Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/printk.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
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
18 */
19
20#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21
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>
51
52#include <linux/uaccess.h>
53#include <asm/sections.h>
54
55#include <trace/events/initcall.h>
56#define CREATE_TRACE_POINTS
57#include <trace/events/printk.h>
58
59#include "console_cmdline.h"
60#include "braille.h"
61#include "internal.h"
62
63int 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 */
68};
69EXPORT_SYMBOL_GPL(console_printk);
70
71atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
72EXPORT_SYMBOL(ignore_console_lock_warning);
73
74/*
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.
77 */
78int oops_in_progress;
79EXPORT_SYMBOL(oops_in_progress);
80
81/*
82 * console_sem protects the console_drivers list, and also
83 * provides serialisation for access to the entire console
84 * driver system.
85 */
86static DEFINE_SEMAPHORE(console_sem);
87struct console *console_drivers;
88EXPORT_SYMBOL_GPL(console_drivers);
89
90/*
91 * System may need to suppress printk message under certain
92 * circumstances, like after kernel panic happens.
93 */
94int __read_mostly suppress_printk;
95
96#ifdef CONFIG_LOCKDEP
97static struct lockdep_map console_lock_dep_map = {
98 .name = "console_lock"
99};
100#endif
101
102enum devkmsg_log_bits {
103 __DEVKMSG_LOG_BIT_ON = 0,
104 __DEVKMSG_LOG_BIT_OFF,
105 __DEVKMSG_LOG_BIT_LOCK,
106};
107
108enum 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),
112};
113
114/* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
115#define DEVKMSG_LOG_MASK_DEFAULT 0
116
117static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
118
119static int __control_devkmsg(char *str)
120{
121 size_t len;
122
123 if (!str)
124 return -EINVAL;
125
126 len = str_has_prefix(str, "on");
127 if (len) {
128 devkmsg_log = DEVKMSG_LOG_MASK_ON;
129 return len;
130 }
131
132 len = str_has_prefix(str, "off");
133 if (len) {
134 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
135 return len;
136 }
137
138 len = str_has_prefix(str, "ratelimit");
139 if (len) {
140 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
141 return len;
142 }
143
144 return -EINVAL;
145}
146
147static int __init control_devkmsg(char *str)
148{
149 if (__control_devkmsg(str) < 0)
150 return 1;
151
152 /*
153 * Set sysctl string accordingly:
154 */
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. */
160
161 /*
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.
166 */
167 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
168
169 return 0;
170}
171__setup("printk.devkmsg=", control_devkmsg);
172
173char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
174
175int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
176 void __user *buffer, size_t *lenp, loff_t *ppos)
177{
178 char old_str[DEVKMSG_STR_MAX_SIZE];
179 unsigned int old;
180 int err;
181
182 if (write) {
183 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
184 return -EINVAL;
185
186 old = devkmsg_log;
187 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
188 }
189
190 err = proc_dostring(table, write, buffer, lenp, ppos);
191 if (err)
192 return err;
193
194 if (write) {
195 err = __control_devkmsg(devkmsg_log_str);
196
197 /*
198 * Do not accept an unknown string OR a known string with
199 * trailing crap...
200 */
201 if (err < 0 || (err + 1 != *lenp)) {
202
203 /* ... and restore old setting. */
204 devkmsg_log = old;
205 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
206
207 return -EINVAL;
208 }
209 }
210
211 return 0;
212}
213
214/* Number of registered extended console drivers. */
215static int nr_ext_console_drivers;
216
217/*
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.
220 */
221#define down_console_sem() do { \
222 down(&console_sem);\
223 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
224} while (0)
225
226static int __down_trylock_console_sem(unsigned long ip)
227{
228 int lock_failed;
229 unsigned long flags;
230
231 /*
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.
235 */
236 printk_safe_enter_irqsave(flags);
237 lock_failed = down_trylock(&console_sem);
238 printk_safe_exit_irqrestore(flags);
239
240 if (lock_failed)
241 return 1;
242 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
243 return 0;
244}
245#define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
246
247static void __up_console_sem(unsigned long ip)
248{
249 unsigned long flags;
250
251 mutex_release(&console_lock_dep_map, 1, ip);
252
253 printk_safe_enter_irqsave(flags);
254 up(&console_sem);
255 printk_safe_exit_irqrestore(flags);
256}
257#define up_console_sem() __up_console_sem(_RET_IP_)
258
259/*
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).
266 */
267static int console_locked, console_suspended;
268
269/*
270 * If exclusive_console is non-NULL then only this console is to be printed to.
271 */
272static struct console *exclusive_console;
273
274/*
275 * Array of consoles built from command line options (console=)
276 */
277
278#define MAX_CMDLINECONSOLES 8
279
280static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
281
282static int preferred_console = -1;
283int console_set_on_cmdline;
284EXPORT_SYMBOL(console_set_on_cmdline);
285
286/* Flag: console code may call schedule() */
287static int console_may_schedule;
288
289enum con_msg_format_flags {
290 MSG_FORMAT_DEFAULT = 0,
291 MSG_FORMAT_SYSLOG = (1 << 0),
292};
293
294static int console_msg_format = MSG_FORMAT_DEFAULT;
295
296/*
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.
300 *
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.
304 *
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.
308 *
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.
314 *
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.
318 *
319 * Optionally, a message can carry a dictionary of properties (key/value pairs),
320 * to provide userspace with a machine-readable message context.
321 *
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
329 *
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.
333 *
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
347 *
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.
351 *
352 * /dev/kmsg exports the structured data in the following line format:
353 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
354 *
355 * Users of the export format should ignore possible additional values
356 * separated by ',', and find the message after the ';' character.
357 *
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.
361 */
362
363enum log_flags {
364 LOG_NEWLINE = 2, /* text ended with a newline */
365 LOG_CONT = 8, /* text is a fragment of a continuation line */
366};
367
368struct 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
379}
380#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
381__packed __aligned(4)
382#endif
383;
384
385/*
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.
389 */
390DEFINE_RAW_SPINLOCK(logbuf_lock);
391
392/*
393 * Helper macros to lock/unlock logbuf_lock and switch between
394 * printk-safe/unsafe modes.
395 */
396#define logbuf_lock_irq() \
397 do { \
398 printk_safe_enter_irq(); \
399 raw_spin_lock(&logbuf_lock); \
400 } while (0)
401
402#define logbuf_unlock_irq() \
403 do { \
404 raw_spin_unlock(&logbuf_lock); \
405 printk_safe_exit_irq(); \
406 } while (0)
407
408#define logbuf_lock_irqsave(flags) \
409 do { \
410 printk_safe_enter_irqsave(flags); \
411 raw_spin_lock(&logbuf_lock); \
412 } while (0)
413
414#define logbuf_unlock_irqrestore(flags) \
415 do { \
416 raw_spin_unlock(&logbuf_lock); \
417 printk_safe_exit_irqrestore(flags); \
418 } while (0)
419
420#ifdef CONFIG_PRINTK
421DECLARE_WAIT_QUEUE_HEAD(log_wait);
422/* the next printk record to read by syslog(READ) or /proc/kmsg */
423static u64 syslog_seq;
424static u32 syslog_idx;
425static size_t syslog_partial;
426static bool syslog_time;
427
428/* index and sequence number of the first record stored in the buffer */
429static u64 log_first_seq;
430static u32 log_first_idx;
431
432/* index and sequence number of the next record to store in the buffer */
433static u64 log_next_seq;
434static u32 log_next_idx;
435
436/* the next printk record to write to the console */
437static u64 console_seq;
438static u32 console_idx;
439static u64 exclusive_console_stop_seq;
440
441/* the next printk record to read after the last 'clear' command */
442static u64 clear_seq;
443static u32 clear_idx;
444
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)
451
452#define LOG_LEVEL(v) ((v) & 0x07)
453#define LOG_FACILITY(v) ((v) >> 3 & 0xff)
454
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)
459static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
460static char *log_buf = __log_buf;
461static u32 log_buf_len = __LOG_BUF_LEN;
462
463/* Return log buffer address */
464char *log_buf_addr_get(void)
465{
466 return log_buf;
467}
468
469/* Return log buffer size */
470u32 log_buf_len_get(void)
471{
472 return log_buf_len;
473}
474
475/* human readable text of the record */
476static char *log_text(const struct printk_log *msg)
477{
478 return (char *)msg + sizeof(struct printk_log);
479}
480
481/* optional key/value pair dictionary attached to the record */
482static char *log_dict(const struct printk_log *msg)
483{
484 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
485}
486
487/* get record by index; idx must point to valid msg */
488static struct printk_log *log_from_idx(u32 idx)
489{
490 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
491
492 /*
493 * A length == 0 record is the end of buffer marker. Wrap around and
494 * read the message at the start of the buffer.
495 */
496 if (!msg->len)
497 return (struct printk_log *)log_buf;
498 return msg;
499}
500
501/* get next record; idx must point to valid msg */
502static u32 log_next(u32 idx)
503{
504 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
505
506 /* length == 0 indicates the end of the buffer; wrap */
507 /*
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.
511 */
512 if (!msg->len) {
513 msg = (struct printk_log *)log_buf;
514 return msg->len;
515 }
516 return idx + msg->len;
517}
518
519/*
520 * Check whether there is enough free space for the given message.
521 *
522 * The same values of first_idx and next_idx mean that the buffer
523 * is either empty or full.
524 *
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.
527 */
528static int logbuf_has_space(u32 msg_size, bool empty)
529{
530 u32 free;
531
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;
536
537 /*
538 * We need space also for an empty header that signalizes wrapping
539 * of the buffer.
540 */
541 return free >= msg_size + sizeof(struct printk_log);
542}
543
544static int log_make_free_space(u32 msg_size)
545{
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++;
551 }
552
553 if (clear_seq < log_first_seq) {
554 clear_seq = log_first_seq;
555 clear_idx = log_first_idx;
556 }
557
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;
561
562 return -ENOMEM;
563}
564
565/* compute the message size including the padding bytes */
566static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
567{
568 u32 size;
569
570 size = sizeof(struct printk_log) + text_len + dict_len;
571 *pad_len = (-size) & (LOG_ALIGN - 1);
572 size += *pad_len;
573
574 return size;
575}
576
577/*
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.
581 */
582#define MAX_LOG_TAKE_PART 4
583static const char trunc_msg[] = "<truncated>";
584
585static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
586 u16 *dict_len, u32 *pad_len)
587{
588 /*
589 * The message should not take the whole buffer. Otherwise, it might
590 * get removed too soon.
591 */
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);
601}
602
603/* insert record into the buffer, discard old ones, update heads */
604static 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)
608{
609 struct printk_log *msg;
610 u32 size, pad_len;
611 u16 trunc_msg_len = 0;
612
613 /* number of '\0' padding bytes to next message */
614 size = msg_used_size(text_len, dict_len, &pad_len);
615
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;
623 }
624
625 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
626 /*
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.
630 */
631 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
632 log_next_idx = 0;
633 }
634
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;
642 }
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;
657
658 /* insert message */
659 log_next_idx += msg->len;
660 log_next_seq++;
661
662 return msg->text_len;
663}
664
665int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
666
667static int syslog_action_restricted(int type)
668{
669 if (dmesg_restrict)
670 return 1;
671 /*
672 * Unless restricted, we allow "read all" and "get buffer size"
673 * for everybody.
674 */
675 return type != SYSLOG_ACTION_READ_ALL &&
676 type != SYSLOG_ACTION_SIZE_BUFFER;
677}
678
679static int check_syslog_permissions(int type, int source)
680{
681 /*
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.
684 */
685 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
686 goto ok;
687
688 if (syslog_action_restricted(type)) {
689 if (capable(CAP_SYSLOG))
690 goto ok;
691 /*
692 * For historical reasons, accept CAP_SYS_ADMIN too, with
693 * a warning.
694 */
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;
701 }
702 return -EPERM;
703 }
704ok:
705 return security_syslog(type);
706}
707
708static void append_char(char **pp, char *e, char c)
709{
710 if (*pp < e)
711 *(*pp)++ = c;
712}
713
714static ssize_t msg_print_ext_header(char *buf, size_t size,
715 struct printk_log *msg, u64 seq)
716{
717 u64 ts_usec = msg->ts_nsec;
718 char caller[20];
719#ifdef CONFIG_PRINTK_CALLER
720 u32 id = msg->caller_id;
721
722 snprintf(caller, sizeof(caller), ",caller=%c%u",
723 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
724#else
725 caller[0] = '\0';
726#endif
727
728 do_div(ts_usec, 1000);
729
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);
733}
734
735static 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)
738{
739 char *p = buf, *e = buf + size;
740 size_t i;
741
742 /* escape non-printable characters */
743 for (i = 0; i < text_len; i++) {
744 unsigned char c = text[i];
745
746 if (c < ' ' || c >= 127 || c == '\\')
747 p += scnprintf(p, e - p, "\\x%02x", c);
748 else
749 append_char(&p, e, c);
750 }
751 append_char(&p, e, '\n');
752
753 if (dict_len) {
754 bool line = true;
755
756 for (i = 0; i < dict_len; i++) {
757 unsigned char c = dict[i];
758
759 if (line) {
760 append_char(&p, e, ' ');
761 line = false;
762 }
763
764 if (c == '\0') {
765 append_char(&p, e, '\n');
766 line = true;
767 continue;
768 }
769
770 if (c < ' ' || c >= 127 || c == '\\') {
771 p += scnprintf(p, e - p, "\\x%02x", c);
772 continue;
773 }
774
775 append_char(&p, e, c);
776 }
777 append_char(&p, e, '\n');
778 }
779
780 return p - buf;
781}
782
783/* /dev/kmsg - userspace message inject/listen interface */
784struct devkmsg_user {
785 u64 seq;
786 u32 idx;
787 struct ratelimit_state rs;
788 struct mutex lock;
789 char buf[CONSOLE_EXT_LOG_MAX];
790};
791
792static __printf(3, 4) __cold
793int devkmsg_emit(int facility, int level, const char *fmt, ...)
794{
795 va_list args;
796 int r;
797
798 va_start(args, fmt);
799 r = vprintk_emit(facility, level, NULL, 0, fmt, args);
800 va_end(args);
801
802 return r;
803}
804
805static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
806{
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;
814
815 if (!user || len > LOG_LINE_MAX)
816 return -EINVAL;
817
818 /* Ignore when user logging is disabled. */
819 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
820 return len;
821
822 /* Ratelimit when not explicitly enabled. */
823 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
824 if (!___ratelimit(&user->rs, current->comm))
825 return ret;
826 }
827
828 buf = kmalloc(len+1, GFP_KERNEL);
829 if (buf == NULL)
830 return -ENOMEM;
831
832 buf[len] = '\0';
833 if (!copy_from_iter_full(buf, len, from)) {
834 kfree(buf);
835 return -EFAULT;
836 }
837
838 /*
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.
842 *
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.
846 */
847 line = buf;
848 if (line[0] == '<') {
849 char *endp = NULL;
850 unsigned int u;
851
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;
860 }
861 }
862
863 devkmsg_emit(facility, level, "%s", line);
864 kfree(buf);
865 return ret;
866}
867
868static ssize_t devkmsg_read(struct file *file, char __user *buf,
869 size_t count, loff_t *ppos)
870{
871 struct devkmsg_user *user = file->private_data;
872 struct printk_log *msg;
873 size_t len;
874 ssize_t ret;
875
876 if (!user)
877 return -EBADF;
878
879 ret = mutex_lock_interruptible(&user->lock);
880 if (ret)
881 return ret;
882
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;
889 }
890
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();
897 }
898
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;
906 }
907
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);
914
915 user->idx = log_next(user->idx);
916 user->seq++;
917 logbuf_unlock_irq();
918
919 if (len > count) {
920 ret = -EINVAL;
921 goto out;
922 }
923
924 if (copy_to_user(buf, user->buf, len)) {
925 ret = -EFAULT;
926 goto out;
927 }
928 ret = len;
929out:
930 mutex_unlock(&user->lock);
931 return ret;
932}
933
934static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
935{
936 struct devkmsg_user *user = file->private_data;
937 loff_t ret = 0;
938
939 if (!user)
940 return -EBADF;
941 if (offset)
942 return -ESPIPE;
943
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:
952 /*
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.
956 */
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;
967 }
968 logbuf_unlock_irq();
969 return ret;
970}
971
972static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
973{
974 struct devkmsg_user *user = file->private_data;
975 __poll_t ret = 0;
976
977 if (!user)
978 return EPOLLERR|EPOLLNVAL;
979
980 poll_wait(file, &log_wait, wait);
981
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;
989 }
990 logbuf_unlock_irq();
991
992 return ret;
993}
994
995static int devkmsg_open(struct inode *inode, struct file *file)
996{
997 struct devkmsg_user *user;
998 int err;
999
1000 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
1001 return -EPERM;
1002
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;
1009 }
1010
1011 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
1012 if (!user)
1013 return -ENOMEM;
1014
1015 ratelimit_default_init(&user->rs);
1016 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
1017
1018 mutex_init(&user->lock);
1019
1020 logbuf_lock_irq();
1021 user->idx = log_first_idx;
1022 user->seq = log_first_seq;
1023 logbuf_unlock_irq();
1024
1025 file->private_data = user;
1026 return 0;
1027}
1028
1029static int devkmsg_release(struct inode *inode, struct file *file)
1030{
1031 struct devkmsg_user *user = file->private_data;
1032
1033 if (!user)
1034 return 0;
1035
1036 ratelimit_state_exit(&user->rs);
1037
1038 mutex_destroy(&user->lock);
1039 kfree(user);
1040 return 0;
1041}
1042
1043const 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,
1050};
1051
1052#ifdef CONFIG_CRASH_CORE
1053/*
1054 * This appends the listed symbols to /proc/vmcore
1055 *
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.
1060 */
1061void log_buf_vmcoreinfo_setup(void)
1062{
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);
1068 /*
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.
1071 */
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
1080}
1081#endif
1082
1083/* requested log_buf_len from kernel cmdline */
1084static unsigned long __initdata new_log_buf_len;
1085
1086/* we practice scaling the ring buffer by powers of 2 */
1087static void __init log_buf_len_update(u64 size)
1088{
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");
1092 }
1093
1094 if (size)
1095 size = roundup_pow_of_two(size);
1096 if (size > log_buf_len)
1097 new_log_buf_len = (unsigned long)size;
1098}
1099
1100/* save requested log_buf_len since it's too early to process it */
1101static int __init log_buf_len_setup(char *str)
1102{
1103 u64 size;
1104
1105 if (!str)
1106 return -EINVAL;
1107
1108 size = memparse(str, &str);
1109
1110 log_buf_len_update(size);
1111
1112 return 0;
1113}
1114early_param("log_buf_len", log_buf_len_setup);
1115
1116#ifdef CONFIG_SMP
1117#define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1118
1119static void __init log_buf_add_cpu(void)
1120{
1121 unsigned int cpu_extra;
1122
1123 /*
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.
