2 * linux/kernel/printk.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * Modified to make sys_syslog() more flexible: added commands to
7 * return the last 4k of kernel messages, regardless of whether
8 * they've been read or not. Added option to suppress kernel printk's
9 * to the console. Added hook for sending the console messages
10 * elsewhere, in preparation for a serial line console (someday).
12 * Modified for sysctl support, 1/8/97, Chris Horn.
13 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
14 * manfred@colorfullife.com
15 * Rewrote bits to get rid of console_lock
16 * 01Mar01 Andrew Morton
19 #include <linux/kernel.h>
21 #include <linux/tty.h>
22 #include <linux/tty_driver.h>
23 #include <linux/console.h>
24 #include <linux/init.h>
25 #include <linux/jiffies.h>
26 #include <linux/nmi.h>
27 #include <linux/module.h>
28 #include <linux/moduleparam.h>
29 #include <linux/delay.h>
30 #include <linux/smp.h>
31 #include <linux/security.h>
32 #include <linux/bootmem.h>
33 #include <linux/memblock.h>
34 #include <linux/syscalls.h>
35 #include <linux/crash_core.h>
36 #include <linux/kdb.h>
37 #include <linux/ratelimit.h>
38 #include <linux/kmsg_dump.h>
39 #include <linux/syslog.h>
40 #include <linux/cpu.h>
41 #include <linux/rculist.h>
42 #include <linux/poll.h>
43 #include <linux/irq_work.h>
44 #include <linux/ctype.h>
45 #include <linux/uio.h>
46 #include <linux/sched/clock.h>
47 #include <linux/sched/debug.h>
48 #include <linux/sched/task_stack.h>
50 #include <linux/uaccess.h>
51 #include <asm/sections.h>
53 #include <trace/events/initcall.h>
54 #define CREATE_TRACE_POINTS
55 #include <trace/events/printk.h>
57 #include "console_cmdline.h"
61 int console_printk[4] = {
62 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
63 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
64 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
65 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
68 atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
69 EXPORT_SYMBOL(ignore_console_lock_warning);
72 * Low level drivers may need that to know if they can schedule in
73 * their unblank() callback or not. So let's export it.
76 EXPORT_SYMBOL(oops_in_progress);
79 * console_sem protects the console_drivers list, and also
80 * provides serialisation for access to the entire console
83 static DEFINE_SEMAPHORE(console_sem);
84 struct console *console_drivers;
85 EXPORT_SYMBOL_GPL(console_drivers);
88 static struct lockdep_map console_lock_dep_map = {
89 .name = "console_lock"
93 enum devkmsg_log_bits {
94 __DEVKMSG_LOG_BIT_ON = 0,
95 __DEVKMSG_LOG_BIT_OFF,
96 __DEVKMSG_LOG_BIT_LOCK,
99 enum devkmsg_log_masks {
100 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
101 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
102 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
105 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
106 #define DEVKMSG_LOG_MASK_DEFAULT 0
108 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
110 static int __control_devkmsg(char *str)
115 if (!strncmp(str, "on", 2)) {
116 devkmsg_log = DEVKMSG_LOG_MASK_ON;
118 } else if (!strncmp(str, "off", 3)) {
119 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
121 } else if (!strncmp(str, "ratelimit", 9)) {
122 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
128 static int __init control_devkmsg(char *str)
130 if (__control_devkmsg(str) < 0)
134 * Set sysctl string accordingly:
136 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
137 strcpy(devkmsg_log_str, "on");
138 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
139 strcpy(devkmsg_log_str, "off");
140 /* else "ratelimit" which is set by default. */
143 * Sysctl cannot change it anymore. The kernel command line setting of
144 * this parameter is to force the setting to be permanent throughout the
145 * runtime of the system. This is a precation measure against userspace
146 * trying to be a smarta** and attempting to change it up on us.
148 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
152 __setup("printk.devkmsg=", control_devkmsg);
154 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
156 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
157 void __user *buffer, size_t *lenp, loff_t *ppos)
159 char old_str[DEVKMSG_STR_MAX_SIZE];
164 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
168 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
171 err = proc_dostring(table, write, buffer, lenp, ppos);
176 err = __control_devkmsg(devkmsg_log_str);
179 * Do not accept an unknown string OR a known string with
182 if (err < 0 || (err + 1 != *lenp)) {
184 /* ... and restore old setting. */
186 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
196 * Number of registered extended console drivers.
198 * If extended consoles are present, in-kernel cont reassembly is disabled
199 * and each fragment is stored as a separate log entry with proper
200 * continuation flag so that every emitted message has full metadata. This
201 * doesn't change the result for regular consoles or /proc/kmsg. For
202 * /dev/kmsg, as long as the reader concatenates messages according to
203 * consecutive continuation flags, the end result should be the same too.
205 static int nr_ext_console_drivers;
208 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
209 * macros instead of functions so that _RET_IP_ contains useful information.
211 #define down_console_sem() do { \
213 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
216 static int __down_trylock_console_sem(unsigned long ip)
222 * Here and in __up_console_sem() we need to be in safe mode,
223 * because spindump/WARN/etc from under console ->lock will
224 * deadlock in printk()->down_trylock_console_sem() otherwise.
226 printk_safe_enter_irqsave(flags);
227 lock_failed = down_trylock(&console_sem);
228 printk_safe_exit_irqrestore(flags);
232 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
235 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
237 static void __up_console_sem(unsigned long ip)
241 mutex_release(&console_lock_dep_map, 1, ip);
243 printk_safe_enter_irqsave(flags);
245 printk_safe_exit_irqrestore(flags);
247 #define up_console_sem() __up_console_sem(_RET_IP_)
250 * This is used for debugging the mess that is the VT code by
251 * keeping track if we have the console semaphore held. It's
252 * definitely not the perfect debug tool (we don't know if _WE_
253 * hold it and are racing, but it helps tracking those weird code
254 * paths in the console code where we end up in places I want
255 * locked without the console sempahore held).
257 static int console_locked, console_suspended;
260 * If exclusive_console is non-NULL then only this console is to be printed to.
262 static struct console *exclusive_console;
265 * Array of consoles built from command line options (console=)
268 #define MAX_CMDLINECONSOLES 8
270 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
272 static int preferred_console = -1;
273 int console_set_on_cmdline;
274 EXPORT_SYMBOL(console_set_on_cmdline);
276 /* Flag: console code may call schedule() */
277 static int console_may_schedule;
279 enum con_msg_format_flags {
280 MSG_FORMAT_DEFAULT = 0,
281 MSG_FORMAT_SYSLOG = (1 << 0),
284 static int console_msg_format = MSG_FORMAT_DEFAULT;
287 * The printk log buffer consists of a chain of concatenated variable
288 * length records. Every record starts with a record header, containing
289 * the overall length of the record.
291 * The heads to the first and last entry in the buffer, as well as the
292 * sequence numbers of these entries are maintained when messages are
295 * If the heads indicate available messages, the length in the header
296 * tells the start next message. A length == 0 for the next message
297 * indicates a wrap-around to the beginning of the buffer.
299 * Every record carries the monotonic timestamp in microseconds, as well as
300 * the standard userspace syslog level and syslog facility. The usual
301 * kernel messages use LOG_KERN; userspace-injected messages always carry
302 * a matching syslog facility, by default LOG_USER. The origin of every
303 * message can be reliably determined that way.
305 * The human readable log message directly follows the message header. The
306 * length of the message text is stored in the header, the stored message
309 * Optionally, a message can carry a dictionary of properties (key/value pairs),
310 * to provide userspace with a machine-readable message context.
312 * Examples for well-defined, commonly used property names are:
313 * DEVICE=b12:8 device identifier
317 * +sound:card0 subsystem:devname
318 * SUBSYSTEM=pci driver-core subsystem name
320 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
321 * follows directly after a '=' character. Every property is terminated by
322 * a '\0' character. The last property is not terminated.
324 * Example of a message structure:
325 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
326 * 0008 34 00 record is 52 bytes long
327 * 000a 0b 00 text is 11 bytes long
328 * 000c 1f 00 dictionary is 23 bytes long
329 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
330 * 0010 69 74 27 73 20 61 20 6c "it's a l"
332 * 001b 44 45 56 49 43 "DEVIC"
333 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
334 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
336 * 0032 00 00 00 padding to next message header
338 * The 'struct printk_log' buffer header must never be directly exported to
339 * userspace, it is a kernel-private implementation detail that might
340 * need to be changed in the future, when the requirements change.
342 * /dev/kmsg exports the structured data in the following line format:
343 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
345 * Users of the export format should ignore possible additional values
346 * separated by ',', and find the message after the ';' character.
348 * The optional key/value pairs are attached as continuation lines starting
349 * with a space character and terminated by a newline. All possible
350 * non-prinatable characters are escaped in the "\xff" notation.
354 LOG_NEWLINE = 2, /* text ended with a newline */
355 LOG_PREFIX = 4, /* text started with a prefix */
356 LOG_CONT = 8, /* text is a fragment of a continuation line */
360 u64 ts_nsec; /* timestamp in nanoseconds */
361 u16 len; /* length of entire record */
362 u16 text_len; /* length of text buffer */
363 u16 dict_len; /* length of dictionary buffer */
364 u8 facility; /* syslog facility */
365 u8 flags:5; /* internal record flags */
366 u8 level:3; /* syslog level */
368 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
369 __packed __aligned(4)
374 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
375 * within the scheduler's rq lock. It must be released before calling
376 * console_unlock() or anything else that might wake up a process.
378 DEFINE_RAW_SPINLOCK(logbuf_lock);
381 * Helper macros to lock/unlock logbuf_lock and switch between
382 * printk-safe/unsafe modes.
