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/kexec.h>
36 #include <linux/kdb.h>
37 #include <linux/ratelimit.h>
38 #include <linux/kmsg_dump.h>
39 #include <linux/syslog.h>
40 #include <linux/cpu.h>
41 #include <linux/notifier.h>
42 #include <linux/rculist.h>
43 #include <linux/poll.h>
44 #include <linux/irq_work.h>
45 #include <linux/utsname.h>
46 #include <linux/ctype.h>
47 #include <linux/uio.h>
49 #include <asm/uaccess.h>
50 #include <asm/sections.h>
52 #define CREATE_TRACE_POINTS
53 #include <trace/events/printk.h>
55 #include "console_cmdline.h"
59 int console_printk[4] = {
60 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
61 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
62 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
63 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
67 * Low level drivers may need that to know if they can schedule in
68 * their unblank() callback or not. So let's export it.
71 EXPORT_SYMBOL(oops_in_progress);
74 * console_sem protects the console_drivers list, and also
75 * provides serialisation for access to the entire console
78 static DEFINE_SEMAPHORE(console_sem);
79 struct console *console_drivers;
80 EXPORT_SYMBOL_GPL(console_drivers);
83 static struct lockdep_map console_lock_dep_map = {
84 .name = "console_lock"
88 enum devkmsg_log_bits {
89 __DEVKMSG_LOG_BIT_ON = 0,
90 __DEVKMSG_LOG_BIT_OFF,
91 __DEVKMSG_LOG_BIT_LOCK,
94 enum devkmsg_log_masks {
95 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
96 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
97 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
100 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
101 #define DEVKMSG_LOG_MASK_DEFAULT 0
103 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
105 static int __control_devkmsg(char *str)
110 if (!strncmp(str, "on", 2)) {
111 devkmsg_log = DEVKMSG_LOG_MASK_ON;
113 } else if (!strncmp(str, "off", 3)) {
114 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
116 } else if (!strncmp(str, "ratelimit", 9)) {
117 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
123 static int __init control_devkmsg(char *str)
125 if (__control_devkmsg(str) < 0)
129 * Set sysctl string accordingly:
131 if (devkmsg_log == DEVKMSG_LOG_MASK_ON) {
132 memset(devkmsg_log_str, 0, DEVKMSG_STR_MAX_SIZE);
133 strncpy(devkmsg_log_str, "on", 2);
134 } else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF) {
135 memset(devkmsg_log_str, 0, DEVKMSG_STR_MAX_SIZE);
136 strncpy(devkmsg_log_str, "off", 3);
138 /* else "ratelimit" which is set by default. */
141 * Sysctl cannot change it anymore. The kernel command line setting of
142 * this parameter is to force the setting to be permanent throughout the
143 * runtime of the system. This is a precation measure against userspace
144 * trying to be a smarta** and attempting to change it up on us.
146 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
150 __setup("printk.devkmsg=", control_devkmsg);
152 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
154 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
155 void __user *buffer, size_t *lenp, loff_t *ppos)
157 char old_str[DEVKMSG_STR_MAX_SIZE];
162 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
166 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
169 err = proc_dostring(table, write, buffer, lenp, ppos);
174 err = __control_devkmsg(devkmsg_log_str);
177 * Do not accept an unknown string OR a known string with
180 if (err < 0 || (err + 1 != *lenp)) {
182 /* ... and restore old setting. */
184 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
194 * Number of registered extended console drivers.
196 * If extended consoles are present, in-kernel cont reassembly is disabled
197 * and each fragment is stored as a separate log entry with proper
198 * continuation flag so that every emitted message has full metadata. This
199 * doesn't change the result for regular consoles or /proc/kmsg. For
200 * /dev/kmsg, as long as the reader concatenates messages according to
201 * consecutive continuation flags, the end result should be the same too.
203 static int nr_ext_console_drivers;
206 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
207 * macros instead of functions so that _RET_IP_ contains useful information.
209 #define down_console_sem() do { \
211 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
214 static int __down_trylock_console_sem(unsigned long ip)
216 if (down_trylock(&console_sem))
218 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
221 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
223 #define up_console_sem() do { \
224 mutex_release(&console_lock_dep_map, 1, _RET_IP_);\
229 * This is used for debugging the mess that is the VT code by
230 * keeping track if we have the console semaphore held. It's
231 * definitely not the perfect debug tool (we don't know if _WE_
232 * hold it and are racing, but it helps tracking those weird code
233 * paths in the console code where we end up in places I want
234 * locked without the console sempahore held).
236 static int console_locked, console_suspended;
239 * If exclusive_console is non-NULL then only this console is to be printed to.
241 static struct console *exclusive_console;
244 * Array of consoles built from command line options (console=)
247 #define MAX_CMDLINECONSOLES 8
249 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
251 static int selected_console = -1;
252 static int preferred_console = -1;
253 int console_set_on_cmdline;
254 EXPORT_SYMBOL(console_set_on_cmdline);
256 /* Flag: console code may call schedule() */
257 static int console_may_schedule;
260 * The printk log buffer consists of a chain of concatenated variable
261 * length records. Every record starts with a record header, containing
262 * the overall length of the record.
264 * The heads to the first and last entry in the buffer, as well as the
265 * sequence numbers of these entries are maintained when messages are
268 * If the heads indicate available messages, the length in the header
269 * tells the start next message. A length == 0 for the next message
270 * indicates a wrap-around to the beginning of the buffer.
272 * Every record carries the monotonic timestamp in microseconds, as well as
273 * the standard userspace syslog level and syslog facility. The usual
274 * kernel messages use LOG_KERN; userspace-injected messages always carry
275 * a matching syslog facility, by default LOG_USER. The origin of every
276 * message can be reliably determined that way.
278 * The human readable log message directly follows the message header. The
279 * length of the message text is stored in the header, the stored message
282 * Optionally, a message can carry a dictionary of properties (key/value pairs),
283 * to provide userspace with a machine-readable message context.
285 * Examples for well-defined, commonly used property names are:
286 * DEVICE=b12:8 device identifier
290 * +sound:card0 subsystem:devname
291 * SUBSYSTEM=pci driver-core subsystem name
293 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
294 * follows directly after a '=' character. Every property is terminated by
295 * a '\0' character. The last property is not terminated.
297 * Example of a message structure:
298 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
299 * 0008 34 00 record is 52 bytes long
300 * 000a 0b 00 text is 11 bytes long
301 * 000c 1f 00 dictionary is 23 bytes long
302 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
303 * 0010 69 74 27 73 20 61 20 6c "it's a l"
305 * 001b 44 45 56 49 43 "DEVIC"
306 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
307 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
309 * 0032 00 00 00 padding to next message header
311 * The 'struct printk_log' buffer header must never be directly exported to
312 * userspace, it is a kernel-private implementation detail that might
313 * need to be changed in the future, when the requirements change.
315 * /dev/kmsg exports the structured data in the following line format:
316 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
318 * Users of the export format should ignore possible additional values
319 * separated by ',', and find the message after the ';' character.
321 * The optional key/value pairs are attached as continuation lines starting
322 * with a space character and terminated by a newline. All possible
323 * non-prinatable characters are escaped in the "\xff" notation.
327 LOG_NOCONS = 1, /* already flushed, do not print to console */
328 LOG_NEWLINE = 2, /* text ended with a newline */
329 LOG_PREFIX = 4, /* text started with a prefix */
330 LOG_CONT = 8, /* text is a fragment of a continuation line */
334 u64 ts_nsec; /* timestamp in nanoseconds */
335 u16 len; /* length of entire record */
336 u16 text_len; /* length of text buffer */
337 u16 dict_len; /* length of dictionary buffer */
338 u8 facility; /* syslog facility */
339 u8 flags:5; /* internal record flags */
340 u8 level:3; /* syslog level */
342 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
343 __packed __aligned(4)
348 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
349 * within the scheduler's rq lock. It must be released before calling
350 * console_unlock() or anything else that might wake up a process.
352 DEFINE_RAW_SPINLOCK(logbuf_lock);
355 DECLARE_WAIT_QUEUE_HEAD(log_wait);
356 /* the next printk record to read by syslog(READ) or /proc/kmsg */
357 static u64 syslog_seq;
358 static u32 syslog_idx;
359 static size_t syslog_partial;
361 /* index and sequence number of the first record stored in the buffer */
362 static u64 log_first_seq;
363 static u32 log_first_idx;
365 /* index and sequence number of the next record to store in the buffer */
366 static u64 log_next_seq;
367 static u32 log_next_idx;
369 /* the next printk record to write to the console */
370 static u64 console_seq;
371 static u32 console_idx;
373 /* the next printk record to read after the last 'clear' command */
374 static u64 clear_seq;
375 static u32 clear_idx;
377 #define PREFIX_MAX 32
378 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
380 #define LOG_LEVEL(v) ((v) & 0x07)
381 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
384 #define LOG_ALIGN __alignof__(struct printk_log)
385 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
386 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
387 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
388 static char *log_buf = __log_buf;
389 static u32 log_buf_len = __LOG_BUF_LEN;
391 /* Return log buffer address */
392 char *log_buf_addr_get(void)
397 /* Return log buffer size */
398 u32 log_buf_len_get(void)
403 /* human readable text of the record */
404 static char *log_text(const struct printk_log *msg)
406 return (char *)msg + sizeof(struct printk_log);
409 /* optional key/value pair dictionary attached to the record */
410 static char *log_dict(const struct printk_log *msg)
412 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
415 /* get record by index; idx must point to valid msg */
416 static struct printk_log *log_from_idx(u32 idx)
418 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
421 * A length == 0 record is the end of buffer marker. Wrap around and
422 * read the message at the start of the buffer.