1127 */
1128 if (num_possible_cpus() == 1)
1129 return;
1130
1131 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1132
1133 /* by default this will only continue through for large > 64 CPUs */
1134 if (cpu_extra <= __LOG_BUF_LEN / 2)
1135 return;
1136
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);
1142
1143 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1144}
1145#else /* !CONFIG_SMP */
1146static inline void log_buf_add_cpu(void) {}
1147#endif /* CONFIG_SMP */
1148
1149void __init setup_log_buf(int early)
1150{
1151 unsigned long flags;
1152 char *new_log_buf;
1153 unsigned int free;
1154
1155 if (log_buf != __log_buf)
1156 return;
1157
1158 if (!early && !new_log_buf_len)
1159 log_buf_add_cpu();
1160
1161 if (!new_log_buf_len)
1162 return;
1163
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;
1169 }
1170
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);
1178
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);
1182}
1183
1184static bool __read_mostly ignore_loglevel;
1185
1186static int __init ignore_loglevel_setup(char *str)
1187{
1188 ignore_loglevel = true;
1189 pr_info("debug: ignoring loglevel setting.\n");
1190
1191 return 0;
1192}
1193
1194early_param("ignore_loglevel", ignore_loglevel_setup);
1195module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1196MODULE_PARM_DESC(ignore_loglevel,
1197 "ignore loglevel setting (prints all kernel messages to the console)");
1198
1199static bool suppress_message_printing(int level)
1200{
1201 return (level >= console_loglevel && !ignore_loglevel);
1202}
1203
1204#ifdef CONFIG_BOOT_PRINTK_DELAY
1205
1206static int boot_delay; /* msecs delay after each printk during bootup */
1207static unsigned long long loops_per_msec; /* based on boot_delay */
1208
1209static int __init boot_delay_setup(char *str)
1210{
1211 unsigned long lpj;
1212
1213 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1214 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1215
1216 get_option(&str, &boot_delay);
1217 if (boot_delay > 10 * 1000)
1218 boot_delay = 0;
1219
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;
1224}
1225early_param("boot_delay", boot_delay_setup);
1226
1227static void boot_delay_msec(int level)
1228{
1229 unsigned long long k;
1230 unsigned long timeout;
1231
1232 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1233 || suppress_message_printing(level)) {
1234 return;
1235 }
1236
1237 k = (unsigned long long)loops_per_msec * boot_delay;
1238
1239 timeout = jiffies + msecs_to_jiffies(boot_delay);
1240 while (k) {
1241 k--;
1242 cpu_relax();
1243 /*
1244 * use (volatile) jiffies to prevent
1245 * compiler reduction; loop termination via jiffies
1246 * is secondary and may or may not happen.
1247 */
1248 if (time_after(jiffies, timeout))
1249 break;
1250 touch_nmi_watchdog();
1251 }
1252}
1253#else
1254static inline void boot_delay_msec(int level)
1255{
1256}
1257#endif
1258
1259static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1260module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1261
1262static size_t print_syslog(unsigned int level, char *buf)
1263{
1264 return sprintf(buf, "<%u>", level);
1265}
1266
1267static size_t print_time(u64 ts, char *buf)
1268{
1269 unsigned long rem_nsec = do_div(ts, 1000000000);
1270
1271 return sprintf(buf, "[%5lu.%06lu]",
1272 (unsigned long)ts, rem_nsec / 1000);
1273}
1274
1275#ifdef CONFIG_PRINTK_CALLER
1276static size_t print_caller(u32 id, char *buf)
1277{
1278 char caller[12];
1279
1280 snprintf(caller, sizeof(caller), "%c%u",
1281 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1282 return sprintf(buf, "[%6s]", caller);
1283}
1284#else
1285#define print_caller(id, buf) 0
1286#endif
1287
1288static size_t print_prefix(const struct printk_log *msg, bool syslog,
1289 bool time, char *buf)
1290{
1291 size_t len = 0;
1292
1293 if (syslog)
1294 len = print_syslog((msg->facility << 3) | msg->level, buf);
1295
1296 if (time)
1297 len += print_time(msg->ts_nsec, buf + len);
1298
1299 len += print_caller(msg->caller_id, buf + len);
1300
1301 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1302 buf[len++] = ' ';
1303 buf[len] = '\0';
1304 }
1305
1306 return len;
1307}
1308
1309static size_t msg_print_text(const struct printk_log *msg, bool syslog,
1310 bool time, char *buf, size_t size)
1311{
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);
1317
1318 do {
1319 const char *next = memchr(text, '\n', text_size);
1320 size_t text_len;
1321
1322 if (next) {
1323 text_len = next - text;
1324 next++;
1325 text_size -= next - text;
1326 } else {
1327 text_len = text_size;
1328 }
1329
1330 if (buf) {
1331 if (prefix_len + text_len + 1 >= size - len)
1332 break;
1333
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;
1342 }
1343
1344 text = next;
1345 } while (text);
1346
1347 return len;
1348}
1349
1350static int syslog_print(char __user *buf, int size)
1351{
1352 char *text;
1353 struct printk_log *msg;
1354 int len = 0;
1355
1356 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1357 if (!text)
1358 return -ENOMEM;
1359
1360 while (size > 0) {
1361 size_t n;
1362 size_t skip;
1363
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;
1370 }
1371 if (syslog_seq == log_next_seq) {
1372 logbuf_unlock_irq();
1373 break;
1374 }
1375
1376 /*
1377 * To keep reading/counting partial line consistent,
1378 * use printk_time value as of the beginning of a line.
1379 */
1380 if (!syslog_partial)
1381 syslog_time = printk_time;
1382
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();
1400
1401 if (!n)
1402 break;
1403
1404 if (copy_to_user(buf, text + skip, n)) {
1405 if (!len)
1406 len = -EFAULT;
1407 break;
1408 }
1409
1410 len += n;
1411 size -= n;
1412 buf += n;
1413 }
1414
1415 kfree(text);
1416 return len;
1417}
1418
1419static int syslog_print_all(char __user *buf, int size, bool clear)
1420{
1421 char *text;
1422 int len = 0;
1423 u64 next_seq;
1424 u64 seq;
1425 u32 idx;
1426 bool time;
1427
1428 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1429 if (!text)
1430 return -ENOMEM;
1431
1432 time = printk_time;
1433 logbuf_lock_irq();
1434 /*
1435 * Find first record that fits, including all following records,
1436 * into the user-provided buffer for this dump.
1437 */
1438 seq = clear_seq;
1439 idx = clear_idx;
1440 while (seq < log_next_seq) {
1441 struct printk_log *msg = log_from_idx(idx);
1442
1443 len += msg_print_text(msg, true, time, NULL, 0);
1444 idx = log_next(idx);
1445 seq++;
1446 }
1447
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);
1453
1454 len -= msg_print_text(msg, true, time, NULL, 0);
1455 idx = log_next(idx);
1456 seq++;
1457 }
1458
1459 /* last message fitting into this dump */
1460 next_seq = log_next_seq;
1461
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);
1467
1468 idx = log_next(idx);
1469 seq++;
1470
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();
1477
1478 if (seq < log_first_seq) {
1479 /* messages are gone, move to next one */
1480 seq = log_first_seq;
1481 idx = log_first_idx;
1482 }
1483 }
1484
1485 if (clear) {
1486 clear_seq = log_next_seq;
1487 clear_idx = log_next_idx;
1488 }
1489 logbuf_unlock_irq();
1490
1491 kfree(text);
1492 return len;
1493}
1494
1495static void syslog_clear(void)
1496{
1497 logbuf_lock_irq();
1498 clear_seq = log_next_seq;
1499 clear_idx = log_next_idx;
1500 logbuf_unlock_irq();
1501}
1502
1503int do_syslog(int type, char __user *buf, int len, int source)
1504{
1505 bool clear = false;
1506 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1507 int error;
1508
1509 error = check_syslog_permissions(type, source);
1510 if (error)
1511 return error;
1512
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;
1560 }
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;
1580 }
1581 if (source == SYSLOG_FROM_PROC) {
1582 /*
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.
1586 */
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;
1592
1593 while (seq < log_next_seq) {
1594 struct printk_log *msg = log_from_idx(idx);
1595
1596 error += msg_print_text(msg, true, time, NULL,
1597 0);
1598 time = printk_time;
1599 idx = log_next(idx);
1600 seq++;
1601 }
1602 error -= syslog_partial;
1603 }
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;
1613 }
1614
1615 return error;
1616}
1617
1618SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1619{
1620 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1621}
1622
1623/*
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.
1626 */
1627
1628#ifdef CONFIG_LOCKDEP
1629static struct lockdep_map console_owner_dep_map = {
1630 .name = "console_owner"
1631};
1632#endif
1633
1634static DEFINE_RAW_SPINLOCK(console_owner_lock);
1635static struct task_struct *console_owner;
1636static bool console_waiter;
1637
1638/**
1639 * console_lock_spinning_enable - mark beginning of code where another
1640 * thread might safely busy wait
1641 *
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.
1646 */
1647static void console_lock_spinning_enable(void)
1648{
1649 raw_spin_lock(&console_owner_lock);
1650 console_owner = current;
1651 raw_spin_unlock(&console_owner_lock);
1652
1653 /* The waiter may spin on us after setting console_owner */
1654 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1655}
1656
1657/**
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
1660 *
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.
1665 *
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.
1669 *
1670 * Return: 1 if the lock rights were passed, 0 otherwise.
1671 */
1672static int console_lock_spinning_disable_and_check(void)
1673{
1674 int waiter;
1675
1676 raw_spin_lock(&console_owner_lock);
1677 waiter = READ_ONCE(console_waiter);
1678 console_owner = NULL;
1679 raw_spin_unlock(&console_owner_lock);
1680
1681 if (!waiter) {
1682 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1683 return 0;
1684 }
1685
1686 /* The waiter is now free to continue */
1687 WRITE_ONCE(console_waiter, false);
1688
1689 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1690
1691 /*
1692 * Hand off console_lock to waiter. The waiter will perform
1693 * the up(). After this, the waiter is the console_lock owner.
1694 */
1695 mutex_release(&console_lock_dep_map, 1, _THIS_IP_);
1696 return 1;
1697}
1698
1699/**
1700 * console_trylock_spinning - try to get console_lock by busy waiting
1701 *
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.
1706 *
1707 * Return: 1 if we got the lock, 0 othrewise
1708 */
1709static int console_trylock_spinning(void)
1710{
1711 struct task_struct *owner = NULL;
1712 bool waiter;
1713 bool spin = false;
1714 unsigned long flags;
1715
1716 if (console_trylock())
1717 return 1;
1718
1719 printk_safe_enter_irqsave(flags);
1720
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;
1727 }
1728 raw_spin_unlock(&console_owner_lock);
1729
1730 /*
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?).
1738 */
1739 if (!spin) {
1740 printk_safe_exit_irqrestore(flags);
1741 return 0;
1742 }
1743
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, 1, _THIS_IP_);
1750
1751 printk_safe_exit_irqrestore(flags);
1752 /*
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.
1757 */
1758 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1759
1760 return 1;
1761}
1762
1763/*
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.
1767 */
1768static void call_console_drivers(const char *ext_text, size_t ext_len,
1769 const char *text, size_t len)
1770{
1771 struct console *con;
1772
1773 trace_console_rcuidle(text, len);
1774
1775 if (!console_drivers)
1776 return;
1777
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);
1792 }
1793}
1794
1795int printk_delay_msec __read_mostly;
1796
1797static inline void printk_delay(void)
1798{
1799 if (unlikely(printk_delay_msec)) {
1800 int m = printk_delay_msec;
1801
1802 while (m--) {
1803 mdelay(1);
1804 touch_nmi_watchdog();
1805 }
1806 }
1807}
1808
1809static inline u32 printk_caller_id(void)
1810{
1811 return in_task() ? task_pid_nr(current) :
1812 0x80000000 + raw_smp_processor_id();
1813}
1814
1815/*
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.
1820 */
1821static 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;
1830
1831static void cont_flush(void)
1832{
1833 if (cont.len == 0)
1834 return;
1835
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;
1839}
1840
1841static bool cont_add(u32 caller_id, int facility, int level,
1842 enum log_flags flags, const char *text, size_t len)
1843{
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;
1848 }
1849
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;
1856 }
1857
1858 memcpy(cont.buf + cont.len, text, len);
1859 cont.len += len;
1860
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();
1866 }
1867
1868 return true;
1869}
1870
1871static size_t log_output(int facility, int level, enum log_flags lflags, const char *dict, size_t dictlen, char *text, size_t text_len)
1872{
1873 const u32 caller_id = printk_caller_id();
1874
1875 /*
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.
1878 */
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;
1883 }
1884 /* Otherwise, make sure it's flushed */
1885 cont_flush();
1886 }
1887
1888 /* Skip empty continuation lines that couldn't be added - they just flush */
1889 if (!text_len && (lflags & LOG_CONT))
1890 return 0;
1891
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;
1896 }
1897
1898 /* Store it in the record log */
1899 return log_store(caller_id, facility, level, lflags, 0,
1900 dict, dictlen, text, text_len);
1901}
1902
1903/* Must be called under logbuf_lock. */
1904int vprintk_store(int facility, int level,
1905 const char *dict, size_t dictlen,
1906 const char *fmt, va_list args)
1907{
1908 static char textbuf[LOG_LINE_MAX];
1909 char *text = textbuf;
1910 size_t text_len;
1911 enum log_flags lflags = 0;
1912
1913 /*
1914 * The printf needs to come first; we need the syslog
1915 * prefix which might be passed-in as a parameter.
1916 */
1917 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1918
1919 /* mark and strip a trailing newline */
1920 if (text_len && text[text_len-1] == '\n') {
1921 text_len--;
1922 lflags |= LOG_NEWLINE;
1923 }
1924
1925 /* strip kernel syslog prefix and extract log level or control flags */
1926 if (facility == 0) {
1927 int kern_level;
1928
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;
1937 }
1938
1939 text_len -= 2;
1940 text += 2;
1941 }
1942 }
1943
1944 if (level == LOGLEVEL_DEFAULT)
1945 level = default_message_loglevel;
1946
1947 if (dict)
1948 lflags |= LOG_NEWLINE;
1949
1950 return log_output(facility, level, lflags,
1951 dict, dictlen, text, text_len);
1952}
1953
1954asmlinkage int vprintk_emit(int facility, int level,
1955 const char *dict, size_t dictlen,
1956 const char *fmt, va_list args)
1957{
1958 int printed_len;
1959 bool in_sched = false, pending_output;
1960 unsigned long flags;
1961 u64 curr_log_seq;
1962
1963 /* Suppress unimportant messages after panic happens */
1964 if (unlikely(suppress_printk))
1965 return 0;
1966
1967 if (level == LOGLEVEL_SCHED) {
1968 level = LOGLEVEL_DEFAULT;
1969 in_sched = true;
1970 }
1971
1972 boot_delay_msec(level);
1973 printk_delay();
1974
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);
1981
1982 /* If called from the scheduler, we can not call up(). */
1983 if (!in_sched && pending_output) {
1984 /*
1985 * Disable preemption to avoid being preempted while holding
1986 * console_sem which would prevent anyone from printing to
1987 * console
1988 */
1989 preempt_disable();
1990 /*
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.
1994 */
1995 if (console_trylock_spinning())
1996 console_unlock();
1997 preempt_enable();
1998 }
1999
2000 if (pending_output)
2001 wake_up_klogd();
2002 return printed_len;
2003}
2004EXPORT_SYMBOL(vprintk_emit);
2005
2006asmlinkage int vprintk(const char *fmt, va_list args)
2007{
2008 return vprintk_func(fmt, args);
2009}
2010EXPORT_SYMBOL(vprintk);
2011
2012int vprintk_default(const char *fmt, va_list args)
2013{
2014 int r;
2015
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;
2021 }
2022#endif
2023 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
2024
2025 return r;
2026}
2027EXPORT_SYMBOL_GPL(vprintk_default);
2028
2029/**
2030 * printk - print a kernel message
2031 * @fmt: format string
2032 *
2033 * This is printk(). It can be called from any context. We want it to work.
2034 *
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.
2040 *
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.
2044 *
2045 * See also:
2046 * printf(3)
2047 *
2048 * See the vsnprintf() documentation for format string extensions over C99.
2049 */
2050asmlinkage __visible int printk(const char *fmt, ...)
2051{
2052 va_list args;
2053 int r;
2054
2055 va_start(args, fmt);
2056 r = vprintk_func(fmt, args);
2057 va_end(args);
2058
2059 return r;
2060}
2061EXPORT_SYMBOL(printk);
2062
2063#else /* CONFIG_PRINTK */
2064
2065#define LOG_LINE_MAX 0
2066#define PREFIX_MAX 0
2067#define printk_time false
2068
2069static u64 syslog_seq;
2070static u32 syslog_idx;
2071static u64 console_seq;
2072static u32 console_idx;
2073static u64 exclusive_console_stop_seq;
2074static u64 log_first_seq;
2075static u32 log_first_idx;
2076static u64 log_next_seq;
2077static char *log_text(const struct printk_log *msg) { return NULL; }
2078static char *log_dict(const struct printk_log *msg) { return NULL; }
2079static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2080static u32 log_next(u32 idx) { return 0; }
2081static ssize_t msg_print_ext_header(char *buf, size_t size,
2082 struct printk_log *msg,
2083 u64 seq) { return 0; }
2084static 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; }
2087static void console_lock_spinning_enable(void) { }
2088static int console_lock_spinning_disable_and_check(void) { return 0; }
2089static void call_console_drivers(const char *ext_text, size_t ext_len,
2090 const char *text, size_t len) {}
2091static size_t msg_print_text(const struct printk_log *msg, bool syslog,
2092 bool time, char *buf, size_t size) { return 0; }
2093static bool suppress_message_printing(int level) { return false; }
2094
2095#endif /* CONFIG_PRINTK */
2096
2097#ifdef CONFIG_EARLY_PRINTK
2098struct console *early_console;
2099
2100asmlinkage __visible void early_printk(const char *fmt, ...)
2101{
2102 va_list ap;
2103 char buf[512];
2104 int n;
2105
2106 if (!early_console)
2107 return;
2108
2109 va_start(ap, fmt);
2110 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2111 va_end(ap);
2112
2113 early_console->write(early_console, buf, n);
2114}
2115#endif
2116
2117static int __add_preferred_console(char *name, int idx, char *options,
2118 char *brl_options)
2119{
2120 struct console_cmdline *c;
2121 int i;
2122
2123 /*
2124 * See if this tty is not yet registered, and
2125 * if we have a slot free.
2126 */
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;
2134 }
2135 }
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);
2143
2144 c->index = idx;
2145 return 0;
2146}
2147
2148static int __init console_msg_format_setup(char *str)
2149{
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;
2155}
2156__setup("console_msg_format=", console_msg_format_setup);
2157
2158/*
2159 * Set up a console. Called via do_early_param() in init/main.c
2160 * for each "console=" parameter in the boot command line.
2161 */
2162static int __init console_setup(char *str)
2163{
2164 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2165 char *s, *options, *brl_options = NULL;
2166 int idx;
2167
2168 if (_braille_console_setup(&str, &brl_options))
2169 return 1;
2170
2171 /*
2172 * Decode str into name, index, options.
2173 */
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);
2179 }
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;
2195
2196 __add_preferred_console(buf, idx, options, brl_options);
2197 console_set_on_cmdline = 1;
2198 return 1;
2199}
2200__setup("console=", console_setup);
2201
2202/**
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
2207 *
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.
