Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/printk.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * Modified to make sys_syslog() more flexible: added commands to
8 * return the last 4k of kernel messages, regardless of whether
9 * they've been read or not. Added option to suppress kernel printk's
10 * to the console. Added hook for sending the console messages
11 * elsewhere, in preparation for a serial line console (someday).
12 * Ted Ts'o, 2/11/93.
13 * Modified for sysctl support, 1/8/97, Chris Horn.
14 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
15 * manfred@colorfullife.com
16 * Rewrote bits to get rid of console_lock
17 * 01Mar01 Andrew Morton
18 */
19
20#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21
22#include <linux/kernel.h>
23#include <linux/mm.h>
24#include <linux/tty.h>
25#include <linux/tty_driver.h>
26#include <linux/console.h>
27#include <linux/init.h>
28#include <linux/jiffies.h>
29#include <linux/nmi.h>
30#include <linux/module.h>
31#include <linux/moduleparam.h>
32#include <linux/delay.h>
33#include <linux/smp.h>
34#include <linux/security.h>
35#include <linux/memblock.h>
36#include <linux/syscalls.h>
37#include <linux/crash_core.h>
38#include <linux/kdb.h>
39#include <linux/ratelimit.h>
40#include <linux/kmsg_dump.h>
41#include <linux/syslog.h>
42#include <linux/cpu.h>
43#include <linux/rculist.h>
44#include <linux/poll.h>
45#include <linux/irq_work.h>
46#include <linux/ctype.h>
47#include <linux/uio.h>
48#include <linux/sched/clock.h>
49#include <linux/sched/debug.h>
50#include <linux/sched/task_stack.h>
51
52#include <linux/uaccess.h>
53#include <asm/sections.h>
54
55#include <trace/events/initcall.h>
56#define CREATE_TRACE_POINTS
57#include <trace/events/printk.h>
58
59#include "console_cmdline.h"
60#include "braille.h"
61#include "internal.h"
62
63int console_printk[4] = {
64 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
65 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
66 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
67 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
68};
69EXPORT_SYMBOL_GPL(console_printk);
70
71atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
72EXPORT_SYMBOL(ignore_console_lock_warning);
73
74/*
75 * Low level drivers may need that to know if they can schedule in
76 * their unblank() callback or not. So let's export it.
77 */
78int oops_in_progress;
79EXPORT_SYMBOL(oops_in_progress);
80
81/*
82 * console_sem protects the console_drivers list, and also
83 * provides serialisation for access to the entire console
84 * driver system.
85 */
86static DEFINE_SEMAPHORE(console_sem);
87struct console *console_drivers;
88EXPORT_SYMBOL_GPL(console_drivers);
89
90/*
91 * System may need to suppress printk message under certain
92 * circumstances, like after kernel panic happens.
93 */
94int __read_mostly suppress_printk;
95
96#ifdef CONFIG_LOCKDEP
97static struct lockdep_map console_lock_dep_map = {
98 .name = "console_lock"
99};
100#endif
101
102enum devkmsg_log_bits {
103 __DEVKMSG_LOG_BIT_ON = 0,
104 __DEVKMSG_LOG_BIT_OFF,
105 __DEVKMSG_LOG_BIT_LOCK,
106};
107
108enum devkmsg_log_masks {
109 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
110 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
111 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
112};
113
114/* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
115#define DEVKMSG_LOG_MASK_DEFAULT 0
116
117static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
118
119static int __control_devkmsg(char *str)
120{
121 size_t len;
122
123 if (!str)
124 return -EINVAL;
125
126 len = str_has_prefix(str, "on");
127 if (len) {
128 devkmsg_log = DEVKMSG_LOG_MASK_ON;
129 return len;
130 }
131
132 len = str_has_prefix(str, "off");
133 if (len) {
134 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
135 return len;
136 }
137
138 len = str_has_prefix(str, "ratelimit");
139 if (len) {
140 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
141 return len;
142 }
143
144 return -EINVAL;
145}
146
147static int __init control_devkmsg(char *str)
148{
149 if (__control_devkmsg(str) < 0)
150 return 1;
151
152 /*
153 * Set sysctl string accordingly:
154 */
155 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
156 strcpy(devkmsg_log_str, "on");
157 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
158 strcpy(devkmsg_log_str, "off");
159 /* else "ratelimit" which is set by default. */
160
161 /*
162 * Sysctl cannot change it anymore. The kernel command line setting of
163 * this parameter is to force the setting to be permanent throughout the
164 * runtime of the system. This is a precation measure against userspace
165 * trying to be a smarta** and attempting to change it up on us.
166 */
167 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
168
169 return 0;
170}
171__setup("printk.devkmsg=", control_devkmsg);
172
173char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
174
175int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
176 void __user *buffer, size_t *lenp, loff_t *ppos)
177{
178 char old_str[DEVKMSG_STR_MAX_SIZE];
179 unsigned int old;
180 int err;
181
182 if (write) {
183 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
184 return -EINVAL;
185
186 old = devkmsg_log;
187 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
188 }
189
190 err = proc_dostring(table, write, buffer, lenp, ppos);
191 if (err)
192 return err;
193
194 if (write) {
195 err = __control_devkmsg(devkmsg_log_str);
196
197 /*
198 * Do not accept an unknown string OR a known string with
199 * trailing crap...
200 */
201 if (err < 0 || (err + 1 != *lenp)) {
202
203 /* ... and restore old setting. */
204 devkmsg_log = old;
205 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
206
207 return -EINVAL;
208 }
209 }
210
211 return 0;
212}
213
214/* Number of registered extended console drivers. */
215static int nr_ext_console_drivers;
216
217/*
218 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
219 * macros instead of functions so that _RET_IP_ contains useful information.
220 */
221#define down_console_sem() do { \
222 down(&console_sem);\
223 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
224} while (0)
225
226static int __down_trylock_console_sem(unsigned long ip)
227{
228 int lock_failed;
229 unsigned long flags;
230
231 /*
232 * Here and in __up_console_sem() we need to be in safe mode,
233 * because spindump/WARN/etc from under console ->lock will
234 * deadlock in printk()->down_trylock_console_sem() otherwise.
235 */
236 printk_safe_enter_irqsave(flags);
237 lock_failed = down_trylock(&console_sem);
238 printk_safe_exit_irqrestore(flags);
239
240 if (lock_failed)
241 return 1;
242 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
243 return 0;
244}
245#define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
246
247static void __up_console_sem(unsigned long ip)
248{
249 unsigned long flags;
250
251 mutex_release(&console_lock_dep_map, 1, ip);
252
253 printk_safe_enter_irqsave(flags);
254 up(&console_sem);
255 printk_safe_exit_irqrestore(flags);
256}
257#define up_console_sem() __up_console_sem(_RET_IP_)
258
259/*
260 * This is used for debugging the mess that is the VT code by
261 * keeping track if we have the console semaphore held. It's
262 * definitely not the perfect debug tool (we don't know if _WE_
263 * hold it and are racing, but it helps tracking those weird code
264 * paths in the console code where we end up in places I want
265 * locked without the console sempahore held).
266 */
267static int console_locked, console_suspended;
268
269/*
270 * If exclusive_console is non-NULL then only this console is to be printed to.
271 */
272static struct console *exclusive_console;
273
274/*
275 * Array of consoles built from command line options (console=)
276 */
277
278#define MAX_CMDLINECONSOLES 8
279
280static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
281
282static int preferred_console = -1;
283int console_set_on_cmdline;
284EXPORT_SYMBOL(console_set_on_cmdline);
285
286/* Flag: console code may call schedule() */
287static int console_may_schedule;
288
289enum con_msg_format_flags {
290 MSG_FORMAT_DEFAULT = 0,
291 MSG_FORMAT_SYSLOG = (1 << 0),
292};
293
294static int console_msg_format = MSG_FORMAT_DEFAULT;
295
296/*
297 * The printk log buffer consists of a chain of concatenated variable
298 * length records. Every record starts with a record header, containing
299 * the overall length of the record.
300 *
301 * The heads to the first and last entry in the buffer, as well as the
302 * sequence numbers of these entries are maintained when messages are
303 * stored.
304 *
305 * If the heads indicate available messages, the length in the header
306 * tells the start next message. A length == 0 for the next message
307 * indicates a wrap-around to the beginning of the buffer.
308 *
309 * Every record carries the monotonic timestamp in microseconds, as well as
310 * the standard userspace syslog level and syslog facility. The usual
311 * kernel messages use LOG_KERN; userspace-injected messages always carry
312 * a matching syslog facility, by default LOG_USER. The origin of every
313 * message can be reliably determined that way.
314 *
315 * The human readable log message directly follows the message header. The
316 * length of the message text is stored in the header, the stored message
317 * is not terminated.
318 *
319 * Optionally, a message can carry a dictionary of properties (key/value pairs),
320 * to provide userspace with a machine-readable message context.
321 *
322 * Examples for well-defined, commonly used property names are:
323 * DEVICE=b12:8 device identifier
324 * b12:8 block dev_t
325 * c127:3 char dev_t
326 * n8 netdev ifindex
327 * +sound:card0 subsystem:devname
328 * SUBSYSTEM=pci driver-core subsystem name
329 *
330 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
331 * follows directly after a '=' character. Every property is terminated by
332 * a '\0' character. The last property is not terminated.
333 *
334 * Example of a message structure:
335 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
336 * 0008 34 00 record is 52 bytes long
337 * 000a 0b 00 text is 11 bytes long
338 * 000c 1f 00 dictionary is 23 bytes long
339 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
340 * 0010 69 74 27 73 20 61 20 6c "it's a l"
341 * 69 6e 65 "ine"
342 * 001b 44 45 56 49 43 "DEVIC"
343 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
344 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
345 * 67 "g"
346 * 0032 00 00 00 padding to next message header
347 *
348 * The 'struct printk_log' buffer header must never be directly exported to
349 * userspace, it is a kernel-private implementation detail that might
350 * need to be changed in the future, when the requirements change.
351 *
352 * /dev/kmsg exports the structured data in the following line format:
353 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
354 *
355 * Users of the export format should ignore possible additional values
356 * separated by ',', and find the message after the ';' character.
357 *
358 * The optional key/value pairs are attached as continuation lines starting
359 * with a space character and terminated by a newline. All possible
360 * non-prinatable characters are escaped in the "\xff" notation.
361 */
362
363enum log_flags {
364 LOG_NEWLINE = 2, /* text ended with a newline */
365 LOG_CONT = 8, /* text is a fragment of a continuation line */
366};
367
368struct printk_log {
369 u64 ts_nsec; /* timestamp in nanoseconds */
370 u16 len; /* length of entire record */
371 u16 text_len; /* length of text buffer */
372 u16 dict_len; /* length of dictionary buffer */
373 u8 facility; /* syslog facility */
374 u8 flags:5; /* internal record flags */
375 u8 level:3; /* syslog level */
376#ifdef CONFIG_PRINTK_CALLER
377 u32 caller_id; /* thread id or processor id */
378#endif
379}
380#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
381__packed __aligned(4)
382#endif
383;
384
385/*
386 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
387 * within the scheduler's rq lock. It must be released before calling
388 * console_unlock() or anything else that might wake up a process.
389 */
390DEFINE_RAW_SPINLOCK(logbuf_lock);
391
392/*
393 * Helper macros to lock/unlock logbuf_lock and switch between
394 * printk-safe/unsafe modes.
