Loading...
Note: File does not exist in v3.1.
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/ratelimit.h>
39#include <linux/kmsg_dump.h>
40#include <linux/syslog.h>
41#include <linux/cpu.h>
42#include <linux/rculist.h>
43#include <linux/poll.h>
44#include <linux/irq_work.h>
45#include <linux/ctype.h>
46#include <linux/uio.h>
47#include <linux/sched/clock.h>
48#include <linux/sched/debug.h>
49#include <linux/sched/task_stack.h>
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};
68EXPORT_SYMBOL_GPL(console_printk);
69
70atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
71EXPORT_SYMBOL(ignore_console_lock_warning);
72
73/*
74 * Low level drivers may need that to know if they can schedule in
75 * their unblank() callback or not. So let's export it.
76 */
77int oops_in_progress;
78EXPORT_SYMBOL(oops_in_progress);
79
80/*
81 * console_sem protects the console_drivers list, and also
82 * provides serialisation for access to the entire console
83 * driver system.
84 */
85static DEFINE_SEMAPHORE(console_sem);
86struct console *console_drivers;
87EXPORT_SYMBOL_GPL(console_drivers);
88
89/*
90 * System may need to suppress printk message under certain
91 * circumstances, like after kernel panic happens.
92 */
93int __read_mostly suppress_printk;
94
95#ifdef CONFIG_LOCKDEP
96static struct lockdep_map console_lock_dep_map = {
97 .name = "console_lock"
98};
99#endif
100
101enum devkmsg_log_bits {
102 __DEVKMSG_LOG_BIT_ON = 0,
103 __DEVKMSG_LOG_BIT_OFF,
104 __DEVKMSG_LOG_BIT_LOCK,
105};
106
107enum devkmsg_log_masks {
108 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
109 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
110 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
111};
112
113/* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
114#define DEVKMSG_LOG_MASK_DEFAULT 0
115
116static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
117
118static int __control_devkmsg(char *str)
119{
120 size_t len;
121
122 if (!str)
123 return -EINVAL;
124
125 len = str_has_prefix(str, "on");
126 if (len) {
127 devkmsg_log = DEVKMSG_LOG_MASK_ON;
128 return len;
129 }
130
131 len = str_has_prefix(str, "off");
132 if (len) {
133 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
134 return len;
135 }
136
137 len = str_has_prefix(str, "ratelimit");
138 if (len) {
139 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
140 return len;
141 }
142
143 return -EINVAL;
144}
145
146static int __init control_devkmsg(char *str)
147{
148 if (__control_devkmsg(str) < 0)
149 return 1;
150
151 /*
152 * Set sysctl string accordingly:
153 */
154 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
155 strcpy(devkmsg_log_str, "on");
156 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
157 strcpy(devkmsg_log_str, "off");
158 /* else "ratelimit" which is set by default. */
159
160 /*
161 * Sysctl cannot change it anymore. The kernel command line setting of
162 * this parameter is to force the setting to be permanent throughout the
163 * runtime of the system. This is a precation measure against userspace
164 * trying to be a smarta** and attempting to change it up on us.
165 */
166 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
167
168 return 0;
169}
170__setup("printk.devkmsg=", control_devkmsg);
171
172char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
173
174int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
175 void *buffer, size_t *lenp, loff_t *ppos)
176{
177 char old_str[DEVKMSG_STR_MAX_SIZE];
178 unsigned int old;
179 int err;
180
181 if (write) {
182 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
183 return -EINVAL;
184
185 old = devkmsg_log;
186 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
187 }
188
189 err = proc_dostring(table, write, buffer, lenp, ppos);
190 if (err)
191 return err;
192
193 if (write) {
194 err = __control_devkmsg(devkmsg_log_str);
195
196 /*
197 * Do not accept an unknown string OR a known string with
198 * trailing crap...
199 */
200 if (err < 0 || (err + 1 != *lenp)) {
201
202 /* ... and restore old setting. */
203 devkmsg_log = old;
204 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
205
206 return -EINVAL;
207 }
208 }
209
210 return 0;
211}
212
213/* Number of registered extended console drivers. */
214static int nr_ext_console_drivers;
215
216/*
217 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
218 * macros instead of functions so that _RET_IP_ contains useful information.
219 */
220#define down_console_sem() do { \
221 down(&console_sem);\
222 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
223} while (0)
224
225static int __down_trylock_console_sem(unsigned long ip)
226{
227 int lock_failed;
228 unsigned long flags;
229
230 /*
231 * Here and in __up_console_sem() we need to be in safe mode,
232 * because spindump/WARN/etc from under console ->lock will
233 * deadlock in printk()->down_trylock_console_sem() otherwise.
234 */
235 printk_safe_enter_irqsave(flags);
236 lock_failed = down_trylock(&console_sem);
237 printk_safe_exit_irqrestore(flags);
238
239 if (lock_failed)
240 return 1;
241 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
242 return 0;
243}
244#define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
245
246static void __up_console_sem(unsigned long ip)
247{
248 unsigned long flags;
249
250 mutex_release(&console_lock_dep_map, ip);
251
252 printk_safe_enter_irqsave(flags);
253 up(&console_sem);
254 printk_safe_exit_irqrestore(flags);
255}
256#define up_console_sem() __up_console_sem(_RET_IP_)
257
258/*
259 * This is used for debugging the mess that is the VT code by
260 * keeping track if we have the console semaphore held. It's
261 * definitely not the perfect debug tool (we don't know if _WE_
262 * hold it and are racing, but it helps tracking those weird code
263 * paths in the console code where we end up in places I want
264 * locked without the console sempahore held).
265 */
266static int console_locked, console_suspended;
267
268/*
269 * If exclusive_console is non-NULL then only this console is to be printed to.
270 */
271static struct console *exclusive_console;
272
273/*
274 * Array of consoles built from command line options (console=)
275 */
276
277#define MAX_CMDLINECONSOLES 8
278
279static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
280
281static int preferred_console = -1;
282static bool has_preferred_console;
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/*
464 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
465 * per_cpu_areas are initialised. This variable is set to true when
466 * it's safe to access per-CPU data.
467 */
468static bool __printk_percpu_data_ready __read_mostly;
469
470bool printk_percpu_data_ready(void)
471{
472 return __printk_percpu_data_ready;
473}
474
475/* Return log buffer address */
476char *log_buf_addr_get(void)
477{
478 return log_buf;
479}
480
481/* Return log buffer size */
482u32 log_buf_len_get(void)
483{
484 return log_buf_len;
485}
486
487/* human readable text of the record */
488static char *log_text(const struct printk_log *msg)
489{
490 return (char *)msg + sizeof(struct printk_log);
491}
492
493/* optional key/value pair dictionary attached to the record */
494static char *log_dict(const struct printk_log *msg)
495{
496 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
497}
498
499/* get record by index; idx must point to valid msg */
500static struct printk_log *log_from_idx(u32 idx)
501{
502 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
503
504 /*
505 * A length == 0 record is the end of buffer marker. Wrap around and
506 * read the message at the start of the buffer.
507 */
508 if (!msg->len)
509 return (struct printk_log *)log_buf;
510 return msg;
511}
512
513/* get next record; idx must point to valid msg */
514static u32 log_next(u32 idx)
515{
516 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
517
518 /* length == 0 indicates the end of the buffer; wrap */
519 /*
520 * A length == 0 record is the end of buffer marker. Wrap around and
521 * read the message at the start of the buffer as *this* one, and
522 * return the one after that.
523 */
524 if (!msg->len) {
525 msg = (struct printk_log *)log_buf;
526 return msg->len;
527 }
528 return idx + msg->len;
529}
530
531/*
532 * Check whether there is enough free space for the given message.
533 *
534 * The same values of first_idx and next_idx mean that the buffer
535 * is either empty or full.
536 *
537 * If the buffer is empty, we must respect the position of the indexes.
538 * They cannot be reset to the beginning of the buffer.
539 */
540static int logbuf_has_space(u32 msg_size, bool empty)
541{
542 u32 free;
543
544 if (log_next_idx > log_first_idx || empty)
545 free = max(log_buf_len - log_next_idx, log_first_idx);
546 else
547 free = log_first_idx - log_next_idx;
548
549 /*
550 * We need space also for an empty header that signalizes wrapping
551 * of the buffer.
552 */
553 return free >= msg_size + sizeof(struct printk_log);
554}
555
556static int log_make_free_space(u32 msg_size)
557{
558 while (log_first_seq < log_next_seq &&
559 !logbuf_has_space(msg_size, false)) {
560 /* drop old messages until we have enough contiguous space */
561 log_first_idx = log_next(log_first_idx);
562 log_first_seq++;
563 }
564
565 if (clear_seq < log_first_seq) {
566 clear_seq = log_first_seq;
567 clear_idx = log_first_idx;
568 }
569
570 /* sequence numbers are equal, so the log buffer is empty */
571 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
572 return 0;
573
574 return -ENOMEM;
575}
576
577/* compute the message size including the padding bytes */
578static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
579{
580 u32 size;
581
582 size = sizeof(struct printk_log) + text_len + dict_len;
583 *pad_len = (-size) & (LOG_ALIGN - 1);
584 size += *pad_len;
585
586 return size;
587}
588
589/*
590 * Define how much of the log buffer we could take at maximum. The value
591 * must be greater than two. Note that only half of the buffer is available
592 * when the index points to the middle.
593 */
594#define MAX_LOG_TAKE_PART 4
595static const char trunc_msg[] = "<truncated>";
596
597static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
598 u16 *dict_len, u32 *pad_len)
599{
600 /*
601 * The message should not take the whole buffer. Otherwise, it might
602 * get removed too soon.
603 */
604 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
605 if (*text_len > max_text_len)
606 *text_len = max_text_len;
607 /* enable the warning message */
608 *trunc_msg_len = strlen(trunc_msg);
609 /* disable the "dict" completely */
610 *dict_len = 0;
611 /* compute the size again, count also the warning message */
612 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
613}
614
615/* insert record into the buffer, discard old ones, update heads */
616static int log_store(u32 caller_id, int facility, int level,
617 enum log_flags flags, u64 ts_nsec,
618 const char *dict, u16 dict_len,
619 const char *text, u16 text_len)
620{
621 struct printk_log *msg;
622 u32 size, pad_len;
623 u16 trunc_msg_len = 0;
624
625 /* number of '\0' padding bytes to next message */
626 size = msg_used_size(text_len, dict_len, &pad_len);
627
628 if (log_make_free_space(size)) {
629 /* truncate the message if it is too long for empty buffer */
630 size = truncate_msg(&text_len, &trunc_msg_len,
631 &dict_len, &pad_len);
632 /* survive when the log buffer is too small for trunc_msg */
633 if (log_make_free_space(size))
634 return 0;
635 }
636
637 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
638 /*
639 * This message + an additional empty header does not fit
640 * at the end of the buffer. Add an empty header with len == 0
641 * to signify a wrap around.
642 */
643 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
644 log_next_idx = 0;
645 }
646
647 /* fill message */
648 msg = (struct printk_log *)(log_buf + log_next_idx);
649 memcpy(log_text(msg), text, text_len);
650 msg->text_len = text_len;
651 if (trunc_msg_len) {
652 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
653 msg->text_len += trunc_msg_len;
654 }
655 memcpy(log_dict(msg), dict, dict_len);
656 msg->dict_len = dict_len;
657 msg->facility = facility;
658 msg->level = level & 7;
659 msg->flags = flags & 0x1f;
660 if (ts_nsec > 0)
661 msg->ts_nsec = ts_nsec;
662 else
663 msg->ts_nsec = local_clock();
664#ifdef CONFIG_PRINTK_CALLER
665 msg->caller_id = caller_id;
666#endif
667 memset(log_dict(msg) + dict_len, 0, pad_len);
668 msg->len = size;
669
670 /* insert message */
671 log_next_idx += msg->len;
672 log_next_seq++;
673
674 return msg->text_len;
675}
676
677int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
678
679static int syslog_action_restricted(int type)
680{
681 if (dmesg_restrict)
682 return 1;
683 /*
684 * Unless restricted, we allow "read all" and "get buffer size"
685 * for everybody.