425 return (struct printk_log *)log_buf;
429 /* get next record; idx must point to valid msg */
430 static u32 log_next(u32 idx)
432 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
434 /* length == 0 indicates the end of the buffer; wrap */
436 * A length == 0 record is the end of buffer marker. Wrap around and
437 * read the message at the start of the buffer as *this* one, and
438 * return the one after that.
441 msg = (struct printk_log *)log_buf;
444 return idx + msg->len;
448 * Check whether there is enough free space for the given message.
450 * The same values of first_idx and next_idx mean that the buffer
451 * is either empty or full.
453 * If the buffer is empty, we must respect the position of the indexes.
454 * They cannot be reset to the beginning of the buffer.
456 static int logbuf_has_space(u32 msg_size, bool empty)
460 if (log_next_idx > log_first_idx || empty)
461 free = max(log_buf_len - log_next_idx, log_first_idx);
463 free = log_first_idx - log_next_idx;
466 * We need space also for an empty header that signalizes wrapping
469 return free >= msg_size + sizeof(struct printk_log);
472 static int log_make_free_space(u32 msg_size)
474 while (log_first_seq < log_next_seq &&
475 !logbuf_has_space(msg_size, false)) {
476 /* drop old messages until we have enough contiguous space */
477 log_first_idx = log_next(log_first_idx);
481 if (clear_seq < log_first_seq) {
482 clear_seq = log_first_seq;
483 clear_idx = log_first_idx;
486 /* sequence numbers are equal, so the log buffer is empty */
487 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
493 /* compute the message size including the padding bytes */
494 static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
498 size = sizeof(struct printk_log) + text_len + dict_len;
499 *pad_len = (-size) & (LOG_ALIGN - 1);
506 * Define how much of the log buffer we could take at maximum. The value
507 * must be greater than two. Note that only half of the buffer is available
508 * when the index points to the middle.
510 #define MAX_LOG_TAKE_PART 4
511 static const char trunc_msg[] = "<truncated>";
513 static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
514 u16 *dict_len, u32 *pad_len)
517 * The message should not take the whole buffer. Otherwise, it might
518 * get removed too soon.
520 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
521 if (*text_len > max_text_len)
522 *text_len = max_text_len;
523 /* enable the warning message */
524 *trunc_msg_len = strlen(trunc_msg);
525 /* disable the "dict" completely */
527 /* compute the size again, count also the warning message */
528 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
531 /* insert record into the buffer, discard old ones, update heads */
532 static int log_store(int facility, int level,
533 enum log_flags flags, u64 ts_nsec,
534 const char *dict, u16 dict_len,
535 const char *text, u16 text_len)
537 struct printk_log *msg;
539 u16 trunc_msg_len = 0;
541 /* number of '\0' padding bytes to next message */
542 size = msg_used_size(text_len, dict_len, &pad_len);
544 if (log_make_free_space(size)) {
545 /* truncate the message if it is too long for empty buffer */
546 size = truncate_msg(&text_len, &trunc_msg_len,
547 &dict_len, &pad_len);
548 /* survive when the log buffer is too small for trunc_msg */
549 if (log_make_free_space(size))
553 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
555 * This message + an additional empty header does not fit
556 * at the end of the buffer. Add an empty header with len == 0
557 * to signify a wrap around.
559 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
564 msg = (struct printk_log *)(log_buf + log_next_idx);
565 memcpy(log_text(msg), text, text_len);
566 msg->text_len = text_len;
568 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
569 msg->text_len += trunc_msg_len;
571 memcpy(log_dict(msg), dict, dict_len);
572 msg->dict_len = dict_len;
573 msg->facility = facility;
574 msg->level = level & 7;
575 msg->flags = flags & 0x1f;
577 msg->ts_nsec = ts_nsec;
579 msg->ts_nsec = local_clock();
580 memset(log_dict(msg) + dict_len, 0, pad_len);
584 log_next_idx += msg->len;
587 return msg->text_len;
590 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
592 static int syslog_action_restricted(int type)
597 * Unless restricted, we allow "read all" and "get buffer size"
600 return type != SYSLOG_ACTION_READ_ALL &&
601 type != SYSLOG_ACTION_SIZE_BUFFER;
604 int check_syslog_permissions(int type, int source)
607 * If this is from /proc/kmsg and we've already opened it, then we've
608 * already done the capabilities checks at open time.
610 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
613 if (syslog_action_restricted(type)) {
614 if (capable(CAP_SYSLOG))
617 * For historical reasons, accept CAP_SYS_ADMIN too, with
620 if (capable(CAP_SYS_ADMIN)) {
621 pr_warn_once("%s (%d): Attempt to access syslog with "
622 "CAP_SYS_ADMIN but no CAP_SYSLOG "
624 current->comm, task_pid_nr(current));
630 return security_syslog(type);
632 EXPORT_SYMBOL_GPL(check_syslog_permissions);
634 static void append_char(char **pp, char *e, char c)
640 static ssize_t msg_print_ext_header(char *buf, size_t size,
641 struct printk_log *msg, u64 seq)
643 u64 ts_usec = msg->ts_nsec;
645 do_div(ts_usec, 1000);
647 return scnprintf(buf, size, "%u,%llu,%llu,%c;",
648 (msg->facility << 3) | msg->level, seq, ts_usec,
649 msg->flags & LOG_CONT ? 'c' : '-');
652 static ssize_t msg_print_ext_body(char *buf, size_t size,
653 char *dict, size_t dict_len,
654 char *text, size_t text_len)
656 char *p = buf, *e = buf + size;
659 /* escape non-printable characters */
660 for (i = 0; i < text_len; i++) {
661 unsigned char c = text[i];
663 if (c < ' ' || c >= 127 || c == '\\')
664 p += scnprintf(p, e - p, "\\x%02x", c);
666 append_char(&p, e, c);
668 append_char(&p, e, '\n');
673 for (i = 0; i < dict_len; i++) {
674 unsigned char c = dict[i];
677 append_char(&p, e, ' ');
682 append_char(&p, e, '\n');
687 if (c < ' ' || c >= 127 || c == '\\') {
688 p += scnprintf(p, e - p, "\\x%02x", c);
692 append_char(&p, e, c);
694 append_char(&p, e, '\n');
700 /* /dev/kmsg - userspace message inject/listen interface */
701 struct devkmsg_user {
704 struct ratelimit_state rs;
706 char buf[CONSOLE_EXT_LOG_MAX];
709 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
712 int level = default_message_loglevel;
713 int facility = 1; /* LOG_USER */
714 struct file *file = iocb->ki_filp;
715 struct devkmsg_user *user = file->private_data;
716 size_t len = iov_iter_count(from);
719 if (!user || len > LOG_LINE_MAX)
722 /* Ignore when user logging is disabled. */
723 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
726 /* Ratelimit when not explicitly enabled. */
727 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
728 if (!___ratelimit(&user->rs, current->comm))
732 buf = kmalloc(len+1, GFP_KERNEL);
737 if (copy_from_iter(buf, len, from) != len) {
743 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
744 * the decimal value represents 32bit, the lower 3 bit are the log
745 * level, the rest are the log facility.
747 * If no prefix or no userspace facility is specified, we
748 * enforce LOG_USER, to be able to reliably distinguish
749 * kernel-generated messages from userspace-injected ones.
752 if (line[0] == '<') {
756 u = simple_strtoul(line + 1, &endp, 10);
757 if (endp && endp[0] == '>') {
758 level = LOG_LEVEL(u);
759 if (LOG_FACILITY(u) != 0)
760 facility = LOG_FACILITY(u);
767 printk_emit(facility, level, NULL, 0, "%s", line);
772 static ssize_t devkmsg_read(struct file *file, char __user *buf,
773 size_t count, loff_t *ppos)
775 struct devkmsg_user *user = file->private_data;
776 struct printk_log *msg;
783 ret = mutex_lock_interruptible(&user->lock);
786 raw_spin_lock_irq(&logbuf_lock);
787 while (user->seq == log_next_seq) {
788 if (file->f_flags & O_NONBLOCK) {
790 raw_spin_unlock_irq(&logbuf_lock);
794 raw_spin_unlock_irq(&logbuf_lock);
795 ret = wait_event_interruptible(log_wait,
796 user->seq != log_next_seq);
799 raw_spin_lock_irq(&logbuf_lock);
802 if (user->seq < log_first_seq) {
803 /* our last seen message is gone, return error and reset */
804 user->idx = log_first_idx;
805 user->seq = log_first_seq;
807 raw_spin_unlock_irq(&logbuf_lock);
811 msg = log_from_idx(user->idx);
812 len = msg_print_ext_header(user->buf, sizeof(user->buf),
814 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
815 log_dict(msg), msg->dict_len,
816 log_text(msg), msg->text_len);
818 user->idx = log_next(user->idx);
820 raw_spin_unlock_irq(&logbuf_lock);
827 if (copy_to_user(buf, user->buf, len)) {
833 mutex_unlock(&user->lock);
837 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
839 struct devkmsg_user *user = file->private_data;
847 raw_spin_lock_irq(&logbuf_lock);
850 /* the first record */
851 user->idx = log_first_idx;
852 user->seq = log_first_seq;
856 * The first record after the last SYSLOG_ACTION_CLEAR,
857 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
858 * changes no global state, and does not clear anything.
