split dev_queue
[cor.git] / kernel / printk / printk.c
blob1ef6f75d92f1f6c186fd1187786023a996a7fea0
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * linux/kernel/printk.c
5 * Copyright (C) 1991, 1992 Linus Torvalds
7 * Modified to make sys_syslog() more flexible: added commands to
8 * return the last 4k of kernel messages, regardless of whether
9 * they've been read or not. Added option to suppress kernel printk's
10 * to the console. Added hook for sending the console messages
11 * elsewhere, in preparation for a serial line console (someday).
12 * Ted Ts'o, 2/11/93.
13 * Modified for sysctl support, 1/8/97, Chris Horn.
14 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
15 * manfred@colorfullife.com
16 * Rewrote bits to get rid of console_lock
17 * 01Mar01 Andrew Morton
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22 #include <linux/kernel.h>
23 #include <linux/mm.h>
24 #include <linux/tty.h>
25 #include <linux/tty_driver.h>
26 #include <linux/console.h>
27 #include <linux/init.h>
28 #include <linux/jiffies.h>
29 #include <linux/nmi.h>
30 #include <linux/module.h>
31 #include <linux/moduleparam.h>
32 #include <linux/delay.h>
33 #include <linux/smp.h>
34 #include <linux/security.h>
35 #include <linux/memblock.h>
36 #include <linux/syscalls.h>
37 #include <linux/crash_core.h>
38 #include <linux/kdb.h>
39 #include <linux/ratelimit.h>
40 #include <linux/kmsg_dump.h>
41 #include <linux/syslog.h>
42 #include <linux/cpu.h>
43 #include <linux/rculist.h>
44 #include <linux/poll.h>
45 #include <linux/irq_work.h>
46 #include <linux/ctype.h>
47 #include <linux/uio.h>
48 #include <linux/sched/clock.h>
49 #include <linux/sched/debug.h>
50 #include <linux/sched/task_stack.h>
52 #include <linux/uaccess.h>
53 #include <asm/sections.h>
55 #include <trace/events/initcall.h>
56 #define CREATE_TRACE_POINTS
57 #include <trace/events/printk.h>
59 #include "console_cmdline.h"
60 #include "braille.h"
61 #include "internal.h"
63 int console_printk[4] = {
64 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
65 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
66 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
67 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
69 EXPORT_SYMBOL_GPL(console_printk);
71 atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
72 EXPORT_SYMBOL(ignore_console_lock_warning);
75 * Low level drivers may need that to know if they can schedule in
76 * their unblank() callback or not. So let's export it.
78 int oops_in_progress;
79 EXPORT_SYMBOL(oops_in_progress);
82 * console_sem protects the console_drivers list, and also
83 * provides serialisation for access to the entire console
84 * driver system.
86 static DEFINE_SEMAPHORE(console_sem);
87 struct console *console_drivers;
88 EXPORT_SYMBOL_GPL(console_drivers);
91 * System may need to suppress printk message under certain
92 * circumstances, like after kernel panic happens.
94 int __read_mostly suppress_printk;
96 #ifdef CONFIG_LOCKDEP
97 static struct lockdep_map console_lock_dep_map = {
98 .name = "console_lock"
100 #endif
102 enum devkmsg_log_bits {
103 __DEVKMSG_LOG_BIT_ON = 0,
104 __DEVKMSG_LOG_BIT_OFF,
105 __DEVKMSG_LOG_BIT_LOCK,
108 enum devkmsg_log_masks {
109 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
110 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
111 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
114 /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
115 #define DEVKMSG_LOG_MASK_DEFAULT 0
117 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
119 static int __control_devkmsg(char *str)
121 size_t len;
123 if (!str)
124 return -EINVAL;
126 len = str_has_prefix(str, "on");
127 if (len) {
128 devkmsg_log = DEVKMSG_LOG_MASK_ON;
129 return len;
132 len = str_has_prefix(str, "off");
133 if (len) {
134 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
135 return len;
138 len = str_has_prefix(str, "ratelimit");
139 if (len) {
140 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
141 return len;
144 return -EINVAL;
147 static int __init control_devkmsg(char *str)
149 if (__control_devkmsg(str) < 0)
150 return 1;
153 * Set sysctl string accordingly:
155 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
156 strcpy(devkmsg_log_str, "on");
157 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
158 strcpy(devkmsg_log_str, "off");
159 /* else "ratelimit" which is set by default. */
162 * Sysctl cannot change it anymore. The kernel command line setting of
163 * this parameter is to force the setting to be permanent throughout the
164 * runtime of the system. This is a precation measure against userspace
165 * trying to be a smarta** and attempting to change it up on us.
167 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
169 return 0;
171 __setup("printk.devkmsg=", control_devkmsg);
173 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
175 int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
176 void __user *buffer, size_t *lenp, loff_t *ppos)
178 char old_str[DEVKMSG_STR_MAX_SIZE];
179 unsigned int old;
180 int err;
182 if (write) {
183 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
184 return -EINVAL;
186 old = devkmsg_log;
187 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
190 err = proc_dostring(table, write, buffer, lenp, ppos);
191 if (err)
192 return err;
194 if (write) {
195 err = __control_devkmsg(devkmsg_log_str);
198 * Do not accept an unknown string OR a known string with
199 * trailing crap...
201 if (err < 0 || (err + 1 != *lenp)) {
203 /* ... and restore old setting. */
204 devkmsg_log = old;
205 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
207 return -EINVAL;
211 return 0;
214 /* Number of registered extended console drivers. */
215 static int nr_ext_console_drivers;
218 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
219 * macros instead of functions so that _RET_IP_ contains useful information.
221 #define down_console_sem() do { \
222 down(&console_sem);\
223 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
224 } while (0)
226 static int __down_trylock_console_sem(unsigned long ip)
228 int lock_failed;
229 unsigned long flags;
232 * Here and in __up_console_sem() we need to be in safe mode,
233 * because spindump/WARN/etc from under console ->lock will
234 * deadlock in printk()->down_trylock_console_sem() otherwise.
236 printk_safe_enter_irqsave(flags);
237 lock_failed = down_trylock(&console_sem);
238 printk_safe_exit_irqrestore(flags);
240 if (lock_failed)
241 return 1;
242 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
243 return 0;
245 #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
247 static void __up_console_sem(unsigned long ip)
249 unsigned long flags;
251 mutex_release(&console_lock_dep_map, ip);
253 printk_safe_enter_irqsave(flags);
254 up(&console_sem);
255 printk_safe_exit_irqrestore(flags);
257 #define up_console_sem() __up_console_sem(_RET_IP_)
260 * This is used for debugging the mess that is the VT code by
261 * keeping track if we have the console semaphore held. It's
262 * definitely not the perfect debug tool (we don't know if _WE_
263 * hold it and are racing, but it helps tracking those weird code
264 * paths in the console code where we end up in places I want
265 * locked without the console sempahore held).
267 static int console_locked, console_suspended;
270 * If exclusive_console is non-NULL then only this console is to be printed to.
272 static struct console *exclusive_console;
275 * Array of consoles built from command line options (console=)
278 #define MAX_CMDLINECONSOLES 8
280 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
282 static int preferred_console = -1;
283 int console_set_on_cmdline;
284 EXPORT_SYMBOL(console_set_on_cmdline);
286 /* Flag: console code may call schedule() */
287 static int console_may_schedule;
289 enum con_msg_format_flags {
290 MSG_FORMAT_DEFAULT = 0,
291 MSG_FORMAT_SYSLOG = (1 << 0),
294 static int console_msg_format = MSG_FORMAT_DEFAULT;
297 * The printk log buffer consists of a chain of concatenated variable
298 * length records. Every record starts with a record header, containing
299 * the overall length of the record.
301 * The heads to the first and last entry in the buffer, as well as the
302 * sequence numbers of these entries are maintained when messages are
303 * stored.
305 * If the heads indicate available messages, the length in the header
306 * tells the start next message. A length == 0 for the next message
307 * indicates a wrap-around to the beginning of the buffer.
309 * Every record carries the monotonic timestamp in microseconds, as well as
310 * the standard userspace syslog level and syslog facility. The usual
311 * kernel messages use LOG_KERN; userspace-injected messages always carry
312 * a matching syslog facility, by default LOG_USER. The origin of every
313 * message can be reliably determined that way.
315 * The human readable log message directly follows the message header. The
316 * length of the message text is stored in the header, the stored message
317 * is not terminated.
319 * Optionally, a message can carry a dictionary of properties (key/value pairs),
320 * to provide userspace with a machine-readable message context.
322 * Examples for well-defined, commonly used property names are:
323 * DEVICE=b12:8 device identifier
324 * b12:8 block dev_t
325 * c127:3 char dev_t
326 * n8 netdev ifindex
327 * +sound:card0 subsystem:devname
328 * SUBSYSTEM=pci driver-core subsystem name
330 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
331 * follows directly after a '=' character. Every property is terminated by
332 * a '\0' character. The last property is not terminated.
334 * Example of a message structure:
335 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
336 * 0008 34 00 record is 52 bytes long
337 * 000a 0b 00 text is 11 bytes long
338 * 000c 1f 00 dictionary is 23 bytes long
339 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
340 * 0010 69 74 27 73 20 61 20 6c "it's a l"
341 * 69 6e 65 "ine"
342 * 001b 44 45 56 49 43 "DEVIC"
343 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
344 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
345 * 67 "g"
346 * 0032 00 00 00 padding to next message header
348 * The 'struct printk_log' buffer header must never be directly exported to
349 * userspace, it is a kernel-private implementation detail that might
350 * need to be changed in the future, when the requirements change.
352 * /dev/kmsg exports the structured data in the following line format:
353 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
355 * Users of the export format should ignore possible additional values
356 * separated by ',', and find the message after the ';' character.
358 * The optional key/value pairs are attached as continuation lines starting
359 * with a space character and terminated by a newline. All possible
360 * non-prinatable characters are escaped in the "\xff" notation.
363 enum log_flags {
364 LOG_NEWLINE = 2, /* text ended with a newline */
365 LOG_CONT = 8, /* text is a fragment of a continuation line */
368 struct printk_log {
369 u64 ts_nsec; /* timestamp in nanoseconds */
370 u16 len; /* length of entire record */
371 u16 text_len; /* length of text buffer */
372 u16 dict_len; /* length of dictionary buffer */
373 u8 facility; /* syslog facility */
374 u8 flags:5; /* internal record flags */
375 u8 level:3; /* syslog level */
376 #ifdef CONFIG_PRINTK_CALLER
377 u32 caller_id; /* thread id or processor id */
378 #endif
380 #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
381 __packed __aligned(4)
382 #endif
386 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
387 * within the scheduler's rq lock. It must be released before calling
388 * console_unlock() or anything else that might wake up a process.
390 DEFINE_RAW_SPINLOCK(logbuf_lock);
393 * Helper macros to lock/unlock logbuf_lock and switch between
394 * printk-safe/unsafe modes.