686 */
687 return type != SYSLOG_ACTION_READ_ALL &&
688 type != SYSLOG_ACTION_SIZE_BUFFER;
689}
690
691static int check_syslog_permissions(int type, int source)
692{
693 /*
694 * If this is from /proc/kmsg and we've already opened it, then we've
695 * already done the capabilities checks at open time.
696 */
697 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
698 goto ok;
699
700 if (syslog_action_restricted(type)) {
701 if (capable(CAP_SYSLOG))
702 goto ok;
703 /*
704 * For historical reasons, accept CAP_SYS_ADMIN too, with
705 * a warning.
706 */
707 if (capable(CAP_SYS_ADMIN)) {
708 pr_warn_once("%s (%d): Attempt to access syslog with "
709 "CAP_SYS_ADMIN but no CAP_SYSLOG "
710 "(deprecated).\n",
711 current->comm, task_pid_nr(current));
712 goto ok;
713 }
714 return -EPERM;
715 }
716ok:
717 return security_syslog(type);
718}
719
720static void append_char(char **pp, char *e, char c)
721{
722 if (*pp < e)
723 *(*pp)++ = c;
724}
725
726static ssize_t msg_print_ext_header(char *buf, size_t size,
727 struct printk_log *msg, u64 seq)
728{
729 u64 ts_usec = msg->ts_nsec;
730 char caller[20];
731#ifdef CONFIG_PRINTK_CALLER
732 u32 id = msg->caller_id;
733
734 snprintf(caller, sizeof(caller), ",caller=%c%u",
735 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
736#else
737 caller[0] = '\0';
738#endif
739
740 do_div(ts_usec, 1000);
741
742 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
743 (msg->facility << 3) | msg->level, seq, ts_usec,
744 msg->flags & LOG_CONT ? 'c' : '-', caller);
745}
746
747static ssize_t msg_print_ext_body(char *buf, size_t size,
748 char *dict, size_t dict_len,
749 char *text, size_t text_len)
750{
751 char *p = buf, *e = buf + size;
752 size_t i;
753
754 /* escape non-printable characters */
755 for (i = 0; i < text_len; i++) {
756 unsigned char c = text[i];
757
758 if (c < ' ' || c >= 127 || c == '\\')
759 p += scnprintf(p, e - p, "\\x%02x", c);
760 else
761 append_char(&p, e, c);
762 }
763 append_char(&p, e, '\n');
764
765 if (dict_len) {
766 bool line = true;
767
768 for (i = 0; i < dict_len; i++) {
769 unsigned char c = dict[i];
770
771 if (line) {
772 append_char(&p, e, ' ');
773 line = false;
774 }
775
776 if (c == '\0') {
777 append_char(&p, e, '\n');
778 line = true;
779 continue;
780 }
781
782 if (c < ' ' || c >= 127 || c == '\\') {
783 p += scnprintf(p, e - p, "\\x%02x", c);
784 continue;
785 }
786
787 append_char(&p, e, c);
788 }
789 append_char(&p, e, '\n');
790 }
791
792 return p - buf;
793}
794
795/* /dev/kmsg - userspace message inject/listen interface */
796struct devkmsg_user {
797 u64 seq;
798 u32 idx;
799 struct ratelimit_state rs;
800 struct mutex lock;
801 char buf[CONSOLE_EXT_LOG_MAX];
802};
803
804static __printf(3, 4) __cold
805int devkmsg_emit(int facility, int level, const char *fmt, ...)
806{
807 va_list args;
808 int r;
809
810 va_start(args, fmt);
811 r = vprintk_emit(facility, level, NULL, 0, fmt, args);
812 va_end(args);
813
814 return r;
815}
816
817static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
818{
819 char *buf, *line;
820 int level = default_message_loglevel;
821 int facility = 1; /* LOG_USER */
822 struct file *file = iocb->ki_filp;
823 struct devkmsg_user *user = file->private_data;
824 size_t len = iov_iter_count(from);
825 ssize_t ret = len;
826
827 if (!user || len > LOG_LINE_MAX)
828 return -EINVAL;
829
830 /* Ignore when user logging is disabled. */
831 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
832 return len;
833
834 /* Ratelimit when not explicitly enabled. */
835 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
836 if (!___ratelimit(&user->rs, current->comm))
837 return ret;
838 }
839
840 buf = kmalloc(len+1, GFP_KERNEL);
841 if (buf == NULL)
842 return -ENOMEM;
843
844 buf[len] = '\0';
845 if (!copy_from_iter_full(buf, len, from)) {
846 kfree(buf);
847 return -EFAULT;
848 }
849
850 /*
851 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
852 * the decimal value represents 32bit, the lower 3 bit are the log
853 * level, the rest are the log facility.
854 *
855 * If no prefix or no userspace facility is specified, we
856 * enforce LOG_USER, to be able to reliably distinguish
857 * kernel-generated messages from userspace-injected ones.
858 */
859 line = buf;
860 if (line[0] == '<') {
861 char *endp = NULL;
862 unsigned int u;
863
864 u = simple_strtoul(line + 1, &endp, 10);
865 if (endp && endp[0] == '>') {
866 level = LOG_LEVEL(u);
867 if (LOG_FACILITY(u) != 0)
868 facility = LOG_FACILITY(u);
869 endp++;
870 len -= endp - line;
871 line = endp;
872 }
873 }
874
875 devkmsg_emit(facility, level, "%s", line);
876 kfree(buf);
877 return ret;
878}
879
880static ssize_t devkmsg_read(struct file *file, char __user *buf,
881 size_t count, loff_t *ppos)
882{
883 struct devkmsg_user *user = file->private_data;
884 struct printk_log *msg;
885 size_t len;
886 ssize_t ret;
887
888 if (!user)
889 return -EBADF;
890
891 ret = mutex_lock_interruptible(&user->lock);
892 if (ret)
893 return ret;
894
895 logbuf_lock_irq();
896 while (user->seq == log_next_seq) {
897 if (file->f_flags & O_NONBLOCK) {
898 ret = -EAGAIN;
899 logbuf_unlock_irq();
900 goto out;
901 }
902
903 logbuf_unlock_irq();
904 ret = wait_event_interruptible(log_wait,
905 user->seq != log_next_seq);
906 if (ret)
907 goto out;
908 logbuf_lock_irq();
909 }
910
911 if (user->seq < log_first_seq) {
912 /* our last seen message is gone, return error and reset */
913 user->idx = log_first_idx;
914 user->seq = log_first_seq;
915 ret = -EPIPE;
916 logbuf_unlock_irq();
917 goto out;
918 }
919
920 msg = log_from_idx(user->idx);
921 len = msg_print_ext_header(user->buf, sizeof(user->buf),
922 msg, user->seq);
923 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
924 log_dict(msg), msg->dict_len,
925 log_text(msg), msg->text_len);
926
927 user->idx = log_next(user->idx);
928 user->seq++;
929 logbuf_unlock_irq();
930
931 if (len > count) {
932 ret = -EINVAL;
933 goto out;
934 }
935
936 if (copy_to_user(buf, user->buf, len)) {
937 ret = -EFAULT;
938 goto out;
939 }
940 ret = len;
941out:
942 mutex_unlock(&user->lock);
943 return ret;
944}
945
946/*
947 * Be careful when modifying this function!!!
948 *
949 * Only few operations are supported because the device works only with the
950 * entire variable length messages (records). Non-standard values are
951 * returned in the other cases and has been this way for quite some time.
952 * User space applications might depend on this behavior.
953 */
954static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
955{
956 struct devkmsg_user *user = file->private_data;
957 loff_t ret = 0;
958
959 if (!user)
960 return -EBADF;
961 if (offset)
962 return -ESPIPE;
963
964 logbuf_lock_irq();
965 switch (whence) {
966 case SEEK_SET:
967 /* the first record */
968 user->idx = log_first_idx;
969 user->seq = log_first_seq;
970 break;
971 case SEEK_DATA:
972 /*
973 * The first record after the last SYSLOG_ACTION_CLEAR,
974 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
975 * changes no global state, and does not clear anything.
976 */
977 user->idx = clear_idx;
978 user->seq = clear_seq;
979 break;
980 case SEEK_END:
981 /* after the last record */
982 user->idx = log_next_idx;
983 user->seq = log_next_seq;
984 break;
985 default:
986 ret = -EINVAL;
987 }
988 logbuf_unlock_irq();
989 return ret;
990}
991
992static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
993{
994 struct devkmsg_user *user = file->private_data;
995 __poll_t ret = 0;
996
997 if (!user)
998 return EPOLLERR|EPOLLNVAL;
999
1000 poll_wait(file, &log_wait, wait);
1001
1002 logbuf_lock_irq();
1003 if (user->seq < log_next_seq) {
1004 /* return error when data has vanished underneath us */
1005 if (user->seq < log_first_seq)
1006 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
1007 else
1008 ret = EPOLLIN|EPOLLRDNORM;
1009 }
1010 logbuf_unlock_irq();
1011
1012 return ret;
1013}
1014
1015static int devkmsg_open(struct inode *inode, struct file *file)
1016{
1017 struct devkmsg_user *user;
1018 int err;
1019
1020 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
1021 return -EPERM;
1022
1023 /* write-only does not need any file context */
1024 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
1025 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
1026 SYSLOG_FROM_READER);
1027 if (err)
1028 return err;
1029 }
1030
1031 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
1032 if (!user)
1033 return -ENOMEM;
1034
1035 ratelimit_default_init(&user->rs);
1036 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
1037
1038 mutex_init(&user->lock);
1039
1040 logbuf_lock_irq();
1041 user->idx = log_first_idx;
1042 user->seq = log_first_seq;
1043 logbuf_unlock_irq();
1044
1045 file->private_data = user;
1046 return 0;
1047}
1048
1049static int devkmsg_release(struct inode *inode, struct file *file)
1050{
1051 struct devkmsg_user *user = file->private_data;
1052
1053 if (!user)
1054 return 0;
1055
1056 ratelimit_state_exit(&user->rs);
1057
1058 mutex_destroy(&user->lock);
1059 kfree(user);
1060 return 0;
1061}
1062
1063const struct file_operations kmsg_fops = {
1064 .open = devkmsg_open,
1065 .read = devkmsg_read,
1066 .write_iter = devkmsg_write,
1067 .llseek = devkmsg_llseek,
1068 .poll = devkmsg_poll,
1069 .release = devkmsg_release,
1070};
1071
1072#ifdef CONFIG_CRASH_CORE
1073/*
1074 * This appends the listed symbols to /proc/vmcore
1075 *
1076 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1077 * obtain access to symbols that are otherwise very difficult to locate. These
1078 * symbols are specifically used so that utilities can access and extract the
1079 * dmesg log from a vmcore file after a crash.
1080 */
1081void log_buf_vmcoreinfo_setup(void)
1082{
1083 VMCOREINFO_SYMBOL(log_buf);
1084 VMCOREINFO_SYMBOL(log_buf_len);
1085 VMCOREINFO_SYMBOL(log_first_idx);
1086 VMCOREINFO_SYMBOL(clear_idx);
1087 VMCOREINFO_SYMBOL(log_next_idx);
1088 /*
1089 * Export struct printk_log size and field offsets. User space tools can
1090 * parse it and detect any changes to structure down the line.