2214 */
2215int add_preferred_console(char *name, int idx, char *options)
2216{
2217 return __add_preferred_console(name, idx, options, NULL);
2218}
2219
2220bool console_suspend_enabled = true;
2221EXPORT_SYMBOL(console_suspend_enabled);
2222
2223static int __init console_suspend_disable(char *str)
2224{
2225 console_suspend_enabled = false;
2226 return 1;
2227}
2228__setup("no_console_suspend", console_suspend_disable);
2229module_param_named(console_suspend, console_suspend_enabled,
2230 bool, S_IRUGO | S_IWUSR);
2231MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2232 " and hibernate operations");
2233
2234/**
2235 * suspend_console - suspend the console subsystem
2236 *
2237 * This disables printk() while we go into suspend states
2238 */
2239void suspend_console(void)
2240{
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();
2247}
2248
2249void resume_console(void)
2250{
2251 if (!console_suspend_enabled)
2252 return;
2253 down_console_sem();
2254 console_suspended = 0;
2255 console_unlock();
2256}
2257
2258/**
2259 * console_cpu_notify - print deferred console messages after CPU hotplug
2260 * @cpu: unused
2261 *
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.
2266 */
2267static int console_cpu_notify(unsigned int cpu)
2268{
2269 if (!cpuhp_tasks_frozen) {
2270 /* If trylock fails, someone else is doing the printing */
2271 if (console_trylock())
2272 console_unlock();
2273 }
2274 return 0;
2275}
2276
2277/**
2278 * console_lock - lock the console system for exclusive use.
2279 *
2280 * Acquires a lock which guarantees that the caller has
2281 * exclusive access to the console system and the console_drivers list.
2282 *
2283 * Can sleep, returns nothing.
2284 */
2285void console_lock(void)
2286{
2287 might_sleep();
2288
2289 down_console_sem();
2290 if (console_suspended)
2291 return;
2292 console_locked = 1;
2293 console_may_schedule = 1;
2294}
2295EXPORT_SYMBOL(console_lock);
2296
2297/**
2298 * console_trylock - try to lock the console system for exclusive use.
2299 *
2300 * Try to acquire a lock which guarantees that the caller has exclusive
2301 * access to the console system and the console_drivers list.
2302 *
2303 * returns 1 on success, and 0 on failure to acquire the lock.
2304 */
2305int console_trylock(void)
2306{
2307 if (down_trylock_console_sem())
2308 return 0;
2309 if (console_suspended) {
2310 up_console_sem();
2311 return 0;
2312 }
2313 console_locked = 1;
2314 console_may_schedule = 0;
2315 return 1;
2316}
2317EXPORT_SYMBOL(console_trylock);
2318
2319int is_console_locked(void)
2320{
2321 return console_locked;
2322}
2323EXPORT_SYMBOL(is_console_locked);
2324
2325/*
2326 * Check if we have any console that is capable of printing while cpu is
2327 * booting or shutting down. Requires console_sem.
2328 */
2329static int have_callable_console(void)
2330{
2331 struct console *con;
2332
2333 for_each_console(con)
2334 if ((con->flags & CON_ENABLED) &&
2335 (con->flags & CON_ANYTIME))
2336 return 1;
2337
2338 return 0;
2339}
2340
2341/*
2342 * Can we actually use the console at this time on this cpu?
2343 *
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.
2347 */
2348static inline int can_use_console(void)
2349{
2350 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2351}
2352
2353/**
2354 * console_unlock - unlock the console system
2355 *
2356 * Releases the console_lock which the caller holds on the console system
2357 * and the console driver list.
2358 *
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.
2362 *
2363 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2364 *
2365 * console_unlock(); may be called from any context.
2366 */
2367void console_unlock(void)
2368{
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;
2373
2374 if (console_suspended) {
2375 up_console_sem();
2376 return;
2377 }
2378
2379 /*
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.
2388 *
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.
2392 */
2393 do_cond_resched = console_may_schedule;
2394again:
2395 console_may_schedule = 0;
2396
2397 /*
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.
2401 */
2402 if (!can_use_console()) {
2403 console_locked = 0;
2404 up_console_sem();
2405 return;
2406 }
2407
2408 for (;;) {
2409 struct printk_log *msg;
2410 size_t ext_len = 0;
2411 size_t len;
2412
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);
2419
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;
2425 }
2426skip:
2427 if (console_seq == log_next_seq)
2428 break;
2429
2430 msg = log_from_idx(console_idx);
2431 if (suppress_message_printing(msg->level)) {
2432 /*
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.
2436 */
2437 console_idx = log_next(console_idx);
2438 console_seq++;
2439 goto skip;
2440 }
2441
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;
2446 }
2447
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);
2459 }
2460 console_idx = log_next(console_idx);
2461 console_seq++;
2462 raw_spin_unlock(&logbuf_lock);
2463
2464 /*
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.
2469 */
2470 console_lock_spinning_enable();
2471
2472 stop_critical_timings(); /* don't trace print latency */
2473 call_console_drivers(ext_text, ext_len, text, len);
2474 start_critical_timings();
2475
2476 if (console_lock_spinning_disable_and_check()) {
2477 printk_safe_exit_irqrestore(flags);
2478 return;
2479 }
2480
2481 printk_safe_exit_irqrestore(flags);
2482
2483 if (do_cond_resched)
2484 cond_resched();
2485 }
2486
2487 console_locked = 0;
2488
2489 raw_spin_unlock(&logbuf_lock);
2490
2491 up_console_sem();
2492
2493 /*
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.
2498 */
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);
2503
2504 if (retry && console_trylock())
2505 goto again;
2506}
2507EXPORT_SYMBOL(console_unlock);
2508
2509/**
2510 * console_conditional_schedule - yield the CPU if required
2511 *
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.
2515 *
2516 * Must be called within console_lock();.
2517 */
2518void __sched console_conditional_schedule(void)
2519{
2520 if (console_may_schedule)
2521 cond_resched();
2522}
2523EXPORT_SYMBOL(console_conditional_schedule);
2524
2525void console_unblank(void)
2526{
2527 struct console *c;
2528
2529 /*
2530 * console_unblank can no longer be called in interrupt context unless
2531 * oops_in_progress is set to 1..
2532 */
2533 if (oops_in_progress) {
2534 if (down_trylock_console_sem() != 0)
2535 return;
2536 } else
2537 console_lock();
2538
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();
2545}
2546
2547/**
2548 * console_flush_on_panic - flush console content on panic
2549 * @mode: flush all messages in buffer or just the pending ones
2550 *
2551 * Immediately output all pending messages no matter what.
2552 */
2553void console_flush_on_panic(enum con_flush_mode mode)
2554{
2555 /*
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.
2561 */
2562 console_trylock();
2563 console_may_schedule = 0;
2564
2565 if (mode == CONSOLE_REPLAY_ALL) {
2566 unsigned long flags;
2567
2568 logbuf_lock_irqsave(flags);
2569 console_seq = log_first_seq;
2570 console_idx = log_first_idx;
2571 logbuf_unlock_irqrestore(flags);
2572 }
2573 console_unlock();
2574}
2575
2576/*
2577 * Return the console tty driver structure and its associated index
2578 */
2579struct tty_driver *console_device(int *index)
2580{
2581 struct console *c;
2582 struct tty_driver *driver = NULL;
2583
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;
2591 }
2592 console_unlock();
2593 return driver;
2594}
2595
2596/*
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.
2600 */
2601void console_stop(struct console *console)
2602{
2603 console_lock();
2604 console->flags &= ~CON_ENABLED;
2605 console_unlock();
2606}
2607EXPORT_SYMBOL(console_stop);
2608
2609void console_start(struct console *console)
2610{
2611 console_lock();
2612 console->flags |= CON_ENABLED;
2613 console_unlock();
2614}
2615EXPORT_SYMBOL(console_start);
2616
2617static int __read_mostly keep_bootcon;
2618
2619static int __init keep_bootcon_setup(char *str)
2620{
2621 keep_bootcon = 1;
2622 pr_info("debug: skip boot console de-registration.\n");
2623
2624 return 0;
2625}
2626
2627early_param("keep_bootcon", keep_bootcon_setup);
2628
2629/*
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.
2634 *
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.
2638 *
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
2647 */
2648void register_console(struct console *newcon)
2649{
2650 int i;
2651 unsigned long flags;
2652 struct console *bcon = NULL;
2653 struct console_cmdline *c;
2654 static bool has_preferred;
2655
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;
2662
2663 /*
2664 * before we register a new CON_BOOT console, make sure we don't
2665 * already have a valid console
2666 */
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;
2674 }
2675 }
2676 }
2677
2678 if (console_drivers && console_drivers->flags & CON_BOOT)
2679 bcon = console_drivers;
2680
2681 if (!has_preferred || bcon || !console_drivers)
2682 has_preferred = preferred_console >= 0;
2683
2684 /*
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.
2688 */
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;
2698 }
2699 }
2700 }
2701
2702 /*
2703 * See if this console matches one we selected on
2704 * the command line.
2705 */
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;
2720
2721 if (_braille_register_console(newcon, c))
2722 return;
2723
2724 if (newcon->setup &&
2725 newcon->setup(newcon, c->options) != 0)
2726 break;
2727 }
2728
2729 newcon->flags |= CON_ENABLED;
2730 if (i == preferred_console) {
2731 newcon->flags |= CON_CONSDEV;
2732 has_preferred = true;
2733 }
2734 break;
2735 }
2736
2737 if (!(newcon->flags & CON_ENABLED))
2738 return;
2739
2740 /*
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
2745 */
2746 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2747 newcon->flags &= ~CON_PRINTBUFFER;
2748
2749 /*
2750 * Put this console in the list - keep the
2751 * preferred driver at the head of the list.
2752 */
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;
2762 }
2763
2764 if (newcon->flags & CON_EXTENDED)
2765 nr_ext_console_drivers++;
2766
2767 if (newcon->flags & CON_PRINTBUFFER) {
2768 /*
2769 * console_unlock(); will print out the buffered messages
2770 * for us.
2771 */
2772 logbuf_lock_irqsave(flags);
2773 console_seq = syslog_seq;
2774 console_idx = syslog_idx;
2775 /*
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.
2779 *
2780 * Set exclusive_console with disabled interrupts to reduce
2781 * race window with eventual console_flush_on_panic() that
2782 * ignores console_lock.
2783 */
2784 exclusive_console = newcon;
2785 exclusive_console_stop_seq = console_seq;
2786 logbuf_unlock_irqrestore(flags);
2787 }
2788 console_unlock();
2789 console_sysfs_notify();
2790
2791 /*
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)
2797 */
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.
2806 */
2807 for_each_console(bcon)
2808 if (bcon->flags & CON_BOOT)
2809 unregister_console(bcon);
2810 }
2811}
2812EXPORT_SYMBOL(register_console);
2813
2814int unregister_console(struct console *console)
2815{
2816 struct console *a, *b;
2817 int res;
2818
2819 pr_info("%sconsole [%s%d] disabled\n",
2820 (console->flags & CON_BOOT) ? "boot" : "" ,
2821 console->name, console->index);
2822
2823 res = _braille_unregister_console(console);
2824 if (res)
2825 return res;
2826
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;
2839 }
2840 }
2841 }
2842
2843 if (!res && (console->flags & CON_EXTENDED))
2844 nr_ext_console_drivers--;
2845
2846 /*
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.
2849 */
2850 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2851 console_drivers->flags |= CON_CONSDEV;
2852
2853 console->flags &= ~CON_ENABLED;
2854 console_unlock();
2855 console_sysfs_notify();
2856 return res;
2857}
2858EXPORT_SYMBOL(unregister_console);
2859
2860/*
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.
2865 */
2866void __init console_init(void)
2867{
2868 int ret;
2869 initcall_t call;
2870 initcall_entry_t *ce;
2871
2872 /* Setup the default TTY line discipline. */
2873 n_tty_init();
2874
2875 /*
2876 * set up the console device so that later boot sequences can
2877 * inform about problems etc..
2878 */
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++;
2887 }
2888}
2889
2890/*
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.
2894 *
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.
2899 *
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.
2903 */
2904static int __init printk_late_init(void)
2905{
2906 struct console *con;
2907 int ret;
2908
2909 for_each_console(con) {
2910 if (!(con->flags & CON_BOOT))
2911 continue;
2912
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)) {
2920 /*
2921 * Please, consider moving the reported consoles out
2922 * of the init section.
2923 */
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);
2927 }
2928 }
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;
2936}
2937late_initcall(printk_late_init);
2938
2939#if defined CONFIG_PRINTK
2940/*
2941 * Delayed printk version, for scheduler-internal messages:
2942 */
2943#define PRINTK_PENDING_WAKEUP 0x01
2944#define PRINTK_PENDING_OUTPUT 0x02
2945
2946static DEFINE_PER_CPU(int, printk_pending);
2947
2948static void wake_up_klogd_work_func(struct irq_work *irq_work)
2949{
2950 int pending = __this_cpu_xchg(printk_pending, 0);
2951
2952 if (pending & PRINTK_PENDING_OUTPUT) {
2953 /* If trylock fails, someone else is doing the printing */
2954 if (console_trylock())
2955 console_unlock();
2956 }
2957
2958 if (pending & PRINTK_PENDING_WAKEUP)
2959 wake_up_interruptible(&log_wait);
2960}
2961
2962static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2963 .func = wake_up_klogd_work_func,
2964 .flags = IRQ_WORK_LAZY,
2965};
2966
2967void wake_up_klogd(void)
2968{
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));
2973 }
2974 preempt_enable();
2975}
2976
2977void defer_console_output(void)
2978{
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();
2983}
2984
2985int vprintk_deferred(const char *fmt, va_list args)
2986{
2987 int r;
2988
2989 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2990 defer_console_output();
2991
2992 return r;
2993}
2994
2995int printk_deferred(const char *fmt, ...)
2996{
2997 va_list args;
2998 int r;
2999
3000 va_start(args, fmt);
3001 r = vprintk_deferred(fmt, args);
3002 va_end(args);
3003
3004 return r;
3005}
3006
3007/*
3008 * printk rate limiting, lifted from the networking subsystem.
3009 *
3010 * This enforces a rate limit: not more than 10 kernel messages
3011 * every 5s to make a denial-of-service attack impossible.
3012 */
3013DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3014
3015int __printk_ratelimit(const char *func)
3016{
3017 return ___ratelimit(&printk_ratelimit_state, func);
3018}
3019EXPORT_SYMBOL(__printk_ratelimit);
3020
3021/**
3022 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3023 * @caller_jiffies: pointer to caller's state
3024 * @interval_msecs: minimum interval between prints
3025 *
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.
3029 */
3030bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3031 unsigned int interval_msecs)
3032{
3033 unsigned long elapsed = jiffies - *caller_jiffies;
3034
3035 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3036 return false;
3037
3038 *caller_jiffies = jiffies;
3039 return true;
3040}
3041EXPORT_SYMBOL(printk_timed_ratelimit);
3042
3043static DEFINE_SPINLOCK(dump_list_lock);
3044static LIST_HEAD(dump_list);
3045
3046/**
3047 * kmsg_dump_register - register a kernel log dumper.
3048 * @dumper: pointer to the kmsg_dumper structure
3049 *
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.
3053 */
3054int kmsg_dump_register(struct kmsg_dumper *dumper)
3055{
3056 unsigned long flags;
3057 int err = -EBUSY;
3058
3059 /* The dump callback needs to be set */
3060 if (!dumper->dump)
3061 return -EINVAL;
3062
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;
3069 }
3070 spin_unlock_irqrestore(&dump_list_lock, flags);
3071
3072 return err;
3073}
3074EXPORT_SYMBOL_GPL(kmsg_dump_register);
3075
3076/**
3077 * kmsg_dump_unregister - unregister a kmsg dumper.
3078 * @dumper: pointer to the kmsg_dumper structure
3079 *
3080 * Removes a dump device from the system. Returns zero on success and
3081 * %-EINVAL otherwise.
3082 */
3083int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3084{
3085 unsigned long flags;
3086 int err = -EINVAL;
3087
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;
3093 }
3094 spin_unlock_irqrestore(&dump_list_lock, flags);
3095 synchronize_rcu();
3096
3097 return err;
3098}
3099EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3100
3101static bool always_kmsg_dump;
3102module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3103
3104/**
3105 * kmsg_dump - dump kernel log to kernel message dumpers.
3106 * @reason: the reason (oops, panic etc) for dumping
3107 *
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().
3111 */
3112void kmsg_dump(enum kmsg_dump_reason reason)
3113{
3114 struct kmsg_dumper *dumper;
3115 unsigned long flags;
3116
3117 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3118 return;
3119
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;
3124
3125 /* initialize iterator with data about the stored records */
3126 dumper->active = true;
3127
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);
3134
3135 /* invoke dumper which will iterate over records */
3136 dumper->dump(dumper, reason);
3137
3138 /* reset iterator */
3139 dumper->active = false;
3140 }
3141 rcu_read_unlock();
3142}
3143
3144/**
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
3151 *
3152 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3153 * record, and copy one record into the provided buffer.
3154 *
3155 * Consecutive calls will return the next available record moving
3156 * towards the end of the buffer with the youngest messages.
3157 *
3158 * A return value of FALSE indicates that there are no more records to
3159 * read.
3160 *
3161 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3162 */
3163bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3164 char *line, size_t size, size_t *len)
3165{
3166 struct printk_log *msg;
3167 size_t l = 0;
3168 bool ret = false;
3169
3170 if (!dumper->active)
3171 goto out;
3172
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;
3177 }
3178
3179 /* last entry */
3180 if (dumper->cur_seq >= log_next_seq)
3181 goto out;
3182
3183 msg = log_from_idx(dumper->cur_idx);
3184 l = msg_print_text(msg, syslog, printk_time, line, size);
3185
3186 dumper->cur_idx = log_next(dumper->cur_idx);
3187 dumper->cur_seq++;
3188 ret = true;
3189out:
3190 if (len)
3191 *len = l;
3192 return ret;
3193}
3194
3195/**
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
3202 *
3203 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3204 * record, and copy one record into the provided buffer.
3205 *
3206 * Consecutive calls will return the next available record moving
3207 * towards the end of the buffer with the youngest messages.
3208 *
3209 * A return value of FALSE indicates that there are no more records to
3210 * read.
3211 */
3212bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3213 char *line, size_t size, size_t *len)
3214{
3215 unsigned long flags;
3216 bool ret;
3217
3218 logbuf_lock_irqsave(flags);
3219 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3220 logbuf_unlock_irqrestore(flags);
3221
3222 return ret;
3223}
3224EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3225
3226/**
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
3233 *
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.
3238 *
3239 * Consecutive calls will fill the buffer with the next block of
3240 * available older records, not including the earlier retrieved ones.
3241 *
3242 * A return value of FALSE indicates that there are no more records to
3243 * read.
3244 */
3245bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3246 char *buf, size_t size, size_t *len)
3247{
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;
3256
3257 if (!dumper->active)
3258 goto out;
3259
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;
3265 }
3266
3267 /* last entry */
3268 if (dumper->cur_seq >= dumper->next_seq) {
3269 logbuf_unlock_irqrestore(flags);
3270 goto out;
3271 }
3272
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);
3278
3279 l += msg_print_text(msg, true, time, NULL, 0);
3280 idx = log_next(idx);
3281 seq++;
3282 }
3283
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);
3289
3290 l -= msg_print_text(msg, true, time, NULL, 0);
3291 idx = log_next(idx);
3292 seq++;
3293 }
3294
3295 /* last message in next interation */
3296 next_seq = seq;
3297 next_idx = idx;
3298
3299 l = 0;
3300 while (seq < dumper->next_seq) {
3301 struct printk_log *msg = log_from_idx(idx);
3302
3303 l += msg_print_text(msg, syslog, time, buf + l, size - l);
3304 idx = log_next(idx);
3305 seq++;
3306 }
3307
3308 dumper->next_seq = next_seq;
3309 dumper->next_idx = next_idx;
3310 ret = true;
3311 logbuf_unlock_irqrestore(flags);
3312out:
3313 if (len)
3314 *len = l;
3315 return ret;
3316}
3317EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3318
3319/**
3320 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3321 * @dumper: registered kmsg dumper
3322 *
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.
