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 */
69 atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
70 EXPORT_SYMBOL(ignore_console_lock_warning);
73 * Low level drivers may need that to know if they can schedule in
74 * their unblank() callback or not. So let's export it.
77 EXPORT_SYMBOL(oops_in_progress);
80 * console_sem protects the console_drivers list, and also
81 * provides serialisation for access to the entire console
84 static DEFINE_SEMAPHORE(console_sem);
85 struct console *console_drivers;
86 EXPORT_SYMBOL_GPL(console_drivers);
89 static struct lockdep_map console_lock_dep_map = {
90 .name = "console_lock"
94 enum devkmsg_log_bits {
95 __DEVKMSG_LOG_BIT_ON = 0,
96 __DEVKMSG_LOG_BIT_OFF,
97 __DEVKMSG_LOG_BIT_LOCK,
100 enum devkmsg_log_masks {
101 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
102 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
103 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
106 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
107 #define DEVKMSG_LOG_MASK_DEFAULT 0
109 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
111 static int __control_devkmsg(char *str)
116 if (!strncmp(str, "on", 2)) {
117 devkmsg_log = DEVKMSG_LOG_MASK_ON;
119 } else if (!strncmp(str, "off", 3)) {
120 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
122 } else if (!strncmp(str, "ratelimit", 9)) {
123 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
129 static int __init control_devkmsg(char *str)
131 if (__control_devkmsg(str) < 0)
135 * Set sysctl string accordingly:
137 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
138 strcpy(devkmsg_log_str, "on");
139 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
140 strcpy(devkmsg_log_str, "off");
141 /* else "ratelimit" which is set by default. */
144 * Sysctl cannot change it anymore. The kernel command line setting of
145 * this parameter is to force the setting to be permanent throughout the
146 * runtime of the system. This is a precation measure against userspace
147 * trying to be a smarta** and attempting to change it up on us.
149 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
153 __setup("printk.devkmsg=", control_devkmsg);
155 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
157 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
158 void __user *buffer, size_t *lenp, loff_t *ppos)
160 char old_str[DEVKMSG_STR_MAX_SIZE];
165 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
169 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
172 err = proc_dostring(table, write, buffer, lenp, ppos);
177 err = __control_devkmsg(devkmsg_log_str);
180 * Do not accept an unknown string OR a known string with
183 if (err < 0 || (err + 1 != *lenp)) {
185 /* ... and restore old setting. */
187 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
196 /* Number of registered extended console drivers. */
197 static int nr_ext_console_drivers;
200 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
201 * macros instead of functions so that _RET_IP_ contains useful information.
203 #define down_console_sem() do { \
205 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
208 static int __down_trylock_console_sem(unsigned long ip)
214 * Here and in __up_console_sem() we need to be in safe mode,
215 * because spindump/WARN/etc from under console ->lock will
216 * deadlock in printk()->down_trylock_console_sem() otherwise.
218 printk_safe_enter_irqsave(flags);
219 lock_failed = down_trylock(&console_sem);
220 printk_safe_exit_irqrestore(flags);
224 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
227 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
229 static void __up_console_sem(unsigned long ip)
233 mutex_release(&console_lock_dep_map, 1, ip);
235 printk_safe_enter_irqsave(flags);
237 printk_safe_exit_irqrestore(flags);
239 #define up_console_sem() __up_console_sem(_RET_IP_)
242 * This is used for debugging the mess that is the VT code by
243 * keeping track if we have the console semaphore held. It's
244 * definitely not the perfect debug tool (we don't know if _WE_
245 * hold it and are racing, but it helps tracking those weird code
246 * paths in the console code where we end up in places I want
247 * locked without the console sempahore held).
249 static int console_locked, console_suspended;
252 * If exclusive_console is non-NULL then only this console is to be printed to.
254 static struct console *exclusive_console;
257 * Array of consoles built from command line options (console=)
260 #define MAX_CMDLINECONSOLES 8
262 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
264 static int preferred_console = -1;
265 int console_set_on_cmdline;
266 EXPORT_SYMBOL(console_set_on_cmdline);
268 /* Flag: console code may call schedule() */
269 static int console_may_schedule;
271 enum con_msg_format_flags {
272 MSG_FORMAT_DEFAULT = 0,
273 MSG_FORMAT_SYSLOG = (1 << 0),
276 static int console_msg_format = MSG_FORMAT_DEFAULT;
279 * The printk log buffer consists of a chain of concatenated variable
280 * length records. Every record starts with a record header, containing
281 * the overall length of the record.
283 * The heads to the first and last entry in the buffer, as well as the
284 * sequence numbers of these entries are maintained when messages are
287 * If the heads indicate available messages, the length in the header
288 * tells the start next message. A length == 0 for the next message
289 * indicates a wrap-around to the beginning of the buffer.
291 * Every record carries the monotonic timestamp in microseconds, as well as
292 * the standard userspace syslog level and syslog facility. The usual
293 * kernel messages use LOG_KERN; userspace-injected messages always carry
294 * a matching syslog facility, by default LOG_USER. The origin of every
295 * message can be reliably determined that way.
297 * The human readable log message directly follows the message header. The
298 * length of the message text is stored in the header, the stored message
301 * Optionally, a message can carry a dictionary of properties (key/value pairs),
302 * to provide userspace with a machine-readable message context.
304 * Examples for well-defined, commonly used property names are:
305 * DEVICE=b12:8 device identifier
309 * +sound:card0 subsystem:devname
310 * SUBSYSTEM=pci driver-core subsystem name
312 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
313 * follows directly after a '=' character. Every property is terminated by
314 * a '\0' character. The last property is not terminated.
316 * Example of a message structure:
317 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
318 * 0008 34 00 record is 52 bytes long
319 * 000a 0b 00 text is 11 bytes long
320 * 000c 1f 00 dictionary is 23 bytes long
321 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
322 * 0010 69 74 27 73 20 61 20 6c "it's a l"
324 * 001b 44 45 56 49 43 "DEVIC"
325 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
326 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
328 * 0032 00 00 00 padding to next message header
330 * The 'struct printk_log' buffer header must never be directly exported to
331 * userspace, it is a kernel-private implementation detail that might
332 * need to be changed in the future, when the requirements change.
334 * /dev/kmsg exports the structured data in the following line format:
335 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
337 * Users of the export format should ignore possible additional values
338 * separated by ',', and find the message after the ';' character.
340 * The optional key/value pairs are attached as continuation lines starting
341 * with a space character and terminated by a newline. All possible
342 * non-prinatable characters are escaped in the "\xff" notation.
346 LOG_NEWLINE = 2, /* text ended with a newline */
347 LOG_PREFIX = 4, /* text started with a prefix */
348 LOG_CONT = 8, /* text is a fragment of a continuation line */
352 u64 ts_nsec; /* timestamp in nanoseconds */
353 u16 len; /* length of entire record */
354 u16 text_len; /* length of text buffer */
355 u16 dict_len; /* length of dictionary buffer */
356 u8 facility; /* syslog facility */
357 u8 flags:5; /* internal record flags */
358 u8 level:3; /* syslog level */
360 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
361 __packed __aligned(4)
366 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
367 * within the scheduler's rq lock. It must be released before calling
368 * console_unlock() or anything else that might wake up a process.
370 DEFINE_RAW_SPINLOCK(logbuf_lock);
373 * Helper macros to lock/unlock logbuf_lock and switch between
374 * printk-safe/unsafe modes.
376 #define logbuf_lock_irq() \
378 printk_safe_enter_irq(); \
379 raw_spin_lock(&logbuf_lock); \
382 #define logbuf_unlock_irq() \
384 raw_spin_unlock(&logbuf_lock); \
385 printk_safe_exit_irq(); \
388 #define logbuf_lock_irqsave(flags) \
390 printk_safe_enter_irqsave(flags); \
391 raw_spin_lock(&logbuf_lock); \
394 #define logbuf_unlock_irqrestore(flags) \
396 raw_spin_unlock(&logbuf_lock); \
397 printk_safe_exit_irqrestore(flags); \
401 DECLARE_WAIT_QUEUE_HEAD(log_wait);
402 /* the next printk record to read by syslog(READ) or /proc/kmsg */
403 static u64 syslog_seq;
404 static u32 syslog_idx;
405 static size_t syslog_partial;
407 /* index and sequence number of the first record stored in the buffer */
408 static u64 log_first_seq;
409 static u32 log_first_idx;
411 /* index and sequence number of the next record to store in the buffer */
412 static u64 log_next_seq;
413 static u32 log_next_idx;
415 /* the next printk record to write to the console */
416 static u64 console_seq;
417 static u32 console_idx;
418 static u64 exclusive_console_stop_seq;
420 /* the next printk record to read after the last 'clear' command */
421 static u64 clear_seq;
422 static u32 clear_idx;
424 #define PREFIX_MAX 32
425 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
427 #define LOG_LEVEL(v) ((v) & 0x07)
428 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
431 #define LOG_ALIGN __alignof__(struct printk_log)
432 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
433 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
434 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
435 static char *log_buf = __log_buf;
436 static u32 log_buf_len = __LOG_BUF_LEN;
438 /* Return log buffer address */
439 char *log_buf_addr_get(void)
444 /* Return log buffer size */
445 u32 log_buf_len_get(void)
450 /* human readable text of the record */
451 static char *log_text(const struct printk_log *msg)
453 return (char *)msg + sizeof(struct printk_log);
456 /* optional key/value pair dictionary attached to the record */
457 static char *log_dict(const struct printk_log *msg)
459 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
462 /* get record by index; idx must point to valid msg */
463 static struct printk_log *log_from_idx(u32 idx)
465 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
468 * A length == 0 record is the end of buffer marker. Wrap around and
469 * read the message at the start of the buffer.