395 */
396#define logbuf_lock_irq() \
397 do { \
398 printk_safe_enter_irq(); \
399 raw_spin_lock(&logbuf_lock); \
400 } while (0)
401
402#define logbuf_unlock_irq() \
403 do { \
404 raw_spin_unlock(&logbuf_lock); \
405 printk_safe_exit_irq(); \
406 } while (0)
407
408#define logbuf_lock_irqsave(flags) \
409 do { \
410 printk_safe_enter_irqsave(flags); \
411 raw_spin_lock(&logbuf_lock); \
412 } while (0)
413
414#define logbuf_unlock_irqrestore(flags) \
415 do { \
416 raw_spin_unlock(&logbuf_lock); \
417 printk_safe_exit_irqrestore(flags); \
418 } while (0)
419
420#ifdef CONFIG_PRINTK
421DECLARE_WAIT_QUEUE_HEAD(log_wait);
422/* the next printk record to read by syslog(READ) or /proc/kmsg */
423static u64 syslog_seq;
424static u32 syslog_idx;
425static size_t syslog_partial;
426static bool syslog_time;
427
428/* index and sequence number of the first record stored in the buffer */
429static u64 log_first_seq;
430static u32 log_first_idx;
431
432/* index and sequence number of the next record to store in the buffer */
433static u64 log_next_seq;
434static u32 log_next_idx;
435
436/* the next printk record to write to the console */
437static u64 console_seq;
438static u32 console_idx;
439static u64 exclusive_console_stop_seq;
440
441/* the next printk record to read after the last 'clear' command */
442static u64 clear_seq;
443static u32 clear_idx;
444
445#ifdef CONFIG_PRINTK_CALLER
446#define PREFIX_MAX 48
447#else
448#define PREFIX_MAX 32
449#endif
450#define LOG_LINE_MAX (1024 - PREFIX_MAX)
451
452#define LOG_LEVEL(v) ((v) & 0x07)
453#define LOG_FACILITY(v) ((v) >> 3 & 0xff)
454
455/* record buffer */
456#define LOG_ALIGN __alignof__(struct printk_log)
457#define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
458#define LOG_BUF_LEN_MAX (u32)(1 << 31)
459static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
460static char *log_buf = __log_buf;
461static u32 log_buf_len = __LOG_BUF_LEN;
462
463/* Return log buffer address */
464char *log_buf_addr_get(void)
465{
466 return log_buf;
467}
468
469/* Return log buffer size */
470u32 log_buf_len_get(void)
471{
472 return log_buf_len;
473}
474
475/* human readable text of the record */
476static char *log_text(const struct printk_log *msg)
477{
478 return (char *)msg + sizeof(struct printk_log);
479}
480
481/* optional key/value pair dictionary attached to the record */
482static char *log_dict(const struct printk_log *msg)
483{
484 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
485}
486
487/* get record by index; idx must point to valid msg */
488static struct printk_log *log_from_idx(u32 idx)
489{
490 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
491
492 /*
493 * A length == 0 record is the end of buffer marker. Wrap around and
494 * read the message at the start of the buffer.
495 */
496 if (!msg->len)
497 return (struct printk_log *)log_buf;
498 return msg;
499}
500
501/* get next record; idx must point to valid msg */
502static u32 log_next(u32 idx)
503{
504 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
505
506 /* length == 0 indicates the end of the buffer; wrap */
507 /*
508 * A length == 0 record is the end of buffer marker. Wrap around and
509 * read the message at the start of the buffer as *this* one, and
510 * return the one after that.
511 */
512 if (!msg->len) {
513 msg = (struct printk_log *)log_buf;
514 return msg->len;
515 }
516 return idx + msg->len;
517}
518
519/*
520 * Check whether there is enough free space for the given message.
521 *
522 * The same values of first_idx and next_idx mean that the buffer
523 * is either empty or full.
524 *
525 * If the buffer is empty, we must respect the position of the indexes.
526 * They cannot be reset to the beginning of the buffer.
527 */
528static int logbuf_has_space(u32 msg_size, bool empty)
529{
530 u32 free;
531
532 if (log_next_idx > log_first_idx || empty)
533 free = max(log_buf_len - log_next_idx, log_first_idx);
534 else
535 free = log_first_idx - log_next_idx;
536
537 /*
538 * We need space also for an empty header that signalizes wrapping
539 * of the buffer.
540 */
541 return free >= msg_size + sizeof(struct printk_log);
542}
543
544static int log_make_free_space(u32 msg_size)
545{
546 while (log_first_seq < log_next_seq &&
547 !logbuf_has_space(msg_size, false)) {
548 /* drop old messages until we have enough contiguous space */
549 log_first_idx = log_next(log_first_idx);
550 log_first_seq++;
551 }
552
553 if (clear_seq < log_first_seq) {
554 clear_seq = log_first_seq;
555 clear_idx = log_first_idx;
556 }
557
558 /* sequence numbers are equal, so the log buffer is empty */
559 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
560 return 0;
561
562 return -ENOMEM;
563}
564
565/* compute the message size including the padding bytes */
566static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
567{
568 u32 size;
569
570 size = sizeof(struct printk_log) + text_len + dict_len;
571 *pad_len = (-size) & (LOG_ALIGN - 1);
572 size += *pad_len;
573
574 return size;
575}
576
577/*
578 * Define how much of the log buffer we could take at maximum. The value
579 * must be greater than two. Note that only half of the buffer is available
580 * when the index points to the middle.
581 */
582#define MAX_LOG_TAKE_PART 4
583static const char trunc_msg[] = "<truncated>";
584
585static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
586 u16 *dict_len, u32 *pad_len)
587{
588 /*
589 * The message should not take the whole buffer. Otherwise, it might
590 * get removed too soon.
591 */
592 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
593 if (*text_len > max_text_len)
594 *text_len = max_text_len;
595 /* enable the warning message */
596 *trunc_msg_len = strlen(trunc_msg);
597 /* disable the "dict" completely */
598 *dict_len = 0;
599 /* compute the size again, count also the warning message */
600 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
601}
602
603/* insert record into the buffer, discard old ones, update heads */
604static int log_store(u32 caller_id, int facility, int level,
605 enum log_flags flags, u64 ts_nsec,
606 const char *dict, u16 dict_len,
607 const char *text, u16 text_len)
608{
609 struct printk_log *msg;
610 u32 size, pad_len;
611 u16 trunc_msg_len = 0;
612
613 /* number of '\0' padding bytes to next message */
614 size = msg_used_size(text_len, dict_len, &pad_len);
615
616 if (log_make_free_space(size)) {
617 /* truncate the message if it is too long for empty buffer */
618 size = truncate_msg(&text_len, &trunc_msg_len,
619 &dict_len, &pad_len);
620 /* survive when the log buffer is too small for trunc_msg */
621 if (log_make_free_space(size))
622 return 0;
623 }
624
625 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
626 /*
627 * This message + an additional empty header does not fit
628 * at the end of the buffer. Add an empty header with len == 0
629 * to signify a wrap around.
630 */
631 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
632 log_next_idx = 0;
633 }
634
635 /* fill message */
636 msg = (struct printk_log *)(log_buf + log_next_idx);
637 memcpy(log_text(msg), text, text_len);
638 msg->text_len = text_len;
639 if (trunc_msg_len) {
640 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
641 msg->text_len += trunc_msg_len;
642 }
643 memcpy(log_dict(msg), dict, dict_len);
644 msg->dict_len = dict_len;
645 msg->facility = facility;
646 msg->level = level & 7;
647 msg->flags = flags & 0x1f;
648 if (ts_nsec > 0)
649 msg->ts_nsec = ts_nsec;
650 else
651 msg->ts_nsec = local_clock();
652#ifdef CONFIG_PRINTK_CALLER
653 msg->caller_id = caller_id;
654#endif
655 memset(log_dict(msg) + dict_len, 0, pad_len);
656 msg->len = size;
657
658 /* insert message */
659 log_next_idx += msg->len;
660 log_next_seq++;
661
662 return msg->text_len;
663}
664
665int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
666
667static int syslog_action_restricted(int type)
668{
669 if (dmesg_restrict)
670 return 1;
671 /*
672 * Unless restricted, we allow "read all" and "get buffer size"
673 * for everybody.
674 */
675 return type != SYSLOG_ACTION_READ_ALL &&
676 type != SYSLOG_ACTION_SIZE_BUFFER;
677}
678
679static int check_syslog_permissions(int type, int source)
680{
681 /*
682 * If this is from /proc/kmsg and we've already opened it, then we've
683 * already done the capabilities checks at open time.
684 */
685 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
686 goto ok;
687
688 if (syslog_action_restricted(type)) {
689 if (capable(CAP_SYSLOG))
690 goto ok;
691 /*
692 * For historical reasons, accept CAP_SYS_ADMIN too, with
693 * a warning.
694 */
695 if (capable(CAP_SYS_ADMIN)) {
696 pr_warn_once("%s (%d): Attempt to access syslog with "
697 "CAP_SYS_ADMIN but no CAP_SYSLOG "
698 "(deprecated).\n",
699 current->comm, task_pid_nr(current));
700 goto ok;
701 }
702 return -EPERM;
703 }
704ok:
705 return security_syslog(type);
706}
707
708static void append_char(char **pp, char *e, char c)
709{
710 if (*pp < e)
711 *(*pp)++ = c;
712}
713
714static ssize_t msg_print_ext_header(char *buf, size_t size,
715 struct printk_log *msg, u64 seq)
716{
717 u64 ts_usec = msg->ts_nsec;
718 char caller[20];
719#ifdef CONFIG_PRINTK_CALLER
720 u32 id = msg->caller_id;
721
722 snprintf(caller, sizeof(caller), ",caller=%c%u",
723 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
724#else
725 caller[0] = '\0';
726#endif
727
728 do_div(ts_usec, 1000);
729
730 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
731 (msg->facility << 3) | msg->level, seq, ts_usec,
732 msg->flags & LOG_CONT ? 'c' : '-', caller);
733}
734
735static ssize_t msg_print_ext_body(char *buf, size_t size,
736 char *dict, size_t dict_len,
737 char *text, size_t text_len)
738{
739 char *p = buf, *e = buf + size;
740 size_t i;
741
742 /* escape non-printable characters */
743 for (i = 0; i < text_len; i++) {
744 unsigned char c = text[i];
745
746 if (c < ' ' || c >= 127 || c == '\\')
747 p += scnprintf(p, e - p, "\\x%02x", c);
748 else
749 append_char(&p, e, c);
750 }
751 append_char(&p, e, '\n');
752
753 if (dict_len) {
754 bool line = true;
755
756 for (i = 0; i < dict_len; i++) {
757 unsigned char c = dict[i];
758
759 if (line) {
760 append_char(&p, e, ' ');
761 line = false;
762 }
763
764 if (c == '\0') {
765 append_char(&p, e, '\n');
766 line = true;
767 continue;
768 }
769
770 if (c < ' ' || c >= 127 || c == '\\') {
771 p += scnprintf(p, e - p, "\\x%02x", c);
772 continue;
773 }
774
775 append_char(&p, e, c);
776 }
777 append_char(&p, e, '\n');
778 }
779
780 return p - buf;
781}
782
783/* /dev/kmsg - userspace message inject/listen interface */
784struct devkmsg_user {
785 u64 seq;
786 u32 idx;
787 struct ratelimit_state rs;
788 struct mutex lock;
789 char buf[CONSOLE_EXT_LOG_MAX];
790};
791
792static __printf(3, 4) __cold
793int devkmsg_emit(int facility, int level, const char *fmt, ...)
794{
795 va_list args;
796 int r;
797
798 va_start(args, fmt);
799 r = vprintk_emit(facility, level, NULL, 0, fmt, args);
800 va_end(args);
801
802 return r;
803}
804
805static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
806{
807 char *buf, *line;
808 int level = default_message_loglevel;
809 int facility = 1; /* LOG_USER */
810 struct file *file = iocb->ki_filp;
811 struct devkmsg_user *user = file->private_data;
812 size_t len = iov_iter_count(from);
813 ssize_t ret = len;
814
815 if (!user || len > LOG_LINE_MAX)
816 return -EINVAL;
817
818 /* Ignore when user logging is disabled. */
819 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
820 return len;
821
822 /* Ratelimit when not explicitly enabled. */
823 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
824 if (!___ratelimit(&user->rs, current->comm))
825 return ret;
826 }
827
828 buf = kmalloc(len+1, GFP_KERNEL);
829 if (buf == NULL)
830 return -ENOMEM;
831
832 buf[len] = '\0';
833 if (!copy_from_iter_full(buf, len, from)) {
834 kfree(buf);
835 return -EFAULT;
836 }
837
838 /*
839 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
840 * the decimal value represents 32bit, the lower 3 bit are the log
841 * level, the rest are the log facility.