3326 *
3327 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3328 */
3329void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3330{
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;
3335}
3336
3337/**
3338 * kmsg_dump_rewind - reset the interator
3339 * @dumper: registered kmsg dumper
3340 *
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.
3344 */
3345void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3346{
3347 unsigned long flags;
3348
3349 logbuf_lock_irqsave(flags);
3350 kmsg_dump_rewind_nolock(dumper);
3351 logbuf_unlock_irqrestore(flags);
3352}
3353EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3354
3355#endif
1/*
2 * linux/kernel/printk.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
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).
11 * Ted Ts'o, 2/11/93.
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
17 */
18
19#include <linux/kernel.h>
20#include <linux/mm.h>
21#include <linux/tty.h>
22#include <linux/tty_driver.h>
23#include <linux/console.h>
24#include <linux/init.h>
25#include <linux/jiffies.h>
26#include <linux/nmi.h>
27#include <linux/module.h>
28#include <linux/moduleparam.h>
29#include <linux/delay.h>
30#include <linux/smp.h>
31#include <linux/security.h>
32#include <linux/bootmem.h>
33#include <linux/memblock.h>
34#include <linux/syscalls.h>
35#include <linux/crash_core.h>
36#include <linux/kdb.h>
37#include <linux/ratelimit.h>
38#include <linux/kmsg_dump.h>
39#include <linux/syslog.h>
40#include <linux/cpu.h>
41#include <linux/notifier.h>
42#include <linux/rculist.h>
43#include <linux/poll.h>
44#include <linux/irq_work.h>
45#include <linux/ctype.h>
46#include <linux/uio.h>
47#include <linux/sched/clock.h>
48#include <linux/sched/debug.h>
49#include <linux/sched/task_stack.h>
50
51#include <linux/uaccess.h>
52#include <asm/sections.h>
53
54#include <trace/events/initcall.h>
55#define CREATE_TRACE_POINTS
56#include <trace/events/printk.h>
57
58#include "console_cmdline.h"
59#include "braille.h"
60#include "internal.h"
61
62int console_printk[4] = {
63 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
64 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
65 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
66 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
67};
68
69/*
70 * Low level drivers may need that to know if they can schedule in
71 * their unblank() callback or not. So let's export it.
72 */
73int oops_in_progress;
74EXPORT_SYMBOL(oops_in_progress);
75
76/*
77 * console_sem protects the console_drivers list, and also
78 * provides serialisation for access to the entire console
79 * driver system.
80 */
81static DEFINE_SEMAPHORE(console_sem);
82struct console *console_drivers;
83EXPORT_SYMBOL_GPL(console_drivers);
84
85#ifdef CONFIG_LOCKDEP
86static struct lockdep_map console_lock_dep_map = {
87 .name = "console_lock"
88};
89#endif
90
91enum devkmsg_log_bits {
92 __DEVKMSG_LOG_BIT_ON = 0,
93 __DEVKMSG_LOG_BIT_OFF,
94 __DEVKMSG_LOG_BIT_LOCK,
95};
96
97enum devkmsg_log_masks {
98 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
99 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
100 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
101};
102
103/* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
104#define DEVKMSG_LOG_MASK_DEFAULT 0
105
106static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
107
108static int __control_devkmsg(char *str)
109{
110 if (!str)
111 return -EINVAL;
112
113 if (!strncmp(str, "on", 2)) {
114 devkmsg_log = DEVKMSG_LOG_MASK_ON;
115 return 2;
116 } else if (!strncmp(str, "off", 3)) {
117 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
118 return 3;
119 } else if (!strncmp(str, "ratelimit", 9)) {
120 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
121 return 9;
122 }
123 return -EINVAL;
124}
125
126static int __init control_devkmsg(char *str)
127{
128 if (__control_devkmsg(str) < 0)
129 return 1;
130
131 /*
132 * Set sysctl string accordingly:
133 */
134 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
135 strcpy(devkmsg_log_str, "on");
136 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
137 strcpy(devkmsg_log_str, "off");
138 /* else "ratelimit" which is set by default. */
139
140 /*
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.
145 */
146 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
147
148 return 0;
149}
150__setup("printk.devkmsg=", control_devkmsg);
151
152char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
153
154int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
155 void __user *buffer, size_t *lenp, loff_t *ppos)
156{
157 char old_str[DEVKMSG_STR_MAX_SIZE];
158 unsigned int old;
159 int err;
160
161 if (write) {
162 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
163 return -EINVAL;
164
165 old = devkmsg_log;
166 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
167 }
168
169 err = proc_dostring(table, write, buffer, lenp, ppos);
170 if (err)
171 return err;
172
173 if (write) {
174 err = __control_devkmsg(devkmsg_log_str);
175
176 /*
177 * Do not accept an unknown string OR a known string with
178 * trailing crap...
179 */
180 if (err < 0 || (err + 1 != *lenp)) {
181
182 /* ... and restore old setting. */
183 devkmsg_log = old;
184 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
185
186 return -EINVAL;
187 }
188 }
189
190 return 0;
191}
192
193/*
194 * Number of registered extended console drivers.
195 *
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.
202 */
203static int nr_ext_console_drivers;
204
205/*
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.
208 */
209#define down_console_sem() do { \
210 down(&console_sem);\
211 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
212} while (0)
213
214static int __down_trylock_console_sem(unsigned long ip)
215{
216 int lock_failed;
217 unsigned long flags;
218
219 /*
220 * Here and in __up_console_sem() we need to be in safe mode,
221 * because spindump/WARN/etc from under console ->lock will
222 * deadlock in printk()->down_trylock_console_sem() otherwise.
223 */
224 printk_safe_enter_irqsave(flags);
225 lock_failed = down_trylock(&console_sem);
226 printk_safe_exit_irqrestore(flags);
227
228 if (lock_failed)
229 return 1;
230 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
231 return 0;
232}
233#define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
234
235static void __up_console_sem(unsigned long ip)
236{
237 unsigned long flags;
238
239 mutex_release(&console_lock_dep_map, 1, ip);
240
241 printk_safe_enter_irqsave(flags);
242 up(&console_sem);
243 printk_safe_exit_irqrestore(flags);
244}
245#define up_console_sem() __up_console_sem(_RET_IP_)
246
247/*
248 * This is used for debugging the mess that is the VT code by
249 * keeping track if we have the console semaphore held. It's
250 * definitely not the perfect debug tool (we don't know if _WE_
251 * hold it and are racing, but it helps tracking those weird code
252 * paths in the console code where we end up in places I want
253 * locked without the console sempahore held).
254 */
255static int console_locked, console_suspended;
256
257/*
258 * If exclusive_console is non-NULL then only this console is to be printed to.
259 */
260static struct console *exclusive_console;
261
262/*
263 * Array of consoles built from command line options (console=)
264 */
265
266#define MAX_CMDLINECONSOLES 8
267
268static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
269
270static int preferred_console = -1;
271int console_set_on_cmdline;
272EXPORT_SYMBOL(console_set_on_cmdline);
273
274/* Flag: console code may call schedule() */
275static int console_may_schedule;
276
277enum con_msg_format_flags {
278 MSG_FORMAT_DEFAULT = 0,
279 MSG_FORMAT_SYSLOG = (1 << 0),
280};
281
282static int console_msg_format = MSG_FORMAT_DEFAULT;
283
284/*
285 * The printk log buffer consists of a chain of concatenated variable
286 * length records. Every record starts with a record header, containing
287 * the overall length of the record.
288 *
289 * The heads to the first and last entry in the buffer, as well as the
290 * sequence numbers of these entries are maintained when messages are
291 * stored.
292 *
293 * If the heads indicate available messages, the length in the header
294 * tells the start next message. A length == 0 for the next message
295 * indicates a wrap-around to the beginning of the buffer.
296 *
297 * Every record carries the monotonic timestamp in microseconds, as well as
298 * the standard userspace syslog level and syslog facility. The usual
299 * kernel messages use LOG_KERN; userspace-injected messages always carry
300 * a matching syslog facility, by default LOG_USER. The origin of every
301 * message can be reliably determined that way.
302 *
303 * The human readable log message directly follows the message header. The
304 * length of the message text is stored in the header, the stored message
305 * is not terminated.
306 *
307 * Optionally, a message can carry a dictionary of properties (key/value pairs),
308 * to provide userspace with a machine-readable message context.
309 *
310 * Examples for well-defined, commonly used property names are:
311 * DEVICE=b12:8 device identifier
312 * b12:8 block dev_t
313 * c127:3 char dev_t
314 * n8 netdev ifindex
315 * +sound:card0 subsystem:devname
316 * SUBSYSTEM=pci driver-core subsystem name
317 *
318 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
319 * follows directly after a '=' character. Every property is terminated by
320 * a '\0' character. The last property is not terminated.
321 *
322 * Example of a message structure:
323 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
324 * 0008 34 00 record is 52 bytes long
325 * 000a 0b 00 text is 11 bytes long
326 * 000c 1f 00 dictionary is 23 bytes long
327 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
328 * 0010 69 74 27 73 20 61 20 6c "it's a l"
329 * 69 6e 65 "ine"
330 * 001b 44 45 56 49 43 "DEVIC"
331 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
332 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
333 * 67 "g"
334 * 0032 00 00 00 padding to next message header
335 *
336 * The 'struct printk_log' buffer header must never be directly exported to
337 * userspace, it is a kernel-private implementation detail that might
338 * need to be changed in the future, when the requirements change.
339 *
340 * /dev/kmsg exports the structured data in the following line format:
341 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
342 *
343 * Users of the export format should ignore possible additional values
344 * separated by ',', and find the message after the ';' character.
345 *
346 * The optional key/value pairs are attached as continuation lines starting
347 * with a space character and terminated by a newline. All possible
348 * non-prinatable characters are escaped in the "\xff" notation.
349 */
350
351enum log_flags {
352 LOG_NOCONS = 1, /* already flushed, do not print to console */
353 LOG_NEWLINE = 2, /* text ended with a newline */
354 LOG_PREFIX = 4, /* text started with a prefix */
355 LOG_CONT = 8, /* text is a fragment of a continuation line */
356};
357
358struct printk_log {
359 u64 ts_nsec; /* timestamp in nanoseconds */
360 u16 len; /* length of entire record */
361 u16 text_len; /* length of text buffer */
362 u16 dict_len; /* length of dictionary buffer */
363 u8 facility; /* syslog facility */
364 u8 flags:5; /* internal record flags */
365 u8 level:3; /* syslog level */
366}
367#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
368__packed __aligned(4)
369#endif
370;
371
372/*
373 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
374 * within the scheduler's rq lock. It must be released before calling
375 * console_unlock() or anything else that might wake up a process.
376 */
377DEFINE_RAW_SPINLOCK(logbuf_lock);
378
379/*
380 * Helper macros to lock/unlock logbuf_lock and switch between
381 * printk-safe/unsafe modes.
382 */
383#define logbuf_lock_irq() \
384 do { \
385 printk_safe_enter_irq(); \
386 raw_spin_lock(&logbuf_lock); \
387 } while (0)
388
389#define logbuf_unlock_irq() \
390 do { \
391 raw_spin_unlock(&logbuf_lock); \
392 printk_safe_exit_irq(); \
393 } while (0)
394
395#define logbuf_lock_irqsave(flags) \
396 do { \
397 printk_safe_enter_irqsave(flags); \
398 raw_spin_lock(&logbuf_lock); \
399 } while (0)
400
401#define logbuf_unlock_irqrestore(flags) \
402 do { \
403 raw_spin_unlock(&logbuf_lock); \
404 printk_safe_exit_irqrestore(flags); \
405 } while (0)
406
407#ifdef CONFIG_PRINTK
408DECLARE_WAIT_QUEUE_HEAD(log_wait);
409/* the next printk record to read by syslog(READ) or /proc/kmsg */
410static u64 syslog_seq;
411static u32 syslog_idx;
412static size_t syslog_partial;
413
414/* index and sequence number of the first record stored in the buffer */
415static u64 log_first_seq;
416static u32 log_first_idx;
417
418/* index and sequence number of the next record to store in the buffer */
419static u64 log_next_seq;
420static u32 log_next_idx;
421
422/* the next printk record to write to the console */
423static u64 console_seq;
424static u32 console_idx;
425
426/* the next printk record to read after the last 'clear' command */
427static u64 clear_seq;
428static u32 clear_idx;
429
430#define PREFIX_MAX 32
431#define LOG_LINE_MAX (1024 - PREFIX_MAX)
432
433#define LOG_LEVEL(v) ((v) & 0x07)
434#define LOG_FACILITY(v) ((v) >> 3 & 0xff)
435
436/* record buffer */
437#define LOG_ALIGN __alignof__(struct printk_log)
438#define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
439static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
440static char *log_buf = __log_buf;
441static u32 log_buf_len = __LOG_BUF_LEN;
442
443/* Return log buffer address */
444char *log_buf_addr_get(void)
445{
446 return log_buf;
447}
448
449/* Return log buffer size */
450u32 log_buf_len_get(void)
451{
452 return log_buf_len;
453}
454
455/* human readable text of the record */
456static char *log_text(const struct printk_log *msg)
457{
458 return (char *)msg + sizeof(struct printk_log);
459}
460
461/* optional key/value pair dictionary attached to the record */
462static char *log_dict(const struct printk_log *msg)
463{
464 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
465}
466
467/* get record by index; idx must point to valid msg */
468static struct printk_log *log_from_idx(u32 idx)
469{
470 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
471
472 /*
473 * A length == 0 record is the end of buffer marker. Wrap around and
474 * read the message at the start of the buffer.
475 */
476 if (!msg->len)
477 return (struct printk_log *)log_buf;
478 return msg;
479}
480
481/* get next record; idx must point to valid msg */
482static u32 log_next(u32 idx)
483{
484 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
485
486 /* length == 0 indicates the end of the buffer; wrap */
487 /*
488 * A length == 0 record is the end of buffer marker. Wrap around and
489 * read the message at the start of the buffer as *this* one, and
490 * return the one after that.
491 */
492 if (!msg->len) {
493 msg = (struct printk_log *)log_buf;
494 return msg->len;
495 }
496 return idx + msg->len;
497}
498
499/*
500 * Check whether there is enough free space for the given message.
501 *
502 * The same values of first_idx and next_idx mean that the buffer
503 * is either empty or full.
504 *
505 * If the buffer is empty, we must respect the position of the indexes.
506 * They cannot be reset to the beginning of the buffer.
507 */
508static int logbuf_has_space(u32 msg_size, bool empty)
509{
510 u32 free;
511
512 if (log_next_idx > log_first_idx || empty)
513 free = max(log_buf_len - log_next_idx, log_first_idx);
514 else
515 free = log_first_idx - log_next_idx;
516
517 /*
518 * We need space also for an empty header that signalizes wrapping
519 * of the buffer.
520 */
521 return free >= msg_size + sizeof(struct printk_log);
522}
523
524static int log_make_free_space(u32 msg_size)
525{
526 while (log_first_seq < log_next_seq &&
527 !logbuf_has_space(msg_size, false)) {
528 /* drop old messages until we have enough contiguous space */
529 log_first_idx = log_next(log_first_idx);
530 log_first_seq++;
531 }
532
533 if (clear_seq < log_first_seq) {
534 clear_seq = log_first_seq;
535 clear_idx = log_first_idx;
536 }
537
538 /* sequence numbers are equal, so the log buffer is empty */
539 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
540 return 0;
541
542 return -ENOMEM;
543}
544
545/* compute the message size including the padding bytes */
546static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
547{
548 u32 size;
549
550 size = sizeof(struct printk_log) + text_len + dict_len;
551 *pad_len = (-size) & (LOG_ALIGN - 1);
552 size += *pad_len;
553
554 return size;
555}
556
557/*
558 * Define how much of the log buffer we could take at maximum. The value
559 * must be greater than two. Note that only half of the buffer is available
560 * when the index points to the middle.
561 */
562#define MAX_LOG_TAKE_PART 4
563static const char trunc_msg[] = "<truncated>";
564
565static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
566 u16 *dict_len, u32 *pad_len)
567{
568 /*
569 * The message should not take the whole buffer. Otherwise, it might
570 * get removed too soon.
571 */
572 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
573 if (*text_len > max_text_len)
574 *text_len = max_text_len;
575 /* enable the warning message */
576 *trunc_msg_len = strlen(trunc_msg);
577 /* disable the "dict" completely */
578 *dict_len = 0;
579 /* compute the size again, count also the warning message */
580 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
581}
582
583/* insert record into the buffer, discard old ones, update heads */
584static int log_store(int facility, int level,
585 enum log_flags flags, u64 ts_nsec,
586 const char *dict, u16 dict_len,
587 const char *text, u16 text_len)
588{
589 struct printk_log *msg;
590 u32 size, pad_len;
591 u16 trunc_msg_len = 0;
592
593 /* number of '\0' padding bytes to next message */
594 size = msg_used_size(text_len, dict_len, &pad_len);
595
596 if (log_make_free_space(size)) {
597 /* truncate the message if it is too long for empty buffer */
598 size = truncate_msg(&text_len, &trunc_msg_len,
599 &dict_len, &pad_len);
600 /* survive when the log buffer is too small for trunc_msg */
601 if (log_make_free_space(size))
602 return 0;
603 }
604
605 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
606 /*
607 * This message + an additional empty header does not fit
608 * at the end of the buffer. Add an empty header with len == 0
609 * to signify a wrap around.
610 */
611 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
612 log_next_idx = 0;
613 }
614
615 /* fill message */
616 msg = (struct printk_log *)(log_buf + log_next_idx);
617 memcpy(log_text(msg), text, text_len);
618 msg->text_len = text_len;
619 if (trunc_msg_len) {
620 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
621 msg->text_len += trunc_msg_len;
622 }
623 memcpy(log_dict(msg), dict, dict_len);
624 msg->dict_len = dict_len;
625 msg->facility = facility;
626 msg->level = level & 7;
627 msg->flags = flags & 0x1f;
628 if (ts_nsec > 0)
629 msg->ts_nsec = ts_nsec;
630 else
631 msg->ts_nsec = local_clock();
632 memset(log_dict(msg) + dict_len, 0, pad_len);
633 msg->len = size;
634
635 /* insert message */
636 log_next_idx += msg->len;
637 log_next_seq++;
638
639 return msg->text_len;
640}
641
642int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
643
644static int syslog_action_restricted(int type)
645{
646 if (dmesg_restrict)
647 return 1;
648 /*
649 * Unless restricted, we allow "read all" and "get buffer size"
650 * for everybody.
651 */
652 return type != SYSLOG_ACTION_READ_ALL &&
653 type != SYSLOG_ACTION_SIZE_BUFFER;
654}
655
656static int check_syslog_permissions(int type, int source)
657{
658 /*
659 * If this is from /proc/kmsg and we've already opened it, then we've
660 * already done the capabilities checks at open time.
661 */
662 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
663 goto ok;
664
665 if (syslog_action_restricted(type)) {
666 if (capable(CAP_SYSLOG))
667 goto ok;
668 /*
669 * For historical reasons, accept CAP_SYS_ADMIN too, with
670 * a warning.