472 return (struct printk_log *)log_buf;
476 /* get next record; idx must point to valid msg */
477 static u32 log_next(u32 idx)
479 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
481 /* length == 0 indicates the end of the buffer; wrap */
483 * A length == 0 record is the end of buffer marker. Wrap around and
484 * read the message at the start of the buffer as *this* one, and
485 * return the one after that.
488 msg = (struct printk_log *)log_buf;
491 return idx + msg->len;
495 * Check whether there is enough free space for the given message.
497 * The same values of first_idx and next_idx mean that the buffer
498 * is either empty or full.
500 * If the buffer is empty, we must respect the position of the indexes.
501 * They cannot be reset to the beginning of the buffer.
503 static int logbuf_has_space(u32 msg_size, bool empty)
507 if (log_next_idx > log_first_idx || empty)
508 free = max(log_buf_len - log_next_idx, log_first_idx);
510 free = log_first_idx - log_next_idx;
513 * We need space also for an empty header that signalizes wrapping
516 return free >= msg_size + sizeof(struct printk_log);
519 static int log_make_free_space(u32 msg_size)
521 while (log_first_seq < log_next_seq &&
522 !logbuf_has_space(msg_size, false)) {
523 /* drop old messages until we have enough contiguous space */
524 log_first_idx = log_next(log_first_idx);
528 if (clear_seq < log_first_seq) {
529 clear_seq = log_first_seq;
530 clear_idx = log_first_idx;
533 /* sequence numbers are equal, so the log buffer is empty */
534 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
540 /* compute the message size including the padding bytes */
541 static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
545 size = sizeof(struct printk_log) + text_len + dict_len;
546 *pad_len = (-size) & (LOG_ALIGN - 1);
553 * Define how much of the log buffer we could take at maximum. The value
554 * must be greater than two. Note that only half of the buffer is available
555 * when the index points to the middle.
557 #define MAX_LOG_TAKE_PART 4
558 static const char trunc_msg[] = "<truncated>";
560 static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
561 u16 *dict_len, u32 *pad_len)
564 * The message should not take the whole buffer. Otherwise, it might
565 * get removed too soon.
567 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
568 if (*text_len > max_text_len)
569 *text_len = max_text_len;
570 /* enable the warning message */
571 *trunc_msg_len = strlen(trunc_msg);
572 /* disable the "dict" completely */
574 /* compute the size again, count also the warning message */
575 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
578 /* insert record into the buffer, discard old ones, update heads */
579 static int log_store(int facility, int level,
580 enum log_flags flags, u64 ts_nsec,
581 const char *dict, u16 dict_len,
582 const char *text, u16 text_len)
584 struct printk_log *msg;
586 u16 trunc_msg_len = 0;
588 /* number of '\0' padding bytes to next message */
589 size = msg_used_size(text_len, dict_len, &pad_len);
591 if (log_make_free_space(size)) {
592 /* truncate the message if it is too long for empty buffer */
593 size = truncate_msg(&text_len, &trunc_msg_len,
594 &dict_len, &pad_len);
595 /* survive when the log buffer is too small for trunc_msg */
596 if (log_make_free_space(size))
600 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
602 * This message + an additional empty header does not fit
603 * at the end of the buffer. Add an empty header with len == 0
604 * to signify a wrap around.
606 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
611 msg = (struct printk_log *)(log_buf + log_next_idx);
612 memcpy(log_text(msg), text, text_len);
613 msg->text_len = text_len;
615 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
616 msg->text_len += trunc_msg_len;
618 memcpy(log_dict(msg), dict, dict_len);
619 msg->dict_len = dict_len;
620 msg->facility = facility;
621 msg->level = level & 7;
622 msg->flags = flags & 0x1f;
624 msg->ts_nsec = ts_nsec;
626 msg->ts_nsec = local_clock();
627 memset(log_dict(msg) + dict_len, 0, pad_len);
631 log_next_idx += msg->len;
634 return msg->text_len;
637 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
639 static int syslog_action_restricted(int type)
644 * Unless restricted, we allow "read all" and "get buffer size"
647 return type != SYSLOG_ACTION_READ_ALL &&
648 type != SYSLOG_ACTION_SIZE_BUFFER;
651 static int check_syslog_permissions(int type, int source)
654 * If this is from /proc/kmsg and we've already opened it, then we've
655 * already done the capabilities checks at open time.
657 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
660 if (syslog_action_restricted(type)) {
661 if (capable(CAP_SYSLOG))
664 * For historical reasons, accept CAP_SYS_ADMIN too, with
667 if (capable(CAP_SYS_ADMIN)) {
668 pr_warn_once("%s (%d): Attempt to access syslog with "
669 "CAP_SYS_ADMIN but no CAP_SYSLOG "
671 current->comm, task_pid_nr(current));
677 return security_syslog(type);
680 static void append_char(char **pp, char *e, char c)
686 static ssize_t msg_print_ext_header(char *buf, size_t size,
687 struct printk_log *msg, u64 seq)
689 u64 ts_usec = msg->ts_nsec;
691 do_div(ts_usec, 1000);
693 return scnprintf(buf, size, "%u,%llu,%llu,%c;",
694 (msg->facility << 3) | msg->level, seq, ts_usec,
695 msg->flags & LOG_CONT ? 'c' : '-');
698 static ssize_t msg_print_ext_body(char *buf, size_t size,
699 char *dict, size_t dict_len,
700 char *text, size_t text_len)
702 char *p = buf, *e = buf + size;
705 /* escape non-printable characters */
706 for (i = 0; i < text_len; i++) {
707 unsigned char c = text[i];
709 if (c < ' ' || c >= 127 || c == '\\')
710 p += scnprintf(p, e - p, "\\x%02x", c);
712 append_char(&p, e, c);
714 append_char(&p, e, '\n');
719 for (i = 0; i < dict_len; i++) {
720 unsigned char c = dict[i];
723 append_char(&p, e, ' ');
728 append_char(&p, e, '\n');
733 if (c < ' ' || c >= 127 || c == '\\') {
734 p += scnprintf(p, e - p, "\\x%02x", c);
738 append_char(&p, e, c);
740 append_char(&p, e, '\n');
746 /* /dev/kmsg - userspace message inject/listen interface */
747 struct devkmsg_user {
750 struct ratelimit_state rs;
752 char buf[CONSOLE_EXT_LOG_MAX];
755 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
758 int level = default_message_loglevel;
759 int facility = 1; /* LOG_USER */
760 struct file *file = iocb->ki_filp;
761 struct devkmsg_user *user = file->private_data;
762 size_t len = iov_iter_count(from);
765 if (!user || len > LOG_LINE_MAX)
768 /* Ignore when user logging is disabled. */
769 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
772 /* Ratelimit when not explicitly enabled. */
773 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
774 if (!___ratelimit(&user->rs, current->comm))
778 buf = kmalloc(len+1, GFP_KERNEL);
783 if (!copy_from_iter_full(buf, len, from)) {
789 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
790 * the decimal value represents 32bit, the lower 3 bit are the log
791 * level, the rest are the log facility.
793 * If no prefix or no userspace facility is specified, we
794 * enforce LOG_USER, to be able to reliably distinguish
795 * kernel-generated messages from userspace-injected ones.
798 if (line[0] == '<') {
802 u = simple_strtoul(line + 1, &endp, 10);
803 if (endp && endp[0] == '>') {
804 level = LOG_LEVEL(u);
805 if (LOG_FACILITY(u) != 0)
806 facility = LOG_FACILITY(u);
813 printk_emit(facility, level, NULL, 0, "%s", line);
818 static ssize_t devkmsg_read(struct file *file, char __user *buf,
819 size_t count, loff_t *ppos)
821 struct devkmsg_user *user = file->private_data;
822 struct printk_log *msg;
829 ret = mutex_lock_interruptible(&user->lock);
834 while (user->seq == log_next_seq) {
835 if (file->f_flags & O_NONBLOCK) {
842 ret = wait_event_interruptible(log_wait,
843 user->seq != log_next_seq);
849 if (user->seq < log_first_seq) {
850 /* our last seen message is gone, return error and reset */
851 user->idx = log_first_idx;
852 user->seq = log_first_seq;
858 msg = log_from_idx(user->idx);
859 len = msg_print_ext_header(user->buf, sizeof(user->buf),
861 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
862 log_dict(msg), msg->dict_len,
863 log_text(msg), msg->text_len);
865 user->idx = log_next(user->idx);
874 if (copy_to_user(buf, user->buf, len)) {
880 mutex_unlock(&user->lock);
884 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
886 struct devkmsg_user *user = file->private_data;
897 /* the first record */
898 user->idx = log_first_idx;
899 user->seq = log_first_seq;
903 * The first record after the last SYSLOG_ACTION_CLEAR,
904 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
905 * changes no global state, and does not clear anything.