396 #define logbuf_lock_irq() \
397 do { \
398 printk_safe_enter_irq(); \
399 raw_spin_lock(&logbuf_lock); \
400 } while (0)
402 #define logbuf_unlock_irq() \
403 do { \
404 raw_spin_unlock(&logbuf_lock); \
405 printk_safe_exit_irq(); \
406 } while (0)
408 #define logbuf_lock_irqsave(flags) \
409 do { \
410 printk_safe_enter_irqsave(flags); \
411 raw_spin_lock(&logbuf_lock); \
412 } while (0)
414 #define logbuf_unlock_irqrestore(flags) \
415 do { \
416 raw_spin_unlock(&logbuf_lock); \
417 printk_safe_exit_irqrestore(flags); \
418 } while (0)
420 #ifdef CONFIG_PRINTK
421 DECLARE_WAIT_QUEUE_HEAD(log_wait);
422 /* the next printk record to read by syslog(READ) or /proc/kmsg */
423 static u64 syslog_seq;
424 static u32 syslog_idx;
425 static size_t syslog_partial;
426 static bool syslog_time;
428 /* index and sequence number of the first record stored in the buffer */
429 static u64 log_first_seq;
430 static u32 log_first_idx;
432 /* index and sequence number of the next record to store in the buffer */
433 static u64 log_next_seq;
434 static u32 log_next_idx;
436 /* the next printk record to write to the console */
437 static u64 console_seq;
438 static u32 console_idx;
439 static u64 exclusive_console_stop_seq;
441 /* the next printk record to read after the last 'clear' command */
442 static u64 clear_seq;
443 static u32 clear_idx;
445 #ifdef CONFIG_PRINTK_CALLER
446 #define PREFIX_MAX 48
447 #else
448 #define PREFIX_MAX 32
449 #endif
450 #define LOG_LINE_MAX (1024 - PREFIX_MAX)
452 #define LOG_LEVEL(v) ((v) & 0x07)
453 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
455 /* record buffer */
456 #define LOG_ALIGN __alignof__(struct printk_log)
457 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
458 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
459 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
460 static char *log_buf = __log_buf;
461 static u32 log_buf_len = __LOG_BUF_LEN;
463 /* Return log buffer address */
464 char *log_buf_addr_get(void)
466 return log_buf;
469 /* Return log buffer size */
470 u32 log_buf_len_get(void)
472 return log_buf_len;
475 /* human readable text of the record */
476 static char *log_text(const struct printk_log *msg)
478 return (char *)msg + sizeof(struct printk_log);
481 /* optional key/value pair dictionary attached to the record */
482 static char *log_dict(const struct printk_log *msg)
484 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
487 /* get record by index; idx must point to valid msg */
488 static struct printk_log *log_from_idx(u32 idx)
490 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
493 * A length == 0 record is the end of buffer marker. Wrap around and
494 * read the message at the start of the buffer.
496 if (!msg->len)
497 return (struct printk_log *)log_buf;
498 return msg;
501 /* get next record; idx must point to valid msg */
502 static u32 log_next(u32 idx)
504 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
506 /* length == 0 indicates the end of the buffer; wrap */
508 * A length == 0 record is the end of buffer marker. Wrap around and
509 * read the message at the start of the buffer as *this* one, and
510 * return the one after that.
512 if (!msg->len) {
513 msg = (struct printk_log *)log_buf;
514 return msg->len;
516 return idx + msg->len;
520 * Check whether there is enough free space for the given message.
522 * The same values of first_idx and next_idx mean that the buffer
523 * is either empty or full.
525 * If the buffer is empty, we must respect the position of the indexes.
526 * They cannot be reset to the beginning of the buffer.
528 static int logbuf_has_space(u32 msg_size, bool empty)
530 u32 free;
532 if (log_next_idx > log_first_idx || empty)
533 free = max(log_buf_len - log_next_idx, log_first_idx);
534 else
535 free = log_first_idx - log_next_idx;
538 * We need space also for an empty header that signalizes wrapping
539 * of the buffer.
541 return free >= msg_size + sizeof(struct printk_log);
544 static int log_make_free_space(u32 msg_size)
546 while (log_first_seq < log_next_seq &&
547 !logbuf_has_space(msg_size, false)) {
548 /* drop old messages until we have enough contiguous space */
549 log_first_idx = log_next(log_first_idx);
550 log_first_seq++;
553 if (clear_seq < log_first_seq) {
554 clear_seq = log_first_seq;
555 clear_idx = log_first_idx;
558 /* sequence numbers are equal, so the log buffer is empty */
559 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
560 return 0;
562 return -ENOMEM;
565 /* compute the message size including the padding bytes */
566 static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
568 u32 size;
570 size = sizeof(struct printk_log) + text_len + dict_len;
571 *pad_len = (-size) & (LOG_ALIGN - 1);
572 size += *pad_len;
574 return size;
578 * Define how much of the log buffer we could take at maximum. The value
579 * must be greater than two. Note that only half of the buffer is available
580 * when the index points to the middle.
582 #define MAX_LOG_TAKE_PART 4
583 static const char trunc_msg[] = "<truncated>";
585 static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
586 u16 *dict_len, u32 *pad_len)
589 * The message should not take the whole buffer. Otherwise, it might
590 * get removed too soon.
592 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
593 if (*text_len > max_text_len)
594 *text_len = max_text_len;
595 /* enable the warning message */
596 *trunc_msg_len = strlen(trunc_msg);
597 /* disable the "dict" completely */
598 *dict_len = 0;
599 /* compute the size again, count also the warning message */
600 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
603 /* insert record into the buffer, discard old ones, update heads */
604 static int log_store(u32 caller_id, int facility, int level,
605 enum log_flags flags, u64 ts_nsec,
606 const char *dict, u16 dict_len,
607 const char *text, u16 text_len)
609 struct printk_log *msg;
610 u32 size, pad_len;
611 u16 trunc_msg_len = 0;
613 /* number of '\0' padding bytes to next message */
614 size = msg_used_size(text_len, dict_len, &pad_len);
616 if (log_make_free_space(size)) {
617 /* truncate the message if it is too long for empty buffer */
618 size = truncate_msg(&text_len, &trunc_msg_len,
619 &dict_len, &pad_len);
620 /* survive when the log buffer is too small for trunc_msg */
621 if (log_make_free_space(size))
622 return 0;
625 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
627 * This message + an additional empty header does not fit
628 * at the end of the buffer. Add an empty header with len == 0
629 * to signify a wrap around.
631 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
632 log_next_idx = 0;
635 /* fill message */
636 msg = (struct printk_log *)(log_buf + log_next_idx);
637 memcpy(log_text(msg), text, text_len);
638 msg->text_len = text_len;
639 if (trunc_msg_len) {
640 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
641 msg->text_len += trunc_msg_len;
643 memcpy(log_dict(msg), dict, dict_len);
644 msg->dict_len = dict_len;
645 msg->facility = facility;
646 msg->level = level & 7;
647 msg->flags = flags & 0x1f;
648 if (ts_nsec > 0)
649 msg->ts_nsec = ts_nsec;
650 else
651 msg->ts_nsec = local_clock();
652 #ifdef CONFIG_PRINTK_CALLER
653 msg->caller_id = caller_id;
654 #endif
655 memset(log_dict(msg) + dict_len, 0, pad_len);
656 msg->len = size;
658 /* insert message */
659 log_next_idx += msg->len;
660 log_next_seq++;
662 return msg->text_len;
665 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
667 static int syslog_action_restricted(int type)
669 if (dmesg_restrict)
670 return 1;
672 * Unless restricted, we allow "read all" and "get buffer size"
673 * for everybody.
675 return type != SYSLOG_ACTION_READ_ALL &&
676 type != SYSLOG_ACTION_SIZE_BUFFER;
679 static int check_syslog_permissions(int type, int source)
682 * If this is from /proc/kmsg and we've already opened it, then we've
683 * already done the capabilities checks at open time.
685 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
686 goto ok;
688 if (syslog_action_restricted(type)) {
689 if (capable(CAP_SYSLOG))
690 goto ok;
692 * For historical reasons, accept CAP_SYS_ADMIN too, with
693 * a warning.
695 if (capable(CAP_SYS_ADMIN)) {
696 pr_warn_once("%s (%d): Attempt to access syslog with "
697 "CAP_SYS_ADMIN but no CAP_SYSLOG "
698 "(deprecated).\n",
699 current->comm, task_pid_nr(current));
700 goto ok;
702 return -EPERM;
705 return security_syslog(type);
708 static void append_char(char **pp, char *e, char c)
710 if (*pp < e)
711 *(*pp)++ = c;
714 static ssize_t msg_print_ext_header(char *buf, size_t size,
715 struct printk_log *msg, u64 seq)
717 u64 ts_usec = msg->ts_nsec;
718 char caller[20];
719 #ifdef CONFIG_PRINTK_CALLER
720 u32 id = msg->caller_id;
722 snprintf(caller, sizeof(caller), ",caller=%c%u",
723 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
724 #else
725 caller[0] = '\0';
726 #endif
728 do_div(ts_usec, 1000);
730 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
731 (msg->facility << 3) | msg->level, seq, ts_usec,
732 msg->flags & LOG_CONT ? 'c' : '-', caller);
735 static ssize_t msg_print_ext_body(char *buf, size_t size,
736 char *dict, size_t dict_len,
737 char *text, size_t text_len)
739 char *p = buf, *e = buf + size;
740 size_t i;
742 /* escape non-printable characters */
743 for (i = 0; i < text_len; i++) {
744 unsigned char c = text[i];
746 if (c < ' ' || c >= 127 || c == '\\')
747 p += scnprintf(p, e - p, "\\x%02x", c);
748 else
749 append_char(&p, e, c);
751 append_char(&p, e, '\n');
753 if (dict_len) {
754 bool line = true;
756 for (i = 0; i < dict_len; i++) {
757 unsigned char c = dict[i];
759 if (line) {
760 append_char(&p, e, ' ');
761 line = false;
764 if (c == '\0') {
765 append_char(&p, e, '\n');
766 line = true;
767 continue;
770 if (c < ' ' || c >= 127 || c == '\\') {
771 p += scnprintf(p, e - p, "\\x%02x", c);
772 continue;
775 append_char(&p, e, c);
777 append_char(&p, e, '\n');
780 return p - buf;
783 /* /dev/kmsg - userspace message inject/listen interface */
784 struct devkmsg_user {
785 u64 seq;
786 u32 idx;
787 struct ratelimit_state rs;
788 struct mutex lock;
789 char buf[CONSOLE_EXT_LOG_MAX];
792 static __printf(3, 4) __cold
793 int devkmsg_emit(int facility, int level, const char *fmt, ...)
795 va_list args;
796 int r;
798 va_start(args, fmt);
799 r = vprintk_emit(facility, level, NULL, 0, fmt, args);
800 va_end(args);
802 return r;
805 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
807 char *buf, *line;
808 int level = default_message_loglevel;
809 int facility = 1; /* LOG_USER */
810 struct file *file = iocb->ki_filp;
811 struct devkmsg_user *user = file->private_data;
812 size_t len = iov_iter_count(from);
813 ssize_t ret = len;
815 if (!user || len > LOG_LINE_MAX)
816 return -EINVAL;
818 /* Ignore when user logging is disabled. */
819 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
820 return len;
822 /* Ratelimit when not explicitly enabled. */
823 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
824 if (!___ratelimit(&user->rs, current->comm))
825 return ret;
828 buf = kmalloc(len+1, GFP_KERNEL);
829 if (buf == NULL)
830 return -ENOMEM;
832 buf[len] = '\0';
833 if (!copy_from_iter_full(buf, len, from)) {
834 kfree(buf);
835 return -EFAULT;
839 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
840 * the decimal value represents 32bit, the lower 3 bit are the log
841 * level, the rest are the log facility.
843 * If no prefix or no userspace facility is specified, we
844 * enforce LOG_USER, to be able to reliably distinguish
845 * kernel-generated messages from userspace-injected ones.