860 user->idx = clear_idx;
861 user->seq = clear_seq;
864 /* after the last record */
865 user->idx = log_next_idx;
866 user->seq = log_next_seq;
871 raw_spin_unlock_irq(&logbuf_lock);
875 static unsigned int devkmsg_poll(struct file *file, poll_table *wait)
877 struct devkmsg_user *user = file->private_data;
881 return POLLERR|POLLNVAL;
883 poll_wait(file, &log_wait, wait);
885 raw_spin_lock_irq(&logbuf_lock);
886 if (user->seq < log_next_seq) {
887 /* return error when data has vanished underneath us */
888 if (user->seq < log_first_seq)
889 ret = POLLIN|POLLRDNORM|POLLERR|POLLPRI;
891 ret = POLLIN|POLLRDNORM;
893 raw_spin_unlock_irq(&logbuf_lock);
898 static int devkmsg_open(struct inode *inode, struct file *file)
900 struct devkmsg_user *user;
903 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
906 /* write-only does not need any file context */
907 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
908 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
914 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
918 ratelimit_default_init(&user->rs);
919 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
921 mutex_init(&user->lock);
923 raw_spin_lock_irq(&logbuf_lock);
924 user->idx = log_first_idx;
925 user->seq = log_first_seq;
926 raw_spin_unlock_irq(&logbuf_lock);
928 file->private_data = user;
932 static int devkmsg_release(struct inode *inode, struct file *file)
934 struct devkmsg_user *user = file->private_data;
939 ratelimit_state_exit(&user->rs);
941 mutex_destroy(&user->lock);
946 const struct file_operations kmsg_fops = {
947 .open = devkmsg_open,
948 .read = devkmsg_read,
949 .write_iter = devkmsg_write,
950 .llseek = devkmsg_llseek,
951 .poll = devkmsg_poll,
952 .release = devkmsg_release,
955 #ifdef CONFIG_KEXEC_CORE
957 * This appends the listed symbols to /proc/vmcore
959 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
960 * obtain access to symbols that are otherwise very difficult to locate. These
961 * symbols are specifically used so that utilities can access and extract the
962 * dmesg log from a vmcore file after a crash.
964 void log_buf_kexec_setup(void)
966 VMCOREINFO_SYMBOL(log_buf);
967 VMCOREINFO_SYMBOL(log_buf_len);
968 VMCOREINFO_SYMBOL(log_first_idx);
969 VMCOREINFO_SYMBOL(clear_idx);
970 VMCOREINFO_SYMBOL(log_next_idx);
972 * Export struct printk_log size and field offsets. User space tools can
973 * parse it and detect any changes to structure down the line.
975 VMCOREINFO_STRUCT_SIZE(printk_log);
976 VMCOREINFO_OFFSET(printk_log, ts_nsec);
977 VMCOREINFO_OFFSET(printk_log, len);
978 VMCOREINFO_OFFSET(printk_log, text_len);
979 VMCOREINFO_OFFSET(printk_log, dict_len);
983 /* requested log_buf_len from kernel cmdline */
984 static unsigned long __initdata new_log_buf_len;
986 /* we practice scaling the ring buffer by powers of 2 */
987 static void __init log_buf_len_update(u64 size)
989 if (size > (u64)LOG_BUF_LEN_MAX) {
990 size = (u64)LOG_BUF_LEN_MAX;
991 pr_err("log_buf over 2G is not supported.\n");
995 size = roundup_pow_of_two(size);
996 if (size > log_buf_len)
997 new_log_buf_len = (unsigned long)size;
1000 /* save requested log_buf_len since it's too early to process it */
1001 static int __init log_buf_len_setup(char *str)
1008 size = memparse(str, &str);
1010 log_buf_len_update(size);
1014 early_param("log_buf_len", log_buf_len_setup);
1017 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1019 static void __init log_buf_add_cpu(void)
1021 unsigned int cpu_extra;
1024 * archs should set up cpu_possible_bits properly with
1025 * set_cpu_possible() after setup_arch() but just in
1026 * case lets ensure this is valid.
1028 if (num_possible_cpus() == 1)
1031 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1033 /* by default this will only continue through for large > 64 CPUs */
1034 if (cpu_extra <= __LOG_BUF_LEN / 2)
1037 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1038 __LOG_CPU_MAX_BUF_LEN);
1039 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1041 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1043 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1045 #else /* !CONFIG_SMP */
1046 static inline void log_buf_add_cpu(void) {}
1047 #endif /* CONFIG_SMP */
1049 void __init setup_log_buf(int early)
1051 unsigned long flags;
1055 if (log_buf != __log_buf)
1058 if (!early && !new_log_buf_len)
1061 if (!new_log_buf_len)
1066 memblock_virt_alloc(new_log_buf_len, LOG_ALIGN);
1068 new_log_buf = memblock_virt_alloc_nopanic(new_log_buf_len,
1072 if (unlikely(!new_log_buf)) {
1073 pr_err("log_buf_len: %lu bytes not available\n",
1078 raw_spin_lock_irqsave(&logbuf_lock, flags);
1079 log_buf_len = new_log_buf_len;
1080 log_buf = new_log_buf;
1081 new_log_buf_len = 0;
1082 free = __LOG_BUF_LEN - log_next_idx;
1083 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1084 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
1086 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1087 pr_info("early log buf free: %u(%u%%)\n",
1088 free, (free * 100) / __LOG_BUF_LEN);
1091 static bool __read_mostly ignore_loglevel;
1093 static int __init ignore_loglevel_setup(char *str)
1095 ignore_loglevel = true;
1096 pr_info("debug: ignoring loglevel setting.\n");
1101 early_param("ignore_loglevel", ignore_loglevel_setup);
1102 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1103 MODULE_PARM_DESC(ignore_loglevel,
1104 "ignore loglevel setting (prints all kernel messages to the console)");
1106 static bool suppress_message_printing(int level)
1108 return (level >= console_loglevel && !ignore_loglevel);
1111 #ifdef CONFIG_BOOT_PRINTK_DELAY
1113 static int boot_delay; /* msecs delay after each printk during bootup */
1114 static unsigned long long loops_per_msec; /* based on boot_delay */
1116 static int __init boot_delay_setup(char *str)
1120 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1121 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1123 get_option(&str, &boot_delay);
1124 if (boot_delay > 10 * 1000)
1127 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1128 "HZ: %d, loops_per_msec: %llu\n",
1129 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1132 early_param("boot_delay", boot_delay_setup);
1134 static void boot_delay_msec(int level)
1136 unsigned long long k;
1137 unsigned long timeout;
1139 if ((boot_delay == 0 || system_state != SYSTEM_BOOTING)
1140 || suppress_message_printing(level)) {
1144 k = (unsigned long long)loops_per_msec * boot_delay;
1146 timeout = jiffies + msecs_to_jiffies(boot_delay);
1151 * use (volatile) jiffies to prevent
1152 * compiler reduction; loop termination via jiffies
1153 * is secondary and may or may not happen.
1155 if (time_after(jiffies, timeout))
1157 touch_nmi_watchdog();
1161 static inline void boot_delay_msec(int level)
1166 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1167 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1169 static size_t print_time(u64 ts, char *buf)
1171 unsigned long rem_nsec;
1176 rem_nsec = do_div(ts, 1000000000);
1179 return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
1181 return sprintf(buf, "[%5lu.%06lu] ",
1182 (unsigned long)ts, rem_nsec / 1000);
1185 static size_t print_prefix(const struct printk_log *msg, bool syslog, char *buf)
1188 unsigned int prefix = (msg->facility << 3) | msg->level;
1192 len += sprintf(buf, "<%u>", prefix);
1197 else if (prefix > 99)
1199 else if (prefix > 9)
1204 len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
1208 static size_t msg_print_text(const struct printk_log *msg, bool syslog, char *buf, size_t size)
1210 const char *text = log_text(msg);
1211 size_t text_size = msg->text_len;
1215 const char *next = memchr(text, '\n', text_size);
1219 text_len = next - text;
1221 text_size -= next - text;
1223 text_len = text_size;
1227 if (print_prefix(msg, syslog, NULL) +
1228 text_len + 1 >= size - len)
1231 len += print_prefix(msg, syslog, buf + len);
1232 memcpy(buf + len, text, text_len);
1236 /* SYSLOG_ACTION_* buffer size only calculation */
1237 len += print_prefix(msg, syslog, NULL);
1248 static int syslog_print(char __user *buf, int size)
1251 struct printk_log *msg;
1254 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1262 raw_spin_lock_irq(&logbuf_lock);
1263 if (syslog_seq < log_first_seq) {
1264 /* messages are gone, move to first one */
1265 syslog_seq = log_first_seq;
1266 syslog_idx = log_first_idx;
1269 if (syslog_seq == log_next_seq) {
1270 raw_spin_unlock_irq(&logbuf_lock);
1274 skip = syslog_partial;
1275 msg = log_from_idx(syslog_idx);
1276 n = msg_print_text(msg, true, text, LOG_LINE_MAX + PREFIX_MAX);
1277 if (n - syslog_partial <= size) {
1278 /* message fits into buffer, move forward */
1279 syslog_idx = log_next(syslog_idx);
1281 n -= syslog_partial;
1284 /* partial read(), remember position */
1286 syslog_partial += n;
1289 raw_spin_unlock_irq(&logbuf_lock);
1294 if (copy_to_user(buf, text + skip, n)) {
1309 static int syslog_print_all(char __user *buf, int size, bool clear)
1314 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1318 raw_spin_lock_irq(&logbuf_lock);
1325 * Find first record that fits, including all following records,
1326 * into the user-provided buffer for this dump.
