Loading...
1/*
2 * linux/kernel/printk.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * Modified to make sys_syslog() more flexible: added commands to
7 * return the last 4k of kernel messages, regardless of whether
8 * they've been read or not. Added option to suppress kernel printk's
9 * to the console. Added hook for sending the console messages
10 * elsewhere, in preparation for a serial line console (someday).
11 * Ted Ts'o, 2/11/93.
12 * Modified for sysctl support, 1/8/97, Chris Horn.
13 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
14 * manfred@colorfullife.com
15 * Rewrote bits to get rid of console_lock
16 * 01Mar01 Andrew Morton
17 */
18
19#include <linux/kernel.h>
20#include <linux/mm.h>
21#include <linux/tty.h>
22#include <linux/tty_driver.h>
23#include <linux/console.h>
24#include <linux/init.h>
25#include <linux/jiffies.h>
26#include <linux/nmi.h>
27#include <linux/module.h>
28#include <linux/moduleparam.h>
29#include <linux/interrupt.h> /* For in_interrupt() */
30#include <linux/delay.h>
31#include <linux/smp.h>
32#include <linux/security.h>
33#include <linux/bootmem.h>
34#include <linux/memblock.h>
35#include <linux/syscalls.h>
36#include <linux/kexec.h>
37#include <linux/kdb.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/notifier.h>
43#include <linux/rculist.h>
44#include <linux/poll.h>
45#include <linux/irq_work.h>
46#include <linux/utsname.h>
47#include <linux/ctype.h>
48#include <linux/uio.h>
49
50#include <asm/uaccess.h>
51#include <asm-generic/sections.h>
52
53#define CREATE_TRACE_POINTS
54#include <trace/events/printk.h>
55
56#include "console_cmdline.h"
57#include "braille.h"
58
59int console_printk[4] = {
60 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
61 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
62 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
63 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
64};
65
66/*
67 * Low level drivers may need that to know if they can schedule in
68 * their unblank() callback or not. So let's export it.
69 */
70int oops_in_progress;
71EXPORT_SYMBOL(oops_in_progress);
72
73/*
74 * console_sem protects the console_drivers list, and also
75 * provides serialisation for access to the entire console
76 * driver system.
77 */
78static DEFINE_SEMAPHORE(console_sem);
79struct console *console_drivers;
80EXPORT_SYMBOL_GPL(console_drivers);
81
82#ifdef CONFIG_LOCKDEP
83static struct lockdep_map console_lock_dep_map = {
84 .name = "console_lock"
85};
86#endif
87
88/*
89 * Number of registered extended console drivers.
90 *
91 * If extended consoles are present, in-kernel cont reassembly is disabled
92 * and each fragment is stored as a separate log entry with proper
93 * continuation flag so that every emitted message has full metadata. This
94 * doesn't change the result for regular consoles or /proc/kmsg. For
95 * /dev/kmsg, as long as the reader concatenates messages according to
96 * consecutive continuation flags, the end result should be the same too.
97 */
98static int nr_ext_console_drivers;
99
100/*
101 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
102 * macros instead of functions so that _RET_IP_ contains useful information.
103 */
104#define down_console_sem() do { \
105 down(&console_sem);\
106 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
107} while (0)
108
109static int __down_trylock_console_sem(unsigned long ip)
110{
111 if (down_trylock(&console_sem))
112 return 1;
113 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
114 return 0;
115}
116#define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
117
118#define up_console_sem() do { \
119 mutex_release(&console_lock_dep_map, 1, _RET_IP_);\
120 up(&console_sem);\
121} while (0)
122
123/*
124 * This is used for debugging the mess that is the VT code by
125 * keeping track if we have the console semaphore held. It's
126 * definitely not the perfect debug tool (we don't know if _WE_
127 * hold it and are racing, but it helps tracking those weird code
128 * paths in the console code where we end up in places I want
129 * locked without the console sempahore held).
130 */
131static int console_locked, console_suspended;
132
133/*
134 * If exclusive_console is non-NULL then only this console is to be printed to.
135 */
136static struct console *exclusive_console;
137
138/*
139 * Array of consoles built from command line options (console=)
140 */
141
142#define MAX_CMDLINECONSOLES 8
143
144static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
145
146static int selected_console = -1;
147static int preferred_console = -1;
148int console_set_on_cmdline;
149EXPORT_SYMBOL(console_set_on_cmdline);
150
151/* Flag: console code may call schedule() */
152static int console_may_schedule;
153
154/*
155 * The printk log buffer consists of a chain of concatenated variable
156 * length records. Every record starts with a record header, containing
157 * the overall length of the record.
158 *
159 * The heads to the first and last entry in the buffer, as well as the
160 * sequence numbers of these entries are maintained when messages are
161 * stored.
162 *
163 * If the heads indicate available messages, the length in the header
164 * tells the start next message. A length == 0 for the next message
165 * indicates a wrap-around to the beginning of the buffer.
166 *
167 * Every record carries the monotonic timestamp in microseconds, as well as
168 * the standard userspace syslog level and syslog facility. The usual
169 * kernel messages use LOG_KERN; userspace-injected messages always carry
170 * a matching syslog facility, by default LOG_USER. The origin of every
171 * message can be reliably determined that way.
172 *
173 * The human readable log message directly follows the message header. The
174 * length of the message text is stored in the header, the stored message
175 * is not terminated.
176 *
177 * Optionally, a message can carry a dictionary of properties (key/value pairs),
178 * to provide userspace with a machine-readable message context.
179 *
180 * Examples for well-defined, commonly used property names are:
181 * DEVICE=b12:8 device identifier
182 * b12:8 block dev_t
183 * c127:3 char dev_t
184 * n8 netdev ifindex
185 * +sound:card0 subsystem:devname
186 * SUBSYSTEM=pci driver-core subsystem name
187 *
188 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
189 * follows directly after a '=' character. Every property is terminated by
190 * a '\0' character. The last property is not terminated.
191 *
192 * Example of a message structure:
193 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
194 * 0008 34 00 record is 52 bytes long
195 * 000a 0b 00 text is 11 bytes long
196 * 000c 1f 00 dictionary is 23 bytes long
197 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
198 * 0010 69 74 27 73 20 61 20 6c "it's a l"
199 * 69 6e 65 "ine"
200 * 001b 44 45 56 49 43 "DEVIC"
201 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
202 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
203 * 67 "g"
204 * 0032 00 00 00 padding to next message header
205 *
206 * The 'struct printk_log' buffer header must never be directly exported to
207 * userspace, it is a kernel-private implementation detail that might
208 * need to be changed in the future, when the requirements change.
209 *
210 * /dev/kmsg exports the structured data in the following line format:
211 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
212 *
213 * Users of the export format should ignore possible additional values
214 * separated by ',', and find the message after the ';' character.
215 *
216 * The optional key/value pairs are attached as continuation lines starting
217 * with a space character and terminated by a newline. All possible
218 * non-prinatable characters are escaped in the "\xff" notation.
219 */
220
221enum log_flags {
222 LOG_NOCONS = 1, /* already flushed, do not print to console */
223 LOG_NEWLINE = 2, /* text ended with a newline */
224 LOG_PREFIX = 4, /* text started with a prefix */
225 LOG_CONT = 8, /* text is a fragment of a continuation line */
226};
227
228struct printk_log {
229 u64 ts_nsec; /* timestamp in nanoseconds */
230 u16 len; /* length of entire record */
231 u16 text_len; /* length of text buffer */
232 u16 dict_len; /* length of dictionary buffer */
233 u8 facility; /* syslog facility */
234 u8 flags:5; /* internal record flags */
235 u8 level:3; /* syslog level */
236}
237#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
238__packed __aligned(4)
239#endif
240;
241
242/*
243 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
244 * within the scheduler's rq lock. It must be released before calling
245 * console_unlock() or anything else that might wake up a process.
246 */
247static DEFINE_RAW_SPINLOCK(logbuf_lock);
248
249#ifdef CONFIG_PRINTK
250DECLARE_WAIT_QUEUE_HEAD(log_wait);
251/* the next printk record to read by syslog(READ) or /proc/kmsg */
252static u64 syslog_seq;
253static u32 syslog_idx;
254static enum log_flags syslog_prev;
255static size_t syslog_partial;
256
257/* index and sequence number of the first record stored in the buffer */
258static u64 log_first_seq;
259static u32 log_first_idx;
260
261/* index and sequence number of the next record to store in the buffer */
262static u64 log_next_seq;
263static u32 log_next_idx;
264
265/* the next printk record to write to the console */
266static u64 console_seq;
267static u32 console_idx;
268static enum log_flags console_prev;
269
270/* the next printk record to read after the last 'clear' command */
271static u64 clear_seq;
272static u32 clear_idx;
273
274#define PREFIX_MAX 32
275#define LOG_LINE_MAX (1024 - PREFIX_MAX)
276
277#define LOG_LEVEL(v) ((v) & 0x07)
278#define LOG_FACILITY(v) ((v) >> 3 & 0xff)
279
280/* record buffer */
281#define LOG_ALIGN __alignof__(struct printk_log)
282#define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
283static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
284static char *log_buf = __log_buf;
285static u32 log_buf_len = __LOG_BUF_LEN;
286
287/* Return log buffer address */
288char *log_buf_addr_get(void)
289{
290 return log_buf;
291}
292
293/* Return log buffer size */
294u32 log_buf_len_get(void)
295{
296 return log_buf_len;
297}
298
299/* human readable text of the record */
300static char *log_text(const struct printk_log *msg)
301{
302 return (char *)msg + sizeof(struct printk_log);
303}
304
305/* optional key/value pair dictionary attached to the record */
306static char *log_dict(const struct printk_log *msg)
307{
308 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
309}
310
311/* get record by index; idx must point to valid msg */
312static struct printk_log *log_from_idx(u32 idx)
313{
314 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
315
316 /*
317 * A length == 0 record is the end of buffer marker. Wrap around and
318 * read the message at the start of the buffer.
319 */
320 if (!msg->len)
321 return (struct printk_log *)log_buf;
322 return msg;
323}
324
325/* get next record; idx must point to valid msg */
326static u32 log_next(u32 idx)
327{
328 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
329
330 /* length == 0 indicates the end of the buffer; wrap */
331 /*
332 * A length == 0 record is the end of buffer marker. Wrap around and
333 * read the message at the start of the buffer as *this* one, and
334 * return the one after that.
335 */
336 if (!msg->len) {
337 msg = (struct printk_log *)log_buf;
338 return msg->len;
339 }
340 return idx + msg->len;
341}
342
343/*
344 * Check whether there is enough free space for the given message.
345 *
346 * The same values of first_idx and next_idx mean that the buffer
347 * is either empty or full.
348 *
349 * If the buffer is empty, we must respect the position of the indexes.
350 * They cannot be reset to the beginning of the buffer.
351 */
352static int logbuf_has_space(u32 msg_size, bool empty)
353{
354 u32 free;
355
356 if (log_next_idx > log_first_idx || empty)
357 free = max(log_buf_len - log_next_idx, log_first_idx);
358 else
359 free = log_first_idx - log_next_idx;
360
361 /*
362 * We need space also for an empty header that signalizes wrapping
363 * of the buffer.
364 */
365 return free >= msg_size + sizeof(struct printk_log);
366}
367
368static int log_make_free_space(u32 msg_size)
369{
370 while (log_first_seq < log_next_seq &&
371 !logbuf_has_space(msg_size, false)) {
372 /* drop old messages until we have enough contiguous space */
373 log_first_idx = log_next(log_first_idx);
374 log_first_seq++;
375 }
376
377 if (clear_seq < log_first_seq) {
378 clear_seq = log_first_seq;
379 clear_idx = log_first_idx;
380 }
381
382 /* sequence numbers are equal, so the log buffer is empty */
383 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
384 return 0;
385
386 return -ENOMEM;
387}
388
389/* compute the message size including the padding bytes */
390static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
391{
392 u32 size;
393
394 size = sizeof(struct printk_log) + text_len + dict_len;
395 *pad_len = (-size) & (LOG_ALIGN - 1);
396 size += *pad_len;
397
398 return size;
399}
400
401/*
402 * Define how much of the log buffer we could take at maximum. The value
403 * must be greater than two. Note that only half of the buffer is available
404 * when the index points to the middle.
405 */
406#define MAX_LOG_TAKE_PART 4
407static const char trunc_msg[] = "<truncated>";
408
409static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
410 u16 *dict_len, u32 *pad_len)
411{
412 /*
413 * The message should not take the whole buffer. Otherwise, it might
414 * get removed too soon.
415 */
416 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
417 if (*text_len > max_text_len)
418 *text_len = max_text_len;
419 /* enable the warning message */
420 *trunc_msg_len = strlen(trunc_msg);
421 /* disable the "dict" completely */
422 *dict_len = 0;
423 /* compute the size again, count also the warning message */
424 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
425}
426
427/* insert record into the buffer, discard old ones, update heads */
428static int log_store(int facility, int level,
429 enum log_flags flags, u64 ts_nsec,
430 const char *dict, u16 dict_len,
431 const char *text, u16 text_len)
432{
433 struct printk_log *msg;
434 u32 size, pad_len;
435 u16 trunc_msg_len = 0;
436
437 /* number of '\0' padding bytes to next message */
438 size = msg_used_size(text_len, dict_len, &pad_len);
439
440 if (log_make_free_space(size)) {
441 /* truncate the message if it is too long for empty buffer */
442 size = truncate_msg(&text_len, &trunc_msg_len,
443 &dict_len, &pad_len);
444 /* survive when the log buffer is too small for trunc_msg */
445 if (log_make_free_space(size))
446 return 0;
447 }
448
449 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
450 /*
451 * This message + an additional empty header does not fit
452 * at the end of the buffer. Add an empty header with len == 0
453 * to signify a wrap around.
454 */
455 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
456 log_next_idx = 0;
457 }
458
459 /* fill message */
460 msg = (struct printk_log *)(log_buf + log_next_idx);
461 memcpy(log_text(msg), text, text_len);
462 msg->text_len = text_len;
463 if (trunc_msg_len) {
464 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
465 msg->text_len += trunc_msg_len;
466 }
467 memcpy(log_dict(msg), dict, dict_len);
468 msg->dict_len = dict_len;
469 msg->facility = facility;
470 msg->level = level & 7;
471 msg->flags = flags & 0x1f;
472 if (ts_nsec > 0)
473 msg->ts_nsec = ts_nsec;
474 else
475 msg->ts_nsec = local_clock();
476 memset(log_dict(msg) + dict_len, 0, pad_len);
477 msg->len = size;
478
479 /* insert message */
480 log_next_idx += msg->len;
481 log_next_seq++;
482
483 return msg->text_len;
484}
485
486int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
487
488static int syslog_action_restricted(int type)
489{
490 if (dmesg_restrict)
491 return 1;
492 /*
493 * Unless restricted, we allow "read all" and "get buffer size"
494 * for everybody.
495 */
496 return type != SYSLOG_ACTION_READ_ALL &&
497 type != SYSLOG_ACTION_SIZE_BUFFER;
498}
499
500int check_syslog_permissions(int type, int source)
501{
502 /*
503 * If this is from /proc/kmsg and we've already opened it, then we've
504 * already done the capabilities checks at open time.
505 */
506 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
507 goto ok;
508
509 if (syslog_action_restricted(type)) {
510 if (capable(CAP_SYSLOG))
511 goto ok;
512 /*
513 * For historical reasons, accept CAP_SYS_ADMIN too, with
514 * a warning.
515 */
516 if (capable(CAP_SYS_ADMIN)) {
517 pr_warn_once("%s (%d): Attempt to access syslog with "
518 "CAP_SYS_ADMIN but no CAP_SYSLOG "
519 "(deprecated).\n",
520 current->comm, task_pid_nr(current));
521 goto ok;
522 }
523 return -EPERM;
524 }
525ok:
526 return security_syslog(type);
527}
528EXPORT_SYMBOL_GPL(check_syslog_permissions);
529
530static void append_char(char **pp, char *e, char c)
531{
532 if (*pp < e)
533 *(*pp)++ = c;
534}
535
536static ssize_t msg_print_ext_header(char *buf, size_t size,
537 struct printk_log *msg, u64 seq,
538 enum log_flags prev_flags)
539{
540 u64 ts_usec = msg->ts_nsec;
541 char cont = '-';
542
543 do_div(ts_usec, 1000);
544
545 /*
546 * If we couldn't merge continuation line fragments during the print,
547 * export the stored flags to allow an optional external merge of the
548 * records. Merging the records isn't always neccessarily correct, like
549 * when we hit a race during printing. In most cases though, it produces
550 * better readable output. 'c' in the record flags mark the first
551 * fragment of a line, '+' the following.
552 */
553 if (msg->flags & LOG_CONT && !(prev_flags & LOG_CONT))
554 cont = 'c';
555 else if ((msg->flags & LOG_CONT) ||
556 ((prev_flags & LOG_CONT) && !(msg->flags & LOG_PREFIX)))
557 cont = '+';
558
559 return scnprintf(buf, size, "%u,%llu,%llu,%c;",
560 (msg->facility << 3) | msg->level, seq, ts_usec, cont);
561}
562
563static ssize_t msg_print_ext_body(char *buf, size_t size,
564 char *dict, size_t dict_len,
565 char *text, size_t text_len)
566{
567 char *p = buf, *e = buf + size;
568 size_t i;
569
570 /* escape non-printable characters */
571 for (i = 0; i < text_len; i++) {
572 unsigned char c = text[i];
573
574 if (c < ' ' || c >= 127 || c == '\\')
575 p += scnprintf(p, e - p, "\\x%02x", c);
576 else
577 append_char(&p, e, c);
578 }
579 append_char(&p, e, '\n');
580
581 if (dict_len) {
582 bool line = true;
583
584 for (i = 0; i < dict_len; i++) {
585 unsigned char c = dict[i];
586
587 if (line) {
588 append_char(&p, e, ' ');
589 line = false;
590 }
591
592 if (c == '\0') {
593 append_char(&p, e, '\n');
594 line = true;
595 continue;
596 }
597
598 if (c < ' ' || c >= 127 || c == '\\') {
599 p += scnprintf(p, e - p, "\\x%02x", c);
600 continue;
601 }
602
603 append_char(&p, e, c);
604 }
605 append_char(&p, e, '\n');
606 }
607
608 return p - buf;
609}
610
611/* /dev/kmsg - userspace message inject/listen interface */
612struct devkmsg_user {
613 u64 seq;
614 u32 idx;
615 enum log_flags prev;
616 struct mutex lock;
617 char buf[CONSOLE_EXT_LOG_MAX];
618};
619
620static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
621{
622 char *buf, *line;
623 int level = default_message_loglevel;
624 int facility = 1; /* LOG_USER */
625 size_t len = iov_iter_count(from);
626 ssize_t ret = len;
627
628 if (len > LOG_LINE_MAX)
629 return -EINVAL;
630 buf = kmalloc(len+1, GFP_KERNEL);
631 if (buf == NULL)
632 return -ENOMEM;
633
634 buf[len] = '\0';
635 if (copy_from_iter(buf, len, from) != len) {
636 kfree(buf);
637 return -EFAULT;
638 }
639
640 /*
641 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
642 * the decimal value represents 32bit, the lower 3 bit are the log
643 * level, the rest are the log facility.
644 *
645 * If no prefix or no userspace facility is specified, we
646 * enforce LOG_USER, to be able to reliably distinguish
647 * kernel-generated messages from userspace-injected ones.
648 */
649 line = buf;
650 if (line[0] == '<') {
651 char *endp = NULL;
652 unsigned int u;
653
654 u = simple_strtoul(line + 1, &endp, 10);
655 if (endp && endp[0] == '>') {
656 level = LOG_LEVEL(u);
657 if (LOG_FACILITY(u) != 0)
658 facility = LOG_FACILITY(u);
659 endp++;
660 len -= endp - line;
661 line = endp;
662 }
663 }
664
665 printk_emit(facility, level, NULL, 0, "%s", line);
666 kfree(buf);
667 return ret;
668}
669
670static ssize_t devkmsg_read(struct file *file, char __user *buf,
671 size_t count, loff_t *ppos)
672{
673 struct devkmsg_user *user = file->private_data;
674 struct printk_log *msg;
675 size_t len;
676 ssize_t ret;
677
678 if (!user)
679 return -EBADF;
680
681 ret = mutex_lock_interruptible(&user->lock);
682 if (ret)
683 return ret;
684 raw_spin_lock_irq(&logbuf_lock);
685 while (user->seq == log_next_seq) {
686 if (file->f_flags & O_NONBLOCK) {
687 ret = -EAGAIN;
688 raw_spin_unlock_irq(&logbuf_lock);
689 goto out;
690 }
691
692 raw_spin_unlock_irq(&logbuf_lock);
693 ret = wait_event_interruptible(log_wait,
694 user->seq != log_next_seq);
695 if (ret)
696 goto out;
697 raw_spin_lock_irq(&logbuf_lock);
698 }
699
700 if (user->seq < log_first_seq) {
701 /* our last seen message is gone, return error and reset */
702 user->idx = log_first_idx;
703 user->seq = log_first_seq;
704 ret = -EPIPE;
705 raw_spin_unlock_irq(&logbuf_lock);
706 goto out;
707 }
708
709 msg = log_from_idx(user->idx);
710 len = msg_print_ext_header(user->buf, sizeof(user->buf),
711 msg, user->seq, user->prev);
712 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
713 log_dict(msg), msg->dict_len,
714 log_text(msg), msg->text_len);
715
716 user->prev = msg->flags;
717 user->idx = log_next(user->idx);
718 user->seq++;
719 raw_spin_unlock_irq(&logbuf_lock);
720
721 if (len > count) {
722 ret = -EINVAL;
723 goto out;
724 }
725
726 if (copy_to_user(buf, user->buf, len)) {
727 ret = -EFAULT;
728 goto out;
729 }
730 ret = len;
731out:
732 mutex_unlock(&user->lock);
733 return ret;
734}
735
736static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
737{
738 struct devkmsg_user *user = file->private_data;
739 loff_t ret = 0;
740
741 if (!user)
742 return -EBADF;
743 if (offset)
744 return -ESPIPE;
745
746 raw_spin_lock_irq(&logbuf_lock);
747 switch (whence) {
748 case SEEK_SET:
749 /* the first record */
750 user->idx = log_first_idx;
751 user->seq = log_first_seq;
752 break;
753 case SEEK_DATA:
754 /*
755 * The first record after the last SYSLOG_ACTION_CLEAR,
756 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
757 * changes no global state, and does not clear anything.
758 */
759 user->idx = clear_idx;
760 user->seq = clear_seq;
761 break;
762 case SEEK_END:
763 /* after the last record */
764 user->idx = log_next_idx;
765 user->seq = log_next_seq;
766 break;
767 default:
768 ret = -EINVAL;
769 }
770 raw_spin_unlock_irq(&logbuf_lock);
771 return ret;
772}
773
774static unsigned int devkmsg_poll(struct file *file, poll_table *wait)
775{
776 struct devkmsg_user *user = file->private_data;
777 int ret = 0;
778
779 if (!user)
780 return POLLERR|POLLNVAL;
781
782 poll_wait(file, &log_wait, wait);
783
784 raw_spin_lock_irq(&logbuf_lock);
785 if (user->seq < log_next_seq) {
786 /* return error when data has vanished underneath us */
787 if (user->seq < log_first_seq)
788 ret = POLLIN|POLLRDNORM|POLLERR|POLLPRI;
789 else
790 ret = POLLIN|POLLRDNORM;
791 }
792 raw_spin_unlock_irq(&logbuf_lock);
793
794 return ret;
795}
796
797static int devkmsg_open(struct inode *inode, struct file *file)
798{
799 struct devkmsg_user *user;
800 int err;
801
802 /* write-only does not need any file context */
803 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
804 return 0;
805
806 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
807 SYSLOG_FROM_READER);
808 if (err)
809 return err;
810
811 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
812 if (!user)
813 return -ENOMEM;
814
815 mutex_init(&user->lock);
816
817 raw_spin_lock_irq(&logbuf_lock);
818 user->idx = log_first_idx;
819 user->seq = log_first_seq;
820 raw_spin_unlock_irq(&logbuf_lock);
821
822 file->private_data = user;
823 return 0;
824}
825
826static int devkmsg_release(struct inode *inode, struct file *file)
827{
828 struct devkmsg_user *user = file->private_data;
829
830 if (!user)
831 return 0;
832
833 mutex_destroy(&user->lock);
834 kfree(user);
835 return 0;
836}
837
838const struct file_operations kmsg_fops = {
839 .open = devkmsg_open,
840 .read = devkmsg_read,
841 .write_iter = devkmsg_write,
842 .llseek = devkmsg_llseek,
843 .poll = devkmsg_poll,
844 .release = devkmsg_release,
845};
846
847#ifdef CONFIG_KEXEC_CORE
848/*
849 * This appends the listed symbols to /proc/vmcore
850 *
851 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
852 * obtain access to symbols that are otherwise very difficult to locate. These
853 * symbols are specifically used so that utilities can access and extract the
854 * dmesg log from a vmcore file after a crash.