847 line = buf;
848 if (line[0] == '<') {
849 char *endp = NULL;
850 unsigned int u;
852 u = simple_strtoul(line + 1, &endp, 10);
853 if (endp && endp[0] == '>') {
854 level = LOG_LEVEL(u);
855 if (LOG_FACILITY(u) != 0)
856 facility = LOG_FACILITY(u);
857 endp++;
858 len -= endp - line;
859 line = endp;
863 devkmsg_emit(facility, level, "%s", line);
864 kfree(buf);
865 return ret;
868 static ssize_t devkmsg_read(struct file *file, char __user *buf,
869 size_t count, loff_t *ppos)
871 struct devkmsg_user *user = file->private_data;
872 struct printk_log *msg;
873 size_t len;
874 ssize_t ret;
876 if (!user)
877 return -EBADF;
879 ret = mutex_lock_interruptible(&user->lock);
880 if (ret)
881 return ret;
883 logbuf_lock_irq();
884 while (user->seq == log_next_seq) {
885 if (file->f_flags & O_NONBLOCK) {
886 ret = -EAGAIN;
887 logbuf_unlock_irq();
888 goto out;
891 logbuf_unlock_irq();
892 ret = wait_event_interruptible(log_wait,
893 user->seq != log_next_seq);
894 if (ret)
895 goto out;
896 logbuf_lock_irq();
899 if (user->seq < log_first_seq) {
900 /* our last seen message is gone, return error and reset */
901 user->idx = log_first_idx;
902 user->seq = log_first_seq;
903 ret = -EPIPE;
904 logbuf_unlock_irq();
905 goto out;
908 msg = log_from_idx(user->idx);
909 len = msg_print_ext_header(user->buf, sizeof(user->buf),
910 msg, user->seq);
911 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
912 log_dict(msg), msg->dict_len,
913 log_text(msg), msg->text_len);
915 user->idx = log_next(user->idx);
916 user->seq++;
917 logbuf_unlock_irq();
919 if (len > count) {
920 ret = -EINVAL;
921 goto out;
924 if (copy_to_user(buf, user->buf, len)) {
925 ret = -EFAULT;
926 goto out;
928 ret = len;
929 out:
930 mutex_unlock(&user->lock);
931 return ret;
934 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
936 struct devkmsg_user *user = file->private_data;
937 loff_t ret = 0;
939 if (!user)
940 return -EBADF;
941 if (offset)
942 return -ESPIPE;
944 logbuf_lock_irq();
945 switch (whence) {
946 case SEEK_SET:
947 /* the first record */
948 user->idx = log_first_idx;
949 user->seq = log_first_seq;
950 break;
951 case SEEK_DATA:
953 * The first record after the last SYSLOG_ACTION_CLEAR,
954 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
955 * changes no global state, and does not clear anything.
957 user->idx = clear_idx;
958 user->seq = clear_seq;
959 break;
960 case SEEK_END:
961 /* after the last record */
962 user->idx = log_next_idx;
963 user->seq = log_next_seq;
964 break;
965 default:
966 ret = -EINVAL;
968 logbuf_unlock_irq();
969 return ret;
972 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
974 struct devkmsg_user *user = file->private_data;
975 __poll_t ret = 0;
977 if (!user)
978 return EPOLLERR|EPOLLNVAL;
980 poll_wait(file, &log_wait, wait);
982 logbuf_lock_irq();
983 if (user->seq < log_next_seq) {
984 /* return error when data has vanished underneath us */
985 if (user->seq < log_first_seq)
986 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
987 else
988 ret = EPOLLIN|EPOLLRDNORM;
990 logbuf_unlock_irq();
992 return ret;
995 static int devkmsg_open(struct inode *inode, struct file *file)
997 struct devkmsg_user *user;
998 int err;
1000 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
1001 return -EPERM;
1003 /* write-only does not need any file context */
1004 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
1005 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
1006 SYSLOG_FROM_READER);
1007 if (err)
1008 return err;
1011 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
1012 if (!user)
1013 return -ENOMEM;
1015 ratelimit_default_init(&user->rs);
1016 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
1018 mutex_init(&user->lock);
1020 logbuf_lock_irq();
1021 user->idx = log_first_idx;
1022 user->seq = log_first_seq;
1023 logbuf_unlock_irq();
1025 file->private_data = user;
1026 return 0;
1029 static int devkmsg_release(struct inode *inode, struct file *file)
1031 struct devkmsg_user *user = file->private_data;
1033 if (!user)
1034 return 0;
1036 ratelimit_state_exit(&user->rs);
1038 mutex_destroy(&user->lock);
1039 kfree(user);
1040 return 0;
1043 const struct file_operations kmsg_fops = {
1044 .open = devkmsg_open,
1045 .read = devkmsg_read,
1046 .write_iter = devkmsg_write,
1047 .llseek = devkmsg_llseek,
1048 .poll = devkmsg_poll,
1049 .release = devkmsg_release,
1052 #ifdef CONFIG_CRASH_CORE
1054 * This appends the listed symbols to /proc/vmcore
1056 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1057 * obtain access to symbols that are otherwise very difficult to locate. These
1058 * symbols are specifically used so that utilities can access and extract the
1059 * dmesg log from a vmcore file after a crash.
1061 void log_buf_vmcoreinfo_setup(void)
1063 VMCOREINFO_SYMBOL(log_buf);
1064 VMCOREINFO_SYMBOL(log_buf_len);
1065 VMCOREINFO_SYMBOL(log_first_idx);
1066 VMCOREINFO_SYMBOL(clear_idx);
1067 VMCOREINFO_SYMBOL(log_next_idx);
1069 * Export struct printk_log size and field offsets. User space tools can
1070 * parse it and detect any changes to structure down the line.
1072 VMCOREINFO_STRUCT_SIZE(printk_log);
1073 VMCOREINFO_OFFSET(printk_log, ts_nsec);
1074 VMCOREINFO_OFFSET(printk_log, len);
1075 VMCOREINFO_OFFSET(printk_log, text_len);
1076 VMCOREINFO_OFFSET(printk_log, dict_len);
1077 #ifdef CONFIG_PRINTK_CALLER
1078 VMCOREINFO_OFFSET(printk_log, caller_id);
1079 #endif
1081 #endif
1083 /* requested log_buf_len from kernel cmdline */
1084 static unsigned long __initdata new_log_buf_len;
1086 /* we practice scaling the ring buffer by powers of 2 */
1087 static void __init log_buf_len_update(u64 size)
1089 if (size > (u64)LOG_BUF_LEN_MAX) {
1090 size = (u64)LOG_BUF_LEN_MAX;
1091 pr_err("log_buf over 2G is not supported.\n");
1094 if (size)
1095 size = roundup_pow_of_two(size);
1096 if (size > log_buf_len)
1097 new_log_buf_len = (unsigned long)size;
1100 /* save requested log_buf_len since it's too early to process it */
1101 static int __init log_buf_len_setup(char *str)
1103 u64 size;
1105 if (!str)
1106 return -EINVAL;
1108 size = memparse(str, &str);
1110 log_buf_len_update(size);
1112 return 0;
1114 early_param("log_buf_len", log_buf_len_setup);
1116 #ifdef CONFIG_SMP
1117 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1119 static void __init log_buf_add_cpu(void)
1121 unsigned int cpu_extra;
1124 * archs should set up cpu_possible_bits properly with
1125 * set_cpu_possible() after setup_arch() but just in
1126 * case lets ensure this is valid.
1128 if (num_possible_cpus() == 1)
1129 return;
1131 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1133 /* by default this will only continue through for large > 64 CPUs */
1134 if (cpu_extra <= __LOG_BUF_LEN / 2)
1135 return;
1137 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1138 __LOG_CPU_MAX_BUF_LEN);
1139 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1140 cpu_extra);
1141 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1143 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1145 #else /* !CONFIG_SMP */
1146 static inline void log_buf_add_cpu(void) {}
1147 #endif /* CONFIG_SMP */
1149 void __init setup_log_buf(int early)
1151 unsigned long flags;
1152 char *new_log_buf;
1153 unsigned int free;
1155 if (log_buf != __log_buf)
1156 return;
1158 if (!early && !new_log_buf_len)
1159 log_buf_add_cpu();
1161 if (!new_log_buf_len)
1162 return;
1164 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1165 if (unlikely(!new_log_buf)) {
1166 pr_err("log_buf_len: %lu bytes not available\n",
1167 new_log_buf_len);
1168 return;
1171 logbuf_lock_irqsave(flags);
1172 log_buf_len = new_log_buf_len;
1173 log_buf = new_log_buf;
1174 new_log_buf_len = 0;
1175 free = __LOG_BUF_LEN - log_next_idx;
1176 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1177 logbuf_unlock_irqrestore(flags);
1179 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1180 pr_info("early log buf free: %u(%u%%)\n",
1181 free, (free * 100) / __LOG_BUF_LEN);
1184 static bool __read_mostly ignore_loglevel;
1186 static int __init ignore_loglevel_setup(char *str)
1188 ignore_loglevel = true;
1189 pr_info("debug: ignoring loglevel setting.\n");
1191 return 0;
1194 early_param("ignore_loglevel", ignore_loglevel_setup);
1195 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1196 MODULE_PARM_DESC(ignore_loglevel,
1197 "ignore loglevel setting (prints all kernel messages to the console)");
1199 static bool suppress_message_printing(int level)
1201 return (level >= console_loglevel && !ignore_loglevel);
1204 #ifdef CONFIG_BOOT_PRINTK_DELAY
1206 static int boot_delay; /* msecs delay after each printk during bootup */
1207 static unsigned long long loops_per_msec; /* based on boot_delay */
1209 static int __init boot_delay_setup(char *str)
1211 unsigned long lpj;
1213 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1214 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1216 get_option(&str, &boot_delay);
1217 if (boot_delay > 10 * 1000)
1218 boot_delay = 0;
1220 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1221 "HZ: %d, loops_per_msec: %llu\n",
1222 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1223 return 0;
1225 early_param("boot_delay", boot_delay_setup);
1227 static void boot_delay_msec(int level)
1229 unsigned long long k;
1230 unsigned long timeout;
1232 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1233 || suppress_message_printing(level)) {
1234 return;
1237 k = (unsigned long long)loops_per_msec * boot_delay;
1239 timeout = jiffies + msecs_to_jiffies(boot_delay);
1240 while (k) {
1241 k--;
1242 cpu_relax();
1244 * use (volatile) jiffies to prevent
1245 * compiler reduction; loop termination via jiffies
1246 * is secondary and may or may not happen.