907 user->idx = clear_idx;
908 user->seq = clear_seq;
911 /* after the last record */
912 user->idx = log_next_idx;
913 user->seq = log_next_seq;
922 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
924 struct devkmsg_user *user = file->private_data;
928 return EPOLLERR|EPOLLNVAL;
930 poll_wait(file, &log_wait, wait);
933 if (user->seq < log_next_seq) {
934 /* return error when data has vanished underneath us */
935 if (user->seq < log_first_seq)
936 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
938 ret = EPOLLIN|EPOLLRDNORM;
945 static int devkmsg_open(struct inode *inode, struct file *file)
947 struct devkmsg_user *user;
950 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
953 /* write-only does not need any file context */
954 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
955 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
961 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
965 ratelimit_default_init(&user->rs);
966 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
968 mutex_init(&user->lock);
971 user->idx = log_first_idx;
972 user->seq = log_first_seq;
975 file->private_data = user;
979 static int devkmsg_release(struct inode *inode, struct file *file)
981 struct devkmsg_user *user = file->private_data;
986 ratelimit_state_exit(&user->rs);
988 mutex_destroy(&user->lock);
993 const struct file_operations kmsg_fops = {
994 .open = devkmsg_open,
995 .read = devkmsg_read,
996 .write_iter = devkmsg_write,
997 .llseek = devkmsg_llseek,
998 .poll = devkmsg_poll,
999 .release = devkmsg_release,
1002 #ifdef CONFIG_CRASH_CORE
1004 * This appends the listed symbols to /proc/vmcore
1006 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1007 * obtain access to symbols that are otherwise very difficult to locate. These
1008 * symbols are specifically used so that utilities can access and extract the
1009 * dmesg log from a vmcore file after a crash.
1011 void log_buf_vmcoreinfo_setup(void)
1013 VMCOREINFO_SYMBOL(log_buf);
1014 VMCOREINFO_SYMBOL(log_buf_len);
1015 VMCOREINFO_SYMBOL(log_first_idx);
1016 VMCOREINFO_SYMBOL(clear_idx);
1017 VMCOREINFO_SYMBOL(log_next_idx);
1019 * Export struct printk_log size and field offsets. User space tools can
1020 * parse it and detect any changes to structure down the line.
1022 VMCOREINFO_STRUCT_SIZE(printk_log);
1023 VMCOREINFO_OFFSET(printk_log, ts_nsec);
1024 VMCOREINFO_OFFSET(printk_log, len);
1025 VMCOREINFO_OFFSET(printk_log, text_len);
1026 VMCOREINFO_OFFSET(printk_log, dict_len);
1030 /* requested log_buf_len from kernel cmdline */
1031 static unsigned long __initdata new_log_buf_len;
1033 /* we practice scaling the ring buffer by powers of 2 */
1034 static void __init log_buf_len_update(u64 size)
1036 if (size > (u64)LOG_BUF_LEN_MAX) {
1037 size = (u64)LOG_BUF_LEN_MAX;
1038 pr_err("log_buf over 2G is not supported.\n");
1042 size = roundup_pow_of_two(size);
1043 if (size > log_buf_len)
1044 new_log_buf_len = (unsigned long)size;
1047 /* save requested log_buf_len since it's too early to process it */
1048 static int __init log_buf_len_setup(char *str)
1055 size = memparse(str, &str);
1057 log_buf_len_update(size);
1061 early_param("log_buf_len", log_buf_len_setup);
1064 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1066 static void __init log_buf_add_cpu(void)
1068 unsigned int cpu_extra;
1071 * archs should set up cpu_possible_bits properly with
1072 * set_cpu_possible() after setup_arch() but just in
1073 * case lets ensure this is valid.
1075 if (num_possible_cpus() == 1)
1078 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1080 /* by default this will only continue through for large > 64 CPUs */
1081 if (cpu_extra <= __LOG_BUF_LEN / 2)
1084 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1085 __LOG_CPU_MAX_BUF_LEN);
1086 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1088 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1090 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1092 #else /* !CONFIG_SMP */
1093 static inline void log_buf_add_cpu(void) {}
1094 #endif /* CONFIG_SMP */
1096 void __init setup_log_buf(int early)
1098 unsigned long flags;
1102 if (log_buf != __log_buf)
1105 if (!early && !new_log_buf_len)
1108 if (!new_log_buf_len)
1113 memblock_alloc(new_log_buf_len, LOG_ALIGN);
1115 new_log_buf = memblock_alloc_nopanic(new_log_buf_len,
1119 if (unlikely(!new_log_buf)) {
1120 pr_err("log_buf_len: %lu bytes not available\n",
1125 logbuf_lock_irqsave(flags);
1126 log_buf_len = new_log_buf_len;
1127 log_buf = new_log_buf;
1128 new_log_buf_len = 0;
1129 free = __LOG_BUF_LEN - log_next_idx;
1130 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1131 logbuf_unlock_irqrestore(flags);
1133 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1134 pr_info("early log buf free: %u(%u%%)\n",
1135 free, (free * 100) / __LOG_BUF_LEN);
1138 static bool __read_mostly ignore_loglevel;
1140 static int __init ignore_loglevel_setup(char *str)
1142 ignore_loglevel = true;
1143 pr_info("debug: ignoring loglevel setting.\n");
1148 early_param("ignore_loglevel", ignore_loglevel_setup);
1149 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1150 MODULE_PARM_DESC(ignore_loglevel,
1151 "ignore loglevel setting (prints all kernel messages to the console)");
1153 static bool suppress_message_printing(int level)
1155 return (level >= console_loglevel && !ignore_loglevel);
1158 #ifdef CONFIG_BOOT_PRINTK_DELAY
1160 static int boot_delay; /* msecs delay after each printk during bootup */
1161 static unsigned long long loops_per_msec; /* based on boot_delay */
1163 static int __init boot_delay_setup(char *str)
1167 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1168 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1170 get_option(&str, &boot_delay);
1171 if (boot_delay > 10 * 1000)
1174 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1175 "HZ: %d, loops_per_msec: %llu\n",
1176 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1179 early_param("boot_delay", boot_delay_setup);
1181 static void boot_delay_msec(int level)
1183 unsigned long long k;
1184 unsigned long timeout;
1186 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1187 || suppress_message_printing(level)) {
1191 k = (unsigned long long)loops_per_msec * boot_delay;
1193 timeout = jiffies + msecs_to_jiffies(boot_delay);
1198 * use (volatile) jiffies to prevent
1199 * compiler reduction; loop termination via jiffies
1200 * is secondary and may or may not happen.
1202 if (time_after(jiffies, timeout))
1204 touch_nmi_watchdog();
1208 static inline void boot_delay_msec(int level)
1213 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1214 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1216 static size_t print_time(u64 ts, char *buf)
1218 unsigned long rem_nsec;
1223 rem_nsec = do_div(ts, 1000000000);
1226 return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
1228 return sprintf(buf, "[%5lu.%06lu] ",
1229 (unsigned long)ts, rem_nsec / 1000);
1232 static size_t print_prefix(const struct printk_log *msg, bool syslog, char *buf)
1235 unsigned int prefix = (msg->facility << 3) | msg->level;
1239 len += sprintf(buf, "<%u>", prefix);
1244 else if (prefix > 99)
1246 else if (prefix > 9)
1251 len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
1255 static size_t msg_print_text(const struct printk_log *msg, bool syslog, char *buf, size_t size)
1257 const char *text = log_text(msg);
1258 size_t text_size = msg->text_len;
1262 const char *next = memchr(text, '\n', text_size);
1266 text_len = next - text;
1268 text_size -= next - text;
1270 text_len = text_size;
1274 if (print_prefix(msg, syslog, NULL) +
1275 text_len + 1 >= size - len)
1278 len += print_prefix(msg, syslog, buf + len);
1279 memcpy(buf + len, text, text_len);
1283 /* SYSLOG_ACTION_* buffer size only calculation */
1284 len += print_prefix(msg, syslog, NULL);
1295 static int syslog_print(char __user *buf, int size)
1298 struct printk_log *msg;
1301 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1310 if (syslog_seq < log_first_seq) {
1311 /* messages are gone, move to first one */
1312 syslog_seq = log_first_seq;
1313 syslog_idx = log_first_idx;
1316 if (syslog_seq == log_next_seq) {
1317 logbuf_unlock_irq();
1321 skip = syslog_partial;
1322 msg = log_from_idx(syslog_idx);
1323 n = msg_print_text(msg, true, text, LOG_LINE_MAX + PREFIX_MAX);
1324 if (n - syslog_partial <= size) {
1325 /* message fits into buffer, move forward */
1326 syslog_idx = log_next(syslog_idx);
1328 n -= syslog_partial;
1331 /* partial read(), remember position */
1333 syslog_partial += n;
1336 logbuf_unlock_irq();
1341 if (copy_to_user(buf, text + skip, n)) {
1356 static int syslog_print_all(char __user *buf, int size, bool clear)
1364 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1370 * Find first record that fits, including all following records,
1371 * into the user-provided buffer for this dump.