1091 */
1092 VMCOREINFO_STRUCT_SIZE(printk_log);
1093 VMCOREINFO_OFFSET(printk_log, ts_nsec);
1094 VMCOREINFO_OFFSET(printk_log, len);
1095 VMCOREINFO_OFFSET(printk_log, text_len);
1096 VMCOREINFO_OFFSET(printk_log, dict_len);
1097#ifdef CONFIG_PRINTK_CALLER
1098 VMCOREINFO_OFFSET(printk_log, caller_id);
1099#endif
1100}
1101#endif
1102
1103/* requested log_buf_len from kernel cmdline */
1104static unsigned long __initdata new_log_buf_len;
1105
1106/* we practice scaling the ring buffer by powers of 2 */
1107static void __init log_buf_len_update(u64 size)
1108{
1109 if (size > (u64)LOG_BUF_LEN_MAX) {
1110 size = (u64)LOG_BUF_LEN_MAX;
1111 pr_err("log_buf over 2G is not supported.\n");
1112 }
1113
1114 if (size)
1115 size = roundup_pow_of_two(size);
1116 if (size > log_buf_len)
1117 new_log_buf_len = (unsigned long)size;
1118}
1119
1120/* save requested log_buf_len since it's too early to process it */
1121static int __init log_buf_len_setup(char *str)
1122{
1123 u64 size;
1124
1125 if (!str)
1126 return -EINVAL;
1127
1128 size = memparse(str, &str);
1129
1130 log_buf_len_update(size);
1131
1132 return 0;
1133}
1134early_param("log_buf_len", log_buf_len_setup);
1135
1136#ifdef CONFIG_SMP
1137#define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1138
1139static void __init log_buf_add_cpu(void)
1140{
1141 unsigned int cpu_extra;
1142
1143 /*
1144 * archs should set up cpu_possible_bits properly with
1145 * set_cpu_possible() after setup_arch() but just in
1146 * case lets ensure this is valid.
1147 */
1148 if (num_possible_cpus() == 1)
1149 return;
1150
1151 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1152
1153 /* by default this will only continue through for large > 64 CPUs */
1154 if (cpu_extra <= __LOG_BUF_LEN / 2)
1155 return;
1156
1157 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1158 __LOG_CPU_MAX_BUF_LEN);
1159 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1160 cpu_extra);
1161 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1162
1163 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1164}
1165#else /* !CONFIG_SMP */
1166static inline void log_buf_add_cpu(void) {}
1167#endif /* CONFIG_SMP */
1168
1169static void __init set_percpu_data_ready(void)
1170{
1171 printk_safe_init();
1172 /* Make sure we set this flag only after printk_safe() init is done */
1173 barrier();
1174 __printk_percpu_data_ready = true;
1175}
1176
1177void __init setup_log_buf(int early)
1178{
1179 unsigned long flags;
1180 char *new_log_buf;
1181 unsigned int free;
1182
1183 /*
1184 * Some archs call setup_log_buf() multiple times - first is very
1185 * early, e.g. from setup_arch(), and second - when percpu_areas
1186 * are initialised.
1187 */
1188 if (!early)
1189 set_percpu_data_ready();
1190
1191 if (log_buf != __log_buf)
1192 return;
1193
1194 if (!early && !new_log_buf_len)
1195 log_buf_add_cpu();
1196
1197 if (!new_log_buf_len)
1198 return;
1199
1200 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1201 if (unlikely(!new_log_buf)) {
1202 pr_err("log_buf_len: %lu bytes not available\n",
1203 new_log_buf_len);
1204 return;
1205 }
1206
1207 logbuf_lock_irqsave(flags);
1208 log_buf_len = new_log_buf_len;
1209 log_buf = new_log_buf;
1210 new_log_buf_len = 0;
1211 free = __LOG_BUF_LEN - log_next_idx;
1212 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1213 logbuf_unlock_irqrestore(flags);
1214
1215 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1216 pr_info("early log buf free: %u(%u%%)\n",
1217 free, (free * 100) / __LOG_BUF_LEN);
1218}
1219
1220static bool __read_mostly ignore_loglevel;
1221
1222static int __init ignore_loglevel_setup(char *str)
1223{
1224 ignore_loglevel = true;
1225 pr_info("debug: ignoring loglevel setting.\n");
1226
1227 return 0;
1228}
1229
1230early_param("ignore_loglevel", ignore_loglevel_setup);
1231module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1232MODULE_PARM_DESC(ignore_loglevel,
1233 "ignore loglevel setting (prints all kernel messages to the console)");
1234
1235static bool suppress_message_printing(int level)
1236{
1237 return (level >= console_loglevel && !ignore_loglevel);
1238}
1239
1240#ifdef CONFIG_BOOT_PRINTK_DELAY
1241
1242static int boot_delay; /* msecs delay after each printk during bootup */
1243static unsigned long long loops_per_msec; /* based on boot_delay */
1244
1245static int __init boot_delay_setup(char *str)
1246{
1247 unsigned long lpj;
1248
1249 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1250 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1251
1252 get_option(&str, &boot_delay);
1253 if (boot_delay > 10 * 1000)
1254 boot_delay = 0;
1255
1256 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1257 "HZ: %d, loops_per_msec: %llu\n",
1258 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1259 return 0;
1260}
1261early_param("boot_delay", boot_delay_setup);
1262
1263static void boot_delay_msec(int level)
1264{
1265 unsigned long long k;
1266 unsigned long timeout;
1267
1268 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1269 || suppress_message_printing(level)) {
1270 return;
1271 }
1272
1273 k = (unsigned long long)loops_per_msec * boot_delay;
1274
1275 timeout = jiffies + msecs_to_jiffies(boot_delay);
1276 while (k) {
1277 k--;
1278 cpu_relax();
1279 /*
1280 * use (volatile) jiffies to prevent
1281 * compiler reduction; loop termination via jiffies
1282 * is secondary and may or may not happen.
1283 */
1284 if (time_after(jiffies, timeout))
1285 break;
1286 touch_nmi_watchdog();
1287 }
1288}
1289#else
1290static inline void boot_delay_msec(int level)
1291{
1292}
1293#endif
1294
1295static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1296module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1297
1298static size_t print_syslog(unsigned int level, char *buf)
1299{
1300 return sprintf(buf, "<%u>", level);
1301}
1302
1303static size_t print_time(u64 ts, char *buf)
1304{
1305 unsigned long rem_nsec = do_div(ts, 1000000000);
1306
1307 return sprintf(buf, "[%5lu.%06lu]",
1308 (unsigned long)ts, rem_nsec / 1000);
1309}
1310
1311#ifdef CONFIG_PRINTK_CALLER
1312static size_t print_caller(u32 id, char *buf)
1313{
1314 char caller[12];
1315
1316 snprintf(caller, sizeof(caller), "%c%u",
1317 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1318 return sprintf(buf, "[%6s]", caller);
1319}
1320#else
1321#define print_caller(id, buf) 0
1322#endif
1323
1324static size_t print_prefix(const struct printk_log *msg, bool syslog,
1325 bool time, char *buf)
1326{
1327 size_t len = 0;
1328
1329 if (syslog)
1330 len = print_syslog((msg->facility << 3) | msg->level, buf);
1331
1332 if (time)
1333 len += print_time(msg->ts_nsec, buf + len);
1334
1335 len += print_caller(msg->caller_id, buf + len);
1336
1337 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1338 buf[len++] = ' ';
1339 buf[len] = '\0';
1340 }
1341
1342 return len;
1343}
1344
1345static size_t msg_print_text(const struct printk_log *msg, bool syslog,
1346 bool time, char *buf, size_t size)
1347{
1348 const char *text = log_text(msg);
1349 size_t text_size = msg->text_len;
1350 size_t len = 0;
1351 char prefix[PREFIX_MAX];
1352 const size_t prefix_len = print_prefix(msg, syslog, time, prefix);
1353
1354 do {
1355 const char *next = memchr(text, '\n', text_size);
1356 size_t text_len;
1357
1358 if (next) {
1359 text_len = next - text;
1360 next++;
1361 text_size -= next - text;
1362 } else {
1363 text_len = text_size;
1364 }
1365
1366 if (buf) {
1367 if (prefix_len + text_len + 1 >= size - len)
1368 break;
1369
1370 memcpy(buf + len, prefix, prefix_len);
1371 len += prefix_len;
1372 memcpy(buf + len, text, text_len);
1373 len += text_len;
1374 buf[len++] = '\n';
1375 } else {
1376 /* SYSLOG_ACTION_* buffer size only calculation */
1377 len += prefix_len + text_len + 1;
1378 }
1379
1380 text = next;
1381 } while (text);
1382
1383 return len;
1384}
1385
1386static int syslog_print(char __user *buf, int size)
1387{
1388 char *text;
1389 struct printk_log *msg;
1390 int len = 0;
1391
1392 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1393 if (!text)
1394 return -ENOMEM;
1395
1396 while (size > 0) {
1397 size_t n;
1398 size_t skip;
1399
1400 logbuf_lock_irq();
1401 if (syslog_seq < log_first_seq) {
1402 /* messages are gone, move to first one */
1403 syslog_seq = log_first_seq;
1404 syslog_idx = log_first_idx;
1405 syslog_partial = 0;
1406 }
1407 if (syslog_seq == log_next_seq) {
1408 logbuf_unlock_irq();
1409 break;
1410 }
1411
1412 /*
1413 * To keep reading/counting partial line consistent,
1414 * use printk_time value as of the beginning of a line.
1415 */
1416 if (!syslog_partial)
1417 syslog_time = printk_time;
1418
1419 skip = syslog_partial;
1420 msg = log_from_idx(syslog_idx);
1421 n = msg_print_text(msg, true, syslog_time, text,
1422 LOG_LINE_MAX + PREFIX_MAX);
1423 if (n - syslog_partial <= size) {
1424 /* message fits into buffer, move forward */
1425 syslog_idx = log_next(syslog_idx);
1426 syslog_seq++;
1427 n -= syslog_partial;
1428 syslog_partial = 0;
1429 } else if (!len){
1430 /* partial read(), remember position */
1431 n = size;
1432 syslog_partial += n;
1433 } else
1434 n = 0;
1435 logbuf_unlock_irq();
1436
1437 if (!n)
1438 break;
1439
1440 if (copy_to_user(buf, text + skip, n)) {
1441 if (!len)
1442 len = -EFAULT;
1443 break;
1444 }
1445
1446 len += n;
1447 size -= n;
1448 buf += n;
1449 }
1450
1451 kfree(text);
1452 return len;
1453}
1454
1455static int syslog_print_all(char __user *buf, int size, bool clear)
1456{
1457 char *text;
1458 int len = 0;
1459 u64 next_seq;
1460 u64 seq;
1461 u32 idx;
1462 bool time;
1463
1464 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1465 if (!text)
1466 return -ENOMEM;
1467
1468 time = printk_time;
1469 logbuf_lock_irq();
1470 /*
1471 * Find first record that fits, including all following records,
1472 * into the user-provided buffer for this dump.