1248 if (time_after(jiffies, timeout))
1249 break;
1250 touch_nmi_watchdog();
1253 #else
1254 static inline void boot_delay_msec(int level)
1257 #endif
1259 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1260 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1262 static size_t print_syslog(unsigned int level, char *buf)
1264 return sprintf(buf, "<%u>", level);
1267 static size_t print_time(u64 ts, char *buf)
1269 unsigned long rem_nsec = do_div(ts, 1000000000);
1271 return sprintf(buf, "[%5lu.%06lu]",
1272 (unsigned long)ts, rem_nsec / 1000);
1275 #ifdef CONFIG_PRINTK_CALLER
1276 static size_t print_caller(u32 id, char *buf)
1278 char caller[12];
1280 snprintf(caller, sizeof(caller), "%c%u",
1281 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1282 return sprintf(buf, "[%6s]", caller);
1284 #else
1285 #define print_caller(id, buf) 0
1286 #endif
1288 static size_t print_prefix(const struct printk_log *msg, bool syslog,
1289 bool time, char *buf)
1291 size_t len = 0;
1293 if (syslog)
1294 len = print_syslog((msg->facility << 3) | msg->level, buf);
1296 if (time)
1297 len += print_time(msg->ts_nsec, buf + len);
1299 len += print_caller(msg->caller_id, buf + len);
1301 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1302 buf[len++] = ' ';
1303 buf[len] = '\0';
1306 return len;
1309 static size_t msg_print_text(const struct printk_log *msg, bool syslog,
1310 bool time, char *buf, size_t size)
1312 const char *text = log_text(msg);
1313 size_t text_size = msg->text_len;
1314 size_t len = 0;
1315 char prefix[PREFIX_MAX];
1316 const size_t prefix_len = print_prefix(msg, syslog, time, prefix);
1318 do {
1319 const char *next = memchr(text, '\n', text_size);
1320 size_t text_len;
1322 if (next) {
1323 text_len = next - text;
1324 next++;
1325 text_size -= next - text;
1326 } else {
1327 text_len = text_size;
1330 if (buf) {
1331 if (prefix_len + text_len + 1 >= size - len)
1332 break;
1334 memcpy(buf + len, prefix, prefix_len);
1335 len += prefix_len;
1336 memcpy(buf + len, text, text_len);
1337 len += text_len;
1338 buf[len++] = '\n';
1339 } else {
1340 /* SYSLOG_ACTION_* buffer size only calculation */
1341 len += prefix_len + text_len + 1;
1344 text = next;
1345 } while (text);
1347 return len;
1350 static int syslog_print(char __user *buf, int size)
1352 char *text;
1353 struct printk_log *msg;
1354 int len = 0;
1356 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1357 if (!text)
1358 return -ENOMEM;
1360 while (size > 0) {
1361 size_t n;
1362 size_t skip;
1364 logbuf_lock_irq();
1365 if (syslog_seq < log_first_seq) {
1366 /* messages are gone, move to first one */
1367 syslog_seq = log_first_seq;
1368 syslog_idx = log_first_idx;
1369 syslog_partial = 0;
1371 if (syslog_seq == log_next_seq) {
1372 logbuf_unlock_irq();
1373 break;
1377 * To keep reading/counting partial line consistent,
1378 * use printk_time value as of the beginning of a line.
1380 if (!syslog_partial)
1381 syslog_time = printk_time;
1383 skip = syslog_partial;
1384 msg = log_from_idx(syslog_idx);
1385 n = msg_print_text(msg, true, syslog_time, text,
1386 LOG_LINE_MAX + PREFIX_MAX);
1387 if (n - syslog_partial <= size) {
1388 /* message fits into buffer, move forward */
1389 syslog_idx = log_next(syslog_idx);
1390 syslog_seq++;
1391 n -= syslog_partial;
1392 syslog_partial = 0;
1393 } else if (!len){
1394 /* partial read(), remember position */
1395 n = size;
1396 syslog_partial += n;
1397 } else
1398 n = 0;
1399 logbuf_unlock_irq();
1401 if (!n)
1402 break;
1404 if (copy_to_user(buf, text + skip, n)) {
1405 if (!len)
1406 len = -EFAULT;
1407 break;
1410 len += n;
1411 size -= n;
1412 buf += n;
1415 kfree(text);
1416 return len;
1419 static int syslog_print_all(char __user *buf, int size, bool clear)
1421 char *text;
1422 int len = 0;
1423 u64 next_seq;
1424 u64 seq;
1425 u32 idx;
1426 bool time;
1428 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1429 if (!text)
1430 return -ENOMEM;
1432 time = printk_time;
1433 logbuf_lock_irq();
1435 * Find first record that fits, including all following records,
1436 * into the user-provided buffer for this dump.
1438 seq = clear_seq;
1439 idx = clear_idx;
1440 while (seq < log_next_seq) {
1441 struct printk_log *msg = log_from_idx(idx);
1443 len += msg_print_text(msg, true, time, NULL, 0);
1444 idx = log_next(idx);
1445 seq++;
1448 /* move first record forward until length fits into the buffer */
1449 seq = clear_seq;
1450 idx = clear_idx;
1451 while (len > size && seq < log_next_seq) {
1452 struct printk_log *msg = log_from_idx(idx);
1454 len -= msg_print_text(msg, true, time, NULL, 0);
1455 idx = log_next(idx);
1456 seq++;
1459 /* last message fitting into this dump */
1460 next_seq = log_next_seq;
1462 len = 0;
1463 while (len >= 0 && seq < next_seq) {
1464 struct printk_log *msg = log_from_idx(idx);
1465 int textlen = msg_print_text(msg, true, time, text,
1466 LOG_LINE_MAX + PREFIX_MAX);
1468 idx = log_next(idx);
1469 seq++;
1471 logbuf_unlock_irq();
1472 if (copy_to_user(buf + len, text, textlen))
1473 len = -EFAULT;
1474 else
1475 len += textlen;
1476 logbuf_lock_irq();
1478 if (seq < log_first_seq) {
1479 /* messages are gone, move to next one */
1480 seq = log_first_seq;
1481 idx = log_first_idx;
1485 if (clear) {
1486 clear_seq = log_next_seq;
1487 clear_idx = log_next_idx;
1489 logbuf_unlock_irq();
1491 kfree(text);
1492 return len;
1495 static void syslog_clear(void)
1497 logbuf_lock_irq();
1498 clear_seq = log_next_seq;
1499 clear_idx = log_next_idx;
1500 logbuf_unlock_irq();
1503 int do_syslog(int type, char __user *buf, int len, int source)
1505 bool clear = false;
1506 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1507 int error;
1509 error = check_syslog_permissions(type, source);
1510 if (error)
1511 return error;
1513 switch (type) {
1514 case SYSLOG_ACTION_CLOSE: /* Close log */
1515 break;
1516 case SYSLOG_ACTION_OPEN: /* Open log */
1517 break;
1518 case SYSLOG_ACTION_READ: /* Read from log */
1519 if (!buf || len < 0)
1520 return -EINVAL;
1521 if (!len)
1522 return 0;
1523 if (!access_ok(buf, len))
1524 return -EFAULT;
1525 error = wait_event_interruptible(log_wait,
1526 syslog_seq != log_next_seq);
1527 if (error)
1528 return error;
1529 error = syslog_print(buf, len);
1530 break;
1531 /* Read/clear last kernel messages */
1532 case SYSLOG_ACTION_READ_CLEAR:
1533 clear = true;
1534 /* FALL THRU */
1535 /* Read last kernel messages */
1536 case SYSLOG_ACTION_READ_ALL:
1537 if (!buf || len < 0)
1538 return -EINVAL;
1539 if (!len)
1540 return 0;
1541 if (!access_ok(buf, len))
1542 return -EFAULT;
1543 error = syslog_print_all(buf, len, clear);
1544 break;
1545 /* Clear ring buffer */
1546 case SYSLOG_ACTION_CLEAR:
1547 syslog_clear();
1548 break;
1549 /* Disable logging to console */
1550 case SYSLOG_ACTION_CONSOLE_OFF:
1551 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1552 saved_console_loglevel = console_loglevel;
1553 console_loglevel = minimum_console_loglevel;
1554 break;
1555 /* Enable logging to console */
1556 case SYSLOG_ACTION_CONSOLE_ON:
1557 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1558 console_loglevel = saved_console_loglevel;
1559 saved_console_loglevel = LOGLEVEL_DEFAULT;
1561 break;
1562 /* Set level of messages printed to console */
1563 case SYSLOG_ACTION_CONSOLE_LEVEL:
1564 if (len < 1 || len > 8)
1565 return -EINVAL;
1566 if (len < minimum_console_loglevel)
1567 len = minimum_console_loglevel;
1568 console_loglevel = len;
1569 /* Implicitly re-enable logging to console */
1570 saved_console_loglevel = LOGLEVEL_DEFAULT;
1571 break;
1572 /* Number of chars in the log buffer */
1573 case SYSLOG_ACTION_SIZE_UNREAD:
1574 logbuf_lock_irq();
1575 if (syslog_seq < log_first_seq) {
1576 /* messages are gone, move to first one */
1577 syslog_seq = log_first_seq;
1578 syslog_idx = log_first_idx;
1579 syslog_partial = 0;
1581 if (source == SYSLOG_FROM_PROC) {
1583 * Short-cut for poll(/"proc/kmsg") which simply checks
1584 * for pending data, not the size; return the count of
1585 * records, not the length.
1587 error = log_next_seq - syslog_seq;
1588 } else {
1589 u64 seq = syslog_seq;
1590 u32 idx = syslog_idx;
1591 bool time = syslog_partial ? syslog_time : printk_time;
1593 while (seq < log_next_seq) {
1594 struct printk_log *msg = log_from_idx(idx);
1596 error += msg_print_text(msg, true, time, NULL,
1598 time = printk_time;
1599 idx = log_next(idx);
1600 seq++;
1602 error -= syslog_partial;
1604 logbuf_unlock_irq();
1605 break;
1606 /* Size of the log buffer */
1607 case SYSLOG_ACTION_SIZE_BUFFER:
1608 error = log_buf_len;
1609 break;
1610 default:
1611 error = -EINVAL;
1612 break;
1615 return error;
1618 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1620 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1624 * Special console_lock variants that help to reduce the risk of soft-lockups.
1625 * They allow to pass console_lock to another printk() call using a busy wait.
1628 #ifdef CONFIG_LOCKDEP
1629 static struct lockdep_map console_owner_dep_map = {
1630 .name = "console_owner"
1632 #endif
1634 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1635 static struct task_struct *console_owner;
1636 static bool console_waiter;
1639 * console_lock_spinning_enable - mark beginning of code where another
1640 * thread might safely busy wait
1642 * This basically converts console_lock into a spinlock. This marks
1643 * the section where the console_lock owner can not sleep, because
1644 * there may be a waiter spinning (like a spinlock). Also it must be
1645 * ready to hand over the lock at the end of the section.
1647 static void console_lock_spinning_enable(void)
1649 raw_spin_lock(&console_owner_lock);
1650 console_owner = current;
1651 raw_spin_unlock(&console_owner_lock);
1653 /* The waiter may spin on us after setting console_owner */
1654 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1658 * console_lock_spinning_disable_and_check - mark end of code where another
1659 * thread was able to busy wait and check if there is a waiter
1661 * This is called at the end of the section where spinning is allowed.
1662 * It has two functions. First, it is a signal that it is no longer
1663 * safe to start busy waiting for the lock. Second, it checks if
1664 * there is a busy waiter and passes the lock rights to her.
1666 * Important: Callers lose the lock if there was a busy waiter.
1667 * They must not touch items synchronized by console_lock
1668 * in this case.
1670 * Return: 1 if the lock rights were passed, 0 otherwise.
1672 static int console_lock_spinning_disable_and_check(void)
1674 int waiter;
1676 raw_spin_lock(&console_owner_lock);
1677 waiter = READ_ONCE(console_waiter);
1678 console_owner = NULL;
1679 raw_spin_unlock(&console_owner_lock);
1681 if (!waiter) {
1682 spin_release(&console_owner_dep_map, _THIS_IP_);
1683 return 0;
1686 /* The waiter is now free to continue */
1687 WRITE_ONCE(console_waiter, false);
1689 spin_release(&console_owner_dep_map, _THIS_IP_);
1692 * Hand off console_lock to waiter. The waiter will perform
1693 * the up(). After this, the waiter is the console_lock owner.
1695 mutex_release(&console_lock_dep_map, _THIS_IP_);
1696 return 1;
1700 * console_trylock_spinning - try to get console_lock by busy waiting
1702 * This allows to busy wait for the console_lock when the current
1703 * owner is running in specially marked sections. It means that
1704 * the current owner is running and cannot reschedule until it
1705 * is ready to lose the lock.
