->read, 0) ||
2896 init_section_contains(con->device, 0) ||
2897 init_section_contains(con->unblank, 0) ||
2898 init_section_contains(con->data, 0)) {
2900 * Please, consider moving the reported consoles out
2901 * of the init section.
2903 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2904 con->name, con->index);
2905 unregister_console(con);
2908 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2909 console_cpu_notify);
2911 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2912 console_cpu_notify, NULL);
2916 late_initcall(printk_late_init);
2918 #if defined CONFIG_PRINTK
2920 * Delayed printk version, for scheduler-internal messages:
2922 #define PRINTK_PENDING_WAKEUP 0x01
2923 #define PRINTK_PENDING_OUTPUT 0x02
2925 static DEFINE_PER_CPU(int, printk_pending);
2927 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2929 int pending = __this_cpu_xchg(printk_pending, 0);
2931 if (pending & PRINTK_PENDING_OUTPUT) {
2932 /* If trylock fails, someone else is doing the printing */
2933 if (console_trylock())
2937 if (pending & PRINTK_PENDING_WAKEUP)
2938 wake_up_interruptible(&log_wait);
2941 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2942 .func = wake_up_klogd_work_func,
2943 .flags = IRQ_WORK_LAZY,
2946 void wake_up_klogd(void)
2949 if (waitqueue_active(&log_wait)) {
2950 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2951 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2956 void defer_console_output(void)
2959 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2960 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2964 int vprintk_deferred(const char *fmt, va_list args)
2968 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2969 defer_console_output();
2974 int printk_deferred(const char *fmt, ...)
2979 va_start(args, fmt);
2980 r = vprintk_deferred(fmt, args);
2987 * printk rate limiting, lifted from the networking subsystem.
2989 * This enforces a rate limit: not more than 10 kernel messages
2990 * every 5s to make a denial-of-service attack impossible.
2992 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2994 int __printk_ratelimit(const char *func)
2996 return ___ratelimit(&printk_ratelimit_state, func);
2998 EXPORT_SYMBOL(__printk_ratelimit);
3001 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3002 * @caller_jiffies: pointer to caller's state
3003 * @interval_msecs: minimum interval between prints
3005 * printk_timed_ratelimit() returns true if more than @interval_msecs
3006 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3009 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3010 unsigned int interval_msecs)
3012 unsigned long elapsed = jiffies - *caller_jiffies;
3014 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3017 *caller_jiffies = jiffies;
3020 EXPORT_SYMBOL(printk_timed_ratelimit);
3022 static DEFINE_SPINLOCK(dump_list_lock);
3023 static LIST_HEAD(dump_list);
3026 * kmsg_dump_register - register a kernel log dumper.
3027 * @dumper: pointer to the kmsg_dumper structure
3029 * Adds a kernel log dumper to the system. The dump callback in the
3030 * structure will be called when the kernel oopses or panics and must be
3031 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3033 int kmsg_dump_register(struct kmsg_dumper *dumper)
3035 unsigned long flags;
3038 /* The dump callback needs to be set */
3042 spin_lock_irqsave(&dump_list_lock, flags);
3043 /* Don't allow registering multiple times */
3044 if (!dumper->registered) {
3045 dumper->registered = 1;
3046 list_add_tail_rcu(&dumper->list, &dump_list);
3049 spin_unlock_irqrestore(&dump_list_lock, flags);
3053 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3056 * kmsg_dump_unregister - unregister a kmsg dumper.
3057 * @dumper: pointer to the kmsg_dumper structure
3059 * Removes a dump device from the system. Returns zero on success and
3060 * %-EINVAL otherwise.
3062 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3064 unsigned long flags;
3067 spin_lock_irqsave(&dump_list_lock, flags);
3068 if (dumper->registered) {
3069 dumper->registered = 0;
3070 list_del_rcu(&dumper->list);
3073 spin_unlock_irqrestore(&dump_list_lock, flags);
3078 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3080 static bool always_kmsg_dump;
3081 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3084 * kmsg_dump - dump kernel log to kernel message dumpers.
3085 * @reason: the reason (oops, panic etc) for dumping
3087 * Call each of the registered dumper's dump() callback, which can
3088 * retrieve the kmsg records with kmsg_dump_get_line() or
3089 * kmsg_dump_get_buffer().
3091 void kmsg_dump(enum kmsg_dump_reason reason)
3093 struct kmsg_dumper *dumper;
3094 unsigned long flags;
3096 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3100 list_for_each_entry_rcu(dumper, &dump_list, list) {
3101 if (dumper->max_reason && reason > dumper->max_reason)
3104 /* initialize iterator with data about the stored records */
3105 dumper->active = true;
3107 logbuf_lock_irqsave(flags);
3108 dumper->cur_seq = clear_seq;
3109 dumper->cur_idx = clear_idx;
3110 dumper->next_seq = log_next_seq;
3111 dumper->next_idx = log_next_idx;
3112 logbuf_unlock_irqrestore(flags);
3114 /* invoke dumper which will iterate over records */
3115 dumper->dump(dumper, reason);
3117 /* reset iterator */
3118 dumper->active = false;
3124 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3125 * @dumper: registered kmsg dumper
3126 * @syslog: include the "<4>" prefixes
3127 * @line: buffer to copy the line to
3128 * @size: maximum size of the buffer
3129 * @len: length of line placed into buffer
3131 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3132 * record, and copy one record into the provided buffer.