671 */
672 if (capable(CAP_SYS_ADMIN)) {
673 pr_warn_once("%s (%d): Attempt to access syslog with "
674 "CAP_SYS_ADMIN but no CAP_SYSLOG "
675 "(deprecated).\n",
676 current->comm, task_pid_nr(current));
677 goto ok;
678 }
679 return -EPERM;
680 }
681ok:
682 return security_syslog(type);
683}
684
685static void append_char(char **pp, char *e, char c)
686{
687 if (*pp < e)
688 *(*pp)++ = c;
689}
690
691static ssize_t msg_print_ext_header(char *buf, size_t size,
692 struct printk_log *msg, u64 seq)
693{
694 u64 ts_usec = msg->ts_nsec;
695
696 do_div(ts_usec, 1000);
697
698 return scnprintf(buf, size, "%u,%llu,%llu,%c;",
699 (msg->facility << 3) | msg->level, seq, ts_usec,
700 msg->flags & LOG_CONT ? 'c' : '-');
701}
702
703static ssize_t msg_print_ext_body(char *buf, size_t size,
704 char *dict, size_t dict_len,
705 char *text, size_t text_len)
706{
707 char *p = buf, *e = buf + size;
708 size_t i;
709
710 /* escape non-printable characters */
711 for (i = 0; i < text_len; i++) {
712 unsigned char c = text[i];
713
714 if (c < ' ' || c >= 127 || c == '\\')
715 p += scnprintf(p, e - p, "\\x%02x", c);
716 else
717 append_char(&p, e, c);
718 }
719 append_char(&p, e, '\n');
720
721 if (dict_len) {
722 bool line = true;
723
724 for (i = 0; i < dict_len; i++) {
725 unsigned char c = dict[i];
726
727 if (line) {
728 append_char(&p, e, ' ');
729 line = false;
730 }
731
732 if (c == '\0') {
733 append_char(&p, e, '\n');
734 line = true;
735 continue;
736 }
737
738 if (c < ' ' || c >= 127 || c == '\\') {
739 p += scnprintf(p, e - p, "\\x%02x", c);
740 continue;
741 }
742
743 append_char(&p, e, c);
744 }
745 append_char(&p, e, '\n');
746 }
747
748 return p - buf;
749}
750
751/* /dev/kmsg - userspace message inject/listen interface */
752struct devkmsg_user {
753 u64 seq;
754 u32 idx;
755 struct ratelimit_state rs;
756 struct mutex lock;
757 char buf[CONSOLE_EXT_LOG_MAX];
758};
759
760static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
761{
762 char *buf, *line;
763 int level = default_message_loglevel;
764 int facility = 1; /* LOG_USER */
765 struct file *file = iocb->ki_filp;
766 struct devkmsg_user *user = file->private_data;
767 size_t len = iov_iter_count(from);
768 ssize_t ret = len;
769
770 if (!user || len > LOG_LINE_MAX)
771 return -EINVAL;
772
773 /* Ignore when user logging is disabled. */
774 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
775 return len;
776
777 /* Ratelimit when not explicitly enabled. */
778 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
779 if (!___ratelimit(&user->rs, current->comm))
780 return ret;
781 }
782
783 buf = kmalloc(len+1, GFP_KERNEL);
784 if (buf == NULL)
785 return -ENOMEM;
786
787 buf[len] = '\0';
788 if (!copy_from_iter_full(buf, len, from)) {
789 kfree(buf);
790 return -EFAULT;
791 }
792
793 /*
794 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
795 * the decimal value represents 32bit, the lower 3 bit are the log
796 * level, the rest are the log facility.
797 *
798 * If no prefix or no userspace facility is specified, we
799 * enforce LOG_USER, to be able to reliably distinguish
800 * kernel-generated messages from userspace-injected ones.
801 */
802 line = buf;
803 if (line[0] == '<') {
804 char *endp = NULL;
805 unsigned int u;
806
807 u = simple_strtoul(line + 1, &endp, 10);
808 if (endp && endp[0] == '>') {
809 level = LOG_LEVEL(u);
810 if (LOG_FACILITY(u) != 0)
811 facility = LOG_FACILITY(u);
812 endp++;
813 len -= endp - line;
814 line = endp;
815 }
816 }
817
818 printk_emit(facility, level, NULL, 0, "%s", line);
819 kfree(buf);
820 return ret;
821}
822
823static ssize_t devkmsg_read(struct file *file, char __user *buf,
824 size_t count, loff_t *ppos)
825{
826 struct devkmsg_user *user = file->private_data;
827 struct printk_log *msg;
828 size_t len;
829 ssize_t ret;
830
831 if (!user)
832 return -EBADF;
833
834 ret = mutex_lock_interruptible(&user->lock);
835 if (ret)
836 return ret;
837
838 logbuf_lock_irq();
839 while (user->seq == log_next_seq) {
840 if (file->f_flags & O_NONBLOCK) {
841 ret = -EAGAIN;
842 logbuf_unlock_irq();
843 goto out;
844 }
845
846 logbuf_unlock_irq();
847 ret = wait_event_interruptible(log_wait,
848 user->seq != log_next_seq);
849 if (ret)
850 goto out;
851 logbuf_lock_irq();
852 }
853
854 if (user->seq < log_first_seq) {
855 /* our last seen message is gone, return error and reset */
856 user->idx = log_first_idx;
857 user->seq = log_first_seq;
858 ret = -EPIPE;
859 logbuf_unlock_irq();
860 goto out;
861 }
862
863 msg = log_from_idx(user->idx);
864 len = msg_print_ext_header(user->buf, sizeof(user->buf),
865 msg, user->seq);
866 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
867 log_dict(msg), msg->dict_len,
868 log_text(msg), msg->text_len);
869
870 user->idx = log_next(user->idx);
871 user->seq++;
872 logbuf_unlock_irq();
873
874 if (len > count) {
875 ret = -EINVAL;
876 goto out;
877 }
878
879 if (copy_to_user(buf, user->buf, len)) {
880 ret = -EFAULT;
881 goto out;
882 }
883 ret = len;
884out:
885 mutex_unlock(&user->lock);
886 return ret;
887}
888
889static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
890{
891 struct devkmsg_user *user = file->private_data;
892 loff_t ret = 0;
893
894 if (!user)
895 return -EBADF;
896 if (offset)
897 return -ESPIPE;
898
899 logbuf_lock_irq();
900 switch (whence) {
901 case SEEK_SET:
902 /* the first record */
903 user->idx = log_first_idx;
904 user->seq = log_first_seq;
905 break;
906 case SEEK_DATA:
907 /*
908 * The first record after the last SYSLOG_ACTION_CLEAR,
909 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
910 * changes no global state, and does not clear anything.
911 */
912 user->idx = clear_idx;
913 user->seq = clear_seq;
914 break;
915 case SEEK_END:
916 /* after the last record */
917 user->idx = log_next_idx;
918 user->seq = log_next_seq;
919 break;
920 default:
921 ret = -EINVAL;
922 }
923 logbuf_unlock_irq();
924 return ret;
925}
926
927static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
928{
929 struct devkmsg_user *user = file->private_data;
930 __poll_t ret = 0;
931
932 if (!user)
933 return EPOLLERR|EPOLLNVAL;
934
935 poll_wait(file, &log_wait, wait);
936
937 logbuf_lock_irq();
938 if (user->seq < log_next_seq) {
939 /* return error when data has vanished underneath us */
940 if (user->seq < log_first_seq)
941 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
942 else
943 ret = EPOLLIN|EPOLLRDNORM;
944 }
945 logbuf_unlock_irq();
946
947 return ret;
948}
949
950static int devkmsg_open(struct inode *inode, struct file *file)
951{
952 struct devkmsg_user *user;
953 int err;
954
955 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
956 return -EPERM;
957
958 /* write-only does not need any file context */
959 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
960 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
961 SYSLOG_FROM_READER);
962 if (err)
963 return err;
964 }
965
966 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
967 if (!user)
968 return -ENOMEM;
969
970 ratelimit_default_init(&user->rs);
971 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
972
973 mutex_init(&user->lock);
974
975 logbuf_lock_irq();
976 user->idx = log_first_idx;
977 user->seq = log_first_seq;
978 logbuf_unlock_irq();
979
980 file->private_data = user;
981 return 0;
982}
983
984static int devkmsg_release(struct inode *inode, struct file *file)
985{
986 struct devkmsg_user *user = file->private_data;
987
988 if (!user)
989 return 0;
990
991 ratelimit_state_exit(&user->rs);
992
993 mutex_destroy(&user->lock);
994 kfree(user);
995 return 0;
996}
997
998const struct file_operations kmsg_fops = {
999 .open = devkmsg_open,
1000 .read = devkmsg_read,
1001 .write_iter = devkmsg_write,
1002 .llseek = devkmsg_llseek,
1003 .poll = devkmsg_poll,
1004 .release = devkmsg_release,
1005};
1006
1007#ifdef CONFIG_CRASH_CORE
1008/*
1009 * This appends the listed symbols to /proc/vmcore
1010 *
1011 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1012 * obtain access to symbols that are otherwise very difficult to locate. These
1013 * symbols are specifically used so that utilities can access and extract the
1014 * dmesg log from a vmcore file after a crash.
1015 */
1016void log_buf_vmcoreinfo_setup(void)
1017{
1018 VMCOREINFO_SYMBOL(log_buf);
1019 VMCOREINFO_SYMBOL(log_buf_len);
1020 VMCOREINFO_SYMBOL(log_first_idx);
1021 VMCOREINFO_SYMBOL(clear_idx);
1022 VMCOREINFO_SYMBOL(log_next_idx);
1023 /*
1024 * Export struct printk_log size and field offsets. User space tools can
1025 * parse it and detect any changes to structure down the line.
1026 */
1027 VMCOREINFO_STRUCT_SIZE(printk_log);
1028 VMCOREINFO_OFFSET(printk_log, ts_nsec);
1029 VMCOREINFO_OFFSET(printk_log, len);
1030 VMCOREINFO_OFFSET(printk_log, text_len);
1031 VMCOREINFO_OFFSET(printk_log, dict_len);
1032}
1033#endif
1034
1035/* requested log_buf_len from kernel cmdline */
1036static unsigned long __initdata new_log_buf_len;
1037
1038/* we practice scaling the ring buffer by powers of 2 */
1039static void __init log_buf_len_update(unsigned size)
1040{
1041 if (size)
1042 size = roundup_pow_of_two(size);
1043 if (size > log_buf_len)
1044 new_log_buf_len = size;
1045}
1046
1047/* save requested log_buf_len since it's too early to process it */
1048static int __init log_buf_len_setup(char *str)
1049{
1050 unsigned size = memparse(str, &str);
1051
1052 log_buf_len_update(size);
1053
1054 return 0;
1055}
1056early_param("log_buf_len", log_buf_len_setup);
1057
1058#ifdef CONFIG_SMP
1059#define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1060
1061static void __init log_buf_add_cpu(void)
1062{
1063 unsigned int cpu_extra;
1064
1065 /*
1066 * archs should set up cpu_possible_bits properly with
1067 * set_cpu_possible() after setup_arch() but just in
1068 * case lets ensure this is valid.
1069 */
1070 if (num_possible_cpus() == 1)
1071 return;
1072
1073 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1074
1075 /* by default this will only continue through for large > 64 CPUs */
1076 if (cpu_extra <= __LOG_BUF_LEN / 2)
1077 return;
1078
1079 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1080 __LOG_CPU_MAX_BUF_LEN);
1081 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1082 cpu_extra);
1083 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1084
1085 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1086}
1087#else /* !CONFIG_SMP */
1088static inline void log_buf_add_cpu(void) {}
1089#endif /* CONFIG_SMP */
1090
1091void __init setup_log_buf(int early)
1092{
1093 unsigned long flags;
1094 char *new_log_buf;
1095 int free;
1096
1097 if (log_buf != __log_buf)
1098 return;
1099
1100 if (!early && !new_log_buf_len)
1101 log_buf_add_cpu();
1102
1103 if (!new_log_buf_len)
1104 return;
1105
1106 if (early) {
1107 new_log_buf =
1108 memblock_virt_alloc(new_log_buf_len, LOG_ALIGN);
1109 } else {
1110 new_log_buf = memblock_virt_alloc_nopanic(new_log_buf_len,
1111 LOG_ALIGN);
1112 }
1113
1114 if (unlikely(!new_log_buf)) {
1115 pr_err("log_buf_len: %ld bytes not available\n",
1116 new_log_buf_len);
1117 return;
1118 }
1119
1120 logbuf_lock_irqsave(flags);
1121 log_buf_len = new_log_buf_len;
1122 log_buf = new_log_buf;
1123 new_log_buf_len = 0;
1124 free = __LOG_BUF_LEN - log_next_idx;
1125 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1126 logbuf_unlock_irqrestore(flags);
1127
1128 pr_info("log_buf_len: %d bytes\n", log_buf_len);
1129 pr_info("early log buf free: %d(%d%%)\n",
1130 free, (free * 100) / __LOG_BUF_LEN);
1131}
1132
1133static bool __read_mostly ignore_loglevel;
1134
1135static int __init ignore_loglevel_setup(char *str)
1136{
1137 ignore_loglevel = true;
1138 pr_info("debug: ignoring loglevel setting.\n");
1139
1140 return 0;
1141}
1142
1143early_param("ignore_loglevel", ignore_loglevel_setup);
1144module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1145MODULE_PARM_DESC(ignore_loglevel,
1146 "ignore loglevel setting (prints all kernel messages to the console)");
1147
1148static bool suppress_message_printing(int level)
1149{
1150 return (level >= console_loglevel && !ignore_loglevel);
1151}
1152
1153#ifdef CONFIG_BOOT_PRINTK_DELAY
1154
1155static int boot_delay; /* msecs delay after each printk during bootup */
1156static unsigned long long loops_per_msec; /* based on boot_delay */
1157
1158static int __init boot_delay_setup(char *str)
1159{
1160 unsigned long lpj;
1161
1162 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1163 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1164
1165 get_option(&str, &boot_delay);
1166 if (boot_delay > 10 * 1000)
1167 boot_delay = 0;
1168
1169 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1170 "HZ: %d, loops_per_msec: %llu\n",
1171 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1172 return 0;
1173}
1174early_param("boot_delay", boot_delay_setup);
1175
1176static void boot_delay_msec(int level)
1177{
1178 unsigned long long k;
1179 unsigned long timeout;
1180
1181 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1182 || suppress_message_printing(level)) {
1183 return;
1184 }
1185
1186 k = (unsigned long long)loops_per_msec * boot_delay;
1187
1188 timeout = jiffies + msecs_to_jiffies(boot_delay);
1189 while (k) {
1190 k--;
1191 cpu_relax();
1192 /*
1193 * use (volatile) jiffies to prevent
1194 * compiler reduction; loop termination via jiffies
1195 * is secondary and may or may not happen.
1196 */
1197 if (time_after(jiffies, timeout))
1198 break;
1199 touch_nmi_watchdog();
1200 }
1201}
1202#else
1203static inline void boot_delay_msec(int level)
1204{
1205}
1206#endif
1207
1208static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1209module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1210
1211static size_t print_time(u64 ts, char *buf)
1212{
1213 unsigned long rem_nsec;
1214
1215 if (!printk_time)
1216 return 0;
1217
1218 rem_nsec = do_div(ts, 1000000000);
1219
1220 if (!buf)
1221 return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
1222
1223 return sprintf(buf, "[%5lu.%06lu] ",
1224 (unsigned long)ts, rem_nsec / 1000);
1225}
1226
1227static size_t print_prefix(const struct printk_log *msg, bool syslog, char *buf)
1228{
1229 size_t len = 0;
1230 unsigned int prefix = (msg->facility << 3) | msg->level;
1231
1232 if (syslog) {
1233 if (buf) {
1234 len += sprintf(buf, "<%u>", prefix);
1235 } else {
1236 len += 3;
1237 if (prefix > 999)
1238 len += 3;
1239 else if (prefix > 99)
1240 len += 2;
1241 else if (prefix > 9)
1242 len++;
1243 }
1244 }
1245
1246 len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
1247 return len;
1248}
1249
1250static size_t msg_print_text(const struct printk_log *msg, bool syslog, char *buf, size_t size)
1251{
1252 const char *text = log_text(msg);
1253 size_t text_size = msg->text_len;
1254 size_t len = 0;
1255
1256 do {
1257 const char *next = memchr(text, '\n', text_size);
1258 size_t text_len;
1259
1260 if (next) {
1261 text_len = next - text;
1262 next++;
1263 text_size -= next - text;
1264 } else {
1265 text_len = text_size;
1266 }
1267
1268 if (buf) {
1269 if (print_prefix(msg, syslog, NULL) +
1270 text_len + 1 >= size - len)
1271 break;
1272
1273 len += print_prefix(msg, syslog, buf + len);
1274 memcpy(buf + len, text, text_len);
1275 len += text_len;
1276 buf[len++] = '\n';
1277 } else {
1278 /* SYSLOG_ACTION_* buffer size only calculation */
1279 len += print_prefix(msg, syslog, NULL);
1280 len += text_len;
1281 len++;
1282 }
1283
1284 text = next;
1285 } while (text);
1286
1287 return len;
1288}
1289
1290static int syslog_print(char __user *buf, int size)
1291{
1292 char *text;
1293 struct printk_log *msg;
1294 int len = 0;
1295
1296 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1297 if (!text)
1298 return -ENOMEM;
1299
1300 while (size > 0) {
1301 size_t n;
1302 size_t skip;
1303
1304 logbuf_lock_irq();
1305 if (syslog_seq < log_first_seq) {
1306 /* messages are gone, move to first one */
1307 syslog_seq = log_first_seq;
1308 syslog_idx = log_first_idx;
1309 syslog_partial = 0;
1310 }
1311 if (syslog_seq == log_next_seq) {
1312 logbuf_unlock_irq();
1313 break;
1314 }
1315
1316 skip = syslog_partial;
1317 msg = log_from_idx(syslog_idx);
1318 n = msg_print_text(msg, true, text, LOG_LINE_MAX + PREFIX_MAX);
1319 if (n - syslog_partial <= size) {
1320 /* message fits into buffer, move forward */
1321 syslog_idx = log_next(syslog_idx);
1322 syslog_seq++;
1323 n -= syslog_partial;
1324 syslog_partial = 0;
1325 } else if (!len){
1326 /* partial read(), remember position */
1327 n = size;
1328 syslog_partial += n;
1329 } else
1330 n = 0;
1331 logbuf_unlock_irq();
1332
1333 if (!n)
1334 break;
1335
1336 if (copy_to_user(buf, text + skip, n)) {
1337 if (!len)
1338 len = -EFAULT;
1339 break;
1340 }
1341
1342 len += n;
1343 size -= n;
1344 buf += n;
1345 }
1346
1347 kfree(text);
1348 return len;
1349}
1350
1351static int syslog_print_all(char __user *buf, int size, bool clear)
1352{
1353 char *text;
1354 int len = 0;
1355
1356 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1357 if (!text)
1358 return -ENOMEM;
1359
1360 logbuf_lock_irq();
1361 if (buf) {
1362 u64 next_seq;
1363 u64 seq;
1364 u32 idx;
1365
1366 /*
1367 * Find first record that fits, including all following records,
1368 * into the user-provided buffer for this dump.