842 *
843 * If no prefix or no userspace facility is specified, we
844 * enforce LOG_USER, to be able to reliably distinguish
845 * kernel-generated messages from userspace-injected ones.
846 */
847 line = buf;
848 if (line[0] == '<') {
849 char *endp = NULL;
850 unsigned int u;
851
852 u = simple_strtoul(line + 1, &endp, 10);
853 if (endp && endp[0] == '>') {
854 level = LOG_LEVEL(u);
855 if (LOG_FACILITY(u) != 0)
856 facility = LOG_FACILITY(u);
857 endp++;
858 len -= endp - line;
859 line = endp;
860 }
861 }
862
863 devkmsg_emit(facility, level, "%s", line);
864 kfree(buf);
865 return ret;
866}
867
868static ssize_t devkmsg_read(struct file *file, char __user *buf,
869 size_t count, loff_t *ppos)
870{
871 struct devkmsg_user *user = file->private_data;
872 struct printk_log *msg;
873 size_t len;
874 ssize_t ret;
875
876 if (!user)
877 return -EBADF;
878
879 ret = mutex_lock_interruptible(&user->lock);
880 if (ret)
881 return ret;
882
883 logbuf_lock_irq();
884 while (user->seq == log_next_seq) {
885 if (file->f_flags & O_NONBLOCK) {
886 ret = -EAGAIN;
887 logbuf_unlock_irq();
888 goto out;
889 }
890
891 logbuf_unlock_irq();
892 ret = wait_event_interruptible(log_wait,
893 user->seq != log_next_seq);
894 if (ret)
895 goto out;
896 logbuf_lock_irq();
897 }
898
899 if (user->seq < log_first_seq) {
900 /* our last seen message is gone, return error and reset */
901 user->idx = log_first_idx;
902 user->seq = log_first_seq;
903 ret = -EPIPE;
904 logbuf_unlock_irq();
905 goto out;
906 }
907
908 msg = log_from_idx(user->idx);
909 len = msg_print_ext_header(user->buf, sizeof(user->buf),
910 msg, user->seq);
911 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
912 log_dict(msg), msg->dict_len,
913 log_text(msg), msg->text_len);
914
915 user->idx = log_next(user->idx);
916 user->seq++;
917 logbuf_unlock_irq();
918
919 if (len > count) {
920 ret = -EINVAL;
921 goto out;
922 }
923
924 if (copy_to_user(buf, user->buf, len)) {
925 ret = -EFAULT;
926 goto out;
927 }
928 ret = len;
929out:
930 mutex_unlock(&user->lock);
931 return ret;
932}
933
934static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
935{
936 struct devkmsg_user *user = file->private_data;
937 loff_t ret = 0;
938
939 if (!user)
940 return -EBADF;
941 if (offset)
942 return -ESPIPE;
943
944 logbuf_lock_irq();
945 switch (whence) {
946 case SEEK_SET:
947 /* the first record */
948 user->idx = log_first_idx;
949 user->seq = log_first_seq;
950 break;
951 case SEEK_DATA:
952 /*
953 * The first record after the last SYSLOG_ACTION_CLEAR,
954 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
955 * changes no global state, and does not clear anything.
956 */
957 user->idx = clear_idx;
958 user->seq = clear_seq;
959 break;
960 case SEEK_END:
961 /* after the last record */
962 user->idx = log_next_idx;
963 user->seq = log_next_seq;
964 break;
965 default:
966 ret = -EINVAL;
967 }
968 logbuf_unlock_irq();
969 return ret;
970}
971
972static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
973{
974 struct devkmsg_user *user = file->private_data;
975 __poll_t ret = 0;
976
977 if (!user)
978 return EPOLLERR|EPOLLNVAL;
979
980 poll_wait(file, &log_wait, wait);
981
982 logbuf_lock_irq();
983 if (user->seq < log_next_seq) {
984 /* return error when data has vanished underneath us */
985 if (user->seq < log_first_seq)
986 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
987 else
988 ret = EPOLLIN|EPOLLRDNORM;
989 }
990 logbuf_unlock_irq();
991
992 return ret;
993}
994
995static int devkmsg_open(struct inode *inode, struct file *file)
996{
997 struct devkmsg_user *user;
998 int err;
999
1000 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
1001 return -EPERM;
1002
1003 /* write-only does not need any file context */
1004 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
1005 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
1006 SYSLOG_FROM_READER);
1007 if (err)
1008 return err;
1009 }
1010
1011 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
1012 if (!user)
1013 return -ENOMEM;
1014
1015 ratelimit_default_init(&user->rs);
1016 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
1017
1018 mutex_init(&user->lock);
1019
1020 logbuf_lock_irq();
1021 user->idx = log_first_idx;
1022 user->seq = log_first_seq;
1023 logbuf_unlock_irq();
1024
1025 file->private_data = user;
1026 return 0;
1027}
1028
1029static int devkmsg_release(struct inode *inode, struct file *file)
1030{
1031 struct devkmsg_user *user = file->private_data;
1032
1033 if (!user)
1034 return 0;
1035
1036 ratelimit_state_exit(&user->rs);
1037
1038 mutex_destroy(&user->lock);
1039 kfree(user);
1040 return 0;
1041}
1042
1043const struct file_operations kmsg_fops = {
1044 .open = devkmsg_open,
1045 .read = devkmsg_read,
1046 .write_iter = devkmsg_write,
1047 .llseek = devkmsg_llseek,
1048 .poll = devkmsg_poll,
1049 .release = devkmsg_release,
1050};
1051
1052#ifdef CONFIG_CRASH_CORE
1053/*
1054 * This appends the listed symbols to /proc/vmcore
1055 *
1056 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1057 * obtain access to symbols that are otherwise very difficult to locate. These
1058 * symbols are specifically used so that utilities can access and extract the
1059 * dmesg log from a vmcore file after a crash.
1060 */
1061void log_buf_vmcoreinfo_setup(void)
1062{
1063 VMCOREINFO_SYMBOL(log_buf);
1064 VMCOREINFO_SYMBOL(log_buf_len);
1065 VMCOREINFO_SYMBOL(log_first_idx);
1066 VMCOREINFO_SYMBOL(clear_idx);
1067 VMCOREINFO_SYMBOL(log_next_idx);
1068 /*
1069 * Export struct printk_log size and field offsets. User space tools can
1070 * parse it and detect any changes to structure down the line.
1071 */
1072 VMCOREINFO_STRUCT_SIZE(printk_log);
1073 VMCOREINFO_OFFSET(printk_log, ts_nsec);
1074 VMCOREINFO_OFFSET(printk_log, len);
1075 VMCOREINFO_OFFSET(printk_log, text_len);
1076 VMCOREINFO_OFFSET(printk_log, dict_len);
1077#ifdef CONFIG_PRINTK_CALLER
1078 VMCOREINFO_OFFSET(printk_log, caller_id);
1079#endif
1080}
1081#endif
1082
1083/* requested log_buf_len from kernel cmdline */
1084static unsigned long __initdata new_log_buf_len;
1085
1086/* we practice scaling the ring buffer by powers of 2 */
1087static void __init log_buf_len_update(u64 size)
1088{
1089 if (size > (u64)LOG_BUF_LEN_MAX) {
1090 size = (u64)LOG_BUF_LEN_MAX;
1091 pr_err("log_buf over 2G is not supported.\n");
1092 }
1093
1094 if (size)
1095 size = roundup_pow_of_two(size);
1096 if (size > log_buf_len)
1097 new_log_buf_len = (unsigned long)size;
1098}
1099
1100/* save requested log_buf_len since it's too early to process it */
1101static int __init log_buf_len_setup(char *str)
1102{
1103 u64 size;
1104
1105 if (!str)
1106 return -EINVAL;
1107
1108 size = memparse(str, &str);
1109
1110 log_buf_len_update(size);
1111
1112 return 0;
1113}
1114early_param("log_buf_len", log_buf_len_setup);
1115
1116#ifdef CONFIG_SMP
1117#define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1118
1119static void __init log_buf_add_cpu(void)
1120{
1121 unsigned int cpu_extra;
1122
1123 /*
1124 * archs should set up cpu_possible_bits properly with
1125 * set_cpu_possible() after setup_arch() but just in
1126 * case lets ensure this is valid.
1127 */
1128 if (num_possible_cpus() == 1)
1129 return;
1130
1131 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1132
1133 /* by default this will only continue through for large > 64 CPUs */
1134 if (cpu_extra <= __LOG_BUF_LEN / 2)
1135 return;
1136
1137 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1138 __LOG_CPU_MAX_BUF_LEN);
1139 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1140 cpu_extra);
1141 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1142
1143 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1144}
1145#else /* !CONFIG_SMP */
1146static inline void log_buf_add_cpu(void) {}
1147#endif /* CONFIG_SMP */
1148
1149void __init setup_log_buf(int early)
1150{
1151 unsigned long flags;
1152 char *new_log_buf;
1153 unsigned int free;
1154
1155 if (log_buf != __log_buf)
1156 return;
1157
1158 if (!early && !new_log_buf_len)
1159 log_buf_add_cpu();
1160
1161 if (!new_log_buf_len)
1162 return;
1163
1164 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1165 if (unlikely(!new_log_buf)) {
1166 pr_err("log_buf_len: %lu bytes not available\n",
1167 new_log_buf_len);
1168 return;
1169 }
1170
1171 logbuf_lock_irqsave(flags);
1172 log_buf_len = new_log_buf_len;
1173 log_buf = new_log_buf;
1174 new_log_buf_len = 0;
1175 free = __LOG_BUF_LEN - log_next_idx;
1176 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1177 logbuf_unlock_irqrestore(flags);
1178
1179 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1180 pr_info("early log buf free: %u(%u%%)\n",
1181 free, (free * 100) / __LOG_BUF_LEN);
1182}
1183
1184static bool __read_mostly ignore_loglevel;
1185
1186static int __init ignore_loglevel_setup(char *str)
1187{
1188 ignore_loglevel = true;
1189 pr_info("debug: ignoring loglevel setting.\n");
1190
1191 return 0;
1192}
1193
1194early_param("ignore_loglevel", ignore_loglevel_setup);
1195module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1196MODULE_PARM_DESC(ignore_loglevel,
1197 "ignore loglevel setting (prints all kernel messages to the console)");
1198
1199static bool suppress_message_printing(int level)
1200{
1201 return (level >= console_loglevel && !ignore_loglevel);
1202}
1203
1204#ifdef CONFIG_BOOT_PRINTK_DELAY
1205
1206static int boot_delay; /* msecs delay after each printk during bootup */
1207static unsigned long long loops_per_msec; /* based on boot_delay */
1208
1209static int __init boot_delay_setup(char *str)
1210{
1211 unsigned long lpj;
1212
1213 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1214 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1215
1216 get_option(&str, &boot_delay);
1217 if (boot_delay > 10 * 1000)
1218 boot_delay = 0;
1219
1220 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1221 "HZ: %d, loops_per_msec: %llu\n",
1222 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1223 return 0;
1224}
1225early_param("boot_delay", boot_delay_setup);
1226
1227static void boot_delay_msec(int level)
1228{
1229 unsigned long long k;
1230 unsigned long timeout;
1231
1232 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1233 || suppress_message_printing(level)) {
1234 return;
1235 }
1236
1237 k = (unsigned long long)loops_per_msec * boot_delay;
1238
1239 timeout = jiffies + msecs_to_jiffies(boot_delay);
1240 while (k) {
1241 k--;
1242 cpu_relax();
1243 /*
1244 * use (volatile) jiffies to prevent
1245 * compiler reduction; loop termination via jiffies
1246 * is secondary and may or may not happen.