855 */
856void log_buf_kexec_setup(void)
857{
858 VMCOREINFO_SYMBOL(log_buf);
859 VMCOREINFO_SYMBOL(log_buf_len);
860 VMCOREINFO_SYMBOL(log_first_idx);
861 VMCOREINFO_SYMBOL(clear_idx);
862 VMCOREINFO_SYMBOL(log_next_idx);
863 /*
864 * Export struct printk_log size and field offsets. User space tools can
865 * parse it and detect any changes to structure down the line.
866 */
867 VMCOREINFO_STRUCT_SIZE(printk_log);
868 VMCOREINFO_OFFSET(printk_log, ts_nsec);
869 VMCOREINFO_OFFSET(printk_log, len);
870 VMCOREINFO_OFFSET(printk_log, text_len);
871 VMCOREINFO_OFFSET(printk_log, dict_len);
872}
873#endif
874
875/* requested log_buf_len from kernel cmdline */
876static unsigned long __initdata new_log_buf_len;
877
878/* we practice scaling the ring buffer by powers of 2 */
879static void __init log_buf_len_update(unsigned size)
880{
881 if (size)
882 size = roundup_pow_of_two(size);
883 if (size > log_buf_len)
884 new_log_buf_len = size;
885}
886
887/* save requested log_buf_len since it's too early to process it */
888static int __init log_buf_len_setup(char *str)
889{
890 unsigned size = memparse(str, &str);
891
892 log_buf_len_update(size);
893
894 return 0;
895}
896early_param("log_buf_len", log_buf_len_setup);
897
898#ifdef CONFIG_SMP
899#define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
900
901static void __init log_buf_add_cpu(void)
902{
903 unsigned int cpu_extra;
904
905 /*
906 * archs should set up cpu_possible_bits properly with
907 * set_cpu_possible() after setup_arch() but just in
908 * case lets ensure this is valid.
909 */
910 if (num_possible_cpus() == 1)
911 return;
912
913 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
914
915 /* by default this will only continue through for large > 64 CPUs */
916 if (cpu_extra <= __LOG_BUF_LEN / 2)
917 return;
918
919 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
920 __LOG_CPU_MAX_BUF_LEN);
921 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
922 cpu_extra);
923 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
924
925 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
926}
927#else /* !CONFIG_SMP */
928static inline void log_buf_add_cpu(void) {}
929#endif /* CONFIG_SMP */
930
931void __init setup_log_buf(int early)
932{
933 unsigned long flags;
934 char *new_log_buf;
935 int free;
936
937 if (log_buf != __log_buf)
938 return;
939
940 if (!early && !new_log_buf_len)
941 log_buf_add_cpu();
942
943 if (!new_log_buf_len)
944 return;
945
946 if (early) {
947 new_log_buf =
948 memblock_virt_alloc(new_log_buf_len, LOG_ALIGN);
949 } else {
950 new_log_buf = memblock_virt_alloc_nopanic(new_log_buf_len,
951 LOG_ALIGN);
952 }
953
954 if (unlikely(!new_log_buf)) {
955 pr_err("log_buf_len: %ld bytes not available\n",
956 new_log_buf_len);
957 return;
958 }
959
960 raw_spin_lock_irqsave(&logbuf_lock, flags);
961 log_buf_len = new_log_buf_len;
962 log_buf = new_log_buf;
963 new_log_buf_len = 0;
964 free = __LOG_BUF_LEN - log_next_idx;
965 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
966 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
967
968 pr_info("log_buf_len: %d bytes\n", log_buf_len);
969 pr_info("early log buf free: %d(%d%%)\n",
970 free, (free * 100) / __LOG_BUF_LEN);
971}
972
973static bool __read_mostly ignore_loglevel;
974
975static int __init ignore_loglevel_setup(char *str)
976{
977 ignore_loglevel = true;
978 pr_info("debug: ignoring loglevel setting.\n");
979
980 return 0;
981}
982
983early_param("ignore_loglevel", ignore_loglevel_setup);
984module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
985MODULE_PARM_DESC(ignore_loglevel,
986 "ignore loglevel setting (prints all kernel messages to the console)");
987
988#ifdef CONFIG_BOOT_PRINTK_DELAY
989
990static int boot_delay; /* msecs delay after each printk during bootup */
991static unsigned long long loops_per_msec; /* based on boot_delay */
992
993static int __init boot_delay_setup(char *str)
994{
995 unsigned long lpj;
996
997 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
998 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
999
1000 get_option(&str, &boot_delay);
1001 if (boot_delay > 10 * 1000)
1002 boot_delay = 0;
1003
1004 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1005 "HZ: %d, loops_per_msec: %llu\n",
1006 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1007 return 0;
1008}
1009early_param("boot_delay", boot_delay_setup);
1010
1011static void boot_delay_msec(int level)
1012{
1013 unsigned long long k;
1014 unsigned long timeout;
1015
1016 if ((boot_delay == 0 || system_state != SYSTEM_BOOTING)
1017 || (level >= console_loglevel && !ignore_loglevel)) {
1018 return;
1019 }
1020
1021 k = (unsigned long long)loops_per_msec * boot_delay;
1022
1023 timeout = jiffies + msecs_to_jiffies(boot_delay);
1024 while (k) {
1025 k--;
1026 cpu_relax();
1027 /*
1028 * use (volatile) jiffies to prevent
1029 * compiler reduction; loop termination via jiffies
1030 * is secondary and may or may not happen.
1031 */
1032 if (time_after(jiffies, timeout))
1033 break;
1034 touch_nmi_watchdog();
1035 }
1036}
1037#else
1038static inline void boot_delay_msec(int level)
1039{
1040}
1041#endif
1042
1043static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1044module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1045
1046static size_t print_time(u64 ts, char *buf)
1047{
1048 unsigned long rem_nsec;
1049
1050 if (!printk_time)
1051 return 0;
1052
1053 rem_nsec = do_div(ts, 1000000000);
1054
1055 if (!buf)
1056 return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
1057
1058 return sprintf(buf, "[%5lu.%06lu] ",
1059 (unsigned long)ts, rem_nsec / 1000);
1060}
1061
1062static size_t print_prefix(const struct printk_log *msg, bool syslog, char *buf)
1063{
1064 size_t len = 0;
1065 unsigned int prefix = (msg->facility << 3) | msg->level;
1066
1067 if (syslog) {
1068 if (buf) {
1069 len += sprintf(buf, "<%u>", prefix);
1070 } else {
1071 len += 3;
1072 if (prefix > 999)
1073 len += 3;
1074 else if (prefix > 99)
1075 len += 2;
1076 else if (prefix > 9)
1077 len++;
1078 }
1079 }
1080
1081 len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
1082 return len;
1083}
1084
1085static size_t msg_print_text(const struct printk_log *msg, enum log_flags prev,
1086 bool syslog, char *buf, size_t size)
1087{
1088 const char *text = log_text(msg);
1089 size_t text_size = msg->text_len;
1090 bool prefix = true;
1091 bool newline = true;
1092 size_t len = 0;
1093
1094 if ((prev & LOG_CONT) && !(msg->flags & LOG_PREFIX))
1095 prefix = false;
1096
1097 if (msg->flags & LOG_CONT) {
1098 if ((prev & LOG_CONT) && !(prev & LOG_NEWLINE))
1099 prefix = false;
1100
1101 if (!(msg->flags & LOG_NEWLINE))
1102 newline = false;
1103 }
1104
1105 do {
1106 const char *next = memchr(text, '\n', text_size);
1107 size_t text_len;
1108
1109 if (next) {
1110 text_len = next - text;
1111 next++;
1112 text_size -= next - text;
1113 } else {
1114 text_len = text_size;
1115 }
1116
1117 if (buf) {
1118 if (print_prefix(msg, syslog, NULL) +
1119 text_len + 1 >= size - len)
1120 break;
1121
1122 if (prefix)
1123 len += print_prefix(msg, syslog, buf + len);
1124 memcpy(buf + len, text, text_len);
1125 len += text_len;
1126 if (next || newline)
1127 buf[len++] = '\n';
1128 } else {
1129 /* SYSLOG_ACTION_* buffer size only calculation */
1130 if (prefix)
1131 len += print_prefix(msg, syslog, NULL);
1132 len += text_len;
1133 if (next || newline)
1134 len++;
1135 }
1136
1137 prefix = true;
1138 text = next;
1139 } while (text);
1140
1141 return len;
1142}
1143
1144static int syslog_print(char __user *buf, int size)
1145{
1146 char *text;
1147 struct printk_log *msg;
1148 int len = 0;
1149
1150 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1151 if (!text)
1152 return -ENOMEM;
1153
1154 while (size > 0) {
1155 size_t n;
1156 size_t skip;
1157
1158 raw_spin_lock_irq(&logbuf_lock);
1159 if (syslog_seq < log_first_seq) {
1160 /* messages are gone, move to first one */
1161 syslog_seq = log_first_seq;
1162 syslog_idx = log_first_idx;
1163 syslog_prev = 0;
1164 syslog_partial = 0;
1165 }
1166 if (syslog_seq == log_next_seq) {
1167 raw_spin_unlock_irq(&logbuf_lock);
1168 break;
1169 }
1170
1171 skip = syslog_partial;
1172 msg = log_from_idx(syslog_idx);
1173 n = msg_print_text(msg, syslog_prev, true, text,
1174 LOG_LINE_MAX + PREFIX_MAX);
1175 if (n - syslog_partial <= size) {
1176 /* message fits into buffer, move forward */
1177 syslog_idx = log_next(syslog_idx);
1178 syslog_seq++;
1179 syslog_prev = msg->flags;
1180 n -= syslog_partial;
1181 syslog_partial = 0;
1182 } else if (!len){
1183 /* partial read(), remember position */
1184 n = size;
1185 syslog_partial += n;
1186 } else
1187 n = 0;
1188 raw_spin_unlock_irq(&logbuf_lock);
1189
1190 if (!n)
1191 break;
1192
1193 if (copy_to_user(buf, text + skip, n)) {
1194 if (!len)
1195 len = -EFAULT;
1196 break;
1197 }
1198
1199 len += n;
1200 size -= n;
1201 buf += n;
1202 }
1203
1204 kfree(text);
1205 return len;
1206}
1207
1208static int syslog_print_all(char __user *buf, int size, bool clear)
1209{
1210 char *text;
1211 int len = 0;
1212
1213 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1214 if (!text)
1215 return -ENOMEM;
1216
1217 raw_spin_lock_irq(&logbuf_lock);
1218 if (buf) {
1219 u64 next_seq;
1220 u64 seq;
1221 u32 idx;
1222 enum log_flags prev;
1223
1224 /*
1225 * Find first record that fits, including all following records,
1226 * into the user-provided buffer for this dump.
1227 */
1228 seq = clear_seq;
1229 idx = clear_idx;
1230 prev = 0;
1231 while (seq < log_next_seq) {
1232 struct printk_log *msg = log_from_idx(idx);
1233
1234 len += msg_print_text(msg, prev, true, NULL, 0);
1235 prev = msg->flags;
1236 idx = log_next(idx);
1237 seq++;
1238 }
1239
1240 /* move first record forward until length fits into the buffer */
1241 seq = clear_seq;
1242 idx = clear_idx;
1243 prev = 0;
1244 while (len > size && seq < log_next_seq) {
1245 struct printk_log *msg = log_from_idx(idx);
1246
1247 len -= msg_print_text(msg, prev, true, NULL, 0);
1248 prev = msg->flags;
1249 idx = log_next(idx);
1250 seq++;
1251 }
1252
1253 /* last message fitting into this dump */
1254 next_seq = log_next_seq;
1255
1256 len = 0;
1257 while (len >= 0 && seq < next_seq) {
1258 struct printk_log *msg = log_from_idx(idx);
1259 int textlen;
1260
1261 textlen = msg_print_text(msg, prev, true, text,
1262 LOG_LINE_MAX + PREFIX_MAX);
1263 if (textlen < 0) {
1264 len = textlen;
1265 break;
1266 }
1267 idx = log_next(idx);
1268 seq++;
1269 prev = msg->flags;
1270
1271 raw_spin_unlock_irq(&logbuf_lock);
1272 if (copy_to_user(buf + len, text, textlen))
1273 len = -EFAULT;
1274 else
1275 len += textlen;
1276 raw_spin_lock_irq(&logbuf_lock);
1277
1278 if (seq < log_first_seq) {
1279 /* messages are gone, move to next one */
1280 seq = log_first_seq;
1281 idx = log_first_idx;
1282 prev = 0;
1283 }
1284 }
1285 }
1286
1287 if (clear) {
1288 clear_seq = log_next_seq;
1289 clear_idx = log_next_idx;
1290 }
1291 raw_spin_unlock_irq(&logbuf_lock);
1292
1293 kfree(text);
1294 return len;
1295}
1296
1297int do_syslog(int type, char __user *buf, int len, int source)
1298{
1299 bool clear = false;
1300 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1301 int error;
1302
1303 error = check_syslog_permissions(type, source);
1304 if (error)
1305 goto out;
1306
1307 switch (type) {
1308 case SYSLOG_ACTION_CLOSE: /* Close log */
1309 break;
1310 case SYSLOG_ACTION_OPEN: /* Open log */
1311 break;
1312 case SYSLOG_ACTION_READ: /* Read from log */
1313 error = -EINVAL;
1314 if (!buf || len < 0)
1315 goto out;
1316 error = 0;
1317 if (!len)
1318 goto out;
1319 if (!access_ok(VERIFY_WRITE, buf, len)) {
1320 error = -EFAULT;
1321 goto out;
1322 }
1323 error = wait_event_interruptible(log_wait,
1324 syslog_seq != log_next_seq);
1325 if (error)
1326 goto out;
1327 error = syslog_print(buf, len);
1328 break;
1329 /* Read/clear last kernel messages */
1330 case SYSLOG_ACTION_READ_CLEAR:
1331 clear = true;
1332 /* FALL THRU */
1333 /* Read last kernel messages */
1334 case SYSLOG_ACTION_READ_ALL:
1335 error = -EINVAL;
1336 if (!buf || len < 0)
1337 goto out;
1338 error = 0;
1339 if (!len)
1340 goto out;
1341 if (!access_ok(VERIFY_WRITE, buf, len)) {
1342 error = -EFAULT;
1343 goto out;
1344 }
1345 error = syslog_print_all(buf, len, clear);
1346 break;
1347 /* Clear ring buffer */
1348 case SYSLOG_ACTION_CLEAR:
1349 syslog_print_all(NULL, 0, true);
1350 break;
1351 /* Disable logging to console */
1352 case SYSLOG_ACTION_CONSOLE_OFF:
1353 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1354 saved_console_loglevel = console_loglevel;
1355 console_loglevel = minimum_console_loglevel;
1356 break;
1357 /* Enable logging to console */
1358 case SYSLOG_ACTION_CONSOLE_ON:
1359 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1360 console_loglevel = saved_console_loglevel;
1361 saved_console_loglevel = LOGLEVEL_DEFAULT;
1362 }
1363 break;
1364 /* Set level of messages printed to console */
1365 case SYSLOG_ACTION_CONSOLE_LEVEL:
1366 error = -EINVAL;
1367 if (len < 1 || len > 8)
1368 goto out;
1369 if (len < minimum_console_loglevel)
1370 len = minimum_console_loglevel;
1371 console_loglevel = len;
1372 /* Implicitly re-enable logging to console */
1373 saved_console_loglevel = LOGLEVEL_DEFAULT;
1374 error = 0;
1375 break;
1376 /* Number of chars in the log buffer */
1377 case SYSLOG_ACTION_SIZE_UNREAD:
1378 raw_spin_lock_irq(&logbuf_lock);
1379 if (syslog_seq < log_first_seq) {
1380 /* messages are gone, move to first one */
1381 syslog_seq = log_first_seq;
1382 syslog_idx = log_first_idx;
1383 syslog_prev = 0;
1384 syslog_partial = 0;
1385 }
1386 if (source == SYSLOG_FROM_PROC) {
1387 /*
1388 * Short-cut for poll(/"proc/kmsg") which simply checks
1389 * for pending data, not the size; return the count of
1390 * records, not the length.
1391 */
1392 error = log_next_seq - syslog_seq;
1393 } else {
1394 u64 seq = syslog_seq;
1395 u32 idx = syslog_idx;
1396 enum log_flags prev = syslog_prev;
1397
1398 error = 0;
1399 while (seq < log_next_seq) {
1400 struct printk_log *msg = log_from_idx(idx);
1401
1402 error += msg_print_text(msg, prev, true, NULL, 0);
1403 idx = log_next(idx);
1404 seq++;
1405 prev = msg->flags;
1406 }
1407 error -= syslog_partial;
1408 }
1409 raw_spin_unlock_irq(&logbuf_lock);
1410 break;
1411 /* Size of the log buffer */
1412 case SYSLOG_ACTION_SIZE_BUFFER:
1413 error = log_buf_len;
1414 break;
1415 default:
1416 error = -EINVAL;
1417 break;
1418 }
1419out:
1420 return error;
1421}
1422
1423SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1424{
1425 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1426}
1427
1428/*
1429 * Call the console drivers, asking them to write out
1430 * log_buf[start] to log_buf[end - 1].
1431 * The console_lock must be held.
1432 */
1433static void call_console_drivers(int level,
1434 const char *ext_text, size_t ext_len,
1435 const char *text, size_t len)
1436{
1437 struct console *con;
1438
1439 trace_console(text, len);
1440
1441 if (level >= console_loglevel && !ignore_loglevel)
1442 return;
1443 if (!console_drivers)
1444 return;
1445
1446 for_each_console(con) {
1447 if (exclusive_console && con != exclusive_console)
1448 continue;
1449 if (!(con->flags & CON_ENABLED))
1450 continue;
1451 if (!con->write)
1452 continue;
1453 if (!cpu_online(smp_processor_id()) &&
1454 !(con->flags & CON_ANYTIME))
1455 continue;
1456 if (con->flags & CON_EXTENDED)
1457 con->write(con, ext_text, ext_len);
1458 else
1459 con->write(con, text, len);
1460 }
1461}
1462
1463/*
1464 * Zap console related locks when oopsing.
1465 * To leave time for slow consoles to print a full oops,
1466 * only zap at most once every 30 seconds.
1467 */
1468static void zap_locks(void)
1469{
1470 static unsigned long oops_timestamp;
1471
1472 if (time_after_eq(jiffies, oops_timestamp) &&
1473 !time_after(jiffies, oops_timestamp + 30 * HZ))
1474 return;
1475
1476 oops_timestamp = jiffies;
1477
1478 debug_locks_off();
1479 /* If a crash is occurring, make sure we can't deadlock */
1480 raw_spin_lock_init(&logbuf_lock);
1481 /* And make sure that we print immediately */
1482 sema_init(&console_sem, 1);
1483}
1484
1485int printk_delay_msec __read_mostly;
1486
1487static inline void printk_delay(void)
1488{
1489 if (unlikely(printk_delay_msec)) {
1490 int m = printk_delay_msec;
1491
1492 while (m--) {
1493 mdelay(1);
1494 touch_nmi_watchdog();
1495 }
1496 }
1497}
1498
1499/*
1500 * Continuation lines are buffered, and not committed to the record buffer
1501 * until the line is complete, or a race forces it. The line fragments
1502 * though, are printed immediately to the consoles to ensure everything has
1503 * reached the console in case of a kernel crash.
1504 */
1505static struct cont {
1506 char buf[LOG_LINE_MAX];
1507 size_t len; /* length == 0 means unused buffer */
1508 size_t cons; /* bytes written to console */
1509 struct task_struct *owner; /* task of first print*/
1510 u64 ts_nsec; /* time of first print */
1511 u8 level; /* log level of first message */
1512 u8 facility; /* log facility of first message */
1513 enum log_flags flags; /* prefix, newline flags */
1514 bool flushed:1; /* buffer sealed and committed */
1515} cont;
1516
1517static void cont_flush(enum log_flags flags)
1518{
1519 if (cont.flushed)
1520 return;
1521 if (cont.len == 0)
1522 return;
1523
1524 if (cont.cons) {
1525 /*
1526 * If a fragment of this line was directly flushed to the
1527 * console; wait for the console to pick up the rest of the
1528 * line. LOG_NOCONS suppresses a duplicated output.
1529 */
1530 log_store(cont.facility, cont.level, flags | LOG_NOCONS,
1531 cont.ts_nsec, NULL, 0, cont.buf, cont.len);
1532 cont.flags = flags;
1533 cont.flushed = true;
1534 } else {
1535 /*
1536 * If no fragment of this line ever reached the console,
1537 * just submit it to the store and free the buffer.
1538 */
1539 log_store(cont.facility, cont.level, flags, 0,
1540 NULL, 0, cont.buf, cont.len);
1541 cont.len = 0;
1542 }
1543}
1544
1545static bool cont_add(int facility, int level, const char *text, size_t len)
1546{
1547 if (cont.len && cont.flushed)
1548 return false;
1549
1550 /*
1551 * If ext consoles are present, flush and skip in-kernel
1552 * continuation. See nr_ext_console_drivers definition. Also, if
1553 * the line gets too long, split it up in separate records.
1554 */
1555 if (nr_ext_console_drivers || cont.len + len > sizeof(cont.buf)) {
1556 cont_flush(LOG_CONT);
1557 return false;
1558 }
1559
1560 if (!cont.len) {
1561 cont.facility = facility;
1562 cont.level = level;
1563 cont.owner = current;
1564 cont.ts_nsec = local_clock();
1565 cont.flags = 0;
1566 cont.cons = 0;
1567 cont.flushed = false;
1568 }
1569
1570 memcpy(cont.buf + cont.len, text, len);
1571 cont.len += len;
1572
1573 if (cont.len > (sizeof(cont.buf) * 80) / 100)
1574 cont_flush(LOG_CONT);
1575
1576 return true;
1577}
1578
1579static size_t cont_print_text(char *text, size_t size)
1580{
1581 size_t textlen = 0;
1582 size_t len;
1583
1584 if (cont.cons == 0 && (console_prev & LOG_NEWLINE)) {
1585 textlen += print_time(cont.ts_nsec, text);
1586 size -= textlen;
1587 }
1588
1589 len = cont.len - cont.cons;
1590 if (len > 0) {
1591 if (len+1 > size)
1592 len = size-1;
1593 memcpy(text + textlen, cont.buf + cont.cons, len);
1594 textlen += len;
1595 cont.cons = cont.len;
1596 }
1597
1598 if (cont.flushed) {
1599 if (cont.flags & LOG_NEWLINE)
1600 text[textlen++] = '\n';
1601 /* got everything, release buffer */
1602 cont.len = 0;
1603 }
1604 return textlen;
1605}
1606
1607asmlinkage int vprintk_emit(int facility, int level,
1608 const char *dict, size_t dictlen,
1609 const char *fmt, va_list args)
1610{
1611 static bool recursion_bug;
1612 static char textbuf[LOG_LINE_MAX];
1613 char *text = textbuf;
1614 size_t text_len = 0;
1615 enum log_flags lflags = 0;
1616 unsigned long flags;
1617 int this_cpu;
1618 int printed_len = 0;
1619 bool in_sched = false;
1620 /* cpu currently holding logbuf_lock in this function */
1621 static unsigned int logbuf_cpu = UINT_MAX;
1622
1623 if (level == LOGLEVEL_SCHED) {
1624 level = LOGLEVEL_DEFAULT;
1625 in_sched = true;
1626 }
1627
1628 boot_delay_msec(level);
1629 printk_delay();
1630
1631 local_irq_save(flags);
1632 this_cpu = smp_processor_id();
1633
1634 /*
1635 * Ouch, printk recursed into itself!
1636 */
1637 if (unlikely(logbuf_cpu == this_cpu)) {
1638 /*
1639 * If a crash is occurring during printk() on this CPU,
1640 * then try to get the crash message out but make sure
1641 * we can't deadlock. Otherwise just return to avoid the
1642 * recursion and return - but flag the recursion so that
1643 * it can be printed at the next appropriate moment:
1644 */
1645 if (!oops_in_progress && !lockdep_recursing(current)) {
1646 recursion_bug = true;
1647 local_irq_restore(flags);
1648 return 0;
1649 }
1650 zap_locks();
1651 }
1652
1653 lockdep_off();
1654 /* This stops the holder of console_sem just where we want him */
1655 raw_spin_lock(&logbuf_lock);
1656 logbuf_cpu = this_cpu;
1657
1658 if (unlikely(recursion_bug)) {
1659 static const char recursion_msg[] =
1660 "BUG: recent printk recursion!";
1661
1662 recursion_bug = false;
1663 /* emit KERN_CRIT message */
1664 printed_len += log_store(0, 2, LOG_PREFIX|LOG_NEWLINE, 0,
1665 NULL, 0, recursion_msg,
1666 strlen(recursion_msg));
1667 }
1668
1669 /*
1670 * The printf needs to come first; we need the syslog
1671 * prefix which might be passed-in as a parameter.