384 #define logbuf_lock_irq() \
386 printk_safe_enter_irq(); \
387 raw_spin_lock(&logbuf_lock); \
390 #define logbuf_unlock_irq() \
392 raw_spin_unlock(&logbuf_lock); \
393 printk_safe_exit_irq(); \
396 #define logbuf_lock_irqsave(flags) \
398 printk_safe_enter_irqsave(flags); \
399 raw_spin_lock(&logbuf_lock); \
402 #define logbuf_unlock_irqrestore(flags) \
404 raw_spin_unlock(&logbuf_lock); \
405 printk_safe_exit_irqrestore(flags); \
409 DECLARE_WAIT_QUEUE_HEAD(log_wait);
410 /* the next printk record to read by syslog(READ) or /proc/kmsg */
411 static u64 syslog_seq;
412 static u32 syslog_idx;
413 static size_t syslog_partial;
415 /* index and sequence number of the first record stored in the buffer */
416 static u64 log_first_seq;
417 static u32 log_first_idx;
419 /* index and sequence number of the next record to store in the buffer */
420 static u64 log_next_seq;
421 static u32 log_next_idx;
423 /* the next printk record to write to the console */
424 static u64 console_seq;
425 static u32 console_idx;
426 static u64 exclusive_console_stop_seq;
428 /* the next printk record to read after the last 'clear' command */
429 static u64 clear_seq;
430 static u32 clear_idx;
432 #define PREFIX_MAX 32
433 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
435 #define LOG_LEVEL(v) ((v) & 0x07)
436 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
439 #define LOG_ALIGN __alignof__(struct printk_log)
440 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
441 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
442 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
443 static char *log_buf = __log_buf;
444 static u32 log_buf_len = __LOG_BUF_LEN;
447 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
448 * per_cpu_areas are initialised. This variable is set to true when
449 * it's safe to access per-CPU data.
451 static bool __printk_percpu_data_ready __read_mostly;
453 bool printk_percpu_data_ready(void)
455 return __printk_percpu_data_ready;
458 /* Return log buffer address */
459 char *log_buf_addr_get(void)
464 /* Return log buffer size */
465 u32 log_buf_len_get(void)
470 /* human readable text of the record */
471 static char *log_text(const struct printk_log *msg)
473 return (char *)msg + sizeof(struct printk_log);
476 /* optional key/value pair dictionary attached to the record */
477 static char *log_dict(const struct printk_log *msg)
479 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
482 /* get record by index; idx must point to valid msg */
483 static struct printk_log *log_from_idx(u32 idx)
485 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
488 * A length == 0 record is the end of buffer marker. Wrap around and
489 * read the message at the start of the buffer.
492 return (struct printk_log *)log_buf;
496 /* get next record; idx must point to valid msg */
497 static u32 log_next(u32 idx)
499 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
501 /* length == 0 indicates the end of the buffer; wrap */
503 * A length == 0 record is the end of buffer marker. Wrap around and
504 * read the message at the start of the buffer as *this* one, and
505 * return the one after that.
508 msg = (struct printk_log *)log_buf;
511 return idx + msg->len;
515 * Check whether there is enough free space for the given message.
517 * The same values of first_idx and next_idx mean that the buffer
518 * is either empty or full.
520 * If the buffer is empty, we must respect the position of the indexes.
521 * They cannot be reset to the beginning of the buffer.
523 static int logbuf_has_space(u32 msg_size, bool empty)
527 if (log_next_idx > log_first_idx || empty)
528 free = max(log_buf_len - log_next_idx, log_first_idx);
530 free = log_first_idx - log_next_idx;
533 * We need space also for an empty header that signalizes wrapping
536 return free >= msg_size + sizeof(struct printk_log);
539 static int log_make_free_space(u32 msg_size)
541 while (log_first_seq < log_next_seq &&
542 !logbuf_has_space(msg_size, false)) {
543 /* drop old messages until we have enough contiguous space */
544 log_first_idx = log_next(log_first_idx);
548 if (clear_seq < log_first_seq) {
549 clear_seq = log_first_seq;
550 clear_idx = log_first_idx;
553 /* sequence numbers are equal, so the log buffer is empty */
554 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
560 /* compute the message size including the padding bytes */
561 static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
565 size = sizeof(struct printk_log) + text_len + dict_len;
566 *pad_len = (-size) & (LOG_ALIGN - 1);
573 * Define how much of the log buffer we could take at maximum. The value
574 * must be greater than two. Note that only half of the buffer is available
575 * when the index points to the middle.
577 #define MAX_LOG_TAKE_PART 4
578 static const char trunc_msg[] = "<truncated>";
580 static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
581 u16 *dict_len, u32 *pad_len)
584 * The message should not take the whole buffer. Otherwise, it might
585 * get removed too soon.
587 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
588 if (*text_len > max_text_len)
589 *text_len = max_text_len;
590 /* enable the warning message */
591 *trunc_msg_len = strlen(trunc_msg);
592 /* disable the "dict" completely */
594 /* compute the size again, count also the warning message */
595 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
598 /* insert record into the buffer, discard old ones, update heads */
599 static int log_store(int facility, int level,
600 enum log_flags flags, u64 ts_nsec,
601 const char *dict, u16 dict_len,
602 const char *text, u16 text_len)
604 struct printk_log *msg;
606 u16 trunc_msg_len = 0;
608 /* number of '\0' padding bytes to next message */
609 size = msg_used_size(text_len, dict_len, &pad_len);
611 if (log_make_free_space(size)) {
612 /* truncate the message if it is too long for empty buffer */
613 size = truncate_msg(&text_len, &trunc_msg_len,
614 &dict_len, &pad_len);
615 /* survive when the log buffer is too small for trunc_msg */
616 if (log_make_free_space(size))
620 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
622 * This message + an additional empty header does not fit
623 * at the end of the buffer. Add an empty header with len == 0
624 * to signify a wrap around.
626 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
631 msg = (struct printk_log *)(log_buf + log_next_idx);
632 memcpy(log_text(msg), text, text_len);
633 msg->text_len = text_len;
635 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
636 msg->text_len += trunc_msg_len;
638 memcpy(log_dict(msg), dict, dict_len);
639 msg->dict_len = dict_len;
640 msg->facility = facility;
641 msg->level = level & 7;
642 msg->flags = flags & 0x1f;
644 msg->ts_nsec = ts_nsec;
646 msg->ts_nsec = local_clock();
647 memset(log_dict(msg) + dict_len, 0, pad_len);
651 log_next_idx += msg->len;
654 return msg->text_len;
657 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
659 static int syslog_action_restricted(int type)
664 * Unless restricted, we allow "read all" and "get buffer size"
667 return type != SYSLOG_ACTION_READ_ALL &&
668 type != SYSLOG_ACTION_SIZE_BUFFER;
671 static int check_syslog_permissions(int type, int source)
674 * If this is from /proc/kmsg and we've already opened it, then we've
675 * already done the capabilities checks at open time.
677 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
680 if (syslog_action_restricted(type)) {
681 if (capable(CAP_SYSLOG))
684 * For historical reasons, accept CAP_SYS_ADMIN too, with
687 if (capable(CAP_SYS_ADMIN)) {
688 pr_warn_once("%s (%d): Attempt to access syslog with "
689 "CAP_SYS_ADMIN but no CAP_SYSLOG "
691 current->comm, task_pid_nr(current));
697 return security_syslog(type);
700 static void append_char(char **pp, char *e, char c)
706 static ssize_t msg_print_ext_header(char *buf, size_t size,
707 struct printk_log *msg, u64 seq)
709 u64 ts_usec = msg->ts_nsec;
711 do_div(ts_usec, 1000);
713 return scnprintf(buf, size, "%u,%llu,%llu,%c;",
714 (msg->facility << 3) | msg->level, seq, ts_usec,
715 msg->flags & LOG_CONT ? 'c' : '-');
718 static ssize_t msg_print_ext_body(char *buf, size_t size,
719 char *dict, size_t dict_len,
720 char *text, size_t text_len)
722 char *p = buf, *e = buf + size;
725 /* escape non-printable characters */
726 for (i = 0; i < text_len; i++) {
727 unsigned char c = text[i];
729 if (c < ' ' || c >= 127 || c == '\\')
730 p += scnprintf(p, e - p, "\\x%02x", c);
732 append_char(&p, e, c);
734 append_char(&p, e, '\n');
739 for (i = 0; i < dict_len; i++) {
740 unsigned char c = dict[i];
743 append_char(&p, e, ' ');
748 append_char(&p, e, '\n');
753 if (c < ' ' || c >= 127 || c == '\\') {
754 p += scnprintf(p, e - p, "\\x%02x", c);
758 append_char(&p, e, c);
760 append_char(&p, e, '\n');
766 /* /dev/kmsg - userspace message inject/listen interface */
767 struct devkmsg_user {
770 struct ratelimit_state rs;
772 char buf[CONSOLE_EXT_LOG_MAX];
775 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
778 int level = default_message_loglevel;
779 int facility = 1; /* LOG_USER */
780 struct file *file = iocb->ki_filp;
781 struct devkmsg_user *user = file->private_data;
782 size_t len = iov_iter_count(from);
785 if (!user || len > LOG_LINE_MAX)
788 /* Ignore when user logging is disabled. */
789 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
792 /* Ratelimit when not explicitly enabled. */
793 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
794 if (!___ratelimit(&user->rs, current->comm))
798 buf = kmalloc(len+1, GFP_KERNEL);
803 if (!copy_from_iter_full(buf, len, from)) {
809 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
810 * the decimal value represents 32bit, the lower 3 bit are the log
811 * level, the rest are the log facility.