1247 */
1248 if (time_after(jiffies, timeout))
1249 break;
1250 touch_nmi_watchdog();
1251 }
1252}
1253#else
1254static inline void boot_delay_msec(int level)
1255{
1256}
1257#endif
1258
1259static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1260module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1261
1262static size_t print_syslog(unsigned int level, char *buf)
1263{
1264 return sprintf(buf, "<%u>", level);
1265}
1266
1267static size_t print_time(u64 ts, char *buf)
1268{
1269 unsigned long rem_nsec = do_div(ts, 1000000000);
1270
1271 return sprintf(buf, "[%5lu.%06lu]",
1272 (unsigned long)ts, rem_nsec / 1000);
1273}
1274
1275#ifdef CONFIG_PRINTK_CALLER
1276static size_t print_caller(u32 id, char *buf)
1277{
1278 char caller[12];
1279
1280 snprintf(caller, sizeof(caller), "%c%u",
1281 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1282 return sprintf(buf, "[%6s]", caller);
1283}
1284#else
1285#define print_caller(id, buf) 0
1286#endif
1287
1288static size_t print_prefix(const struct printk_log *msg, bool syslog,
1289 bool time, char *buf)
1290{
1291 size_t len = 0;
1292
1293 if (syslog)
1294 len = print_syslog((msg->facility << 3) | msg->level, buf);
1295
1296 if (time)
1297 len += print_time(msg->ts_nsec, buf + len);
1298
1299 len += print_caller(msg->caller_id, buf + len);
1300
1301 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1302 buf[len++] = ' ';
1303 buf[len] = '\0';
1304 }
1305
1306 return len;
1307}
1308
1309static size_t msg_print_text(const struct printk_log *msg, bool syslog,
1310 bool time, char *buf, size_t size)
1311{
1312 const char *text = log_text(msg);
1313 size_t text_size = msg->text_len;
1314 size_t len = 0;
1315 char prefix[PREFIX_MAX];
1316 const size_t prefix_len = print_prefix(msg, syslog, time, prefix);
1317
1318 do {
1319 const char *next = memchr(text, '\n', text_size);
1320 size_t text_len;
1321
1322 if (next) {
1323 text_len = next - text;
1324 next++;
1325 text_size -= next - text;
1326 } else {
1327 text_len = text_size;
1328 }
1329
1330 if (buf) {
1331 if (prefix_len + text_len + 1 >= size - len)
1332 break;
1333
1334 memcpy(buf + len, prefix, prefix_len);
1335 len += prefix_len;
1336 memcpy(buf + len, text, text_len);
1337 len += text_len;
1338 buf[len++] = '\n';
1339 } else {
1340 /* SYSLOG_ACTION_* buffer size only calculation */
1341 len += prefix_len + text_len + 1;
1342 }
1343
1344 text = next;
1345 } while (text);
1346
1347 return len;
1348}
1349
1350static int syslog_print(char __user *buf, int size)
1351{
1352 char *text;
1353 struct printk_log *msg;
1354 int len = 0;
1355
1356 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1357 if (!text)
1358 return -ENOMEM;
1359
1360 while (size > 0) {
1361 size_t n;
1362 size_t skip;
1363
1364 logbuf_lock_irq();
1365 if (syslog_seq < log_first_seq) {
1366 /* messages are gone, move to first one */
1367 syslog_seq = log_first_seq;
1368 syslog_idx = log_first_idx;
1369 syslog_partial = 0;
1370 }
1371 if (syslog_seq == log_next_seq) {
1372 logbuf_unlock_irq();
1373 break;
1374 }
1375
1376 /*
1377 * To keep reading/counting partial line consistent,
1378 * use printk_time value as of the beginning of a line.
1379 */
1380 if (!syslog_partial)
1381 syslog_time = printk_time;
1382
1383 skip = syslog_partial;
1384 msg = log_from_idx(syslog_idx);
1385 n = msg_print_text(msg, true, syslog_time, text,
1386 LOG_LINE_MAX + PREFIX_MAX);
1387 if (n - syslog_partial <= size) {
1388 /* message fits into buffer, move forward */
1389 syslog_idx = log_next(syslog_idx);
1390 syslog_seq++;
1391 n -= syslog_partial;
1392 syslog_partial = 0;
1393 } else if (!len){
1394 /* partial read(), remember position */
1395 n = size;
1396 syslog_partial += n;
1397 } else
1398 n = 0;
1399 logbuf_unlock_irq();
1400
1401 if (!n)
1402 break;
1403
1404 if (copy_to_user(buf, text + skip, n)) {
1405 if (!len)
1406 len = -EFAULT;
1407 break;
1408 }
1409
1410 len += n;
1411 size -= n;
1412 buf += n;
1413 }
1414
1415 kfree(text);
1416 return len;
1417}
1418
1419static int syslog_print_all(char __user *buf, int size, bool clear)
1420{
1421 char *text;
1422 int len = 0;
1423 u64 next_seq;
1424 u64 seq;
1425 u32 idx;
1426 bool time;
1427
1428 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1429 if (!text)
1430 return -ENOMEM;
1431
1432 time = printk_time;
1433 logbuf_lock_irq();
1434 /*
1435 * Find first record that fits, including all following records,
1436 * into the user-provided buffer for this dump.
1437 */
1438 seq = clear_seq;
1439 idx = clear_idx;
1440 while (seq < log_next_seq) {
1441 struct printk_log *msg = log_from_idx(idx);
1442
1443 len += msg_print_text(msg, true, time, NULL, 0);
1444 idx = log_next(idx);
1445 seq++;
1446 }
1447
1448 /* move first record forward until length fits into the buffer */
1449 seq = clear_seq;
1450 idx = clear_idx;
1451 while (len > size && seq < log_next_seq) {
1452 struct printk_log *msg = log_from_idx(idx);
1453
1454 len -= msg_print_text(msg, true, time, NULL, 0);
1455 idx = log_next(idx);
1456 seq++;
1457 }
1458
1459 /* last message fitting into this dump */
1460 next_seq = log_next_seq;
1461
1462 len = 0;
1463 while (len >= 0 && seq < next_seq) {
1464 struct printk_log *msg = log_from_idx(idx);
1465 int textlen = msg_print_text(msg, true, time, text,
1466 LOG_LINE_MAX + PREFIX_MAX);
1467
1468 idx = log_next(idx);
1469 seq++;
1470
1471 logbuf_unlock_irq();
1472 if (copy_to_user(buf + len, text, textlen))
1473 len = -EFAULT;
1474 else
1475 len += textlen;
1476 logbuf_lock_irq();
1477
1478 if (seq < log_first_seq) {
1479 /* messages are gone, move to next one */
1480 seq = log_first_seq;
1481 idx = log_first_idx;
1482 }
1483 }
1484
1485 if (clear) {
1486 clear_seq = log_next_seq;
1487 clear_idx = log_next_idx;
1488 }
1489 logbuf_unlock_irq();
1490
1491 kfree(text);
1492 return len;
1493}
1494
1495static void syslog_clear(void)
1496{
1497 logbuf_lock_irq();
1498 clear_seq = log_next_seq;
1499 clear_idx = log_next_idx;
1500 logbuf_unlock_irq();
1501}
1502
1503int do_syslog(int type, char __user *buf, int len, int source)
1504{
1505 bool clear = false;
1506 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1507 int error;
1508
1509 error = check_syslog_permissions(type, source);
1510 if (error)
1511 return error;
1512
1513 switch (type) {
1514 case SYSLOG_ACTION_CLOSE: /* Close log */
1515 break;
1516 case SYSLOG_ACTION_OPEN: /* Open log */
1517 break;
1518 case SYSLOG_ACTION_READ: /* Read from log */
1519 if (!buf || len < 0)
1520 return -EINVAL;
1521 if (!len)
1522 return 0;
1523 if (!access_ok(buf, len))
1524 return -EFAULT;
1525 error = wait_event_interruptible(log_wait,
1526 syslog_seq != log_next_seq);
1527 if (error)
1528 return error;
1529 error = syslog_print(buf, len);
1530 break;
1531 /* Read/clear last kernel messages */
1532 case SYSLOG_ACTION_READ_CLEAR:
1533 clear = true;
1534 /* FALL THRU */
1535 /* Read last kernel messages */
1536 case SYSLOG_ACTION_READ_ALL:
1537 if (!buf || len < 0)
1538 return -EINVAL;
1539 if (!len)
1540 return 0;
1541 if (!access_ok(buf, len))
1542 return -EFAULT;
1543 error = syslog_print_all(buf, len, clear);
1544 break;
1545 /* Clear ring buffer */
1546 case SYSLOG_ACTION_CLEAR:
1547 syslog_clear();
1548 break;
1549 /* Disable logging to console */
1550 case SYSLOG_ACTION_CONSOLE_OFF:
1551 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1552 saved_console_loglevel = console_loglevel;
1553 console_loglevel = minimum_console_loglevel;
1554 break;
1555 /* Enable logging to console */
1556 case SYSLOG_ACTION_CONSOLE_ON:
1557 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1558 console_loglevel = saved_console_loglevel;
1559 saved_console_loglevel = LOGLEVEL_DEFAULT;
1560 }
1561 break;
1562 /* Set level of messages printed to console */
1563 case SYSLOG_ACTION_CONSOLE_LEVEL:
1564 if (len < 1 || len > 8)
1565 return -EINVAL;
1566 if (len < minimum_console_loglevel)
1567 len = minimum_console_loglevel;
1568 console_loglevel = len;
1569 /* Implicitly re-enable logging to console */
1570 saved_console_loglevel = LOGLEVEL_DEFAULT;
1571 break;
1572 /* Number of chars in the log buffer */
1573 case SYSLOG_ACTION_SIZE_UNREAD:
1574 logbuf_lock_irq();
1575 if (syslog_seq < log_first_seq) {
1576 /* messages are gone, move to first one */
1577 syslog_seq = log_first_seq;
1578 syslog_idx = log_first_idx;
1579 syslog_partial = 0;
1580 }
1581 if (source == SYSLOG_FROM_PROC) {
1582 /*
1583 * Short-cut for poll(/"proc/kmsg") which simply checks
1584 * for pending data, not the size; return the count of
1585 * records, not the length.
1586 */
1587 error = log_next_seq - syslog_seq;
1588 } else {
1589 u64 seq = syslog_seq;
1590 u32 idx = syslog_idx;
1591 bool time = syslog_partial ? syslog_time : printk_time;
1592
1593 while (seq < log_next_seq) {
1594 struct printk_log *msg = log_from_idx(idx);
1595
1596 error += msg_print_text(msg, true, time, NULL,
1597 0);
1598 time = printk_time;
1599 idx = log_next(idx);
1600 seq++;
1601 }
1602 error -= syslog_partial;
1603 }
1604 logbuf_unlock_irq();
1605 break;
1606 /* Size of the log buffer */
1607 case SYSLOG_ACTION_SIZE_BUFFER:
1608 error = log_buf_len;
1609 break;
1610 default:
1611 error = -EINVAL;
1612 break;
1613 }
1614
1615 return error;
1616}
1617
1618SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1619{
1620 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1621}
1622
1623/*
1624 * Special console_lock variants that help to reduce the risk of soft-lockups.
1625 * They allow to pass console_lock to another printk() call using a busy wait.