1330 while (seq < log_next_seq) {
1331 struct printk_log *msg = log_from_idx(idx);
1333 len += msg_print_text(msg, true, NULL, 0);
1334 idx = log_next(idx);
1338 /* move first record forward until length fits into the buffer */
1341 while (len > size && seq < log_next_seq) {
1342 struct printk_log *msg = log_from_idx(idx);
1344 len -= msg_print_text(msg, true, NULL, 0);
1345 idx = log_next(idx);
1349 /* last message fitting into this dump */
1350 next_seq = log_next_seq;
1353 while (len >= 0 && seq < next_seq) {
1354 struct printk_log *msg = log_from_idx(idx);
1357 textlen = msg_print_text(msg, true, text,
1358 LOG_LINE_MAX + PREFIX_MAX);
1363 idx = log_next(idx);
1366 raw_spin_unlock_irq(&logbuf_lock);
1367 if (copy_to_user(buf + len, text, textlen))
1371 raw_spin_lock_irq(&logbuf_lock);
1373 if (seq < log_first_seq) {
1374 /* messages are gone, move to next one */
1375 seq = log_first_seq;
1376 idx = log_first_idx;
1382 clear_seq = log_next_seq;
1383 clear_idx = log_next_idx;
1385 raw_spin_unlock_irq(&logbuf_lock);
1391 int do_syslog(int type, char __user *buf, int len, int source)
1394 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1397 error = check_syslog_permissions(type, source);
1402 case SYSLOG_ACTION_CLOSE: /* Close log */
1404 case SYSLOG_ACTION_OPEN: /* Open log */
1406 case SYSLOG_ACTION_READ: /* Read from log */
1408 if (!buf || len < 0)
1413 if (!access_ok(VERIFY_WRITE, buf, len)) {
1417 error = wait_event_interruptible(log_wait,
1418 syslog_seq != log_next_seq);
1421 error = syslog_print(buf, len);
1423 /* Read/clear last kernel messages */
1424 case SYSLOG_ACTION_READ_CLEAR:
1427 /* Read last kernel messages */
1428 case SYSLOG_ACTION_READ_ALL:
1430 if (!buf || len < 0)
1435 if (!access_ok(VERIFY_WRITE, buf, len)) {
1439 error = syslog_print_all(buf, len, clear);
1441 /* Clear ring buffer */
1442 case SYSLOG_ACTION_CLEAR:
1443 syslog_print_all(NULL, 0, true);
1445 /* Disable logging to console */
1446 case SYSLOG_ACTION_CONSOLE_OFF:
1447 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1448 saved_console_loglevel = console_loglevel;
1449 console_loglevel = minimum_console_loglevel;
1451 /* Enable logging to console */
1452 case SYSLOG_ACTION_CONSOLE_ON:
1453 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1454 console_loglevel = saved_console_loglevel;
1455 saved_console_loglevel = LOGLEVEL_DEFAULT;
1458 /* Set level of messages printed to console */
1459 case SYSLOG_ACTION_CONSOLE_LEVEL:
1461 if (len < 1 || len > 8)
1463 if (len < minimum_console_loglevel)
1464 len = minimum_console_loglevel;
1465 console_loglevel = len;
1466 /* Implicitly re-enable logging to console */
1467 saved_console_loglevel = LOGLEVEL_DEFAULT;
1470 /* Number of chars in the log buffer */
1471 case SYSLOG_ACTION_SIZE_UNREAD:
1472 raw_spin_lock_irq(&logbuf_lock);
1473 if (syslog_seq < log_first_seq) {
1474 /* messages are gone, move to first one */
1475 syslog_seq = log_first_seq;
1476 syslog_idx = log_first_idx;
1479 if (source == SYSLOG_FROM_PROC) {
1481 * Short-cut for poll(/"proc/kmsg") which simply checks
1482 * for pending data, not the size; return the count of
1483 * records, not the length.
1485 error = log_next_seq - syslog_seq;
1487 u64 seq = syslog_seq;
1488 u32 idx = syslog_idx;
1491 while (seq < log_next_seq) {
1492 struct printk_log *msg = log_from_idx(idx);
1494 error += msg_print_text(msg, true, NULL, 0);
1495 idx = log_next(idx);
1498 error -= syslog_partial;
1500 raw_spin_unlock_irq(&logbuf_lock);
1502 /* Size of the log buffer */
1503 case SYSLOG_ACTION_SIZE_BUFFER:
1504 error = log_buf_len;
1514 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1516 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1520 * Call the console drivers, asking them to write out
1521 * log_buf[start] to log_buf[end - 1].
1522 * The console_lock must be held.
1524 static void call_console_drivers(int level,
1525 const char *ext_text, size_t ext_len,
1526 const char *text, size_t len)
1528 struct console *con;
1530 trace_console_rcuidle(text, len);
1532 if (!console_drivers)
1535 for_each_console(con) {
1536 if (exclusive_console && con != exclusive_console)
1538 if (!(con->flags & CON_ENABLED))
1542 if (!cpu_online(smp_processor_id()) &&
1543 !(con->flags & CON_ANYTIME))
1545 if (con->flags & CON_EXTENDED)
1546 con->write(con, ext_text, ext_len);
1548 con->write(con, text, len);
1553 * Zap console related locks when oopsing.
1554 * To leave time for slow consoles to print a full oops,
1555 * only zap at most once every 30 seconds.
1557 static void zap_locks(void)
1559 static unsigned long oops_timestamp;
1561 if (time_after_eq(jiffies, oops_timestamp) &&
1562 !time_after(jiffies, oops_timestamp + 30 * HZ))
1565 oops_timestamp = jiffies;
1568 /* If a crash is occurring, make sure we can't deadlock */
1569 raw_spin_lock_init(&logbuf_lock);
1570 /* And make sure that we print immediately */
1571 sema_init(&console_sem, 1);
1574 int printk_delay_msec __read_mostly;
1576 static inline void printk_delay(void)
1578 if (unlikely(printk_delay_msec)) {
1579 int m = printk_delay_msec;
1583 touch_nmi_watchdog();
1589 * Continuation lines are buffered, and not committed to the record buffer
1590 * until the line is complete, or a race forces it. The line fragments
1591 * though, are printed immediately to the consoles to ensure everything has
1592 * reached the console in case of a kernel crash.
1594 static struct cont {
1595 char buf[LOG_LINE_MAX];
1596 size_t len; /* length == 0 means unused buffer */
1597 size_t cons; /* bytes written to console */
1598 struct task_struct *owner; /* task of first print*/
1599 u64 ts_nsec; /* time of first print */
1600 u8 level; /* log level of first message */
1601 u8 facility; /* log facility of first message */
1602 enum log_flags flags; /* prefix, newline flags */
1603 bool flushed:1; /* buffer sealed and committed */
1606 static void cont_flush(void)
1614 * If a fragment of this line was directly flushed to the
1615 * console; wait for the console to pick up the rest of the
1616 * line. LOG_NOCONS suppresses a duplicated output.
1618 log_store(cont.facility, cont.level, cont.flags | LOG_NOCONS,
1619 cont.ts_nsec, NULL, 0, cont.buf, cont.len);
1620 cont.flushed = true;
1623 * If no fragment of this line ever reached the console,
1624 * just submit it to the store and free the buffer.
1626 log_store(cont.facility, cont.level, cont.flags, 0,
1627 NULL, 0, cont.buf, cont.len);
1632 static bool cont_add(int facility, int level, enum log_flags flags, const char *text, size_t len)
1634 if (cont.len && cont.flushed)
1638 * If ext consoles are present, flush and skip in-kernel
1639 * continuation. See nr_ext_console_drivers definition. Also, if
1640 * the line gets too long, split it up in separate records.
1642 if (nr_ext_console_drivers || cont.len + len > sizeof(cont.buf)) {
1648 cont.facility = facility;
1650 cont.owner = current;
1651 cont.ts_nsec = local_clock();
1654 cont.flushed = false;
1657 memcpy(cont.buf + cont.len, text, len);
1660 // The original flags come from the first line,
1661 // but later continuations can add a newline.
1662 if (flags & LOG_NEWLINE) {
1663 cont.flags |= LOG_NEWLINE;
1667 if (cont.len > (sizeof(cont.buf) * 80) / 100)
1673 static size_t cont_print_text(char *text, size_t size)
1678 if (cont.cons == 0) {
1679 textlen += print_time(cont.ts_nsec, text);
1683 len = cont.len - cont.cons;
1687 memcpy(text + textlen, cont.buf + cont.cons, len);
1689 cont.cons = cont.len;
1693 if (cont.flags & LOG_NEWLINE)
1694 text[textlen++] = '\n';
1695 /* got everything, release buffer */
1701 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)
1704 * If an earlier line was buffered, and we're a continuation
1705 * write from the same process, try to add it to the buffer.
1708 if (cont.owner == current && (lflags & LOG_CONT)) {
1709 if (cont_add(facility, level, lflags, text, text_len))
1712 /* Otherwise, make sure it's flushed */
1716 /* Skip empty continuation lines that couldn't be added - they just flush */
1717 if (!text_len && (lflags & LOG_CONT))
1720 /* If it doesn't end in a newline, try to buffer the current line */
1721 if (!(lflags & LOG_NEWLINE)) {
1722 if (cont_add(facility, level, lflags, text, text_len))
1726 /* Store it in the record log */
1727 return log_store(facility, level, lflags, 0, dict, dictlen, text, text_len);
1730 asmlinkage int vprintk_emit(int facility, int level,
1731 const char *dict, size_t dictlen,
1732 const char *fmt, va_list args)
1734 static bool recursion_bug;
1735 static char textbuf[LOG_LINE_MAX];
1736 char *text = textbuf;
1737 size_t text_len = 0;
1738 enum log_flags lflags = 0;
1739 unsigned long flags;
1741 int printed_len = 0;
1742 int nmi_message_lost;
1743 bool in_sched = false;
1744 /* cpu currently holding logbuf_lock in this function */
1745 static unsigned int logbuf_cpu = UINT_MAX;
1747 if (level == LOGLEVEL_SCHED) {
1748 level = LOGLEVEL_DEFAULT;
1752 boot_delay_msec(level);
1755 local_irq_save(flags);
1756 this_cpu = smp_processor_id();
1759 * Ouch, printk recursed into itself!