1375 while (seq < log_next_seq) {
1376 struct printk_log *msg = log_from_idx(idx);
1378 len += msg_print_text(msg, true, NULL, 0);
1379 idx = log_next(idx);
1383 /* move first record forward until length fits into the buffer */
1386 while (len > size && seq < log_next_seq) {
1387 struct printk_log *msg = log_from_idx(idx);
1389 len -= msg_print_text(msg, true, NULL, 0);
1390 idx = log_next(idx);
1394 /* last message fitting into this dump */
1395 next_seq = log_next_seq;
1398 while (len >= 0 && seq < next_seq) {
1399 struct printk_log *msg = log_from_idx(idx);
1402 textlen = msg_print_text(msg, true, text,
1403 LOG_LINE_MAX + PREFIX_MAX);
1408 idx = log_next(idx);
1411 logbuf_unlock_irq();
1412 if (copy_to_user(buf + len, text, textlen))
1418 if (seq < log_first_seq) {
1419 /* messages are gone, move to next one */
1420 seq = log_first_seq;
1421 idx = log_first_idx;
1426 clear_seq = log_next_seq;
1427 clear_idx = log_next_idx;
1429 logbuf_unlock_irq();
1435 static void syslog_clear(void)
1438 clear_seq = log_next_seq;
1439 clear_idx = log_next_idx;
1440 logbuf_unlock_irq();
1443 int do_syslog(int type, char __user *buf, int len, int source)
1446 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1449 error = check_syslog_permissions(type, source);
1454 case SYSLOG_ACTION_CLOSE: /* Close log */
1456 case SYSLOG_ACTION_OPEN: /* Open log */
1458 case SYSLOG_ACTION_READ: /* Read from log */
1459 if (!buf || len < 0)
1463 if (!access_ok(VERIFY_WRITE, buf, len))
1465 error = wait_event_interruptible(log_wait,
1466 syslog_seq != log_next_seq);
1469 error = syslog_print(buf, len);
1471 /* Read/clear last kernel messages */
1472 case SYSLOG_ACTION_READ_CLEAR:
1475 /* Read last kernel messages */
1476 case SYSLOG_ACTION_READ_ALL:
1477 if (!buf || len < 0)
1481 if (!access_ok(VERIFY_WRITE, buf, len))
1483 error = syslog_print_all(buf, len, clear);
1485 /* Clear ring buffer */
1486 case SYSLOG_ACTION_CLEAR:
1489 /* Disable logging to console */
1490 case SYSLOG_ACTION_CONSOLE_OFF:
1491 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1492 saved_console_loglevel = console_loglevel;
1493 console_loglevel = minimum_console_loglevel;
1495 /* Enable logging to console */
1496 case SYSLOG_ACTION_CONSOLE_ON:
1497 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1498 console_loglevel = saved_console_loglevel;
1499 saved_console_loglevel = LOGLEVEL_DEFAULT;
1502 /* Set level of messages printed to console */
1503 case SYSLOG_ACTION_CONSOLE_LEVEL:
1504 if (len < 1 || len > 8)
1506 if (len < minimum_console_loglevel)
1507 len = minimum_console_loglevel;
1508 console_loglevel = len;
1509 /* Implicitly re-enable logging to console */
1510 saved_console_loglevel = LOGLEVEL_DEFAULT;
1512 /* Number of chars in the log buffer */
1513 case SYSLOG_ACTION_SIZE_UNREAD:
1515 if (syslog_seq < log_first_seq) {
1516 /* messages are gone, move to first one */
1517 syslog_seq = log_first_seq;
1518 syslog_idx = log_first_idx;
1521 if (source == SYSLOG_FROM_PROC) {
1523 * Short-cut for poll(/"proc/kmsg") which simply checks
1524 * for pending data, not the size; return the count of
1525 * records, not the length.
1527 error = log_next_seq - syslog_seq;
1529 u64 seq = syslog_seq;
1530 u32 idx = syslog_idx;
1532 while (seq < log_next_seq) {
1533 struct printk_log *msg = log_from_idx(idx);
1535 error += msg_print_text(msg, true, NULL, 0);
1536 idx = log_next(idx);
1539 error -= syslog_partial;
1541 logbuf_unlock_irq();
1543 /* Size of the log buffer */
1544 case SYSLOG_ACTION_SIZE_BUFFER:
1545 error = log_buf_len;
1555 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1557 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1561 * Special console_lock variants that help to reduce the risk of soft-lockups.
1562 * They allow to pass console_lock to another printk() call using a busy wait.
1565 #ifdef CONFIG_LOCKDEP
1566 static struct lockdep_map console_owner_dep_map = {
1567 .name = "console_owner"
1571 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1572 static struct task_struct *console_owner;
1573 static bool console_waiter;
1576 * console_lock_spinning_enable - mark beginning of code where another
1577 * thread might safely busy wait
1579 * This basically converts console_lock into a spinlock. This marks
1580 * the section where the console_lock owner can not sleep, because
1581 * there may be a waiter spinning (like a spinlock). Also it must be
1582 * ready to hand over the lock at the end of the section.
1584 static void console_lock_spinning_enable(void)
1586 raw_spin_lock(&console_owner_lock);
1587 console_owner = current;
1588 raw_spin_unlock(&console_owner_lock);
1590 /* The waiter may spin on us after setting console_owner */
1591 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1595 * console_lock_spinning_disable_and_check - mark end of code where another
1596 * thread was able to busy wait and check if there is a waiter
1598 * This is called at the end of the section where spinning is allowed.
1599 * It has two functions. First, it is a signal that it is no longer
1600 * safe to start busy waiting for the lock. Second, it checks if
1601 * there is a busy waiter and passes the lock rights to her.
1603 * Important: Callers lose the lock if there was a busy waiter.
1604 * They must not touch items synchronized by console_lock
1607 * Return: 1 if the lock rights were passed, 0 otherwise.
1609 static int console_lock_spinning_disable_and_check(void)
1613 raw_spin_lock(&console_owner_lock);
1614 waiter = READ_ONCE(console_waiter);
1615 console_owner = NULL;
1616 raw_spin_unlock(&console_owner_lock);
1619 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1623 /* The waiter is now free to continue */
1624 WRITE_ONCE(console_waiter, false);
1626 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1629 * Hand off console_lock to waiter. The waiter will perform
1630 * the up(). After this, the waiter is the console_lock owner.
1632 mutex_release(&console_lock_dep_map, 1, _THIS_IP_);
1637 * console_trylock_spinning - try to get console_lock by busy waiting
1639 * This allows to busy wait for the console_lock when the current
1640 * owner is running in specially marked sections. It means that
1641 * the current owner is running and cannot reschedule until it
1642 * is ready to lose the lock.
1644 * Return: 1 if we got the lock, 0 othrewise
1646 static int console_trylock_spinning(void)
1648 struct task_struct *owner = NULL;
1651 unsigned long flags;
1653 if (console_trylock())
1656 printk_safe_enter_irqsave(flags);
1658 raw_spin_lock(&console_owner_lock);
1659 owner = READ_ONCE(console_owner);
1660 waiter = READ_ONCE(console_waiter);
1661 if (!waiter && owner && owner != current) {
1662 WRITE_ONCE(console_waiter, true);
1665 raw_spin_unlock(&console_owner_lock);
1668 * If there is an active printk() writing to the
1669 * consoles, instead of having it write our data too,
1670 * see if we can offload that load from the active
1671 * printer, and do some printing ourselves.
1672 * Go into a spin only if there isn't already a waiter
1673 * spinning, and there is an active printer, and
1674 * that active printer isn't us (recursive printk?).
1677 printk_safe_exit_irqrestore(flags);
1681 /* We spin waiting for the owner to release us */
1682 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1683 /* Owner will clear console_waiter on hand off */
1684 while (READ_ONCE(console_waiter))
1686 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1688 printk_safe_exit_irqrestore(flags);
1690 * The owner passed the console lock to us.
1691 * Since we did not spin on console lock, annotate
1692 * this as a trylock. Otherwise lockdep will
1695 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1701 * Call the console drivers, asking them to write out
1702 * log_buf[start] to log_buf[end - 1].
1703 * The console_lock must be held.
1705 static void call_console_drivers(const char *ext_text, size_t ext_len,
1706 const char *text, size_t len)
1708 struct console *con;
1710 trace_console_rcuidle(text, len);
1712 if (!console_drivers)
1715 for_each_console(con) {
1716 if (exclusive_console && con != exclusive_console)
1718 if (!(con->flags & CON_ENABLED))
1722 if (!cpu_online(smp_processor_id()) &&
1723 !(con->flags & CON_ANYTIME))
1725 if (con->flags & CON_EXTENDED)
1726 con->write(con, ext_text, ext_len);
1728 con->write(con, text, len);
1732 int printk_delay_msec __read_mostly;
1734 static inline void printk_delay(void)
1736 if (unlikely(printk_delay_msec)) {
1737 int m = printk_delay_msec;
1741 touch_nmi_watchdog();
1747 * Continuation lines are buffered, and not committed to the record buffer
1748 * until the line is complete, or a race forces it. The line fragments
1749 * though, are printed immediately to the consoles to ensure everything has
1750 * reached the console in case of a kernel crash.