1626 */
1627
1628#ifdef CONFIG_LOCKDEP
1629static struct lockdep_map console_owner_dep_map = {
1630 .name = "console_owner"
1631};
1632#endif
1633
1634static DEFINE_RAW_SPINLOCK(console_owner_lock);
1635static struct task_struct *console_owner;
1636static bool console_waiter;
1637
1638/**
1639 * console_lock_spinning_enable - mark beginning of code where another
1640 * thread might safely busy wait
1641 *
1642 * This basically converts console_lock into a spinlock. This marks
1643 * the section where the console_lock owner can not sleep, because
1644 * there may be a waiter spinning (like a spinlock). Also it must be
1645 * ready to hand over the lock at the end of the section.
1646 */
1647static void console_lock_spinning_enable(void)
1648{
1649 raw_spin_lock(&console_owner_lock);
1650 console_owner = current;
1651 raw_spin_unlock(&console_owner_lock);
1652
1653 /* The waiter may spin on us after setting console_owner */
1654 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1655}
1656
1657/**
1658 * console_lock_spinning_disable_and_check - mark end of code where another
1659 * thread was able to busy wait and check if there is a waiter
1660 *
1661 * This is called at the end of the section where spinning is allowed.
1662 * It has two functions. First, it is a signal that it is no longer
1663 * safe to start busy waiting for the lock. Second, it checks if
1664 * there is a busy waiter and passes the lock rights to her.
1665 *
1666 * Important: Callers lose the lock if there was a busy waiter.
1667 * They must not touch items synchronized by console_lock
1668 * in this case.
1669 *
1670 * Return: 1 if the lock rights were passed, 0 otherwise.
1671 */
1672static int console_lock_spinning_disable_and_check(void)
1673{
1674 int waiter;
1675
1676 raw_spin_lock(&console_owner_lock);
1677 waiter = READ_ONCE(console_waiter);
1678 console_owner = NULL;
1679 raw_spin_unlock(&console_owner_lock);
1680
1681 if (!waiter) {
1682 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1683 return 0;
1684 }
1685
1686 /* The waiter is now free to continue */
1687 WRITE_ONCE(console_waiter, false);
1688
1689 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1690
1691 /*
1692 * Hand off console_lock to waiter. The waiter will perform
1693 * the up(). After this, the waiter is the console_lock owner.
1694 */
1695 mutex_release(&console_lock_dep_map, 1, _THIS_IP_);
1696 return 1;
1697}
1698
1699/**
1700 * console_trylock_spinning - try to get console_lock by busy waiting
1701 *
1702 * This allows to busy wait for the console_lock when the current
1703 * owner is running in specially marked sections. It means that
1704 * the current owner is running and cannot reschedule until it
1705 * is ready to lose the lock.
1706 *
1707 * Return: 1 if we got the lock, 0 othrewise
1708 */
1709static int console_trylock_spinning(void)
1710{
1711 struct task_struct *owner = NULL;
1712 bool waiter;
1713 bool spin = false;
1714 unsigned long flags;
1715
1716 if (console_trylock())
1717 return 1;
1718
1719 printk_safe_enter_irqsave(flags);
1720
1721 raw_spin_lock(&console_owner_lock);
1722 owner = READ_ONCE(console_owner);
1723 waiter = READ_ONCE(console_waiter);
1724 if (!waiter && owner && owner != current) {
1725 WRITE_ONCE(console_waiter, true);
1726 spin = true;
1727 }
1728 raw_spin_unlock(&console_owner_lock);
1729
1730 /*
1731 * If there is an active printk() writing to the
1732 * consoles, instead of having it write our data too,
1733 * see if we can offload that load from the active
1734 * printer, and do some printing ourselves.
1735 * Go into a spin only if there isn't already a waiter
1736 * spinning, and there is an active printer, and
1737 * that active printer isn't us (recursive printk?).
1738 */
1739 if (!spin) {
1740 printk_safe_exit_irqrestore(flags);
1741 return 0;
1742 }
1743
1744 /* We spin waiting for the owner to release us */
1745 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1746 /* Owner will clear console_waiter on hand off */
1747 while (READ_ONCE(console_waiter))
1748 cpu_relax();
1749 spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1750
1751 printk_safe_exit_irqrestore(flags);
1752 /*
1753 * The owner passed the console lock to us.
1754 * Since we did not spin on console lock, annotate
1755 * this as a trylock. Otherwise lockdep will
1756 * complain.
1757 */
1758 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1759
1760 return 1;
1761}
1762
1763/*
1764 * Call the console drivers, asking them to write out
1765 * log_buf[start] to log_buf[end - 1].
1766 * The console_lock must be held.
1767 */
1768static void call_console_drivers(const char *ext_text, size_t ext_len,
1769 const char *text, size_t len)
1770{
1771 struct console *con;
1772
1773 trace_console_rcuidle(text, len);
1774
1775 if (!console_drivers)
1776 return;
1777
1778 for_each_console(con) {
1779 if (exclusive_console && con != exclusive_console)
1780 continue;
1781 if (!(con->flags & CON_ENABLED))
1782 continue;
1783 if (!con->write)
1784 continue;
1785 if (!cpu_online(smp_processor_id()) &&
1786 !(con->flags & CON_ANYTIME))
1787 continue;
1788 if (con->flags & CON_EXTENDED)
1789 con->write(con, ext_text, ext_len);
1790 else
1791 con->write(con, text, len);
1792 }
1793}
1794
1795int printk_delay_msec __read_mostly;
1796
1797static inline void printk_delay(void)
1798{
1799 if (unlikely(printk_delay_msec)) {
1800 int m = printk_delay_msec;
1801
1802 while (m--) {
1803 mdelay(1);
1804 touch_nmi_watchdog();
1805 }
1806 }
1807}
1808
1809static inline u32 printk_caller_id(void)
1810{
1811 return in_task() ? task_pid_nr(current) :
1812 0x80000000 + raw_smp_processor_id();
1813}
1814
1815/*
1816 * Continuation lines are buffered, and not committed to the record buffer
1817 * until the line is complete, or a race forces it. The line fragments
1818 * though, are printed immediately to the consoles to ensure everything has
1819 * reached the console in case of a kernel crash.
1820 */
1821static struct cont {
1822 char buf[LOG_LINE_MAX];
1823 size_t len; /* length == 0 means unused buffer */
1824 u32 caller_id; /* printk_caller_id() of first print */
1825 u64 ts_nsec; /* time of first print */
1826 u8 level; /* log level of first message */
1827 u8 facility; /* log facility of first message */
1828 enum log_flags flags; /* prefix, newline flags */
1829} cont;
1830
1831static void cont_flush(void)
1832{
1833 if (cont.len == 0)
1834 return;
1835
1836 log_store(cont.caller_id, cont.facility, cont.level, cont.flags,
1837 cont.ts_nsec, NULL, 0, cont.buf, cont.len);
1838 cont.len = 0;
1839}
1840
1841static bool cont_add(u32 caller_id, int facility, int level,
1842 enum log_flags flags, const char *text, size_t len)
1843{
1844 /* If the line gets too long, split it up in separate records. */
1845 if (cont.len + len > sizeof(cont.buf)) {
1846 cont_flush();
1847 return false;
1848 }
1849
1850 if (!cont.len) {
1851 cont.facility = facility;
1852 cont.level = level;
1853 cont.caller_id = caller_id;
1854 cont.ts_nsec = local_clock();
1855 cont.flags = flags;
1856 }
1857
1858 memcpy(cont.buf + cont.len, text, len);
1859 cont.len += len;
1860
1861 // The original flags come from the first line,
1862 // but later continuations can add a newline.
1863 if (flags & LOG_NEWLINE) {
1864 cont.flags |= LOG_NEWLINE;
1865 cont_flush();
1866 }
1867
1868 return true;
1869}
1870
1871static size_t log_output(int facility, int level, enum log_flags lflags, const char *dict, size_t dictlen, char *text, size_t text_len)
1872{
1873 const u32 caller_id = printk_caller_id();
1874
1875 /*
1876 * If an earlier line was buffered, and we're a continuation
1877 * write from the same context, try to add it to the buffer.
1878 */
1879 if (cont.len) {
1880 if (cont.caller_id == caller_id && (lflags & LOG_CONT)) {
1881 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1882 return text_len;
1883 }
1884 /* Otherwise, make sure it's flushed */
1885 cont_flush();
1886 }
1887
1888 /* Skip empty continuation lines that couldn't be added - they just flush */
1889 if (!text_len && (lflags & LOG_CONT))
1890 return 0;
1891
1892 /* If it doesn't end in a newline, try to buffer the current line */
1893 if (!(lflags & LOG_NEWLINE)) {
1894 if (cont_add(caller_id, facility, level, lflags, text, text_len))
1895 return text_len;
1896 }
1897
1898 /* Store it in the record log */
1899 return log_store(caller_id, facility, level, lflags, 0,
1900 dict, dictlen, text, text_len);
1901}
1902
1903/* Must be called under logbuf_lock. */
1904int vprintk_store(int facility, int level,
1905 const char *dict, size_t dictlen,
1906 const char *fmt, va_list args)
1907{
1908 static char textbuf[LOG_LINE_MAX];
1909 char *text = textbuf;
1910 size_t text_len;
1911 enum log_flags lflags = 0;
1912
1913 /*
1914 * The printf needs to come first; we need the syslog
1915 * prefix which might be passed-in as a parameter.
1916 */
1917 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1918
1919 /* mark and strip a trailing newline */
1920 if (text_len && text[text_len-1] == '\n') {
1921 text_len--;
1922 lflags |= LOG_NEWLINE;
1923 }
1924
1925 /* strip kernel syslog prefix and extract log level or control flags */
1926 if (facility == 0) {
1927 int kern_level;
1928
1929 while ((kern_level = printk_get_level(text)) != 0) {
1930 switch (kern_level) {
1931 case '0' ... '7':
1932 if (level == LOGLEVEL_DEFAULT)
1933 level = kern_level - '0';
1934 break;
1935 case 'c': /* KERN_CONT */
1936 lflags |= LOG_CONT;
1937 }
1938
1939 text_len -= 2;
1940 text += 2;
1941 }
1942 }
1943
1944 if (level == LOGLEVEL_DEFAULT)
1945 level = default_message_loglevel;
1946
1947 if (dict)
1948 lflags |= LOG_NEWLINE;
1949
1950 return log_output(facility, level, lflags,
1951 dict, dictlen, text, text_len);
1952}
1953
1954asmlinkage int vprintk_emit(int facility, int level,
1955 const char *dict, size_t dictlen,
1956 const char *fmt, va_list args)
1957{
1958 int printed_len;
1959 bool in_sched = false, pending_output;
1960 unsigned long flags;
1961 u64 curr_log_seq;
1962
1963 /* Suppress unimportant messages after panic happens */
1964 if (unlikely(suppress_printk))
1965 return 0;
1966
1967 if (level == LOGLEVEL_SCHED) {
1968 level = LOGLEVEL_DEFAULT;
1969 in_sched = true;
1970 }
1971
1972 boot_delay_msec(level);
1973 printk_delay();
1974
1975 /* This stops the holder of console_sem just where we want him */
1976 logbuf_lock_irqsave(flags);
1977 curr_log_seq = log_next_seq;
1978 printed_len = vprintk_store(facility, level, dict, dictlen, fmt, args);
1979 pending_output = (curr_log_seq != log_next_seq);
1980 logbuf_unlock_irqrestore(flags);
1981
1982 /* If called from the scheduler, we can not call up(). */
1983 if (!in_sched && pending_output) {
1984 /*
1985 * Disable preemption to avoid being preempted while holding
1986 * console_sem which would prevent anyone from printing to
1987 * console
1988 */
1989 preempt_disable();
1990 /*
1991 * Try to acquire and then immediately release the console
1992 * semaphore. The release will print out buffers and wake up
1993 * /dev/kmsg and syslog() users.