1707 * Return: 1 if we got the lock, 0 othrewise
1709 static int console_trylock_spinning(void)
1711 struct task_struct *owner = NULL;
1712 bool waiter;
1713 bool spin = false;
1714 unsigned long flags;
1716 if (console_trylock())
1717 return 1;
1719 printk_safe_enter_irqsave(flags);
1721 raw_spin_lock(&console_owner_lock);
1722 owner = READ_ONCE(console_owner);
1723 waiter = READ_ONCE(console_waiter);
1724 if (!waiter && owner && owner != current) {
1725 WRITE_ONCE(console_waiter, true);
1726 spin = true;
1728 raw_spin_unlock(&console_owner_lock);
1731 * If there is an active printk() writing to the
1732 * consoles, instead of having it write our data too,
1733 * see if we can offload that load from the active
1734 * printer, and do some printing ourselves.
1735 * Go into a spin only if there isn't already a waiter
1736 * spinning, and there is an active printer, and
1737 * that active printer isn't us (recursive printk?).
1739 if (!spin) {
1740 printk_safe_exit_irqrestore(flags);
1741 return 0;
1744 /* We spin waiting for the owner to release us */
1745 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1746 /* Owner will clear console_waiter on hand off */
1747 while (READ_ONCE(console_waiter))
1748 cpu_relax();
1749 spin_release(&console_owner_dep_map, _THIS_IP_);
1751 printk_safe_exit_irqrestore(flags);
1753 * The owner passed the console lock to us.
1754 * Since we did not spin on console lock, annotate
1755 * this as a trylock. Otherwise lockdep will
1756 * complain.
1758 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1760 return 1;
1764 * Call the console drivers, asking them to write out
1765 * log_buf[start] to log_buf[end - 1].
1766 * The console_lock must be held.
1768 static void call_console_drivers(const char *ext_text, size_t ext_len,
1769 const char *text, size_t len)
1771 struct console *con;
1773 trace_console_rcuidle(text, len);
1775 if (!console_drivers)
1776 return;
1778 for_each_console(con) {
1779 if (exclusive_console && con != exclusive_console)
1780 continue;
1781 if (!(con->flags & CON_ENABLED))
1782 continue;
1783 if (!con->write)
1784 continue;
1785 if (!cpu_online(smp_processor_id()) &&
1786 !(con->flags & CON_ANYTIME))
1787 continue;
1788 if (con->flags & CON_EXTENDED)
1789 con->write(con, ext_text, ext_len);
1790 else
1791 con->write(con, text, len);
1795 int printk_delay_msec __read_mostly;
1797 static inline void printk_delay(void)
1799 if (unlikely(printk_delay_msec)) {
1800 int m = printk_delay_msec;
1802 while (m--) {
1803 mdelay(1);
1804 touch_nmi_watchdog();
1809 static inline u32 printk_caller_id(void)
1811 return in_task() ? task_pid_nr(current) :
1812 0x80000000 + raw_smp_processor_id();
1816 * Continuation lines are buffered, and not committed to the record buffer
1817 * until the line is complete, or a race forces it. The line fragments
1818 * though, are printed immediately to the consoles to ensure everything has
1819 * reached the console in case of a kernel crash.
1821 static struct cont {
1822 char buf[LOG_LINE_MAX];
1823 size_t len; /* length == 0 means unused buffer */
1824 u32 caller_id; /* printk_caller_id() of first print */
1825 u64 ts_nsec; /* time of first print */
1826 u8 level; /* log level of first message */
1827 u8 facility; /* log facility of first message */
1828 enum log_flags flags; /* prefix, newline flags */
1829 } cont;
1831 static void cont_flush(void)
1833 if (cont.len == 0)
1834 return;
1836 log_store(cont.caller_id, cont.facility, cont.level, cont.flags,
1837 cont.ts_nsec, NULL, 0, cont.buf, cont.len);
1838 cont.len = 0;
1841 static bool cont_add(u32 caller_id, int facility, int level,
1842 enum log_flags flags, const char *text, size_t len)
1844 /* If the line gets too long, split it up in separate records. */
1845 if (cont.len + len > sizeof(cont.buf)) {
1846 cont_flush();
1847 return false;
1850 if (!cont.len) {
1851 cont.facility = facility;
1852 cont.level = level;
1853 cont.caller_id = caller_id;
1854 cont.ts_nsec = local_clock();
1855 cont.flags = flags;
1858 memcpy(cont.buf + cont.len, text, len);
1859 cont.len += len;
1861 // The original flags come from the first line,
1862 // but later continuations can add a newline.
1863 if (flags & LOG_NEWLINE) {
1864 cont.flags |= LOG_NEWLINE;
1865 cont_flush();
1868 return true;
1871 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)
1873 const u32 caller_id = printk_caller_id();
1876 * If an earlier line was buffered, and we're a continuation
1877 * write from the same context, try to add it to the buffer.
1879 if (cont.len) {
1880 if (cont.caller_id == caller_id && (lflags & LOG_CONT)) {
1881 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1882 return text_len;
1884 /* Otherwise, make sure it's flushed */
1885 cont_flush();
1888 /* Skip empty continuation lines that couldn't be added - they just flush */
1889 if (!text_len && (lflags & LOG_CONT))
1890 return 0;
1892 /* If it doesn't end in a newline, try to buffer the current line */
1893 if (!(lflags & LOG_NEWLINE)) {
1894 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1895 return text_len;
1898 /* Store it in the record log */
1899 return log_store(caller_id, facility, level, lflags, 0,
1900 dict, dictlen, text, text_len);
1903 /* Must be called under logbuf_lock. */
1904 int vprintk_store(int facility, int level,
1905 const char *dict, size_t dictlen,
1906 const char *fmt, va_list args)
1908 static char textbuf[LOG_LINE_MAX];
1909 char *text = textbuf;
1910 size_t text_len;
1911 enum log_flags lflags = 0;
1914 * The printf needs to come first; we need the syslog
1915 * prefix which might be passed-in as a parameter.
1917 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1919 /* mark and strip a trailing newline */
1920 if (text_len && text[text_len-1] == '\n') {
1921 text_len--;
1922 lflags |= LOG_NEWLINE;
1925 /* strip kernel syslog prefix and extract log level or control flags */
1926 if (facility == 0) {
1927 int kern_level;
1929 while ((kern_level = printk_get_level(text)) != 0) {
1930 switch (kern_level) {
1931 case '0' ... '7':
1932 if (level == LOGLEVEL_DEFAULT)
1933 level = kern_level - '0';
1934 break;
1935 case 'c': /* KERN_CONT */
1936 lflags |= LOG_CONT;
1939 text_len -= 2;
1940 text += 2;
1944 if (level == LOGLEVEL_DEFAULT)
1945 level = default_message_loglevel;
1947 if (dict)
1948 lflags |= LOG_NEWLINE;
1950 return log_output(facility, level, lflags,
1951 dict, dictlen, text, text_len);
1954 asmlinkage int vprintk_emit(int facility, int level,
1955 const char *dict, size_t dictlen,
1956 const char *fmt, va_list args)
1958 int printed_len;
1959 bool in_sched = false, pending_output;
1960 unsigned long flags;
1961 u64 curr_log_seq;
1963 /* Suppress unimportant messages after panic happens */
1964 if (unlikely(suppress_printk))
1965 return 0;
1967 if (level == LOGLEVEL_SCHED) {
1968 level = LOGLEVEL_DEFAULT;
1969 in_sched = true;
1972 boot_delay_msec(level);
1973 printk_delay();
1975 /* This stops the holder of console_sem just where we want him */
1976 logbuf_lock_irqsave(flags);
1977 curr_log_seq = log_next_seq;
1978 printed_len = vprintk_store(facility, level, dict, dictlen, fmt, args);
1979 pending_output = (curr_log_seq != log_next_seq);
1980 logbuf_unlock_irqrestore(flags);
1982 /* If called from the scheduler, we can not call up(). */
1983 if (!in_sched && pending_output) {
1985 * Disable preemption to avoid being preempted while holding
1986 * console_sem which would prevent anyone from printing to
1987 * console
1989 preempt_disable();
1991 * Try to acquire and then immediately release the console
1992 * semaphore. The release will print out buffers and wake up
1993 * /dev/kmsg and syslog() users.
1995 if (console_trylock_spinning())
1996 console_unlock();
1997 preempt_enable();
2000 if (pending_output)
2001 wake_up_klogd();
2002 return printed_len;
2004 EXPORT_SYMBOL(vprintk_emit);
2006 asmlinkage int vprintk(const char *fmt, va_list args)
2008 return vprintk_func(fmt, args);
2010 EXPORT_SYMBOL(vprintk);
2012 int vprintk_default(const char *fmt, va_list args)
2014 int r;
2016 #ifdef CONFIG_KGDB_KDB
2017 /* Allow to pass printk() to kdb but avoid a recursion. */
2018 if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
2019 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
2020 return r;
2022 #endif
2023 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
2025 return r;
2027 EXPORT_SYMBOL_GPL(vprintk_default);
2030 * printk - print a kernel message
2031 * @fmt: format string
2033 * This is printk(). It can be called from any context. We want it to work.
2035 * We try to grab the console_lock. If we succeed, it's easy - we log the
2036 * output and call the console drivers. If we fail to get the semaphore, we
2037 * place the output into the log buffer and return. The current holder of
2038 * the console_sem will notice the new output in console_unlock(); and will
2039 * send it to the consoles before releasing the lock.
2041 * One effect of this deferred printing is that code which calls printk() and
2042 * then changes console_loglevel may break. This is because console_loglevel
2043 * is inspected when the actual printing occurs.
2045 * See also:
2046 * printf(3)
2048 * See the vsnprintf() documentation for format string extensions over C99.
2050 asmlinkage __visible int printk(const char *fmt, ...)
2052 va_list args;
2053 int r;
2055 va_start(args, fmt);
2056 r = vprintk_func(fmt, args);
2057 va_end(args);
2059 return r;
2061 EXPORT_SYMBOL(printk);
2063 #else /* CONFIG_PRINTK */
2065 #define LOG_LINE_MAX 0
2066 #define PREFIX_MAX 0
2067 #define printk_time false
2069 static u64 syslog_seq;
2070 static u32 syslog_idx;
2071 static u64 console_seq;
2072 static u32 console_idx;
2073 static u64 exclusive_console_stop_seq;
2074 static u64 log_first_seq;
2075 static u32 log_first_idx;
2076 static u64 log_next_seq;
2077 static char *log_text(const struct printk_log *msg) { return NULL; }
2078 static char *log_dict(const struct printk_log *msg) { return NULL; }
2079 static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2080 static u32 log_next(u32 idx) { return 0; }
2081 static ssize_t msg_print_ext_header(char *buf, size_t size,
2082 struct printk_log *msg,
2083 u64 seq) { return 0; }
2084 static ssize_t msg_print_ext_body(char *buf, size_t size,
2085 char *dict, size_t dict_len,
2086 char *text, size_t text_len) { return 0; }
2087 static void console_lock_spinning_enable(void) { }
2088 static int console_lock_spinning_disable_and_check(void) { return 0; }
2089 static void call_console_drivers(const char *ext_text, size_t ext_len,
2090 const char *text, size_t len) {}
2091 static size_t msg_print_text(const struct printk_log *msg, bool syslog,
2092 bool time, char *buf, size_t size) { return 0; }
2093 static bool suppress_message_printing(int level) { return false; }
2095 #endif /* CONFIG_PRINTK */
2097 #ifdef CONFIG_EARLY_PRINTK
2098 struct console *early_console;
2100 asmlinkage __visible void early_printk(const char *fmt, ...)