1369 */
1370 seq = clear_seq;
1371 idx = clear_idx;
1372 while (seq < log_next_seq) {
1373 struct printk_log *msg = log_from_idx(idx);
1374
1375 len += msg_print_text(msg, true, NULL, 0);
1376 idx = log_next(idx);
1377 seq++;
1378 }
1379
1380 /* move first record forward until length fits into the buffer */
1381 seq = clear_seq;
1382 idx = clear_idx;
1383 while (len > size && seq < log_next_seq) {
1384 struct printk_log *msg = log_from_idx(idx);
1385
1386 len -= msg_print_text(msg, true, NULL, 0);
1387 idx = log_next(idx);
1388 seq++;
1389 }
1390
1391 /* last message fitting into this dump */
1392 next_seq = log_next_seq;
1393
1394 len = 0;
1395 while (len >= 0 && seq < next_seq) {
1396 struct printk_log *msg = log_from_idx(idx);
1397 int textlen;
1398
1399 textlen = msg_print_text(msg, true, text,
1400 LOG_LINE_MAX + PREFIX_MAX);
1401 if (textlen < 0) {
1402 len = textlen;
1403 break;
1404 }
1405 idx = log_next(idx);
1406 seq++;
1407
1408 logbuf_unlock_irq();
1409 if (copy_to_user(buf + len, text, textlen))
1410 len = -EFAULT;
1411 else
1412 len += textlen;
1413 logbuf_lock_irq();
1414
1415 if (seq < log_first_seq) {
1416 /* messages are gone, move to next one */
1417 seq = log_first_seq;
1418 idx = log_first_idx;
1419 }
1420 }
1421 }
1422
1423 if (clear) {
1424 clear_seq = log_next_seq;
1425 clear_idx = log_next_idx;
1426 }
1427 logbuf_unlock_irq();
1428
1429 kfree(text);
1430 return len;
1431}
1432
1433int do_syslog(int type, char __user *buf, int len, int source)
1434{
1435 bool clear = false;
1436 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1437 int error;
1438
1439 error = check_syslog_permissions(type, source);
1440 if (error)
1441 return error;
1442
1443 switch (type) {
1444 case SYSLOG_ACTION_CLOSE: /* Close log */
1445 break;
1446 case SYSLOG_ACTION_OPEN: /* Open log */
1447 break;
1448 case SYSLOG_ACTION_READ: /* Read from log */
1449 if (!buf || len < 0)
1450 return -EINVAL;
1451 if (!len)
1452 return 0;
1453 if (!access_ok(VERIFY_WRITE, buf, len))
1454 return -EFAULT;
1455 error = wait_event_interruptible(log_wait,
1456 syslog_seq != log_next_seq);
1457 if (error)
1458 return error;
1459 error = syslog_print(buf, len);
1460 break;
1461 /* Read/clear last kernel messages */
1462 case SYSLOG_ACTION_READ_CLEAR:
1463 clear = true;
1464 /* FALL THRU */
1465 /* Read last kernel messages */
1466 case SYSLOG_ACTION_READ_ALL:
1467 if (!buf || len < 0)
1468 return -EINVAL;
1469 if (!len)
1470 return 0;
1471 if (!access_ok(VERIFY_WRITE, buf, len))
1472 return -EFAULT;
1473 error = syslog_print_all(buf, len, clear);
1474 break;
1475 /* Clear ring buffer */
1476 case SYSLOG_ACTION_CLEAR:
1477 syslog_print_all(NULL, 0, true);
1478 break;
1479 /* Disable logging to console */
1480 case SYSLOG_ACTION_CONSOLE_OFF:
1481 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1482 saved_console_loglevel = console_loglevel;
1483 console_loglevel = minimum_console_loglevel;
1484 break;
1485 /* Enable logging to console */
1486 case SYSLOG_ACTION_CONSOLE_ON:
1487 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1488 console_loglevel = saved_console_loglevel;
1489 saved_console_loglevel = LOGLEVEL_DEFAULT;
1490 }
1491 break;
1492 /* Set level of messages printed to console */
1493 case SYSLOG_ACTION_CONSOLE_LEVEL:
1494 if (len < 1 || len > 8)
1495 return -EINVAL;
1496 if (len < minimum_console_loglevel)
1497 len = minimum_console_loglevel;
1498 console_loglevel = len;
1499 /* Implicitly re-enable logging to console */
1500 saved_console_loglevel = LOGLEVEL_DEFAULT;
1501 break;
1502 /* Number of chars in the log buffer */
1503 case SYSLOG_ACTION_SIZE_UNREAD:
1504 logbuf_lock_irq();
1505 if (syslog_seq < log_first_seq) {
1506 /* messages are gone, move to first one */
1507 syslog_seq = log_first_seq;
1508 syslog_idx = log_first_idx;
1509 syslog_partial = 0;
1510 }
1511 if (source == SYSLOG_FROM_PROC) {
1512 /*
1513 * Short-cut for poll(/"proc/kmsg") which simply checks
1514 * for pending data, not the size; return the count of
1515 * records, not the length.
1516 */
1517 error = log_next_seq - syslog_seq;
1518 } else {
1519 u64 seq = syslog_seq;
1520 u32 idx = syslog_idx;
1521
1522 while (seq < log_next_seq) {
1523 struct printk_log *msg = log_from_idx(idx);
1524
1525 error += msg_print_text(msg, true, NULL, 0);
1526 idx = log_next(idx);
1527 seq++;
1528 }
1529 error -= syslog_partial;
1530 }
1531 logbuf_unlock_irq();
1532 break;
1533 /* Size of the log buffer */
1534 case SYSLOG_ACTION_SIZE_BUFFER:
1535 error = log_buf_len;
1536 break;
1537 default:
1538 error = -EINVAL;
1539 break;
1540 }
1541
1542 return error;
1543}
1544
1545SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1546{
1547 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1548}
1549
1550/*
1551 * Special console_lock variants that help to reduce the risk of soft-lockups.
1552 * They allow to pass console_lock to another printk() call using a busy wait.
1553 */
1554
1555#ifdef CONFIG_LOCKDEP
1556static struct lockdep_map console_owner_dep_map = {
1557 .name = "console_owner"
1558};
1559#endif
1560
1561static DEFINE_RAW_SPINLOCK(console_owner_lock);
1562static struct task_struct *console_owner;
1563static bool console_waiter;
1564
1565/**
1566 * console_lock_spinning_enable - mark beginning of code where another
1567 * thread might safely busy wait
1568 *
1569 * This basically converts console_lock into a spinlock. This marks
1570 * the section where the console_lock owner can not sleep, because
1571 * there may be a waiter spinning (like a spinlock). Also it must be
1572 * ready to hand over the lock at the end of the section.
1573 */
1574static void console_lock_spinning_enable(void)
1575{
1576 raw_spin_lock(&console_owner_lock);
1577 console_owner = current;
1578 raw_spin_unlock(&console_owner_lock);
1579
1580 /* The waiter may spin on us after setting console_owner */
1581 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1582}
1583
1584/**
1585 * console_lock_spinning_disable_and_check - mark end of code where another
1586 * thread was able to busy wait and check if there is a waiter
1587 *
1588 * This is called at the end of the section where spinning is allowed.
1589 * It has two functions. First, it is a signal that it is no longer
1590 * safe to start busy waiting for the lock. Second, it checks if
1591 * there is a busy waiter and passes the lock rights to her.
1592 *
1593 * Important: Callers lose the lock if there was a busy waiter.
1594 * They must not touch items synchronized by console_lock
1595 * in this case.
1596 *
1597 * Return: 1 if the lock rights were passed, 0 otherwise.
1598 */
1599static int console_lock_spinning_disable_and_check(void)
1600{
1601 int waiter;
1602
1603 raw_spin_lock(&console_owner_lock);
1604 waiter = READ_ONCE(console_waiter);
1605 console_owner = NULL;
1606 raw_spin_unlock(&console_owner_lock);
1607
1608 if (!waiter) {
1609 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1610 return 0;
1611 }
1612
1613 /* The waiter is now free to continue */
1614 WRITE_ONCE(console_waiter, false);
1615
1616 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1617
1618 /*
1619 * Hand off console_lock to waiter. The waiter will perform
1620 * the up(). After this, the waiter is the console_lock owner.
1621 */
1622 mutex_release(&console_lock_dep_map, 1, _THIS_IP_);
1623 return 1;
1624}
1625
1626/**
1627 * console_trylock_spinning - try to get console_lock by busy waiting
1628 *
1629 * This allows to busy wait for the console_lock when the current
1630 * owner is running in specially marked sections. It means that
1631 * the current owner is running and cannot reschedule until it
1632 * is ready to lose the lock.
1633 *
1634 * Return: 1 if we got the lock, 0 othrewise
1635 */
1636static int console_trylock_spinning(void)
1637{
1638 struct task_struct *owner = NULL;
1639 bool waiter;
1640 bool spin = false;
1641 unsigned long flags;
1642
1643 if (console_trylock())
1644 return 1;
1645
1646 printk_safe_enter_irqsave(flags);
1647
1648 raw_spin_lock(&console_owner_lock);
1649 owner = READ_ONCE(console_owner);
1650 waiter = READ_ONCE(console_waiter);
1651 if (!waiter && owner && owner != current) {
1652 WRITE_ONCE(console_waiter, true);
1653 spin = true;
1654 }
1655 raw_spin_unlock(&console_owner_lock);
1656
1657 /*
1658 * If there is an active printk() writing to the
1659 * consoles, instead of having it write our data too,
1660 * see if we can offload that load from the active
1661 * printer, and do some printing ourselves.
1662 * Go into a spin only if there isn't already a waiter
1663 * spinning, and there is an active printer, and
1664 * that active printer isn't us (recursive printk?).
1665 */
1666 if (!spin) {
1667 printk_safe_exit_irqrestore(flags);
1668 return 0;
1669 }
1670
1671 /* We spin waiting for the owner to release us */
1672 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1673 /* Owner will clear console_waiter on hand off */
1674 while (READ_ONCE(console_waiter))
1675 cpu_relax();
1676 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1677
1678 printk_safe_exit_irqrestore(flags);
1679 /*
1680 * The owner passed the console lock to us.
1681 * Since we did not spin on console lock, annotate
1682 * this as a trylock. Otherwise lockdep will
1683 * complain.
1684 */
1685 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1686
1687 return 1;
1688}
1689
1690/*
1691 * Call the console drivers, asking them to write out
1692 * log_buf[start] to log_buf[end - 1].
1693 * The console_lock must be held.
1694 */
1695static void call_console_drivers(const char *ext_text, size_t ext_len,
1696 const char *text, size_t len)
1697{
1698 struct console *con;
1699
1700 trace_console_rcuidle(text, len);
1701
1702 if (!console_drivers)
1703 return;
1704
1705 for_each_console(con) {
1706 if (exclusive_console && con != exclusive_console)
1707 continue;
1708 if (!(con->flags & CON_ENABLED))
1709 continue;
1710 if (!con->write)
1711 continue;
1712 if (!cpu_online(smp_processor_id()) &&
1713 !(con->flags & CON_ANYTIME))
1714 continue;
1715 if (con->flags & CON_EXTENDED)
1716 con->write(con, ext_text, ext_len);
1717 else
1718 con->write(con, text, len);
1719 }
1720}
1721
1722int printk_delay_msec __read_mostly;
1723
1724static inline void printk_delay(void)
1725{
1726 if (unlikely(printk_delay_msec)) {
1727 int m = printk_delay_msec;
1728
1729 while (m--) {
1730 mdelay(1);
1731 touch_nmi_watchdog();
1732 }
1733 }
1734}
1735
1736/*
1737 * Continuation lines are buffered, and not committed to the record buffer
1738 * until the line is complete, or a race forces it. The line fragments
1739 * though, are printed immediately to the consoles to ensure everything has
1740 * reached the console in case of a kernel crash.
1741 */
1742static struct cont {
1743 char buf[LOG_LINE_MAX];
1744 size_t len; /* length == 0 means unused buffer */
1745 struct task_struct *owner; /* task of first print*/
1746 u64 ts_nsec; /* time of first print */
1747 u8 level; /* log level of first message */
1748 u8 facility; /* log facility of first message */
1749 enum log_flags flags; /* prefix, newline flags */
1750} cont;
1751
1752static void cont_flush(void)
1753{
1754 if (cont.len == 0)
1755 return;
1756
1757 log_store(cont.facility, cont.level, cont.flags, cont.ts_nsec,
1758 NULL, 0, cont.buf, cont.len);
1759 cont.len = 0;
1760}
1761
1762static bool cont_add(int facility, int level, enum log_flags flags, const char *text, size_t len)
1763{
1764 /*
1765 * If ext consoles are present, flush and skip in-kernel
1766 * continuation. See nr_ext_console_drivers definition. Also, if
1767 * the line gets too long, split it up in separate records.
1768 */
1769 if (nr_ext_console_drivers || cont.len + len > sizeof(cont.buf)) {
1770 cont_flush();
1771 return false;
1772 }
1773
1774 if (!cont.len) {
1775 cont.facility = facility;
1776 cont.level = level;
1777 cont.owner = current;
1778 cont.ts_nsec = local_clock();
1779 cont.flags = flags;
1780 }
1781
1782 memcpy(cont.buf + cont.len, text, len);
1783 cont.len += len;
1784
1785 // The original flags come from the first line,
1786 // but later continuations can add a newline.
1787 if (flags & LOG_NEWLINE) {
1788 cont.flags |= LOG_NEWLINE;
1789 cont_flush();
1790 }
1791
1792 if (cont.len > (sizeof(cont.buf) * 80) / 100)
1793 cont_flush();
1794
1795 return true;
1796}
1797
1798static size_t log_output(int facility, int level, enum log_flags lflags, const char *dict, size_t dictlen, char *text, size_t text_len)
1799{
1800 /*
1801 * If an earlier line was buffered, and we're a continuation
1802 * write from the same process, try to add it to the buffer.
1803 */
1804 if (cont.len) {
1805 if (cont.owner == current && (lflags & LOG_CONT)) {
1806 if (cont_add(facility, level, lflags, text, text_len))
1807 return text_len;
1808 }
1809 /* Otherwise, make sure it's flushed */
1810 cont_flush();
1811 }
1812
1813 /* Skip empty continuation lines that couldn't be added - they just flush */
1814 if (!text_len && (lflags & LOG_CONT))
1815 return 0;
1816
1817 /* If it doesn't end in a newline, try to buffer the current line */
1818 if (!(lflags & LOG_NEWLINE)) {
1819 if (cont_add(facility, level, lflags, text, text_len))
1820 return text_len;
1821 }
1822
1823 /* Store it in the record log */
1824 return log_store(facility, level, lflags, 0, dict, dictlen, text, text_len);
1825}
1826
1827asmlinkage int vprintk_emit(int facility, int level,
1828 const char *dict, size_t dictlen,
1829 const char *fmt, va_list args)
1830{
1831 static char textbuf[LOG_LINE_MAX];
1832 char *text = textbuf;
1833 size_t text_len;
1834 enum log_flags lflags = 0;
1835 unsigned long flags;
1836 int printed_len;
1837 bool in_sched = false;
1838
1839 if (level == LOGLEVEL_SCHED) {
1840 level = LOGLEVEL_DEFAULT;
1841 in_sched = true;
1842 }
1843
1844 boot_delay_msec(level);
1845 printk_delay();
1846
1847 /* This stops the holder of console_sem just where we want him */
1848 logbuf_lock_irqsave(flags);
1849 /*
1850 * The printf needs to come first; we need the syslog
1851 * prefix which might be passed-in as a parameter.
1852 */
1853 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1854
1855 /* mark and strip a trailing newline */
1856 if (text_len && text[text_len-1] == '\n') {
1857 text_len--;
1858 lflags |= LOG_NEWLINE;
1859 }
1860
1861 /* strip kernel syslog prefix and extract log level or control flags */
1862 if (facility == 0) {
1863 int kern_level;
1864
1865 while ((kern_level = printk_get_level(text)) != 0) {
1866 switch (kern_level) {
1867 case '0' ... '7':
1868 if (level == LOGLEVEL_DEFAULT)
1869 level = kern_level - '0';
1870 /* fallthrough */
1871 case 'd': /* KERN_DEFAULT */
1872 lflags |= LOG_PREFIX;
1873 break;
1874 case 'c': /* KERN_CONT */
1875 lflags |= LOG_CONT;
1876 }
1877
1878 text_len -= 2;
1879 text += 2;
1880 }
1881 }
1882
1883 if (level == LOGLEVEL_DEFAULT)
1884 level = default_message_loglevel;
1885
1886 if (dict)
1887 lflags |= LOG_PREFIX|LOG_NEWLINE;
1888
1889 printed_len = log_output(facility, level, lflags, dict, dictlen, text, text_len);
1890
1891 logbuf_unlock_irqrestore(flags);
1892
1893 /* If called from the scheduler, we can not call up(). */
1894 if (!in_sched) {
1895 /*
1896 * Disable preemption to avoid being preempted while holding
1897 * console_sem which would prevent anyone from printing to
1898 * console
1899 */
1900 preempt_disable();
1901 /*
1902 * Try to acquire and then immediately release the console
1903 * semaphore. The release will print out buffers and wake up
1904 * /dev/kmsg and syslog() users.
1905 */
1906 if (console_trylock_spinning())
1907 console_unlock();
1908 preempt_enable();
1909 }
1910
1911 return printed_len;
1912}
1913EXPORT_SYMBOL(vprintk_emit);
1914
1915asmlinkage int vprintk(const char *fmt, va_list args)
1916{
1917 return vprintk_func(fmt, args);
1918}
1919EXPORT_SYMBOL(vprintk);
1920
1921asmlinkage int printk_emit(int facility, int level,
1922 const char *dict, size_t dictlen,
1923 const char *fmt, ...)
1924{
1925 va_list args;
1926 int r;
1927
1928 va_start(args, fmt);
1929 r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1930 va_end(args);
1931
1932 return r;
1933}
1934EXPORT_SYMBOL(printk_emit);
1935
1936int vprintk_default(const char *fmt, va_list args)
1937{
1938 int r;
1939
1940#ifdef CONFIG_KGDB_KDB
1941 /* Allow to pass printk() to kdb but avoid a recursion. */
1942 if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
1943 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
1944 return r;
1945 }
1946#endif
1947 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1948
1949 return r;
1950}
1951EXPORT_SYMBOL_GPL(vprintk_default);
1952
1953/**
1954 * printk - print a kernel message
1955 * @fmt: format string
1956 *
1957 * This is printk(). It can be called from any context. We want it to work.
1958 *
1959 * We try to grab the console_lock. If we succeed, it's easy - we log the
1960 * output and call the console drivers. If we fail to get the semaphore, we
1961 * place the output into the log buffer and return. The current holder of
1962 * the console_sem will notice the new output in console_unlock(); and will
1963 * send it to the consoles before releasing the lock.
1964 *
1965 * One effect of this deferred printing is that code which calls printk() and
1966 * then changes console_loglevel may break. This is because console_loglevel
1967 * is inspected when the actual printing occurs.
1968 *
1969 * See also:
1970 * printf(3)
1971 *
1972 * See the vsnprintf() documentation for format string extensions over C99.
1973 */
1974asmlinkage __visible int printk(const char *fmt, ...)