813 * If no prefix or no userspace facility is specified, we
814 * enforce LOG_USER, to be able to reliably distinguish
815 * kernel-generated messages from userspace-injected ones.
818 if (line[0] == '<') {
822 u = simple_strtoul(line + 1, &endp, 10);
823 if (endp && endp[0] == '>') {
824 level = LOG_LEVEL(u);
825 if (LOG_FACILITY(u) != 0)
826 facility = LOG_FACILITY(u);
833 printk_emit(facility, level, NULL, 0, "%s", line);
838 static ssize_t devkmsg_read(struct file *file, char __user *buf,
839 size_t count, loff_t *ppos)
841 struct devkmsg_user *user = file->private_data;
842 struct printk_log *msg;
849 ret = mutex_lock_interruptible(&user->lock);
854 while (user->seq == log_next_seq) {
855 if (file->f_flags & O_NONBLOCK) {
862 ret = wait_event_interruptible(log_wait,
863 user->seq != log_next_seq);
869 if (user->seq < log_first_seq) {
870 /* our last seen message is gone, return error and reset */
871 user->idx = log_first_idx;
872 user->seq = log_first_seq;
878 msg = log_from_idx(user->idx);
879 len = msg_print_ext_header(user->buf, sizeof(user->buf),
881 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
882 log_dict(msg), msg->dict_len,
883 log_text(msg), msg->text_len);
885 user->idx = log_next(user->idx);
894 if (copy_to_user(buf, user->buf, len)) {
900 mutex_unlock(&user->lock);
904 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
906 struct devkmsg_user *user = file->private_data;
917 /* the first record */
918 user->idx = log_first_idx;
919 user->seq = log_first_seq;
923 * The first record after the last SYSLOG_ACTION_CLEAR,
924 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
925 * changes no global state, and does not clear anything.
927 user->idx = clear_idx;
928 user->seq = clear_seq;
931 /* after the last record */
932 user->idx = log_next_idx;
933 user->seq = log_next_seq;
942 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
944 struct devkmsg_user *user = file->private_data;
948 return EPOLLERR|EPOLLNVAL;
950 poll_wait(file, &log_wait, wait);
953 if (user->seq < log_next_seq) {
954 /* return error when data has vanished underneath us */
955 if (user->seq < log_first_seq)
956 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
958 ret = EPOLLIN|EPOLLRDNORM;
965 static int devkmsg_open(struct inode *inode, struct file *file)
967 struct devkmsg_user *user;
970 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
973 /* write-only does not need any file context */
974 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
975 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
981 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
985 ratelimit_default_init(&user->rs);
986 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
988 mutex_init(&user->lock);
991 user->idx = log_first_idx;
992 user->seq = log_first_seq;
995 file->private_data = user;
999 static int devkmsg_release(struct inode *inode, struct file *file)
1001 struct devkmsg_user *user = file->private_data;
1006 ratelimit_state_exit(&user->rs);
1008 mutex_destroy(&user->lock);
1013 const struct file_operations kmsg_fops = {
1014 .open = devkmsg_open,
1015 .read = devkmsg_read,
1016 .write_iter = devkmsg_write,
1017 .llseek = devkmsg_llseek,
1018 .poll = devkmsg_poll,
1019 .release = devkmsg_release,
1022 #ifdef CONFIG_CRASH_CORE
1024 * This appends the listed symbols to /proc/vmcore
1026 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1027 * obtain access to symbols that are otherwise very difficult to locate. These
1028 * symbols are specifically used so that utilities can access and extract the
1029 * dmesg log from a vmcore file after a crash.
1031 void log_buf_vmcoreinfo_setup(void)
1033 VMCOREINFO_SYMBOL(log_buf);
1034 VMCOREINFO_SYMBOL(log_buf_len);
1035 VMCOREINFO_SYMBOL(log_first_idx);
1036 VMCOREINFO_SYMBOL(clear_idx);
1037 VMCOREINFO_SYMBOL(log_next_idx);
1039 * Export struct printk_log size and field offsets. User space tools can
1040 * parse it and detect any changes to structure down the line.
1042 VMCOREINFO_STRUCT_SIZE(printk_log);
1043 VMCOREINFO_OFFSET(printk_log, ts_nsec);
1044 VMCOREINFO_OFFSET(printk_log, len);
1045 VMCOREINFO_OFFSET(printk_log, text_len);
1046 VMCOREINFO_OFFSET(printk_log, dict_len);
1050 /* requested log_buf_len from kernel cmdline */
1051 static unsigned long __initdata new_log_buf_len;
1053 /* we practice scaling the ring buffer by powers of 2 */
1054 static void __init log_buf_len_update(u64 size)
1056 if (size > (u64)LOG_BUF_LEN_MAX) {
1057 size = (u64)LOG_BUF_LEN_MAX;
1058 pr_err("log_buf over 2G is not supported.\n");
1062 size = roundup_pow_of_two(size);
1063 if (size > log_buf_len)
1064 new_log_buf_len = (unsigned long)size;
1067 /* save requested log_buf_len since it's too early to process it */
1068 static int __init log_buf_len_setup(char *str)
1075 size = memparse(str, &str);
1077 log_buf_len_update(size);
1081 early_param("log_buf_len", log_buf_len_setup);
1084 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1086 static void __init log_buf_add_cpu(void)
1088 unsigned int cpu_extra;
1091 * archs should set up cpu_possible_bits properly with
1092 * set_cpu_possible() after setup_arch() but just in
1093 * case lets ensure this is valid.
1095 if (num_possible_cpus() == 1)
1098 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1100 /* by default this will only continue through for large > 64 CPUs */
1101 if (cpu_extra <= __LOG_BUF_LEN / 2)
1104 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1105 __LOG_CPU_MAX_BUF_LEN);
1106 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1108 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1110 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1112 #else /* !CONFIG_SMP */
1113 static inline void log_buf_add_cpu(void) {}
1114 #endif /* CONFIG_SMP */
1116 static void __init set_percpu_data_ready(void)
1119 /* Make sure we set this flag only after printk_safe() init is done */
1121 __printk_percpu_data_ready = true;
1124 void __init setup_log_buf(int early)
1126 unsigned long flags;
1131 * Some archs call setup_log_buf() multiple times - first is very
1132 * early, e.g. from setup_arch(), and second - when percpu_areas
1136 set_percpu_data_ready();
1138 if (log_buf != __log_buf)
1141 if (!early && !new_log_buf_len)
1144 if (!new_log_buf_len)
1149 memblock_virt_alloc(new_log_buf_len, LOG_ALIGN);
1151 new_log_buf = memblock_virt_alloc_nopanic(new_log_buf_len,
1155 if (unlikely(!new_log_buf)) {
1156 pr_err("log_buf_len: %lu bytes not available\n",
1161 logbuf_lock_irqsave(flags);
1162 log_buf_len = new_log_buf_len;
1163 log_buf = new_log_buf;
1164 new_log_buf_len = 0;
1165 free = __LOG_BUF_LEN - log_next_idx;
1166 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1167 logbuf_unlock_irqrestore(flags);
1169 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1170 pr_info("early log buf free: %u(%u%%)\n",
1171 free, (free * 100) / __LOG_BUF_LEN);
1174 static bool __read_mostly ignore_loglevel;
1176 static int __init ignore_loglevel_setup(char *str)
1178 ignore_loglevel = true;
1179 pr_info("debug: ignoring loglevel setting.\n");
1184 early_param("ignore_loglevel", ignore_loglevel_setup);
1185 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1186 MODULE_PARM_DESC(ignore_loglevel,
1187 "ignore loglevel setting (prints all kernel messages to the console)");
1189 static bool suppress_message_printing(int level)
1191 return (level >= console_loglevel && !ignore_loglevel);
1194 #ifdef CONFIG_BOOT_PRINTK_DELAY
1196 static int boot_delay; /* msecs delay after each printk during bootup */
1197 static unsigned long long loops_per_msec; /* based on boot_delay */
1199 static int __init boot_delay_setup(char *str)
1203 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1204 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1206 get_option(&str, &boot_delay);
1207 if (boot_delay > 10 * 1000)
1210 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1211 "HZ: %d, loops_per_msec: %llu\n",
1212 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1215 early_param("boot_delay", boot_delay_setup);
1217 static void boot_delay_msec(int level)
1219 unsigned long long k;
1220 unsigned long timeout;
1222 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1223 || suppress_message_printing(level)) {
1227 k = (unsigned long long)loops_per_msec * boot_delay;
1229 timeout = jiffies + msecs_to_jiffies(boot_delay);
1234 * use (volatile) jiffies to prevent
1235 * compiler reduction; loop termination via jiffies
1236 * is secondary and may or may not happen.