1473 */
1474 seq = clear_seq;
1475 idx = clear_idx;
1476 while (seq < log_next_seq) {
1477 struct printk_log *msg = log_from_idx(idx);
1478
1479 len += msg_print_text(msg, true, time, NULL, 0);
1480 idx = log_next(idx);
1481 seq++;
1482 }
1483
1484 /* move first record forward until length fits into the buffer */
1485 seq = clear_seq;
1486 idx = clear_idx;
1487 while (len > size && seq < log_next_seq) {
1488 struct printk_log *msg = log_from_idx(idx);
1489
1490 len -= msg_print_text(msg, true, time, NULL, 0);
1491 idx = log_next(idx);
1492 seq++;
1493 }
1494
1495 /* last message fitting into this dump */
1496 next_seq = log_next_seq;
1497
1498 len = 0;
1499 while (len >= 0 && seq < next_seq) {
1500 struct printk_log *msg = log_from_idx(idx);
1501 int textlen = msg_print_text(msg, true, time, text,
1502 LOG_LINE_MAX + PREFIX_MAX);
1503
1504 idx = log_next(idx);
1505 seq++;
1506
1507 logbuf_unlock_irq();
1508 if (copy_to_user(buf + len, text, textlen))
1509 len = -EFAULT;
1510 else
1511 len += textlen;
1512 logbuf_lock_irq();
1513
1514 if (seq < log_first_seq) {
1515 /* messages are gone, move to next one */
1516 seq = log_first_seq;
1517 idx = log_first_idx;
1518 }
1519 }
1520
1521 if (clear) {
1522 clear_seq = log_next_seq;
1523 clear_idx = log_next_idx;
1524 }
1525 logbuf_unlock_irq();
1526
1527 kfree(text);
1528 return len;
1529}
1530
1531static void syslog_clear(void)
1532{
1533 logbuf_lock_irq();
1534 clear_seq = log_next_seq;
1535 clear_idx = log_next_idx;
1536 logbuf_unlock_irq();
1537}
1538
1539int do_syslog(int type, char __user *buf, int len, int source)
1540{
1541 bool clear = false;
1542 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1543 int error;
1544
1545 error = check_syslog_permissions(type, source);
1546 if (error)
1547 return error;
1548
1549 switch (type) {
1550 case SYSLOG_ACTION_CLOSE: /* Close log */
1551 break;
1552 case SYSLOG_ACTION_OPEN: /* Open log */
1553 break;
1554 case SYSLOG_ACTION_READ: /* Read from log */
1555 if (!buf || len < 0)
1556 return -EINVAL;
1557 if (!len)
1558 return 0;
1559 if (!access_ok(buf, len))
1560 return -EFAULT;
1561 error = wait_event_interruptible(log_wait,
1562 syslog_seq != log_next_seq);
1563 if (error)
1564 return error;
1565 error = syslog_print(buf, len);
1566 break;
1567 /* Read/clear last kernel messages */
1568 case SYSLOG_ACTION_READ_CLEAR:
1569 clear = true;
1570 /* FALL THRU */
1571 /* Read last kernel messages */
1572 case SYSLOG_ACTION_READ_ALL:
1573 if (!buf || len < 0)
1574 return -EINVAL;
1575 if (!len)
1576 return 0;
1577 if (!access_ok(buf, len))
1578 return -EFAULT;
1579 error = syslog_print_all(buf, len, clear);
1580 break;
1581 /* Clear ring buffer */
1582 case SYSLOG_ACTION_CLEAR:
1583 syslog_clear();
1584 break;
1585 /* Disable logging to console */
1586 case SYSLOG_ACTION_CONSOLE_OFF:
1587 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1588 saved_console_loglevel = console_loglevel;
1589 console_loglevel = minimum_console_loglevel;
1590 break;
1591 /* Enable logging to console */
1592 case SYSLOG_ACTION_CONSOLE_ON:
1593 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1594 console_loglevel = saved_console_loglevel;
1595 saved_console_loglevel = LOGLEVEL_DEFAULT;
1596 }
1597 break;
1598 /* Set level of messages printed to console */
1599 case SYSLOG_ACTION_CONSOLE_LEVEL:
1600 if (len < 1 || len > 8)
1601 return -EINVAL;
1602 if (len < minimum_console_loglevel)
1603 len = minimum_console_loglevel;
1604 console_loglevel = len;
1605 /* Implicitly re-enable logging to console */
1606 saved_console_loglevel = LOGLEVEL_DEFAULT;
1607 break;
1608 /* Number of chars in the log buffer */
1609 case SYSLOG_ACTION_SIZE_UNREAD:
1610 logbuf_lock_irq();
1611 if (syslog_seq < log_first_seq) {
1612 /* messages are gone, move to first one */
1613 syslog_seq = log_first_seq;
1614 syslog_idx = log_first_idx;
1615 syslog_partial = 0;
1616 }
1617 if (source == SYSLOG_FROM_PROC) {
1618 /*
1619 * Short-cut for poll(/"proc/kmsg") which simply checks
1620 * for pending data, not the size; return the count of
1621 * records, not the length.
1622 */
1623 error = log_next_seq - syslog_seq;
1624 } else {
1625 u64 seq = syslog_seq;
1626 u32 idx = syslog_idx;
1627 bool time = syslog_partial ? syslog_time : printk_time;
1628
1629 while (seq < log_next_seq) {
1630 struct printk_log *msg = log_from_idx(idx);
1631
1632 error += msg_print_text(msg, true, time, NULL,
1633 0);
1634 time = printk_time;
1635 idx = log_next(idx);
1636 seq++;
1637 }
1638 error -= syslog_partial;
1639 }
1640 logbuf_unlock_irq();
1641 break;
1642 /* Size of the log buffer */
1643 case SYSLOG_ACTION_SIZE_BUFFER:
1644 error = log_buf_len;
1645 break;
1646 default:
1647 error = -EINVAL;
1648 break;
1649 }
1650
1651 return error;
1652}
1653
1654SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1655{
1656 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1657}
1658
1659/*
1660 * Special console_lock variants that help to reduce the risk of soft-lockups.
1661 * They allow to pass console_lock to another printk() call using a busy wait.
1662 */
1663
1664#ifdef CONFIG_LOCKDEP
1665static struct lockdep_map console_owner_dep_map = {
1666 .name = "console_owner"
1667};
1668#endif
1669
1670static DEFINE_RAW_SPINLOCK(console_owner_lock);
1671static struct task_struct *console_owner;
1672static bool console_waiter;
1673
1674/**
1675 * console_lock_spinning_enable - mark beginning of code where another
1676 * thread might safely busy wait
1677 *
1678 * This basically converts console_lock into a spinlock. This marks
1679 * the section where the console_lock owner can not sleep, because
1680 * there may be a waiter spinning (like a spinlock). Also it must be
1681 * ready to hand over the lock at the end of the section.
1682 */
1683static void console_lock_spinning_enable(void)
1684{
1685 raw_spin_lock(&console_owner_lock);
1686 console_owner = current;
1687 raw_spin_unlock(&console_owner_lock);
1688
1689 /* The waiter may spin on us after setting console_owner */
1690 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1691}
1692
1693/**
1694 * console_lock_spinning_disable_and_check - mark end of code where another
1695 * thread was able to busy wait and check if there is a waiter
1696 *
1697 * This is called at the end of the section where spinning is allowed.
1698 * It has two functions. First, it is a signal that it is no longer
1699 * safe to start busy waiting for the lock. Second, it checks if
1700 * there is a busy waiter and passes the lock rights to her.
1701 *
1702 * Important: Callers lose the lock if there was a busy waiter.
1703 * They must not touch items synchronized by console_lock
1704 * in this case.
1705 *
1706 * Return: 1 if the lock rights were passed, 0 otherwise.
1707 */
1708static int console_lock_spinning_disable_and_check(void)
1709{
1710 int waiter;
1711
1712 raw_spin_lock(&console_owner_lock);
1713 waiter = READ_ONCE(console_waiter);
1714 console_owner = NULL;
1715 raw_spin_unlock(&console_owner_lock);
1716
1717 if (!waiter) {
1718 spin_release(&console_owner_dep_map, _THIS_IP_);
1719 return 0;
1720 }
1721
1722 /* The waiter is now free to continue */
1723 WRITE_ONCE(console_waiter, false);
1724
1725 spin_release(&console_owner_dep_map, _THIS_IP_);
1726
1727 /*
1728 * Hand off console_lock to waiter. The waiter will perform
1729 * the up(). After this, the waiter is the console_lock owner.
1730 */
1731 mutex_release(&console_lock_dep_map, _THIS_IP_);
1732 return 1;
1733}
1734
1735/**
1736 * console_trylock_spinning - try to get console_lock by busy waiting
1737 *
1738 * This allows to busy wait for the console_lock when the current
1739 * owner is running in specially marked sections. It means that
1740 * the current owner is running and cannot reschedule until it
1741 * is ready to lose the lock.
1742 *
1743 * Return: 1 if we got the lock, 0 othrewise
1744 */
1745static int console_trylock_spinning(void)
1746{
1747 struct task_struct *owner = NULL;
1748 bool waiter;
1749 bool spin = false;
1750 unsigned long flags;
1751
1752 if (console_trylock())
1753 return 1;
1754
1755 printk_safe_enter_irqsave(flags);
1756
1757 raw_spin_lock(&console_owner_lock);
1758 owner = READ_ONCE(console_owner);
1759 waiter = READ_ONCE(console_waiter);
1760 if (!waiter && owner && owner != current) {
1761 WRITE_ONCE(console_waiter, true);
1762 spin = true;
1763 }
1764 raw_spin_unlock(&console_owner_lock);
1765
1766 /*
1767 * If there is an active printk() writing to the
1768 * consoles, instead of having it write our data too,
1769 * see if we can offload that load from the active
1770 * printer, and do some printing ourselves.
1771 * Go into a spin only if there isn't already a waiter
1772 * spinning, and there is an active printer, and
1773 * that active printer isn't us (recursive printk?).
1774 */
1775 if (!spin) {
1776 printk_safe_exit_irqrestore(flags);
1777 return 0;
1778 }
1779
1780 /* We spin waiting for the owner to release us */
1781 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1782 /* Owner will clear console_waiter on hand off */
1783 while (READ_ONCE(console_waiter))
1784 cpu_relax();
1785 spin_release(&console_owner_dep_map, _THIS_IP_);
1786
1787 printk_safe_exit_irqrestore(flags);
1788 /*
1789 * The owner passed the console lock to us.
1790 * Since we did not spin on console lock, annotate
1791 * this as a trylock. Otherwise lockdep will
1792 * complain.
1793 */
1794 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1795
1796 return 1;
1797}
1798
1799/*
1800 * Call the console drivers, asking them to write out
1801 * log_buf[start] to log_buf[end - 1].
1802 * The console_lock must be held.
1803 */
1804static void call_console_drivers(const char *ext_text, size_t ext_len,
1805 const char *text, size_t len)
1806{
1807 struct console *con;
1808
1809 trace_console_rcuidle(text, len);
1810
1811 for_each_console(con) {
1812 if (exclusive_console && con != exclusive_console)
1813 continue;
1814 if (!(con->flags & CON_ENABLED))
1815 continue;
1816 if (!con->write)
1817 continue;
1818 if (!cpu_online(smp_processor_id()) &&
1819 !(con->flags & CON_ANYTIME))
1820 continue;
1821 if (con->flags & CON_EXTENDED)
1822 con->write(con, ext_text, ext_len);
1823 else
1824 con->write(con, text, len);
1825 }
1826}
1827
1828int printk_delay_msec __read_mostly;
1829
1830static inline void printk_delay(void)
1831{
1832 if (unlikely(printk_delay_msec)) {
1833 int m = printk_delay_msec;
1834
1835 while (m--) {
1836 mdelay(1);
1837 touch_nmi_watchdog();
1838 }
1839 }
1840}
1841
1842static inline u32 printk_caller_id(void)
1843{
1844 return in_task() ? task_pid_nr(current) :
1845 0x80000000 + raw_smp_processor_id();
1846}
1847
1848/*
1849 * Continuation lines are buffered, and not committed to the record buffer
1850 * until the line is complete, or a race forces it. The line fragments
1851 * though, are printed immediately to the consoles to ensure everything has
1852 * reached the console in case of a kernel crash.
1853 */
1854static struct cont {
1855 char buf[LOG_LINE_MAX];
1856 size_t len; /* length == 0 means unused buffer */
1857 u32 caller_id; /* printk_caller_id() of first print */
1858 u64 ts_nsec; /* time of first print */
1859 u8 level; /* log level of first message */
1860 u8 facility; /* log facility of first message */
1861 enum log_flags flags; /* prefix, newline flags */
1862} cont;
1863
1864static void cont_flush(void)
1865{
1866 if (cont.len == 0)
1867 return;
1868
1869 log_store(cont.caller_id, cont.facility, cont.level, cont.flags,
1870 cont.ts_nsec, NULL, 0, cont.buf, cont.len);
1871 cont.len = 0;
1872}
1873
1874static bool cont_add(u32 caller_id, int facility, int level,
1875 enum log_flags flags, const char *text, size_t len)
1876{
1877 /* If the line gets too long, split it up in separate records. */
1878 if (cont.len + len > sizeof(cont.buf)) {
1879 cont_flush();
1880 return false;
1881 }
1882
1883 if (!cont.len) {
1884 cont.facility = facility;
1885 cont.level = level;
1886 cont.caller_id = caller_id;
1887 cont.ts_nsec = local_clock();
1888 cont.flags = flags;
1889 }
1890
1891 memcpy(cont.buf + cont.len, text, len);
1892 cont.len += len;
1893
1894 // The original flags come from the first line,
1895 // but later continuations can add a newline.