1761 if (unlikely(logbuf_cpu == this_cpu)) {
1763 * If a crash is occurring during printk() on this CPU,
1764 * then try to get the crash message out but make sure
1765 * we can't deadlock. Otherwise just return to avoid the
1766 * recursion and return - but flag the recursion so that
1767 * it can be printed at the next appropriate moment:
1769 if (!oops_in_progress && !lockdep_recursing(current)) {
1770 recursion_bug = true;
1771 local_irq_restore(flags);
1778 /* This stops the holder of console_sem just where we want him */
1779 raw_spin_lock(&logbuf_lock);
1780 logbuf_cpu = this_cpu;
1782 if (unlikely(recursion_bug)) {
1783 static const char recursion_msg[] =
1784 "BUG: recent printk recursion!";
1786 recursion_bug = false;
1787 /* emit KERN_CRIT message */
1788 printed_len += log_store(0, 2, LOG_PREFIX|LOG_NEWLINE, 0,
1789 NULL, 0, recursion_msg,
1790 strlen(recursion_msg));
1793 nmi_message_lost = get_nmi_message_lost();
1794 if (unlikely(nmi_message_lost)) {
1795 text_len = scnprintf(textbuf, sizeof(textbuf),
1796 "BAD LUCK: lost %d message(s) from NMI context!",
1798 printed_len += log_store(0, 2, LOG_PREFIX|LOG_NEWLINE, 0,
1799 NULL, 0, textbuf, text_len);
1803 * The printf needs to come first; we need the syslog
1804 * prefix which might be passed-in as a parameter.
1806 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1808 /* mark and strip a trailing newline */
1809 if (text_len && text[text_len-1] == '\n') {
1811 lflags |= LOG_NEWLINE;
1814 /* strip kernel syslog prefix and extract log level or control flags */
1815 if (facility == 0) {
1818 while ((kern_level = printk_get_level(text)) != 0) {
1819 switch (kern_level) {
1821 if (level == LOGLEVEL_DEFAULT)
1822 level = kern_level - '0';
1824 case 'd': /* KERN_DEFAULT */
1825 lflags |= LOG_PREFIX;
1827 case 'c': /* KERN_CONT */
1836 if (level == LOGLEVEL_DEFAULT)
1837 level = default_message_loglevel;
1840 lflags |= LOG_PREFIX|LOG_NEWLINE;
1842 printed_len += log_output(facility, level, lflags, dict, dictlen, text, text_len);
1844 logbuf_cpu = UINT_MAX;
1845 raw_spin_unlock(&logbuf_lock);
1847 local_irq_restore(flags);
1849 /* If called from the scheduler, we can not call up(). */
1853 * Try to acquire and then immediately release the console
1854 * semaphore. The release will print out buffers and wake up
1855 * /dev/kmsg and syslog() users.
1857 if (console_trylock())
1864 EXPORT_SYMBOL(vprintk_emit);
1866 asmlinkage int vprintk(const char *fmt, va_list args)
1868 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1870 EXPORT_SYMBOL(vprintk);
1872 asmlinkage int printk_emit(int facility, int level,
1873 const char *dict, size_t dictlen,
1874 const char *fmt, ...)
1879 va_start(args, fmt);
1880 r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1885 EXPORT_SYMBOL(printk_emit);
1887 int vprintk_default(const char *fmt, va_list args)
1891 #ifdef CONFIG_KGDB_KDB
1892 if (unlikely(kdb_trap_printk)) {
1893 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
1897 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1901 EXPORT_SYMBOL_GPL(vprintk_default);
1904 * printk - print a kernel message
1905 * @fmt: format string
1907 * This is printk(). It can be called from any context. We want it to work.
1909 * We try to grab the console_lock. If we succeed, it's easy - we log the
1910 * output and call the console drivers. If we fail to get the semaphore, we
1911 * place the output into the log buffer and return. The current holder of
1912 * the console_sem will notice the new output in console_unlock(); and will
1913 * send it to the consoles before releasing the lock.
1915 * One effect of this deferred printing is that code which calls printk() and
1916 * then changes console_loglevel may break. This is because console_loglevel
1917 * is inspected when the actual printing occurs.
1922 * See the vsnprintf() documentation for format string extensions over C99.
1924 asmlinkage __visible int printk(const char *fmt, ...)
1929 va_start(args, fmt);
1930 r = vprintk_func(fmt, args);
1935 EXPORT_SYMBOL(printk);
1937 #else /* CONFIG_PRINTK */
1939 #define LOG_LINE_MAX 0
1940 #define PREFIX_MAX 0
1942 static u64 syslog_seq;
1943 static u32 syslog_idx;
1944 static u64 console_seq;
1945 static u32 console_idx;
1946 static u64 log_first_seq;
1947 static u32 log_first_idx;
1948 static u64 log_next_seq;
1949 static struct cont {
1955 static char *log_text(const struct printk_log *msg) { return NULL; }
1956 static char *log_dict(const struct printk_log *msg) { return NULL; }
1957 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
1958 static u32 log_next(u32 idx) { return 0; }
1959 static ssize_t msg_print_ext_header(char *buf, size_t size,
1960 struct printk_log *msg,
1961 u64 seq) { return 0; }
1962 static ssize_t msg_print_ext_body(char *buf, size_t size,
1963 char *dict, size_t dict_len,
1964 char *text, size_t text_len) { return 0; }
1965 static void call_console_drivers(int level,
1966 const char *ext_text, size_t ext_len,
1967 const char *text, size_t len) {}
1968 static size_t msg_print_text(const struct printk_log *msg,
1969 bool syslog, char *buf, size_t size) { return 0; }
1970 static size_t cont_print_text(char *text, size_t size) { return 0; }
1971 static bool suppress_message_printing(int level) { return false; }
1973 /* Still needs to be defined for users */
1974 DEFINE_PER_CPU(printk_func_t, printk_func);
1976 #endif /* CONFIG_PRINTK */
1978 #ifdef CONFIG_EARLY_PRINTK
1979 struct console *early_console;
1981 asmlinkage __visible void early_printk(const char *fmt, ...)
1991 n = vscnprintf(buf, sizeof(buf), fmt, ap);
1994 early_console->write(early_console, buf, n);
1998 static int __add_preferred_console(char *name, int idx, char *options,
2001 struct console_cmdline *c;
2005 * See if this tty is not yet registered, and
2006 * if we have a slot free.
2008 for (i = 0, c = console_cmdline;
2009 i < MAX_CMDLINECONSOLES && c->name[0];
2011 if (strcmp(c->name, name) == 0 && c->index == idx) {
2013 selected_console = i;
2017 if (i == MAX_CMDLINECONSOLES)
2020 selected_console = i;
2021 strlcpy(c->name, name, sizeof(c->name));
2022 c->options = options;
2023 braille_set_options(c, brl_options);
2029 * Set up a console. Called via do_early_param() in init/main.c
2030 * for each "console=" parameter in the boot command line.
2032 static int __init console_setup(char *str)
2034 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2035 char *s, *options, *brl_options = NULL;
2039 * console="" or console=null have been suggested as a way to
2040 * disable console output. Use ttynull that has been created
2041 * for exacly this purpose.
2043 if (str[0] == 0 || strcmp(str, "null") == 0) {
2044 __add_preferred_console("ttynull", 0, NULL, NULL);
2048 if (_braille_console_setup(&str, &brl_options))
2052 * Decode str into name, index, options.
2054 if (str[0] >= '0' && str[0] <= '9') {
2055 strcpy(buf, "ttyS");
2056 strncpy(buf + 4, str, sizeof(buf) - 5);
2058 strncpy(buf, str, sizeof(buf) - 1);
2060 buf[sizeof(buf) - 1] = 0;
2061 options = strchr(str, ',');
2065 if (!strcmp(str, "ttya"))
2066 strcpy(buf, "ttyS0");
2067 if (!strcmp(str, "ttyb"))
2068 strcpy(buf, "ttyS1");
2070 for (s = buf; *s; s++)
2071 if (isdigit(*s) || *s == ',')
2073 idx = simple_strtoul(s, NULL, 10);
2076 __add_preferred_console(buf, idx, options, brl_options);
2077 console_set_on_cmdline = 1;
2080 __setup("console=", console_setup);
2083 * add_preferred_console - add a device to the list of preferred consoles.
2084 * @name: device name
2085 * @idx: device index
2086 * @options: options for this console
2088 * The last preferred console added will be used for kernel messages
2089 * and stdin/out/err for init. Normally this is used by console_setup
2090 * above to handle user-supplied console arguments; however it can also
2091 * be used by arch-specific code either to override the user or more
2092 * commonly to provide a default console (ie from PROM variables) when
2093 * the user has not supplied one.