1238 if (time_after(jiffies, timeout))
1240 touch_nmi_watchdog();
1244 static inline void boot_delay_msec(int level)
1249 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1250 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1252 static size_t print_time(u64 ts, char *buf)
1254 unsigned long rem_nsec;
1259 rem_nsec = do_div(ts, 1000000000);
1262 return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
1264 return sprintf(buf, "[%5lu.%06lu] ",
1265 (unsigned long)ts, rem_nsec / 1000);
1268 static size_t print_prefix(const struct printk_log *msg, bool syslog, char *buf)
1271 unsigned int prefix = (msg->facility << 3) | msg->level;
1275 len += sprintf(buf, "<%u>", prefix);
1280 else if (prefix > 99)
1282 else if (prefix > 9)
1287 len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
1291 static size_t msg_print_text(const struct printk_log *msg, bool syslog, char *buf, size_t size)
1293 const char *text = log_text(msg);
1294 size_t text_size = msg->text_len;
1298 const char *next = memchr(text, '\n', text_size);
1302 text_len = next - text;
1304 text_size -= next - text;
1306 text_len = text_size;
1310 if (print_prefix(msg, syslog, NULL) +
1311 text_len + 1 >= size - len)
1314 len += print_prefix(msg, syslog, buf + len);
1315 memcpy(buf + len, text, text_len);
1319 /* SYSLOG_ACTION_* buffer size only calculation */
1320 len += print_prefix(msg, syslog, NULL);
1331 static int syslog_print(char __user *buf, int size)
1334 struct printk_log *msg;
1337 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1346 if (syslog_seq < log_first_seq) {
1347 /* messages are gone, move to first one */
1348 syslog_seq = log_first_seq;
1349 syslog_idx = log_first_idx;
1352 if (syslog_seq == log_next_seq) {
1353 logbuf_unlock_irq();
1357 skip = syslog_partial;
1358 msg = log_from_idx(syslog_idx);
1359 n = msg_print_text(msg, true, text, LOG_LINE_MAX + PREFIX_MAX);
1360 if (n - syslog_partial <= size) {
1361 /* message fits into buffer, move forward */
1362 syslog_idx = log_next(syslog_idx);
1364 n -= syslog_partial;
1367 /* partial read(), remember position */
1369 syslog_partial += n;
1372 logbuf_unlock_irq();
1377 if (copy_to_user(buf, text + skip, n)) {
1392 static int syslog_print_all(char __user *buf, int size, bool clear)
1400 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1406 * Find first record that fits, including all following records,
1407 * into the user-provided buffer for this dump.
1411 while (seq < log_next_seq) {
1412 struct printk_log *msg = log_from_idx(idx);
1414 len += msg_print_text(msg, true, NULL, 0);
1415 idx = log_next(idx);
1419 /* move first record forward until length fits into the buffer */
1422 while (len > size && seq < log_next_seq) {
1423 struct printk_log *msg = log_from_idx(idx);
1425 len -= msg_print_text(msg, true, NULL, 0);
1426 idx = log_next(idx);
1430 /* last message fitting into this dump */
1431 next_seq = log_next_seq;
1434 while (len >= 0 && seq < next_seq) {
1435 struct printk_log *msg = log_from_idx(idx);
1438 textlen = msg_print_text(msg, true, text,
1439 LOG_LINE_MAX + PREFIX_MAX);
1444 idx = log_next(idx);
1447 logbuf_unlock_irq();
1448 if (copy_to_user(buf + len, text, textlen))
1454 if (seq < log_first_seq) {
1455 /* messages are gone, move to next one */
1456 seq = log_first_seq;
1457 idx = log_first_idx;
1462 clear_seq = log_next_seq;
1463 clear_idx = log_next_idx;
1465 logbuf_unlock_irq();
1471 static void syslog_clear(void)
1474 clear_seq = log_next_seq;
1475 clear_idx = log_next_idx;
1476 logbuf_unlock_irq();
1479 int do_syslog(int type, char __user *buf, int len, int source)
1482 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1485 error = check_syslog_permissions(type, source);
1490 case SYSLOG_ACTION_CLOSE: /* Close log */
1492 case SYSLOG_ACTION_OPEN: /* Open log */
1494 case SYSLOG_ACTION_READ: /* Read from log */
1495 if (!buf || len < 0)
1499 if (!access_ok(VERIFY_WRITE, buf, len))
1501 error = wait_event_interruptible(log_wait,
1502 syslog_seq != log_next_seq);
1505 error = syslog_print(buf, len);
1507 /* Read/clear last kernel messages */
1508 case SYSLOG_ACTION_READ_CLEAR:
1511 /* Read last kernel messages */
1512 case SYSLOG_ACTION_READ_ALL:
1513 if (!buf || len < 0)
1517 if (!access_ok(VERIFY_WRITE, buf, len))
1519 error = syslog_print_all(buf, len, clear);
1521 /* Clear ring buffer */
1522 case SYSLOG_ACTION_CLEAR:
1525 /* Disable logging to console */
1526 case SYSLOG_ACTION_CONSOLE_OFF:
1527 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1528 saved_console_loglevel = console_loglevel;
1529 console_loglevel = minimum_console_loglevel;
1531 /* Enable logging to console */
1532 case SYSLOG_ACTION_CONSOLE_ON:
1533 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1534 console_loglevel = saved_console_loglevel;
1535 saved_console_loglevel = LOGLEVEL_DEFAULT;
1538 /* Set level of messages printed to console */
1539 case SYSLOG_ACTION_CONSOLE_LEVEL:
1540 if (len < 1 || len > 8)
1542 if (len < minimum_console_loglevel)
1543 len = minimum_console_loglevel;
1544 console_loglevel = len;
1545 /* Implicitly re-enable logging to console */
1546 saved_console_loglevel = LOGLEVEL_DEFAULT;
1548 /* Number of chars in the log buffer */
1549 case SYSLOG_ACTION_SIZE_UNREAD:
1551 if (syslog_seq < log_first_seq) {
1552 /* messages are gone, move to first one */
1553 syslog_seq = log_first_seq;
1554 syslog_idx = log_first_idx;
1557 if (source == SYSLOG_FROM_PROC) {
1559 * Short-cut for poll(/"proc/kmsg") which simply checks
1560 * for pending data, not the size; return the count of
1561 * records, not the length.
1563 error = log_next_seq - syslog_seq;
1565 u64 seq = syslog_seq;
1566 u32 idx = syslog_idx;
1568 while (seq < log_next_seq) {
1569 struct printk_log *msg = log_from_idx(idx);
1571 error += msg_print_text(msg, true, NULL, 0);
1572 idx = log_next(idx);
1575 error -= syslog_partial;
1577 logbuf_unlock_irq();
1579 /* Size of the log buffer */
1580 case SYSLOG_ACTION_SIZE_BUFFER:
1581 error = log_buf_len;
1591 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1593 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1597 * Special console_lock variants that help to reduce the risk of soft-lockups.
1598 * They allow to pass console_lock to another printk() call using a busy wait.
1601 #ifdef CONFIG_LOCKDEP
1602 static struct lockdep_map console_owner_dep_map = {
1603 .name = "console_owner"
1607 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1608 static struct task_struct *console_owner;
1609 static bool console_waiter;
1612 * console_lock_spinning_enable - mark beginning of code where another
1613 * thread might safely busy wait
1615 * This basically converts console_lock into a spinlock. This marks
1616 * the section where the console_lock owner can not sleep, because
1617 * there may be a waiter spinning (like a spinlock). Also it must be
1618 * ready to hand over the lock at the end of the section.
1620 static void console_lock_spinning_enable(void)
1622 raw_spin_lock(&console_owner_lock);
1623 console_owner = current;
1624 raw_spin_unlock(&console_owner_lock);
1626 /* The waiter may spin on us after setting console_owner */
1627 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1631 * console_lock_spinning_disable_and_check - mark end of code where another
1632 * thread was able to busy wait and check if there is a waiter
1634 * This is called at the end of the section where spinning is allowed.
1635 * It has two functions. First, it is a signal that it is no longer
1636 * safe to start busy waiting for the lock. Second, it checks if
1637 * there is a busy waiter and passes the lock rights to her.
1639 * Important: Callers lose the lock if there was a busy waiter.
1640 * They must not touch items synchronized by console_lock
1643 * Return: 1 if the lock rights were passed, 0 otherwise.
1645 static int console_lock_spinning_disable_and_check(void)
1649 raw_spin_lock(&console_owner_lock);
1650 waiter = READ_ONCE(console_waiter);
1651 console_owner = NULL;
1652 raw_spin_unlock(&console_owner_lock);
1655 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1659 /* The waiter is now free to continue */
1660 WRITE_ONCE(console_waiter, false);
1662 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1665 * Hand off console_lock to waiter. The waiter will perform
1666 * the up(). After this, the waiter is the console_lock owner.
1668 mutex_release(&console_lock_dep_map, 1, _THIS_IP_);
1673 * console_trylock_spinning - try to get console_lock by busy waiting
1675 * This allows to busy wait for the console_lock when the current
1676 * owner is running in specially marked sections. It means that
1677 * the current owner is running and cannot reschedule until it
1678 * is ready to lose the lock.
1680 * Return: 1 if we got the lock, 0 othrewise
1682 static int console_trylock_spinning(void)
1684 struct task_struct *owner = NULL;
1687 unsigned long flags;
1689 if (console_trylock())
1692 printk_safe_enter_irqsave(flags);
1694 raw_spin_lock(&console_owner_lock);
1695 owner = READ_ONCE(console_owner);
1696 waiter = READ_ONCE(console_waiter);
1697 if (!waiter && owner && owner != current) {
1698 WRITE_ONCE(console_waiter, true);
1701 raw_spin_unlock(&console_owner_lock);
1704 * If there is an active printk() writing to the
1705 * consoles, instead of having it write our data too,
1706 * see if we can offload that load from the active
1707 * printer, and do some printing ourselves.
1708 * Go into a spin only if there isn't already a waiter
1709 * spinning, and there is an active printer, and
1710 * that active printer isn't us (recursive printk?).
1713 printk_safe_exit_irqrestore(flags);
1717 /* We spin waiting for the owner to release us */
1718 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1719 /* Owner will clear console_waiter on hand off */
1720 while (READ_ONCE(console_waiter))
1722 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1724 printk_safe_exit_irqrestore(flags);
1726 * The owner passed the console lock to us.
1727 * Since we did not spin on console lock, annotate
1728 * this as a trylock. Otherwise lockdep will
1731 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1737 * Call the console drivers, asking them to write out
1738 * log_buf[start] to log_buf[end - 1].