1896 if (flags & LOG_NEWLINE) {
1897 cont.flags |= LOG_NEWLINE;
1898 cont_flush();
1899 }
1900
1901 return true;
1902}
1903
1904static size_t log_output(int facility, int level, enum log_flags lflags, const char *dict, size_t dictlen, char *text, size_t text_len)
1905{
1906 const u32 caller_id = printk_caller_id();
1907
1908 /*
1909 * If an earlier line was buffered, and we're a continuation
1910 * write from the same context, try to add it to the buffer.
1911 */
1912 if (cont.len) {
1913 if (cont.caller_id == caller_id && (lflags & LOG_CONT)) {
1914 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1915 return text_len;
1916 }
1917 /* Otherwise, make sure it's flushed */
1918 cont_flush();
1919 }
1920
1921 /* Skip empty continuation lines that couldn't be added - they just flush */
1922 if (!text_len && (lflags & LOG_CONT))
1923 return 0;
1924
1925 /* If it doesn't end in a newline, try to buffer the current line */
1926 if (!(lflags & LOG_NEWLINE)) {
1927 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1928 return text_len;
1929 }
1930
1931 /* Store it in the record log */
1932 return log_store(caller_id, facility, level, lflags, 0,
1933 dict, dictlen, text, text_len);
1934}
1935
1936/* Must be called under logbuf_lock. */
1937int vprintk_store(int facility, int level,
1938 const char *dict, size_t dictlen,
1939 const char *fmt, va_list args)
1940{
1941 static char textbuf[LOG_LINE_MAX];
1942 char *text = textbuf;
1943 size_t text_len;
1944 enum log_flags lflags = 0;
1945
1946 /*
1947 * The printf needs to come first; we need the syslog
1948 * prefix which might be passed-in as a parameter.
1949 */
1950 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1951
1952 /* mark and strip a trailing newline */
1953 if (text_len && text[text_len-1] == '\n') {
1954 text_len--;
1955 lflags |= LOG_NEWLINE;
1956 }
1957
1958 /* strip kernel syslog prefix and extract log level or control flags */
1959 if (facility == 0) {
1960 int kern_level;
1961
1962 while ((kern_level = printk_get_level(text)) != 0) {
1963 switch (kern_level) {
1964 case '0' ... '7':
1965 if (level == LOGLEVEL_DEFAULT)
1966 level = kern_level - '0';
1967 break;
1968 case 'c': /* KERN_CONT */
1969 lflags |= LOG_CONT;
1970 }
1971
1972 text_len -= 2;
1973 text += 2;
1974 }
1975 }
1976
1977 if (level == LOGLEVEL_DEFAULT)
1978 level = default_message_loglevel;
1979
1980 if (dict)
1981 lflags |= LOG_NEWLINE;
1982
1983 return log_output(facility, level, lflags,
1984 dict, dictlen, text, text_len);
1985}
1986
1987asmlinkage int vprintk_emit(int facility, int level,
1988 const char *dict, size_t dictlen,
1989 const char *fmt, va_list args)
1990{
1991 int printed_len;
1992 bool in_sched = false, pending_output;
1993 unsigned long flags;
1994 u64 curr_log_seq;
1995
1996 /* Suppress unimportant messages after panic happens */
1997 if (unlikely(suppress_printk))
1998 return 0;
1999
2000 if (level == LOGLEVEL_SCHED) {
2001 level = LOGLEVEL_DEFAULT;
2002 in_sched = true;
2003 }
2004
2005 boot_delay_msec(level);
2006 printk_delay();
2007
2008 /* This stops the holder of console_sem just where we want him */
2009 logbuf_lock_irqsave(flags);
2010 curr_log_seq = log_next_seq;
2011 printed_len = vprintk_store(facility, level, dict, dictlen, fmt, args);
2012 pending_output = (curr_log_seq != log_next_seq);
2013 logbuf_unlock_irqrestore(flags);
2014
2015 /* If called from the scheduler, we can not call up(). */
2016 if (!in_sched && pending_output) {
2017 /*
2018 * Disable preemption to avoid being preempted while holding
2019 * console_sem which would prevent anyone from printing to
2020 * console
2021 */
2022 preempt_disable();
2023 /*
2024 * Try to acquire and then immediately release the console
2025 * semaphore. The release will print out buffers and wake up
2026 * /dev/kmsg and syslog() users.
2027 */
2028 if (console_trylock_spinning())
2029 console_unlock();
2030 preempt_enable();
2031 }
2032
2033 if (pending_output)
2034 wake_up_klogd();
2035 return printed_len;
2036}
2037EXPORT_SYMBOL(vprintk_emit);
2038
2039asmlinkage int vprintk(const char *fmt, va_list args)
2040{
2041 return vprintk_func(fmt, args);
2042}
2043EXPORT_SYMBOL(vprintk);
2044
2045int vprintk_default(const char *fmt, va_list args)
2046{
2047 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
2048}
2049EXPORT_SYMBOL_GPL(vprintk_default);
2050
2051/**
2052 * printk - print a kernel message
2053 * @fmt: format string
2054 *
2055 * This is printk(). It can be called from any context. We want it to work.
2056 *
2057 * We try to grab the console_lock. If we succeed, it's easy - we log the
2058 * output and call the console drivers. If we fail to get the semaphore, we
2059 * place the output into the log buffer and return. The current holder of
2060 * the console_sem will notice the new output in console_unlock(); and will
2061 * send it to the consoles before releasing the lock.
2062 *
2063 * One effect of this deferred printing is that code which calls printk() and
2064 * then changes console_loglevel may break. This is because console_loglevel
2065 * is inspected when the actual printing occurs.
2066 *
2067 * See also:
2068 * printf(3)
2069 *
2070 * See the vsnprintf() documentation for format string extensions over C99.
2071 */
2072asmlinkage __visible int printk(const char *fmt, ...)
2073{
2074 va_list args;
2075 int r;
2076
2077 va_start(args, fmt);
2078 r = vprintk_func(fmt, args);
2079 va_end(args);
2080
2081 return r;
2082}
2083EXPORT_SYMBOL(printk);
2084
2085#else /* CONFIG_PRINTK */
2086
2087#define LOG_LINE_MAX 0
2088#define PREFIX_MAX 0
2089#define printk_time false
2090
2091static u64 syslog_seq;
2092static u32 syslog_idx;
2093static u64 console_seq;
2094static u32 console_idx;
2095static u64 exclusive_console_stop_seq;
2096static u64 log_first_seq;
2097static u32 log_first_idx;
2098static u64 log_next_seq;
2099static char *log_text(const struct printk_log *msg) { return NULL; }
2100static char *log_dict(const struct printk_log *msg) { return NULL; }
2101static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2102static u32 log_next(u32 idx) { return 0; }
2103static ssize_t msg_print_ext_header(char *buf, size_t size,
2104 struct printk_log *msg,
2105 u64 seq) { return 0; }
2106static ssize_t msg_print_ext_body(char *buf, size_t size,
2107 char *dict, size_t dict_len,
2108 char *text, size_t text_len) { return 0; }
2109static void console_lock_spinning_enable(void) { }
2110static int console_lock_spinning_disable_and_check(void) { return 0; }
2111static void call_console_drivers(const char *ext_text, size_t ext_len,
2112 const char *text, size_t len) {}
2113static size_t msg_print_text(const struct printk_log *msg, bool syslog,
2114 bool time, char *buf, size_t size) { return 0; }
2115static bool suppress_message_printing(int level) { return false; }
2116
2117#endif /* CONFIG_PRINTK */
2118
2119#ifdef CONFIG_EARLY_PRINTK
2120struct console *early_console;
2121
2122asmlinkage __visible void early_printk(const char *fmt, ...)
2123{
2124 va_list ap;
2125 char buf[512];
2126 int n;
2127
2128 if (!early_console)
2129 return;
2130
2131 va_start(ap, fmt);
2132 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2133 va_end(ap);
2134
2135 early_console->write(early_console, buf, n);
2136}
2137#endif
2138
2139static int __add_preferred_console(char *name, int idx, char *options,
2140 char *brl_options, bool user_specified)
2141{
2142 struct console_cmdline *c;
2143 int i;
2144
2145 /*
2146 * See if this tty is not yet registered, and
2147 * if we have a slot free.
2148 */
2149 for (i = 0, c = console_cmdline;
2150 i < MAX_CMDLINECONSOLES && c->name[0];
2151 i++, c++) {
2152 if (strcmp(c->name, name) == 0 && c->index == idx) {
2153 if (!brl_options)
2154 preferred_console = i;
2155 if (user_specified)
2156 c->user_specified = true;
2157 return 0;
2158 }
2159 }
2160 if (i == MAX_CMDLINECONSOLES)
2161 return -E2BIG;
2162 if (!brl_options)
2163 preferred_console = i;
2164 strlcpy(c->name, name, sizeof(c->name));
2165 c->options = options;
2166 c->user_specified = user_specified;
2167 braille_set_options(c, brl_options);
2168
2169 c->index = idx;
2170 return 0;
2171}
2172
2173static int __init console_msg_format_setup(char *str)
2174{
2175 if (!strcmp(str, "syslog"))
2176 console_msg_format = MSG_FORMAT_SYSLOG;
2177 if (!strcmp(str, "default"))
2178 console_msg_format = MSG_FORMAT_DEFAULT;
2179 return 1;
2180}
2181__setup("console_msg_format=", console_msg_format_setup);
2182
2183/*
2184 * Set up a console. Called via do_early_param() in init/main.c
2185 * for each "console=" parameter in the boot command line.
2186 */
2187static int __init console_setup(char *str)
2188{
2189 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2190 char *s, *options, *brl_options = NULL;
2191 int idx;
2192
2193 if (str[0] == 0)
2194 return 1;
2195
2196 if (_braille_console_setup(&str, &brl_options))
2197 return 1;
2198
2199 /*
2200 * Decode str into name, index, options.
2201 */
2202 if (str[0] >= '0' && str[0] <= '9') {
2203 strcpy(buf, "ttyS");
2204 strncpy(buf + 4, str, sizeof(buf) - 5);
2205 } else {
2206 strncpy(buf, str, sizeof(buf) - 1);
2207 }
2208 buf[sizeof(buf) - 1] = 0;
2209 options = strchr(str, ',');
2210 if (options)
2211 *(options++) = 0;
2212#ifdef __sparc__
2213 if (!strcmp(str, "ttya"))
2214 strcpy(buf, "ttyS0");
2215 if (!strcmp(str, "ttyb"))
2216 strcpy(buf, "ttyS1");
2217#endif
2218 for (s = buf; *s; s++)
2219 if (isdigit(*s) || *s == ',')
2220 break;
2221 idx = simple_strtoul(s, NULL, 10);
2222 *s = 0;
2223
2224 __add_preferred_console(buf, idx, options, brl_options, true);
2225 console_set_on_cmdline = 1;
2226 return 1;
2227}
2228__setup("console=", console_setup);
2229
2230/**
2231 * add_preferred_console - add a device to the list of preferred consoles.
2232 * @name: device name
2233 * @idx: device index
2234 * @options: options for this console
2235 *
2236 * The last preferred console added will be used for kernel messages
2237 * and stdin/out/err for init. Normally this is used by console_setup
2238 * above to handle user-supplied console arguments; however it can also
2239 * be used by arch-specific code either to override the user or more
2240 * commonly to provide a default console (ie from PROM variables) when
2241 * the user has not supplied one.
2242 */
2243int add_preferred_console(char *name, int idx, char *options)
2244{
2245 return __add_preferred_console(name, idx, options, NULL, false);
2246}
2247
2248bool console_suspend_enabled = true;
2249EXPORT_SYMBOL(console_suspend_enabled);
2250
2251static int __init console_suspend_disable(char *str)
2252{
2253 console_suspend_enabled = false;
2254 return 1;
2255}
2256__setup("no_console_suspend", console_suspend_disable);
2257module_param_named(console_suspend, console_suspend_enabled,
2258 bool, S_IRUGO | S_IWUSR);
2259MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2260 " and hibernate operations");
2261
2262/**
2263 * suspend_console - suspend the console subsystem
2264 *
2265 * This disables printk() while we go into suspend states
2266 */
2267void suspend_console(void)
2268{
2269 if (!console_suspend_enabled)
2270 return;
2271 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2272 console_lock();
2273 console_suspended = 1;
2274 up_console_sem();
2275}
2276
2277void resume_console(void)
2278{
2279 if (!console_suspend_enabled)
2280 return;
2281 down_console_sem();
2282 console_suspended = 0;
2283 console_unlock();
2284}
2285
2286/**
2287 * console_cpu_notify - print deferred console messages after CPU hotplug
2288 * @cpu: unused
2289 *
2290 * If printk() is called from a CPU that is not online yet, the messages
2291 * will be printed on the console only if there are CON_ANYTIME consoles.