2095 int add_preferred_console(char *name, int idx, char *options)
2097 return __add_preferred_console(name, idx, options, NULL);
2100 bool console_suspend_enabled = true;
2101 EXPORT_SYMBOL(console_suspend_enabled);
2103 static int __init console_suspend_disable(char *str)
2105 console_suspend_enabled = false;
2108 __setup("no_console_suspend", console_suspend_disable);
2109 module_param_named(console_suspend, console_suspend_enabled,
2110 bool, S_IRUGO | S_IWUSR);
2111 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2112 " and hibernate operations");
2115 * suspend_console - suspend the console subsystem
2117 * This disables printk() while we go into suspend states
2119 void suspend_console(void)
2121 if (!console_suspend_enabled)
2123 printk("Suspending console(s) (use no_console_suspend to debug)\n");
2125 console_suspended = 1;
2129 void resume_console(void)
2131 if (!console_suspend_enabled)
2134 console_suspended = 0;
2139 * console_cpu_notify - print deferred console messages after CPU hotplug
2140 * @self: notifier struct
2141 * @action: CPU hotplug event
2144 * If printk() is called from a CPU that is not online yet, the messages
2145 * will be spooled but will not show up on the console. This function is
2146 * called when a new CPU comes online (or fails to come up), and ensures
2147 * that any such output gets printed.
2149 static int console_cpu_notify(struct notifier_block *self,
2150 unsigned long action, void *hcpu)
2155 case CPU_DOWN_FAILED:
2156 case CPU_UP_CANCELED:
2164 * console_lock - lock the console system for exclusive use.
2166 * Acquires a lock which guarantees that the caller has
2167 * exclusive access to the console system and the console_drivers list.
2169 * Can sleep, returns nothing.
2171 void console_lock(void)
2176 if (console_suspended)
2179 console_may_schedule = 1;
2181 EXPORT_SYMBOL(console_lock);
2184 * console_trylock - try to lock the console system for exclusive use.
2186 * Try to acquire a lock which guarantees that the caller has exclusive
2187 * access to the console system and the console_drivers list.
2189 * returns 1 on success, and 0 on failure to acquire the lock.
2191 int console_trylock(void)
2193 if (down_trylock_console_sem())
2195 if (console_suspended) {
2201 * When PREEMPT_COUNT disabled we can't reliably detect if it's
2202 * safe to schedule (e.g. calling printk while holding a spin_lock),
2203 * because preempt_disable()/preempt_enable() are just barriers there
2204 * and preempt_count() is always 0.
2206 * RCU read sections have a separate preemption counter when
2207 * PREEMPT_RCU enabled thus we must take extra care and check
2208 * rcu_preempt_depth(), otherwise RCU read sections modify
2211 console_may_schedule = !oops_in_progress &&
2213 !rcu_preempt_depth();
2216 EXPORT_SYMBOL(console_trylock);
2218 int is_console_locked(void)
2220 return console_locked;
2224 * Check if we have any console that is capable of printing while cpu is
2225 * booting or shutting down. Requires console_sem.
2227 static int have_callable_console(void)
2229 struct console *con;
2231 for_each_console(con)
2232 if ((con->flags & CON_ENABLED) &&
2233 (con->flags & CON_ANYTIME))
2240 * Can we actually use the console at this time on this cpu?
2242 * Console drivers may assume that per-cpu resources have been allocated. So
2243 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2244 * call them until this CPU is officially up.
2246 static inline int can_use_console(void)
2248 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2251 static void console_cont_flush(char *text, size_t size)
2253 unsigned long flags;
2256 raw_spin_lock_irqsave(&logbuf_lock, flags);
2261 if (suppress_message_printing(cont.level)) {
2262 cont.cons = cont.len;
2269 * We still queue earlier records, likely because the console was
2270 * busy. The earlier ones need to be printed before this one, we
2271 * did not flush any fragment so far, so just let it queue up.
2273 if (console_seq < log_next_seq && !cont.cons)
2276 len = cont_print_text(text, size);
2277 raw_spin_unlock(&logbuf_lock);
2278 stop_critical_timings();
2279 call_console_drivers(cont.level, NULL, 0, text, len);
2280 start_critical_timings();
2281 local_irq_restore(flags);
2284 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2288 * console_unlock - unlock the console system
2290 * Releases the console_lock which the caller holds on the console system
2291 * and the console driver list.
2293 * While the console_lock was held, console output may have been buffered
2294 * by printk(). If this is the case, console_unlock(); emits
2295 * the output prior to releasing the lock.
2297 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2299 * console_unlock(); may be called from any context.
2301 void console_unlock(void)
2303 static char ext_text[CONSOLE_EXT_LOG_MAX];
2304 static char text[LOG_LINE_MAX + PREFIX_MAX];
2305 static u64 seen_seq;
2306 unsigned long flags;
2307 bool wake_klogd = false;
2308 bool do_cond_resched, retry;
2310 if (console_suspended) {
2316 * Console drivers are called with interrupts disabled, so
2317 * @console_may_schedule should be cleared before; however, we may
2318 * end up dumping a lot of lines, for example, if called from
2319 * console registration path, and should invoke cond_resched()
2320 * between lines if allowable. Not doing so can cause a very long
2321 * scheduling stall on a slow console leading to RCU stall and
2322 * softlockup warnings which exacerbate the issue with more
2323 * messages practically incapacitating the system.
2325 * console_trylock() is not able to detect the preemptive
2326 * context reliably. Therefore the value must be stored before
2327 * and cleared after the the "again" goto label.
2329 do_cond_resched = console_may_schedule;
2331 console_may_schedule = 0;
2334 * We released the console_sem lock, so we need to recheck if
2335 * cpu is online and (if not) is there at least one CON_ANYTIME
2338 if (!can_use_console()) {
2344 /* flush buffered message fragment immediately to console */
2345 console_cont_flush(text, sizeof(text));
2348 struct printk_log *msg;
2353 raw_spin_lock_irqsave(&logbuf_lock, flags);
2354 if (seen_seq != log_next_seq) {
2356 seen_seq = log_next_seq;
2359 if (console_seq < log_first_seq) {
2360 len = sprintf(text, "** %u printk messages dropped ** ",
2361 (unsigned)(log_first_seq - console_seq));
2363 /* messages are gone, move to first one */
2364 console_seq = log_first_seq;
2365 console_idx = log_first_idx;
2370 if (console_seq == log_next_seq)
2373 msg = log_from_idx(console_idx);
2375 if ((msg->flags & LOG_NOCONS) ||
2376 suppress_message_printing(level)) {
2378 * Skip record we have buffered and already printed
2379 * directly to the console when we received it, and
2380 * record that has level above the console loglevel.
2382 console_idx = log_next(console_idx);
2385 * We will get here again when we register a new
2386 * CON_PRINTBUFFER console. Clear the flag so we
2387 * will properly dump everything later.
2389 msg->flags &= ~LOG_NOCONS;
2393 len += msg_print_text(msg, false, text + len, sizeof(text) - len);
2394 if (nr_ext_console_drivers) {
2395 ext_len = msg_print_ext_header(ext_text,
2398 ext_len += msg_print_ext_body(ext_text + ext_len,
2399 sizeof(ext_text) - ext_len,
2400 log_dict(msg), msg->dict_len,
2401 log_text(msg), msg->text_len);
2403 console_idx = log_next(console_idx);
2405 raw_spin_unlock(&logbuf_lock);
2407 stop_critical_timings(); /* don't trace print latency */
2408 call_console_drivers(level, ext_text, ext_len, text, len);
2409 start_critical_timings();
2410 local_irq_restore(flags);
2412 if (do_cond_resched)
2417 /* Release the exclusive_console once it is used */
2418 if (unlikely(exclusive_console))
2419 exclusive_console = NULL;
2421 raw_spin_unlock(&logbuf_lock);
2426 * Someone could have filled up the buffer again, so re-check if there's
2427 * something to flush. In case we cannot trylock the console_sem again,
2428 * there's a new owner and the console_unlock() from them will do the
2429 * flush, no worries.
2431 raw_spin_lock(&logbuf_lock);
2432 retry = console_seq != log_next_seq;
2433 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2435 if (retry && console_trylock())
2441 EXPORT_SYMBOL(console_unlock);
2444 * console_conditional_schedule - yield the CPU if required
2446 * If the console code is currently allowed to sleep, and
2447 * if this CPU should yield the CPU to another task, do
2450 * Must be called within console_lock();.
2452 void __sched console_conditional_schedule(void)
2454 if (console_may_schedule)
2457 EXPORT_SYMBOL(console_conditional_schedule);
2459 void console_unblank(void)
2464 * console_unblank can no longer be called in interrupt context unless
2465 * oops_in_progress is set to 1..
2467 if (oops_in_progress) {
2468 if (down_trylock_console_sem() != 0)
2474 console_may_schedule = 0;
2476 if ((c->flags & CON_ENABLED) && c->unblank)
2482 * console_flush_on_panic - flush console content on panic
2484 * Immediately output all pending messages no matter what.
2486 void console_flush_on_panic(void)
2489 * If someone else is holding the console lock, trylock will fail
2490 * and may_schedule may be set. Ignore and proceed to unlock so
2491 * that messages are flushed out. As this can be called from any
2492 * context and we don't want to get preempted while flushing,
2493 * ensure may_schedule is cleared.
2496 console_may_schedule = 0;
2501 * Return the console tty driver structure and its associated index
2503 struct tty_driver *console_device(int *index)
2506 struct tty_driver *driver = NULL;
2509 for_each_console(c) {
2512 driver = c->device(c, index);
2521 * Prevent further output on the passed console device so that (for example)
2522 * serial drivers can disable console output before suspending a port, and can
2523 * re-enable output afterwards.