1739 * The console_lock must be held.
1741 static void call_console_drivers(const char *ext_text, size_t ext_len,
1742 const char *text, size_t len)
1744 struct console *con;
1746 trace_console_rcuidle(text, len);
1748 if (!console_drivers)
1751 for_each_console(con) {
1752 if (exclusive_console && con != exclusive_console)
1754 if (!(con->flags & CON_ENABLED))
1758 if (!cpu_online(smp_processor_id()) &&
1759 !(con->flags & CON_ANYTIME))
1761 if (con->flags & CON_EXTENDED)
1762 con->write(con, ext_text, ext_len);
1764 con->write(con, text, len);
1768 int printk_delay_msec __read_mostly;
1770 static inline void printk_delay(void)
1772 if (unlikely(printk_delay_msec)) {
1773 int m = printk_delay_msec;
1777 touch_nmi_watchdog();
1783 * Continuation lines are buffered, and not committed to the record buffer
1784 * until the line is complete, or a race forces it. The line fragments
1785 * though, are printed immediately to the consoles to ensure everything has
1786 * reached the console in case of a kernel crash.
1788 static struct cont {
1789 char buf[LOG_LINE_MAX];
1790 size_t len; /* length == 0 means unused buffer */
1791 struct task_struct *owner; /* task of first print*/
1792 u64 ts_nsec; /* time of first print */
1793 u8 level; /* log level of first message */
1794 u8 facility; /* log facility of first message */
1795 enum log_flags flags; /* prefix, newline flags */
1798 static void cont_flush(void)
1803 log_store(cont.facility, cont.level, cont.flags, cont.ts_nsec,
1804 NULL, 0, cont.buf, cont.len);
1808 static bool cont_add(int facility, int level, enum log_flags flags, const char *text, size_t len)
1811 * If ext consoles are present, flush and skip in-kernel
1812 * continuation. See nr_ext_console_drivers definition. Also, if
1813 * the line gets too long, split it up in separate records.
1815 if (nr_ext_console_drivers || cont.len + len > sizeof(cont.buf)) {
1821 cont.facility = facility;
1823 cont.owner = current;
1824 cont.ts_nsec = local_clock();
1828 memcpy(cont.buf + cont.len, text, len);
1831 // The original flags come from the first line,
1832 // but later continuations can add a newline.
1833 if (flags & LOG_NEWLINE) {
1834 cont.flags |= LOG_NEWLINE;
1838 if (cont.len > (sizeof(cont.buf) * 80) / 100)
1844 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)
1847 * If an earlier line was buffered, and we're a continuation
1848 * write from the same process, try to add it to the buffer.
1851 if (cont.owner == current && (lflags & LOG_CONT)) {
1852 if (cont_add(facility, level, lflags, text, text_len))
1855 /* Otherwise, make sure it's flushed */
1859 /* Skip empty continuation lines that couldn't be added - they just flush */
1860 if (!text_len && (lflags & LOG_CONT))
1863 /* If it doesn't end in a newline, try to buffer the current line */
1864 if (!(lflags & LOG_NEWLINE)) {
1865 if (cont_add(facility, level, lflags, text, text_len))
1869 /* Store it in the record log */
1870 return log_store(facility, level, lflags, 0, dict, dictlen, text, text_len);
1873 /* Must be called under logbuf_lock. */
1874 int vprintk_store(int facility, int level,
1875 const char *dict, size_t dictlen,
1876 const char *fmt, va_list args)
1878 static char textbuf[LOG_LINE_MAX];
1879 char *text = textbuf;
1881 enum log_flags lflags = 0;
1884 * The printf needs to come first; we need the syslog
1885 * prefix which might be passed-in as a parameter.
1887 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1889 /* mark and strip a trailing newline */
1890 if (text_len && text[text_len-1] == '\n') {
1892 lflags |= LOG_NEWLINE;
1895 /* strip kernel syslog prefix and extract log level or control flags */
1896 if (facility == 0) {
1899 while ((kern_level = printk_get_level(text)) != 0) {
1900 switch (kern_level) {
1902 if (level == LOGLEVEL_DEFAULT)
1903 level = kern_level - '0';
1905 case 'd': /* KERN_DEFAULT */
1906 lflags |= LOG_PREFIX;
1908 case 'c': /* KERN_CONT */
1917 if (level == LOGLEVEL_DEFAULT)
1918 level = default_message_loglevel;
1921 lflags |= LOG_PREFIX|LOG_NEWLINE;
1923 return log_output(facility, level, lflags,
1924 dict, dictlen, text, text_len);
1927 asmlinkage int vprintk_emit(int facility, int level,
1928 const char *dict, size_t dictlen,
1929 const char *fmt, va_list args)
1932 bool in_sched = false, pending_output;
1933 unsigned long flags;
1936 if (level == LOGLEVEL_SCHED) {
1937 level = LOGLEVEL_DEFAULT;
1941 boot_delay_msec(level);
1944 /* This stops the holder of console_sem just where we want him */
1945 logbuf_lock_irqsave(flags);
1946 curr_log_seq = log_next_seq;
1947 printed_len = vprintk_store(facility, level, dict, dictlen, fmt, args);
1948 pending_output = (curr_log_seq != log_next_seq);
1949 logbuf_unlock_irqrestore(flags);
1951 /* If called from the scheduler, we can not call up(). */
1952 if (!in_sched && pending_output) {
1954 * Disable preemption to avoid being preempted while holding
1955 * console_sem which would prevent anyone from printing to
1960 * Try to acquire and then immediately release the console
1961 * semaphore. The release will print out buffers and wake up
1962 * /dev/kmsg and syslog() users.
1964 if (console_trylock_spinning())
1973 EXPORT_SYMBOL(vprintk_emit);
1975 asmlinkage int vprintk(const char *fmt, va_list args)
1977 return vprintk_func(fmt, args);
1979 EXPORT_SYMBOL(vprintk);
1981 asmlinkage int printk_emit(int facility, int level,
1982 const char *dict, size_t dictlen,
1983 const char *fmt, ...)
1988 va_start(args, fmt);
1989 r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1994 EXPORT_SYMBOL(printk_emit);
1996 int vprintk_default(const char *fmt, va_list args)
2000 #ifdef CONFIG_KGDB_KDB
2001 /* Allow to pass printk() to kdb but avoid a recursion. */
2002 if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
2003 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
2007 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
2011 EXPORT_SYMBOL_GPL(vprintk_default);
2014 * printk - print a kernel message
2015 * @fmt: format string
2017 * This is printk(). It can be called from any context. We want it to work.
2019 * We try to grab the console_lock. If we succeed, it's easy - we log the
2020 * output and call the console drivers. If we fail to get the semaphore, we
2021 * place the output into the log buffer and return. The current holder of
2022 * the console_sem will notice the new output in console_unlock(); and will
2023 * send it to the consoles before releasing the lock.
2025 * One effect of this deferred printing is that code which calls printk() and
2026 * then changes console_loglevel may break. This is because console_loglevel
2027 * is inspected when the actual printing occurs.
2032 * See the vsnprintf() documentation for format string extensions over C99.
2034 asmlinkage __visible int printk(const char *fmt, ...)
2039 va_start(args, fmt);
2040 r = vprintk_func(fmt, args);
2045 EXPORT_SYMBOL(printk);
2047 #else /* CONFIG_PRINTK */
2049 #define LOG_LINE_MAX 0
2050 #define PREFIX_MAX 0
2052 static u64 syslog_seq;
2053 static u32 syslog_idx;
2054 static u64 console_seq;
2055 static u32 console_idx;
2056 static u64 exclusive_console_stop_seq;
2057 static u64 log_first_seq;
2058 static u32 log_first_idx;
2059 static u64 log_next_seq;
2060 static char *log_text(const struct printk_log *msg) { return NULL; }
2061 static char *log_dict(const struct printk_log *msg) { return NULL; }
2062 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2063 static u32 log_next(u32 idx) { return 0; }
2064 static ssize_t msg_print_ext_header(char *buf, size_t size,
2065 struct printk_log *msg,
2066 u64 seq) { return 0; }
2067 static ssize_t msg_print_ext_body(char *buf, size_t size,
2068 char *dict, size_t dict_len,
2069 char *text, size_t text_len) { return 0; }
2070 static void console_lock_spinning_enable(void) { }
2071 static int console_lock_spinning_disable_and_check(void) { return 0; }
2072 static void call_console_drivers(const char *ext_text, size_t ext_len,
2073 const char *text, size_t len) {}
2074 static size_t msg_print_text(const struct printk_log *msg,
2075 bool syslog, char *buf, size_t size) { return 0; }
2076 static bool suppress_message_printing(int level) { return false; }
2078 #endif /* CONFIG_PRINTK */
2080 #ifdef CONFIG_EARLY_PRINTK
2081 struct console *early_console;
2083 asmlinkage __visible void early_printk(const char *fmt, ...)
2093 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2096 early_console->write(early_console, buf, n);
2100 static int __add_preferred_console(char *name, int idx, char *options,
2103 struct console_cmdline *c;
2107 * See if this tty is not yet registered, and
2108 * if we have a slot free.
2110 for (i = 0, c = console_cmdline;
2111 i < MAX_CMDLINECONSOLES && c->name[0];
2113 if (strcmp(c->name, name) == 0 && c->index == idx) {
2115 preferred_console = i;
2119 if (i == MAX_CMDLINECONSOLES)
2122 preferred_console = i;
2123 strlcpy(c->name, name, sizeof(c->name));
2124 c->options = options;
2125 braille_set_options(c, brl_options);
2131 static int __init console_msg_format_setup(char *str)
2133 if (!strcmp(str, "syslog"))
2134 console_msg_format = MSG_FORMAT_SYSLOG;
2135 if (!strcmp(str, "default"))
2136 console_msg_format = MSG_FORMAT_DEFAULT;
2139 __setup("console_msg_format=", console_msg_format_setup);
2142 * Set up a console. Called via do_early_param() in init/main.c
2143 * for each "console=" parameter in the boot command line.