1994 */
1995 if (console_trylock_spinning())
1996 console_unlock();
1997 preempt_enable();
1998 }
1999
2000 if (pending_output)
2001 wake_up_klogd();
2002 return printed_len;
2003}
2004EXPORT_SYMBOL(vprintk_emit);
2005
2006asmlinkage int vprintk(const char *fmt, va_list args)
2007{
2008 return vprintk_func(fmt, args);
2009}
2010EXPORT_SYMBOL(vprintk);
2011
2012int vprintk_default(const char *fmt, va_list args)
2013{
2014 int r;
2015
2016#ifdef CONFIG_KGDB_KDB
2017 /* Allow to pass printk() to kdb but avoid a recursion. */
2018 if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
2019 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
2020 return r;
2021 }
2022#endif
2023 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
2024
2025 return r;
2026}
2027EXPORT_SYMBOL_GPL(vprintk_default);
2028
2029/**
2030 * printk - print a kernel message
2031 * @fmt: format string
2032 *
2033 * This is printk(). It can be called from any context. We want it to work.
2034 *
2035 * We try to grab the console_lock. If we succeed, it's easy - we log the
2036 * output and call the console drivers. If we fail to get the semaphore, we
2037 * place the output into the log buffer and return. The current holder of
2038 * the console_sem will notice the new output in console_unlock(); and will
2039 * send it to the consoles before releasing the lock.
2040 *
2041 * One effect of this deferred printing is that code which calls printk() and
2042 * then changes console_loglevel may break. This is because console_loglevel
2043 * is inspected when the actual printing occurs.
2044 *
2045 * See also:
2046 * printf(3)
2047 *
2048 * See the vsnprintf() documentation for format string extensions over C99.
2049 */
2050asmlinkage __visible int printk(const char *fmt, ...)
2051{
2052 va_list args;
2053 int r;
2054
2055 va_start(args, fmt);
2056 r = vprintk_func(fmt, args);
2057 va_end(args);
2058
2059 return r;
2060}
2061EXPORT_SYMBOL(printk);
2062
2063#else /* CONFIG_PRINTK */
2064
2065#define LOG_LINE_MAX 0
2066#define PREFIX_MAX 0
2067#define printk_time false
2068
2069static u64 syslog_seq;
2070static u32 syslog_idx;
2071static u64 console_seq;
2072static u32 console_idx;
2073static u64 exclusive_console_stop_seq;
2074static u64 log_first_seq;
2075static u32 log_first_idx;
2076static u64 log_next_seq;
2077static char *log_text(const struct printk_log *msg) { return NULL; }
2078static char *log_dict(const struct printk_log *msg) { return NULL; }
2079static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2080static u32 log_next(u32 idx) { return 0; }
2081static ssize_t msg_print_ext_header(char *buf, size_t size,
2082 struct printk_log *msg,
2083 u64 seq) { return 0; }
2084static ssize_t msg_print_ext_body(char *buf, size_t size,
2085 char *dict, size_t dict_len,
2086 char *text, size_t text_len) { return 0; }
2087static void console_lock_spinning_enable(void) { }
2088static int console_lock_spinning_disable_and_check(void) { return 0; }
2089static void call_console_drivers(const char *ext_text, size_t ext_len,
2090 const char *text, size_t len) {}
2091static size_t msg_print_text(const struct printk_log *msg, bool syslog,
2092 bool time, char *buf, size_t size) { return 0; }
2093static bool suppress_message_printing(int level) { return false; }
2094
2095#endif /* CONFIG_PRINTK */
2096
2097#ifdef CONFIG_EARLY_PRINTK
2098struct console *early_console;
2099
2100asmlinkage __visible void early_printk(const char *fmt, ...)
2101{
2102 va_list ap;
2103 char buf[512];
2104 int n;
2105
2106 if (!early_console)
2107 return;
2108
2109 va_start(ap, fmt);
2110 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2111 va_end(ap);
2112
2113 early_console->write(early_console, buf, n);
2114}
2115#endif
2116
2117static int __add_preferred_console(char *name, int idx, char *options,
2118 char *brl_options)
2119{
2120 struct console_cmdline *c;
2121 int i;
2122
2123 /*
2124 * See if this tty is not yet registered, and
2125 * if we have a slot free.
2126 */
2127 for (i = 0, c = console_cmdline;
2128 i < MAX_CMDLINECONSOLES && c->name[0];
2129 i++, c++) {
2130 if (strcmp(c->name, name) == 0 && c->index == idx) {
2131 if (!brl_options)
2132 preferred_console = i;
2133 return 0;
2134 }
2135 }
2136 if (i == MAX_CMDLINECONSOLES)
2137 return -E2BIG;
2138 if (!brl_options)
2139 preferred_console = i;
2140 strlcpy(c->name, name, sizeof(c->name));
2141 c->options = options;
2142 braille_set_options(c, brl_options);
2143
2144 c->index = idx;
2145 return 0;
2146}
2147
2148static int __init console_msg_format_setup(char *str)
2149{
2150 if (!strcmp(str, "syslog"))
2151 console_msg_format = MSG_FORMAT_SYSLOG;
2152 if (!strcmp(str, "default"))
2153 console_msg_format = MSG_FORMAT_DEFAULT;
2154 return 1;
2155}
2156__setup("console_msg_format=", console_msg_format_setup);
2157
2158/*
2159 * Set up a console. Called via do_early_param() in init/main.c
2160 * for each "console=" parameter in the boot command line.
2161 */
2162static int __init console_setup(char *str)
2163{
2164 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2165 char *s, *options, *brl_options = NULL;
2166 int idx;
2167
2168 if (_braille_console_setup(&str, &brl_options))
2169 return 1;
2170
2171 /*
2172 * Decode str into name, index, options.
2173 */
2174 if (str[0] >= '0' && str[0] <= '9') {
2175 strcpy(buf, "ttyS");
2176 strncpy(buf + 4, str, sizeof(buf) - 5);
2177 } else {
2178 strncpy(buf, str, sizeof(buf) - 1);
2179 }
2180 buf[sizeof(buf) - 1] = 0;
2181 options = strchr(str, ',');
2182 if (options)
2183 *(options++) = 0;
2184#ifdef __sparc__
2185 if (!strcmp(str, "ttya"))
2186 strcpy(buf, "ttyS0");
2187 if (!strcmp(str, "ttyb"))
2188 strcpy(buf, "ttyS1");
2189#endif
2190 for (s = buf; *s; s++)
2191 if (isdigit(*s) || *s == ',')
2192 break;
2193 idx = simple_strtoul(s, NULL, 10);
2194 *s = 0;
2195
2196 __add_preferred_console(buf, idx, options, brl_options);
2197 console_set_on_cmdline = 1;
2198 return 1;
2199}
2200__setup("console=", console_setup);
2201
2202/**
2203 * add_preferred_console - add a device to the list of preferred consoles.
2204 * @name: device name
2205 * @idx: device index
2206 * @options: options for this console
2207 *
2208 * The last preferred console added will be used for kernel messages
2209 * and stdin/out/err for init. Normally this is used by console_setup
2210 * above to handle user-supplied console arguments; however it can also
2211 * be used by arch-specific code either to override the user or more
2212 * commonly to provide a default console (ie from PROM variables) when
2213 * the user has not supplied one.
2214 */
2215int add_preferred_console(char *name, int idx, char *options)
2216{
2217 return __add_preferred_console(name, idx, options, NULL);
2218}
2219
2220bool console_suspend_enabled = true;
2221EXPORT_SYMBOL(console_suspend_enabled);
2222
2223static int __init console_suspend_disable(char *str)
2224{
2225 console_suspend_enabled = false;
2226 return 1;
2227}
2228__setup("no_console_suspend", console_suspend_disable);
2229module_param_named(console_suspend, console_suspend_enabled,
2230 bool, S_IRUGO | S_IWUSR);
2231MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2232 " and hibernate operations");
2233
2234/**
2235 * suspend_console - suspend the console subsystem
2236 *
2237 * This disables printk() while we go into suspend states
2238 */
2239void suspend_console(void)
2240{
2241 if (!console_suspend_enabled)
2242 return;
2243 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2244 console_lock();
2245 console_suspended = 1;
2246 up_console_sem();
2247}
2248
2249void resume_console(void)
2250{
2251 if (!console_suspend_enabled)
2252 return;
2253 down_console_sem();
2254 console_suspended = 0;
2255 console_unlock();
2256}
2257
2258/**
2259 * console_cpu_notify - print deferred console messages after CPU hotplug
2260 * @cpu: unused
2261 *
2262 * If printk() is called from a CPU that is not online yet, the messages
2263 * will be printed on the console only if there are CON_ANYTIME consoles.
2264 * This function is called when a new CPU comes online (or fails to come
2265 * up) or goes offline.
2266 */
2267static int console_cpu_notify(unsigned int cpu)
2268{
2269 if (!cpuhp_tasks_frozen) {
2270 /* If trylock fails, someone else is doing the printing */
2271 if (console_trylock())
2272 console_unlock();
2273 }
2274 return 0;
2275}
2276
2277/**
2278 * console_lock - lock the console system for exclusive use.
2279 *
2280 * Acquires a lock which guarantees that the caller has
2281 * exclusive access to the console system and the console_drivers list.
2282 *
2283 * Can sleep, returns nothing.
2284 */
2285void console_lock(void)
2286{
2287 might_sleep();
2288
2289 down_console_sem();
2290 if (console_suspended)
2291 return;
2292 console_locked = 1;
2293 console_may_schedule = 1;
2294}
2295EXPORT_SYMBOL(console_lock);
2296
2297/**
2298 * console_trylock - try to lock the console system for exclusive use.
2299 *
2300 * Try to acquire a lock which guarantees that the caller has exclusive
2301 * access to the console system and the console_drivers list.
2302 *
2303 * returns 1 on success, and 0 on failure to acquire the lock.
2304 */
2305int console_trylock(void)
2306{
2307 if (down_trylock_console_sem())
2308 return 0;
2309 if (console_suspended) {
2310 up_console_sem();
2311 return 0;
2312 }
2313 console_locked = 1;
2314 console_may_schedule = 0;
2315 return 1;
2316}
2317EXPORT_SYMBOL(console_trylock);
2318
2319int is_console_locked(void)
2320{
2321 return console_locked;
2322}
2323EXPORT_SYMBOL(is_console_locked);
2324
2325/*
2326 * Check if we have any console that is capable of printing while cpu is
2327 * booting or shutting down. Requires console_sem.
2328 */
2329static int have_callable_console(void)
2330{
2331 struct console *con;
2332
2333 for_each_console(con)
2334 if ((con->flags & CON_ENABLED) &&
2335 (con->flags & CON_ANYTIME))
2336 return 1;
2337
2338 return 0;
2339}
2340
2341/*
2342 * Can we actually use the console at this time on this cpu?
2343 *
2344 * Console drivers may assume that per-cpu resources have been allocated. So
2345 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2346 * call them until this CPU is officially up.
2347 */
2348static inline int can_use_console(void)
2349{
2350 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2351}
2352
2353/**
2354 * console_unlock - unlock the console system
2355 *
2356 * Releases the console_lock which the caller holds on the console system
2357 * and the console driver list.
2358 *
2359 * While the console_lock was held, console output may have been buffered
2360 * by printk(). If this is the case, console_unlock(); emits
2361 * the output prior to releasing the lock.
2362 *
2363 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2364 *
2365 * console_unlock(); may be called from any context.
2366 */
2367void console_unlock(void)
2368{
2369 static char ext_text[CONSOLE_EXT_LOG_MAX];
2370 static char text[LOG_LINE_MAX + PREFIX_MAX];
2371 unsigned long flags;
2372 bool do_cond_resched, retry;
2373
2374 if (console_suspended) {
2375 up_console_sem();
2376 return;
2377 }
2378
2379 /*
2380 * Console drivers are called with interrupts disabled, so
2381 * @console_may_schedule should be cleared before; however, we may
2382 * end up dumping a lot of lines, for example, if called from
2383 * console registration path, and should invoke cond_resched()
2384 * between lines if allowable. Not doing so can cause a very long
2385 * scheduling stall on a slow console leading to RCU stall and
2386 * softlockup warnings which exacerbate the issue with more
2387 * messages practically incapacitating the system.