2525 void console_stop(struct console *console)
2528 console->flags &= ~CON_ENABLED;
2531 EXPORT_SYMBOL(console_stop);
2533 void console_start(struct console *console)
2536 console->flags |= CON_ENABLED;
2539 EXPORT_SYMBOL(console_start);
2541 static int __read_mostly keep_bootcon;
2543 static int __init keep_bootcon_setup(char *str)
2546 pr_info("debug: skip boot console de-registration.\n");
2551 early_param("keep_bootcon", keep_bootcon_setup);
2554 * The console driver calls this routine during kernel initialization
2555 * to register the console printing procedure with printk() and to
2556 * print any messages that were printed by the kernel before the
2557 * console driver was initialized.
2559 * This can happen pretty early during the boot process (because of
2560 * early_printk) - sometimes before setup_arch() completes - be careful
2561 * of what kernel features are used - they may not be initialised yet.
2563 * There are two types of consoles - bootconsoles (early_printk) and
2564 * "real" consoles (everything which is not a bootconsole) which are
2565 * handled differently.
2566 * - Any number of bootconsoles can be registered at any time.
2567 * - As soon as a "real" console is registered, all bootconsoles
2568 * will be unregistered automatically.
2569 * - Once a "real" console is registered, any attempt to register a
2570 * bootconsoles will be rejected
2572 void register_console(struct console *newcon)
2575 unsigned long flags;
2576 struct console *bcon = NULL;
2577 struct console_cmdline *c;
2579 if (console_drivers)
2580 for_each_console(bcon)
2581 if (WARN(bcon == newcon,
2582 "console '%s%d' already registered\n",
2583 bcon->name, bcon->index))
2587 * before we register a new CON_BOOT console, make sure we don't
2588 * already have a valid console
2590 if (console_drivers && newcon->flags & CON_BOOT) {
2591 /* find the last or real console */
2592 for_each_console(bcon) {
2593 if (!(bcon->flags & CON_BOOT)) {
2594 pr_info("Too late to register bootconsole %s%d\n",
2595 newcon->name, newcon->index);
2601 if (console_drivers && console_drivers->flags & CON_BOOT)
2602 bcon = console_drivers;
2604 if (preferred_console < 0 || bcon || !console_drivers)
2605 preferred_console = selected_console;
2608 * See if we want to use this console driver. If we
2609 * didn't select a console we take the first one
2610 * that registers here.
2612 if (preferred_console < 0) {
2613 if (newcon->index < 0)
2615 if (newcon->setup == NULL ||
2616 newcon->setup(newcon, NULL) == 0) {
2617 newcon->flags |= CON_ENABLED;
2618 if (newcon->device) {
2619 newcon->flags |= CON_CONSDEV;
2620 preferred_console = 0;
2626 * See if this console matches one we selected on
2629 for (i = 0, c = console_cmdline;
2630 i < MAX_CMDLINECONSOLES && c->name[0];
2632 if (!newcon->match ||
2633 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2634 /* default matching */
2635 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2636 if (strcmp(c->name, newcon->name) != 0)
2638 if (newcon->index >= 0 &&
2639 newcon->index != c->index)
2641 if (newcon->index < 0)
2642 newcon->index = c->index;
2644 if (_braille_register_console(newcon, c))
2647 if (newcon->setup &&
2648 newcon->setup(newcon, c->options) != 0)
2652 newcon->flags |= CON_ENABLED;
2653 if (i == selected_console) {
2654 newcon->flags |= CON_CONSDEV;
2655 preferred_console = selected_console;
2660 if (!(newcon->flags & CON_ENABLED))
2664 * If we have a bootconsole, and are switching to a real console,
2665 * don't print everything out again, since when the boot console, and
2666 * the real console are the same physical device, it's annoying to
2667 * see the beginning boot messages twice
2669 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2670 newcon->flags &= ~CON_PRINTBUFFER;
2673 * Put this console in the list - keep the
2674 * preferred driver at the head of the list.
2677 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2678 newcon->next = console_drivers;
2679 console_drivers = newcon;
2681 newcon->next->flags &= ~CON_CONSDEV;
2683 newcon->next = console_drivers->next;
2684 console_drivers->next = newcon;
2687 if (newcon->flags & CON_EXTENDED)
2688 if (!nr_ext_console_drivers++)
2689 pr_info("printk: continuation disabled due to ext consoles, expect more fragments in /dev/kmsg\n");
2691 if (newcon->flags & CON_PRINTBUFFER) {
2693 * console_unlock(); will print out the buffered messages
2696 raw_spin_lock_irqsave(&logbuf_lock, flags);
2697 console_seq = syslog_seq;
2698 console_idx = syslog_idx;
2699 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2701 * We're about to replay the log buffer. Only do this to the
2702 * just-registered console to avoid excessive message spam to
2703 * the already-registered consoles.
2705 exclusive_console = newcon;
2708 console_sysfs_notify();
2711 * By unregistering the bootconsoles after we enable the real console
2712 * we get the "console xxx enabled" message on all the consoles -
2713 * boot consoles, real consoles, etc - this is to ensure that end
2714 * users know there might be something in the kernel's log buffer that
2715 * went to the bootconsole (that they do not see on the real console)
2717 pr_info("%sconsole [%s%d] enabled\n",
2718 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2719 newcon->name, newcon->index);
2721 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2723 /* We need to iterate through all boot consoles, to make
2724 * sure we print everything out, before we unregister them.
2726 for_each_console(bcon)
2727 if (bcon->flags & CON_BOOT)
2728 unregister_console(bcon);
2731 EXPORT_SYMBOL(register_console);
2733 int unregister_console(struct console *console)
2735 struct console *a, *b;
2738 pr_info("%sconsole [%s%d] disabled\n",
2739 (console->flags & CON_BOOT) ? "boot" : "" ,
2740 console->name, console->index);
2742 res = _braille_unregister_console(console);
2748 if (console_drivers == console) {
2749 console_drivers=console->next;
2751 } else if (console_drivers) {
2752 for (a=console_drivers->next, b=console_drivers ;
2753 a; b=a, a=b->next) {
2762 if (!res && (console->flags & CON_EXTENDED))
2763 nr_ext_console_drivers--;
2766 * If this isn't the last console and it has CON_CONSDEV set, we
2767 * need to set it on the next preferred console.
2769 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2770 console_drivers->flags |= CON_CONSDEV;
2772 console->flags &= ~CON_ENABLED;
2774 console_sysfs_notify();
2777 EXPORT_SYMBOL(unregister_console);
2780 * Some boot consoles access data that is in the init section and which will
2781 * be discarded after the initcalls have been run. To make sure that no code
2782 * will access this data, unregister the boot consoles in a late initcall.
2784 * If for some reason, such as deferred probe or the driver being a loadable
2785 * module, the real console hasn't registered yet at this point, there will
2786 * be a brief interval in which no messages are logged to the console, which
2787 * makes it difficult to diagnose problems that occur during this time.
2789 * To mitigate this problem somewhat, only unregister consoles whose memory
2790 * intersects with the init section. Note that code exists elsewhere to get
2791 * rid of the boot console as soon as the proper console shows up, so there
2792 * won't be side-effects from postponing the removal.
2794 static int __init printk_late_init(void)
2796 struct console *con;
2798 for_each_console(con) {
2799 if (!keep_bootcon && con->flags & CON_BOOT) {
2801 * Make sure to unregister boot consoles whose data
2802 * resides in the init section before the init section
2803 * is discarded. Boot consoles whose data will stick
2804 * around will automatically be unregistered when the
2805 * proper console replaces them.
2807 if (init_section_intersects(con, sizeof(*con)))
2808 unregister_console(con);
2811 hotcpu_notifier(console_cpu_notify, 0);
2814 late_initcall(printk_late_init);
2816 #if defined CONFIG_PRINTK
2818 * Delayed printk version, for scheduler-internal messages:
2820 #define PRINTK_PENDING_WAKEUP 0x01
2821 #define PRINTK_PENDING_OUTPUT 0x02
2823 static DEFINE_PER_CPU(int, printk_pending);
2825 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2827 int pending = __this_cpu_xchg(printk_pending, 0);
2829 if (pending & PRINTK_PENDING_OUTPUT) {
2830 /* If trylock fails, someone else is doing the printing */
2831 if (console_trylock())
2835 if (pending & PRINTK_PENDING_WAKEUP)
2836 wake_up_interruptible(&log_wait);
2839 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2840 .func = wake_up_klogd_work_func,
2841 .flags = IRQ_WORK_LAZY,
2844 void wake_up_klogd(void)
2847 if (waitqueue_active(&log_wait)) {
2848 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2849 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2854 int printk_deferred(const char *fmt, ...)
2860 va_start(args, fmt);
2861 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2864 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2865 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2872 * printk rate limiting, lifted from the networking subsystem.
2874 * This enforces a rate limit: not more than 10 kernel messages
2875 * every 5s to make a denial-of-service attack impossible.
2877 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2879 int __printk_ratelimit(const char *func)
2881 return ___ratelimit(&printk_ratelimit_state, func);
2883 EXPORT_SYMBOL(__printk_ratelimit);
2886 * printk_timed_ratelimit - caller-controlled printk ratelimiting
2887 * @caller_jiffies: pointer to caller's state
2888 * @interval_msecs: minimum interval between prints
2890 * printk_timed_ratelimit() returns true if more than @interval_msecs
2891 * milliseconds have elapsed since the last time printk_timed_ratelimit()
2894 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
2895 unsigned int interval_msecs)
2897 unsigned long elapsed = jiffies - *caller_jiffies;
2899 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
2902 *caller_jiffies = jiffies;
2905 EXPORT_SYMBOL(printk_timed_ratelimit);
2907 static DEFINE_SPINLOCK(dump_list_lock);
2908 static LIST_HEAD(dump_list);
2911 * kmsg_dump_register - register a kernel log dumper.