1672 */
1673 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1674
1675 /* mark and strip a trailing newline */
1676 if (text_len && text[text_len-1] == '\n') {
1677 text_len--;
1678 lflags |= LOG_NEWLINE;
1679 }
1680
1681 /* strip kernel syslog prefix and extract log level or control flags */
1682 if (facility == 0) {
1683 int kern_level = printk_get_level(text);
1684
1685 if (kern_level) {
1686 const char *end_of_header = printk_skip_level(text);
1687 switch (kern_level) {
1688 case '0' ... '7':
1689 if (level == LOGLEVEL_DEFAULT)
1690 level = kern_level - '0';
1691 /* fallthrough */
1692 case 'd': /* KERN_DEFAULT */
1693 lflags |= LOG_PREFIX;
1694 }
1695 /*
1696 * No need to check length here because vscnprintf
1697 * put '\0' at the end of the string. Only valid and
1698 * newly printed level is detected.
1699 */
1700 text_len -= end_of_header - text;
1701 text = (char *)end_of_header;
1702 }
1703 }
1704
1705 if (level == LOGLEVEL_DEFAULT)
1706 level = default_message_loglevel;
1707
1708 if (dict)
1709 lflags |= LOG_PREFIX|LOG_NEWLINE;
1710
1711 if (!(lflags & LOG_NEWLINE)) {
1712 /*
1713 * Flush the conflicting buffer. An earlier newline was missing,
1714 * or another task also prints continuation lines.
1715 */
1716 if (cont.len && (lflags & LOG_PREFIX || cont.owner != current))
1717 cont_flush(LOG_NEWLINE);
1718
1719 /* buffer line if possible, otherwise store it right away */
1720 if (cont_add(facility, level, text, text_len))
1721 printed_len += text_len;
1722 else
1723 printed_len += log_store(facility, level,
1724 lflags | LOG_CONT, 0,
1725 dict, dictlen, text, text_len);
1726 } else {
1727 bool stored = false;
1728
1729 /*
1730 * If an earlier newline was missing and it was the same task,
1731 * either merge it with the current buffer and flush, or if
1732 * there was a race with interrupts (prefix == true) then just
1733 * flush it out and store this line separately.
1734 * If the preceding printk was from a different task and missed
1735 * a newline, flush and append the newline.
1736 */
1737 if (cont.len) {
1738 if (cont.owner == current && !(lflags & LOG_PREFIX))
1739 stored = cont_add(facility, level, text,
1740 text_len);
1741 cont_flush(LOG_NEWLINE);
1742 }
1743
1744 if (stored)
1745 printed_len += text_len;
1746 else
1747 printed_len += log_store(facility, level, lflags, 0,
1748 dict, dictlen, text, text_len);
1749 }
1750
1751 logbuf_cpu = UINT_MAX;
1752 raw_spin_unlock(&logbuf_lock);
1753 lockdep_on();
1754 local_irq_restore(flags);
1755
1756 /* If called from the scheduler, we can not call up(). */
1757 if (!in_sched) {
1758 lockdep_off();
1759 /*
1760 * Try to acquire and then immediately release the console
1761 * semaphore. The release will print out buffers and wake up
1762 * /dev/kmsg and syslog() users.
1763 */
1764 if (console_trylock())
1765 console_unlock();
1766 lockdep_on();
1767 }
1768
1769 return printed_len;
1770}
1771EXPORT_SYMBOL(vprintk_emit);
1772
1773asmlinkage int vprintk(const char *fmt, va_list args)
1774{
1775 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1776}
1777EXPORT_SYMBOL(vprintk);
1778
1779asmlinkage int printk_emit(int facility, int level,
1780 const char *dict, size_t dictlen,
1781 const char *fmt, ...)
1782{
1783 va_list args;
1784 int r;
1785
1786 va_start(args, fmt);
1787 r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1788 va_end(args);
1789
1790 return r;
1791}
1792EXPORT_SYMBOL(printk_emit);
1793
1794int vprintk_default(const char *fmt, va_list args)
1795{
1796 int r;
1797
1798#ifdef CONFIG_KGDB_KDB
1799 if (unlikely(kdb_trap_printk)) {
1800 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
1801 return r;
1802 }
1803#endif
1804 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1805
1806 return r;
1807}
1808EXPORT_SYMBOL_GPL(vprintk_default);
1809
1810/*
1811 * This allows printk to be diverted to another function per cpu.
1812 * This is useful for calling printk functions from within NMI
1813 * without worrying about race conditions that can lock up the
1814 * box.
1815 */
1816DEFINE_PER_CPU(printk_func_t, printk_func) = vprintk_default;
1817
1818/**
1819 * printk - print a kernel message
1820 * @fmt: format string
1821 *
1822 * This is printk(). It can be called from any context. We want it to work.
1823 *
1824 * We try to grab the console_lock. If we succeed, it's easy - we log the
1825 * output and call the console drivers. If we fail to get the semaphore, we
1826 * place the output into the log buffer and return. The current holder of
1827 * the console_sem will notice the new output in console_unlock(); and will
1828 * send it to the consoles before releasing the lock.
1829 *
1830 * One effect of this deferred printing is that code which calls printk() and
1831 * then changes console_loglevel may break. This is because console_loglevel
1832 * is inspected when the actual printing occurs.
1833 *
1834 * See also:
1835 * printf(3)
1836 *
1837 * See the vsnprintf() documentation for format string extensions over C99.
1838 */
1839asmlinkage __visible int printk(const char *fmt, ...)
1840{
1841 printk_func_t vprintk_func;
1842 va_list args;
1843 int r;
1844
1845 va_start(args, fmt);
1846
1847 /*
1848 * If a caller overrides the per_cpu printk_func, then it needs
1849 * to disable preemption when calling printk(). Otherwise
1850 * the printk_func should be set to the default. No need to
1851 * disable preemption here.
1852 */
1853 vprintk_func = this_cpu_read(printk_func);
1854 r = vprintk_func(fmt, args);
1855
1856 va_end(args);
1857
1858 return r;
1859}
1860EXPORT_SYMBOL(printk);
1861
1862#else /* CONFIG_PRINTK */
1863
1864#define LOG_LINE_MAX 0
1865#define PREFIX_MAX 0
1866
1867static u64 syslog_seq;
1868static u32 syslog_idx;
1869static u64 console_seq;
1870static u32 console_idx;
1871static enum log_flags syslog_prev;
1872static u64 log_first_seq;
1873static u32 log_first_idx;
1874static u64 log_next_seq;
1875static enum log_flags console_prev;
1876static struct cont {
1877 size_t len;
1878 size_t cons;
1879 u8 level;
1880 bool flushed:1;
1881} cont;
1882static char *log_text(const struct printk_log *msg) { return NULL; }
1883static char *log_dict(const struct printk_log *msg) { return NULL; }
1884static struct printk_log *log_from_idx(u32 idx) { return NULL; }
1885static u32 log_next(u32 idx) { return 0; }
1886static ssize_t msg_print_ext_header(char *buf, size_t size,
1887 struct printk_log *msg, u64 seq,
1888 enum log_flags prev_flags) { return 0; }
1889static ssize_t msg_print_ext_body(char *buf, size_t size,
1890 char *dict, size_t dict_len,
1891 char *text, size_t text_len) { return 0; }
1892static void call_console_drivers(int level,
1893 const char *ext_text, size_t ext_len,
1894 const char *text, size_t len) {}
1895static size_t msg_print_text(const struct printk_log *msg, enum log_flags prev,
1896 bool syslog, char *buf, size_t size) { return 0; }
1897static size_t cont_print_text(char *text, size_t size) { return 0; }
1898
1899/* Still needs to be defined for users */
1900DEFINE_PER_CPU(printk_func_t, printk_func);
1901
1902#endif /* CONFIG_PRINTK */
1903
1904#ifdef CONFIG_EARLY_PRINTK
1905struct console *early_console;
1906
1907asmlinkage __visible void early_printk(const char *fmt, ...)
1908{
1909 va_list ap;
1910 char buf[512];
1911 int n;
1912
1913 if (!early_console)
1914 return;
1915
1916 va_start(ap, fmt);
1917 n = vscnprintf(buf, sizeof(buf), fmt, ap);
1918 va_end(ap);
1919
1920 early_console->write(early_console, buf, n);
1921}
1922#endif
1923
1924static int __add_preferred_console(char *name, int idx, char *options,
1925 char *brl_options)
1926{
1927 struct console_cmdline *c;
1928 int i;
1929
1930 /*
1931 * See if this tty is not yet registered, and
1932 * if we have a slot free.
1933 */
1934 for (i = 0, c = console_cmdline;
1935 i < MAX_CMDLINECONSOLES && c->name[0];
1936 i++, c++) {
1937 if (strcmp(c->name, name) == 0 && c->index == idx) {
1938 if (!brl_options)
1939 selected_console = i;
1940 return 0;
1941 }
1942 }
1943 if (i == MAX_CMDLINECONSOLES)
1944 return -E2BIG;
1945 if (!brl_options)
1946 selected_console = i;
1947 strlcpy(c->name, name, sizeof(c->name));
1948 c->options = options;
1949 braille_set_options(c, brl_options);
1950
1951 c->index = idx;
1952 return 0;
1953}
1954/*
1955 * Set up a console. Called via do_early_param() in init/main.c
1956 * for each "console=" parameter in the boot command line.
1957 */
1958static int __init console_setup(char *str)
1959{
1960 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
1961 char *s, *options, *brl_options = NULL;
1962 int idx;
1963
1964 if (_braille_console_setup(&str, &brl_options))
1965 return 1;
1966
1967 /*
1968 * Decode str into name, index, options.
1969 */
1970 if (str[0] >= '0' && str[0] <= '9') {
1971 strcpy(buf, "ttyS");
1972 strncpy(buf + 4, str, sizeof(buf) - 5);
1973 } else {
1974 strncpy(buf, str, sizeof(buf) - 1);
1975 }
1976 buf[sizeof(buf) - 1] = 0;
1977 options = strchr(str, ',');
1978 if (options)
1979 *(options++) = 0;
1980#ifdef __sparc__
1981 if (!strcmp(str, "ttya"))
1982 strcpy(buf, "ttyS0");
1983 if (!strcmp(str, "ttyb"))
1984 strcpy(buf, "ttyS1");
1985#endif
1986 for (s = buf; *s; s++)
1987 if (isdigit(*s) || *s == ',')
1988 break;
1989 idx = simple_strtoul(s, NULL, 10);
1990 *s = 0;
1991
1992 __add_preferred_console(buf, idx, options, brl_options);
1993 console_set_on_cmdline = 1;
1994 return 1;
1995}
1996__setup("console=", console_setup);
1997
1998/**
1999 * add_preferred_console - add a device to the list of preferred consoles.
2000 * @name: device name
2001 * @idx: device index
2002 * @options: options for this console
2003 *
2004 * The last preferred console added will be used for kernel messages
2005 * and stdin/out/err for init. Normally this is used by console_setup
2006 * above to handle user-supplied console arguments; however it can also
2007 * be used by arch-specific code either to override the user or more
2008 * commonly to provide a default console (ie from PROM variables) when
2009 * the user has not supplied one.
2010 */
2011int add_preferred_console(char *name, int idx, char *options)
2012{
2013 return __add_preferred_console(name, idx, options, NULL);
2014}
2015
2016bool console_suspend_enabled = true;
2017EXPORT_SYMBOL(console_suspend_enabled);
2018
2019static int __init console_suspend_disable(char *str)
2020{
2021 console_suspend_enabled = false;
2022 return 1;
2023}
2024__setup("no_console_suspend", console_suspend_disable);
2025module_param_named(console_suspend, console_suspend_enabled,
2026 bool, S_IRUGO | S_IWUSR);
2027MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2028 " and hibernate operations");
2029
2030/**
2031 * suspend_console - suspend the console subsystem
2032 *
2033 * This disables printk() while we go into suspend states
2034 */
2035void suspend_console(void)
2036{
2037 if (!console_suspend_enabled)
2038 return;
2039 printk("Suspending console(s) (use no_console_suspend to debug)\n");
2040 console_lock();
2041 console_suspended = 1;
2042 up_console_sem();
2043}
2044
2045void resume_console(void)
2046{
2047 if (!console_suspend_enabled)
2048 return;
2049 down_console_sem();
2050 console_suspended = 0;
2051 console_unlock();
2052}
2053
2054/**
2055 * console_cpu_notify - print deferred console messages after CPU hotplug
2056 * @self: notifier struct
2057 * @action: CPU hotplug event
2058 * @hcpu: unused
2059 *
2060 * If printk() is called from a CPU that is not online yet, the messages
2061 * will be spooled but will not show up on the console. This function is
2062 * called when a new CPU comes online (or fails to come up), and ensures
2063 * that any such output gets printed.
2064 */
2065static int console_cpu_notify(struct notifier_block *self,
2066 unsigned long action, void *hcpu)
2067{
2068 switch (action) {
2069 case CPU_ONLINE:
2070 case CPU_DEAD:
2071 case CPU_DOWN_FAILED:
2072 case CPU_UP_CANCELED:
2073 console_lock();
2074 console_unlock();
2075 }
2076 return NOTIFY_OK;
2077}
2078
2079/**
2080 * console_lock - lock the console system for exclusive use.
2081 *
2082 * Acquires a lock which guarantees that the caller has
2083 * exclusive access to the console system and the console_drivers list.
2084 *
2085 * Can sleep, returns nothing.
2086 */
2087void console_lock(void)
2088{
2089 might_sleep();
2090
2091 down_console_sem();
2092 if (console_suspended)
2093 return;
2094 console_locked = 1;
2095 console_may_schedule = 1;
2096}
2097EXPORT_SYMBOL(console_lock);
2098
2099/**
2100 * console_trylock - try to lock the console system for exclusive use.
2101 *
2102 * Try to acquire a lock which guarantees that the caller has exclusive
2103 * access to the console system and the console_drivers list.
2104 *
2105 * returns 1 on success, and 0 on failure to acquire the lock.
2106 */
2107int console_trylock(void)
2108{
2109 if (down_trylock_console_sem())
2110 return 0;
2111 if (console_suspended) {
2112 up_console_sem();
2113 return 0;
2114 }
2115 console_locked = 1;
2116 /*
2117 * When PREEMPT_COUNT disabled we can't reliably detect if it's
2118 * safe to schedule (e.g. calling printk while holding a spin_lock),
2119 * because preempt_disable()/preempt_enable() are just barriers there
2120 * and preempt_count() is always 0.
2121 *
2122 * RCU read sections have a separate preemption counter when
2123 * PREEMPT_RCU enabled thus we must take extra care and check
2124 * rcu_preempt_depth(), otherwise RCU read sections modify
2125 * preempt_count().
2126 */
2127 console_may_schedule = !oops_in_progress &&
2128 preemptible() &&
2129 !rcu_preempt_depth();
2130 return 1;
2131}
2132EXPORT_SYMBOL(console_trylock);
2133
2134int is_console_locked(void)
2135{
2136 return console_locked;
2137}
2138
2139/*
2140 * Check if we have any console that is capable of printing while cpu is
2141 * booting or shutting down. Requires console_sem.
2142 */
2143static int have_callable_console(void)
2144{
2145 struct console *con;
2146
2147 for_each_console(con)
2148 if ((con->flags & CON_ENABLED) &&
2149 (con->flags & CON_ANYTIME))
2150 return 1;
2151
2152 return 0;
2153}
2154
2155/*
2156 * Can we actually use the console at this time on this cpu?
2157 *
2158 * Console drivers may assume that per-cpu resources have been allocated. So
2159 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2160 * call them until this CPU is officially up.
2161 */
2162static inline int can_use_console(void)
2163{
2164 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2165}
2166
2167static void console_cont_flush(char *text, size_t size)
2168{
2169 unsigned long flags;
2170 size_t len;
2171
2172 raw_spin_lock_irqsave(&logbuf_lock, flags);
2173
2174 if (!cont.len)
2175 goto out;
2176
2177 /*
2178 * We still queue earlier records, likely because the console was
2179 * busy. The earlier ones need to be printed before this one, we
2180 * did not flush any fragment so far, so just let it queue up.
2181 */
2182 if (console_seq < log_next_seq && !cont.cons)
2183 goto out;
2184
2185 len = cont_print_text(text, size);
2186 raw_spin_unlock(&logbuf_lock);
2187 stop_critical_timings();
2188 call_console_drivers(cont.level, NULL, 0, text, len);
2189 start_critical_timings();
2190 local_irq_restore(flags);
2191 return;
2192out:
2193 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2194}
2195
2196/**
2197 * console_unlock - unlock the console system
2198 *
2199 * Releases the console_lock which the caller holds on the console system
2200 * and the console driver list.
2201 *
2202 * While the console_lock was held, console output may have been buffered
2203 * by printk(). If this is the case, console_unlock(); emits
2204 * the output prior to releasing the lock.
2205 *
2206 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2207 *
2208 * console_unlock(); may be called from any context.
2209 */
2210void console_unlock(void)
2211{
2212 static char ext_text[CONSOLE_EXT_LOG_MAX];
2213 static char text[LOG_LINE_MAX + PREFIX_MAX];
2214 static u64 seen_seq;
2215 unsigned long flags;
2216 bool wake_klogd = false;
2217 bool do_cond_resched, retry;
2218
2219 if (console_suspended) {
2220 up_console_sem();
2221 return;
2222 }
2223
2224 /*
2225 * Console drivers are called under logbuf_lock, so
2226 * @console_may_schedule should be cleared before; however, we may
2227 * end up dumping a lot of lines, for example, if called from
2228 * console registration path, and should invoke cond_resched()
2229 * between lines if allowable. Not doing so can cause a very long
2230 * scheduling stall on a slow console leading to RCU stall and
2231 * softlockup warnings which exacerbate the issue with more
2232 * messages practically incapacitating the system.
2233 */
2234 do_cond_resched = console_may_schedule;
2235 console_may_schedule = 0;
2236
2237again:
2238 /*
2239 * We released the console_sem lock, so we need to recheck if
2240 * cpu is online and (if not) is there at least one CON_ANYTIME
2241 * console.
2242 */
2243 if (!can_use_console()) {
2244 console_locked = 0;
2245 up_console_sem();
2246 return;
2247 }
2248
2249 /* flush buffered message fragment immediately to console */
2250 console_cont_flush(text, sizeof(text));
2251
2252 for (;;) {
2253 struct printk_log *msg;
2254 size_t ext_len = 0;
2255 size_t len;
2256 int level;
2257
2258 raw_spin_lock_irqsave(&logbuf_lock, flags);
2259 if (seen_seq != log_next_seq) {
2260 wake_klogd = true;
2261 seen_seq = log_next_seq;
2262 }
2263
2264 if (console_seq < log_first_seq) {
2265 len = sprintf(text, "** %u printk messages dropped ** ",
2266 (unsigned)(log_first_seq - console_seq));
2267
2268 /* messages are gone, move to first one */
2269 console_seq = log_first_seq;
2270 console_idx = log_first_idx;
2271 console_prev = 0;
2272 } else {
2273 len = 0;
2274 }
2275skip:
2276 if (console_seq == log_next_seq)
2277 break;
2278
2279 msg = log_from_idx(console_idx);
2280 if (msg->flags & LOG_NOCONS) {
2281 /*
2282 * Skip record we have buffered and already printed
2283 * directly to the console when we received it.
2284 */
2285 console_idx = log_next(console_idx);
2286 console_seq++;
2287 /*
2288 * We will get here again when we register a new
2289 * CON_PRINTBUFFER console. Clear the flag so we
2290 * will properly dump everything later.
2291 */
2292 msg->flags &= ~LOG_NOCONS;
2293 console_prev = msg->flags;
2294 goto skip;
2295 }
2296
2297 level = msg->level;
2298 len += msg_print_text(msg, console_prev, false,
2299 text + len, sizeof(text) - len);
2300 if (nr_ext_console_drivers) {
2301 ext_len = msg_print_ext_header(ext_text,
2302 sizeof(ext_text),
2303 msg, console_seq, console_prev);
2304 ext_len += msg_print_ext_body(ext_text + ext_len,
2305 sizeof(ext_text) - ext_len,
2306 log_dict(msg), msg->dict_len,
2307 log_text(msg), msg->text_len);
2308 }
2309 console_idx = log_next(console_idx);
2310 console_seq++;
2311 console_prev = msg->flags;
2312 raw_spin_unlock(&logbuf_lock);
2313
2314 stop_critical_timings(); /* don't trace print latency */
2315 call_console_drivers(level, ext_text, ext_len, text, len);
2316 start_critical_timings();
2317 local_irq_restore(flags);
2318
2319 if (do_cond_resched)
2320 cond_resched();
2321 }
2322 console_locked = 0;
2323
2324 /* Release the exclusive_console once it is used */
2325 if (unlikely(exclusive_console))
2326 exclusive_console = NULL;
2327
2328 raw_spin_unlock(&logbuf_lock);
2329
2330 up_console_sem();
2331
2332 /*
2333 * Someone could have filled up the buffer again, so re-check if there's
2334 * something to flush. In case we cannot trylock the console_sem again,
2335 * there's a new owner and the console_unlock() from them will do the
2336 * flush, no worries.
2337 */
2338 raw_spin_lock(&logbuf_lock);
2339 retry = console_seq != log_next_seq;
2340 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2341
2342 if (retry && console_trylock())
2343 goto again;
2344
2345 if (wake_klogd)
2346 wake_up_klogd();
2347}
2348EXPORT_SYMBOL(console_unlock);
2349
2350/**
2351 * console_conditional_schedule - yield the CPU if required
2352 *
2353 * If the console code is currently allowed to sleep, and
2354 * if this CPU should yield the CPU to another task, do
2355 * so here.
2356 *
2357 * Must be called within console_lock();.
2358 */
2359void __sched console_conditional_schedule(void)
2360{
2361 if (console_may_schedule)
2362 cond_resched();
2363}
2364EXPORT_SYMBOL(console_conditional_schedule);
2365
2366void console_unblank(void)
2367{
2368 struct console *c;
2369
2370 /*
2371 * console_unblank can no longer be called in interrupt context unless
2372 * oops_in_progress is set to 1..
2373 */
2374 if (oops_in_progress) {
2375 if (down_trylock_console_sem() != 0)
2376 return;
2377 } else
2378 console_lock();
2379
2380 console_locked = 1;
2381 console_may_schedule = 0;
2382 for_each_console(c)
2383 if ((c->flags & CON_ENABLED) && c->unblank)
2384 c->unblank();
2385 console_unlock();
2386}
2387
2388/**
2389 * console_flush_on_panic - flush console content on panic
2390 *
2391 * Immediately output all pending messages no matter what.
2392 */
2393void console_flush_on_panic(void)
2394{
2395 /*
2396 * If someone else is holding the console lock, trylock will fail
2397 * and may_schedule may be set. Ignore and proceed to unlock so
2398 * that messages are flushed out. As this can be called from any
2399 * context and we don't want to get preempted while flushing,
2400 * ensure may_schedule is cleared.
2401 */
2402 console_trylock();
2403 console_may_schedule = 0;
2404 console_unlock();
2405}
2406
2407/*
2408 * Return the console tty driver structure and its associated index
2409 */
2410struct tty_driver *console_device(int *index)
2411{
2412 struct console *c;
2413 struct tty_driver *driver = NULL;
2414
2415 console_lock();
2416 for_each_console(c) {
2417 if (!c->device)
2418 continue;
2419 driver = c->device(c, index);
2420 if (driver)
2421 break;
2422 }
2423 console_unlock();
2424 return driver;
2425}
2426
2427/*
2428 * Prevent further output on the passed console device so that (for example)
2429 * serial drivers can disable console output before suspending a port, and can
2430 * re-enable output afterwards.