2388 *
2389 * console_trylock() is not able to detect the preemptive
2390 * context reliably. Therefore the value must be stored before
2391 * and cleared after the the "again" goto label.
2392 */
2393 do_cond_resched = console_may_schedule;
2394again:
2395 console_may_schedule = 0;
2396
2397 /*
2398 * We released the console_sem lock, so we need to recheck if
2399 * cpu is online and (if not) is there at least one CON_ANYTIME
2400 * console.
2401 */
2402 if (!can_use_console()) {
2403 console_locked = 0;
2404 up_console_sem();
2405 return;
2406 }
2407
2408 for (;;) {
2409 struct printk_log *msg;
2410 size_t ext_len = 0;
2411 size_t len;
2412
2413 printk_safe_enter_irqsave(flags);
2414 raw_spin_lock(&logbuf_lock);
2415 if (console_seq < log_first_seq) {
2416 len = sprintf(text,
2417 "** %llu printk messages dropped **\n",
2418 log_first_seq - console_seq);
2419
2420 /* messages are gone, move to first one */
2421 console_seq = log_first_seq;
2422 console_idx = log_first_idx;
2423 } else {
2424 len = 0;
2425 }
2426skip:
2427 if (console_seq == log_next_seq)
2428 break;
2429
2430 msg = log_from_idx(console_idx);
2431 if (suppress_message_printing(msg->level)) {
2432 /*
2433 * Skip record we have buffered and already printed
2434 * directly to the console when we received it, and
2435 * record that has level above the console loglevel.
2436 */
2437 console_idx = log_next(console_idx);
2438 console_seq++;
2439 goto skip;
2440 }
2441
2442 /* Output to all consoles once old messages replayed. */
2443 if (unlikely(exclusive_console &&
2444 console_seq >= exclusive_console_stop_seq)) {
2445 exclusive_console = NULL;
2446 }
2447
2448 len += msg_print_text(msg,
2449 console_msg_format & MSG_FORMAT_SYSLOG,
2450 printk_time, text + len, sizeof(text) - len);
2451 if (nr_ext_console_drivers) {
2452 ext_len = msg_print_ext_header(ext_text,
2453 sizeof(ext_text),
2454 msg, console_seq);
2455 ext_len += msg_print_ext_body(ext_text + ext_len,
2456 sizeof(ext_text) - ext_len,
2457 log_dict(msg), msg->dict_len,
2458 log_text(msg), msg->text_len);
2459 }
2460 console_idx = log_next(console_idx);
2461 console_seq++;
2462 raw_spin_unlock(&logbuf_lock);
2463
2464 /*
2465 * While actively printing out messages, if another printk()
2466 * were to occur on another CPU, it may wait for this one to
2467 * finish. This task can not be preempted if there is a
2468 * waiter waiting to take over.
2469 */
2470 console_lock_spinning_enable();
2471
2472 stop_critical_timings(); /* don't trace print latency */
2473 call_console_drivers(ext_text, ext_len, text, len);
2474 start_critical_timings();
2475
2476 if (console_lock_spinning_disable_and_check()) {
2477 printk_safe_exit_irqrestore(flags);
2478 return;
2479 }
2480
2481 printk_safe_exit_irqrestore(flags);
2482
2483 if (do_cond_resched)
2484 cond_resched();
2485 }
2486
2487 console_locked = 0;
2488
2489 raw_spin_unlock(&logbuf_lock);
2490
2491 up_console_sem();
2492
2493 /*
2494 * Someone could have filled up the buffer again, so re-check if there's
2495 * something to flush. In case we cannot trylock the console_sem again,
2496 * there's a new owner and the console_unlock() from them will do the
2497 * flush, no worries.
2498 */
2499 raw_spin_lock(&logbuf_lock);
2500 retry = console_seq != log_next_seq;
2501 raw_spin_unlock(&logbuf_lock);
2502 printk_safe_exit_irqrestore(flags);
2503
2504 if (retry && console_trylock())
2505 goto again;
2506}
2507EXPORT_SYMBOL(console_unlock);
2508
2509/**
2510 * console_conditional_schedule - yield the CPU if required
2511 *
2512 * If the console code is currently allowed to sleep, and
2513 * if this CPU should yield the CPU to another task, do
2514 * so here.
2515 *
2516 * Must be called within console_lock();.
2517 */
2518void __sched console_conditional_schedule(void)
2519{
2520 if (console_may_schedule)
2521 cond_resched();
2522}
2523EXPORT_SYMBOL(console_conditional_schedule);
2524
2525void console_unblank(void)
2526{
2527 struct console *c;
2528
2529 /*
2530 * console_unblank can no longer be called in interrupt context unless
2531 * oops_in_progress is set to 1..
2532 */
2533 if (oops_in_progress) {
2534 if (down_trylock_console_sem() != 0)
2535 return;
2536 } else
2537 console_lock();
2538
2539 console_locked = 1;
2540 console_may_schedule = 0;
2541 for_each_console(c)
2542 if ((c->flags & CON_ENABLED) && c->unblank)
2543 c->unblank();
2544 console_unlock();
2545}
2546
2547/**
2548 * console_flush_on_panic - flush console content on panic
2549 * @mode: flush all messages in buffer or just the pending ones
2550 *
2551 * Immediately output all pending messages no matter what.
2552 */
2553void console_flush_on_panic(enum con_flush_mode mode)
2554{
2555 /*
2556 * If someone else is holding the console lock, trylock will fail
2557 * and may_schedule may be set. Ignore and proceed to unlock so
2558 * that messages are flushed out. As this can be called from any
2559 * context and we don't want to get preempted while flushing,
2560 * ensure may_schedule is cleared.
2561 */
2562 console_trylock();
2563 console_may_schedule = 0;
2564
2565 if (mode == CONSOLE_REPLAY_ALL) {
2566 unsigned long flags;
2567
2568 logbuf_lock_irqsave(flags);
2569 console_seq = log_first_seq;
2570 console_idx = log_first_idx;
2571 logbuf_unlock_irqrestore(flags);
2572 }
2573 console_unlock();
2574}
2575
2576/*
2577 * Return the console tty driver structure and its associated index
2578 */
2579struct tty_driver *console_device(int *index)
2580{
2581 struct console *c;
2582 struct tty_driver *driver = NULL;
2583
2584 console_lock();
2585 for_each_console(c) {
2586 if (!c->device)
2587 continue;
2588 driver = c->device(c, index);
2589 if (driver)
2590 break;
2591 }
2592 console_unlock();
2593 return driver;
2594}
2595
2596/*
2597 * Prevent further output on the passed console device so that (for example)
2598 * serial drivers can disable console output before suspending a port, and can
2599 * re-enable output afterwards.
2600 */
2601void console_stop(struct console *console)
2602{
2603 console_lock();
2604 console->flags &= ~CON_ENABLED;
2605 console_unlock();
2606}
2607EXPORT_SYMBOL(console_stop);
2608
2609void console_start(struct console *console)
2610{
2611 console_lock();
2612 console->flags |= CON_ENABLED;
2613 console_unlock();
2614}
2615EXPORT_SYMBOL(console_start);
2616
2617static int __read_mostly keep_bootcon;
2618
2619static int __init keep_bootcon_setup(char *str)
2620{
2621 keep_bootcon = 1;
2622 pr_info("debug: skip boot console de-registration.\n");
2623
2624 return 0;
2625}
2626
2627early_param("keep_bootcon", keep_bootcon_setup);
2628
2629/*
2630 * The console driver calls this routine during kernel initialization
2631 * to register the console printing procedure with printk() and to
2632 * print any messages that were printed by the kernel before the
2633 * console driver was initialized.
2634 *
2635 * This can happen pretty early during the boot process (because of
2636 * early_printk) - sometimes before setup_arch() completes - be careful
2637 * of what kernel features are used - they may not be initialised yet.
2638 *
2639 * There are two types of consoles - bootconsoles (early_printk) and
2640 * "real" consoles (everything which is not a bootconsole) which are
2641 * handled differently.
2642 * - Any number of bootconsoles can be registered at any time.
2643 * - As soon as a "real" console is registered, all bootconsoles
2644 * will be unregistered automatically.
2645 * - Once a "real" console is registered, any attempt to register a
2646 * bootconsoles will be rejected
2647 */
2648void register_console(struct console *newcon)
2649{
2650 int i;
2651 unsigned long flags;
2652 struct console *bcon = NULL;
2653 struct console_cmdline *c;
2654 static bool has_preferred;
2655
2656 if (console_drivers)
2657 for_each_console(bcon)
2658 if (WARN(bcon == newcon,
2659 "console '%s%d' already registered\n",
2660 bcon->name, bcon->index))
2661 return;
2662
2663 /*
2664 * before we register a new CON_BOOT console, make sure we don't
2665 * already have a valid console
2666 */
2667 if (console_drivers && newcon->flags & CON_BOOT) {
2668 /* find the last or real console */
2669 for_each_console(bcon) {
2670 if (!(bcon->flags & CON_BOOT)) {
2671 pr_info("Too late to register bootconsole %s%d\n",
2672 newcon->name, newcon->index);
2673 return;
2674 }
2675 }
2676 }
2677
2678 if (console_drivers && console_drivers->flags & CON_BOOT)
2679 bcon = console_drivers;
2680
2681 if (!has_preferred || bcon || !console_drivers)
2682 has_preferred = preferred_console >= 0;
2683
2684 /*
2685 * See if we want to use this console driver. If we
2686 * didn't select a console we take the first one
2687 * that registers here.
2688 */
2689 if (!has_preferred) {
2690 if (newcon->index < 0)
2691 newcon->index = 0;
2692 if (newcon->setup == NULL ||
2693 newcon->setup(newcon, NULL) == 0) {
2694 newcon->flags |= CON_ENABLED;
2695 if (newcon->device) {
2696 newcon->flags |= CON_CONSDEV;
2697 has_preferred = true;
2698 }
2699 }
2700 }
2701
2702 /*
2703 * See if this console matches one we selected on
2704 * the command line.
2705 */
2706 for (i = 0, c = console_cmdline;
2707 i < MAX_CMDLINECONSOLES && c->name[0];
2708 i++, c++) {
2709 if (!newcon->match ||
2710 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2711 /* default matching */
2712 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2713 if (strcmp(c->name, newcon->name) != 0)
2714 continue;
2715 if (newcon->index >= 0 &&
2716 newcon->index != c->index)
2717 continue;
2718 if (newcon->index < 0)
2719 newcon->index = c->index;
2720
2721 if (_braille_register_console(newcon, c))
2722 return;
2723
2724 if (newcon->setup &&
2725 newcon->setup(newcon, c->options) != 0)
2726 break;
2727 }
2728
2729 newcon->flags |= CON_ENABLED;
2730 if (i == preferred_console) {
2731 newcon->flags |= CON_CONSDEV;
2732 has_preferred = true;
2733 }
2734 break;
2735 }
2736
2737 if (!(newcon->flags & CON_ENABLED))
2738 return;
2739
2740 /*
2741 * If we have a bootconsole, and are switching to a real console,
2742 * don't print everything out again, since when the boot console, and
2743 * the real console are the same physical device, it's annoying to
2744 * see the beginning boot messages twice
2745 */
2746 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2747 newcon->flags &= ~CON_PRINTBUFFER;
2748
2749 /*
2750 * Put this console in the list - keep the
2751 * preferred driver at the head of the list.