2145 static int __init console_setup(char *str)
2147 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2148 char *s, *options, *brl_options = NULL;
2152 * console="" or console=null have been suggested as a way to
2153 * disable console output. Use ttynull that has been created
2154 * for exacly this purpose.
2156 if (str[0] == 0 || strcmp(str, "null") == 0) {
2157 __add_preferred_console("ttynull", 0, NULL, NULL);
2161 if (_braille_console_setup(&str, &brl_options))
2165 * Decode str into name, index, options.
2167 if (str[0] >= '0' && str[0] <= '9') {
2168 strcpy(buf, "ttyS");
2169 strncpy(buf + 4, str, sizeof(buf) - 5);
2171 strncpy(buf, str, sizeof(buf) - 1);
2173 buf[sizeof(buf) - 1] = 0;
2174 options = strchr(str, ',');
2178 if (!strcmp(str, "ttya"))
2179 strcpy(buf, "ttyS0");
2180 if (!strcmp(str, "ttyb"))
2181 strcpy(buf, "ttyS1");
2183 for (s = buf; *s; s++)
2184 if (isdigit(*s) || *s == ',')
2186 idx = simple_strtoul(s, NULL, 10);
2189 __add_preferred_console(buf, idx, options, brl_options);
2190 console_set_on_cmdline = 1;
2193 __setup("console=", console_setup);
2196 * add_preferred_console - add a device to the list of preferred consoles.
2197 * @name: device name
2198 * @idx: device index
2199 * @options: options for this console
2201 * The last preferred console added will be used for kernel messages
2202 * and stdin/out/err for init. Normally this is used by console_setup
2203 * above to handle user-supplied console arguments; however it can also
2204 * be used by arch-specific code either to override the user or more
2205 * commonly to provide a default console (ie from PROM variables) when
2206 * the user has not supplied one.
2208 int add_preferred_console(char *name, int idx, char *options)
2210 return __add_preferred_console(name, idx, options, NULL);
2213 bool console_suspend_enabled = true;
2214 EXPORT_SYMBOL(console_suspend_enabled);
2216 static int __init console_suspend_disable(char *str)
2218 console_suspend_enabled = false;
2221 __setup("no_console_suspend", console_suspend_disable);
2222 module_param_named(console_suspend, console_suspend_enabled,
2223 bool, S_IRUGO | S_IWUSR);
2224 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2225 " and hibernate operations");
2228 * suspend_console - suspend the console subsystem
2230 * This disables printk() while we go into suspend states
2232 void suspend_console(void)
2234 if (!console_suspend_enabled)
2236 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2238 console_suspended = 1;
2242 void resume_console(void)
2244 if (!console_suspend_enabled)
2247 console_suspended = 0;
2252 * console_cpu_notify - print deferred console messages after CPU hotplug
2255 * If printk() is called from a CPU that is not online yet, the messages
2256 * will be printed on the console only if there are CON_ANYTIME consoles.
2257 * This function is called when a new CPU comes online (or fails to come
2258 * up) or goes offline.
2260 static int console_cpu_notify(unsigned int cpu)
2262 if (!cpuhp_tasks_frozen) {
2263 /* If trylock fails, someone else is doing the printing */
2264 if (console_trylock())
2271 * console_lock - lock the console system for exclusive use.
2273 * Acquires a lock which guarantees that the caller has
2274 * exclusive access to the console system and the console_drivers list.
2276 * Can sleep, returns nothing.
2278 void console_lock(void)
2283 if (console_suspended)
2286 console_may_schedule = 1;
2288 EXPORT_SYMBOL(console_lock);
2291 * console_trylock - try to lock the console system for exclusive use.
2293 * Try to acquire a lock which guarantees that the caller has exclusive
2294 * access to the console system and the console_drivers list.
2296 * returns 1 on success, and 0 on failure to acquire the lock.
2298 int console_trylock(void)
2300 if (down_trylock_console_sem())
2302 if (console_suspended) {
2307 console_may_schedule = 0;
2310 EXPORT_SYMBOL(console_trylock);
2312 int is_console_locked(void)
2314 return console_locked;
2316 EXPORT_SYMBOL(is_console_locked);
2319 * Check if we have any console that is capable of printing while cpu is
2320 * booting or shutting down. Requires console_sem.
2322 static int have_callable_console(void)
2324 struct console *con;
2326 for_each_console(con)
2327 if ((con->flags & CON_ENABLED) &&
2328 (con->flags & CON_ANYTIME))
2335 * Can we actually use the console at this time on this cpu?
2337 * Console drivers may assume that per-cpu resources have been allocated. So
2338 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2339 * call them until this CPU is officially up.
2341 static inline int can_use_console(void)
2343 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2347 * console_unlock - unlock the console system
2349 * Releases the console_lock which the caller holds on the console system
2350 * and the console driver list.
2352 * While the console_lock was held, console output may have been buffered
2353 * by printk(). If this is the case, console_unlock(); emits
2354 * the output prior to releasing the lock.
2356 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2358 * console_unlock(); may be called from any context.
2360 void console_unlock(void)
2362 static char ext_text[CONSOLE_EXT_LOG_MAX];
2363 static char text[LOG_LINE_MAX + PREFIX_MAX];
2364 unsigned long flags;
2365 bool do_cond_resched, retry;
2367 if (console_suspended) {
2373 * Console drivers are called with interrupts disabled, so
2374 * @console_may_schedule should be cleared before; however, we may
2375 * end up dumping a lot of lines, for example, if called from
2376 * console registration path, and should invoke cond_resched()
2377 * between lines if allowable. Not doing so can cause a very long
2378 * scheduling stall on a slow console leading to RCU stall and
2379 * softlockup warnings which exacerbate the issue with more
2380 * messages practically incapacitating the system.
2382 * console_trylock() is not able to detect the preemptive
2383 * context reliably. Therefore the value must be stored before
2384 * and cleared after the the "again" goto label.
2386 do_cond_resched = console_may_schedule;
2388 console_may_schedule = 0;
2391 * We released the console_sem lock, so we need to recheck if
2392 * cpu is online and (if not) is there at least one CON_ANYTIME
2395 if (!can_use_console()) {
2402 struct printk_log *msg;
2406 printk_safe_enter_irqsave(flags);
2407 raw_spin_lock(&logbuf_lock);
2408 if (console_seq < log_first_seq) {
2410 "** %llu printk messages dropped **\n",
2411 log_first_seq - console_seq);
2413 /* messages are gone, move to first one */
2414 console_seq = log_first_seq;
2415 console_idx = log_first_idx;
2420 if (console_seq == log_next_seq)
2423 msg = log_from_idx(console_idx);
2424 if (suppress_message_printing(msg->level)) {
2426 * Skip record we have buffered and already printed
2427 * directly to the console when we received it, and
2428 * record that has level above the console loglevel.
2430 console_idx = log_next(console_idx);
2435 /* Output to all consoles once old messages replayed. */
2436 if (unlikely(exclusive_console &&
2437 console_seq >= exclusive_console_stop_seq)) {
2438 exclusive_console = NULL;
2441 len += msg_print_text(msg,
2442 console_msg_format & MSG_FORMAT_SYSLOG,
2444 sizeof(text) - len);
2445 if (nr_ext_console_drivers) {
2446 ext_len = msg_print_ext_header(ext_text,
2449 ext_len += msg_print_ext_body(ext_text + ext_len,
2450 sizeof(ext_text) - ext_len,
2451 log_dict(msg), msg->dict_len,
2452 log_text(msg), msg->text_len);
2454 console_idx = log_next(console_idx);
2456 raw_spin_unlock(&logbuf_lock);
2459 * While actively printing out messages, if another printk()
2460 * were to occur on another CPU, it may wait for this one to
2461 * finish. This task can not be preempted if there is a
2462 * waiter waiting to take over.
2464 console_lock_spinning_enable();
2466 stop_critical_timings(); /* don't trace print latency */
2467 call_console_drivers(ext_text, ext_len, text, len);
2468 start_critical_timings();
2470 if (console_lock_spinning_disable_and_check()) {
2471 printk_safe_exit_irqrestore(flags);
2475 printk_safe_exit_irqrestore(flags);
2477 if (do_cond_resched)
2483 raw_spin_unlock(&logbuf_lock);
2488 * Someone could have filled up the buffer again, so re-check if there's
2489 * something to flush. In case we cannot trylock the console_sem again,
2490 * there's a new owner and the console_unlock() from them will do the
2491 * flush, no worries.
2493 raw_spin_lock(&logbuf_lock);
2494 retry = console_seq != log_next_seq;
2495 raw_spin_unlock(&logbuf_lock);
2496 printk_safe_exit_irqrestore(flags);
2498 if (retry && console_trylock())
2501 EXPORT_SYMBOL(console_unlock);
2504 * console_conditional_schedule - yield the CPU if required
2506 * If the console code is currently allowed to sleep, and
2507 * if this CPU should yield the CPU to another task, do
2510 * Must be called within console_lock();.
2512 void __sched console_conditional_schedule(void)
2514 if (console_may_schedule)
2517 EXPORT_SYMBOL(console_conditional_schedule);
2519 void console_unblank(void)
2524 * console_unblank can no longer be called in interrupt context unless
2525 * oops_in_progress is set to 1..