1752 static struct cont {
1753 char buf[LOG_LINE_MAX];
1754 size_t len; /* length == 0 means unused buffer */
1755 struct task_struct *owner; /* task of first print*/
1756 u64 ts_nsec; /* time of first print */
1757 u8 level; /* log level of first message */
1758 u8 facility; /* log facility of first message */
1759 enum log_flags flags; /* prefix, newline flags */
1762 static void cont_flush(void)
1767 log_store(cont.facility, cont.level, cont.flags, cont.ts_nsec,
1768 NULL, 0, cont.buf, cont.len);
1772 static bool cont_add(int facility, int level, enum log_flags flags, const char *text, size_t len)
1774 /* If the line gets too long, split it up in separate records. */
1775 if (cont.len + len > sizeof(cont.buf)) {
1781 cont.facility = facility;
1783 cont.owner = current;
1784 cont.ts_nsec = local_clock();
1788 memcpy(cont.buf + cont.len, text, len);
1791 // The original flags come from the first line,
1792 // but later continuations can add a newline.
1793 if (flags & LOG_NEWLINE) {
1794 cont.flags |= LOG_NEWLINE;
1801 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)
1804 * If an earlier line was buffered, and we're a continuation
1805 * write from the same process, try to add it to the buffer.
1808 if (cont.owner == current && (lflags & LOG_CONT)) {
1809 if (cont_add(facility, level, lflags, text, text_len))
1812 /* Otherwise, make sure it's flushed */
1816 /* Skip empty continuation lines that couldn't be added - they just flush */
1817 if (!text_len && (lflags & LOG_CONT))
1820 /* If it doesn't end in a newline, try to buffer the current line */
1821 if (!(lflags & LOG_NEWLINE)) {
1822 if (cont_add(facility, level, lflags, text, text_len))
1826 /* Store it in the record log */
1827 return log_store(facility, level, lflags, 0, dict, dictlen, text, text_len);
1830 /* Must be called under logbuf_lock. */
1831 int vprintk_store(int facility, int level,
1832 const char *dict, size_t dictlen,
1833 const char *fmt, va_list args)
1835 static char textbuf[LOG_LINE_MAX];
1836 char *text = textbuf;
1838 enum log_flags lflags = 0;
1841 * The printf needs to come first; we need the syslog
1842 * prefix which might be passed-in as a parameter.
1844 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1846 /* mark and strip a trailing newline */
1847 if (text_len && text[text_len-1] == '\n') {
1849 lflags |= LOG_NEWLINE;
1852 /* strip kernel syslog prefix and extract log level or control flags */
1853 if (facility == 0) {
1856 while ((kern_level = printk_get_level(text)) != 0) {
1857 switch (kern_level) {
1859 if (level == LOGLEVEL_DEFAULT)
1860 level = kern_level - '0';
1862 case 'd': /* KERN_DEFAULT */
1863 lflags |= LOG_PREFIX;
1865 case 'c': /* KERN_CONT */
1874 if (level == LOGLEVEL_DEFAULT)
1875 level = default_message_loglevel;
1878 lflags |= LOG_PREFIX|LOG_NEWLINE;
1880 return log_output(facility, level, lflags,
1881 dict, dictlen, text, text_len);
1884 asmlinkage int vprintk_emit(int facility, int level,
1885 const char *dict, size_t dictlen,
1886 const char *fmt, va_list args)
1889 bool in_sched = false, pending_output;
1890 unsigned long flags;
1893 if (level == LOGLEVEL_SCHED) {
1894 level = LOGLEVEL_DEFAULT;
1898 boot_delay_msec(level);
1901 /* This stops the holder of console_sem just where we want him */
1902 logbuf_lock_irqsave(flags);
1903 curr_log_seq = log_next_seq;
1904 printed_len = vprintk_store(facility, level, dict, dictlen, fmt, args);
1905 pending_output = (curr_log_seq != log_next_seq);
1906 logbuf_unlock_irqrestore(flags);
1908 /* If called from the scheduler, we can not call up(). */
1909 if (!in_sched && pending_output) {
1911 * Disable preemption to avoid being preempted while holding
1912 * console_sem which would prevent anyone from printing to
1917 * Try to acquire and then immediately release the console
1918 * semaphore. The release will print out buffers and wake up
1919 * /dev/kmsg and syslog() users.
1921 if (console_trylock_spinning())
1930 EXPORT_SYMBOL(vprintk_emit);
1932 asmlinkage int vprintk(const char *fmt, va_list args)
1934 return vprintk_func(fmt, args);
1936 EXPORT_SYMBOL(vprintk);
1938 asmlinkage int printk_emit(int facility, int level,
1939 const char *dict, size_t dictlen,
1940 const char *fmt, ...)
1945 va_start(args, fmt);
1946 r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1951 EXPORT_SYMBOL(printk_emit);
1953 int vprintk_default(const char *fmt, va_list args)
1957 #ifdef CONFIG_KGDB_KDB
1958 /* Allow to pass printk() to kdb but avoid a recursion. */
1959 if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
1960 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
1964 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1968 EXPORT_SYMBOL_GPL(vprintk_default);
1971 * printk - print a kernel message
1972 * @fmt: format string
1974 * This is printk(). It can be called from any context. We want it to work.
1976 * We try to grab the console_lock. If we succeed, it's easy - we log the
1977 * output and call the console drivers. If we fail to get the semaphore, we
1978 * place the output into the log buffer and return. The current holder of
1979 * the console_sem will notice the new output in console_unlock(); and will
1980 * send it to the consoles before releasing the lock.
1982 * One effect of this deferred printing is that code which calls printk() and
1983 * then changes console_loglevel may break. This is because console_loglevel
1984 * is inspected when the actual printing occurs.
1989 * See the vsnprintf() documentation for format string extensions over C99.
1991 asmlinkage __visible int printk(const char *fmt, ...)
1996 va_start(args, fmt);
1997 r = vprintk_func(fmt, args);
2002 EXPORT_SYMBOL(printk);
2004 #else /* CONFIG_PRINTK */
2006 #define LOG_LINE_MAX 0
2007 #define PREFIX_MAX 0
2009 static u64 syslog_seq;
2010 static u32 syslog_idx;
2011 static u64 console_seq;
2012 static u32 console_idx;
2013 static u64 exclusive_console_stop_seq;
2014 static u64 log_first_seq;
2015 static u32 log_first_idx;
2016 static u64 log_next_seq;
2017 static char *log_text(const struct printk_log *msg) { return NULL; }
2018 static char *log_dict(const struct printk_log *msg) { return NULL; }
2019 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2020 static u32 log_next(u32 idx) { return 0; }
2021 static ssize_t msg_print_ext_header(char *buf, size_t size,
2022 struct printk_log *msg,
2023 u64 seq) { return 0; }
2024 static ssize_t msg_print_ext_body(char *buf, size_t size,
2025 char *dict, size_t dict_len,
2026 char *text, size_t text_len) { return 0; }
2027 static void console_lock_spinning_enable(void) { }
2028 static int console_lock_spinning_disable_and_check(void) { return 0; }
2029 static void call_console_drivers(const char *ext_text, size_t ext_len,
2030 const char *text, size_t len) {}
2031 static size_t msg_print_text(const struct printk_log *msg,
2032 bool syslog, char *buf, size_t size) { return 0; }
2033 static bool suppress_message_printing(int level) { return false; }
2035 #endif /* CONFIG_PRINTK */
2037 #ifdef CONFIG_EARLY_PRINTK
2038 struct console *early_console;
2040 asmlinkage __visible void early_printk(const char *fmt, ...)
2050 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2053 early_console->write(early_console, buf, n);
2057 static int __add_preferred_console(char *name, int idx, char *options,
2060 struct console_cmdline *c;
2064 * See if this tty is not yet registered, and
2065 * if we have a slot free.
2067 for (i = 0, c = console_cmdline;
2068 i < MAX_CMDLINECONSOLES && c->name[0];
2070 if (strcmp(c->name, name) == 0 && c->index == idx) {
2072 preferred_console = i;
2076 if (i == MAX_CMDLINECONSOLES)
2079 preferred_console = i;
2080 strlcpy(c->name, name, sizeof(c->name));
2081 c->options = options;
2082 braille_set_options(c, brl_options);
2088 static int __init console_msg_format_setup(char *str)
2090 if (!strcmp(str, "syslog"))
2091 console_msg_format = MSG_FORMAT_SYSLOG;
2092 if (!strcmp(str, "default"))
2093 console_msg_format = MSG_FORMAT_DEFAULT;
2096 __setup("console_msg_format=", console_msg_format_setup);
2099 * Set up a console. Called via do_early_param() in init/main.c
2100 * for each "console=" parameter in the boot command line.
2102 static int __init console_setup(char *str)
2104 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2105 char *s, *options, *brl_options = NULL;
2108 if (_braille_console_setup(&str, &brl_options))
2112 * Decode str into name, index, options.