2431 */
2432void console_stop(struct console *console)
2433{
2434 console_lock();
2435 console->flags &= ~CON_ENABLED;
2436 console_unlock();
2437}
2438EXPORT_SYMBOL(console_stop);
2439
2440void console_start(struct console *console)
2441{
2442 console_lock();
2443 console->flags |= CON_ENABLED;
2444 console_unlock();
2445}
2446EXPORT_SYMBOL(console_start);
2447
2448static int __read_mostly keep_bootcon;
2449
2450static int __init keep_bootcon_setup(char *str)
2451{
2452 keep_bootcon = 1;
2453 pr_info("debug: skip boot console de-registration.\n");
2454
2455 return 0;
2456}
2457
2458early_param("keep_bootcon", keep_bootcon_setup);
2459
2460/*
2461 * The console driver calls this routine during kernel initialization
2462 * to register the console printing procedure with printk() and to
2463 * print any messages that were printed by the kernel before the
2464 * console driver was initialized.
2465 *
2466 * This can happen pretty early during the boot process (because of
2467 * early_printk) - sometimes before setup_arch() completes - be careful
2468 * of what kernel features are used - they may not be initialised yet.
2469 *
2470 * There are two types of consoles - bootconsoles (early_printk) and
2471 * "real" consoles (everything which is not a bootconsole) which are
2472 * handled differently.
2473 * - Any number of bootconsoles can be registered at any time.
2474 * - As soon as a "real" console is registered, all bootconsoles
2475 * will be unregistered automatically.
2476 * - Once a "real" console is registered, any attempt to register a
2477 * bootconsoles will be rejected
2478 */
2479void register_console(struct console *newcon)
2480{
2481 int i;
2482 unsigned long flags;
2483 struct console *bcon = NULL;
2484 struct console_cmdline *c;
2485
2486 if (console_drivers)
2487 for_each_console(bcon)
2488 if (WARN(bcon == newcon,
2489 "console '%s%d' already registered\n",
2490 bcon->name, bcon->index))
2491 return;
2492
2493 /*
2494 * before we register a new CON_BOOT console, make sure we don't
2495 * already have a valid console
2496 */
2497 if (console_drivers && newcon->flags & CON_BOOT) {
2498 /* find the last or real console */
2499 for_each_console(bcon) {
2500 if (!(bcon->flags & CON_BOOT)) {
2501 pr_info("Too late to register bootconsole %s%d\n",
2502 newcon->name, newcon->index);
2503 return;
2504 }
2505 }
2506 }
2507
2508 if (console_drivers && console_drivers->flags & CON_BOOT)
2509 bcon = console_drivers;
2510
2511 if (preferred_console < 0 || bcon || !console_drivers)
2512 preferred_console = selected_console;
2513
2514 /*
2515 * See if we want to use this console driver. If we
2516 * didn't select a console we take the first one
2517 * that registers here.
2518 */
2519 if (preferred_console < 0) {
2520 if (newcon->index < 0)
2521 newcon->index = 0;
2522 if (newcon->setup == NULL ||
2523 newcon->setup(newcon, NULL) == 0) {
2524 newcon->flags |= CON_ENABLED;
2525 if (newcon->device) {
2526 newcon->flags |= CON_CONSDEV;
2527 preferred_console = 0;
2528 }
2529 }
2530 }
2531
2532 /*
2533 * See if this console matches one we selected on
2534 * the command line.
2535 */
2536 for (i = 0, c = console_cmdline;
2537 i < MAX_CMDLINECONSOLES && c->name[0];
2538 i++, c++) {
2539 if (!newcon->match ||
2540 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2541 /* default matching */
2542 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2543 if (strcmp(c->name, newcon->name) != 0)
2544 continue;
2545 if (newcon->index >= 0 &&
2546 newcon->index != c->index)
2547 continue;
2548 if (newcon->index < 0)
2549 newcon->index = c->index;
2550
2551 if (_braille_register_console(newcon, c))
2552 return;
2553
2554 if (newcon->setup &&
2555 newcon->setup(newcon, c->options) != 0)
2556 break;
2557 }
2558
2559 newcon->flags |= CON_ENABLED;
2560 if (i == selected_console) {
2561 newcon->flags |= CON_CONSDEV;
2562 preferred_console = selected_console;
2563 }
2564 break;
2565 }
2566
2567 if (!(newcon->flags & CON_ENABLED))
2568 return;
2569
2570 /*
2571 * If we have a bootconsole, and are switching to a real console,
2572 * don't print everything out again, since when the boot console, and
2573 * the real console are the same physical device, it's annoying to
2574 * see the beginning boot messages twice
2575 */
2576 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2577 newcon->flags &= ~CON_PRINTBUFFER;
2578
2579 /*
2580 * Put this console in the list - keep the
2581 * preferred driver at the head of the list.
2582 */
2583 console_lock();
2584 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2585 newcon->next = console_drivers;
2586 console_drivers = newcon;
2587 if (newcon->next)
2588 newcon->next->flags &= ~CON_CONSDEV;
2589 } else {
2590 newcon->next = console_drivers->next;
2591 console_drivers->next = newcon;
2592 }
2593
2594 if (newcon->flags & CON_EXTENDED)
2595 if (!nr_ext_console_drivers++)
2596 pr_info("printk: continuation disabled due to ext consoles, expect more fragments in /dev/kmsg\n");
2597
2598 if (newcon->flags & CON_PRINTBUFFER) {
2599 /*
2600 * console_unlock(); will print out the buffered messages
2601 * for us.
2602 */
2603 raw_spin_lock_irqsave(&logbuf_lock, flags);
2604 console_seq = syslog_seq;
2605 console_idx = syslog_idx;
2606 console_prev = syslog_prev;
2607 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2608 /*
2609 * We're about to replay the log buffer. Only do this to the
2610 * just-registered console to avoid excessive message spam to
2611 * the already-registered consoles.
2612 */
2613 exclusive_console = newcon;
2614 }
2615 console_unlock();
2616 console_sysfs_notify();
2617
2618 /*
2619 * By unregistering the bootconsoles after we enable the real console
2620 * we get the "console xxx enabled" message on all the consoles -
2621 * boot consoles, real consoles, etc - this is to ensure that end
2622 * users know there might be something in the kernel's log buffer that
2623 * went to the bootconsole (that they do not see on the real console)
2624 */
2625 pr_info("%sconsole [%s%d] enabled\n",
2626 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2627 newcon->name, newcon->index);
2628 if (bcon &&
2629 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2630 !keep_bootcon) {
2631 /* We need to iterate through all boot consoles, to make
2632 * sure we print everything out, before we unregister them.
2633 */
2634 for_each_console(bcon)
2635 if (bcon->flags & CON_BOOT)
2636 unregister_console(bcon);
2637 }
2638}
2639EXPORT_SYMBOL(register_console);
2640
2641int unregister_console(struct console *console)
2642{
2643 struct console *a, *b;
2644 int res;
2645
2646 pr_info("%sconsole [%s%d] disabled\n",
2647 (console->flags & CON_BOOT) ? "boot" : "" ,
2648 console->name, console->index);
2649
2650 res = _braille_unregister_console(console);
2651 if (res)
2652 return res;
2653
2654 res = 1;
2655 console_lock();
2656 if (console_drivers == console) {
2657 console_drivers=console->next;
2658 res = 0;
2659 } else if (console_drivers) {
2660 for (a=console_drivers->next, b=console_drivers ;
2661 a; b=a, a=b->next) {
2662 if (a == console) {
2663 b->next = a->next;
2664 res = 0;
2665 break;
2666 }
2667 }
2668 }
2669
2670 if (!res && (console->flags & CON_EXTENDED))
2671 nr_ext_console_drivers--;
2672
2673 /*
2674 * If this isn't the last console and it has CON_CONSDEV set, we
2675 * need to set it on the next preferred console.
2676 */
2677 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2678 console_drivers->flags |= CON_CONSDEV;
2679
2680 console->flags &= ~CON_ENABLED;
2681 console_unlock();
2682 console_sysfs_notify();
2683 return res;
2684}
2685EXPORT_SYMBOL(unregister_console);
2686
2687/*
2688 * Some boot consoles access data that is in the init section and which will
2689 * be discarded after the initcalls have been run. To make sure that no code
2690 * will access this data, unregister the boot consoles in a late initcall.
2691 *
2692 * If for some reason, such as deferred probe or the driver being a loadable
2693 * module, the real console hasn't registered yet at this point, there will
2694 * be a brief interval in which no messages are logged to the console, which
2695 * makes it difficult to diagnose problems that occur during this time.
2696 *
2697 * To mitigate this problem somewhat, only unregister consoles whose memory
2698 * intersects with the init section. Note that code exists elsewhere to get
2699 * rid of the boot console as soon as the proper console shows up, so there
2700 * won't be side-effects from postponing the removal.
2701 */
2702static int __init printk_late_init(void)
2703{
2704 struct console *con;
2705
2706 for_each_console(con) {
2707 if (!keep_bootcon && con->flags & CON_BOOT) {
2708 /*
2709 * Make sure to unregister boot consoles whose data
2710 * resides in the init section before the init section
2711 * is discarded. Boot consoles whose data will stick
2712 * around will automatically be unregistered when the
2713 * proper console replaces them.
2714 */
2715 if (init_section_intersects(con, sizeof(*con)))
2716 unregister_console(con);
2717 }
2718 }
2719 hotcpu_notifier(console_cpu_notify, 0);
2720 return 0;
2721}
2722late_initcall(printk_late_init);
2723
2724#if defined CONFIG_PRINTK
2725/*
2726 * Delayed printk version, for scheduler-internal messages:
2727 */
2728#define PRINTK_PENDING_WAKEUP 0x01
2729#define PRINTK_PENDING_OUTPUT 0x02
2730
2731static DEFINE_PER_CPU(int, printk_pending);
2732
2733static void wake_up_klogd_work_func(struct irq_work *irq_work)
2734{
2735 int pending = __this_cpu_xchg(printk_pending, 0);
2736
2737 if (pending & PRINTK_PENDING_OUTPUT) {
2738 /* If trylock fails, someone else is doing the printing */
2739 if (console_trylock())
2740 console_unlock();
2741 }
2742
2743 if (pending & PRINTK_PENDING_WAKEUP)
2744 wake_up_interruptible(&log_wait);
2745}
2746
2747static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2748 .func = wake_up_klogd_work_func,
2749 .flags = IRQ_WORK_LAZY,
2750};
2751
2752void wake_up_klogd(void)
2753{
2754 preempt_disable();
2755 if (waitqueue_active(&log_wait)) {
2756 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2757 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2758 }
2759 preempt_enable();
2760}
2761
2762int printk_deferred(const char *fmt, ...)
2763{
2764 va_list args;
2765 int r;
2766
2767 preempt_disable();
2768 va_start(args, fmt);
2769 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2770 va_end(args);
2771
2772 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2773 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2774 preempt_enable();
2775
2776 return r;
2777}
2778
2779/*
2780 * printk rate limiting, lifted from the networking subsystem.
2781 *
2782 * This enforces a rate limit: not more than 10 kernel messages
2783 * every 5s to make a denial-of-service attack impossible.
2784 */
2785DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2786
2787int __printk_ratelimit(const char *func)
2788{
2789 return ___ratelimit(&printk_ratelimit_state, func);
2790}
2791EXPORT_SYMBOL(__printk_ratelimit);
2792
2793/**
2794 * printk_timed_ratelimit - caller-controlled printk ratelimiting
2795 * @caller_jiffies: pointer to caller's state
2796 * @interval_msecs: minimum interval between prints
2797 *
2798 * printk_timed_ratelimit() returns true if more than @interval_msecs
2799 * milliseconds have elapsed since the last time printk_timed_ratelimit()
2800 * returned true.
2801 */
2802bool printk_timed_ratelimit(unsigned long *caller_jiffies,
2803 unsigned int interval_msecs)
2804{
2805 unsigned long elapsed = jiffies - *caller_jiffies;
2806
2807 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
2808 return false;
2809
2810 *caller_jiffies = jiffies;
2811 return true;
2812}
2813EXPORT_SYMBOL(printk_timed_ratelimit);
2814
2815static DEFINE_SPINLOCK(dump_list_lock);
2816static LIST_HEAD(dump_list);
2817
2818/**
2819 * kmsg_dump_register - register a kernel log dumper.
2820 * @dumper: pointer to the kmsg_dumper structure
2821 *
2822 * Adds a kernel log dumper to the system. The dump callback in the
2823 * structure will be called when the kernel oopses or panics and must be
2824 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
2825 */
2826int kmsg_dump_register(struct kmsg_dumper *dumper)
2827{
2828 unsigned long flags;
2829 int err = -EBUSY;
2830
2831 /* The dump callback needs to be set */
2832 if (!dumper->dump)
2833 return -EINVAL;
2834
2835 spin_lock_irqsave(&dump_list_lock, flags);
2836 /* Don't allow registering multiple times */
2837 if (!dumper->registered) {
2838 dumper->registered = 1;
2839 list_add_tail_rcu(&dumper->list, &dump_list);
2840 err = 0;
2841 }
2842 spin_unlock_irqrestore(&dump_list_lock, flags);
2843
2844 return err;
2845}
2846EXPORT_SYMBOL_GPL(kmsg_dump_register);
2847
2848/**
2849 * kmsg_dump_unregister - unregister a kmsg dumper.
2850 * @dumper: pointer to the kmsg_dumper structure
2851 *
2852 * Removes a dump device from the system. Returns zero on success and
2853 * %-EINVAL otherwise.
2854 */
2855int kmsg_dump_unregister(struct kmsg_dumper *dumper)
2856{
2857 unsigned long flags;
2858 int err = -EINVAL;
2859
2860 spin_lock_irqsave(&dump_list_lock, flags);
2861 if (dumper->registered) {
2862 dumper->registered = 0;
2863 list_del_rcu(&dumper->list);
2864 err = 0;
2865 }
2866 spin_unlock_irqrestore(&dump_list_lock, flags);
2867 synchronize_rcu();
2868
2869 return err;
2870}
2871EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
2872
2873static bool always_kmsg_dump;
2874module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
2875
2876/**
2877 * kmsg_dump - dump kernel log to kernel message dumpers.
2878 * @reason: the reason (oops, panic etc) for dumping
2879 *
2880 * Call each of the registered dumper's dump() callback, which can
2881 * retrieve the kmsg records with kmsg_dump_get_line() or
2882 * kmsg_dump_get_buffer().
2883 */
2884void kmsg_dump(enum kmsg_dump_reason reason)
2885{
2886 struct kmsg_dumper *dumper;
2887 unsigned long flags;
2888
2889 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
2890 return;
2891
2892 rcu_read_lock();
2893 list_for_each_entry_rcu(dumper, &dump_list, list) {
2894 if (dumper->max_reason && reason > dumper->max_reason)
2895 continue;
2896
2897 /* initialize iterator with data about the stored records */
2898 dumper->active = true;
2899
2900 raw_spin_lock_irqsave(&logbuf_lock, flags);
2901 dumper->cur_seq = clear_seq;
2902 dumper->cur_idx = clear_idx;
2903 dumper->next_seq = log_next_seq;
2904 dumper->next_idx = log_next_idx;
2905 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2906
2907 /* invoke dumper which will iterate over records */
2908 dumper->dump(dumper, reason);
2909
2910 /* reset iterator */
2911 dumper->active = false;
2912 }
2913 rcu_read_unlock();
2914}
2915
2916/**
2917 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
2918 * @dumper: registered kmsg dumper
2919 * @syslog: include the "<4>" prefixes
2920 * @line: buffer to copy the line to
2921 * @size: maximum size of the buffer
2922 * @len: length of line placed into buffer
2923 *
2924 * Start at the beginning of the kmsg buffer, with the oldest kmsg
2925 * record, and copy one record into the provided buffer.
2926 *
2927 * Consecutive calls will return the next available record moving
2928 * towards the end of the buffer with the youngest messages.
2929 *
2930 * A return value of FALSE indicates that there are no more records to
2931 * read.
2932 *
2933 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
2934 */
2935bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
2936 char *line, size_t size, size_t *len)
2937{
2938 struct printk_log *msg;
2939 size_t l = 0;
2940 bool ret = false;
2941
2942 if (!dumper->active)
2943 goto out;
2944
2945 if (dumper->cur_seq < log_first_seq) {
2946 /* messages are gone, move to first available one */
2947 dumper->cur_seq = log_first_seq;
2948 dumper->cur_idx = log_first_idx;
2949 }
2950
2951 /* last entry */
2952 if (dumper->cur_seq >= log_next_seq)
2953 goto out;
2954
2955 msg = log_from_idx(dumper->cur_idx);
2956 l = msg_print_text(msg, 0, syslog, line, size);
2957
2958 dumper->cur_idx = log_next(dumper->cur_idx);
2959 dumper->cur_seq++;
2960 ret = true;
2961out:
2962 if (len)
2963 *len = l;
2964 return ret;
2965}
2966
2967/**
2968 * kmsg_dump_get_line - retrieve one kmsg log line
2969 * @dumper: registered kmsg dumper
2970 * @syslog: include the "<4>" prefixes
2971 * @line: buffer to copy the line to
2972 * @size: maximum size of the buffer
2973 * @len: length of line placed into buffer
2974 *
2975 * Start at the beginning of the kmsg buffer, with the oldest kmsg
2976 * record, and copy one record into the provided buffer.
2977 *
2978 * Consecutive calls will return the next available record moving
2979 * towards the end of the buffer with the youngest messages.
2980 *
2981 * A return value of FALSE indicates that there are no more records to
2982 * read.
2983 */
2984bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
2985 char *line, size_t size, size_t *len)
2986{
2987 unsigned long flags;
2988 bool ret;
2989
2990 raw_spin_lock_irqsave(&logbuf_lock, flags);
2991 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
2992 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2993
2994 return ret;
2995}
2996EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
2997
2998/**
2999 * kmsg_dump_get_buffer - copy kmsg log lines
3000 * @dumper: registered kmsg dumper
3001 * @syslog: include the "<4>" prefixes
3002 * @buf: buffer to copy the line to
3003 * @size: maximum size of the buffer
3004 * @len: length of line placed into buffer
3005 *
3006 * Start at the end of the kmsg buffer and fill the provided buffer
3007 * with as many of the the *youngest* kmsg records that fit into it.
3008 * If the buffer is large enough, all available kmsg records will be
3009 * copied with a single call.
3010 *
3011 * Consecutive calls will fill the buffer with the next block of
3012 * available older records, not including the earlier retrieved ones.
3013 *
3014 * A return value of FALSE indicates that there are no more records to
3015 * read.
3016 */
3017bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3018 char *buf, size_t size, size_t *len)
3019{
3020 unsigned long flags;
3021 u64 seq;
3022 u32 idx;
3023 u64 next_seq;
3024 u32 next_idx;
3025 enum log_flags prev;
3026 size_t l = 0;
3027 bool ret = false;
3028
3029 if (!dumper->active)
3030 goto out;
3031
3032 raw_spin_lock_irqsave(&logbuf_lock, flags);
3033 if (dumper->cur_seq < log_first_seq) {
3034 /* messages are gone, move to first available one */
3035 dumper->cur_seq = log_first_seq;
3036 dumper->cur_idx = log_first_idx;
3037 }
3038
3039 /* last entry */
3040 if (dumper->cur_seq >= dumper->next_seq) {
3041 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3042 goto out;
3043 }
3044
3045 /* calculate length of entire buffer */
3046 seq = dumper->cur_seq;
3047 idx = dumper->cur_idx;
3048 prev = 0;
3049 while (seq < dumper->next_seq) {
3050 struct printk_log *msg = log_from_idx(idx);
3051
3052 l += msg_print_text(msg, prev, true, NULL, 0);
3053 idx = log_next(idx);
3054 seq++;
3055 prev = msg->flags;
3056 }
3057
3058 /* move first record forward until length fits into the buffer */
3059 seq = dumper->cur_seq;
3060 idx = dumper->cur_idx;
3061 prev = 0;
3062 while (l > size && seq < dumper->next_seq) {
3063 struct printk_log *msg = log_from_idx(idx);
3064
3065 l -= msg_print_text(msg, prev, true, NULL, 0);
3066 idx = log_next(idx);
3067 seq++;
3068 prev = msg->flags;
3069 }
3070
3071 /* last message in next interation */
3072 next_seq = seq;
3073 next_idx = idx;
3074
3075 l = 0;
3076 while (seq < dumper->next_seq) {
3077 struct printk_log *msg = log_from_idx(idx);
3078
3079 l += msg_print_text(msg, prev, syslog, buf + l, size - l);
3080 idx = log_next(idx);
3081 seq++;
3082 prev = msg->flags;
3083 }
3084
3085 dumper->next_seq = next_seq;
3086 dumper->next_idx = next_idx;
3087 ret = true;
3088 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3089out:
3090 if (len)
3091 *len = l;
3092 return ret;
3093}
3094EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3095
3096/**
3097 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3098 * @dumper: registered kmsg dumper
3099 *
3100 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3101 * kmsg_dump_get_buffer() can be called again and used multiple
3102 * times within the same dumper.dump() callback.
3103 *
3104 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3105 */
3106void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3107{
3108 dumper->cur_seq = clear_seq;
3109 dumper->cur_idx = clear_idx;
3110 dumper->next_seq = log_next_seq;
3111 dumper->next_idx = log_next_idx;
3112}
3113
3114/**
3115 * kmsg_dump_rewind - reset the interator
3116 * @dumper: registered kmsg dumper
3117 *
3118 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3119 * kmsg_dump_get_buffer() can be called again and used multiple
3120 * times within the same dumper.dump() callback.
3121 */
3122void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3123{
3124 unsigned long flags;
3125
3126 raw_spin_lock_irqsave(&logbuf_lock, flags);
3127 kmsg_dump_rewind_nolock(dumper);
3128 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3129}
3130EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3131
3132static char dump_stack_arch_desc_str[128];
3133
3134/**
3135 * dump_stack_set_arch_desc - set arch-specific str to show with task dumps
3136 * @fmt: printf-style format string
3137 * @...: arguments for the format string
3138 *
3139 * The configured string will be printed right after utsname during task
3140 * dumps. Usually used to add arch-specific system identifiers. If an
3141 * arch wants to make use of such an ID string, it should initialize this
3142 * as soon as possible during boot.
3143 */
3144void __init dump_stack_set_arch_desc(const char *fmt, ...)
3145{
3146 va_list args;
3147
3148 va_start(args, fmt);
3149 vsnprintf(dump_stack_arch_desc_str, sizeof(dump_stack_arch_desc_str),
3150 fmt, args);
3151 va_end(args);
3152}
3153
3154/**
3155 * dump_stack_print_info - print generic debug info for dump_stack()
3156 * @log_lvl: log level
3157 *
3158 * Arch-specific dump_stack() implementations can use this function to
3159 * print out the same debug information as the generic dump_stack().
3160 */
3161void dump_stack_print_info(const char *log_lvl)
3162{
3163 printk("%sCPU: %d PID: %d Comm: %.20s %s %s %.*s\n",
3164 log_lvl, raw_smp_processor_id(), current->pid, current->comm,
3165 print_tainted(), init_utsname()->release,
3166 (int)strcspn(init_utsname()->version, " "),
3167 init_utsname()->version);
3168
3169 if (dump_stack_arch_desc_str[0] != '\0')
3170 printk("%sHardware name: %s\n",
3171 log_lvl, dump_stack_arch_desc_str);
3172
3173 print_worker_info(log_lvl, current);
3174}
3175
3176/**
3177 * show_regs_print_info - print generic debug info for show_regs()
3178 * @log_lvl: log level
3179 *
3180 * show_regs() implementations can use this function to print out generic
3181 * debug information.
3182 */
3183void show_regs_print_info(const char *log_lvl)
3184{
3185 dump_stack_print_info(log_lvl);
3186
3187 printk("%stask: %p ti: %p task.ti: %p\n",
3188 log_lvl, current, current_thread_info(),
3189 task_thread_info(current));
3190}
3191
3192#endif
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/printk.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * Modified to make sys_syslog() more flexible: added commands to
8 * return the last 4k of kernel messages, regardless of whether
9 * they've been read or not. Added option to suppress kernel printk's
10 * to the console. Added hook for sending the console messages
11 * elsewhere, in preparation for a serial line console (someday).
12 * Ted Ts'o, 2/11/93.
13 * Modified for sysctl support, 1/8/97, Chris Horn.