2102 va_list ap;
2103 char buf[512];
2104 int n;
2106 if (!early_console)
2107 return;
2109 va_start(ap, fmt);
2110 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2111 va_end(ap);
2113 early_console->write(early_console, buf, n);
2115 #endif
2117 static int __add_preferred_console(char *name, int idx, char *options,
2118 char *brl_options)
2120 struct console_cmdline *c;
2121 int i;
2124 * See if this tty is not yet registered, and
2125 * if we have a slot free.
2127 for (i = 0, c = console_cmdline;
2128 i < MAX_CMDLINECONSOLES && c->name[0];
2129 i++, c++) {
2130 if (strcmp(c->name, name) == 0 && c->index == idx) {
2131 if (!brl_options)
2132 preferred_console = i;
2133 return 0;
2136 if (i == MAX_CMDLINECONSOLES)
2137 return -E2BIG;
2138 if (!brl_options)
2139 preferred_console = i;
2140 strlcpy(c->name, name, sizeof(c->name));
2141 c->options = options;
2142 braille_set_options(c, brl_options);
2144 c->index = idx;
2145 return 0;
2148 static int __init console_msg_format_setup(char *str)
2150 if (!strcmp(str, "syslog"))
2151 console_msg_format = MSG_FORMAT_SYSLOG;
2152 if (!strcmp(str, "default"))
2153 console_msg_format = MSG_FORMAT_DEFAULT;
2154 return 1;
2156 __setup("console_msg_format=", console_msg_format_setup);
2159 * Set up a console. Called via do_early_param() in init/main.c
2160 * for each "console=" parameter in the boot command line.
2162 static int __init console_setup(char *str)
2164 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2165 char *s, *options, *brl_options = NULL;
2166 int idx;
2168 if (_braille_console_setup(&str, &brl_options))
2169 return 1;
2172 * Decode str into name, index, options.
2174 if (str[0] >= '0' && str[0] <= '9') {
2175 strcpy(buf, "ttyS");
2176 strncpy(buf + 4, str, sizeof(buf) - 5);
2177 } else {
2178 strncpy(buf, str, sizeof(buf) - 1);
2180 buf[sizeof(buf) - 1] = 0;
2181 options = strchr(str, ',');
2182 if (options)
2183 *(options++) = 0;
2184 #ifdef __sparc__
2185 if (!strcmp(str, "ttya"))
2186 strcpy(buf, "ttyS0");
2187 if (!strcmp(str, "ttyb"))
2188 strcpy(buf, "ttyS1");
2189 #endif
2190 for (s = buf; *s; s++)
2191 if (isdigit(*s) || *s == ',')
2192 break;
2193 idx = simple_strtoul(s, NULL, 10);
2194 *s = 0;
2196 __add_preferred_console(buf, idx, options, brl_options);
2197 console_set_on_cmdline = 1;
2198 return 1;
2200 __setup("console=", console_setup);
2203 * add_preferred_console - add a device to the list of preferred consoles.
2204 * @name: device name
2205 * @idx: device index
2206 * @options: options for this console
2208 * The last preferred console added will be used for kernel messages
2209 * and stdin/out/err for init. Normally this is used by console_setup
2210 * above to handle user-supplied console arguments; however it can also
2211 * be used by arch-specific code either to override the user or more
2212 * commonly to provide a default console (ie from PROM variables) when
2213 * the user has not supplied one.
2215 int add_preferred_console(char *name, int idx, char *options)
2217 return __add_preferred_console(name, idx, options, NULL);
2220 bool console_suspend_enabled = true;
2221 EXPORT_SYMBOL(console_suspend_enabled);
2223 static int __init console_suspend_disable(char *str)
2225 console_suspend_enabled = false;
2226 return 1;
2228 __setup("no_console_suspend", console_suspend_disable);
2229 module_param_named(console_suspend, console_suspend_enabled,
2230 bool, S_IRUGO | S_IWUSR);
2231 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2232 " and hibernate operations");
2235 * suspend_console - suspend the console subsystem
2237 * This disables printk() while we go into suspend states
2239 void suspend_console(void)
2241 if (!console_suspend_enabled)
2242 return;
2243 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2244 console_lock();
2245 console_suspended = 1;
2246 up_console_sem();
2249 void resume_console(void)
2251 if (!console_suspend_enabled)
2252 return;
2253 down_console_sem();
2254 console_suspended = 0;
2255 console_unlock();
2259 * console_cpu_notify - print deferred console messages after CPU hotplug
2260 * @cpu: unused
2262 * If printk() is called from a CPU that is not online yet, the messages
2263 * will be printed on the console only if there are CON_ANYTIME consoles.
2264 * This function is called when a new CPU comes online (or fails to come
2265 * up) or goes offline.
2267 static int console_cpu_notify(unsigned int cpu)
2269 if (!cpuhp_tasks_frozen) {
2270 /* If trylock fails, someone else is doing the printing */
2271 if (console_trylock())
2272 console_unlock();
2274 return 0;
2278 * console_lock - lock the console system for exclusive use.
2280 * Acquires a lock which guarantees that the caller has
2281 * exclusive access to the console system and the console_drivers list.
2283 * Can sleep, returns nothing.
2285 void console_lock(void)
2287 might_sleep();
2289 down_console_sem();
2290 if (console_suspended)
2291 return;
2292 console_locked = 1;
2293 console_may_schedule = 1;
2295 EXPORT_SYMBOL(console_lock);
2298 * console_trylock - try to lock the console system for exclusive use.
2300 * Try to acquire a lock which guarantees that the caller has exclusive
2301 * access to the console system and the console_drivers list.
2303 * returns 1 on success, and 0 on failure to acquire the lock.
2305 int console_trylock(void)
2307 if (down_trylock_console_sem())
2308 return 0;
2309 if (console_suspended) {
2310 up_console_sem();
2311 return 0;
2313 console_locked = 1;
2314 console_may_schedule = 0;
2315 return 1;
2317 EXPORT_SYMBOL(console_trylock);
2319 int is_console_locked(void)
2321 return console_locked;
2323 EXPORT_SYMBOL(is_console_locked);
2326 * Check if we have any console that is capable of printing while cpu is
2327 * booting or shutting down. Requires console_sem.
2329 static int have_callable_console(void)
2331 struct console *con;
2333 for_each_console(con)
2334 if ((con->flags & CON_ENABLED) &&
2335 (con->flags & CON_ANYTIME))
2336 return 1;
2338 return 0;
2342 * Can we actually use the console at this time on this cpu?
2344 * Console drivers may assume that per-cpu resources have been allocated. So
2345 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2346 * call them until this CPU is officially up.
2348 static inline int can_use_console(void)
2350 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2354 * console_unlock - unlock the console system
2356 * Releases the console_lock which the caller holds on the console system
2357 * and the console driver list.
2359 * While the console_lock was held, console output may have been buffered
2360 * by printk(). If this is the case, console_unlock(); emits
2361 * the output prior to releasing the lock.
2363 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2365 * console_unlock(); may be called from any context.
2367 void console_unlock(void)
2369 static char ext_text[CONSOLE_EXT_LOG_MAX];
2370 static char text[LOG_LINE_MAX + PREFIX_MAX];
2371 unsigned long flags;
2372 bool do_cond_resched, retry;
2374 if (console_suspended) {
2375 up_console_sem();
2376 return;
2380 * Console drivers are called with interrupts disabled, so
2381 * @console_may_schedule should be cleared before; however, we may
2382 * end up dumping a lot of lines, for example, if called from
2383 * console registration path, and should invoke cond_resched()
2384 * between lines if allowable. Not doing so can cause a very long
2385 * scheduling stall on a slow console leading to RCU stall and
2386 * softlockup warnings which exacerbate the issue with more
2387 * messages practically incapacitating the system.
2389 * console_trylock() is not able to detect the preemptive
2390 * context reliably. Therefore the value must be stored before
2391 * and cleared after the the "again" goto label.
2393 do_cond_resched = console_may_schedule;
2394 again:
2395 console_may_schedule = 0;
2398 * We released the console_sem lock, so we need to recheck if
2399 * cpu is online and (if not) is there at least one CON_ANYTIME
2400 * console.
2402 if (!can_use_console()) {
2403 console_locked = 0;
2404 up_console_sem();
2405 return;
2408 for (;;) {
2409 struct printk_log *msg;
2410 size_t ext_len = 0;
2411 size_t len;
2413 printk_safe_enter_irqsave(flags);
2414 raw_spin_lock(&logbuf_lock);
2415 if (console_seq < log_first_seq) {
2416 len = sprintf(text,
2417 "** %llu printk messages dropped **\n",
2418 log_first_seq - console_seq);
2420 /* messages are gone, move to first one */
2421 console_seq = log_first_seq;
2422 console_idx = log_first_idx;
2423 } else {
2424 len = 0;
2426 skip:
2427 if (console_seq == log_next_seq)
2428 break;
2430 msg = log_from_idx(console_idx);
2431 if (suppress_message_printing(msg->level)) {
2433 * Skip record we have buffered and already printed
2434 * directly to the console when we received it, and
2435 * record that has level above the console loglevel.
2437 console_idx = log_next(console_idx);
2438 console_seq++;
2439 goto skip;
2442 /* Output to all consoles once old messages replayed. */
2443 if (unlikely(exclusive_console &&
2444 console_seq >= exclusive_console_stop_seq)) {
2445 exclusive_console = NULL;
2448 len += msg_print_text(msg,
2449 console_msg_format & MSG_FORMAT_SYSLOG,
2450 printk_time, text + len, sizeof(text) - len);
2451 if (nr_ext_console_drivers) {
2452 ext_len = msg_print_ext_header(ext_text,
2453 sizeof(ext_text),
2454 msg, console_seq);
2455 ext_len += msg_print_ext_body(ext_text + ext_len,
2456 sizeof(ext_text) - ext_len,
2457 log_dict(msg), msg->dict_len,
2458 log_text(msg), msg->text_len);
2460 console_idx = log_next(console_idx);
2461 console_seq++;
2462 raw_spin_unlock(&logbuf_lock);
2465 * While actively printing out messages, if another printk()
2466 * were to occur on another CPU, it may wait for this one to
2467 * finish. This task can not be preempted if there is a
2468 * waiter waiting to take over.
2470 console_lock_spinning_enable();
2472 stop_critical_timings(); /* don't trace print latency */
2473 call_console_drivers(ext_text, ext_len, text, len);
2474 start_critical_timings();
2476 if (console_lock_spinning_disable_and_check()) {
2477 printk_safe_exit_irqrestore(flags);
2478 return;
2481 printk_safe_exit_irqrestore(flags);
2483 if (do_cond_resched)
2484 cond_resched();
2487 console_locked = 0;
2489 raw_spin_unlock(&logbuf_lock);
2491 up_console_sem();
2494 * Someone could have filled up the buffer again, so re-check if there's
2495 * something to flush. In case we cannot trylock the console_sem again,
2496 * there's a new owner and the console_unlock() from them will do the
2497 * flush, no worries.
2499 raw_spin_lock(&logbuf_lock);
2500 retry = console_seq != log_next_seq;
2501 raw_spin_unlock(&logbuf_lock);
2502 printk_safe_exit_irqrestore(flags);
2504 if (retry && console_trylock())
2505 goto again;
2507 EXPORT_SYMBOL(console_unlock);
2510 * console_conditional_schedule - yield the CPU if required
2512 * If the console code is currently allowed to sleep, and
2513 * if this CPU should yield the CPU to another task, do
2514 * so here.