2527 if (oops_in_progress) {
2528 if (down_trylock_console_sem() != 0)
2534 console_may_schedule = 0;
2536 if ((c->flags & CON_ENABLED) && c->unblank)
2542 * console_flush_on_panic - flush console content on panic
2544 * Immediately output all pending messages no matter what.
2546 void console_flush_on_panic(void)
2549 * If someone else is holding the console lock, trylock will fail
2550 * and may_schedule may be set. Ignore and proceed to unlock so
2551 * that messages are flushed out. As this can be called from any
2552 * context and we don't want to get preempted while flushing,
2553 * ensure may_schedule is cleared.
2556 console_may_schedule = 0;
2561 * Return the console tty driver structure and its associated index
2563 struct tty_driver *console_device(int *index)
2566 struct tty_driver *driver = NULL;
2569 for_each_console(c) {
2572 driver = c->device(c, index);
2581 * Prevent further output on the passed console device so that (for example)
2582 * serial drivers can disable console output before suspending a port, and can
2583 * re-enable output afterwards.
2585 void console_stop(struct console *console)
2588 console->flags &= ~CON_ENABLED;
2591 EXPORT_SYMBOL(console_stop);
2593 void console_start(struct console *console)
2596 console->flags |= CON_ENABLED;
2599 EXPORT_SYMBOL(console_start);
2601 static int __read_mostly keep_bootcon;
2603 static int __init keep_bootcon_setup(char *str)
2606 pr_info("debug: skip boot console de-registration.\n");
2611 early_param("keep_bootcon", keep_bootcon_setup);
2614 * The console driver calls this routine during kernel initialization
2615 * to register the console printing procedure with printk() and to
2616 * print any messages that were printed by the kernel before the
2617 * console driver was initialized.
2619 * This can happen pretty early during the boot process (because of
2620 * early_printk) - sometimes before setup_arch() completes - be careful
2621 * of what kernel features are used - they may not be initialised yet.
2623 * There are two types of consoles - bootconsoles (early_printk) and
2624 * "real" consoles (everything which is not a bootconsole) which are
2625 * handled differently.
2626 * - Any number of bootconsoles can be registered at any time.
2627 * - As soon as a "real" console is registered, all bootconsoles
2628 * will be unregistered automatically.
2629 * - Once a "real" console is registered, any attempt to register a
2630 * bootconsoles will be rejected
2632 void register_console(struct console *newcon)
2635 unsigned long flags;
2636 struct console *bcon = NULL;
2637 struct console_cmdline *c;
2638 static bool has_preferred;
2640 if (console_drivers)
2641 for_each_console(bcon)
2642 if (WARN(bcon == newcon,
2643 "console '%s%d' already registered\n",
2644 bcon->name, bcon->index))
2648 * before we register a new CON_BOOT console, make sure we don't
2649 * already have a valid console
2651 if (console_drivers && newcon->flags & CON_BOOT) {
2652 /* find the last or real console */
2653 for_each_console(bcon) {
2654 if (!(bcon->flags & CON_BOOT)) {
2655 pr_info("Too late to register bootconsole %s%d\n",
2656 newcon->name, newcon->index);
2662 if (console_drivers && console_drivers->flags & CON_BOOT)
2663 bcon = console_drivers;
2665 if (!has_preferred || bcon || !console_drivers)
2666 has_preferred = preferred_console >= 0;
2669 * See if we want to use this console driver. If we
2670 * didn't select a console we take the first one
2671 * that registers here.
2673 if (!has_preferred) {
2674 if (newcon->index < 0)
2676 if (newcon->setup == NULL ||
2677 newcon->setup(newcon, NULL) == 0) {
2678 newcon->flags |= CON_ENABLED;
2679 if (newcon->device) {
2680 newcon->flags |= CON_CONSDEV;
2681 has_preferred = true;
2687 * See if this console matches one we selected on
2690 for (i = 0, c = console_cmdline;
2691 i < MAX_CMDLINECONSOLES && c->name[0];
2693 if (!newcon->match ||
2694 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2695 /* default matching */
2696 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2697 if (strcmp(c->name, newcon->name) != 0)
2699 if (newcon->index >= 0 &&
2700 newcon->index != c->index)
2702 if (newcon->index < 0)
2703 newcon->index = c->index;
2705 if (_braille_register_console(newcon, c))
2708 if (newcon->setup &&
2709 newcon->setup(newcon, c->options) != 0)
2713 newcon->flags |= CON_ENABLED;
2714 if (i == preferred_console) {
2715 newcon->flags |= CON_CONSDEV;
2716 has_preferred = true;
2721 if (!(newcon->flags & CON_ENABLED))
2725 * If we have a bootconsole, and are switching to a real console,
2726 * don't print everything out again, since when the boot console, and
2727 * the real console are the same physical device, it's annoying to
2728 * see the beginning boot messages twice
2730 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2731 newcon->flags &= ~CON_PRINTBUFFER;
2734 * Put this console in the list - keep the
2735 * preferred driver at the head of the list.
2738 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2739 newcon->next = console_drivers;
2740 console_drivers = newcon;
2742 newcon->next->flags &= ~CON_CONSDEV;
2744 newcon->next = console_drivers->next;
2745 console_drivers->next = newcon;
2748 if (newcon->flags & CON_EXTENDED)
2749 if (!nr_ext_console_drivers++)
2750 pr_info("printk: continuation disabled due to ext consoles, expect more fragments in /dev/kmsg\n");
2752 if (newcon->flags & CON_PRINTBUFFER) {
2754 * console_unlock(); will print out the buffered messages
2757 logbuf_lock_irqsave(flags);
2759 * We're about to replay the log buffer. Only do this to the
2760 * just-registered console to avoid excessive message spam to
2761 * the already-registered consoles.
2763 * Set exclusive_console with disabled interrupts to reduce
2764 * race window with eventual console_flush_on_panic() that
2765 * ignores console_lock.
2767 exclusive_console = newcon;
2768 exclusive_console_stop_seq = console_seq;
2769 console_seq = syslog_seq;
2770 console_idx = syslog_idx;
2771 logbuf_unlock_irqrestore(flags);
2774 console_sysfs_notify();
2777 * By unregistering the bootconsoles after we enable the real console
2778 * we get the "console xxx enabled" message on all the consoles -
2779 * boot consoles, real consoles, etc - this is to ensure that end
2780 * users know there might be something in the kernel's log buffer that
2781 * went to the bootconsole (that they do not see on the real console)
2783 pr_info("%sconsole [%s%d] enabled\n",
2784 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2785 newcon->name, newcon->index);
2787 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2789 /* We need to iterate through all boot consoles, to make
2790 * sure we print everything out, before we unregister them.
2792 for_each_console(bcon)
2793 if (bcon->flags & CON_BOOT)
2794 unregister_console(bcon);
2797 EXPORT_SYMBOL(register_console);
2799 int unregister_console(struct console *console)
2801 struct console *a, *b;
2804 pr_info("%sconsole [%s%d] disabled\n",
2805 (console->flags & CON_BOOT) ? "boot" : "" ,
2806 console->name, console->index);
2808 res = _braille_unregister_console(console);
2814 if (console_drivers == console) {
2815 console_drivers=console->next;
2817 } else if (console_drivers) {
2818 for (a=console_drivers->next, b=console_drivers ;
2819 a; b=a, a=b->next) {
2828 if (!res && (console->flags & CON_EXTENDED))
2829 nr_ext_console_drivers--;
2832 * If this isn't the last console and it has CON_CONSDEV set, we
2833 * need to set it on the next preferred console.
2835 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2836 console_drivers->flags |= CON_CONSDEV;
2838 console->flags &= ~CON_ENABLED;
2840 console_sysfs_notify();
2843 EXPORT_SYMBOL(unregister_console);
2846 * Initialize the console device. This is called *early*, so
2847 * we can't necessarily depend on lots of kernel help here.
2848 * Just do some early initializations, and do the complex setup
2851 void __init console_init(void)
2855 initcall_entry_t *ce;
2857 /* Setup the default TTY line discipline. */
2861 * set up the console device so that later boot sequences can
2862 * inform about problems etc..
2864 ce = __con_initcall_start;
2865 trace_initcall_level("console");
2866 while (ce < __con_initcall_end) {
2867 call = initcall_from_entry(ce);
2868 trace_initcall_start(call);
2870 trace_initcall_finish(call, ret);
2876 * Some boot consoles access data that is in the init section and which will
2877 * be discarded after the initcalls have been run. To make sure that no code
2878 * will access this data, unregister the boot consoles in a late initcall.
2880 * If for some reason, such as deferred probe or the driver being a loadable
2881 * module, the real console hasn't registered yet at this point, there will
2882 * be a brief interval in which no messages are logged to the console, which
2883 * makes it difficult to diagnose problems that occur during this time.
2885 * To mitigate this problem somewhat, only unregister consoles whose memory
2886 * intersects with the init section. Note that all other boot consoles will
2887 * get unregistred when the real preferred console is registered.
2889 static int __init printk_late_init(void)
2891 struct console *con;
2894 for_each_console(con) {
2895 if (!(con->flags & CON_BOOT))
2898 /* Check addresses that might be used for enabled consoles. */
2899 if (init_section_intersects(con, sizeof(*con)) ||
2900 init_section_contains(con->write, 0) ||
2901 init_section_contains(con->read, 0) ||
2902 init_section_contains(con->device, 0) ||
2903 init_section_contains(con->unblank, 0) ||
2904 init_section_contains(con->data, 0)) {
2906 * Please, consider moving the reported consoles out
2907 * of the init section.