14 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
15 * manfred@colorfullife.com
16 * Rewrote bits to get rid of console_lock
17 * 01Mar01 Andrew Morton
18 */
19
20#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21
22#include <linux/kernel.h>
23#include <linux/mm.h>
24#include <linux/tty.h>
25#include <linux/tty_driver.h>
26#include <linux/console.h>
27#include <linux/init.h>
28#include <linux/jiffies.h>
29#include <linux/nmi.h>
30#include <linux/module.h>
31#include <linux/moduleparam.h>
32#include <linux/delay.h>
33#include <linux/smp.h>
34#include <linux/security.h>
35#include <linux/memblock.h>
36#include <linux/syscalls.h>
37#include <linux/crash_core.h>
38#include <linux/kdb.h>
39#include <linux/ratelimit.h>
40#include <linux/kmsg_dump.h>
41#include <linux/syslog.h>
42#include <linux/cpu.h>
43#include <linux/rculist.h>
44#include <linux/poll.h>
45#include <linux/irq_work.h>
46#include <linux/ctype.h>
47#include <linux/uio.h>
48#include <linux/sched/clock.h>
49#include <linux/sched/debug.h>
50#include <linux/sched/task_stack.h>
51
52#include <linux/uaccess.h>
53#include <asm/sections.h>
54
55#include <trace/events/initcall.h>
56#define CREATE_TRACE_POINTS
57#include <trace/events/printk.h>
58
59#include "console_cmdline.h"
60#include "braille.h"
61#include "internal.h"
62
63int console_printk[4] = {
64 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
65 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
66 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
67 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
68};
69EXPORT_SYMBOL_GPL(console_printk);
70
71atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
72EXPORT_SYMBOL(ignore_console_lock_warning);
73
74/*
75 * Low level drivers may need that to know if they can schedule in
76 * their unblank() callback or not. So let's export it.
77 */
78int oops_in_progress;
79EXPORT_SYMBOL(oops_in_progress);
80
81/*
82 * console_sem protects the console_drivers list, and also
83 * provides serialisation for access to the entire console
84 * driver system.
85 */
86static DEFINE_SEMAPHORE(console_sem);
87struct console *console_drivers;
88EXPORT_SYMBOL_GPL(console_drivers);
89
90/*
91 * System may need to suppress printk message under certain
92 * circumstances, like after kernel panic happens.
93 */
94int __read_mostly suppress_printk;
95
96#ifdef CONFIG_LOCKDEP
97static struct lockdep_map console_lock_dep_map = {
98 .name = "console_lock"
99};
100#endif
101
102enum devkmsg_log_bits {
103 __DEVKMSG_LOG_BIT_ON = 0,
104 __DEVKMSG_LOG_BIT_OFF,
105 __DEVKMSG_LOG_BIT_LOCK,
106};
107
108enum devkmsg_log_masks {
109 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
110 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
111 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
112};
113
114/* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
115#define DEVKMSG_LOG_MASK_DEFAULT 0
116
117static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
118
119static int __control_devkmsg(char *str)
120{
121 size_t len;
122
123 if (!str)
124 return -EINVAL;
125
126 len = str_has_prefix(str, "on");
127 if (len) {
128 devkmsg_log = DEVKMSG_LOG_MASK_ON;
129 return len;
130 }
131
132 len = str_has_prefix(str, "off");
133 if (len) {
134 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
135 return len;
136 }
137
138 len = str_has_prefix(str, "ratelimit");
139 if (len) {
140 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
141 return len;
142 }
143
144 return -EINVAL;
145}
146
147static int __init control_devkmsg(char *str)
148{
149 if (__control_devkmsg(str) < 0)
150 return 1;
151
152 /*
153 * Set sysctl string accordingly:
154 */
155 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
156 strcpy(devkmsg_log_str, "on");
157 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
158 strcpy(devkmsg_log_str, "off");
159 /* else "ratelimit" which is set by default. */
160
161 /*
162 * Sysctl cannot change it anymore. The kernel command line setting of
163 * this parameter is to force the setting to be permanent throughout the
164 * runtime of the system. This is a precation measure against userspace
165 * trying to be a smarta** and attempting to change it up on us.
166 */
167 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
168
169 return 0;
170}
171__setup("printk.devkmsg=", control_devkmsg);
172
173char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
174
175int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
176 void __user *buffer, size_t *lenp, loff_t *ppos)
177{
178 char old_str[DEVKMSG_STR_MAX_SIZE];
179 unsigned int old;
180 int err;
181
182 if (write) {
183 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
184 return -EINVAL;
185
186 old = devkmsg_log;
187 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
188 }
189
190 err = proc_dostring(table, write, buffer, lenp, ppos);
191 if (err)
192 return err;
193
194 if (write) {
195 err = __control_devkmsg(devkmsg_log_str);
196
197 /*
198 * Do not accept an unknown string OR a known string with
199 * trailing crap...
200 */
201 if (err < 0 || (err + 1 != *lenp)) {
202
203 /* ... and restore old setting. */
204 devkmsg_log = old;
205 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
206
207 return -EINVAL;
208 }
209 }
210
211 return 0;
212}
213
214/* Number of registered extended console drivers. */
215static int nr_ext_console_drivers;
216
217/*
218 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
219 * macros instead of functions so that _RET_IP_ contains useful information.
220 */
221#define down_console_sem() do { \
222 down(&console_sem);\
223 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
224} while (0)
225
226static int __down_trylock_console_sem(unsigned long ip)
227{
228 int lock_failed;
229 unsigned long flags;
230
231 /*
232 * Here and in __up_console_sem() we need to be in safe mode,
233 * because spindump/WARN/etc from under console ->lock will
234 * deadlock in printk()->down_trylock_console_sem() otherwise.
235 */
236 printk_safe_enter_irqsave(flags);
237 lock_failed = down_trylock(&console_sem);
238 printk_safe_exit_irqrestore(flags);
239
240 if (lock_failed)
241 return 1;
242 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
243 return 0;
244}
245#define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
246
247static void __up_console_sem(unsigned long ip)
248{
249 unsigned long flags;
250
251 mutex_release(&console_lock_dep_map, 1, ip);
252
253 printk_safe_enter_irqsave(flags);
254 up(&console_sem);
255 printk_safe_exit_irqrestore(flags);
256}
257#define up_console_sem() __up_console_sem(_RET_IP_)
258
259/*
260 * This is used for debugging the mess that is the VT code by
261 * keeping track if we have the console semaphore held. It's
262 * definitely not the perfect debug tool (we don't know if _WE_
263 * hold it and are racing, but it helps tracking those weird code
264 * paths in the console code where we end up in places I want
265 * locked without the console sempahore held).
266 */
267static int console_locked, console_suspended;
268
269/*
270 * If exclusive_console is non-NULL then only this console is to be printed to.
271 */
272static struct console *exclusive_console;
273
274/*
275 * Array of consoles built from command line options (console=)
276 */
277
278#define MAX_CMDLINECONSOLES 8
279
280static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
281
282static int preferred_console = -1;
283int console_set_on_cmdline;
284EXPORT_SYMBOL(console_set_on_cmdline);
285
286/* Flag: console code may call schedule() */
287static int console_may_schedule;
288
289enum con_msg_format_flags {
290 MSG_FORMAT_DEFAULT = 0,
291 MSG_FORMAT_SYSLOG = (1 << 0),
292};
293
294static int console_msg_format = MSG_FORMAT_DEFAULT;
295
296/*
297 * The printk log buffer consists of a chain of concatenated variable
298 * length records. Every record starts with a record header, containing
299 * the overall length of the record.
300 *
301 * The heads to the first and last entry in the buffer, as well as the
302 * sequence numbers of these entries are maintained when messages are
303 * stored.
304 *
305 * If the heads indicate available messages, the length in the header
306 * tells the start next message. A length == 0 for the next message
307 * indicates a wrap-around to the beginning of the buffer.
308 *
309 * Every record carries the monotonic timestamp in microseconds, as well as
310 * the standard userspace syslog level and syslog facility. The usual
311 * kernel messages use LOG_KERN; userspace-injected messages always carry
312 * a matching syslog facility, by default LOG_USER. The origin of every
313 * message can be reliably determined that way.
314 *
315 * The human readable log message directly follows the message header. The
316 * length of the message text is stored in the header, the stored message
317 * is not terminated.
318 *
319 * Optionally, a message can carry a dictionary of properties (key/value pairs),
320 * to provide userspace with a machine-readable message context.
321 *
322 * Examples for well-defined, commonly used property names are:
323 * DEVICE=b12:8 device identifier
324 * b12:8 block dev_t
325 * c127:3 char dev_t
326 * n8 netdev ifindex
327 * +sound:card0 subsystem:devname
328 * SUBSYSTEM=pci driver-core subsystem name
329 *
330 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
331 * follows directly after a '=' character. Every property is terminated by
332 * a '\0' character. The last property is not terminated.
333 *
334 * Example of a message structure:
335 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
336 * 0008 34 00 record is 52 bytes long
337 * 000a 0b 00 text is 11 bytes long
338 * 000c 1f 00 dictionary is 23 bytes long
339 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
340 * 0010 69 74 27 73 20 61 20 6c "it's a l"
341 * 69 6e 65 "ine"
342 * 001b 44 45 56 49 43 "DEVIC"
343 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
344 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
345 * 67 "g"
346 * 0032 00 00 00 padding to next message header
347 *
348 * The 'struct printk_log' buffer header must never be directly exported to
349 * userspace, it is a kernel-private implementation detail that might
350 * need to be changed in the future, when the requirements change.
351 *
352 * /dev/kmsg exports the structured data in the following line format:
353 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
354 *
355 * Users of the export format should ignore possible additional values
356 * separated by ',', and find the message after the ';' character.
357 *
358 * The optional key/value pairs are attached as continuation lines starting
359 * with a space character and terminated by a newline. All possible
360 * non-prinatable characters are escaped in the "\xff" notation.
361 */
362
363enum log_flags {
364 LOG_NEWLINE = 2, /* text ended with a newline */
365 LOG_CONT = 8, /* text is a fragment of a continuation line */
366};
367
368struct printk_log {
369 u64 ts_nsec; /* timestamp in nanoseconds */
370 u16 len; /* length of entire record */
371 u16 text_len; /* length of text buffer */
372 u16 dict_len; /* length of dictionary buffer */
373 u8 facility; /* syslog facility */
374 u8 flags:5; /* internal record flags */
375 u8 level:3; /* syslog level */
376#ifdef CONFIG_PRINTK_CALLER
377 u32 caller_id; /* thread id or processor id */
378#endif
379}
380#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
381__packed __aligned(4)
382#endif
383;
384
385/*
386 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
387 * within the scheduler's rq lock. It must be released before calling
388 * console_unlock() or anything else that might wake up a process.
389 */
390DEFINE_RAW_SPINLOCK(logbuf_lock);
391
392/*
393 * Helper macros to lock/unlock logbuf_lock and switch between
394 * printk-safe/unsafe modes.
395 */
396#define logbuf_lock_irq() \
397 do { \
398 printk_safe_enter_irq(); \
399 raw_spin_lock(&logbuf_lock); \
400 } while (0)
401
402#define logbuf_unlock_irq() \
403 do { \
404 raw_spin_unlock(&logbuf_lock); \
405 printk_safe_exit_irq(); \
406 } while (0)
407
408#define logbuf_lock_irqsave(flags) \
409 do { \
410 printk_safe_enter_irqsave(flags); \
411 raw_spin_lock(&logbuf_lock); \
412 } while (0)
413
414#define logbuf_unlock_irqrestore(flags) \
415 do { \
416 raw_spin_unlock(&logbuf_lock); \
417 printk_safe_exit_irqrestore(flags); \
418 } while (0)
419
420#ifdef CONFIG_PRINTK
421DECLARE_WAIT_QUEUE_HEAD(log_wait);
422/* the next printk record to read by syslog(READ) or /proc/kmsg */
423static u64 syslog_seq;
424static u32 syslog_idx;
425static size_t syslog_partial;
426static bool syslog_time;
427
428/* index and sequence number of the first record stored in the buffer */
429static u64 log_first_seq;
430static u32 log_first_idx;
431
432/* index and sequence number of the next record to store in the buffer */
433static u64 log_next_seq;
434static u32 log_next_idx;
435
436/* the next printk record to write to the console */
437static u64 console_seq;
438static u32 console_idx;
439static u64 exclusive_console_stop_seq;
440
441/* the next printk record to read after the last 'clear' command */
442static u64 clear_seq;
443static u32 clear_idx;
444
445#ifdef CONFIG_PRINTK_CALLER
446#define PREFIX_MAX 48
447#else
448#define PREFIX_MAX 32
449#endif
450#define LOG_LINE_MAX (1024 - PREFIX_MAX)
451
452#define LOG_LEVEL(v) ((v) & 0x07)
453#define LOG_FACILITY(v) ((v) >> 3 & 0xff)
454
455/* record buffer */
456#define LOG_ALIGN __alignof__(struct printk_log)
457#define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
458#define LOG_BUF_LEN_MAX (u32)(1 << 31)
459static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
460static char *log_buf = __log_buf;
461static u32 log_buf_len = __LOG_BUF_LEN;
462
463/* Return log buffer address */
464char *log_buf_addr_get(void)
465{
466 return log_buf;
467}
468
469/* Return log buffer size */
470u32 log_buf_len_get(void)
471{
472 return log_buf_len;
473}
474
475/* human readable text of the record */
476static char *log_text(const struct printk_log *msg)
477{
478 return (char *)msg + sizeof(struct printk_log);
479}
480
481/* optional key/value pair dictionary attached to the record */
482static char *log_dict(const struct printk_log *msg)
483{
484 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
485}
486
487/* get record by index; idx must point to valid msg */
488static struct printk_log *log_from_idx(u32 idx)
489{
490 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
491
492 /*
493 * A length == 0 record is the end of buffer marker. Wrap around and
494 * read the message at the start of the buffer.
495 */
496 if (!msg->len)
497 return (struct printk_log *)log_buf;
498 return msg;
499}
500
501/* get next record; idx must point to valid msg */
502static u32 log_next(u32 idx)
503{
504 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
505
506 /* length == 0 indicates the end of the buffer; wrap */
507 /*
508 * A length == 0 record is the end of buffer marker. Wrap around and
509 * read the message at the start of the buffer as *this* one, and
510 * return the one after that.
511 */
512 if (!msg->len) {
513 msg = (struct printk_log *)log_buf;
514 return msg->len;
515 }
516 return idx + msg->len;
517}
518
519/*
520 * Check whether there is enough free space for the given message.
521 *
522 * The same values of first_idx and next_idx mean that the buffer
523 * is either empty or full.
524 *
525 * If the buffer is empty, we must respect the position of the indexes.
526 * They cannot be reset to the beginning of the buffer.
527 */
528static int logbuf_has_space(u32 msg_size, bool empty)
529{
530 u32 free;
531
532 if (log_next_idx > log_first_idx || empty)
533 free = max(log_buf_len - log_next_idx, log_first_idx);
534 else
535 free = log_first_idx - log_next_idx;
536
537 /*
538 * We need space also for an empty header that signalizes wrapping
539 * of the buffer.
540 */
541 return free >= msg_size + sizeof(struct printk_log);
542}
543
544static int log_make_free_space(u32 msg_size)
545{
546 while (log_first_seq < log_next_seq &&
547 !logbuf_has_space(msg_size, false)) {
548 /* drop old messages until we have enough contiguous space */
549 log_first_idx = log_next(log_first_idx);
550 log_first_seq++;
551 }
552
553 if (clear_seq < log_first_seq) {
554 clear_seq = log_first_seq;
555 clear_idx = log_first_idx;
556 }
557
558 /* sequence numbers are equal, so the log buffer is empty */
559 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
560 return 0;
561
562 return -ENOMEM;
563}
564
565/* compute the message size including the padding bytes */
566static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
567{
568 u32 size;
569
570 size = sizeof(struct printk_log) + text_len + dict_len;
571 *pad_len = (-size) & (LOG_ALIGN - 1);
572 size += *pad_len;
573
574 return size;
575}
576
577/*
578 * Define how much of the log buffer we could take at maximum. The value
579 * must be greater than two. Note that only half of the buffer is available
580 * when the index points to the middle.
581 */
582#define MAX_LOG_TAKE_PART 4
583static const char trunc_msg[] = "<truncated>";
584
585static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
586 u16 *dict_len, u32 *pad_len)
587{
588 /*
589 * The message should not take the whole buffer. Otherwise, it might
590 * get removed too soon.
591 */
592 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
593 if (*text_len > max_text_len)
594 *text_len = max_text_len;
595 /* enable the warning message */
596 *trunc_msg_len = strlen(trunc_msg);
597 /* disable the "dict" completely */
598 *dict_len = 0;
599 /* compute the size again, count also the warning message */
600 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
601}
602
603/* insert record into the buffer, discard old ones, update heads */
604static int log_store(u32 caller_id, int facility, int level,
605 enum log_flags flags, u64 ts_nsec,
606 const char *dict, u16 dict_len,
607 const char *text, u16 text_len)
608{
609 struct printk_log *msg;
610 u32 size, pad_len;
611 u16 trunc_msg_len = 0;
612
613 /* number of '\0' padding bytes to next message */
614 size = msg_used_size(text_len, dict_len, &pad_len);
615
616 if (log_make_free_space(size)) {
617 /* truncate the message if it is too long for empty buffer */
618 size = truncate_msg(&text_len, &trunc_msg_len,
619 &dict_len, &pad_len);
620 /* survive when the log buffer is too small for trunc_msg */
621 if (log_make_free_space(size))
622 return 0;
623 }
624
625 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
626 /*
627 * This message + an additional empty header does not fit
628 * at the end of the buffer. Add an empty header with len == 0
629 * to signify a wrap around.
630 */
631 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
632 log_next_idx = 0;
633 }
634
635 /* fill message */
636 msg = (struct printk_log *)(log_buf + log_next_idx);
637 memcpy(log_text(msg), text, text_len);
638 msg->text_len = text_len;
639 if (trunc_msg_len) {
640 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
641 msg->text_len += trunc_msg_len;
642 }
643 memcpy(log_dict(msg), dict, dict_len);
644 msg->dict_len = dict_len;
645 msg->facility = facility;
646 msg->level = level & 7;
647 msg->flags = flags & 0x1f;
648 if (ts_nsec > 0)
649 msg->ts_nsec = ts_nsec;
650 else
651 msg->ts_nsec = local_clock();
652#ifdef CONFIG_PRINTK_CALLER
653 msg->caller_id = caller_id;
654#endif
655 memset(log_dict(msg) + dict_len, 0, pad_len);
656 msg->len = size;
657
658 /* insert message */
659 log_next_idx += msg->len;
660 log_next_seq++;
661
662 return msg->text_len;
663}
664
665int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
666
667static int syslog_action_restricted(int type)
668{
669 if (dmesg_restrict)
670 return 1;
671 /*
672 * Unless restricted, we allow "read all" and "get buffer size"
673 * for everybody.
674 */
675 return type != SYSLOG_ACTION_READ_ALL &&
676 type != SYSLOG_ACTION_SIZE_BUFFER;
677}
678
679static int check_syslog_permissions(int type, int source)
680{
681 /*
682 * If this is from /proc/kmsg and we've already opened it, then we've
683 * already done the capabilities checks at open time.
684 */
685 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
686 goto ok;
687
688 if (syslog_action_restricted(type)) {
689 if (capable(CAP_SYSLOG))
690 goto ok;
691 /*
692 * For historical reasons, accept CAP_SYS_ADMIN too, with
693 * a warning.
694 */
695 if (capable(CAP_SYS_ADMIN)) {
696 pr_warn_once("%s (%d): Attempt to access syslog with "
697 "CAP_SYS_ADMIN but no CAP_SYSLOG "
698 "(deprecated).\n",
699 current->comm, task_pid_nr(current));
700 goto ok;
701 }
702 return -EPERM;
703 }
704ok:
705 return security_syslog(type);
706}
707
708static void append_char(char **pp, char *e, char c)
709{
710 if (*pp < e)
711 *(*pp)++ = c;
712}
713
714static ssize_t msg_print_ext_header(char *buf, size_t size,
715 struct printk_log *msg, u64 seq)
716{
717 u64 ts_usec = msg->ts_nsec;
718 char caller[20];
719#ifdef CONFIG_PRINTK_CALLER
720 u32 id = msg->caller_id;
721
722 snprintf(caller, sizeof(caller), ",caller=%c%u",
723 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
724#else
725 caller[0] = '\0';
726#endif
727
728 do_div(ts_usec, 1000);
729
730 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
731 (msg->facility << 3) | msg->level, seq, ts_usec,
732 msg->flags & LOG_CONT ? 'c' : '-', caller);
733}
734
735static ssize_t msg_print_ext_body(char *buf, size_t size,
736 char *dict, size_t dict_len,
737 char *text, size_t text_len)
738{
739 char *p = buf, *e = buf + size;
740 size_t i;
741
742 /* escape non-printable characters */
743 for (i = 0; i < text_len; i++) {
744 unsigned char c = text[i];
745
746 if (c < ' ' || c >= 127 || c == '\\')
747 p += scnprintf(p, e - p, "\\x%02x", c);
748 else
749 append_char(&p, e, c);
750 }
751 append_char(&p, e, '\n');
752
753 if (dict_len) {
754 bool line = true;
755
756 for (i = 0; i < dict_len; i++) {
757 unsigned char c = dict[i];
758
759 if (line) {
760 append_char(&p, e, ' ');
761 line = false;
762 }
763
764 if (c == '\0') {
765 append_char(&p, e, '\n');
766 line = true;
767 continue;
768 }
769
770 if (c < ' ' || c >= 127 || c == '\\') {
771 p += scnprintf(p, e - p, "\\x%02x", c);
772 continue;
773 }
774
775 append_char(&p, e, c);
776 }
777 append_char(&p, e, '\n');
778 }
779
780 return p - buf;
781}
782
783/* /dev/kmsg - userspace message inject/listen interface */
784struct devkmsg_user {
785 u64 seq;
786 u32 idx;
787 struct ratelimit_state rs;
788 struct mutex lock;
789 char buf[CONSOLE_EXT_LOG_MAX];
790};
791
792static __printf(3, 4) __cold
793int devkmsg_emit(int facility, int level, const char *fmt, ...)
794{
795 va_list args;
796 int r;
797
798 va_start(args, fmt);
799 r = vprintk_emit(facility, level, NULL, 0, fmt, args);
800 va_end(args);
801
802 return r;
803}
804
805static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
806{
807 char *buf, *line;
808 int level = default_message_loglevel;
809 int facility = 1; /* LOG_USER */
810 struct file *file = iocb->ki_filp;
811 struct devkmsg_user *user = file->private_data;
812 size_t len = iov_iter_count(from);
813 ssize_t ret = len;
814
815 if (!user || len > LOG_LINE_MAX)
816 return -EINVAL;
817
818 /* Ignore when user logging is disabled. */
819 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
820 return len;
821
822 /* Ratelimit when not explicitly enabled. */
823 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
824 if (!___ratelimit(&user->rs, current->comm))
825 return ret;
826 }
827
828 buf = kmalloc(len+1, GFP_KERNEL);
829 if (buf == NULL)
830 return -ENOMEM;
831
832 buf[len] = '\0';
833 if (!copy_from_iter_full(buf, len, from)) {
834 kfree(buf);
835 return -EFAULT;
836 }
837
838 /*
839 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
840 * the decimal value represents 32bit, the lower 3 bit are the log
841 * level, the rest are the log facility.
842 *
843 * If no prefix or no userspace facility is specified, we
844 * enforce LOG_USER, to be able to reliably distinguish
845 * kernel-generated messages from userspace-injected ones.
846 */
847 line = buf;
848 if (line[0] == '<') {
849 char *endp = NULL;
850 unsigned int u;
851
852 u = simple_strtoul(line + 1, &endp, 10);
853 if (endp && endp[0] == '>') {
854 level = LOG_LEVEL(u);
855 if (LOG_FACILITY(u) != 0)
856 facility = LOG_FACILITY(u);
857 endp++;
858 len -= endp - line;
859 line = endp;
860 }
861 }
862
863 devkmsg_emit(facility, level, "%s", line);
864 kfree(buf);
865 return ret;
866}
867
868static ssize_t devkmsg_read(struct file *file, char __user *buf,
869 size_t count, loff_t *ppos)
870{
871 struct devkmsg_user *user = file->private_data;
872 struct printk_log *msg;
873 size_t len;
874 ssize_t ret;
875
876 if (!user)
877 return -EBADF;
878
879 ret = mutex_lock_interruptible(&user->lock);
880 if (ret)
881 return ret;
882
883 logbuf_lock_irq();
884 while (user->seq == log_next_seq) {
885 if (file->f_flags & O_NONBLOCK) {
886 ret = -EAGAIN;
887 logbuf_unlock_irq();
888 goto out;
889 }
890
891 logbuf_unlock_irq();
892 ret = wait_event_interruptible(log_wait,
893 user->seq != log_next_seq);
894 if (ret)
895 goto out;
896 logbuf_lock_irq();
897 }
898
899 if (user->seq < log_first_seq) {
900 /* our last seen message is gone, return error and reset */
901 user->idx = log_first_idx;
902 user->seq = log_first_seq;
903 ret = -EPIPE;
904 logbuf_unlock_irq();
905 goto out;
906 }
907
908 msg = log_from_idx(user->idx);
909 len = msg_print_ext_header(user->buf, sizeof(user->buf),
910 msg, user->seq);
911 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
912 log_dict(msg), msg->dict_len,
913 log_text(msg), msg->text_len);
914
915 user->idx = log_next(user->idx);
916 user->seq++;
917 logbuf_unlock_irq();
918
919 if (len > count) {
920 ret = -EINVAL;
921 goto out;
922 }
923
924 if (copy_to_user(buf, user->buf, len)) {
925 ret = -EFAULT;
926 goto out;
927 }
928 ret = len;
929out:
930 mutex_unlock(&user->lock);
931 return ret;
932}
933
934static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
935{
936 struct devkmsg_user *user = file->private_data;
937 loff_t ret = 0;
938
939 if (!user)
940 return -EBADF;
941 if (offset)
942 return -ESPIPE;
943
944 logbuf_lock_irq();
945 switch (whence) {
946 case SEEK_SET:
947 /* the first record */
948 user->idx = log_first_idx;
949 user->seq = log_first_seq;
950 break;
951 case SEEK_DATA:
952 /*
953 * The first record after the last SYSLOG_ACTION_CLEAR,
954 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
955 * changes no global state, and does not clear anything.