3134 * Consecutive calls will return the next available record moving
3135 * towards the end of the buffer with the youngest messages.
3137 * A return value of FALSE indicates that there are no more records to
3140 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3142 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3143 char *line, size_t size, size_t *len)
3145 struct printk_log *msg;
3149 if (!dumper->active)
3152 if (dumper->cur_seq < log_first_seq) {
3153 /* messages are gone, move to first available one */
3154 dumper->cur_seq = log_first_seq;
3155 dumper->cur_idx = log_first_idx;
3159 if (dumper->cur_seq >= log_next_seq)
3162 msg = log_from_idx(dumper->cur_idx);
3163 l = msg_print_text(msg, syslog, printk_time, line, size);
3165 dumper->cur_idx = log_next(dumper->cur_idx);
3175 * kmsg_dump_get_line - retrieve one kmsg log line
3176 * @dumper: registered kmsg dumper
3177 * @syslog: include the "<4>" prefixes
3178 * @line: buffer to copy the line to
3179 * @size: maximum size of the buffer
3180 * @len: length of line placed into buffer
3182 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3183 * record, and copy one record into the provided buffer.
3185 * Consecutive calls will return the next available record moving
3186 * towards the end of the buffer with the youngest messages.
3188 * A return value of FALSE indicates that there are no more records to
3191 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3192 char *line, size_t size, size_t *len)
3194 unsigned long flags;
3197 logbuf_lock_irqsave(flags);
3198 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3199 logbuf_unlock_irqrestore(flags);
3203 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3206 * kmsg_dump_get_buffer - copy kmsg log lines
3207 * @dumper: registered kmsg dumper
3208 * @syslog: include the "<4>" prefixes
3209 * @buf: buffer to copy the line to
3210 * @size: maximum size of the buffer
3211 * @len: length of line placed into buffer
3213 * Start at the end of the kmsg buffer and fill the provided buffer
3214 * with as many of the the *youngest* kmsg records that fit into it.
3215 * If the buffer is large enough, all available kmsg records will be
3216 * copied with a single call.
3218 * Consecutive calls will fill the buffer with the next block of
3219 * available older records, not including the earlier retrieved ones.
3221 * A return value of FALSE indicates that there are no more records to
3224 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3225 char *buf, size_t size, size_t *len)
3227 unsigned long flags;
3234 bool time = printk_time;
3236 if (!dumper->active)
3239 logbuf_lock_irqsave(flags);
3240 if (dumper->cur_seq < log_first_seq) {
3241 /* messages are gone, move to first available one */
3242 dumper->cur_seq = log_first_seq;
3243 dumper->cur_idx = log_first_idx;
3247 if (dumper->cur_seq >= dumper->next_seq) {
3248 logbuf_unlock_irqrestore(flags);
3252 /* calculate length of entire buffer */
3253 seq = dumper->cur_seq;
3254 idx = dumper->cur_idx;
3255 while (seq < dumper->next_seq) {
3256 struct printk_log *msg = log_from_idx(idx);
3258 l += msg_print_text(msg, true, time, NULL, 0);
3259 idx = log_next(idx);
3263 /* move first record forward until length fits into the buffer */
3264 seq = dumper->cur_seq;
3265 idx = dumper->cur_idx;
3266 while (l > size && seq < dumper->next_seq) {
3267 struct printk_log *msg = log_from_idx(idx);
3269 l -= msg_print_text(msg, true, time, NULL, 0);
3270 idx = log_next(idx);
3274 /* last message in next interation */
3279 while (seq < dumper->next_seq) {
3280 struct printk_log *msg = log_from_idx(idx);
3282 l += msg_print_text(msg, syslog, time, buf + l, size - l);
3283 idx = log_next(idx);
3287 dumper->next_seq = next_seq;
3288 dumper->next_idx = next_idx;
3290 logbuf_unlock_irqrestore(flags);
3296 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3299 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3300 * @dumper: registered kmsg dumper
3302 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3303 * kmsg_dump_get_buffer() can be called again and used multiple
3304 * times within the same dumper.dump() callback.
3306 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3308 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3310 dumper->cur_seq = clear_seq;
3311 dumper->cur_idx = clear_idx;
3312 dumper->next_seq = log_next_seq;
3313 dumper->next_idx = log_next_idx;
3317 * kmsg_dump_rewind - reset the interator
3318 * @dumper: registered kmsg dumper
3320 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3321 * kmsg_dump_get_buffer() can be called again and used multiple
3322 * times within the same dumper.dump() callback.
3324 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3326 unsigned long flags;
3328 logbuf_lock_irqsave(flags);
3329 kmsg_dump_rewind_nolock(dumper);
3330 logbuf_unlock_irqrestore(flags);
3332 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);