1975{
1976 va_list args;
1977 int r;
1978
1979 va_start(args, fmt);
1980 r = vprintk_func(fmt, args);
1981 va_end(args);
1982
1983 return r;
1984}
1985EXPORT_SYMBOL(printk);
1986
1987#else /* CONFIG_PRINTK */
1988
1989#define LOG_LINE_MAX 0
1990#define PREFIX_MAX 0
1991
1992static u64 syslog_seq;
1993static u32 syslog_idx;
1994static u64 console_seq;
1995static u32 console_idx;
1996static u64 log_first_seq;
1997static u32 log_first_idx;
1998static u64 log_next_seq;
1999static char *log_text(const struct printk_log *msg) { return NULL; }
2000static char *log_dict(const struct printk_log *msg) { return NULL; }
2001static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2002static u32 log_next(u32 idx) { return 0; }
2003static ssize_t msg_print_ext_header(char *buf, size_t size,
2004 struct printk_log *msg,
2005 u64 seq) { return 0; }
2006static ssize_t msg_print_ext_body(char *buf, size_t size,
2007 char *dict, size_t dict_len,
2008 char *text, size_t text_len) { return 0; }
2009static void console_lock_spinning_enable(void) { }
2010static int console_lock_spinning_disable_and_check(void) { return 0; }
2011static void call_console_drivers(const char *ext_text, size_t ext_len,
2012 const char *text, size_t len) {}
2013static size_t msg_print_text(const struct printk_log *msg,
2014 bool syslog, char *buf, size_t size) { return 0; }
2015static bool suppress_message_printing(int level) { return false; }
2016
2017#endif /* CONFIG_PRINTK */
2018
2019#ifdef CONFIG_EARLY_PRINTK
2020struct console *early_console;
2021
2022asmlinkage __visible void early_printk(const char *fmt, ...)
2023{
2024 va_list ap;
2025 char buf[512];
2026 int n;
2027
2028 if (!early_console)
2029 return;
2030
2031 va_start(ap, fmt);
2032 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2033 va_end(ap);
2034
2035 early_console->write(early_console, buf, n);
2036}
2037#endif
2038
2039static int __add_preferred_console(char *name, int idx, char *options,
2040 char *brl_options)
2041{
2042 struct console_cmdline *c;
2043 int i;
2044
2045 /*
2046 * See if this tty is not yet registered, and
2047 * if we have a slot free.
2048 */
2049 for (i = 0, c = console_cmdline;
2050 i < MAX_CMDLINECONSOLES && c->name[0];
2051 i++, c++) {
2052 if (strcmp(c->name, name) == 0 && c->index == idx) {
2053 if (!brl_options)
2054 preferred_console = i;
2055 return 0;
2056 }
2057 }
2058 if (i == MAX_CMDLINECONSOLES)
2059 return -E2BIG;
2060 if (!brl_options)
2061 preferred_console = i;
2062 strlcpy(c->name, name, sizeof(c->name));
2063 c->options = options;
2064 braille_set_options(c, brl_options);
2065
2066 c->index = idx;
2067 return 0;
2068}
2069
2070static int __init console_msg_format_setup(char *str)
2071{
2072 if (!strcmp(str, "syslog"))
2073 console_msg_format = MSG_FORMAT_SYSLOG;
2074 if (!strcmp(str, "default"))
2075 console_msg_format = MSG_FORMAT_DEFAULT;
2076 return 1;
2077}
2078__setup("console_msg_format=", console_msg_format_setup);
2079
2080/*
2081 * Set up a console. Called via do_early_param() in init/main.c
2082 * for each "console=" parameter in the boot command line.
2083 */
2084static int __init console_setup(char *str)
2085{
2086 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2087 char *s, *options, *brl_options = NULL;
2088 int idx;
2089
2090 if (_braille_console_setup(&str, &brl_options))
2091 return 1;
2092
2093 /*
2094 * Decode str into name, index, options.
2095 */
2096 if (str[0] >= '0' && str[0] <= '9') {
2097 strcpy(buf, "ttyS");
2098 strncpy(buf + 4, str, sizeof(buf) - 5);
2099 } else {
2100 strncpy(buf, str, sizeof(buf) - 1);
2101 }
2102 buf[sizeof(buf) - 1] = 0;
2103 options = strchr(str, ',');
2104 if (options)
2105 *(options++) = 0;
2106#ifdef __sparc__
2107 if (!strcmp(str, "ttya"))
2108 strcpy(buf, "ttyS0");
2109 if (!strcmp(str, "ttyb"))
2110 strcpy(buf, "ttyS1");
2111#endif
2112 for (s = buf; *s; s++)
2113 if (isdigit(*s) || *s == ',')
2114 break;
2115 idx = simple_strtoul(s, NULL, 10);
2116 *s = 0;
2117
2118 __add_preferred_console(buf, idx, options, brl_options);
2119 console_set_on_cmdline = 1;
2120 return 1;
2121}
2122__setup("console=", console_setup);
2123
2124/**
2125 * add_preferred_console - add a device to the list of preferred consoles.
2126 * @name: device name
2127 * @idx: device index
2128 * @options: options for this console
2129 *
2130 * The last preferred console added will be used for kernel messages
2131 * and stdin/out/err for init. Normally this is used by console_setup
2132 * above to handle user-supplied console arguments; however it can also
2133 * be used by arch-specific code either to override the user or more
2134 * commonly to provide a default console (ie from PROM variables) when
2135 * the user has not supplied one.
2136 */
2137int add_preferred_console(char *name, int idx, char *options)
2138{
2139 return __add_preferred_console(name, idx, options, NULL);
2140}
2141
2142bool console_suspend_enabled = true;
2143EXPORT_SYMBOL(console_suspend_enabled);
2144
2145static int __init console_suspend_disable(char *str)
2146{
2147 console_suspend_enabled = false;
2148 return 1;
2149}
2150__setup("no_console_suspend", console_suspend_disable);
2151module_param_named(console_suspend, console_suspend_enabled,
2152 bool, S_IRUGO | S_IWUSR);
2153MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2154 " and hibernate operations");
2155
2156/**
2157 * suspend_console - suspend the console subsystem
2158 *
2159 * This disables printk() while we go into suspend states
2160 */
2161void suspend_console(void)
2162{
2163 if (!console_suspend_enabled)
2164 return;
2165 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2166 console_lock();
2167 console_suspended = 1;
2168 up_console_sem();
2169}
2170
2171void resume_console(void)
2172{
2173 if (!console_suspend_enabled)
2174 return;
2175 down_console_sem();
2176 console_suspended = 0;
2177 console_unlock();
2178}
2179
2180/**
2181 * console_cpu_notify - print deferred console messages after CPU hotplug
2182 * @cpu: unused
2183 *
2184 * If printk() is called from a CPU that is not online yet, the messages
2185 * will be printed on the console only if there are CON_ANYTIME consoles.
2186 * This function is called when a new CPU comes online (or fails to come
2187 * up) or goes offline.
2188 */
2189static int console_cpu_notify(unsigned int cpu)
2190{
2191 if (!cpuhp_tasks_frozen) {
2192 /* If trylock fails, someone else is doing the printing */
2193 if (console_trylock())
2194 console_unlock();
2195 }
2196 return 0;
2197}
2198
2199/**
2200 * console_lock - lock the console system for exclusive use.
2201 *
2202 * Acquires a lock which guarantees that the caller has
2203 * exclusive access to the console system and the console_drivers list.
2204 *
2205 * Can sleep, returns nothing.
2206 */
2207void console_lock(void)
2208{
2209 might_sleep();
2210
2211 down_console_sem();
2212 if (console_suspended)
2213 return;
2214 console_locked = 1;
2215 console_may_schedule = 1;
2216}
2217EXPORT_SYMBOL(console_lock);
2218
2219/**
2220 * console_trylock - try to lock the console system for exclusive use.
2221 *
2222 * Try to acquire a lock which guarantees that the caller has exclusive
2223 * access to the console system and the console_drivers list.
2224 *
2225 * returns 1 on success, and 0 on failure to acquire the lock.
2226 */
2227int console_trylock(void)
2228{
2229 if (down_trylock_console_sem())
2230 return 0;
2231 if (console_suspended) {
2232 up_console_sem();
2233 return 0;
2234 }
2235 console_locked = 1;
2236 console_may_schedule = 0;
2237 return 1;
2238}
2239EXPORT_SYMBOL(console_trylock);
2240
2241int is_console_locked(void)
2242{
2243 return console_locked;
2244}
2245
2246/*
2247 * Check if we have any console that is capable of printing while cpu is
2248 * booting or shutting down. Requires console_sem.
2249 */
2250static int have_callable_console(void)
2251{
2252 struct console *con;
2253
2254 for_each_console(con)
2255 if ((con->flags & CON_ENABLED) &&
2256 (con->flags & CON_ANYTIME))
2257 return 1;
2258
2259 return 0;
2260}
2261
2262/*
2263 * Can we actually use the console at this time on this cpu?
2264 *
2265 * Console drivers may assume that per-cpu resources have been allocated. So
2266 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2267 * call them until this CPU is officially up.
2268 */
2269static inline int can_use_console(void)
2270{
2271 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2272}
2273
2274/**
2275 * console_unlock - unlock the console system
2276 *
2277 * Releases the console_lock which the caller holds on the console system
2278 * and the console driver list.
2279 *
2280 * While the console_lock was held, console output may have been buffered
2281 * by printk(). If this is the case, console_unlock(); emits
2282 * the output prior to releasing the lock.
2283 *
2284 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2285 *
2286 * console_unlock(); may be called from any context.
2287 */
2288void console_unlock(void)
2289{
2290 static char ext_text[CONSOLE_EXT_LOG_MAX];
2291 static char text[LOG_LINE_MAX + PREFIX_MAX];
2292 static u64 seen_seq;
2293 unsigned long flags;
2294 bool wake_klogd = false;
2295 bool do_cond_resched, retry;
2296
2297 if (console_suspended) {
2298 up_console_sem();
2299 return;
2300 }
2301
2302 /*
2303 * Console drivers are called with interrupts disabled, so
2304 * @console_may_schedule should be cleared before; however, we may
2305 * end up dumping a lot of lines, for example, if called from
2306 * console registration path, and should invoke cond_resched()
2307 * between lines if allowable. Not doing so can cause a very long
2308 * scheduling stall on a slow console leading to RCU stall and
2309 * softlockup warnings which exacerbate the issue with more
2310 * messages practically incapacitating the system.
2311 *
2312 * console_trylock() is not able to detect the preemptive
2313 * context reliably. Therefore the value must be stored before
2314 * and cleared after the the "again" goto label.
2315 */
2316 do_cond_resched = console_may_schedule;
2317again:
2318 console_may_schedule = 0;
2319
2320 /*
2321 * We released the console_sem lock, so we need to recheck if
2322 * cpu is online and (if not) is there at least one CON_ANYTIME
2323 * console.
2324 */
2325 if (!can_use_console()) {
2326 console_locked = 0;
2327 up_console_sem();
2328 return;
2329 }
2330
2331 for (;;) {
2332 struct printk_log *msg;
2333 size_t ext_len = 0;
2334 size_t len;
2335
2336 printk_safe_enter_irqsave(flags);
2337 raw_spin_lock(&logbuf_lock);
2338 if (seen_seq != log_next_seq) {
2339 wake_klogd = true;
2340 seen_seq = log_next_seq;
2341 }
2342
2343 if (console_seq < log_first_seq) {
2344 len = sprintf(text, "** %u printk messages dropped **\n",
2345 (unsigned)(log_first_seq - console_seq));
2346
2347 /* messages are gone, move to first one */
2348 console_seq = log_first_seq;
2349 console_idx = log_first_idx;
2350 } else {
2351 len = 0;
2352 }
2353skip:
2354 if (console_seq == log_next_seq)
2355 break;
2356
2357 msg = log_from_idx(console_idx);
2358 if (suppress_message_printing(msg->level)) {
2359 /*
2360 * Skip record we have buffered and already printed
2361 * directly to the console when we received it, and
2362 * record that has level above the console loglevel.
2363 */
2364 console_idx = log_next(console_idx);
2365 console_seq++;
2366 goto skip;
2367 }
2368
2369 len += msg_print_text(msg,
2370 console_msg_format & MSG_FORMAT_SYSLOG,
2371 text + len,
2372 sizeof(text) - len);
2373 if (nr_ext_console_drivers) {
2374 ext_len = msg_print_ext_header(ext_text,
2375 sizeof(ext_text),
2376 msg, console_seq);
2377 ext_len += msg_print_ext_body(ext_text + ext_len,
2378 sizeof(ext_text) - ext_len,
2379 log_dict(msg), msg->dict_len,
2380 log_text(msg), msg->text_len);
2381 }
2382 console_idx = log_next(console_idx);
2383 console_seq++;
2384 raw_spin_unlock(&logbuf_lock);
2385
2386 /*
2387 * While actively printing out messages, if another printk()
2388 * were to occur on another CPU, it may wait for this one to
2389 * finish. This task can not be preempted if there is a
2390 * waiter waiting to take over.
2391 */
2392 console_lock_spinning_enable();
2393
2394 stop_critical_timings(); /* don't trace print latency */
2395 call_console_drivers(ext_text, ext_len, text, len);
2396 start_critical_timings();
2397
2398 if (console_lock_spinning_disable_and_check()) {
2399 printk_safe_exit_irqrestore(flags);
2400 goto out;
2401 }
2402
2403 printk_safe_exit_irqrestore(flags);
2404
2405 if (do_cond_resched)
2406 cond_resched();
2407 }
2408
2409 console_locked = 0;
2410
2411 /* Release the exclusive_console once it is used */
2412 if (unlikely(exclusive_console))
2413 exclusive_console = NULL;
2414
2415 raw_spin_unlock(&logbuf_lock);
2416
2417 up_console_sem();
2418
2419 /*
2420 * Someone could have filled up the buffer again, so re-check if there's
2421 * something to flush. In case we cannot trylock the console_sem again,
2422 * there's a new owner and the console_unlock() from them will do the
2423 * flush, no worries.
2424 */
2425 raw_spin_lock(&logbuf_lock);
2426 retry = console_seq != log_next_seq;
2427 raw_spin_unlock(&logbuf_lock);
2428 printk_safe_exit_irqrestore(flags);
2429
2430 if (retry && console_trylock())
2431 goto again;
2432
2433out:
2434 if (wake_klogd)
2435 wake_up_klogd();
2436}
2437EXPORT_SYMBOL(console_unlock);
2438
2439/**
2440 * console_conditional_schedule - yield the CPU if required
2441 *
2442 * If the console code is currently allowed to sleep, and
2443 * if this CPU should yield the CPU to another task, do
2444 * so here.
2445 *
2446 * Must be called within console_lock();.
2447 */
2448void __sched console_conditional_schedule(void)
2449{
2450 if (console_may_schedule)
2451 cond_resched();
2452}
2453EXPORT_SYMBOL(console_conditional_schedule);
2454
2455void console_unblank(void)
2456{
2457 struct console *c;
2458
2459 /*
2460 * console_unblank can no longer be called in interrupt context unless
2461 * oops_in_progress is set to 1..
2462 */
2463 if (oops_in_progress) {
2464 if (down_trylock_console_sem() != 0)
2465 return;
2466 } else
2467 console_lock();
2468
2469 console_locked = 1;
2470 console_may_schedule = 0;
2471 for_each_console(c)
2472 if ((c->flags & CON_ENABLED) && c->unblank)
2473 c->unblank();
2474 console_unlock();
2475}
2476
2477/**
2478 * console_flush_on_panic - flush console content on panic
2479 *
2480 * Immediately output all pending messages no matter what.
2481 */
2482void console_flush_on_panic(void)
2483{
2484 /*
2485 * If someone else is holding the console lock, trylock will fail
2486 * and may_schedule may be set. Ignore and proceed to unlock so
2487 * that messages are flushed out. As this can be called from any
2488 * context and we don't want to get preempted while flushing,
2489 * ensure may_schedule is cleared.
2490 */
2491 console_trylock();
2492 console_may_schedule = 0;
2493 console_unlock();
2494}
2495
2496/*
2497 * Return the console tty driver structure and its associated index
2498 */
2499struct tty_driver *console_device(int *index)
2500{
2501 struct console *c;
2502 struct tty_driver *driver = NULL;
2503
2504 console_lock();
2505 for_each_console(c) {
2506 if (!c->device)
2507 continue;
2508 driver = c->device(c, index);
2509 if (driver)
2510 break;
2511 }
2512 console_unlock();
2513 return driver;
2514}
2515
2516/*
2517 * Prevent further output on the passed console device so that (for example)
2518 * serial drivers can disable console output before suspending a port, and can
2519 * re-enable output afterwards.
2520 */
2521void console_stop(struct console *console)
2522{
2523 console_lock();
2524 console->flags &= ~CON_ENABLED;
2525 console_unlock();
2526}
2527EXPORT_SYMBOL(console_stop);
2528
2529void console_start(struct console *console)
2530{
2531 console_lock();
2532 console->flags |= CON_ENABLED;
2533 console_unlock();
2534}
2535EXPORT_SYMBOL(console_start);
2536
2537static int __read_mostly keep_bootcon;
2538
2539static int __init keep_bootcon_setup(char *str)
2540{
2541 keep_bootcon = 1;
2542 pr_info("debug: skip boot console de-registration.\n");
2543
2544 return 0;
2545}
2546
2547early_param("keep_bootcon", keep_bootcon_setup);
2548
2549/*
2550 * The console driver calls this routine during kernel initialization
2551 * to register the console printing procedure with printk() and to
2552 * print any messages that were printed by the kernel before the
2553 * console driver was initialized.
2554 *
2555 * This can happen pretty early during the boot process (because of
2556 * early_printk) - sometimes before setup_arch() completes - be careful
2557 * of what kernel features are used - they may not be initialised yet.
2558 *
2559 * There are two types of consoles - bootconsoles (early_printk) and
2560 * "real" consoles (everything which is not a bootconsole) which are
2561 * handled differently.
2562 * - Any number of bootconsoles can be registered at any time.
2563 * - As soon as a "real" console is registered, all bootconsoles
2564 * will be unregistered automatically.
2565 * - Once a "real" console is registered, any attempt to register a
2566 * bootconsoles will be rejected
2567 */
2568void register_console(struct console *newcon)
2569{
2570 int i;
2571 unsigned long flags;
2572 struct console *bcon = NULL;
2573 struct console_cmdline *c;
2574 static bool has_preferred;
2575
2576 if (console_drivers)
2577 for_each_console(bcon)
2578 if (WARN(bcon == newcon,
2579 "console '%s%d' already registered\n",
2580 bcon->name, bcon->index))
2581 return;
2582
2583 /*
2584 * before we register a new CON_BOOT console, make sure we don't
2585 * already have a valid console
2586 */
2587 if (console_drivers && newcon->flags & CON_BOOT) {
2588 /* find the last or real console */
2589 for_each_console(bcon) {
2590 if (!(bcon->flags & CON_BOOT)) {
2591 pr_info("Too late to register bootconsole %s%d\n",
2592 newcon->name, newcon->index);
2593 return;
2594 }
2595 }
2596 }
2597
2598 if (console_drivers && console_drivers->flags & CON_BOOT)
2599 bcon = console_drivers;
2600
2601 if (!has_preferred || bcon || !console_drivers)
2602 has_preferred = preferred_console >= 0;
2603
2604 /*
2605 * See if we want to use this console driver. If we
2606 * didn't select a console we take the first one
2607 * that registers here.
2608 */
2609 if (!has_preferred) {
2610 if (newcon->index < 0)
2611 newcon->index = 0;
2612 if (newcon->setup == NULL ||
2613 newcon->setup(newcon, NULL) == 0) {
2614 newcon->flags |= CON_ENABLED;
2615 if (newcon->device) {
2616 newcon->flags |= CON_CONSDEV;
2617 has_preferred = true;
2618 }
2619 }
2620 }
2621
2622 /*
2623 * See if this console matches one we selected on
2624 * the command line.