956 */
957 user->idx = clear_idx;
958 user->seq = clear_seq;
959 break;
960 case SEEK_END:
961 /* after the last record */
962 user->idx = log_next_idx;
963 user->seq = log_next_seq;
964 break;
965 default:
966 ret = -EINVAL;
967 }
968 logbuf_unlock_irq();
969 return ret;
970}
971
972static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
973{
974 struct devkmsg_user *user = file->private_data;
975 __poll_t ret = 0;
976
977 if (!user)
978 return EPOLLERR|EPOLLNVAL;
979
980 poll_wait(file, &log_wait, wait);
981
982 logbuf_lock_irq();
983 if (user->seq < log_next_seq) {
984 /* return error when data has vanished underneath us */
985 if (user->seq < log_first_seq)
986 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
987 else
988 ret = EPOLLIN|EPOLLRDNORM;
989 }
990 logbuf_unlock_irq();
991
992 return ret;
993}
994
995static int devkmsg_open(struct inode *inode, struct file *file)
996{
997 struct devkmsg_user *user;
998 int err;
999
1000 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
1001 return -EPERM;
1002
1003 /* write-only does not need any file context */
1004 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
1005 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
1006 SYSLOG_FROM_READER);
1007 if (err)
1008 return err;
1009 }
1010
1011 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
1012 if (!user)
1013 return -ENOMEM;
1014
1015 ratelimit_default_init(&user->rs);
1016 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
1017
1018 mutex_init(&user->lock);
1019
1020 logbuf_lock_irq();
1021 user->idx = log_first_idx;
1022 user->seq = log_first_seq;
1023 logbuf_unlock_irq();
1024
1025 file->private_data = user;
1026 return 0;
1027}
1028
1029static int devkmsg_release(struct inode *inode, struct file *file)
1030{
1031 struct devkmsg_user *user = file->private_data;
1032
1033 if (!user)
1034 return 0;
1035
1036 ratelimit_state_exit(&user->rs);
1037
1038 mutex_destroy(&user->lock);
1039 kfree(user);
1040 return 0;
1041}
1042
1043const struct file_operations kmsg_fops = {
1044 .open = devkmsg_open,
1045 .read = devkmsg_read,
1046 .write_iter = devkmsg_write,
1047 .llseek = devkmsg_llseek,
1048 .poll = devkmsg_poll,
1049 .release = devkmsg_release,
1050};
1051
1052#ifdef CONFIG_CRASH_CORE
1053/*
1054 * This appends the listed symbols to /proc/vmcore
1055 *
1056 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1057 * obtain access to symbols that are otherwise very difficult to locate. These
1058 * symbols are specifically used so that utilities can access and extract the
1059 * dmesg log from a vmcore file after a crash.
1060 */
1061void log_buf_vmcoreinfo_setup(void)
1062{
1063 VMCOREINFO_SYMBOL(log_buf);
1064 VMCOREINFO_SYMBOL(log_buf_len);
1065 VMCOREINFO_SYMBOL(log_first_idx);
1066 VMCOREINFO_SYMBOL(clear_idx);
1067 VMCOREINFO_SYMBOL(log_next_idx);
1068 /*
1069 * Export struct printk_log size and field offsets. User space tools can
1070 * parse it and detect any changes to structure down the line.
1071 */
1072 VMCOREINFO_STRUCT_SIZE(printk_log);
1073 VMCOREINFO_OFFSET(printk_log, ts_nsec);
1074 VMCOREINFO_OFFSET(printk_log, len);
1075 VMCOREINFO_OFFSET(printk_log, text_len);
1076 VMCOREINFO_OFFSET(printk_log, dict_len);
1077#ifdef CONFIG_PRINTK_CALLER
1078 VMCOREINFO_OFFSET(printk_log, caller_id);
1079#endif
1080}
1081#endif
1082
1083/* requested log_buf_len from kernel cmdline */
1084static unsigned long __initdata new_log_buf_len;
1085
1086/* we practice scaling the ring buffer by powers of 2 */
1087static void __init log_buf_len_update(u64 size)
1088{
1089 if (size > (u64)LOG_BUF_LEN_MAX) {
1090 size = (u64)LOG_BUF_LEN_MAX;
1091 pr_err("log_buf over 2G is not supported.\n");
1092 }
1093
1094 if (size)
1095 size = roundup_pow_of_two(size);
1096 if (size > log_buf_len)
1097 new_log_buf_len = (unsigned long)size;
1098}
1099
1100/* save requested log_buf_len since it's too early to process it */
1101static int __init log_buf_len_setup(char *str)
1102{
1103 u64 size;
1104
1105 if (!str)
1106 return -EINVAL;
1107
1108 size = memparse(str, &str);
1109
1110 log_buf_len_update(size);
1111
1112 return 0;
1113}
1114early_param("log_buf_len", log_buf_len_setup);
1115
1116#ifdef CONFIG_SMP
1117#define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1118
1119static void __init log_buf_add_cpu(void)
1120{
1121 unsigned int cpu_extra;
1122
1123 /*
1124 * archs should set up cpu_possible_bits properly with
1125 * set_cpu_possible() after setup_arch() but just in
1126 * case lets ensure this is valid.
1127 */
1128 if (num_possible_cpus() == 1)
1129 return;
1130
1131 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1132
1133 /* by default this will only continue through for large > 64 CPUs */
1134 if (cpu_extra <= __LOG_BUF_LEN / 2)
1135 return;
1136
1137 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1138 __LOG_CPU_MAX_BUF_LEN);
1139 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1140 cpu_extra);
1141 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1142
1143 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1144}
1145#else /* !CONFIG_SMP */
1146static inline void log_buf_add_cpu(void) {}
1147#endif /* CONFIG_SMP */
1148
1149void __init setup_log_buf(int early)
1150{
1151 unsigned long flags;
1152 char *new_log_buf;
1153 unsigned int free;
1154
1155 if (log_buf != __log_buf)
1156 return;
1157
1158 if (!early && !new_log_buf_len)
1159 log_buf_add_cpu();
1160
1161 if (!new_log_buf_len)
1162 return;
1163
1164 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1165 if (unlikely(!new_log_buf)) {
1166 pr_err("log_buf_len: %lu bytes not available\n",
1167 new_log_buf_len);
1168 return;
1169 }
1170
1171 logbuf_lock_irqsave(flags);
1172 log_buf_len = new_log_buf_len;
1173 log_buf = new_log_buf;
1174 new_log_buf_len = 0;
1175 free = __LOG_BUF_LEN - log_next_idx;
1176 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1177 logbuf_unlock_irqrestore(flags);
1178
1179 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1180 pr_info("early log buf free: %u(%u%%)\n",
1181 free, (free * 100) / __LOG_BUF_LEN);
1182}
1183
1184static bool __read_mostly ignore_loglevel;
1185
1186static int __init ignore_loglevel_setup(char *str)
1187{
1188 ignore_loglevel = true;
1189 pr_info("debug: ignoring loglevel setting.\n");
1190
1191 return 0;
1192}
1193
1194early_param("ignore_loglevel", ignore_loglevel_setup);
1195module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1196MODULE_PARM_DESC(ignore_loglevel,
1197 "ignore loglevel setting (prints all kernel messages to the console)");
1198
1199static bool suppress_message_printing(int level)
1200{
1201 return (level >= console_loglevel && !ignore_loglevel);
1202}
1203
1204#ifdef CONFIG_BOOT_PRINTK_DELAY
1205
1206static int boot_delay; /* msecs delay after each printk during bootup */
1207static unsigned long long loops_per_msec; /* based on boot_delay */
1208
1209static int __init boot_delay_setup(char *str)
1210{
1211 unsigned long lpj;
1212
1213 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1214 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1215
1216 get_option(&str, &boot_delay);
1217 if (boot_delay > 10 * 1000)
1218 boot_delay = 0;
1219
1220 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1221 "HZ: %d, loops_per_msec: %llu\n",
1222 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1223 return 0;
1224}
1225early_param("boot_delay", boot_delay_setup);
1226
1227static void boot_delay_msec(int level)
1228{
1229 unsigned long long k;
1230 unsigned long timeout;
1231
1232 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1233 || suppress_message_printing(level)) {
1234 return;
1235 }
1236
1237 k = (unsigned long long)loops_per_msec * boot_delay;
1238
1239 timeout = jiffies + msecs_to_jiffies(boot_delay);
1240 while (k) {
1241 k--;
1242 cpu_relax();
1243 /*
1244 * use (volatile) jiffies to prevent
1245 * compiler reduction; loop termination via jiffies
1246 * is secondary and may or may not happen.
1247 */
1248 if (time_after(jiffies, timeout))
1249 break;
1250 touch_nmi_watchdog();
1251 }
1252}
1253#else
1254static inline void boot_delay_msec(int level)
1255{
1256}
1257#endif
1258
1259static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1260module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1261
1262static size_t print_syslog(unsigned int level, char *buf)
1263{
1264 return sprintf(buf, "<%u>", level);
1265}
1266
1267static size_t print_time(u64 ts, char *buf)
1268{
1269 unsigned long rem_nsec = do_div(ts, 1000000000);
1270
1271 return sprintf(buf, "[%5lu.%06lu]",
1272 (unsigned long)ts, rem_nsec / 1000);
1273}
1274
1275#ifdef CONFIG_PRINTK_CALLER
1276static size_t print_caller(u32 id, char *buf)
1277{
1278 char caller[12];
1279
1280 snprintf(caller, sizeof(caller), "%c%u",
1281 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1282 return sprintf(buf, "[%6s]", caller);
1283}
1284#else
1285#define print_caller(id, buf) 0
1286#endif
1287
1288static size_t print_prefix(const struct printk_log *msg, bool syslog,
1289 bool time, char *buf)
1290{
1291 size_t len = 0;
1292
1293 if (syslog)
1294 len = print_syslog((msg->facility << 3) | msg->level, buf);
1295
1296 if (time)
1297 len += print_time(msg->ts_nsec, buf + len);
1298
1299 len += print_caller(msg->caller_id, buf + len);
1300
1301 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1302 buf[len++] = ' ';
1303 buf[len] = '\0';
1304 }
1305
1306 return len;
1307}
1308
1309static size_t msg_print_text(const struct printk_log *msg, bool syslog,
1310 bool time, char *buf, size_t size)
1311{
1312 const char *text = log_text(msg);
1313 size_t text_size = msg->text_len;
1314 size_t len = 0;
1315 char prefix[PREFIX_MAX];
1316 const size_t prefix_len = print_prefix(msg, syslog, time, prefix);
1317
1318 do {
1319 const char *next = memchr(text, '\n', text_size);
1320 size_t text_len;
1321
1322 if (next) {
1323 text_len = next - text;
1324 next++;
1325 text_size -= next - text;
1326 } else {
1327 text_len = text_size;
1328 }
1329
1330 if (buf) {
1331 if (prefix_len + text_len + 1 >= size - len)
1332 break;
1333
1334 memcpy(buf + len, prefix, prefix_len);
1335 len += prefix_len;
1336 memcpy(buf + len, text, text_len);
1337 len += text_len;
1338 buf[len++] = '\n';
1339 } else {
1340 /* SYSLOG_ACTION_* buffer size only calculation */
1341 len += prefix_len + text_len + 1;
1342 }
1343
1344 text = next;
1345 } while (text);
1346
1347 return len;
1348}
1349
1350static int syslog_print(char __user *buf, int size)
1351{
1352 char *text;
1353 struct printk_log *msg;
1354 int len = 0;
1355
1356 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1357 if (!text)
1358 return -ENOMEM;
1359
1360 while (size > 0) {
1361 size_t n;
1362 size_t skip;
1363
1364 logbuf_lock_irq();
1365 if (syslog_seq < log_first_seq) {
1366 /* messages are gone, move to first one */
1367 syslog_seq = log_first_seq;
1368 syslog_idx = log_first_idx;
1369 syslog_partial = 0;
1370 }
1371 if (syslog_seq == log_next_seq) {
1372 logbuf_unlock_irq();
1373 break;
1374 }
1375
1376 /*
1377 * To keep reading/counting partial line consistent,
1378 * use printk_time value as of the beginning of a line.
1379 */
1380 if (!syslog_partial)
1381 syslog_time = printk_time;
1382
1383 skip = syslog_partial;
1384 msg = log_from_idx(syslog_idx);
1385 n = msg_print_text(msg, true, syslog_time, text,
1386 LOG_LINE_MAX + PREFIX_MAX);
1387 if (n - syslog_partial <= size) {
1388 /* message fits into buffer, move forward */
1389 syslog_idx = log_next(syslog_idx);
1390 syslog_seq++;
1391 n -= syslog_partial;
1392 syslog_partial = 0;
1393 } else if (!len){
1394 /* partial read(), remember position */
1395 n = size;
1396 syslog_partial += n;
1397 } else
1398 n = 0;
1399 logbuf_unlock_irq();
1400
1401 if (!n)
1402 break;
1403
1404 if (copy_to_user(buf, text + skip, n)) {
1405 if (!len)
1406 len = -EFAULT;
1407 break;
1408 }
1409
1410 len += n;
1411 size -= n;
1412 buf += n;
1413 }
1414
1415 kfree(text);
1416 return len;
1417}
1418
1419static int syslog_print_all(char __user *buf, int size, bool clear)
1420{
1421 char *text;
1422 int len = 0;
1423 u64 next_seq;
1424 u64 seq;
1425 u32 idx;
1426 bool time;
1427
1428 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1429 if (!text)
1430 return -ENOMEM;
1431
1432 time = printk_time;
1433 logbuf_lock_irq();
1434 /*
1435 * Find first record that fits, including all following records,
1436 * into the user-provided buffer for this dump.
1437 */
1438 seq = clear_seq;
1439 idx = clear_idx;
1440 while (seq < log_next_seq) {
1441 struct printk_log *msg = log_from_idx(idx);
1442
1443 len += msg_print_text(msg, true, time, NULL, 0);
1444 idx = log_next(idx);
1445 seq++;
1446 }
1447
1448 /* move first record forward until length fits into the buffer */
1449 seq = clear_seq;
1450 idx = clear_idx;
1451 while (len > size && seq < log_next_seq) {
1452 struct printk_log *msg = log_from_idx(idx);
1453
1454 len -= msg_print_text(msg, true, time, NULL, 0);
1455 idx = log_next(idx);
1456 seq++;
1457 }
1458
1459 /* last message fitting into this dump */
1460 next_seq = log_next_seq;
1461
1462 len = 0;
1463 while (len >= 0 && seq < next_seq) {
1464 struct printk_log *msg = log_from_idx(idx);
1465 int textlen = msg_print_text(msg, true, time, text,
1466 LOG_LINE_MAX + PREFIX_MAX);
1467
1468 idx = log_next(idx);
1469 seq++;
1470
1471 logbuf_unlock_irq();
1472 if (copy_to_user(buf + len, text, textlen))
1473 len = -EFAULT;
1474 else
1475 len += textlen;
1476 logbuf_lock_irq();
1477
1478 if (seq < log_first_seq) {
1479 /* messages are gone, move to next one */
1480 seq = log_first_seq;
1481 idx = log_first_idx;
1482 }
1483 }
1484
1485 if (clear) {
1486 clear_seq = log_next_seq;
1487 clear_idx = log_next_idx;
1488 }
1489 logbuf_unlock_irq();
1490
1491 kfree(text);
1492 return len;
1493}
1494
1495static void syslog_clear(void)
1496{
1497 logbuf_lock_irq();
1498 clear_seq = log_next_seq;
1499 clear_idx = log_next_idx;
1500 logbuf_unlock_irq();
1501}
1502
1503int do_syslog(int type, char __user *buf, int len, int source)
1504{
1505 bool clear = false;
1506 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1507 int error;
1508
1509 error = check_syslog_permissions(type, source);
1510 if (error)
1511 return error;
1512
1513 switch (type) {
1514 case SYSLOG_ACTION_CLOSE: /* Close log */
1515 break;
1516 case SYSLOG_ACTION_OPEN: /* Open log */
1517 break;
1518 case SYSLOG_ACTION_READ: /* Read from log */
1519 if (!buf || len < 0)
1520 return -EINVAL;
1521 if (!len)
1522 return 0;
1523 if (!access_ok(buf, len))
1524 return -EFAULT;
1525 error = wait_event_interruptible(log_wait,
1526 syslog_seq != log_next_seq);
1527 if (error)
1528 return error;
1529 error = syslog_print(buf, len);
1530 break;
1531 /* Read/clear last kernel messages */
1532 case SYSLOG_ACTION_READ_CLEAR:
1533 clear = true;
1534 /* FALL THRU */
1535 /* Read last kernel messages */
1536 case SYSLOG_ACTION_READ_ALL:
1537 if (!buf || len < 0)
1538 return -EINVAL;
1539 if (!len)
1540 return 0;
1541 if (!access_ok(buf, len))
1542 return -EFAULT;
1543 error = syslog_print_all(buf, len, clear);
1544 break;
1545 /* Clear ring buffer */
1546 case SYSLOG_ACTION_CLEAR:
1547 syslog_clear();
1548 break;
1549 /* Disable logging to console */
1550 case SYSLOG_ACTION_CONSOLE_OFF:
1551 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1552 saved_console_loglevel = console_loglevel;
1553 console_loglevel = minimum_console_loglevel;
1554 break;
1555 /* Enable logging to console */
1556 case SYSLOG_ACTION_CONSOLE_ON:
1557 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1558 console_loglevel = saved_console_loglevel;
1559 saved_console_loglevel = LOGLEVEL_DEFAULT;
1560 }
1561 break;
1562 /* Set level of messages printed to console */
1563 case SYSLOG_ACTION_CONSOLE_LEVEL:
1564 if (len < 1 || len > 8)
1565 return -EINVAL;
1566 if (len < minimum_console_loglevel)
1567 len = minimum_console_loglevel;
1568 console_loglevel = len;
1569 /* Implicitly re-enable logging to console */
1570 saved_console_loglevel = LOGLEVEL_DEFAULT;
1571 break;
1572 /* Number of chars in the log buffer */
1573 case SYSLOG_ACTION_SIZE_UNREAD:
1574 logbuf_lock_irq();
1575 if (syslog_seq < log_first_seq) {
1576 /* messages are gone, move to first one */
1577 syslog_seq = log_first_seq;
1578 syslog_idx = log_first_idx;
1579 syslog_partial = 0;
1580 }
1581 if (source == SYSLOG_FROM_PROC) {
1582 /*
1583 * Short-cut for poll(/"proc/kmsg") which simply checks
1584 * for pending data, not the size; return the count of
1585 * records, not the length.
1586 */
1587 error = log_next_seq - syslog_seq;
1588 } else {
1589 u64 seq = syslog_seq;
1590 u32 idx = syslog_idx;
1591 bool time = syslog_partial ? syslog_time : printk_time;
1592
1593 while (seq < log_next_seq) {
1594 struct printk_log *msg = log_from_idx(idx);
1595
1596 error += msg_print_text(msg, true, time, NULL,
1597 0);
1598 time = printk_time;
1599 idx = log_next(idx);
1600 seq++;
1601 }
1602 error -= syslog_partial;
1603 }
1604 logbuf_unlock_irq();
1605 break;
1606 /* Size of the log buffer */
1607 case SYSLOG_ACTION_SIZE_BUFFER:
1608 error = log_buf_len;
1609 break;
1610 default:
1611 error = -EINVAL;
1612 break;
1613 }
1614
1615 return error;
1616}
1617
1618SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1619{
1620 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1621}
1622
1623/*
1624 * Special console_lock variants that help to reduce the risk of soft-lockups.
1625 * They allow to pass console_lock to another printk() call using a busy wait.
1626 */
1627
1628#ifdef CONFIG_LOCKDEP
1629static struct lockdep_map console_owner_dep_map = {
1630 .name = "console_owner"
1631};
1632#endif
1633
1634static DEFINE_RAW_SPINLOCK(console_owner_lock);
1635static struct task_struct *console_owner;
1636static bool console_waiter;
1637
1638/**
1639 * console_lock_spinning_enable - mark beginning of code where another
1640 * thread might safely busy wait
1641 *
1642 * This basically converts console_lock into a spinlock. This marks
1643 * the section where the console_lock owner can not sleep, because
1644 * there may be a waiter spinning (like a spinlock). Also it must be
1645 * ready to hand over the lock at the end of the section.
1646 */
1647static void console_lock_spinning_enable(void)
1648{
1649 raw_spin_lock(&console_owner_lock);
1650 console_owner = current;
1651 raw_spin_unlock(&console_owner_lock);
1652
1653 /* The waiter may spin on us after setting console_owner */
1654 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1655}
1656
1657/**
1658 * console_lock_spinning_disable_and_check - mark end of code where another
1659 * thread was able to busy wait and check if there is a waiter
1660 *
1661 * This is called at the end of the section where spinning is allowed.
1662 * It has two functions. First, it is a signal that it is no longer
1663 * safe to start busy waiting for the lock. Second, it checks if
1664 * there is a busy waiter and passes the lock rights to her.
1665 *
1666 * Important: Callers lose the lock if there was a busy waiter.
1667 * They must not touch items synchronized by console_lock
1668 * in this case.
1669 *
1670 * Return: 1 if the lock rights were passed, 0 otherwise.
1671 */
1672static int console_lock_spinning_disable_and_check(void)
1673{
1674 int waiter;
1675
1676 raw_spin_lock(&console_owner_lock);
1677 waiter = READ_ONCE(console_waiter);
1678 console_owner = NULL;
1679 raw_spin_unlock(&console_owner_lock);
1680
1681 if (!waiter) {
1682 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1683 return 0;
1684 }
1685
1686 /* The waiter is now free to continue */
1687 WRITE_ONCE(console_waiter, false);
1688
1689 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1690
1691 /*
1692 * Hand off console_lock to waiter. The waiter will perform
1693 * the up(). After this, the waiter is the console_lock owner.
1694 */
1695 mutex_release(&console_lock_dep_map, 1, _THIS_IP_);
1696 return 1;
1697}
1698
1699/**
1700 * console_trylock_spinning - try to get console_lock by busy waiting
1701 *
1702 * This allows to busy wait for the console_lock when the current
1703 * owner is running in specially marked sections. It means that
1704 * the current owner is running and cannot reschedule until it
1705 * is ready to lose the lock.
1706 *
1707 * Return: 1 if we got the lock, 0 othrewise
1708 */
1709static int console_trylock_spinning(void)
1710{
1711 struct task_struct *owner = NULL;
1712 bool waiter;
1713 bool spin = false;
1714 unsigned long flags;
1715
1716 if (console_trylock())
1717 return 1;
1718
1719 printk_safe_enter_irqsave(flags);
1720
1721 raw_spin_lock(&console_owner_lock);
1722 owner = READ_ONCE(console_owner);
1723 waiter = READ_ONCE(console_waiter);
1724 if (!waiter && owner && owner != current) {
1725 WRITE_ONCE(console_waiter, true);
1726 spin = true;
1727 }
1728 raw_spin_unlock(&console_owner_lock);
1729
1730 /*
1731 * If there is an active printk() writing to the
1732 * consoles, instead of having it write our data too,
1733 * see if we can offload that load from the active
1734 * printer, and do some printing ourselves.
1735 * Go into a spin only if there isn't already a waiter
1736 * spinning, and there is an active printer, and
1737 * that active printer isn't us (recursive printk?).