2114 if (str[0] >= '0' && str[0] <= '9') {
2115 strcpy(buf, "ttyS");
2116 strncpy(buf + 4, str, sizeof(buf) - 5);
2118 strncpy(buf, str, sizeof(buf) - 1);
2120 buf[sizeof(buf) - 1] = 0;
2121 options = strchr(str, ',');
2125 if (!strcmp(str, "ttya"))
2126 strcpy(buf, "ttyS0");
2127 if (!strcmp(str, "ttyb"))
2128 strcpy(buf, "ttyS1");
2130 for (s = buf; *s; s++)
2131 if (isdigit(*s) || *s == ',')
2133 idx = simple_strtoul(s, NULL, 10);
2136 __add_preferred_console(buf, idx, options, brl_options);
2137 console_set_on_cmdline = 1;
2140 __setup("console=", console_setup);
2143 * add_preferred_console - add a device to the list of preferred consoles.
2144 * @name: device name
2145 * @idx: device index
2146 * @options: options for this console
2148 * The last preferred console added will be used for kernel messages
2149 * and stdin/out/err for init. Normally this is used by console_setup
2150 * above to handle user-supplied console arguments; however it can also
2151 * be used by arch-specific code either to override the user or more
2152 * commonly to provide a default console (ie from PROM variables) when
2153 * the user has not supplied one.
2155 int add_preferred_console(char *name, int idx, char *options)
2157 return __add_preferred_console(name, idx, options, NULL);
2160 bool console_suspend_enabled = true;
2161 EXPORT_SYMBOL(console_suspend_enabled);
2163 static int __init console_suspend_disable(char *str)
2165 console_suspend_enabled = false;
2168 __setup("no_console_suspend", console_suspend_disable);
2169 module_param_named(console_suspend, console_suspend_enabled,
2170 bool, S_IRUGO | S_IWUSR);
2171 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2172 " and hibernate operations");
2175 * suspend_console - suspend the console subsystem
2177 * This disables printk() while we go into suspend states
2179 void suspend_console(void)
2181 if (!console_suspend_enabled)
2183 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2185 console_suspended = 1;
2189 void resume_console(void)
2191 if (!console_suspend_enabled)
2194 console_suspended = 0;
2199 * console_cpu_notify - print deferred console messages after CPU hotplug
2202 * If printk() is called from a CPU that is not online yet, the messages
2203 * will be printed on the console only if there are CON_ANYTIME consoles.
2204 * This function is called when a new CPU comes online (or fails to come
2205 * up) or goes offline.
2207 static int console_cpu_notify(unsigned int cpu)
2209 if (!cpuhp_tasks_frozen) {
2210 /* If trylock fails, someone else is doing the printing */
2211 if (console_trylock())
2218 * console_lock - lock the console system for exclusive use.
2220 * Acquires a lock which guarantees that the caller has
2221 * exclusive access to the console system and the console_drivers list.
2223 * Can sleep, returns nothing.
2225 void console_lock(void)
2230 if (console_suspended)
2233 console_may_schedule = 1;
2235 EXPORT_SYMBOL(console_lock);
2238 * console_trylock - try to lock the console system for exclusive use.
2240 * Try to acquire a lock which guarantees that the caller has exclusive
2241 * access to the console system and the console_drivers list.
2243 * returns 1 on success, and 0 on failure to acquire the lock.
2245 int console_trylock(void)
2247 if (down_trylock_console_sem())
2249 if (console_suspended) {
2254 console_may_schedule = 0;
2257 EXPORT_SYMBOL(console_trylock);
2259 int is_console_locked(void)
2261 return console_locked;
2263 EXPORT_SYMBOL(is_console_locked);
2266 * Check if we have any console that is capable of printing while cpu is
2267 * booting or shutting down. Requires console_sem.
2269 static int have_callable_console(void)
2271 struct console *con;
2273 for_each_console(con)
2274 if ((con->flags & CON_ENABLED) &&
2275 (con->flags & CON_ANYTIME))
2282 * Can we actually use the console at this time on this cpu?
2284 * Console drivers may assume that per-cpu resources have been allocated. So
2285 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2286 * call them until this CPU is officially up.
2288 static inline int can_use_console(void)
2290 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2294 * console_unlock - unlock the console system
2296 * Releases the console_lock which the caller holds on the console system
2297 * and the console driver list.
2299 * While the console_lock was held, console output may have been buffered
2300 * by printk(). If this is the case, console_unlock(); emits
2301 * the output prior to releasing the lock.
2303 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2305 * console_unlock(); may be called from any context.
2307 void console_unlock(void)
2309 static char ext_text[CONSOLE_EXT_LOG_MAX];
2310 static char text[LOG_LINE_MAX + PREFIX_MAX];
2311 unsigned long flags;
2312 bool do_cond_resched, retry;
2314 if (console_suspended) {
2320 * Console drivers are called with interrupts disabled, so
2321 * @console_may_schedule should be cleared before; however, we may
2322 * end up dumping a lot of lines, for example, if called from
2323 * console registration path, and should invoke cond_resched()
2324 * between lines if allowable. Not doing so can cause a very long
2325 * scheduling stall on a slow console leading to RCU stall and
2326 * softlockup warnings which exacerbate the issue with more
2327 * messages practically incapacitating the system.
2329 * console_trylock() is not able to detect the preemptive
2330 * context reliably. Therefore the value must be stored before
2331 * and cleared after the the "again" goto label.
2333 do_cond_resched = console_may_schedule;
2335 console_may_schedule = 0;
2338 * We released the console_sem lock, so we need to recheck if
2339 * cpu is online and (if not) is there at least one CON_ANYTIME
2342 if (!can_use_console()) {
2349 struct printk_log *msg;
2353 printk_safe_enter_irqsave(flags);
2354 raw_spin_lock(&logbuf_lock);
2355 if (console_seq < log_first_seq) {
2357 "** %llu printk messages dropped **\n",
2358 log_first_seq - console_seq);
2360 /* messages are gone, move to first one */
2361 console_seq = log_first_seq;
2362 console_idx = log_first_idx;
2367 if (console_seq == log_next_seq)
2370 msg = log_from_idx(console_idx);
2371 if (suppress_message_printing(msg->level)) {
2373 * Skip record we have buffered and already printed
2374 * directly to the console when we received it, and
2375 * record that has level above the console loglevel.
2377 console_idx = log_next(console_idx);
2382 /* Output to all consoles once old messages replayed. */
2383 if (unlikely(exclusive_console &&
2384 console_seq >= exclusive_console_stop_seq)) {
2385 exclusive_console = NULL;
2388 len += msg_print_text(msg,
2389 console_msg_format & MSG_FORMAT_SYSLOG,
2391 sizeof(text) - len);
2392 if (nr_ext_console_drivers) {
2393 ext_len = msg_print_ext_header(ext_text,
2396 ext_len += msg_print_ext_body(ext_text + ext_len,
2397 sizeof(ext_text) - ext_len,
2398 log_dict(msg), msg->dict_len,
2399 log_text(msg), msg->text_len);
2401 console_idx = log_next(console_idx);
2403 raw_spin_unlock(&logbuf_lock);
2406 * While actively printing out messages, if another printk()
2407 * were to occur on another CPU, it may wait for this one to
2408 * finish. This task can not be preempted if there is a
2409 * waiter waiting to take over.
2411 console_lock_spinning_enable();
2413 stop_critical_timings(); /* don't trace print latency */
2414 call_console_drivers(ext_text, ext_len, text, len);
2415 start_critical_timings();
2417 if (console_lock_spinning_disable_and_check()) {
2418 printk_safe_exit_irqrestore(flags);
2422 printk_safe_exit_irqrestore(flags);
2424 if (do_cond_resched)
2430 raw_spin_unlock(&logbuf_lock);
2435 * Someone could have filled up the buffer again, so re-check if there's
2436 * something to flush. In case we cannot trylock the console_sem again,
2437 * there's a new owner and the console_unlock() from them will do the
2438 * flush, no worries.
2440 raw_spin_lock(&logbuf_lock);
2441 retry = console_seq != log_next_seq;
2442 raw_spin_unlock(&logbuf_lock);
2443 printk_safe_exit_irqrestore(flags);
2445 if (retry && console_trylock())
2448 EXPORT_SYMBOL(console_unlock);
2451 * console_conditional_schedule - yield the CPU if required
2453 * If the console code is currently allowed to sleep, and
2454 * if this CPU should yield the CPU to another task, do
2457 * Must be called within console_lock();.
2459 void __sched console_conditional_schedule(void)
2461 if (console_may_schedule)
2464 EXPORT_SYMBOL(console_conditional_schedule);
2466 void console_unblank(void)
2471 * console_unblank can no longer be called in interrupt context unless
2472 * oops_in_progress is set to 1..
2474 if (oops_in_progress) {
2475 if (down_trylock_console_sem() != 0)
2481 console_may_schedule = 0;
2483 if ((c->flags & CON_ENABLED) && c->unblank)
2489 * console_flush_on_panic - flush console content on panic
2491 * Immediately output all pending messages no matter what.
2493 void console_flush_on_panic(void)
2496 * If someone else is holding the console lock, trylock will fail
2497 * and may_schedule may be set. Ignore and proceed to unlock so
2498 * that messages are flushed out. As this can be called from any
2499 * context and we don't want to get preempted while flushing,
2500 * ensure may_schedule is cleared.
2503 console_may_schedule = 0;
2508 * Return the console tty driver structure and its associated index
2510 struct tty_driver *console_device(int *index)
2513 struct tty_driver *driver = NULL;
2516 for_each_console(c) {
2519 driver = c->device(c, index);
2528 * Prevent further output on the passed console device so that (for example)
2529 * serial drivers can disable console output before suspending a port, and can
2530 * re-enable output afterwards.