2909 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2910 con->name, con->index);
2911 unregister_console(con);
2914 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2915 console_cpu_notify);
2917 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2918 console_cpu_notify, NULL);
2922 late_initcall(printk_late_init);
2924 #if defined CONFIG_PRINTK
2926 * Delayed printk version, for scheduler-internal messages:
2928 #define PRINTK_PENDING_WAKEUP 0x01
2929 #define PRINTK_PENDING_OUTPUT 0x02
2931 static DEFINE_PER_CPU(int, printk_pending);
2933 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2935 int pending = __this_cpu_xchg(printk_pending, 0);
2937 if (pending & PRINTK_PENDING_OUTPUT) {
2938 /* If trylock fails, someone else is doing the printing */
2939 if (console_trylock())
2943 if (pending & PRINTK_PENDING_WAKEUP)
2944 wake_up_interruptible(&log_wait);
2947 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2948 .func = wake_up_klogd_work_func,
2949 .flags = IRQ_WORK_LAZY,
2952 void wake_up_klogd(void)
2954 if (!printk_percpu_data_ready())
2958 if (waitqueue_active(&log_wait)) {
2959 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2960 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2965 void defer_console_output(void)
2967 if (!printk_percpu_data_ready())
2971 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2972 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2976 int vprintk_deferred(const char *fmt, va_list args)
2980 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2981 defer_console_output();
2986 int printk_deferred(const char *fmt, ...)
2991 va_start(args, fmt);
2992 r = vprintk_deferred(fmt, args);
2999 * printk rate limiting, lifted from the networking subsystem.
3001 * This enforces a rate limit: not more than 10 kernel messages
3002 * every 5s to make a denial-of-service attack impossible.
3004 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3006 int __printk_ratelimit(const char *func)
3008 return ___ratelimit(&printk_ratelimit_state, func);
3010 EXPORT_SYMBOL(__printk_ratelimit);
3013 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3014 * @caller_jiffies: pointer to caller's state
3015 * @interval_msecs: minimum interval between prints
3017 * printk_timed_ratelimit() returns true if more than @interval_msecs
3018 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3021 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3022 unsigned int interval_msecs)
3024 unsigned long elapsed = jiffies - *caller_jiffies;
3026 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3029 *caller_jiffies = jiffies;
3032 EXPORT_SYMBOL(printk_timed_ratelimit);
3034 static DEFINE_SPINLOCK(dump_list_lock);
3035 static LIST_HEAD(dump_list);
3038 * kmsg_dump_register - register a kernel log dumper.
3039 * @dumper: pointer to the kmsg_dumper structure
3041 * Adds a kernel log dumper to the system. The dump callback in the
3042 * structure will be called when the kernel oopses or panics and must be
3043 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3045 int kmsg_dump_register(struct kmsg_dumper *dumper)
3047 unsigned long flags;
3050 /* The dump callback needs to be set */
3054 spin_lock_irqsave(&dump_list_lock, flags);
3055 /* Don't allow registering multiple times */
3056 if (!dumper->registered) {
3057 dumper->registered = 1;
3058 list_add_tail_rcu(&dumper->list, &dump_list);
3061 spin_unlock_irqrestore(&dump_list_lock, flags);
3065 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3068 * kmsg_dump_unregister - unregister a kmsg dumper.
3069 * @dumper: pointer to the kmsg_dumper structure
3071 * Removes a dump device from the system. Returns zero on success and
3072 * %-EINVAL otherwise.
3074 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3076 unsigned long flags;
3079 spin_lock_irqsave(&dump_list_lock, flags);
3080 if (dumper->registered) {
3081 dumper->registered = 0;
3082 list_del_rcu(&dumper->list);
3085 spin_unlock_irqrestore(&dump_list_lock, flags);
3090 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3092 static bool always_kmsg_dump;
3093 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3096 * kmsg_dump - dump kernel log to kernel message dumpers.
3097 * @reason: the reason (oops, panic etc) for dumping
3099 * Call each of the registered dumper's dump() callback, which can
3100 * retrieve the kmsg records with kmsg_dump_get_line() or
3101 * kmsg_dump_get_buffer().
3103 void kmsg_dump(enum kmsg_dump_reason reason)
3105 struct kmsg_dumper *dumper;
3106 unsigned long flags;
3108 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3112 list_for_each_entry_rcu(dumper, &dump_list, list) {
3113 if (dumper->max_reason && reason > dumper->max_reason)
3116 /* initialize iterator with data about the stored records */
3117 dumper->active = true;
3119 logbuf_lock_irqsave(flags);
3120 dumper->cur_seq = clear_seq;
3121 dumper->cur_idx = clear_idx;
3122 dumper->next_seq = log_next_seq;
3123 dumper->next_idx = log_next_idx;
3124 logbuf_unlock_irqrestore(flags);
3126 /* invoke dumper which will iterate over records */
3127 dumper->dump(dumper, reason);
3129 /* reset iterator */
3130 dumper->active = false;
3136 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3137 * @dumper: registered kmsg dumper
3138 * @syslog: include the "<4>" prefixes
3139 * @line: buffer to copy the line to
3140 * @size: maximum size of the buffer
3141 * @len: length of line placed into buffer
3143 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3144 * record, and copy one record into the provided buffer.
3146 * Consecutive calls will return the next available record moving
3147 * towards the end of the buffer with the youngest messages.
3149 * A return value of FALSE indicates that there are no more records to
3152 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3154 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3155 char *line, size_t size, size_t *len)
3157 struct printk_log *msg;
3161 if (!dumper->active)
3164 if (dumper->cur_seq < log_first_seq) {
3165 /* messages are gone, move to first available one */
3166 dumper->cur_seq = log_first_seq;
3167 dumper->cur_idx = log_first_idx;
3171 if (dumper->cur_seq >= log_next_seq)
3174 msg = log_from_idx(dumper->cur_idx);
3175 l = msg_print_text(msg, syslog, line, size);
3177 dumper->cur_idx = log_next(dumper->cur_idx);
3187 * kmsg_dump_get_line - retrieve one kmsg log line
3188 * @dumper: registered kmsg dumper
3189 * @syslog: include the "<4>" prefixes
3190 * @line: buffer to copy the line to
3191 * @size: maximum size of the buffer
3192 * @len: length of line placed into buffer
3194 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3195 * record, and copy one record into the provided buffer.
3197 * Consecutive calls will return the next available record moving
3198 * towards the end of the buffer with the youngest messages.
3200 * A return value of FALSE indicates that there are no more records to
3203 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3204 char *line, size_t size, size_t *len)
3206 unsigned long flags;
3209 logbuf_lock_irqsave(flags);
3210 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3211 logbuf_unlock_irqrestore(flags);
3215 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3218 * kmsg_dump_get_buffer - copy kmsg log lines
3219 * @dumper: registered kmsg dumper
3220 * @syslog: include the "<4>" prefixes
3221 * @buf: buffer to copy the line to
3222 * @size: maximum size of the buffer
3223 * @len: length of line placed into buffer
3225 * Start at the end of the kmsg buffer and fill the provided buffer
3226 * with as many of the the *youngest* kmsg records that fit into it.
3227 * If the buffer is large enough, all available kmsg records will be
3228 * copied with a single call.
3230 * Consecutive calls will fill the buffer with the next block of
3231 * available older records, not including the earlier retrieved ones.
3233 * A return value of FALSE indicates that there are no more records to
3236 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3237 char *buf, size_t size, size_t *len)
3239 unsigned long flags;
3247 if (!dumper->active)
3250 logbuf_lock_irqsave(flags);
3251 if (dumper->cur_seq < log_first_seq) {
3252 /* messages are gone, move to first available one */
3253 dumper->cur_seq = log_first_seq;
3254 dumper->cur_idx = log_first_idx;
3258 if (dumper->cur_seq >= dumper->next_seq) {
3259 logbuf_unlock_irqrestore(flags);
3263 /* calculate length of entire buffer */
3264 seq = dumper->cur_seq;
3265 idx = dumper->cur_idx;
3266 while (seq < dumper->next_seq) {
3267 struct printk_log *msg = log_from_idx(idx);
3269 l += msg_print_text(msg, true, NULL, 0);
3270 idx = log_next(idx);
3274 /* move first record forward until length fits into the buffer */
3275 seq = dumper->cur_seq;
3276 idx = dumper->cur_idx;
3277 while (l >= size && seq < dumper->next_seq) {
3278 struct printk_log *msg = log_from_idx(idx);
3280 l -= msg_print_text(msg, true, NULL, 0);
3281 idx = log_next(idx);
3285 /* last message in next interation */
3290 while (seq < dumper->next_seq) {
3291 struct printk_log *msg = log_from_idx(idx);
3293 l += msg_print_text(msg, syslog, buf + l, size - l);
3294 idx = log_next(idx);
3298 dumper->next_seq = next_seq;
3299 dumper->next_idx = next_idx;
3301 logbuf_unlock_irqrestore(flags);
3307 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3310 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3311 * @dumper: registered kmsg dumper
3313 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3314 * kmsg_dump_get_buffer() can be called again and used multiple
3315 * times within the same dumper.dump() callback.
3317 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3319 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3321 dumper->cur_seq = clear_seq;
3322 dumper->cur_idx = clear_idx;
3323 dumper->next_seq = log_next_seq;
3324 dumper->next_idx = log_next_idx;
3328 * kmsg_dump_rewind - reset the interator
3329 * @dumper: registered kmsg dumper
3331 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3332 * kmsg_dump_get_buffer() can be called again and used multiple
3333 * times within the same dumper.dump() callback.
3335 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3337 unsigned long flags;
3339 logbuf_lock_irqsave(flags);
3340 kmsg_dump_rewind_nolock(dumper);
3341 logbuf_unlock_irqrestore(flags);
3343 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);