1738 */
1739 if (!spin) {
1740 printk_safe_exit_irqrestore(flags);
1741 return 0;
1742 }
1743
1744 /* We spin waiting for the owner to release us */
1745 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1746 /* Owner will clear console_waiter on hand off */
1747 while (READ_ONCE(console_waiter))
1748 cpu_relax();
1749 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1750
1751 printk_safe_exit_irqrestore(flags);
1752 /*
1753 * The owner passed the console lock to us.
1754 * Since we did not spin on console lock, annotate
1755 * this as a trylock. Otherwise lockdep will
1756 * complain.
1757 */
1758 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1759
1760 return 1;
1761}
1762
1763/*
1764 * Call the console drivers, asking them to write out
1765 * log_buf[start] to log_buf[end - 1].
1766 * The console_lock must be held.
1767 */
1768static void call_console_drivers(const char *ext_text, size_t ext_len,
1769 const char *text, size_t len)
1770{
1771 struct console *con;
1772
1773 trace_console_rcuidle(text, len);
1774
1775 if (!console_drivers)
1776 return;
1777
1778 for_each_console(con) {
1779 if (exclusive_console && con != exclusive_console)
1780 continue;
1781 if (!(con->flags & CON_ENABLED))
1782 continue;
1783 if (!con->write)
1784 continue;
1785 if (!cpu_online(smp_processor_id()) &&
1786 !(con->flags & CON_ANYTIME))
1787 continue;
1788 if (con->flags & CON_EXTENDED)
1789 con->write(con, ext_text, ext_len);
1790 else
1791 con->write(con, text, len);
1792 }
1793}
1794
1795int printk_delay_msec __read_mostly;
1796
1797static inline void printk_delay(void)
1798{
1799 if (unlikely(printk_delay_msec)) {
1800 int m = printk_delay_msec;
1801
1802 while (m--) {
1803 mdelay(1);
1804 touch_nmi_watchdog();
1805 }
1806 }
1807}
1808
1809static inline u32 printk_caller_id(void)
1810{
1811 return in_task() ? task_pid_nr(current) :
1812 0x80000000 + raw_smp_processor_id();
1813}
1814
1815/*
1816 * Continuation lines are buffered, and not committed to the record buffer
1817 * until the line is complete, or a race forces it. The line fragments
1818 * though, are printed immediately to the consoles to ensure everything has
1819 * reached the console in case of a kernel crash.
1820 */
1821static struct cont {
1822 char buf[LOG_LINE_MAX];
1823 size_t len; /* length == 0 means unused buffer */
1824 u32 caller_id; /* printk_caller_id() of first print */
1825 u64 ts_nsec; /* time of first print */
1826 u8 level; /* log level of first message */
1827 u8 facility; /* log facility of first message */
1828 enum log_flags flags; /* prefix, newline flags */
1829} cont;
1830
1831static void cont_flush(void)
1832{
1833 if (cont.len == 0)
1834 return;
1835
1836 log_store(cont.caller_id, cont.facility, cont.level, cont.flags,
1837 cont.ts_nsec, NULL, 0, cont.buf, cont.len);
1838 cont.len = 0;
1839}
1840
1841static bool cont_add(u32 caller_id, int facility, int level,
1842 enum log_flags flags, const char *text, size_t len)
1843{
1844 /* If the line gets too long, split it up in separate records. */
1845 if (cont.len + len > sizeof(cont.buf)) {
1846 cont_flush();
1847 return false;
1848 }
1849
1850 if (!cont.len) {
1851 cont.facility = facility;
1852 cont.level = level;
1853 cont.caller_id = caller_id;
1854 cont.ts_nsec = local_clock();
1855 cont.flags = flags;
1856 }
1857
1858 memcpy(cont.buf + cont.len, text, len);
1859 cont.len += len;
1860
1861 // The original flags come from the first line,
1862 // but later continuations can add a newline.
1863 if (flags & LOG_NEWLINE) {
1864 cont.flags |= LOG_NEWLINE;
1865 cont_flush();
1866 }
1867
1868 return true;
1869}
1870
1871static size_t log_output(int facility, int level, enum log_flags lflags, const char *dict, size_t dictlen, char *text, size_t text_len)
1872{
1873 const u32 caller_id = printk_caller_id();
1874
1875 /*
1876 * If an earlier line was buffered, and we're a continuation
1877 * write from the same context, try to add it to the buffer.
1878 */
1879 if (cont.len) {
1880 if (cont.caller_id == caller_id && (lflags & LOG_CONT)) {
1881 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1882 return text_len;
1883 }
1884 /* Otherwise, make sure it's flushed */
1885 cont_flush();
1886 }
1887
1888 /* Skip empty continuation lines that couldn't be added - they just flush */
1889 if (!text_len && (lflags & LOG_CONT))
1890 return 0;
1891
1892 /* If it doesn't end in a newline, try to buffer the current line */
1893 if (!(lflags & LOG_NEWLINE)) {
1894 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1895 return text_len;
1896 }
1897
1898 /* Store it in the record log */
1899 return log_store(caller_id, facility, level, lflags, 0,
1900 dict, dictlen, text, text_len);
1901}
1902
1903/* Must be called under logbuf_lock. */
1904int vprintk_store(int facility, int level,
1905 const char *dict, size_t dictlen,
1906 const char *fmt, va_list args)
1907{
1908 static char textbuf[LOG_LINE_MAX];
1909 char *text = textbuf;
1910 size_t text_len;
1911 enum log_flags lflags = 0;
1912
1913 /*
1914 * The printf needs to come first; we need the syslog
1915 * prefix which might be passed-in as a parameter.
1916 */
1917 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1918
1919 /* mark and strip a trailing newline */
1920 if (text_len && text[text_len-1] == '\n') {
1921 text_len--;
1922 lflags |= LOG_NEWLINE;
1923 }
1924
1925 /* strip kernel syslog prefix and extract log level or control flags */
1926 if (facility == 0) {
1927 int kern_level;
1928
1929 while ((kern_level = printk_get_level(text)) != 0) {
1930 switch (kern_level) {
1931 case '0' ... '7':
1932 if (level == LOGLEVEL_DEFAULT)
1933 level = kern_level - '0';
1934 break;
1935 case 'c': /* KERN_CONT */
1936 lflags |= LOG_CONT;
1937 }
1938
1939 text_len -= 2;
1940 text += 2;
1941 }
1942 }
1943
1944 if (level == LOGLEVEL_DEFAULT)
1945 level = default_message_loglevel;
1946
1947 if (dict)
1948 lflags |= LOG_NEWLINE;
1949
1950 return log_output(facility, level, lflags,
1951 dict, dictlen, text, text_len);
1952}
1953
1954asmlinkage int vprintk_emit(int facility, int level,
1955 const char *dict, size_t dictlen,
1956 const char *fmt, va_list args)
1957{
1958 int printed_len;
1959 bool in_sched = false, pending_output;
1960 unsigned long flags;
1961 u64 curr_log_seq;
1962
1963 /* Suppress unimportant messages after panic happens */
1964 if (unlikely(suppress_printk))
1965 return 0;
1966
1967 if (level == LOGLEVEL_SCHED) {
1968 level = LOGLEVEL_DEFAULT;
1969 in_sched = true;
1970 }
1971
1972 boot_delay_msec(level);
1973 printk_delay();
1974
1975 /* This stops the holder of console_sem just where we want him */
1976 logbuf_lock_irqsave(flags);
1977 curr_log_seq = log_next_seq;
1978 printed_len = vprintk_store(facility, level, dict, dictlen, fmt, args);
1979 pending_output = (curr_log_seq != log_next_seq);
1980 logbuf_unlock_irqrestore(flags);
1981
1982 /* If called from the scheduler, we can not call up(). */
1983 if (!in_sched && pending_output) {
1984 /*
1985 * Disable preemption to avoid being preempted while holding
1986 * console_sem which would prevent anyone from printing to
1987 * console
1988 */
1989 preempt_disable();
1990 /*
1991 * Try to acquire and then immediately release the console
1992 * semaphore. The release will print out buffers and wake up
1993 * /dev/kmsg and syslog() users.
1994 */
1995 if (console_trylock_spinning())
1996 console_unlock();
1997 preempt_enable();
1998 }
1999
2000 if (pending_output)
2001 wake_up_klogd();
2002 return printed_len;
2003}
2004EXPORT_SYMBOL(vprintk_emit);
2005
2006asmlinkage int vprintk(const char *fmt, va_list args)
2007{
2008 return vprintk_func(fmt, args);
2009}
2010EXPORT_SYMBOL(vprintk);
2011
2012int vprintk_default(const char *fmt, va_list args)
2013{
2014 int r;
2015
2016#ifdef CONFIG_KGDB_KDB
2017 /* Allow to pass printk() to kdb but avoid a recursion. */
2018 if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
2019 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
2020 return r;
2021 }
2022#endif
2023 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
2024
2025 return r;
2026}
2027EXPORT_SYMBOL_GPL(vprintk_default);
2028
2029/**
2030 * printk - print a kernel message
2031 * @fmt: format string
2032 *
2033 * This is printk(). It can be called from any context. We want it to work.
2034 *
2035 * We try to grab the console_lock. If we succeed, it's easy - we log the
2036 * output and call the console drivers. If we fail to get the semaphore, we
2037 * place the output into the log buffer and return. The current holder of
2038 * the console_sem will notice the new output in console_unlock(); and will
2039 * send it to the consoles before releasing the lock.
2040 *
2041 * One effect of this deferred printing is that code which calls printk() and
2042 * then changes console_loglevel may break. This is because console_loglevel
2043 * is inspected when the actual printing occurs.
2044 *
2045 * See also:
2046 * printf(3)
2047 *
2048 * See the vsnprintf() documentation for format string extensions over C99.
2049 */
2050asmlinkage __visible int printk(const char *fmt, ...)
2051{
2052 va_list args;
2053 int r;
2054
2055 va_start(args, fmt);
2056 r = vprintk_func(fmt, args);
2057 va_end(args);
2058
2059 return r;
2060}
2061EXPORT_SYMBOL(printk);
2062
2063#else /* CONFIG_PRINTK */
2064
2065#define LOG_LINE_MAX 0
2066#define PREFIX_MAX 0
2067#define printk_time false
2068
2069static u64 syslog_seq;
2070static u32 syslog_idx;
2071static u64 console_seq;
2072static u32 console_idx;
2073static u64 exclusive_console_stop_seq;
2074static u64 log_first_seq;
2075static u32 log_first_idx;
2076static u64 log_next_seq;
2077static char *log_text(const struct printk_log *msg) { return NULL; }
2078static char *log_dict(const struct printk_log *msg) { return NULL; }
2079static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2080static u32 log_next(u32 idx) { return 0; }
2081static ssize_t msg_print_ext_header(char *buf, size_t size,
2082 struct printk_log *msg,
2083 u64 seq) { return 0; }
2084static ssize_t msg_print_ext_body(char *buf, size_t size,
2085 char *dict, size_t dict_len,
2086 char *text, size_t text_len) { return 0; }
2087static void console_lock_spinning_enable(void) { }
2088static int console_lock_spinning_disable_and_check(void) { return 0; }
2089static void call_console_drivers(const char *ext_text, size_t ext_len,
2090 const char *text, size_t len) {}
2091static size_t msg_print_text(const struct printk_log *msg, bool syslog,
2092 bool time, char *buf, size_t size) { return 0; }
2093static bool suppress_message_printing(int level) { return false; }
2094
2095#endif /* CONFIG_PRINTK */
2096
2097#ifdef CONFIG_EARLY_PRINTK
2098struct console *early_console;
2099
2100asmlinkage __visible void early_printk(const char *fmt, ...)
2101{
2102 va_list ap;
2103 char buf[512];
2104 int n;
2105
2106 if (!early_console)
2107 return;
2108
2109 va_start(ap, fmt);
2110 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2111 va_end(ap);
2112
2113 early_console->write(early_console, buf, n);
2114}
2115#endif
2116
2117static int __add_preferred_console(char *name, int idx, char *options,
2118 char *brl_options)
2119{
2120 struct console_cmdline *c;
2121 int i;
2122
2123 /*
2124 * See if this tty is not yet registered, and
2125 * if we have a slot free.
2126 */
2127 for (i = 0, c = console_cmdline;
2128 i < MAX_CMDLINECONSOLES && c->name[0];
2129 i++, c++) {
2130 if (strcmp(c->name, name) == 0 && c->index == idx) {
2131 if (!brl_options)
2132 preferred_console = i;
2133 return 0;
2134 }
2135 }
2136 if (i == MAX_CMDLINECONSOLES)
2137 return -E2BIG;
2138 if (!brl_options)
2139 preferred_console = i;
2140 strlcpy(c->name, name, sizeof(c->name));
2141 c->options = options;
2142 braille_set_options(c, brl_options);
2143
2144 c->index = idx;
2145 return 0;
2146}
2147
2148static int __init console_msg_format_setup(char *str)
2149{
2150 if (!strcmp(str, "syslog"))
2151 console_msg_format = MSG_FORMAT_SYSLOG;
2152 if (!strcmp(str, "default"))
2153 console_msg_format = MSG_FORMAT_DEFAULT;
2154 return 1;
2155}
2156__setup("console_msg_format=", console_msg_format_setup);
2157
2158/*
2159 * Set up a console. Called via do_early_param() in init/main.c
2160 * for each "console=" parameter in the boot command line.
2161 */
2162static int __init console_setup(char *str)
2163{
2164 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2165 char *s, *options, *brl_options = NULL;
2166 int idx;
2167
2168 if (_braille_console_setup(&str, &brl_options))
2169 return 1;
2170
2171 /*
2172 * Decode str into name, index, options.
2173 */
2174 if (str[0] >= '0' && str[0] <= '9') {
2175 strcpy(buf, "ttyS");
2176 strncpy(buf + 4, str, sizeof(buf) - 5);
2177 } else {
2178 strncpy(buf, str, sizeof(buf) - 1);
2179 }
2180 buf[sizeof(buf) - 1] = 0;
2181 options = strchr(str, ',');
2182 if (options)
2183 *(options++) = 0;
2184#ifdef __sparc__
2185 if (!strcmp(str, "ttya"))
2186 strcpy(buf, "ttyS0");
2187 if (!strcmp(str, "ttyb"))
2188 strcpy(buf, "ttyS1");
2189#endif
2190 for (s = buf; *s; s++)
2191 if (isdigit(*s) || *s == ',')
2192 break;
2193 idx = simple_strtoul(s, NULL, 10);
2194 *s = 0;
2195
2196 __add_preferred_console(buf, idx, options, brl_options);
2197 console_set_on_cmdline = 1;
2198 return 1;
2199}
2200__setup("console=", console_setup);
2201
2202/**
2203 * add_preferred_console - add a device to the list of preferred consoles.
2204 * @name: device name
2205 * @idx: device index
2206 * @options: options for this console
2207 *
2208 * The last preferred console added will be used for kernel messages
2209 * and stdin/out/err for init. Normally this is used by console_setup
2210 * above to handle user-supplied console arguments; however it can also
2211 * be used by arch-specific code either to override the user or more
2212 * commonly to provide a default console (ie from PROM variables) when
2213 * the user has not supplied one.
2214 */
2215int add_preferred_console(char *name, int idx, char *options)
2216{
2217 return __add_preferred_console(name, idx, options, NULL);
2218}
2219
2220bool console_suspend_enabled = true;
2221EXPORT_SYMBOL(console_suspend_enabled);
2222
2223static int __init console_suspend_disable(char *str)
2224{
2225 console_suspend_enabled = false;
2226 return 1;
2227}
2228__setup("no_console_suspend", console_suspend_disable);
2229module_param_named(console_suspend, console_suspend_enabled,
2230 bool, S_IRUGO | S_IWUSR);
2231MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2232 " and hibernate operations");
2233
2234/**
2235 * suspend_console - suspend the console subsystem
2236 *
2237 * This disables printk() while we go into suspend states
2238 */
2239void suspend_console(void)
2240{
2241 if (!console_suspend_enabled)
2242 return;
2243 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2244 console_lock();
2245 console_suspended = 1;
2246 up_console_sem();
2247}
2248
2249void resume_console(void)
2250{
2251 if (!console_suspend_enabled)
2252 return;
2253 down_console_sem();
2254 console_suspended = 0;
2255 console_unlock();
2256}
2257
2258/**
2259 * console_cpu_notify - print deferred console messages after CPU hotplug
2260 * @cpu: unused
2261 *
2262 * If printk() is called from a CPU that is not online yet, the messages
2263 * will be printed on the console only if there are CON_ANYTIME consoles.
2264 * This function is called when a new CPU comes online (or fails to come
2265 * up) or goes offline.
2266 */
2267static int console_cpu_notify(unsigned int cpu)
2268{
2269 if (!cpuhp_tasks_frozen) {
2270 /* If trylock fails, someone else is doing the printing */
2271 if (console_trylock())
2272 console_unlock();
2273 }
2274 return 0;
2275}
2276
2277/**
2278 * console_lock - lock the console system for exclusive use.
2279 *
2280 * Acquires a lock which guarantees that the caller has
2281 * exclusive access to the console system and the console_drivers list.
2282 *
2283 * Can sleep, returns nothing.
2284 */
2285void console_lock(void)
2286{
2287 might_sleep();
2288
2289 down_console_sem();
2290 if (console_suspended)
2291 return;
2292 console_locked = 1;
2293 console_may_schedule = 1;
2294}
2295EXPORT_SYMBOL(console_lock);
2296
2297/**
2298 * console_trylock - try to lock the console system for exclusive use.
2299 *
2300 * Try to acquire a lock which guarantees that the caller has exclusive
2301 * access to the console system and the console_drivers list.
2302 *
2303 * returns 1 on success, and 0 on failure to acquire the lock.
2304 */
2305int console_trylock(void)
2306{
2307 if (down_trylock_console_sem())
2308 return 0;
2309 if (console_suspended) {
2310 up_console_sem();
2311 return 0;
2312 }
2313 console_locked = 1;
2314 console_may_schedule = 0;
2315 return 1;
2316}
2317EXPORT_SYMBOL(console_trylock);
2318
2319int is_console_locked(void)
2320{
2321 return console_locked;
2322}
2323EXPORT_SYMBOL(is_console_locked);
2324
2325/*
2326 * Check if we have any console that is capable of printing while cpu is
2327 * booting or shutting down. Requires console_sem.
2328 */
2329static int have_callable_console(void)
2330{
2331 struct console *con;
2332
2333 for_each_console(con)
2334 if ((con->flags & CON_ENABLED) &&
2335 (con->flags & CON_ANYTIME))
2336 return 1;
2337
2338 return 0;
2339}
2340
2341/*
2342 * Can we actually use the console at this time on this cpu?
2343 *
2344 * Console drivers may assume that per-cpu resources have been allocated. So
2345 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2346 * call them until this CPU is officially up.
2347 */
2348static inline int can_use_console(void)
2349{
2350 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2351}
2352
2353/**
2354 * console_unlock - unlock the console system
2355 *
2356 * Releases the console_lock which the caller holds on the console system
2357 * and the console driver list.
2358 *
2359 * While the console_lock was held, console output may have been buffered
2360 * by printk(). If this is the case, console_unlock(); emits
2361 * the output prior to releasing the lock.
2362 *
2363 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2364 *
2365 * console_unlock(); may be called from any context.
2366 */
2367void console_unlock(void)
2368{
2369 static char ext_text[CONSOLE_EXT_LOG_MAX];
2370 static char text[LOG_LINE_MAX + PREFIX_MAX];
2371 unsigned long flags;
2372 bool do_cond_resched, retry;
2373
2374 if (console_suspended) {
2375 up_console_sem();
2376 return;
2377 }
2378
2379 /*
2380 * Console drivers are called with interrupts disabled, so
2381 * @console_may_schedule should be cleared before; however, we may
2382 * end up dumping a lot of lines, for example, if called from
2383 * console registration path, and should invoke cond_resched()
2384 * between lines if allowable. Not doing so can cause a very long
2385 * scheduling stall on a slow console leading to RCU stall and
2386 * softlockup warnings which exacerbate the issue with more
2387 * messages practically incapacitating the system.
2388 *
2389 * console_trylock() is not able to detect the preemptive
2390 * context reliably. Therefore the value must be stored before
2391 * and cleared after the the "again" goto label.
2392 */
2393 do_cond_resched = console_may_schedule;
2394again:
2395 console_may_schedule = 0;
2396
2397 /*
2398 * We released the console_sem lock, so we need to recheck if
2399 * cpu is online and (if not) is there at least one CON_ANYTIME
2400 * console.
2401 */
2402 if (!can_use_console()) {
2403 console_locked = 0;
2404 up_console_sem();
2405 return;
2406 }
2407
2408 for (;;) {
2409 struct printk_log *msg;
2410 size_t ext_len = 0;
2411 size_t len;
2412
2413 printk_safe_enter_irqsave(flags);
2414 raw_spin_lock(&logbuf_lock);
2415 if (console_seq < log_first_seq) {
2416 len = sprintf(text,
2417 "** %llu printk messages dropped **\n",
2418 log_first_seq - console_seq);
2419
2420 /* messages are gone, move to first one */
2421 console_seq = log_first_seq;
2422 console_idx = log_first_idx;
2423 } else {
2424 len = 0;
2425 }
2426skip:
2427 if (console_seq == log_next_seq)
2428 break;
2429
2430 msg = log_from_idx(console_idx);
2431 if (suppress_message_printing(msg->level)) {
2432 /*
2433 * Skip record we have buffered and already printed
2434 * directly to the console when we received it, and
2435 * record that has level above the console loglevel.
2436 */
2437 console_idx = log_next(console_idx);
2438 console_seq++;
2439 goto skip;
2440 }
2441
2442 /* Output to all consoles once old messages replayed. */
2443 if (unlikely(exclusive_console &&
2444 console_seq >= exclusive_console_stop_seq)) {
2445 exclusive_console = NULL;
2446 }
2447
2448 len += msg_print_text(msg,
2449 console_msg_format & MSG_FORMAT_SYSLOG,
2450 printk_time, text + len, sizeof(text) - len);
2451 if (nr_ext_console_drivers) {
2452 ext_len = msg_print_ext_header(ext_text,
2453 sizeof(ext_text),
2454 msg, console_seq);
2455 ext_len += msg_print_ext_body(ext_text + ext_len,
2456 sizeof(ext_text) - ext_len,
2457 log_dict(msg), msg->dict_len,
2458 log_text(msg), msg->text_len);
2459 }
2460 console_idx = log_next(console_idx);
2461 console_seq++;
2462 raw_spin_unlock(&logbuf_lock);
2463
2464 /*
2465 * While actively printing out messages, if another printk()
2466 * were to occur on another CPU, it may wait for this one to
2467 * finish. This task can not be preempted if there is a
2468 * waiter waiting to take over.
2469 */
2470 console_lock_spinning_enable();
2471
2472 stop_critical_timings(); /* don't trace print latency */
2473 call_console_drivers(ext_text, ext_len, text, len);
2474 start_critical_timings();
2475
2476 if (console_lock_spinning_disable_and_check()) {
2477 printk_safe_exit_irqrestore(flags);
2478 return;
2479 }
2480
2481 printk_safe_exit_irqrestore(flags);
2482
2483 if (do_cond_resched)
2484 cond_resched();
2485 }
2486
2487 console_locked = 0;
2488
2489 raw_spin_unlock(&logbuf_lock);
2490
2491 up_console_sem();
2492
2493 /*
2494 * Someone could have filled up the buffer again, so re-check if there's
2495 * something to flush. In case we cannot trylock the console_sem again,
2496 * there's a new owner and the console_unlock() from them will do the
2497 * flush, no worries.
2498 */
2499 raw_spin_lock(&logbuf_lock);
2500 retry = console_seq != log_next_seq;
2501 raw_spin_unlock(&logbuf_lock);
2502 printk_safe_exit_irqrestore(flags);
2503
2504 if (retry && console_trylock())
2505 goto again;
2506}
2507EXPORT_SYMBOL(console_unlock);
2508
2509/**
2510 * console_conditional_schedule - yield the CPU if required
2511 *
2512 * If the console code is currently allowed to sleep, and
2513 * if this CPU should yield the CPU to another task, do
2514 * so here.
2515 *
2516 * Must be called within console_lock();.
2517 */
2518void __sched console_conditional_schedule(void)
2519{
2520 if (console_may_schedule)
2521 cond_resched();
2522}
2523EXPORT_SYMBOL(console_conditional_schedule);
2524
2525void console_unblank(void)
2526{
2527 struct console *c;
2528
2529 /*
2530 * console_unblank can no longer be called in interrupt context unless
2531 * oops_in_progress is set to 1..