2532 void console_stop(struct console *console)
2535 console->flags &= ~CON_ENABLED;
2538 EXPORT_SYMBOL(console_stop);
2540 void console_start(struct console *console)
2543 console->flags |= CON_ENABLED;
2546 EXPORT_SYMBOL(console_start);
2548 static int __read_mostly keep_bootcon;
2550 static int __init keep_bootcon_setup(char *str)
2553 pr_info("debug: skip boot console de-registration.\n");
2558 early_param("keep_bootcon", keep_bootcon_setup);
2561 * The console driver calls this routine during kernel initialization
2562 * to register the console printing procedure with printk() and to
2563 * print any messages that were printed by the kernel before the
2564 * console driver was initialized.
2566 * This can happen pretty early during the boot process (because of
2567 * early_printk) - sometimes before setup_arch() completes - be careful
2568 * of what kernel features are used - they may not be initialised yet.
2570 * There are two types of consoles - bootconsoles (early_printk) and
2571 * "real" consoles (everything which is not a bootconsole) which are
2572 * handled differently.
2573 * - Any number of bootconsoles can be registered at any time.
2574 * - As soon as a "real" console is registered, all bootconsoles
2575 * will be unregistered automatically.
2576 * - Once a "real" console is registered, any attempt to register a
2577 * bootconsoles will be rejected
2579 void register_console(struct console *newcon)
2582 unsigned long flags;
2583 struct console *bcon = NULL;
2584 struct console_cmdline *c;
2585 static bool has_preferred;
2587 if (console_drivers)
2588 for_each_console(bcon)
2589 if (WARN(bcon == newcon,
2590 "console '%s%d' already registered\n",
2591 bcon->name, bcon->index))
2595 * before we register a new CON_BOOT console, make sure we don't
2596 * already have a valid console
2598 if (console_drivers && newcon->flags & CON_BOOT) {
2599 /* find the last or real console */
2600 for_each_console(bcon) {
2601 if (!(bcon->flags & CON_BOOT)) {
2602 pr_info("Too late to register bootconsole %s%d\n",
2603 newcon->name, newcon->index);
2609 if (console_drivers && console_drivers->flags & CON_BOOT)
2610 bcon = console_drivers;
2612 if (!has_preferred || bcon || !console_drivers)
2613 has_preferred = preferred_console >= 0;
2616 * See if we want to use this console driver. If we
2617 * didn't select a console we take the first one
2618 * that registers here.
2620 if (!has_preferred) {
2621 if (newcon->index < 0)
2623 if (newcon->setup == NULL ||
2624 newcon->setup(newcon, NULL) == 0) {
2625 newcon->flags |= CON_ENABLED;
2626 if (newcon->device) {
2627 newcon->flags |= CON_CONSDEV;
2628 has_preferred = true;
2634 * See if this console matches one we selected on
2637 for (i = 0, c = console_cmdline;
2638 i < MAX_CMDLINECONSOLES && c->name[0];
2640 if (!newcon->match ||
2641 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2642 /* default matching */
2643 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2644 if (strcmp(c->name, newcon->name) != 0)
2646 if (newcon->index >= 0 &&
2647 newcon->index != c->index)
2649 if (newcon->index < 0)
2650 newcon->index = c->index;
2652 if (_braille_register_console(newcon, c))
2655 if (newcon->setup &&
2656 newcon->setup(newcon, c->options) != 0)
2660 newcon->flags |= CON_ENABLED;
2661 if (i == preferred_console) {
2662 newcon->flags |= CON_CONSDEV;
2663 has_preferred = true;
2668 if (!(newcon->flags & CON_ENABLED))
2672 * If we have a bootconsole, and are switching to a real console,
2673 * don't print everything out again, since when the boot console, and
2674 * the real console are the same physical device, it's annoying to
2675 * see the beginning boot messages twice
2677 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2678 newcon->flags &= ~CON_PRINTBUFFER;
2681 * Put this console in the list - keep the
2682 * preferred driver at the head of the list.
2685 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2686 newcon->next = console_drivers;
2687 console_drivers = newcon;
2689 newcon->next->flags &= ~CON_CONSDEV;
2691 newcon->next = console_drivers->next;
2692 console_drivers->next = newcon;
2695 if (newcon->flags & CON_EXTENDED)
2696 nr_ext_console_drivers++;
2698 if (newcon->flags & CON_PRINTBUFFER) {
2700 * console_unlock(); will print out the buffered messages
2703 logbuf_lock_irqsave(flags);
2704 console_seq = syslog_seq;
2705 console_idx = syslog_idx;
2707 * We're about to replay the log buffer. Only do this to the
2708 * just-registered console to avoid excessive message spam to
2709 * the already-registered consoles.
2711 * Set exclusive_console with disabled interrupts to reduce
2712 * race window with eventual console_flush_on_panic() that
2713 * ignores console_lock.
2715 exclusive_console = newcon;
2716 exclusive_console_stop_seq = console_seq;
2717 logbuf_unlock_irqrestore(flags);
2720 console_sysfs_notify();
2723 * By unregistering the bootconsoles after we enable the real console
2724 * we get the "console xxx enabled" message on all the consoles -
2725 * boot consoles, real consoles, etc - this is to ensure that end
2726 * users know there might be something in the kernel's log buffer that
2727 * went to the bootconsole (that they do not see on the real console)
2729 pr_info("%sconsole [%s%d] enabled\n",
2730 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2731 newcon->name, newcon->index);
2733 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2735 /* We need to iterate through all boot consoles, to make
2736 * sure we print everything out, before we unregister them.
2738 for_each_console(bcon)
2739 if (bcon->flags & CON_BOOT)
2740 unregister_console(bcon);
2743 EXPORT_SYMBOL(register_console);
2745 int unregister_console(struct console *console)
2747 struct console *a, *b;
2750 pr_info("%sconsole [%s%d] disabled\n",
2751 (console->flags & CON_BOOT) ? "boot" : "" ,
2752 console->name, console->index);
2754 res = _braille_unregister_console(console);
2760 if (console_drivers == console) {
2761 console_drivers=console->next;
2763 } else if (console_drivers) {
2764 for (a=console_drivers->next, b=console_drivers ;
2765 a; b=a, a=b->next) {
2774 if (!res && (console->flags & CON_EXTENDED))
2775 nr_ext_console_drivers--;
2778 * If this isn't the last console and it has CON_CONSDEV set, we
2779 * need to set it on the next preferred console.
2781 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2782 console_drivers->flags |= CON_CONSDEV;
2784 console->flags &= ~CON_ENABLED;
2786 console_sysfs_notify();
2789 EXPORT_SYMBOL(unregister_console);
2792 * Initialize the console device. This is called *early*, so
2793 * we can't necessarily depend on lots of kernel help here.
2794 * Just do some early initializations, and do the complex setup
2797 void __init console_init(void)
2801 initcall_entry_t *ce;
2803 /* Setup the default TTY line discipline. */
2807 * set up the console device so that later boot sequences can
2808 * inform about problems etc..
2810 ce = __con_initcall_start;
2811 trace_initcall_level("console");
2812 while (ce < __con_initcall_end) {
2813 call = initcall_from_entry(ce);
2814 trace_initcall_start(call);
2816 trace_initcall_finish(call, ret);
2822 * Some boot consoles access data that is in the init section and which will
2823 * be discarded after the initcalls have been run. To make sure that no code
2824 * will access this data, unregister the boot consoles in a late initcall.
2826 * If for some reason, such as deferred probe or the driver being a loadable
2827 * module, the real console hasn't registered yet at this point, there will
2828 * be a brief interval in which no messages are logged to the console, which
2829 * makes it difficult to diagnose problems that occur during this time.
2831 * To mitigate this problem somewhat, only unregister consoles whose memory
2832 * intersects with the init section. Note that all other boot consoles will
2833 * get unregistred when the real preferred console is registered.
2835 static int __init printk_late_init(void)
2837 struct console *con;
2840 for_each_console(con) {
2841 if (!(con->flags & CON_BOOT))
2844 /* Check addresses that might be used for enabled consoles. */
2845 if (init_section_intersects(con, sizeof(*con)) ||
2846 init_section_contains(con->write, 0) ||
2847 init_section_contains(con->read, 0) ||
2848 init_section_contains(con->device, 0) ||
2849 init_section_contains(con->unblank, 0) ||
2850 init_section_contains(con->data, 0)) {
2852 * Please, consider moving the reported consoles out
2853 * of the init section.