2516 * Must be called within console_lock();.
2518 void __sched console_conditional_schedule(void)
2520 if (console_may_schedule)
2521 cond_resched();
2523 EXPORT_SYMBOL(console_conditional_schedule);
2525 void console_unblank(void)
2527 struct console *c;
2530 * console_unblank can no longer be called in interrupt context unless
2531 * oops_in_progress is set to 1..
2533 if (oops_in_progress) {
2534 if (down_trylock_console_sem() != 0)
2535 return;
2536 } else
2537 console_lock();
2539 console_locked = 1;
2540 console_may_schedule = 0;
2541 for_each_console(c)
2542 if ((c->flags & CON_ENABLED) && c->unblank)
2543 c->unblank();
2544 console_unlock();
2548 * console_flush_on_panic - flush console content on panic
2549 * @mode: flush all messages in buffer or just the pending ones
2551 * Immediately output all pending messages no matter what.
2553 void console_flush_on_panic(enum con_flush_mode mode)
2556 * If someone else is holding the console lock, trylock will fail
2557 * and may_schedule may be set. Ignore and proceed to unlock so
2558 * that messages are flushed out. As this can be called from any
2559 * context and we don't want to get preempted while flushing,
2560 * ensure may_schedule is cleared.
2562 console_trylock();
2563 console_may_schedule = 0;
2565 if (mode == CONSOLE_REPLAY_ALL) {
2566 unsigned long flags;
2568 logbuf_lock_irqsave(flags);
2569 console_seq = log_first_seq;
2570 console_idx = log_first_idx;
2571 logbuf_unlock_irqrestore(flags);
2573 console_unlock();
2577 * Return the console tty driver structure and its associated index
2579 struct tty_driver *console_device(int *index)
2581 struct console *c;
2582 struct tty_driver *driver = NULL;
2584 console_lock();
2585 for_each_console(c) {
2586 if (!c->device)
2587 continue;
2588 driver = c->device(c, index);
2589 if (driver)
2590 break;
2592 console_unlock();
2593 return driver;
2597 * Prevent further output on the passed console device so that (for example)
2598 * serial drivers can disable console output before suspending a port, and can
2599 * re-enable output afterwards.
2601 void console_stop(struct console *console)
2603 console_lock();
2604 console->flags &= ~CON_ENABLED;
2605 console_unlock();
2607 EXPORT_SYMBOL(console_stop);
2609 void console_start(struct console *console)
2611 console_lock();
2612 console->flags |= CON_ENABLED;
2613 console_unlock();
2615 EXPORT_SYMBOL(console_start);
2617 static int __read_mostly keep_bootcon;
2619 static int __init keep_bootcon_setup(char *str)
2621 keep_bootcon = 1;
2622 pr_info("debug: skip boot console de-registration.\n");
2624 return 0;
2627 early_param("keep_bootcon", keep_bootcon_setup);
2630 * The console driver calls this routine during kernel initialization
2631 * to register the console printing procedure with printk() and to
2632 * print any messages that were printed by the kernel before the
2633 * console driver was initialized.
2635 * This can happen pretty early during the boot process (because of
2636 * early_printk) - sometimes before setup_arch() completes - be careful
2637 * of what kernel features are used - they may not be initialised yet.
2639 * There are two types of consoles - bootconsoles (early_printk) and
2640 * "real" consoles (everything which is not a bootconsole) which are
2641 * handled differently.
2642 * - Any number of bootconsoles can be registered at any time.
2643 * - As soon as a "real" console is registered, all bootconsoles
2644 * will be unregistered automatically.
2645 * - Once a "real" console is registered, any attempt to register a
2646 * bootconsoles will be rejected
2648 void register_console(struct console *newcon)
2650 int i;
2651 unsigned long flags;
2652 struct console *bcon = NULL;
2653 struct console_cmdline *c;
2654 static bool has_preferred;
2656 if (console_drivers)
2657 for_each_console(bcon)
2658 if (WARN(bcon == newcon,
2659 "console '%s%d' already registered\n",
2660 bcon->name, bcon->index))
2661 return;
2664 * before we register a new CON_BOOT console, make sure we don't
2665 * already have a valid console
2667 if (console_drivers && newcon->flags & CON_BOOT) {
2668 /* find the last or real console */
2669 for_each_console(bcon) {
2670 if (!(bcon->flags & CON_BOOT)) {
2671 pr_info("Too late to register bootconsole %s%d\n",
2672 newcon->name, newcon->index);
2673 return;
2678 if (console_drivers && console_drivers->flags & CON_BOOT)
2679 bcon = console_drivers;
2681 if (!has_preferred || bcon || !console_drivers)
2682 has_preferred = preferred_console >= 0;
2685 * See if we want to use this console driver. If we
2686 * didn't select a console we take the first one
2687 * that registers here.
2689 if (!has_preferred) {
2690 if (newcon->index < 0)
2691 newcon->index = 0;
2692 if (newcon->setup == NULL ||
2693 newcon->setup(newcon, NULL) == 0) {
2694 newcon->flags |= CON_ENABLED;
2695 if (newcon->device) {
2696 newcon->flags |= CON_CONSDEV;
2697 has_preferred = true;
2703 * See if this console matches one we selected on
2704 * the command line.
2706 for (i = 0, c = console_cmdline;
2707 i < MAX_CMDLINECONSOLES && c->name[0];
2708 i++, c++) {
2709 if (!newcon->match ||
2710 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2711 /* default matching */
2712 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2713 if (strcmp(c->name, newcon->name) != 0)
2714 continue;
2715 if (newcon->index >= 0 &&
2716 newcon->index != c->index)
2717 continue;
2718 if (newcon->index < 0)
2719 newcon->index = c->index;
2721 if (_braille_register_console(newcon, c))
2722 return;
2724 if (newcon->setup &&
2725 newcon->setup(newcon, c->options) != 0)
2726 break;
2729 newcon->flags |= CON_ENABLED;
2730 if (i == preferred_console) {
2731 newcon->flags |= CON_CONSDEV;
2732 has_preferred = true;
2734 break;
2737 if (!(newcon->flags & CON_ENABLED))
2738 return;
2741 * If we have a bootconsole, and are switching to a real console,
2742 * don't print everything out again, since when the boot console, and
2743 * the real console are the same physical device, it's annoying to
2744 * see the beginning boot messages twice
2746 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2747 newcon->flags &= ~CON_PRINTBUFFER;
2750 * Put this console in the list - keep the
2751 * preferred driver at the head of the list.
2753 console_lock();
2754 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2755 newcon->next = console_drivers;
2756 console_drivers = newcon;
2757 if (newcon->next)
2758 newcon->next->flags &= ~CON_CONSDEV;
2759 } else {
2760 newcon->next = console_drivers->next;
2761 console_drivers->next = newcon;
2764 if (newcon->flags & CON_EXTENDED)
2765 nr_ext_console_drivers++;
2767 if (newcon->flags & CON_PRINTBUFFER) {
2769 * console_unlock(); will print out the buffered messages
2770 * for us.
2772 logbuf_lock_irqsave(flags);
2773 console_seq = syslog_seq;
2774 console_idx = syslog_idx;
2776 * We're about to replay the log buffer. Only do this to the
2777 * just-registered console to avoid excessive message spam to
2778 * the already-registered consoles.
2780 * Set exclusive_console with disabled interrupts to reduce
2781 * race window with eventual console_flush_on_panic() that
2782 * ignores console_lock.
2784 exclusive_console = newcon;
2785 exclusive_console_stop_seq = console_seq;
2786 logbuf_unlock_irqrestore(flags);
2788 console_unlock();
2789 console_sysfs_notify();
2792 * By unregistering the bootconsoles after we enable the real console
2793 * we get the "console xxx enabled" message on all the consoles -
2794 * boot consoles, real consoles, etc - this is to ensure that end
2795 * users know there might be something in the kernel's log buffer that
2796 * went to the bootconsole (that they do not see on the real console)
2798 pr_info("%sconsole [%s%d] enabled\n",
2799 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2800 newcon->name, newcon->index);
2801 if (bcon &&
2802 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2803 !keep_bootcon) {
2804 /* We need to iterate through all boot consoles, to make
2805 * sure we print everything out, before we unregister them.
2807 for_each_console(bcon)
2808 if (bcon->flags & CON_BOOT)
2809 unregister_console(bcon);
2812 EXPORT_SYMBOL(register_console);
2814 int unregister_console(struct console *console)
2816 struct console *a, *b;
2817 int res;
2819 pr_info("%sconsole [%s%d] disabled\n",
2820 (console->flags & CON_BOOT) ? "boot" : "" ,
2821 console->name, console->index);
2823 res = _braille_unregister_console(console);
2824 if (res)
2825 return res;
2827 res = 1;
2828 console_lock();
2829 if (console_drivers == console) {
2830 console_drivers=console->next;
2831 res = 0;
2832 } else if (console_drivers) {
2833 for (a=console_drivers->next, b=console_drivers ;
2834 a; b=a, a=b->next) {
2835 if (a == console) {
2836 b->next = a->next;
2837 res = 0;
2838 break;
2843 if (!res && (console->flags & CON_EXTENDED))
2844 nr_ext_console_drivers--;
2847 * If this isn't the last console and it has CON_CONSDEV set, we
2848 * need to set it on the next preferred console.
2850 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2851 console_drivers->flags |= CON_CONSDEV;
2853 console->flags &= ~CON_ENABLED;
2854 console_unlock();
2855 console_sysfs_notify();
2856 return res;
2858 EXPORT_SYMBOL(unregister_console);
2861 * Initialize the console device. This is called *early*, so
2862 * we can't necessarily depend on lots of kernel help here.
2863 * Just do some early initializations, and do the complex setup
2864 * later.
2866 void __init console_init(void)
2868 int ret;
2869 initcall_t call;
2870 initcall_entry_t *ce;
2872 /* Setup the default TTY line discipline. */
2873 n_tty_init();
2876 * set up the console device so that later boot sequences can
2877 * inform about problems etc..
2879 ce = __con_initcall_start;
2880 trace_initcall_level("console");
2881 while (ce < __con_initcall_end) {
2882 call = initcall_from_entry(ce);
2883 trace_initcall_start(call);
2884 ret = call();
2885 trace_initcall_finish(call, ret);
2886 ce++;
2891 * Some boot consoles access data that is in the init section and which will
2892 * be discarded after the initcalls have been run. To make sure that no code
2893 * will access this data, unregister the boot consoles in a late initcall.
2895 * If for some reason, such as deferred probe or the driver being a loadable
2896 * module, the real console hasn't registered yet at this point, there will
2897 * be a brief interval in which no messages are logged to the console, which
2898 * makes it difficult to diagnose problems that occur during this time.
2900 * To mitigate this problem somewhat, only unregister consoles whose memory
2901 * intersects with the init section. Note that all other boot consoles will
2902 * get unregistred when the real preferred console is registered.
2904 static int __init printk_late_init(void)
2906 struct console *con;
2907 int ret;
2909 for_each_console(con) {
2910 if (!(con->flags & CON_BOOT))
2911 continue;
2913 /* Check addresses that might be used for enabled consoles. */
2914 if (init_section_intersects(con, sizeof(*con)) ||
2915 init_section_contains(con->write, 0) ||
2916 init_section_contains(con->read, 0) ||
2917 init_section_contains(con->device, 0) ||
2918 init_section_contains(con->unblank, 0) ||
2919 init_section_contains(con->data, 0)) {
2921 * Please, consider moving the reported consoles out
2922 * of the init section.