2532 */
2533 if (oops_in_progress) {
2534 if (down_trylock_console_sem() != 0)
2535 return;
2536 } else
2537 console_lock();
2538
2539 console_locked = 1;
2540 console_may_schedule = 0;
2541 for_each_console(c)
2542 if ((c->flags & CON_ENABLED) && c->unblank)
2543 c->unblank();
2544 console_unlock();
2545}
2546
2547/**
2548 * console_flush_on_panic - flush console content on panic
2549 * @mode: flush all messages in buffer or just the pending ones
2550 *
2551 * Immediately output all pending messages no matter what.
2552 */
2553void console_flush_on_panic(enum con_flush_mode mode)
2554{
2555 /*
2556 * If someone else is holding the console lock, trylock will fail
2557 * and may_schedule may be set. Ignore and proceed to unlock so
2558 * that messages are flushed out. As this can be called from any
2559 * context and we don't want to get preempted while flushing,
2560 * ensure may_schedule is cleared.
2561 */
2562 console_trylock();
2563 console_may_schedule = 0;
2564
2565 if (mode == CONSOLE_REPLAY_ALL) {
2566 unsigned long flags;
2567
2568 logbuf_lock_irqsave(flags);
2569 console_seq = log_first_seq;
2570 console_idx = log_first_idx;
2571 logbuf_unlock_irqrestore(flags);
2572 }
2573 console_unlock();
2574}
2575
2576/*
2577 * Return the console tty driver structure and its associated index
2578 */
2579struct tty_driver *console_device(int *index)
2580{
2581 struct console *c;
2582 struct tty_driver *driver = NULL;
2583
2584 console_lock();
2585 for_each_console(c) {
2586 if (!c->device)
2587 continue;
2588 driver = c->device(c, index);
2589 if (driver)
2590 break;
2591 }
2592 console_unlock();
2593 return driver;
2594}
2595
2596/*
2597 * Prevent further output on the passed console device so that (for example)
2598 * serial drivers can disable console output before suspending a port, and can
2599 * re-enable output afterwards.
2600 */
2601void console_stop(struct console *console)
2602{
2603 console_lock();
2604 console->flags &= ~CON_ENABLED;
2605 console_unlock();
2606}
2607EXPORT_SYMBOL(console_stop);
2608
2609void console_start(struct console *console)
2610{
2611 console_lock();
2612 console->flags |= CON_ENABLED;
2613 console_unlock();
2614}
2615EXPORT_SYMBOL(console_start);
2616
2617static int __read_mostly keep_bootcon;
2618
2619static int __init keep_bootcon_setup(char *str)
2620{
2621 keep_bootcon = 1;
2622 pr_info("debug: skip boot console de-registration.\n");
2623
2624 return 0;
2625}
2626
2627early_param("keep_bootcon", keep_bootcon_setup);
2628
2629/*
2630 * The console driver calls this routine during kernel initialization
2631 * to register the console printing procedure with printk() and to
2632 * print any messages that were printed by the kernel before the
2633 * console driver was initialized.
2634 *
2635 * This can happen pretty early during the boot process (because of
2636 * early_printk) - sometimes before setup_arch() completes - be careful
2637 * of what kernel features are used - they may not be initialised yet.
2638 *
2639 * There are two types of consoles - bootconsoles (early_printk) and
2640 * "real" consoles (everything which is not a bootconsole) which are
2641 * handled differently.
2642 * - Any number of bootconsoles can be registered at any time.
2643 * - As soon as a "real" console is registered, all bootconsoles
2644 * will be unregistered automatically.
2645 * - Once a "real" console is registered, any attempt to register a
2646 * bootconsoles will be rejected
2647 */
2648void register_console(struct console *newcon)
2649{
2650 int i;
2651 unsigned long flags;
2652 struct console *bcon = NULL;
2653 struct console_cmdline *c;
2654 static bool has_preferred;
2655
2656 if (console_drivers)
2657 for_each_console(bcon)
2658 if (WARN(bcon == newcon,
2659 "console '%s%d' already registered\n",
2660 bcon->name, bcon->index))
2661 return;
2662
2663 /*
2664 * before we register a new CON_BOOT console, make sure we don't
2665 * already have a valid console
2666 */
2667 if (console_drivers && newcon->flags & CON_BOOT) {
2668 /* find the last or real console */
2669 for_each_console(bcon) {
2670 if (!(bcon->flags & CON_BOOT)) {
2671 pr_info("Too late to register bootconsole %s%d\n",
2672 newcon->name, newcon->index);
2673 return;
2674 }
2675 }
2676 }
2677
2678 if (console_drivers && console_drivers->flags & CON_BOOT)
2679 bcon = console_drivers;
2680
2681 if (!has_preferred || bcon || !console_drivers)
2682 has_preferred = preferred_console >= 0;
2683
2684 /*
2685 * See if we want to use this console driver. If we
2686 * didn't select a console we take the first one
2687 * that registers here.
2688 */
2689 if (!has_preferred) {
2690 if (newcon->index < 0)
2691 newcon->index = 0;
2692 if (newcon->setup == NULL ||
2693 newcon->setup(newcon, NULL) == 0) {
2694 newcon->flags |= CON_ENABLED;
2695 if (newcon->device) {
2696 newcon->flags |= CON_CONSDEV;
2697 has_preferred = true;
2698 }
2699 }
2700 }
2701
2702 /*
2703 * See if this console matches one we selected on
2704 * the command line.
2705 */
2706 for (i = 0, c = console_cmdline;
2707 i < MAX_CMDLINECONSOLES && c->name[0];
2708 i++, c++) {
2709 if (!newcon->match ||
2710 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2711 /* default matching */
2712 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2713 if (strcmp(c->name, newcon->name) != 0)
2714 continue;
2715 if (newcon->index >= 0 &&
2716 newcon->index != c->index)
2717 continue;
2718 if (newcon->index < 0)
2719 newcon->index = c->index;
2720
2721 if (_braille_register_console(newcon, c))
2722 return;
2723
2724 if (newcon->setup &&
2725 newcon->setup(newcon, c->options) != 0)
2726 break;
2727 }
2728
2729 newcon->flags |= CON_ENABLED;
2730 if (i == preferred_console) {
2731 newcon->flags |= CON_CONSDEV;
2732 has_preferred = true;
2733 }
2734 break;
2735 }
2736
2737 if (!(newcon->flags & CON_ENABLED))
2738 return;
2739
2740 /*
2741 * If we have a bootconsole, and are switching to a real console,
2742 * don't print everything out again, since when the boot console, and
2743 * the real console are the same physical device, it's annoying to
2744 * see the beginning boot messages twice
2745 */
2746 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2747 newcon->flags &= ~CON_PRINTBUFFER;
2748
2749 /*
2750 * Put this console in the list - keep the
2751 * preferred driver at the head of the list.
2752 */
2753 console_lock();
2754 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2755 newcon->next = console_drivers;
2756 console_drivers = newcon;
2757 if (newcon->next)
2758 newcon->next->flags &= ~CON_CONSDEV;
2759 } else {
2760 newcon->next = console_drivers->next;
2761 console_drivers->next = newcon;
2762 }
2763
2764 if (newcon->flags & CON_EXTENDED)
2765 nr_ext_console_drivers++;
2766
2767 if (newcon->flags & CON_PRINTBUFFER) {
2768 /*
2769 * console_unlock(); will print out the buffered messages
2770 * for us.
2771 */
2772 logbuf_lock_irqsave(flags);
2773 console_seq = syslog_seq;
2774 console_idx = syslog_idx;
2775 /*
2776 * We're about to replay the log buffer. Only do this to the
2777 * just-registered console to avoid excessive message spam to
2778 * the already-registered consoles.
2779 *
2780 * Set exclusive_console with disabled interrupts to reduce
2781 * race window with eventual console_flush_on_panic() that
2782 * ignores console_lock.
2783 */
2784 exclusive_console = newcon;
2785 exclusive_console_stop_seq = console_seq;
2786 logbuf_unlock_irqrestore(flags);
2787 }
2788 console_unlock();
2789 console_sysfs_notify();
2790
2791 /*
2792 * By unregistering the bootconsoles after we enable the real console
2793 * we get the "console xxx enabled" message on all the consoles -
2794 * boot consoles, real consoles, etc - this is to ensure that end
2795 * users know there might be something in the kernel's log buffer that
2796 * went to the bootconsole (that they do not see on the real console)
2797 */
2798 pr_info("%sconsole [%s%d] enabled\n",
2799 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2800 newcon->name, newcon->index);
2801 if (bcon &&
2802 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2803 !keep_bootcon) {
2804 /* We need to iterate through all boot consoles, to make
2805 * sure we print everything out, before we unregister them.
2806 */
2807 for_each_console(bcon)
2808 if (bcon->flags & CON_BOOT)
2809 unregister_console(bcon);
2810 }
2811}
2812EXPORT_SYMBOL(register_console);
2813
2814int unregister_console(struct console *console)
2815{
2816 struct console *a, *b;
2817 int res;
2818
2819 pr_info("%sconsole [%s%d] disabled\n",
2820 (console->flags & CON_BOOT) ? "boot" : "" ,
2821 console->name, console->index);
2822
2823 res = _braille_unregister_console(console);
2824 if (res)
2825 return res;
2826
2827 res = 1;
2828 console_lock();
2829 if (console_drivers == console) {
2830 console_drivers=console->next;
2831 res = 0;
2832 } else if (console_drivers) {
2833 for (a=console_drivers->next, b=console_drivers ;
2834 a; b=a, a=b->next) {
2835 if (a == console) {
2836 b->next = a->next;
2837 res = 0;
2838 break;
2839 }
2840 }
2841 }
2842
2843 if (!res && (console->flags & CON_EXTENDED))
2844 nr_ext_console_drivers--;
2845
2846 /*
2847 * If this isn't the last console and it has CON_CONSDEV set, we
2848 * need to set it on the next preferred console.
2849 */
2850 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2851 console_drivers->flags |= CON_CONSDEV;
2852
2853 console->flags &= ~CON_ENABLED;
2854 console_unlock();
2855 console_sysfs_notify();
2856 return res;
2857}
2858EXPORT_SYMBOL(unregister_console);
2859
2860/*
2861 * Initialize the console device. This is called *early*, so
2862 * we can't necessarily depend on lots of kernel help here.
2863 * Just do some early initializations, and do the complex setup
2864 * later.
2865 */
2866void __init console_init(void)
2867{
2868 int ret;
2869 initcall_t call;
2870 initcall_entry_t *ce;
2871
2872 /* Setup the default TTY line discipline. */
2873 n_tty_init();
2874
2875 /*
2876 * set up the console device so that later boot sequences can
2877 * inform about problems etc..
2878 */
2879 ce = __con_initcall_start;
2880 trace_initcall_level("console");
2881 while (ce < __con_initcall_end) {
2882 call = initcall_from_entry(ce);
2883 trace_initcall_start(call);
2884 ret = call();
2885 trace_initcall_finish(call, ret);
2886 ce++;
2887 }
2888}
2889
2890/*
2891 * Some boot consoles access data that is in the init section and which will
2892 * be discarded after the initcalls have been run. To make sure that no code
2893 * will access this data, unregister the boot consoles in a late initcall.
2894 *
2895 * If for some reason, such as deferred probe or the driver being a loadable
2896 * module, the real console hasn't registered yet at this point, there will
2897 * be a brief interval in which no messages are logged to the console, which
2898 * makes it difficult to diagnose problems that occur during this time.
2899 *
2900 * To mitigate this problem somewhat, only unregister consoles whose memory
2901 * intersects with the init section. Note that all other boot consoles will
2902 * get unregistred when the real preferred console is registered.
2903 */
2904static int __init printk_late_init(void)
2905{
2906 struct console *con;
2907 int ret;
2908
2909 for_each_console(con) {
2910 if (!(con->flags & CON_BOOT))
2911 continue;
2912
2913 /* Check addresses that might be used for enabled consoles. */
2914 if (init_section_intersects(con, sizeof(*con)) ||
2915 init_section_contains(con->write, 0) ||
2916 init_section_contains(con->read, 0) ||
2917 init_section_contains(con->device, 0) ||
2918 init_section_contains(con->unblank, 0) ||
2919 init_section_contains(con->data, 0)) {
2920 /*
2921 * Please, consider moving the reported consoles out
2922 * of the init section.
2923 */
2924 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2925 con->name, con->index);
2926 unregister_console(con);
2927 }
2928 }
2929 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2930 console_cpu_notify);
2931 WARN_ON(ret < 0);
2932 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2933 console_cpu_notify, NULL);
2934 WARN_ON(ret < 0);
2935 return 0;
2936}
2937late_initcall(printk_late_init);
2938
2939#if defined CONFIG_PRINTK
2940/*
2941 * Delayed printk version, for scheduler-internal messages:
2942 */
2943#define PRINTK_PENDING_WAKEUP 0x01
2944#define PRINTK_PENDING_OUTPUT 0x02
2945
2946static DEFINE_PER_CPU(int, printk_pending);
2947
2948static void wake_up_klogd_work_func(struct irq_work *irq_work)
2949{
2950 int pending = __this_cpu_xchg(printk_pending, 0);
2951
2952 if (pending & PRINTK_PENDING_OUTPUT) {
2953 /* If trylock fails, someone else is doing the printing */
2954 if (console_trylock())
2955 console_unlock();
2956 }
2957
2958 if (pending & PRINTK_PENDING_WAKEUP)
2959 wake_up_interruptible(&log_wait);
2960}
2961
2962static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2963 .func = wake_up_klogd_work_func,
2964 .flags = IRQ_WORK_LAZY,
2965};
2966
2967void wake_up_klogd(void)
2968{
2969 preempt_disable();
2970 if (waitqueue_active(&log_wait)) {
2971 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2972 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2973 }
2974 preempt_enable();
2975}
2976
2977void defer_console_output(void)
2978{
2979 preempt_disable();
2980 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2981 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2982 preempt_enable();
2983}
2984
2985int vprintk_deferred(const char *fmt, va_list args)
2986{
2987 int r;
2988
2989 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2990 defer_console_output();
2991
2992 return r;
2993}
2994
2995int printk_deferred(const char *fmt, ...)
2996{
2997 va_list args;
2998 int r;
2999
3000 va_start(args, fmt);
3001 r = vprintk_deferred(fmt, args);
3002 va_end(args);
3003
3004 return r;
3005}
3006
3007/*
3008 * printk rate limiting, lifted from the networking subsystem.
3009 *
3010 * This enforces a rate limit: not more than 10 kernel messages
3011 * every 5s to make a denial-of-service attack impossible.
3012 */
3013DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3014
3015int __printk_ratelimit(const char *func)
3016{
3017 return ___ratelimit(&printk_ratelimit_state, func);
3018}
3019EXPORT_SYMBOL(__printk_ratelimit);
3020
3021/**
3022 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3023 * @caller_jiffies: pointer to caller's state
3024 * @interval_msecs: minimum interval between prints
3025 *
3026 * printk_timed_ratelimit() returns true if more than @interval_msecs
3027 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3028 * returned true.
3029 */
3030bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3031 unsigned int interval_msecs)
3032{
3033 unsigned long elapsed = jiffies - *caller_jiffies;
3034
3035 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3036 return false;
3037
3038 *caller_jiffies = jiffies;
3039 return true;
3040}
3041EXPORT_SYMBOL(printk_timed_ratelimit);
3042
3043static DEFINE_SPINLOCK(dump_list_lock);
3044static LIST_HEAD(dump_list);
3045
3046/**
3047 * kmsg_dump_register - register a kernel log dumper.
3048 * @dumper: pointer to the kmsg_dumper structure
3049 *
3050 * Adds a kernel log dumper to the system. The dump callback in the
3051 * structure will be called when the kernel oopses or panics and must be
3052 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3053 */
3054int kmsg_dump_register(struct kmsg_dumper *dumper)
3055{
3056 unsigned long flags;
3057 int err = -EBUSY;
3058
3059 /* The dump callback needs to be set */
3060 if (!dumper->dump)
3061 return -EINVAL;
3062
3063 spin_lock_irqsave(&dump_list_lock, flags);
3064 /* Don't allow registering multiple times */
3065 if (!dumper->registered) {
3066 dumper->registered = 1;
3067 list_add_tail_rcu(&dumper->list, &dump_list);
3068 err = 0;
3069 }
3070 spin_unlock_irqrestore(&dump_list_lock, flags);
3071
3072 return err;
3073}
3074EXPORT_SYMBOL_GPL(kmsg_dump_register);
3075
3076/**
3077 * kmsg_dump_unregister - unregister a kmsg dumper.
3078 * @dumper: pointer to the kmsg_dumper structure
3079 *
3080 * Removes a dump device from the system. Returns zero on success and
3081 * %-EINVAL otherwise.
3082 */
3083int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3084{
3085 unsigned long flags;
3086 int err = -EINVAL;
3087
3088 spin_lock_irqsave(&dump_list_lock, flags);
3089 if (dumper->registered) {
3090 dumper->registered = 0;
3091 list_del_rcu(&dumper->list);
3092 err = 0;
3093 }
3094 spin_unlock_irqrestore(&dump_list_lock, flags);
3095 synchronize_rcu();
3096
3097 return err;
3098}
3099EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3100
3101static bool always_kmsg_dump;
3102module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3103
3104/**
3105 * kmsg_dump - dump kernel log to kernel message dumpers.
3106 * @reason: the reason (oops, panic etc) for dumping
3107 *
3108 * Call each of the registered dumper's dump() callback, which can
3109 * retrieve the kmsg records with kmsg_dump_get_line() or
3110 * kmsg_dump_get_buffer().
3111 */
3112void kmsg_dump(enum kmsg_dump_reason reason)
3113{
3114 struct kmsg_dumper *dumper;
3115 unsigned long flags;
3116
3117 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3118 return;
3119
3120 rcu_read_lock();
3121 list_for_each_entry_rcu(dumper, &dump_list, list) {
3122 if (dumper->max_reason && reason > dumper->max_reason)
3123 continue;
3124
3125 /* initialize iterator with data about the stored records */
3126 dumper->active = true;
3127
3128 logbuf_lock_irqsave(flags);
3129 dumper->cur_seq = clear_seq;
3130 dumper->cur_idx = clear_idx;
3131 dumper->next_seq = log_next_seq;
3132 dumper->next_idx = log_next_idx;
3133 logbuf_unlock_irqrestore(flags);
3134
3135 /* invoke dumper which will iterate over records */
3136 dumper->dump(dumper, reason);
3137
3138 /* reset iterator */
3139 dumper->active = false;
3140 }
3141 rcu_read_unlock();
3142}
3143
3144/**
3145 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3146 * @dumper: registered kmsg dumper
3147 * @syslog: include the "<4>" prefixes
3148 * @line: buffer to copy the line to
3149 * @size: maximum size of the buffer
3150 * @len: length of line placed into buffer
3151 *
3152 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3153 * record, and copy one record into the provided buffer.
3154 *
3155 * Consecutive calls will return the next available record moving
3156 * towards the end of the buffer with the youngest messages.
3157 *
3158 * A return value of FALSE indicates that there are no more records to
3159 * read.
3160 *
3161 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3162 */
3163bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3164 char *line, size_t size, size_t *len)
3165{
3166 struct printk_log *msg;
3167 size_t l = 0;
3168 bool ret = false;
3169
3170 if (!dumper->active)
3171 goto out;
3172
3173 if (dumper->cur_seq < log_first_seq) {
3174 /* messages are gone, move to first available one */
3175 dumper->cur_seq = log_first_seq;
3176 dumper->cur_idx = log_first_idx;
3177 }
3178
3179 /* last entry */
3180 if (dumper->cur_seq >= log_next_seq)
3181 goto out;
3182
3183 msg = log_from_idx(dumper->cur_idx);
3184 l = msg_print_text(msg, syslog, printk_time, line, size);
3185
3186 dumper->cur_idx = log_next(dumper->cur_idx);
3187 dumper->cur_seq++;
3188 ret = true;
3189out:
3190 if (len)
3191 *len = l;
3192 return ret;
3193}
3194
3195/**
3196 * kmsg_dump_get_line - retrieve one kmsg log line
3197 * @dumper: registered kmsg dumper
3198 * @syslog: include the "<4>" prefixes
3199 * @line: buffer to copy the line to
3200 * @size: maximum size of the buffer
3201 * @len: length of line placed into buffer
3202 *
3203 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3204 * record, and copy one record into the provided buffer.
3205 *
3206 * Consecutive calls will return the next available record moving
3207 * towards the end of the buffer with the youngest messages.
3208 *
3209 * A return value of FALSE indicates that there are no more records to
3210 * read.
3211 */
3212bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3213 char *line, size_t size, size_t *len)
3214{
3215 unsigned long flags;
3216 bool ret;
3217
3218 logbuf_lock_irqsave(flags);
3219 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3220 logbuf_unlock_irqrestore(flags);
3221
3222 return ret;
3223}
3224EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3225
3226/**
3227 * kmsg_dump_get_buffer - copy kmsg log lines
3228 * @dumper: registered kmsg dumper
3229 * @syslog: include the "<4>" prefixes
3230 * @buf: buffer to copy the line to
3231 * @size: maximum size of the buffer
3232 * @len: length of line placed into buffer
3233 *
3234 * Start at the end of the kmsg buffer and fill the provided buffer
3235 * with as many of the the *youngest* kmsg records that fit into it.
3236 * If the buffer is large enough, all available kmsg records will be
3237 * copied with a single call.
3238 *
3239 * Consecutive calls will fill the buffer with the next block of
3240 * available older records, not including the earlier retrieved ones.
3241 *
3242 * A return value of FALSE indicates that there are no more records to
3243 * read.
3244 */
3245bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3246 char *buf, size_t size, size_t *len)
3247{
3248 unsigned long flags;
3249 u64 seq;
3250 u32 idx;
3251 u64 next_seq;
3252 u32 next_idx;
3253 size_t l = 0;
3254 bool ret = false;
3255 bool time = printk_time;
3256
3257 if (!dumper->active)
3258 goto out;
3259
3260 logbuf_lock_irqsave(flags);
3261 if (dumper->cur_seq < log_first_seq) {
3262 /* messages are gone, move to first available one */
3263 dumper->cur_seq = log_first_seq;
3264 dumper->cur_idx = log_first_idx;
3265 }
3266
3267 /* last entry */
3268 if (dumper->cur_seq >= dumper->next_seq) {
3269 logbuf_unlock_irqrestore(flags);
3270 goto out;
3271 }
3272
3273 /* calculate length of entire buffer */
3274 seq = dumper->cur_seq;
3275 idx = dumper->cur_idx;
3276 while (seq < dumper->next_seq) {
3277 struct printk_log *msg = log_from_idx(idx);
3278
3279 l += msg_print_text(msg, true, time, NULL, 0);
3280 idx = log_next(idx);
3281 seq++;
3282 }
3283
3284 /* move first record forward until length fits into the buffer */
3285 seq = dumper->cur_seq;
3286 idx = dumper->cur_idx;
3287 while (l >= size && seq < dumper->next_seq) {
3288 struct printk_log *msg = log_from_idx(idx);
3289
3290 l -= msg_print_text(msg, true, time, NULL, 0);
3291 idx = log_next(idx);
3292 seq++;
3293 }
3294
3295 /* last message in next interation */
3296 next_seq = seq;
3297 next_idx = idx;
3298
3299 l = 0;
3300 while (seq < dumper->next_seq) {
3301 struct printk_log *msg = log_from_idx(idx);
3302
3303 l += msg_print_text(msg, syslog, time, buf + l, size - l);
3304 idx = log_next(idx);
3305 seq++;
3306 }
3307
3308 dumper->next_seq = next_seq;
3309 dumper->next_idx = next_idx;
3310 ret = true;
3311 logbuf_unlock_irqrestore(flags);
3312out:
3313 if (len)
3314 *len = l;
3315 return ret;
3316}
3317EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3318
3319/**
3320 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3321 * @dumper: registered kmsg dumper
3322 *
3323 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3324 * kmsg_dump_get_buffer() can be called again and used multiple
3325 * times within the same dumper.dump() callback.
3326 *
3327 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3328 */
3329void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3330{
3331 dumper->cur_seq = clear_seq;
3332 dumper->cur_idx = clear_idx;
3333 dumper->next_seq = log_next_seq;
3334 dumper->next_idx = log_next_idx;
3335}
3336
3337/**
3338 * kmsg_dump_rewind - reset the interator
3339 * @dumper: registered kmsg dumper
3340 *
3341 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3342 * kmsg_dump_get_buffer() can be called again and used multiple
3343 * times within the same dumper.dump() callback.
3344 */
3345void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3346{
3347 unsigned long flags;
3348
3349 logbuf_lock_irqsave(flags);
3350 kmsg_dump_rewind_nolock(dumper);
3351 logbuf_unlock_irqrestore(flags);
3352}
3353EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3354
3355#endif