2292 * This function is called when a new CPU comes online (or fails to come
2293 * up) or goes offline.
2294 */
2295static int console_cpu_notify(unsigned int cpu)
2296{
2297 if (!cpuhp_tasks_frozen) {
2298 /* If trylock fails, someone else is doing the printing */
2299 if (console_trylock())
2300 console_unlock();
2301 }
2302 return 0;
2303}
2304
2305/**
2306 * console_lock - lock the console system for exclusive use.
2307 *
2308 * Acquires a lock which guarantees that the caller has
2309 * exclusive access to the console system and the console_drivers list.
2310 *
2311 * Can sleep, returns nothing.
2312 */
2313void console_lock(void)
2314{
2315 might_sleep();
2316
2317 down_console_sem();
2318 if (console_suspended)
2319 return;
2320 console_locked = 1;
2321 console_may_schedule = 1;
2322}
2323EXPORT_SYMBOL(console_lock);
2324
2325/**
2326 * console_trylock - try to lock the console system for exclusive use.
2327 *
2328 * Try to acquire a lock which guarantees that the caller has exclusive
2329 * access to the console system and the console_drivers list.
2330 *
2331 * returns 1 on success, and 0 on failure to acquire the lock.
2332 */
2333int console_trylock(void)
2334{
2335 if (down_trylock_console_sem())
2336 return 0;
2337 if (console_suspended) {
2338 up_console_sem();
2339 return 0;
2340 }
2341 console_locked = 1;
2342 console_may_schedule = 0;
2343 return 1;
2344}
2345EXPORT_SYMBOL(console_trylock);
2346
2347int is_console_locked(void)
2348{
2349 return console_locked;
2350}
2351EXPORT_SYMBOL(is_console_locked);
2352
2353/*
2354 * Check if we have any console that is capable of printing while cpu is
2355 * booting or shutting down. Requires console_sem.
2356 */
2357static int have_callable_console(void)
2358{
2359 struct console *con;
2360
2361 for_each_console(con)
2362 if ((con->flags & CON_ENABLED) &&
2363 (con->flags & CON_ANYTIME))
2364 return 1;
2365
2366 return 0;
2367}
2368
2369/*
2370 * Can we actually use the console at this time on this cpu?
2371 *
2372 * Console drivers may assume that per-cpu resources have been allocated. So
2373 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2374 * call them until this CPU is officially up.
2375 */
2376static inline int can_use_console(void)
2377{
2378 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2379}
2380
2381/**
2382 * console_unlock - unlock the console system
2383 *
2384 * Releases the console_lock which the caller holds on the console system
2385 * and the console driver list.
2386 *
2387 * While the console_lock was held, console output may have been buffered
2388 * by printk(). If this is the case, console_unlock(); emits
2389 * the output prior to releasing the lock.
2390 *
2391 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2392 *
2393 * console_unlock(); may be called from any context.
2394 */
2395void console_unlock(void)
2396{
2397 static char ext_text[CONSOLE_EXT_LOG_MAX];
2398 static char text[LOG_LINE_MAX + PREFIX_MAX];
2399 unsigned long flags;
2400 bool do_cond_resched, retry;
2401
2402 if (console_suspended) {
2403 up_console_sem();
2404 return;
2405 }
2406
2407 /*
2408 * Console drivers are called with interrupts disabled, so
2409 * @console_may_schedule should be cleared before; however, we may
2410 * end up dumping a lot of lines, for example, if called from
2411 * console registration path, and should invoke cond_resched()
2412 * between lines if allowable. Not doing so can cause a very long
2413 * scheduling stall on a slow console leading to RCU stall and
2414 * softlockup warnings which exacerbate the issue with more
2415 * messages practically incapacitating the system.
2416 *
2417 * console_trylock() is not able to detect the preemptive
2418 * context reliably. Therefore the value must be stored before
2419 * and cleared after the the "again" goto label.
2420 */
2421 do_cond_resched = console_may_schedule;
2422again:
2423 console_may_schedule = 0;
2424
2425 /*
2426 * We released the console_sem lock, so we need to recheck if
2427 * cpu is online and (if not) is there at least one CON_ANYTIME
2428 * console.
2429 */
2430 if (!can_use_console()) {
2431 console_locked = 0;
2432 up_console_sem();
2433 return;
2434 }
2435
2436 for (;;) {
2437 struct printk_log *msg;
2438 size_t ext_len = 0;
2439 size_t len;
2440
2441 printk_safe_enter_irqsave(flags);
2442 raw_spin_lock(&logbuf_lock);
2443 if (console_seq < log_first_seq) {
2444 len = snprintf(text, sizeof(text),
2445 "** %llu printk messages dropped **\n",
2446 log_first_seq - console_seq);
2447
2448 /* messages are gone, move to first one */
2449 console_seq = log_first_seq;
2450 console_idx = log_first_idx;
2451 } else {
2452 len = 0;
2453 }
2454skip:
2455 if (console_seq == log_next_seq)
2456 break;
2457
2458 msg = log_from_idx(console_idx);
2459 if (suppress_message_printing(msg->level)) {
2460 /*
2461 * Skip record we have buffered and already printed
2462 * directly to the console when we received it, and
2463 * record that has level above the console loglevel.
2464 */
2465 console_idx = log_next(console_idx);
2466 console_seq++;
2467 goto skip;
2468 }
2469
2470 /* Output to all consoles once old messages replayed. */
2471 if (unlikely(exclusive_console &&
2472 console_seq >= exclusive_console_stop_seq)) {
2473 exclusive_console = NULL;
2474 }
2475
2476 len += msg_print_text(msg,
2477 console_msg_format & MSG_FORMAT_SYSLOG,
2478 printk_time, text + len, sizeof(text) - len);
2479 if (nr_ext_console_drivers) {
2480 ext_len = msg_print_ext_header(ext_text,
2481 sizeof(ext_text),
2482 msg, console_seq);
2483 ext_len += msg_print_ext_body(ext_text + ext_len,
2484 sizeof(ext_text) - ext_len,
2485 log_dict(msg), msg->dict_len,
2486 log_text(msg), msg->text_len);
2487 }
2488 console_idx = log_next(console_idx);
2489 console_seq++;
2490 raw_spin_unlock(&logbuf_lock);
2491
2492 /*
2493 * While actively printing out messages, if another printk()
2494 * were to occur on another CPU, it may wait for this one to
2495 * finish. This task can not be preempted if there is a
2496 * waiter waiting to take over.
2497 */
2498 console_lock_spinning_enable();
2499
2500 stop_critical_timings(); /* don't trace print latency */
2501 call_console_drivers(ext_text, ext_len, text, len);
2502 start_critical_timings();
2503
2504 if (console_lock_spinning_disable_and_check()) {
2505 printk_safe_exit_irqrestore(flags);
2506 return;
2507 }
2508
2509 printk_safe_exit_irqrestore(flags);
2510
2511 if (do_cond_resched)
2512 cond_resched();
2513 }
2514
2515 console_locked = 0;
2516
2517 raw_spin_unlock(&logbuf_lock);
2518
2519 up_console_sem();
2520
2521 /*
2522 * Someone could have filled up the buffer again, so re-check if there's
2523 * something to flush. In case we cannot trylock the console_sem again,
2524 * there's a new owner and the console_unlock() from them will do the
2525 * flush, no worries.
2526 */
2527 raw_spin_lock(&logbuf_lock);
2528 retry = console_seq != log_next_seq;
2529 raw_spin_unlock(&logbuf_lock);
2530 printk_safe_exit_irqrestore(flags);
2531
2532 if (retry && console_trylock())
2533 goto again;
2534}
2535EXPORT_SYMBOL(console_unlock);
2536
2537/**
2538 * console_conditional_schedule - yield the CPU if required
2539 *
2540 * If the console code is currently allowed to sleep, and
2541 * if this CPU should yield the CPU to another task, do
2542 * so here.
2543 *
2544 * Must be called within console_lock();.
2545 */
2546void __sched console_conditional_schedule(void)
2547{
2548 if (console_may_schedule)
2549 cond_resched();
2550}
2551EXPORT_SYMBOL(console_conditional_schedule);
2552
2553void console_unblank(void)
2554{
2555 struct console *c;
2556
2557 /*
2558 * console_unblank can no longer be called in interrupt context unless
2559 * oops_in_progress is set to 1..
2560 */
2561 if (oops_in_progress) {
2562 if (down_trylock_console_sem() != 0)
2563 return;
2564 } else
2565 console_lock();
2566
2567 console_locked = 1;
2568 console_may_schedule = 0;
2569 for_each_console(c)
2570 if ((c->flags & CON_ENABLED) && c->unblank)
2571 c->unblank();
2572 console_unlock();
2573}
2574
2575/**
2576 * console_flush_on_panic - flush console content on panic
2577 * @mode: flush all messages in buffer or just the pending ones
2578 *
2579 * Immediately output all pending messages no matter what.
2580 */
2581void console_flush_on_panic(enum con_flush_mode mode)
2582{
2583 /*
2584 * If someone else is holding the console lock, trylock will fail
2585 * and may_schedule may be set. Ignore and proceed to unlock so
2586 * that messages are flushed out. As this can be called from any
2587 * context and we don't want to get preempted while flushing,
2588 * ensure may_schedule is cleared.
2589 */
2590 console_trylock();
2591 console_may_schedule = 0;
2592
2593 if (mode == CONSOLE_REPLAY_ALL) {
2594 unsigned long flags;
2595
2596 logbuf_lock_irqsave(flags);
2597 console_seq = log_first_seq;
2598 console_idx = log_first_idx;
2599 logbuf_unlock_irqrestore(flags);
2600 }
2601 console_unlock();
2602}
2603
2604/*
2605 * Return the console tty driver structure and its associated index
2606 */
2607struct tty_driver *console_device(int *index)
2608{
2609 struct console *c;
2610 struct tty_driver *driver = NULL;
2611
2612 console_lock();
2613 for_each_console(c) {
2614 if (!c->device)
2615 continue;
2616 driver = c->device(c, index);
2617 if (driver)
2618 break;
2619 }
2620 console_unlock();
2621 return driver;
2622}
2623
2624/*
2625 * Prevent further output on the passed console device so that (for example)
2626 * serial drivers can disable console output before suspending a port, and can
2627 * re-enable output afterwards.
2628 */
2629void console_stop(struct console *console)
2630{
2631 console_lock();
2632 console->flags &= ~CON_ENABLED;
2633 console_unlock();
2634}
2635EXPORT_SYMBOL(console_stop);
2636
2637void console_start(struct console *console)
2638{
2639 console_lock();
2640 console->flags |= CON_ENABLED;
2641 console_unlock();
2642}
2643EXPORT_SYMBOL(console_start);
2644
2645static int __read_mostly keep_bootcon;
2646
2647static int __init keep_bootcon_setup(char *str)
2648{
2649 keep_bootcon = 1;
2650 pr_info("debug: skip boot console de-registration.\n");
2651
2652 return 0;
2653}
2654
2655early_param("keep_bootcon", keep_bootcon_setup);
2656
2657/*
2658 * This is called by register_console() to try to match
2659 * the newly registered console with any of the ones selected
2660 * by either the command line or add_preferred_console() and
2661 * setup/enable it.