2924 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2925 con->name, con->index);
2926 unregister_console(con);
2929 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2930 console_cpu_notify);
2931 WARN_ON(ret < 0);
2932 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2933 console_cpu_notify, NULL);
2934 WARN_ON(ret < 0);
2935 return 0;
2937 late_initcall(printk_late_init);
2939 #if defined CONFIG_PRINTK
2941 * Delayed printk version, for scheduler-internal messages:
2943 #define PRINTK_PENDING_WAKEUP 0x01
2944 #define PRINTK_PENDING_OUTPUT 0x02
2946 static DEFINE_PER_CPU(int, printk_pending);
2948 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2950 int pending = __this_cpu_xchg(printk_pending, 0);
2952 if (pending & PRINTK_PENDING_OUTPUT) {
2953 /* If trylock fails, someone else is doing the printing */
2954 if (console_trylock())
2955 console_unlock();
2958 if (pending & PRINTK_PENDING_WAKEUP)
2959 wake_up_interruptible(&log_wait);
2962 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2963 .func = wake_up_klogd_work_func,
2964 .flags = ATOMIC_INIT(IRQ_WORK_LAZY),
2967 void wake_up_klogd(void)
2969 preempt_disable();
2970 if (waitqueue_active(&log_wait)) {
2971 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2972 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2974 preempt_enable();
2977 void defer_console_output(void)
2979 preempt_disable();
2980 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2981 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2982 preempt_enable();
2985 int vprintk_deferred(const char *fmt, va_list args)
2987 int r;
2989 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2990 defer_console_output();
2992 return r;
2995 int printk_deferred(const char *fmt, ...)
2997 va_list args;
2998 int r;
3000 va_start(args, fmt);
3001 r = vprintk_deferred(fmt, args);
3002 va_end(args);
3004 return r;
3008 * printk rate limiting, lifted from the networking subsystem.
3010 * This enforces a rate limit: not more than 10 kernel messages
3011 * every 5s to make a denial-of-service attack impossible.
3013 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3015 int __printk_ratelimit(const char *func)
3017 return ___ratelimit(&printk_ratelimit_state, func);
3019 EXPORT_SYMBOL(__printk_ratelimit);
3022 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3023 * @caller_jiffies: pointer to caller's state
3024 * @interval_msecs: minimum interval between prints
3026 * printk_timed_ratelimit() returns true if more than @interval_msecs
3027 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3028 * returned true.
3030 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3031 unsigned int interval_msecs)
3033 unsigned long elapsed = jiffies - *caller_jiffies;
3035 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3036 return false;
3038 *caller_jiffies = jiffies;
3039 return true;
3041 EXPORT_SYMBOL(printk_timed_ratelimit);
3043 static DEFINE_SPINLOCK(dump_list_lock);
3044 static LIST_HEAD(dump_list);
3047 * kmsg_dump_register - register a kernel log dumper.
3048 * @dumper: pointer to the kmsg_dumper structure
3050 * Adds a kernel log dumper to the system. The dump callback in the
3051 * structure will be called when the kernel oopses or panics and must be
3052 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3054 int kmsg_dump_register(struct kmsg_dumper *dumper)
3056 unsigned long flags;
3057 int err = -EBUSY;
3059 /* The dump callback needs to be set */
3060 if (!dumper->dump)
3061 return -EINVAL;
3063 spin_lock_irqsave(&dump_list_lock, flags);
3064 /* Don't allow registering multiple times */
3065 if (!dumper->registered) {
3066 dumper->registered = 1;
3067 list_add_tail_rcu(&dumper->list, &dump_list);
3068 err = 0;
3070 spin_unlock_irqrestore(&dump_list_lock, flags);
3072 return err;
3074 EXPORT_SYMBOL_GPL(kmsg_dump_register);
3077 * kmsg_dump_unregister - unregister a kmsg dumper.
3078 * @dumper: pointer to the kmsg_dumper structure
3080 * Removes a dump device from the system. Returns zero on success and
3081 * %-EINVAL otherwise.
3083 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3085 unsigned long flags;
3086 int err = -EINVAL;
3088 spin_lock_irqsave(&dump_list_lock, flags);
3089 if (dumper->registered) {
3090 dumper->registered = 0;
3091 list_del_rcu(&dumper->list);
3092 err = 0;
3094 spin_unlock_irqrestore(&dump_list_lock, flags);
3095 synchronize_rcu();
3097 return err;
3099 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3101 static bool always_kmsg_dump;
3102 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3105 * kmsg_dump - dump kernel log to kernel message dumpers.
3106 * @reason: the reason (oops, panic etc) for dumping
3108 * Call each of the registered dumper's dump() callback, which can
3109 * retrieve the kmsg records with kmsg_dump_get_line() or
3110 * kmsg_dump_get_buffer().
3112 void kmsg_dump(enum kmsg_dump_reason reason)
3114 struct kmsg_dumper *dumper;
3115 unsigned long flags;
3117 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3118 return;
3120 rcu_read_lock();
3121 list_for_each_entry_rcu(dumper, &dump_list, list) {
3122 if (dumper->max_reason && reason > dumper->max_reason)
3123 continue;
3125 /* initialize iterator with data about the stored records */
3126 dumper->active = true;
3128 logbuf_lock_irqsave(flags);
3129 dumper->cur_seq = clear_seq;
3130 dumper->cur_idx = clear_idx;
3131 dumper->next_seq = log_next_seq;
3132 dumper->next_idx = log_next_idx;
3133 logbuf_unlock_irqrestore(flags);
3135 /* invoke dumper which will iterate over records */
3136 dumper->dump(dumper, reason);
3138 /* reset iterator */
3139 dumper->active = false;
3141 rcu_read_unlock();
3145 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3146 * @dumper: registered kmsg dumper
3147 * @syslog: include the "<4>" prefixes
3148 * @line: buffer to copy the line to
3149 * @size: maximum size of the buffer
3150 * @len: length of line placed into buffer
3152 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3153 * record, and copy one record into the provided buffer.
3155 * Consecutive calls will return the next available record moving
3156 * towards the end of the buffer with the youngest messages.
3158 * A return value of FALSE indicates that there are no more records to
3159 * read.
3161 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3163 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3164 char *line, size_t size, size_t *len)
3166 struct printk_log *msg;
3167 size_t l = 0;
3168 bool ret = false;
3170 if (!dumper->active)
3171 goto out;
3173 if (dumper->cur_seq < log_first_seq) {
3174 /* messages are gone, move to first available one */
3175 dumper->cur_seq = log_first_seq;
3176 dumper->cur_idx = log_first_idx;
3179 /* last entry */
3180 if (dumper->cur_seq >= log_next_seq)
3181 goto out;
3183 msg = log_from_idx(dumper->cur_idx);
3184 l = msg_print_text(msg, syslog, printk_time, line, size);
3186 dumper->cur_idx = log_next(dumper->cur_idx);
3187 dumper->cur_seq++;
3188 ret = true;
3189 out:
3190 if (len)
3191 *len = l;
3192 return ret;
3196 * kmsg_dump_get_line - retrieve one kmsg log line
3197 * @dumper: registered kmsg dumper
3198 * @syslog: include the "<4>" prefixes
3199 * @line: buffer to copy the line to
3200 * @size: maximum size of the buffer
3201 * @len: length of line placed into buffer
3203 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3204 * record, and copy one record into the provided buffer.
3206 * Consecutive calls will return the next available record moving
3207 * towards the end of the buffer with the youngest messages.
3209 * A return value of FALSE indicates that there are no more records to
3210 * read.
3212 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3213 char *line, size_t size, size_t *len)
3215 unsigned long flags;
3216 bool ret;
3218 logbuf_lock_irqsave(flags);
3219 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3220 logbuf_unlock_irqrestore(flags);
3222 return ret;
3224 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3227 * kmsg_dump_get_buffer - copy kmsg log lines
3228 * @dumper: registered kmsg dumper
3229 * @syslog: include the "<4>" prefixes
3230 * @buf: buffer to copy the line to
3231 * @size: maximum size of the buffer
3232 * @len: length of line placed into buffer
3234 * Start at the end of the kmsg buffer and fill the provided buffer
3235 * with as many of the the *youngest* kmsg records that fit into it.
3236 * If the buffer is large enough, all available kmsg records will be
3237 * copied with a single call.
3239 * Consecutive calls will fill the buffer with the next block of
3240 * available older records, not including the earlier retrieved ones.
3242 * A return value of FALSE indicates that there are no more records to
3243 * read.
3245 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3246 char *buf, size_t size, size_t *len)
3248 unsigned long flags;
3249 u64 seq;
3250 u32 idx;
3251 u64 next_seq;
3252 u32 next_idx;
3253 size_t l = 0;
3254 bool ret = false;
3255 bool time = printk_time;
3257 if (!dumper->active)
3258 goto out;
3260 logbuf_lock_irqsave(flags);
3261 if (dumper->cur_seq < log_first_seq) {
3262 /* messages are gone, move to first available one */
3263 dumper->cur_seq = log_first_seq;
3264 dumper->cur_idx = log_first_idx;
3267 /* last entry */
3268 if (dumper->cur_seq >= dumper->next_seq) {
3269 logbuf_unlock_irqrestore(flags);
3270 goto out;
3273 /* calculate length of entire buffer */
3274 seq = dumper->cur_seq;
3275 idx = dumper->cur_idx;
3276 while (seq < dumper->next_seq) {
3277 struct printk_log *msg = log_from_idx(idx);
3279 l += msg_print_text(msg, true, time, NULL, 0);
3280 idx = log_next(idx);
3281 seq++;
3284 /* move first record forward until length fits into the buffer */
3285 seq = dumper->cur_seq;
3286 idx = dumper->cur_idx;
3287 while (l >= size && seq < dumper->next_seq) {
3288 struct printk_log *msg = log_from_idx(idx);
3290 l -= msg_print_text(msg, true, time, NULL, 0);
3291 idx = log_next(idx);
3292 seq++;
3295 /* last message in next interation */
3296 next_seq = seq;
3297 next_idx = idx;
3299 l = 0;
3300 while (seq < dumper->next_seq) {
3301 struct printk_log *msg = log_from_idx(idx);
3303 l += msg_print_text(msg, syslog, time, buf + l, size - l);
3304 idx = log_next(idx);
3305 seq++;
3308 dumper->next_seq = next_seq;
3309 dumper->next_idx = next_idx;
3310 ret = true;
3311 logbuf_unlock_irqrestore(flags);
3312 out:
3313 if (len)
3314 *len = l;
3315 return ret;
3317 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3320 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3321 * @dumper: registered kmsg dumper
3323 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3324 * kmsg_dump_get_buffer() can be called again and used multiple
3325 * times within the same dumper.dump() callback.
3327 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3329 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3331 dumper->cur_seq = clear_seq;
3332 dumper->cur_idx = clear_idx;
3333 dumper->next_seq = log_next_seq;
3334 dumper->next_idx = log_next_idx;
3338 * kmsg_dump_rewind - reset the interator
3339 * @dumper: registered kmsg dumper
3341 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3342 * kmsg_dump_get_buffer() can be called again and used multiple
3343 * times within the same dumper.dump() callback.
3345 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3347 unsigned long flags;
3349 logbuf_lock_irqsave(flags);
3350 kmsg_dump_rewind_nolock(dumper);
3351 logbuf_unlock_irqrestore(flags);
3353 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3355 #endif