2912 * @dumper: pointer to the kmsg_dumper structure
2914 * Adds a kernel log dumper to the system. The dump callback in the
2915 * structure will be called when the kernel oopses or panics and must be
2916 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
2918 int kmsg_dump_register(struct kmsg_dumper *dumper)
2920 unsigned long flags;
2923 /* The dump callback needs to be set */
2927 spin_lock_irqsave(&dump_list_lock, flags);
2928 /* Don't allow registering multiple times */
2929 if (!dumper->registered) {
2930 dumper->registered = 1;
2931 list_add_tail_rcu(&dumper->list, &dump_list);
2934 spin_unlock_irqrestore(&dump_list_lock, flags);
2938 EXPORT_SYMBOL_GPL(kmsg_dump_register);
2941 * kmsg_dump_unregister - unregister a kmsg dumper.
2942 * @dumper: pointer to the kmsg_dumper structure
2944 * Removes a dump device from the system. Returns zero on success and
2945 * %-EINVAL otherwise.
2947 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
2949 unsigned long flags;
2952 spin_lock_irqsave(&dump_list_lock, flags);
2953 if (dumper->registered) {
2954 dumper->registered = 0;
2955 list_del_rcu(&dumper->list);
2958 spin_unlock_irqrestore(&dump_list_lock, flags);
2963 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
2965 static bool always_kmsg_dump;
2966 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
2969 * kmsg_dump - dump kernel log to kernel message dumpers.
2970 * @reason: the reason (oops, panic etc) for dumping
2972 * Call each of the registered dumper's dump() callback, which can
2973 * retrieve the kmsg records with kmsg_dump_get_line() or
2974 * kmsg_dump_get_buffer().
2976 void kmsg_dump(enum kmsg_dump_reason reason)
2978 struct kmsg_dumper *dumper;
2979 unsigned long flags;
2981 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
2985 list_for_each_entry_rcu(dumper, &dump_list, list) {
2986 if (dumper->max_reason && reason > dumper->max_reason)
2989 /* initialize iterator with data about the stored records */
2990 dumper->active = true;
2992 raw_spin_lock_irqsave(&logbuf_lock, flags);
2993 dumper->cur_seq = clear_seq;
2994 dumper->cur_idx = clear_idx;
2995 dumper->next_seq = log_next_seq;
2996 dumper->next_idx = log_next_idx;
2997 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2999 /* invoke dumper which will iterate over records */
3000 dumper->dump(dumper, reason);
3002 /* reset iterator */
3003 dumper->active = false;
3009 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3010 * @dumper: registered kmsg dumper
3011 * @syslog: include the "<4>" prefixes
3012 * @line: buffer to copy the line to
3013 * @size: maximum size of the buffer
3014 * @len: length of line placed into buffer
3016 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3017 * record, and copy one record into the provided buffer.
3019 * Consecutive calls will return the next available record moving
3020 * towards the end of the buffer with the youngest messages.
3022 * A return value of FALSE indicates that there are no more records to
3025 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3027 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3028 char *line, size_t size, size_t *len)
3030 struct printk_log *msg;
3034 if (!dumper->active)
3037 if (dumper->cur_seq < log_first_seq) {
3038 /* messages are gone, move to first available one */
3039 dumper->cur_seq = log_first_seq;
3040 dumper->cur_idx = log_first_idx;
3044 if (dumper->cur_seq >= log_next_seq)
3047 msg = log_from_idx(dumper->cur_idx);
3048 l = msg_print_text(msg, syslog, line, size);
3050 dumper->cur_idx = log_next(dumper->cur_idx);
3060 * kmsg_dump_get_line - retrieve one kmsg log line
3061 * @dumper: registered kmsg dumper
3062 * @syslog: include the "<4>" prefixes
3063 * @line: buffer to copy the line to
3064 * @size: maximum size of the buffer
3065 * @len: length of line placed into buffer
3067 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3068 * record, and copy one record into the provided buffer.
3070 * Consecutive calls will return the next available record moving
3071 * towards the end of the buffer with the youngest messages.
3073 * A return value of FALSE indicates that there are no more records to
3076 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3077 char *line, size_t size, size_t *len)
3079 unsigned long flags;
3082 raw_spin_lock_irqsave(&logbuf_lock, flags);
3083 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3084 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3088 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3091 * kmsg_dump_get_buffer - copy kmsg log lines
3092 * @dumper: registered kmsg dumper
3093 * @syslog: include the "<4>" prefixes
3094 * @buf: buffer to copy the line to
3095 * @size: maximum size of the buffer
3096 * @len: length of line placed into buffer
3098 * Start at the end of the kmsg buffer and fill the provided buffer
3099 * with as many of the the *youngest* kmsg records that fit into it.
3100 * If the buffer is large enough, all available kmsg records will be
3101 * copied with a single call.
3103 * Consecutive calls will fill the buffer with the next block of
3104 * available older records, not including the earlier retrieved ones.
3106 * A return value of FALSE indicates that there are no more records to
3109 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3110 char *buf, size_t size, size_t *len)
3112 unsigned long flags;
3120 if (!dumper->active)
3123 raw_spin_lock_irqsave(&logbuf_lock, flags);
3124 if (dumper->cur_seq < log_first_seq) {
3125 /* messages are gone, move to first available one */
3126 dumper->cur_seq = log_first_seq;
3127 dumper->cur_idx = log_first_idx;
3131 if (dumper->cur_seq >= dumper->next_seq) {
3132 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3136 /* calculate length of entire buffer */
3137 seq = dumper->cur_seq;
3138 idx = dumper->cur_idx;
3139 while (seq < dumper->next_seq) {
3140 struct printk_log *msg = log_from_idx(idx);
3142 l += msg_print_text(msg, true, NULL, 0);
3143 idx = log_next(idx);
3147 /* move first record forward until length fits into the buffer */
3148 seq = dumper->cur_seq;
3149 idx = dumper->cur_idx;
3150 while (l >= size && seq < dumper->next_seq) {
3151 struct printk_log *msg = log_from_idx(idx);
3153 l -= msg_print_text(msg, true, NULL, 0);
3154 idx = log_next(idx);
3158 /* last message in next interation */
3163 while (seq < dumper->next_seq) {
3164 struct printk_log *msg = log_from_idx(idx);
3166 l += msg_print_text(msg, syslog, buf + l, size - l);
3167 idx = log_next(idx);
3171 dumper->next_seq = next_seq;
3172 dumper->next_idx = next_idx;
3174 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3180 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3183 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3184 * @dumper: registered kmsg dumper
3186 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3187 * kmsg_dump_get_buffer() can be called again and used multiple
3188 * times within the same dumper.dump() callback.
3190 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3192 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3194 dumper->cur_seq = clear_seq;
3195 dumper->cur_idx = clear_idx;
3196 dumper->next_seq = log_next_seq;
3197 dumper->next_idx = log_next_idx;
3201 * kmsg_dump_rewind - reset the interator
3202 * @dumper: registered kmsg dumper
3204 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3205 * kmsg_dump_get_buffer() can be called again and used multiple
3206 * times within the same dumper.dump() callback.
3208 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3210 unsigned long flags;
3212 raw_spin_lock_irqsave(&logbuf_lock, flags);
3213 kmsg_dump_rewind_nolock(dumper);
3214 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3216 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3218 static char dump_stack_arch_desc_str[128];
3221 * dump_stack_set_arch_desc - set arch-specific str to show with task dumps
3222 * @fmt: printf-style format string
3223 * @...: arguments for the format string
3225 * The configured string will be printed right after utsname during task
3226 * dumps. Usually used to add arch-specific system identifiers. If an
3227 * arch wants to make use of such an ID string, it should initialize this
3228 * as soon as possible during boot.
3230 void __init dump_stack_set_arch_desc(const char *fmt, ...)
3234 va_start(args, fmt);
3235 vsnprintf(dump_stack_arch_desc_str, sizeof(dump_stack_arch_desc_str),
3241 * dump_stack_print_info - print generic debug info for dump_stack()
3242 * @log_lvl: log level
3244 * Arch-specific dump_stack() implementations can use this function to
3245 * print out the same debug information as the generic dump_stack().
3247 void dump_stack_print_info(const char *log_lvl)
3249 printk("%sCPU: %d PID: %d Comm: %.20s %s %s %.*s\n",
3250 log_lvl, raw_smp_processor_id(), current->pid, current->comm,
3251 print_tainted(), init_utsname()->release,
3252 (int)strcspn(init_utsname()->version, " "),
3253 init_utsname()->version);
3255 if (dump_stack_arch_desc_str[0] != '\0')
3256 printk("%sHardware name: %s\n",
3257 log_lvl, dump_stack_arch_desc_str);
3259 print_worker_info(log_lvl, current);
3263 * show_regs_print_info - print generic debug info for show_regs()
3264 * @log_lvl: log level
3266 * show_regs() implementations can use this function to print out generic
3267 * debug information.
3269 void show_regs_print_info(const char *log_lvl)
3271 dump_stack_print_info(log_lvl);
3273 printk("%stask: %p task.stack: %p\n",
3274 log_lvl, current, task_stack_page(current));