2662 *
2663 * Care need to be taken with consoles that are statically
2664 * enabled such as netconsole
2665 */
2666static int try_enable_new_console(struct console *newcon, bool user_specified)
2667{
2668 struct console_cmdline *c;
2669 int i, err;
2670
2671 for (i = 0, c = console_cmdline;
2672 i < MAX_CMDLINECONSOLES && c->name[0];
2673 i++, c++) {
2674 if (c->user_specified != user_specified)
2675 continue;
2676 if (!newcon->match ||
2677 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2678 /* default matching */
2679 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2680 if (strcmp(c->name, newcon->name) != 0)
2681 continue;
2682 if (newcon->index >= 0 &&
2683 newcon->index != c->index)
2684 continue;
2685 if (newcon->index < 0)
2686 newcon->index = c->index;
2687
2688 if (_braille_register_console(newcon, c))
2689 return 0;
2690
2691 if (newcon->setup &&
2692 (err = newcon->setup(newcon, c->options)) != 0)
2693 return err;
2694 }
2695 newcon->flags |= CON_ENABLED;
2696 if (i == preferred_console) {
2697 newcon->flags |= CON_CONSDEV;
2698 has_preferred_console = true;
2699 }
2700 return 0;
2701 }
2702
2703 /*
2704 * Some consoles, such as pstore and netconsole, can be enabled even
2705 * without matching. Accept the pre-enabled consoles only when match()
2706 * and setup() had a chance to be called.
2707 */
2708 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
2709 return 0;
2710
2711 return -ENOENT;
2712}
2713
2714/*
2715 * The console driver calls this routine during kernel initialization
2716 * to register the console printing procedure with printk() and to
2717 * print any messages that were printed by the kernel before the
2718 * console driver was initialized.
2719 *
2720 * This can happen pretty early during the boot process (because of
2721 * early_printk) - sometimes before setup_arch() completes - be careful
2722 * of what kernel features are used - they may not be initialised yet.
2723 *
2724 * There are two types of consoles - bootconsoles (early_printk) and
2725 * "real" consoles (everything which is not a bootconsole) which are
2726 * handled differently.
2727 * - Any number of bootconsoles can be registered at any time.
2728 * - As soon as a "real" console is registered, all bootconsoles
2729 * will be unregistered automatically.
2730 * - Once a "real" console is registered, any attempt to register a
2731 * bootconsoles will be rejected
2732 */
2733void register_console(struct console *newcon)
2734{
2735 unsigned long flags;
2736 struct console *bcon = NULL;
2737 int err;
2738
2739 for_each_console(bcon) {
2740 if (WARN(bcon == newcon, "console '%s%d' already registered\n",
2741 bcon->name, bcon->index))
2742 return;
2743 }
2744
2745 /*
2746 * before we register a new CON_BOOT console, make sure we don't
2747 * already have a valid console
2748 */
2749 if (newcon->flags & CON_BOOT) {
2750 for_each_console(bcon) {
2751 if (!(bcon->flags & CON_BOOT)) {
2752 pr_info("Too late to register bootconsole %s%d\n",
2753 newcon->name, newcon->index);
2754 return;
2755 }
2756 }
2757 }
2758
2759 if (console_drivers && console_drivers->flags & CON_BOOT)
2760 bcon = console_drivers;
2761
2762 if (!has_preferred_console || bcon || !console_drivers)
2763 has_preferred_console = preferred_console >= 0;
2764
2765 /*
2766 * See if we want to use this console driver. If we
2767 * didn't select a console we take the first one
2768 * that registers here.
2769 */
2770 if (!has_preferred_console) {
2771 if (newcon->index < 0)
2772 newcon->index = 0;
2773 if (newcon->setup == NULL ||
2774 newcon->setup(newcon, NULL) == 0) {
2775 newcon->flags |= CON_ENABLED;
2776 if (newcon->device) {
2777 newcon->flags |= CON_CONSDEV;
2778 has_preferred_console = true;
2779 }
2780 }
2781 }
2782
2783 /* See if this console matches one we selected on the command line */
2784 err = try_enable_new_console(newcon, true);
2785
2786 /* If not, try to match against the platform default(s) */
2787 if (err == -ENOENT)
2788 err = try_enable_new_console(newcon, false);
2789
2790 /* printk() messages are not printed to the Braille console. */
2791 if (err || newcon->flags & CON_BRL)
2792 return;
2793
2794 /*
2795 * If we have a bootconsole, and are switching to a real console,
2796 * don't print everything out again, since when the boot console, and
2797 * the real console are the same physical device, it's annoying to
2798 * see the beginning boot messages twice
2799 */
2800 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2801 newcon->flags &= ~CON_PRINTBUFFER;
2802
2803 /*
2804 * Put this console in the list - keep the
2805 * preferred driver at the head of the list.
2806 */
2807 console_lock();
2808 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2809 newcon->next = console_drivers;
2810 console_drivers = newcon;
2811 if (newcon->next)
2812 newcon->next->flags &= ~CON_CONSDEV;
2813 /* Ensure this flag is always set for the head of the list */
2814 newcon->flags |= CON_CONSDEV;
2815 } else {
2816 newcon->next = console_drivers->next;
2817 console_drivers->next = newcon;
2818 }
2819
2820 if (newcon->flags & CON_EXTENDED)
2821 nr_ext_console_drivers++;
2822
2823 if (newcon->flags & CON_PRINTBUFFER) {
2824 /*
2825 * console_unlock(); will print out the buffered messages
2826 * for us.
2827 */
2828 logbuf_lock_irqsave(flags);
2829 /*
2830 * We're about to replay the log buffer. Only do this to the
2831 * just-registered console to avoid excessive message spam to
2832 * the already-registered consoles.
2833 *
2834 * Set exclusive_console with disabled interrupts to reduce
2835 * race window with eventual console_flush_on_panic() that
2836 * ignores console_lock.
2837 */
2838 exclusive_console = newcon;
2839 exclusive_console_stop_seq = console_seq;
2840 console_seq = syslog_seq;
2841 console_idx = syslog_idx;
2842 logbuf_unlock_irqrestore(flags);
2843 }
2844 console_unlock();
2845 console_sysfs_notify();
2846
2847 /*
2848 * By unregistering the bootconsoles after we enable the real console
2849 * we get the "console xxx enabled" message on all the consoles -
2850 * boot consoles, real consoles, etc - this is to ensure that end
2851 * users know there might be something in the kernel's log buffer that
2852 * went to the bootconsole (that they do not see on the real console)
2853 */
2854 pr_info("%sconsole [%s%d] enabled\n",
2855 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2856 newcon->name, newcon->index);
2857 if (bcon &&
2858 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2859 !keep_bootcon) {
2860 /* We need to iterate through all boot consoles, to make
2861 * sure we print everything out, before we unregister them.
2862 */
2863 for_each_console(bcon)
2864 if (bcon->flags & CON_BOOT)
2865 unregister_console(bcon);
2866 }
2867}
2868EXPORT_SYMBOL(register_console);
2869
2870int unregister_console(struct console *console)
2871{
2872 struct console *con;
2873 int res;
2874
2875 pr_info("%sconsole [%s%d] disabled\n",
2876 (console->flags & CON_BOOT) ? "boot" : "" ,
2877 console->name, console->index);
2878
2879 res = _braille_unregister_console(console);
2880 if (res < 0)
2881 return res;
2882 if (res > 0)
2883 return 0;
2884
2885 res = -ENODEV;
2886 console_lock();
2887 if (console_drivers == console) {
2888 console_drivers=console->next;
2889 res = 0;
2890 } else {
2891 for_each_console(con) {
2892 if (con->next == console) {
2893 con->next = console->next;
2894 res = 0;
2895 break;
2896 }
2897 }
2898 }
2899
2900 if (res)
2901 goto out_disable_unlock;
2902
2903 if (console->flags & CON_EXTENDED)
2904 nr_ext_console_drivers--;
2905
2906 /*
2907 * If this isn't the last console and it has CON_CONSDEV set, we
2908 * need to set it on the next preferred console.
2909 */
2910 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2911 console_drivers->flags |= CON_CONSDEV;
2912
2913 console->flags &= ~CON_ENABLED;
2914 console_unlock();
2915 console_sysfs_notify();
2916
2917 if (console->exit)
2918 res = console->exit(console);
2919
2920 return res;
2921
2922out_disable_unlock:
2923 console->flags &= ~CON_ENABLED;
2924 console_unlock();
2925
2926 return res;
2927}
2928EXPORT_SYMBOL(unregister_console);
2929
2930/*
2931 * Initialize the console device. This is called *early*, so
2932 * we can't necessarily depend on lots of kernel help here.
2933 * Just do some early initializations, and do the complex setup
2934 * later.
2935 */
2936void __init console_init(void)
2937{
2938 int ret;
2939 initcall_t call;
2940 initcall_entry_t *ce;
2941
2942 /* Setup the default TTY line discipline. */
2943 n_tty_init();
2944
2945 /*
2946 * set up the console device so that later boot sequences can
2947 * inform about problems etc..
2948 */
2949 ce = __con_initcall_start;
2950 trace_initcall_level("console");
2951 while (ce < __con_initcall_end) {
2952 call = initcall_from_entry(ce);
2953 trace_initcall_start(call);
2954 ret = call();
2955 trace_initcall_finish(call, ret);
2956 ce++;
2957 }
2958}
2959
2960/*
2961 * Some boot consoles access data that is in the init section and which will
2962 * be discarded after the initcalls have been run. To make sure that no code
2963 * will access this data, unregister the boot consoles in a late initcall.
2964 *
2965 * If for some reason, such as deferred probe or the driver being a loadable
2966 * module, the real console hasn't registered yet at this point, there will
2967 * be a brief interval in which no messages are logged to the console, which
2968 * makes it difficult to diagnose problems that occur during this time.
2969 *
2970 * To mitigate this problem somewhat, only unregister consoles whose memory
2971 * intersects with the init section. Note that all other boot consoles will
2972 * get unregistred when the real preferred console is registered.
2973 */
2974static int __init printk_late_init(void)
2975{
2976 struct console *con;
2977 int ret;
2978
2979 for_each_console(con) {
2980 if (!(con->flags & CON_BOOT))
2981 continue;
2982
2983 /* Check addresses that might be used for enabled consoles. */
2984 if (init_section_intersects(con, sizeof(*con)) ||
2985 init_section_contains(con->write, 0) ||
2986 init_section_contains(con->read, 0) ||
2987 init_section_contains(con->device, 0) ||
2988 init_section_contains(con->unblank, 0) ||
2989 init_section_contains(con->data, 0)) {
2990 /*
2991 * Please, consider moving the reported consoles out
2992 * of the init section.
2993 */
2994 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2995 con->name, con->index);
2996 unregister_console(con);
2997 }
2998 }
2999 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3000 console_cpu_notify);
3001 WARN_ON(ret < 0);
3002 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3003 console_cpu_notify, NULL);
3004 WARN_ON(ret < 0);
3005 return 0;
3006}
3007late_initcall(printk_late_init);
3008
3009#if defined CONFIG_PRINTK
3010/*
3011 * Delayed printk version, for scheduler-internal messages:
3012 */
3013#define PRINTK_PENDING_WAKEUP 0x01
3014#define PRINTK_PENDING_OUTPUT 0x02
3015
3016static DEFINE_PER_CPU(int, printk_pending);
3017
3018static void wake_up_klogd_work_func(struct irq_work *irq_work)
3019{
3020 int pending = __this_cpu_xchg(printk_pending, 0);
3021
3022 if (pending & PRINTK_PENDING_OUTPUT) {
3023 /* If trylock fails, someone else is doing the printing */
3024 if (console_trylock())
3025 console_unlock();
3026 }
3027
3028 if (pending & PRINTK_PENDING_WAKEUP)
3029 wake_up_interruptible(&log_wait);
3030}
3031
3032static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
3033 .func = wake_up_klogd_work_func,
3034 .flags = ATOMIC_INIT(IRQ_WORK_LAZY),
3035};
3036
3037void wake_up_klogd(void)
3038{
3039 if (!printk_percpu_data_ready())
3040 return;
3041
3042 preempt_disable();
3043 if (waitqueue_active(&log_wait)) {
3044 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
3045 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3046 }
3047 preempt_enable();
3048}
3049
3050void defer_console_output(void)
3051{
3052 if (!printk_percpu_data_ready())
3053 return;
3054
3055 preempt_disable();
3056 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
3057 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3058 preempt_enable();
3059}
3060
3061int vprintk_deferred(const char *fmt, va_list args)
3062{
3063 int r;
3064
3065 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
3066 defer_console_output();
3067
3068 return r;
3069}
3070
3071int printk_deferred(const char *fmt, ...)