2855 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2856 con->name, con->index);
2857 unregister_console(con);
2860 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2861 console_cpu_notify);
2863 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2864 console_cpu_notify, NULL);
2868 late_initcall(printk_late_init);
2870 #if defined CONFIG_PRINTK
2872 * Delayed printk version, for scheduler-internal messages:
2874 #define PRINTK_PENDING_WAKEUP 0x01
2875 #define PRINTK_PENDING_OUTPUT 0x02
2877 static DEFINE_PER_CPU(int, printk_pending);
2879 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2881 int pending = __this_cpu_xchg(printk_pending, 0);
2883 if (pending & PRINTK_PENDING_OUTPUT) {
2884 /* If trylock fails, someone else is doing the printing */
2885 if (console_trylock())
2889 if (pending & PRINTK_PENDING_WAKEUP)
2890 wake_up_interruptible(&log_wait);
2893 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2894 .func = wake_up_klogd_work_func,
2895 .flags = IRQ_WORK_LAZY,
2898 void wake_up_klogd(void)
2901 if (waitqueue_active(&log_wait)) {
2902 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2903 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2908 void defer_console_output(void)
2911 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2912 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2916 int vprintk_deferred(const char *fmt, va_list args)
2920 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2921 defer_console_output();
2926 int printk_deferred(const char *fmt, ...)
2931 va_start(args, fmt);
2932 r = vprintk_deferred(fmt, args);
2939 * printk rate limiting, lifted from the networking subsystem.
2941 * This enforces a rate limit: not more than 10 kernel messages
2942 * every 5s to make a denial-of-service attack impossible.
2944 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2946 int __printk_ratelimit(const char *func)
2948 return ___ratelimit(&printk_ratelimit_state, func);
2950 EXPORT_SYMBOL(__printk_ratelimit);
2953 * printk_timed_ratelimit - caller-controlled printk ratelimiting
2954 * @caller_jiffies: pointer to caller's state
2955 * @interval_msecs: minimum interval between prints
2957 * printk_timed_ratelimit() returns true if more than @interval_msecs
2958 * milliseconds have elapsed since the last time printk_timed_ratelimit()
2961 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
2962 unsigned int interval_msecs)
2964 unsigned long elapsed = jiffies - *caller_jiffies;
2966 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
2969 *caller_jiffies = jiffies;
2972 EXPORT_SYMBOL(printk_timed_ratelimit);
2974 static DEFINE_SPINLOCK(dump_list_lock);
2975 static LIST_HEAD(dump_list);
2978 * kmsg_dump_register - register a kernel log dumper.
2979 * @dumper: pointer to the kmsg_dumper structure
2981 * Adds a kernel log dumper to the system. The dump callback in the
2982 * structure will be called when the kernel oopses or panics and must be
2983 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
2985 int kmsg_dump_register(struct kmsg_dumper *dumper)
2987 unsigned long flags;
2990 /* The dump callback needs to be set */
2994 spin_lock_irqsave(&dump_list_lock, flags);
2995 /* Don't allow registering multiple times */
2996 if (!dumper->registered) {
2997 dumper->registered = 1;
2998 list_add_tail_rcu(&dumper->list, &dump_list);
3001 spin_unlock_irqrestore(&dump_list_lock, flags);
3005 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3008 * kmsg_dump_unregister - unregister a kmsg dumper.
3009 * @dumper: pointer to the kmsg_dumper structure
3011 * Removes a dump device from the system. Returns zero on success and
3012 * %-EINVAL otherwise.
3014 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3016 unsigned long flags;
3019 spin_lock_irqsave(&dump_list_lock, flags);
3020 if (dumper->registered) {
3021 dumper->registered = 0;
3022 list_del_rcu(&dumper->list);
3025 spin_unlock_irqrestore(&dump_list_lock, flags);
3030 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3032 static bool always_kmsg_dump;
3033 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3036 * kmsg_dump - dump kernel log to kernel message dumpers.
3037 * @reason: the reason (oops, panic etc) for dumping
3039 * Call each of the registered dumper's dump() callback, which can
3040 * retrieve the kmsg records with kmsg_dump_get_line() or
3041 * kmsg_dump_get_buffer().
3043 void kmsg_dump(enum kmsg_dump_reason reason)
3045 struct kmsg_dumper *dumper;
3046 unsigned long flags;
3048 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3052 list_for_each_entry_rcu(dumper, &dump_list, list) {
3053 if (dumper->max_reason && reason > dumper->max_reason)
3056 /* initialize iterator with data about the stored records */
3057 dumper->active = true;
3059 logbuf_lock_irqsave(flags);
3060 dumper->cur_seq = clear_seq;
3061 dumper->cur_idx = clear_idx;
3062 dumper->next_seq = log_next_seq;
3063 dumper->next_idx = log_next_idx;
3064 logbuf_unlock_irqrestore(flags);
3066 /* invoke dumper which will iterate over records */
3067 dumper->dump(dumper, reason);
3069 /* reset iterator */
3070 dumper->active = false;
3076 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3077 * @dumper: registered kmsg dumper
3078 * @syslog: include the "<4>" prefixes
3079 * @line: buffer to copy the line to
3080 * @size: maximum size of the buffer
3081 * @len: length of line placed into buffer
3083 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3084 * record, and copy one record into the provided buffer.
3086 * Consecutive calls will return the next available record moving
3087 * towards the end of the buffer with the youngest messages.
3089 * A return value of FALSE indicates that there are no more records to
3092 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3094 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3095 char *line, size_t size, size_t *len)
3097 struct printk_log *msg;
3101 if (!dumper->active)
3104 if (dumper->cur_seq < log_first_seq) {
3105 /* messages are gone, move to first available one */
3106 dumper->cur_seq = log_first_seq;
3107 dumper->cur_idx = log_first_idx;
3111 if (dumper->cur_seq >= log_next_seq)
3114 msg = log_from_idx(dumper->cur_idx);
3115 l = msg_print_text(msg, syslog, line, size);
3117 dumper->cur_idx = log_next(dumper->cur_idx);
3127 * kmsg_dump_get_line - retrieve one kmsg log line
3128 * @dumper: registered kmsg dumper
3129 * @syslog: include the "<4>" prefixes
3130 * @line: buffer to copy the line to
3131 * @size: maximum size of the buffer
3132 * @len: length of line placed into buffer
3134 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3135 * record, and copy one record into the provided buffer.
3137 * Consecutive calls will return the next available record moving
3138 * towards the end of the buffer with the youngest messages.
3140 * A return value of FALSE indicates that there are no more records to
3143 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3144 char *line, size_t size, size_t *len)
3146 unsigned long flags;
3149 logbuf_lock_irqsave(flags);
3150 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3151 logbuf_unlock_irqrestore(flags);
3155 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3158 * kmsg_dump_get_buffer - copy kmsg log lines
3159 * @dumper: registered kmsg dumper
3160 * @syslog: include the "<4>" prefixes
3161 * @buf: buffer to copy the line to
3162 * @size: maximum size of the buffer
3163 * @len: length of line placed into buffer
3165 * Start at the end of the kmsg buffer and fill the provided buffer
3166 * with as many of the the *youngest* kmsg records that fit into it.
3167 * If the buffer is large enough, all available kmsg records will be
3168 * copied with a single call.
3170 * Consecutive calls will fill the buffer with the next block of
3171 * available older records, not including the earlier retrieved ones.
3173 * A return value of FALSE indicates that there are no more records to
3176 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3177 char *buf, size_t size, size_t *len)
3179 unsigned long flags;
3187 if (!dumper->active)
3190 logbuf_lock_irqsave(flags);
3191 if (dumper->cur_seq < log_first_seq) {
3192 /* messages are gone, move to first available one */
3193 dumper->cur_seq = log_first_seq;
3194 dumper->cur_idx = log_first_idx;
3198 if (dumper->cur_seq >= dumper->next_seq) {
3199 logbuf_unlock_irqrestore(flags);
3203 /* calculate length of entire buffer */
3204 seq = dumper->cur_seq;
3205 idx = dumper->cur_idx;
3206 while (seq < dumper->next_seq) {
3207 struct printk_log *msg = log_from_idx(idx);
3209 l += msg_print_text(msg, true, NULL, 0);
3210 idx = log_next(idx);
3214 /* move first record forward until length fits into the buffer */
3215 seq = dumper->cur_seq;
3216 idx = dumper->cur_idx;
3217 while (l > size && seq < dumper->next_seq) {
3218 struct printk_log *msg = log_from_idx(idx);
3220 l -= msg_print_text(msg, true, NULL, 0);
3221 idx = log_next(idx);
3225 /* last message in next interation */
3230 while (seq < dumper->next_seq) {
3231 struct printk_log *msg = log_from_idx(idx);
3233 l += msg_print_text(msg, syslog, buf + l, size - l);
3234 idx = log_next(idx);
3238 dumper->next_seq = next_seq;
3239 dumper->next_idx = next_idx;
3241 logbuf_unlock_irqrestore(flags);
3247 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3250 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3251 * @dumper: registered kmsg dumper
3253 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3254 * kmsg_dump_get_buffer() can be called again and used multiple
3255 * times within the same dumper.dump() callback.
3257 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3259 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3261 dumper->cur_seq = clear_seq;
3262 dumper->cur_idx = clear_idx;
3263 dumper->next_seq = log_next_seq;
3264 dumper->next_idx = log_next_idx;
3268 * kmsg_dump_rewind - reset the interator
3269 * @dumper: registered kmsg dumper
3271 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3272 * kmsg_dump_get_buffer() can be called again and used multiple
3273 * times within the same dumper.dump() callback.
3275 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3277 unsigned long flags;
3279 logbuf_lock_irqsave(flags);
3280 kmsg_dump_rewind_nolock(dumper);
3281 logbuf_unlock_irqrestore(flags);
3283 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);