2625 */
2626 for (i = 0, c = console_cmdline;
2627 i < MAX_CMDLINECONSOLES && c->name[0];
2628 i++, c++) {
2629 if (!newcon->match ||
2630 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2631 /* default matching */
2632 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2633 if (strcmp(c->name, newcon->name) != 0)
2634 continue;
2635 if (newcon->index >= 0 &&
2636 newcon->index != c->index)
2637 continue;
2638 if (newcon->index < 0)
2639 newcon->index = c->index;
2640
2641 if (_braille_register_console(newcon, c))
2642 return;
2643
2644 if (newcon->setup &&
2645 newcon->setup(newcon, c->options) != 0)
2646 break;
2647 }
2648
2649 newcon->flags |= CON_ENABLED;
2650 if (i == preferred_console) {
2651 newcon->flags |= CON_CONSDEV;
2652 has_preferred = true;
2653 }
2654 break;
2655 }
2656
2657 if (!(newcon->flags & CON_ENABLED))
2658 return;
2659
2660 /*
2661 * If we have a bootconsole, and are switching to a real console,
2662 * don't print everything out again, since when the boot console, and
2663 * the real console are the same physical device, it's annoying to
2664 * see the beginning boot messages twice
2665 */
2666 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2667 newcon->flags &= ~CON_PRINTBUFFER;
2668
2669 /*
2670 * Put this console in the list - keep the
2671 * preferred driver at the head of the list.
2672 */
2673 console_lock();
2674 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2675 newcon->next = console_drivers;
2676 console_drivers = newcon;
2677 if (newcon->next)
2678 newcon->next->flags &= ~CON_CONSDEV;
2679 } else {
2680 newcon->next = console_drivers->next;
2681 console_drivers->next = newcon;
2682 }
2683
2684 if (newcon->flags & CON_EXTENDED)
2685 if (!nr_ext_console_drivers++)
2686 pr_info("printk: continuation disabled due to ext consoles, expect more fragments in /dev/kmsg\n");
2687
2688 if (newcon->flags & CON_PRINTBUFFER) {
2689 /*
2690 * console_unlock(); will print out the buffered messages
2691 * for us.
2692 */
2693 logbuf_lock_irqsave(flags);
2694 console_seq = syslog_seq;
2695 console_idx = syslog_idx;
2696 logbuf_unlock_irqrestore(flags);
2697 /*
2698 * We're about to replay the log buffer. Only do this to the
2699 * just-registered console to avoid excessive message spam to
2700 * the already-registered consoles.
2701 */
2702 exclusive_console = newcon;
2703 }
2704 console_unlock();
2705 console_sysfs_notify();
2706
2707 /*
2708 * By unregistering the bootconsoles after we enable the real console
2709 * we get the "console xxx enabled" message on all the consoles -
2710 * boot consoles, real consoles, etc - this is to ensure that end
2711 * users know there might be something in the kernel's log buffer that
2712 * went to the bootconsole (that they do not see on the real console)
2713 */
2714 pr_info("%sconsole [%s%d] enabled\n",
2715 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2716 newcon->name, newcon->index);
2717 if (bcon &&
2718 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2719 !keep_bootcon) {
2720 /* We need to iterate through all boot consoles, to make
2721 * sure we print everything out, before we unregister them.
2722 */
2723 for_each_console(bcon)
2724 if (bcon->flags & CON_BOOT)
2725 unregister_console(bcon);
2726 }
2727}
2728EXPORT_SYMBOL(register_console);
2729
2730int unregister_console(struct console *console)
2731{
2732 struct console *a, *b;
2733 int res;
2734
2735 pr_info("%sconsole [%s%d] disabled\n",
2736 (console->flags & CON_BOOT) ? "boot" : "" ,
2737 console->name, console->index);
2738
2739 res = _braille_unregister_console(console);
2740 if (res)
2741 return res;
2742
2743 res = 1;
2744 console_lock();
2745 if (console_drivers == console) {
2746 console_drivers=console->next;
2747 res = 0;
2748 } else if (console_drivers) {
2749 for (a=console_drivers->next, b=console_drivers ;
2750 a; b=a, a=b->next) {
2751 if (a == console) {
2752 b->next = a->next;
2753 res = 0;
2754 break;
2755 }
2756 }
2757 }
2758
2759 if (!res && (console->flags & CON_EXTENDED))
2760 nr_ext_console_drivers--;
2761
2762 /*
2763 * If this isn't the last console and it has CON_CONSDEV set, we
2764 * need to set it on the next preferred console.
2765 */
2766 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2767 console_drivers->flags |= CON_CONSDEV;
2768
2769 console->flags &= ~CON_ENABLED;
2770 console_unlock();
2771 console_sysfs_notify();
2772 return res;
2773}
2774EXPORT_SYMBOL(unregister_console);
2775
2776/*
2777 * Initialize the console device. This is called *early*, so
2778 * we can't necessarily depend on lots of kernel help here.
2779 * Just do some early initializations, and do the complex setup
2780 * later.
2781 */
2782void __init console_init(void)
2783{
2784 int ret;
2785 initcall_t *call;
2786
2787 /* Setup the default TTY line discipline. */
2788 n_tty_init();
2789
2790 /*
2791 * set up the console device so that later boot sequences can
2792 * inform about problems etc..
2793 */
2794 call = __con_initcall_start;
2795 trace_initcall_level("console");
2796 while (call < __con_initcall_end) {
2797 trace_initcall_start((*call));
2798 ret = (*call)();
2799 trace_initcall_finish((*call), ret);
2800 call++;
2801 }
2802}
2803
2804/*
2805 * Some boot consoles access data that is in the init section and which will
2806 * be discarded after the initcalls have been run. To make sure that no code
2807 * will access this data, unregister the boot consoles in a late initcall.
2808 *
2809 * If for some reason, such as deferred probe or the driver being a loadable
2810 * module, the real console hasn't registered yet at this point, there will
2811 * be a brief interval in which no messages are logged to the console, which
2812 * makes it difficult to diagnose problems that occur during this time.
2813 *
2814 * To mitigate this problem somewhat, only unregister consoles whose memory
2815 * intersects with the init section. Note that all other boot consoles will
2816 * get unregistred when the real preferred console is registered.
2817 */
2818static int __init printk_late_init(void)
2819{
2820 struct console *con;
2821 int ret;
2822
2823 for_each_console(con) {
2824 if (!(con->flags & CON_BOOT))
2825 continue;
2826
2827 /* Check addresses that might be used for enabled consoles. */
2828 if (init_section_intersects(con, sizeof(*con)) ||
2829 init_section_contains(con->write, 0) ||
2830 init_section_contains(con->read, 0) ||
2831 init_section_contains(con->device, 0) ||
2832 init_section_contains(con->unblank, 0) ||
2833 init_section_contains(con->data, 0)) {
2834 /*
2835 * Please, consider moving the reported consoles out
2836 * of the init section.
2837 */
2838 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2839 con->name, con->index);
2840 unregister_console(con);
2841 }
2842 }
2843 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2844 console_cpu_notify);
2845 WARN_ON(ret < 0);
2846 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2847 console_cpu_notify, NULL);
2848 WARN_ON(ret < 0);
2849 return 0;
2850}
2851late_initcall(printk_late_init);
2852
2853#if defined CONFIG_PRINTK
2854/*
2855 * Delayed printk version, for scheduler-internal messages:
2856 */
2857#define PRINTK_PENDING_WAKEUP 0x01
2858#define PRINTK_PENDING_OUTPUT 0x02
2859
2860static DEFINE_PER_CPU(int, printk_pending);
2861
2862static void wake_up_klogd_work_func(struct irq_work *irq_work)
2863{
2864 int pending = __this_cpu_xchg(printk_pending, 0);
2865
2866 if (pending & PRINTK_PENDING_OUTPUT) {
2867 /* If trylock fails, someone else is doing the printing */
2868 if (console_trylock())
2869 console_unlock();
2870 }
2871
2872 if (pending & PRINTK_PENDING_WAKEUP)
2873 wake_up_interruptible(&log_wait);
2874}
2875
2876static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2877 .func = wake_up_klogd_work_func,
2878 .flags = IRQ_WORK_LAZY,
2879};
2880
2881void wake_up_klogd(void)
2882{
2883 preempt_disable();
2884 if (waitqueue_active(&log_wait)) {
2885 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2886 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2887 }
2888 preempt_enable();
2889}
2890
2891int vprintk_deferred(const char *fmt, va_list args)
2892{
2893 int r;
2894
2895 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2896
2897 preempt_disable();
2898 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2899 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2900 preempt_enable();
2901
2902 return r;
2903}
2904
2905int printk_deferred(const char *fmt, ...)
2906{
2907 va_list args;
2908 int r;
2909
2910 va_start(args, fmt);
2911 r = vprintk_deferred(fmt, args);
2912 va_end(args);
2913
2914 return r;
2915}
2916
2917/*
2918 * printk rate limiting, lifted from the networking subsystem.
2919 *
2920 * This enforces a rate limit: not more than 10 kernel messages
2921 * every 5s to make a denial-of-service attack impossible.
2922 */
2923DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2924
2925int __printk_ratelimit(const char *func)
2926{
2927 return ___ratelimit(&printk_ratelimit_state, func);
2928}
2929EXPORT_SYMBOL(__printk_ratelimit);
2930
2931/**
2932 * printk_timed_ratelimit - caller-controlled printk ratelimiting
2933 * @caller_jiffies: pointer to caller's state
2934 * @interval_msecs: minimum interval between prints
2935 *
2936 * printk_timed_ratelimit() returns true if more than @interval_msecs
2937 * milliseconds have elapsed since the last time printk_timed_ratelimit()
2938 * returned true.
2939 */
2940bool printk_timed_ratelimit(unsigned long *caller_jiffies,
2941 unsigned int interval_msecs)
2942{
2943 unsigned long elapsed = jiffies - *caller_jiffies;
2944
2945 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
2946 return false;
2947
2948 *caller_jiffies = jiffies;
2949 return true;
2950}
2951EXPORT_SYMBOL(printk_timed_ratelimit);
2952
2953static DEFINE_SPINLOCK(dump_list_lock);
2954static LIST_HEAD(dump_list);
2955
2956/**
2957 * kmsg_dump_register - register a kernel log dumper.
2958 * @dumper: pointer to the kmsg_dumper structure
2959 *
2960 * Adds a kernel log dumper to the system. The dump callback in the
2961 * structure will be called when the kernel oopses or panics and must be
2962 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
2963 */
2964int kmsg_dump_register(struct kmsg_dumper *dumper)
2965{
2966 unsigned long flags;
2967 int err = -EBUSY;
2968
2969 /* The dump callback needs to be set */
2970 if (!dumper->dump)
2971 return -EINVAL;
2972
2973 spin_lock_irqsave(&dump_list_lock, flags);
2974 /* Don't allow registering multiple times */
2975 if (!dumper->registered) {
2976 dumper->registered = 1;
2977 list_add_tail_rcu(&dumper->list, &dump_list);
2978 err = 0;
2979 }
2980 spin_unlock_irqrestore(&dump_list_lock, flags);
2981
2982 return err;
2983}
2984EXPORT_SYMBOL_GPL(kmsg_dump_register);
2985
2986/**
2987 * kmsg_dump_unregister - unregister a kmsg dumper.
2988 * @dumper: pointer to the kmsg_dumper structure
2989 *
2990 * Removes a dump device from the system. Returns zero on success and
2991 * %-EINVAL otherwise.
2992 */
2993int kmsg_dump_unregister(struct kmsg_dumper *dumper)
2994{
2995 unsigned long flags;
2996 int err = -EINVAL;
2997
2998 spin_lock_irqsave(&dump_list_lock, flags);
2999 if (dumper->registered) {
3000 dumper->registered = 0;
3001 list_del_rcu(&dumper->list);
3002 err = 0;
3003 }
3004 spin_unlock_irqrestore(&dump_list_lock, flags);
3005 synchronize_rcu();
3006
3007 return err;
3008}
3009EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3010
3011static bool always_kmsg_dump;
3012module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3013
3014/**
3015 * kmsg_dump - dump kernel log to kernel message dumpers.
3016 * @reason: the reason (oops, panic etc) for dumping
3017 *
3018 * Call each of the registered dumper's dump() callback, which can
3019 * retrieve the kmsg records with kmsg_dump_get_line() or
3020 * kmsg_dump_get_buffer().
3021 */
3022void kmsg_dump(enum kmsg_dump_reason reason)
3023{
3024 struct kmsg_dumper *dumper;
3025 unsigned long flags;
3026
3027 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3028 return;
3029
3030 rcu_read_lock();
3031 list_for_each_entry_rcu(dumper, &dump_list, list) {
3032 if (dumper->max_reason && reason > dumper->max_reason)
3033 continue;
3034
3035 /* initialize iterator with data about the stored records */
3036 dumper->active = true;
3037
3038 logbuf_lock_irqsave(flags);
3039 dumper->cur_seq = clear_seq;
3040 dumper->cur_idx = clear_idx;
3041 dumper->next_seq = log_next_seq;
3042 dumper->next_idx = log_next_idx;
3043 logbuf_unlock_irqrestore(flags);
3044
3045 /* invoke dumper which will iterate over records */
3046 dumper->dump(dumper, reason);
3047
3048 /* reset iterator */
3049 dumper->active = false;
3050 }
3051 rcu_read_unlock();
3052}
3053
3054/**
3055 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3056 * @dumper: registered kmsg dumper
3057 * @syslog: include the "<4>" prefixes
3058 * @line: buffer to copy the line to
3059 * @size: maximum size of the buffer
3060 * @len: length of line placed into buffer
3061 *
3062 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3063 * record, and copy one record into the provided buffer.
3064 *
3065 * Consecutive calls will return the next available record moving
3066 * towards the end of the buffer with the youngest messages.
3067 *
3068 * A return value of FALSE indicates that there are no more records to
3069 * read.
3070 *
3071 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3072 */
3073bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3074 char *line, size_t size, size_t *len)
3075{
3076 struct printk_log *msg;
3077 size_t l = 0;
3078 bool ret = false;
3079
3080 if (!dumper->active)
3081 goto out;
3082
3083 if (dumper->cur_seq < log_first_seq) {
3084 /* messages are gone, move to first available one */
3085 dumper->cur_seq = log_first_seq;
3086 dumper->cur_idx = log_first_idx;
3087 }
3088
3089 /* last entry */
3090 if (dumper->cur_seq >= log_next_seq)
3091 goto out;
3092
3093 msg = log_from_idx(dumper->cur_idx);
3094 l = msg_print_text(msg, syslog, line, size);
3095
3096 dumper->cur_idx = log_next(dumper->cur_idx);
3097 dumper->cur_seq++;
3098 ret = true;
3099out:
3100 if (len)
3101 *len = l;
3102 return ret;
3103}
3104
3105/**
3106 * kmsg_dump_get_line - retrieve one kmsg log line
3107 * @dumper: registered kmsg dumper
3108 * @syslog: include the "<4>" prefixes
3109 * @line: buffer to copy the line to
3110 * @size: maximum size of the buffer
3111 * @len: length of line placed into buffer
3112 *
3113 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3114 * record, and copy one record into the provided buffer.
3115 *
3116 * Consecutive calls will return the next available record moving
3117 * towards the end of the buffer with the youngest messages.
3118 *
3119 * A return value of FALSE indicates that there are no more records to
3120 * read.
3121 */
3122bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3123 char *line, size_t size, size_t *len)
3124{
3125 unsigned long flags;
3126 bool ret;
3127
3128 logbuf_lock_irqsave(flags);
3129 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3130 logbuf_unlock_irqrestore(flags);
3131
3132 return ret;
3133}
3134EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3135
3136/**
3137 * kmsg_dump_get_buffer - copy kmsg log lines
3138 * @dumper: registered kmsg dumper
3139 * @syslog: include the "<4>" prefixes
3140 * @buf: buffer to copy the line to
3141 * @size: maximum size of the buffer
3142 * @len: length of line placed into buffer
3143 *
3144 * Start at the end of the kmsg buffer and fill the provided buffer
3145 * with as many of the the *youngest* kmsg records that fit into it.
3146 * If the buffer is large enough, all available kmsg records will be
3147 * copied with a single call.
3148 *
3149 * Consecutive calls will fill the buffer with the next block of
3150 * available older records, not including the earlier retrieved ones.
3151 *
3152 * A return value of FALSE indicates that there are no more records to
3153 * read.
3154 */
3155bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3156 char *buf, size_t size, size_t *len)
3157{
3158 unsigned long flags;
3159 u64 seq;
3160 u32 idx;
3161 u64 next_seq;
3162 u32 next_idx;
3163 size_t l = 0;
3164 bool ret = false;
3165
3166 if (!dumper->active)
3167 goto out;
3168
3169 logbuf_lock_irqsave(flags);
3170 if (dumper->cur_seq < log_first_seq) {
3171 /* messages are gone, move to first available one */
3172 dumper->cur_seq = log_first_seq;
3173 dumper->cur_idx = log_first_idx;
3174 }
3175
3176 /* last entry */
3177 if (dumper->cur_seq >= dumper->next_seq) {
3178 logbuf_unlock_irqrestore(flags);
3179 goto out;
3180 }
3181
3182 /* calculate length of entire buffer */
3183 seq = dumper->cur_seq;
3184 idx = dumper->cur_idx;
3185 while (seq < dumper->next_seq) {
3186 struct printk_log *msg = log_from_idx(idx);
3187
3188 l += msg_print_text(msg, true, NULL, 0);
3189 idx = log_next(idx);
3190 seq++;
3191 }
3192
3193 /* move first record forward until length fits into the buffer */
3194 seq = dumper->cur_seq;
3195 idx = dumper->cur_idx;
3196 while (l > size && seq < dumper->next_seq) {
3197 struct printk_log *msg = log_from_idx(idx);
3198
3199 l -= msg_print_text(msg, true, NULL, 0);
3200 idx = log_next(idx);
3201 seq++;
3202 }
3203
3204 /* last message in next interation */
3205 next_seq = seq;
3206 next_idx = idx;
3207
3208 l = 0;
3209 while (seq < dumper->next_seq) {
3210 struct printk_log *msg = log_from_idx(idx);
3211
3212 l += msg_print_text(msg, syslog, buf + l, size - l);
3213 idx = log_next(idx);
3214 seq++;
3215 }
3216
3217 dumper->next_seq = next_seq;
3218 dumper->next_idx = next_idx;
3219 ret = true;
3220 logbuf_unlock_irqrestore(flags);
3221out:
3222 if (len)
3223 *len = l;
3224 return ret;
3225}
3226EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3227
3228/**
3229 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3230 * @dumper: registered kmsg dumper
3231 *
3232 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3233 * kmsg_dump_get_buffer() can be called again and used multiple
3234 * times within the same dumper.dump() callback.
3235 *
3236 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3237 */
3238void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3239{
3240 dumper->cur_seq = clear_seq;
3241 dumper->cur_idx = clear_idx;
3242 dumper->next_seq = log_next_seq;
3243 dumper->next_idx = log_next_idx;
3244}
3245
3246/**
3247 * kmsg_dump_rewind - reset the interator
3248 * @dumper: registered kmsg dumper
3249 *
3250 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3251 * kmsg_dump_get_buffer() can be called again and used multiple
3252 * times within the same dumper.dump() callback.
3253 */
3254void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3255{
3256 unsigned long flags;
3257
3258 logbuf_lock_irqsave(flags);
3259 kmsg_dump_rewind_nolock(dumper);
3260 logbuf_unlock_irqrestore(flags);
3261}
3262EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3263
3264#endif