3072{
3073 va_list args;
3074 int r;
3075
3076 va_start(args, fmt);
3077 r = vprintk_deferred(fmt, args);
3078 va_end(args);
3079
3080 return r;
3081}
3082
3083/*
3084 * printk rate limiting, lifted from the networking subsystem.
3085 *
3086 * This enforces a rate limit: not more than 10 kernel messages
3087 * every 5s to make a denial-of-service attack impossible.
3088 */
3089DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3090
3091int __printk_ratelimit(const char *func)
3092{
3093 return ___ratelimit(&printk_ratelimit_state, func);
3094}
3095EXPORT_SYMBOL(__printk_ratelimit);
3096
3097/**
3098 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3099 * @caller_jiffies: pointer to caller's state
3100 * @interval_msecs: minimum interval between prints
3101 *
3102 * printk_timed_ratelimit() returns true if more than @interval_msecs
3103 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3104 * returned true.
3105 */
3106bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3107 unsigned int interval_msecs)
3108{
3109 unsigned long elapsed = jiffies - *caller_jiffies;
3110
3111 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3112 return false;
3113
3114 *caller_jiffies = jiffies;
3115 return true;
3116}
3117EXPORT_SYMBOL(printk_timed_ratelimit);
3118
3119static DEFINE_SPINLOCK(dump_list_lock);
3120static LIST_HEAD(dump_list);
3121
3122/**
3123 * kmsg_dump_register - register a kernel log dumper.
3124 * @dumper: pointer to the kmsg_dumper structure
3125 *
3126 * Adds a kernel log dumper to the system. The dump callback in the
3127 * structure will be called when the kernel oopses or panics and must be
3128 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3129 */
3130int kmsg_dump_register(struct kmsg_dumper *dumper)
3131{
3132 unsigned long flags;
3133 int err = -EBUSY;
3134
3135 /* The dump callback needs to be set */
3136 if (!dumper->dump)
3137 return -EINVAL;
3138
3139 spin_lock_irqsave(&dump_list_lock, flags);
3140 /* Don't allow registering multiple times */
3141 if (!dumper->registered) {
3142 dumper->registered = 1;
3143 list_add_tail_rcu(&dumper->list, &dump_list);
3144 err = 0;
3145 }
3146 spin_unlock_irqrestore(&dump_list_lock, flags);
3147
3148 return err;
3149}
3150EXPORT_SYMBOL_GPL(kmsg_dump_register);
3151
3152/**
3153 * kmsg_dump_unregister - unregister a kmsg dumper.
3154 * @dumper: pointer to the kmsg_dumper structure
3155 *
3156 * Removes a dump device from the system. Returns zero on success and
3157 * %-EINVAL otherwise.
3158 */
3159int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3160{
3161 unsigned long flags;
3162 int err = -EINVAL;
3163
3164 spin_lock_irqsave(&dump_list_lock, flags);
3165 if (dumper->registered) {
3166 dumper->registered = 0;
3167 list_del_rcu(&dumper->list);
3168 err = 0;
3169 }
3170 spin_unlock_irqrestore(&dump_list_lock, flags);
3171 synchronize_rcu();
3172
3173 return err;
3174}
3175EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3176
3177static bool always_kmsg_dump;
3178module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3179
3180const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
3181{
3182 switch (reason) {
3183 case KMSG_DUMP_PANIC:
3184 return "Panic";
3185 case KMSG_DUMP_OOPS:
3186 return "Oops";
3187 case KMSG_DUMP_EMERG:
3188 return "Emergency";
3189 case KMSG_DUMP_SHUTDOWN:
3190 return "Shutdown";
3191 default:
3192 return "Unknown";
3193 }
3194}
3195EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
3196
3197/**
3198 * kmsg_dump - dump kernel log to kernel message dumpers.
3199 * @reason: the reason (oops, panic etc) for dumping
3200 *
3201 * Call each of the registered dumper's dump() callback, which can
3202 * retrieve the kmsg records with kmsg_dump_get_line() or
3203 * kmsg_dump_get_buffer().
3204 */
3205void kmsg_dump(enum kmsg_dump_reason reason)
3206{
3207 struct kmsg_dumper *dumper;
3208 unsigned long flags;
3209
3210 rcu_read_lock();
3211 list_for_each_entry_rcu(dumper, &dump_list, list) {
3212 enum kmsg_dump_reason max_reason = dumper->max_reason;
3213
3214 /*
3215 * If client has not provided a specific max_reason, default
3216 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
3217 */
3218 if (max_reason == KMSG_DUMP_UNDEF) {
3219 max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
3220 KMSG_DUMP_OOPS;
3221 }
3222 if (reason > max_reason)
3223 continue;
3224
3225 /* initialize iterator with data about the stored records */
3226 dumper->active = true;
3227
3228 logbuf_lock_irqsave(flags);
3229 dumper->cur_seq = clear_seq;
3230 dumper->cur_idx = clear_idx;
3231 dumper->next_seq = log_next_seq;
3232 dumper->next_idx = log_next_idx;
3233 logbuf_unlock_irqrestore(flags);
3234
3235 /* invoke dumper which will iterate over records */
3236 dumper->dump(dumper, reason);
3237
3238 /* reset iterator */
3239 dumper->active = false;
3240 }
3241 rcu_read_unlock();
3242}
3243
3244/**
3245 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3246 * @dumper: registered kmsg dumper
3247 * @syslog: include the "<4>" prefixes
3248 * @line: buffer to copy the line to
3249 * @size: maximum size of the buffer
3250 * @len: length of line placed into buffer
3251 *
3252 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3253 * record, and copy one record into the provided buffer.
3254 *
3255 * Consecutive calls will return the next available record moving
3256 * towards the end of the buffer with the youngest messages.
3257 *
3258 * A return value of FALSE indicates that there are no more records to
3259 * read.
3260 *
3261 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3262 */
3263bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3264 char *line, size_t size, size_t *len)
3265{
3266 struct printk_log *msg;
3267 size_t l = 0;
3268 bool ret = false;
3269
3270 if (!dumper->active)
3271 goto out;
3272
3273 if (dumper->cur_seq < log_first_seq) {
3274 /* messages are gone, move to first available one */
3275 dumper->cur_seq = log_first_seq;
3276 dumper->cur_idx = log_first_idx;
3277 }
3278
3279 /* last entry */
3280 if (dumper->cur_seq >= log_next_seq)
3281 goto out;
3282
3283 msg = log_from_idx(dumper->cur_idx);
3284 l = msg_print_text(msg, syslog, printk_time, line, size);
3285
3286 dumper->cur_idx = log_next(dumper->cur_idx);
3287 dumper->cur_seq++;
3288 ret = true;
3289out:
3290 if (len)
3291 *len = l;
3292 return ret;
3293}
3294
3295/**
3296 * kmsg_dump_get_line - retrieve one kmsg log line
3297 * @dumper: registered kmsg dumper
3298 * @syslog: include the "<4>" prefixes
3299 * @line: buffer to copy the line to
3300 * @size: maximum size of the buffer
3301 * @len: length of line placed into buffer
3302 *
3303 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3304 * record, and copy one record into the provided buffer.
3305 *
3306 * Consecutive calls will return the next available record moving
3307 * towards the end of the buffer with the youngest messages.
3308 *
3309 * A return value of FALSE indicates that there are no more records to
3310 * read.
3311 */
3312bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3313 char *line, size_t size, size_t *len)
3314{
3315 unsigned long flags;
3316 bool ret;
3317
3318 logbuf_lock_irqsave(flags);
3319 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3320 logbuf_unlock_irqrestore(flags);
3321
3322 return ret;
3323}
3324EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3325
3326/**
3327 * kmsg_dump_get_buffer - copy kmsg log lines
3328 * @dumper: registered kmsg dumper
3329 * @syslog: include the "<4>" prefixes
3330 * @buf: buffer to copy the line to
3331 * @size: maximum size of the buffer
3332 * @len: length of line placed into buffer
3333 *
3334 * Start at the end of the kmsg buffer and fill the provided buffer
3335 * with as many of the the *youngest* kmsg records that fit into it.
3336 * If the buffer is large enough, all available kmsg records will be
3337 * copied with a single call.
3338 *
3339 * Consecutive calls will fill the buffer with the next block of
3340 * available older records, not including the earlier retrieved ones.
3341 *
3342 * A return value of FALSE indicates that there are no more records to
3343 * read.
3344 */
3345bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3346 char *buf, size_t size, size_t *len)
3347{
3348 unsigned long flags;
3349 u64 seq;
3350 u32 idx;
3351 u64 next_seq;
3352 u32 next_idx;
3353 size_t l = 0;
3354 bool ret = false;
3355 bool time = printk_time;
3356
3357 if (!dumper->active)
3358 goto out;
3359
3360 logbuf_lock_irqsave(flags);
3361 if (dumper->cur_seq < log_first_seq) {
3362 /* messages are gone, move to first available one */
3363 dumper->cur_seq = log_first_seq;
3364 dumper->cur_idx = log_first_idx;
3365 }
3366
3367 /* last entry */
3368 if (dumper->cur_seq >= dumper->next_seq) {
3369 logbuf_unlock_irqrestore(flags);
3370 goto out;
3371 }
3372
3373 /* calculate length of entire buffer */
3374 seq = dumper->cur_seq;
3375 idx = dumper->cur_idx;
3376 while (seq < dumper->next_seq) {
3377 struct printk_log *msg = log_from_idx(idx);
3378
3379 l += msg_print_text(msg, true, time, NULL, 0);
3380 idx = log_next(idx);
3381 seq++;
3382 }
3383
3384 /* move first record forward until length fits into the buffer */
3385 seq = dumper->cur_seq;
3386 idx = dumper->cur_idx;
3387 while (l >= size && seq < dumper->next_seq) {
3388 struct printk_log *msg = log_from_idx(idx);
3389
3390 l -= msg_print_text(msg, true, time, NULL, 0);
3391 idx = log_next(idx);
3392 seq++;
3393 }
3394
3395 /* last message in next interation */
3396 next_seq = seq;
3397 next_idx = idx;
3398
3399 l = 0;
3400 while (seq < dumper->next_seq) {
3401 struct printk_log *msg = log_from_idx(idx);
3402
3403 l += msg_print_text(msg, syslog, time, buf + l, size - l);
3404 idx = log_next(idx);
3405 seq++;
3406 }
3407
3408 dumper->next_seq = next_seq;
3409 dumper->next_idx = next_idx;
3410 ret = true;
3411 logbuf_unlock_irqrestore(flags);
3412out:
3413 if (len)
3414 *len = l;
3415 return ret;
3416}
3417EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3418
3419/**
3420 * kmsg_dump_rewind_nolock - reset the iterator (unlocked version)
3421 * @dumper: registered kmsg dumper
3422 *
3423 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3424 * kmsg_dump_get_buffer() can be called again and used multiple
3425 * times within the same dumper.dump() callback.
3426 *
3427 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3428 */
3429void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3430{
3431 dumper->cur_seq = clear_seq;
3432 dumper->cur_idx = clear_idx;
3433 dumper->next_seq = log_next_seq;
3434 dumper->next_idx = log_next_idx;
3435}
3436
3437/**
3438 * kmsg_dump_rewind - reset the iterator
3439 * @dumper: registered kmsg dumper
3440 *
3441 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3442 * kmsg_dump_get_buffer() can be called again and used multiple
3443 * times within the same dumper.dump() callback.
3444 */
3445void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3446{
3447 unsigned long flags;
3448
3449 logbuf_lock_irqsave(flags);
3450 kmsg_dump_rewind_nolock(dumper);
3451 logbuf_unlock_irqrestore(flags);
3452}
3453EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3454
3455#endif