2752 */
2753 console_lock();
2754 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2755 newcon->next = console_drivers;
2756 console_drivers = newcon;
2757 if (newcon->next)
2758 newcon->next->flags &= ~CON_CONSDEV;
2759 } else {
2760 newcon->next = console_drivers->next;
2761 console_drivers->next = newcon;
2762 }
2763
2764 if (newcon->flags & CON_EXTENDED)
2765 nr_ext_console_drivers++;
2766
2767 if (newcon->flags & CON_PRINTBUFFER) {
2768 /*
2769 * console_unlock(); will print out the buffered messages
2770 * for us.
2771 */
2772 logbuf_lock_irqsave(flags);
2773 console_seq = syslog_seq;
2774 console_idx = syslog_idx;
2775 /*
2776 * We're about to replay the log buffer. Only do this to the
2777 * just-registered console to avoid excessive message spam to
2778 * the already-registered consoles.
2779 *
2780 * Set exclusive_console with disabled interrupts to reduce
2781 * race window with eventual console_flush_on_panic() that
2782 * ignores console_lock.
2783 */
2784 exclusive_console = newcon;
2785 exclusive_console_stop_seq = console_seq;
2786 logbuf_unlock_irqrestore(flags);
2787 }
2788 console_unlock();
2789 console_sysfs_notify();
2790
2791 /*
2792 * By unregistering the bootconsoles after we enable the real console
2793 * we get the "console xxx enabled" message on all the consoles -
2794 * boot consoles, real consoles, etc - this is to ensure that end
2795 * users know there might be something in the kernel's log buffer that
2796 * went to the bootconsole (that they do not see on the real console)
2797 */
2798 pr_info("%sconsole [%s%d] enabled\n",
2799 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2800 newcon->name, newcon->index);
2801 if (bcon &&
2802 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2803 !keep_bootcon) {
2804 /* We need to iterate through all boot consoles, to make
2805 * sure we print everything out, before we unregister them.
2806 */
2807 for_each_console(bcon)
2808 if (bcon->flags & CON_BOOT)
2809 unregister_console(bcon);
2810 }
2811}
2812EXPORT_SYMBOL(register_console);
2813
2814int unregister_console(struct console *console)
2815{
2816 struct console *a, *b;
2817 int res;
2818
2819 pr_info("%sconsole [%s%d] disabled\n",
2820 (console->flags & CON_BOOT) ? "boot" : "" ,
2821 console->name, console->index);
2822
2823 res = _braille_unregister_console(console);
2824 if (res)
2825 return res;
2826
2827 res = 1;
2828 console_lock();
2829 if (console_drivers == console) {
2830 console_drivers=console->next;
2831 res = 0;
2832 } else if (console_drivers) {
2833 for (a=console_drivers->next, b=console_drivers ;
2834 a; b=a, a=b->next) {
2835 if (a == console) {
2836 b->next = a->next;
2837 res = 0;
2838 break;
2839 }
2840 }
2841 }
2842
2843 if (!res && (console->flags & CON_EXTENDED))
2844 nr_ext_console_drivers--;
2845
2846 /*
2847 * If this isn't the last console and it has CON_CONSDEV set, we
2848 * need to set it on the next preferred console.
2849 */
2850 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2851 console_drivers->flags |= CON_CONSDEV;
2852
2853 console->flags &= ~CON_ENABLED;
2854 console_unlock();
2855 console_sysfs_notify();
2856 return res;
2857}
2858EXPORT_SYMBOL(unregister_console);
2859
2860/*
2861 * Initialize the console device. This is called *early*, so
2862 * we can't necessarily depend on lots of kernel help here.
2863 * Just do some early initializations, and do the complex setup
2864 * later.
2865 */
2866void __init console_init(void)
2867{
2868 int ret;
2869 initcall_t call;
2870 initcall_entry_t *ce;
2871
2872 /* Setup the default TTY line discipline. */
2873 n_tty_init();
2874
2875 /*
2876 * set up the console device so that later boot sequences can
2877 * inform about problems etc..
2878 */
2879 ce = __con_initcall_start;
2880 trace_initcall_level("console");
2881 while (ce < __con_initcall_end) {
2882 call = initcall_from_entry(ce);
2883 trace_initcall_start(call);
2884 ret = call();
2885 trace_initcall_finish(call, ret);
2886 ce++;
2887 }
2888}
2889
2890/*
2891 * Some boot consoles access data that is in the init section and which will
2892 * be discarded after the initcalls have been run. To make sure that no code
2893 * will access this data, unregister the boot consoles in a late initcall.
2894 *
2895 * If for some reason, such as deferred probe or the driver being a loadable
2896 * module, the real console hasn't registered yet at this point, there will
2897 * be a brief interval in which no messages are logged to the console, which
2898 * makes it difficult to diagnose problems that occur during this time.
2899 *
2900 * To mitigate this problem somewhat, only unregister consoles whose memory
2901 * intersects with the init section. Note that all other boot consoles will
2902 * get unregistred when the real preferred console is registered.
2903 */
2904static int __init printk_late_init(void)
2905{
2906 struct console *con;
2907 int ret;
2908
2909 for_each_console(con) {
2910 if (!(con->flags & CON_BOOT))
2911 continue;
2912
2913 /* Check addresses that might be used for enabled consoles. */
2914 if (init_section_intersects(con, sizeof(*con)) ||
2915 init_section_contains(con->write, 0) ||
2916 init_section_contains(con->read, 0) ||
2917 init_section_contains(con->device, 0) ||
2918 init_section_contains(con->unblank, 0) ||
2919 init_section_contains(con->data, 0)) {
2920 /*
2921 * Please, consider moving the reported consoles out
2922 * of the init section.
2923 */
2924 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2925 con->name, con->index);
2926 unregister_console(con);
2927 }
2928 }
2929 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2930 console_cpu_notify);
2931 WARN_ON(ret < 0);
2932 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2933 console_cpu_notify, NULL);
2934 WARN_ON(ret < 0);
2935 return 0;
2936}
2937late_initcall(printk_late_init);
2938
2939#if defined CONFIG_PRINTK
2940/*
2941 * Delayed printk version, for scheduler-internal messages:
2942 */
2943#define PRINTK_PENDING_WAKEUP 0x01
2944#define PRINTK_PENDING_OUTPUT 0x02
2945
2946static DEFINE_PER_CPU(int, printk_pending);
2947
2948static void wake_up_klogd_work_func(struct irq_work *irq_work)
2949{
2950 int pending = __this_cpu_xchg(printk_pending, 0);
2951
2952 if (pending & PRINTK_PENDING_OUTPUT) {
2953 /* If trylock fails, someone else is doing the printing */
2954 if (console_trylock())
2955 console_unlock();
2956 }
2957
2958 if (pending & PRINTK_PENDING_WAKEUP)
2959 wake_up_interruptible(&log_wait);
2960}
2961
2962static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2963 .func = wake_up_klogd_work_func,
2964 .flags = IRQ_WORK_LAZY,
2965};
2966
2967void wake_up_klogd(void)
2968{
2969 preempt_disable();
2970 if (waitqueue_active(&log_wait)) {
2971 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2972 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2973 }
2974 preempt_enable();
2975}
2976
2977void defer_console_output(void)
2978{
2979 preempt_disable();
2980 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2981 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2982 preempt_enable();
2983}
2984
2985int vprintk_deferred(const char *fmt, va_list args)
2986{
2987 int r;
2988
2989 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2990 defer_console_output();
2991
2992 return r;
2993}
2994
2995int printk_deferred(const char *fmt, ...)
2996{
2997 va_list args;
2998 int r;
2999
3000 va_start(args, fmt);
3001 r = vprintk_deferred(fmt, args);
3002 va_end(args);
3003
3004 return r;
3005}
3006
3007/*
3008 * printk rate limiting, lifted from the networking subsystem.
3009 *
3010 * This enforces a rate limit: not more than 10 kernel messages
3011 * every 5s to make a denial-of-service attack impossible.
3012 */
3013DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3014
3015int __printk_ratelimit(const char *func)
3016{
3017 return ___ratelimit(&printk_ratelimit_state, func);
3018}
3019EXPORT_SYMBOL(__printk_ratelimit);
3020
3021/**
3022 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3023 * @caller_jiffies: pointer to caller's state
3024 * @interval_msecs: minimum interval between prints
3025 *
3026 * printk_timed_ratelimit() returns true if more than @interval_msecs
3027 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3028 * returned true.
3029 */
3030bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3031 unsigned int interval_msecs)
3032{
3033 unsigned long elapsed = jiffies - *caller_jiffies;
3034
3035 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3036 return false;
3037
3038 *caller_jiffies = jiffies;
3039 return true;
3040}
3041EXPORT_SYMBOL(printk_timed_ratelimit);
3042
3043static DEFINE_SPINLOCK(dump_list_lock);
3044static LIST_HEAD(dump_list);
3045
3046/**
3047 * kmsg_dump_register - register a kernel log dumper.
3048 * @dumper: pointer to the kmsg_dumper structure
3049 *
3050 * Adds a kernel log dumper to the system. The dump callback in the
3051 * structure will be called when the kernel oopses or panics and must be
3052 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3053 */
3054int kmsg_dump_register(struct kmsg_dumper *dumper)
3055{
3056 unsigned long flags;
3057 int err = -EBUSY;
3058
3059 /* The dump callback needs to be set */
3060 if (!dumper->dump)
3061 return -EINVAL;
3062
3063 spin_lock_irqsave(&dump_list_lock, flags);
3064 /* Don't allow registering multiple times */
3065 if (!dumper->registered) {
3066 dumper->registered = 1;
3067 list_add_tail_rcu(&dumper->list, &dump_list);
3068 err = 0;
3069 }
3070 spin_unlock_irqrestore(&dump_list_lock, flags);
3071
3072 return err;
3073}
3074EXPORT_SYMBOL_GPL(kmsg_dump_register);
3075
3076/**
3077 * kmsg_dump_unregister - unregister a kmsg dumper.
3078 * @dumper: pointer to the kmsg_dumper structure
3079 *
3080 * Removes a dump device from the system. Returns zero on success and
3081 * %-EINVAL otherwise.
3082 */
3083int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3084{
3085 unsigned long flags;
3086 int err = -EINVAL;
3087
3088 spin_lock_irqsave(&dump_list_lock, flags);
3089 if (dumper->registered) {
3090 dumper->registered = 0;
3091 list_del_rcu(&dumper->list);
3092 err = 0;
3093 }
3094 spin_unlock_irqrestore(&dump_list_lock, flags);
3095 synchronize_rcu();
3096
3097 return err;
3098}
3099EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3100
3101static bool always_kmsg_dump;
3102module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3103
3104/**
3105 * kmsg_dump - dump kernel log to kernel message dumpers.
3106 * @reason: the reason (oops, panic etc) for dumping
3107 *
3108 * Call each of the registered dumper's dump() callback, which can
3109 * retrieve the kmsg records with kmsg_dump_get_line() or
3110 * kmsg_dump_get_buffer().
3111 */
3112void kmsg_dump(enum kmsg_dump_reason reason)
3113{
3114 struct kmsg_dumper *dumper;
3115 unsigned long flags;
3116
3117 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3118 return;
3119
3120 rcu_read_lock();
3121 list_for_each_entry_rcu(dumper, &dump_list, list) {
3122 if (dumper->max_reason && reason > dumper->max_reason)
3123 continue;
3124
3125 /* initialize iterator with data about the stored records */
3126 dumper->active = true;
3127
3128 logbuf_lock_irqsave(flags);
3129 dumper->cur_seq = clear_seq;
3130 dumper->cur_idx = clear_idx;
3131 dumper->next_seq = log_next_seq;
3132 dumper->next_idx = log_next_idx;
3133 logbuf_unlock_irqrestore(flags);
3134
3135 /* invoke dumper which will iterate over records */
3136 dumper->dump(dumper, reason);
3137
3138 /* reset iterator */
3139 dumper->active = false;
3140 }
3141 rcu_read_unlock();
3142}
3143
3144/**
3145 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3146 * @dumper: registered kmsg dumper
3147 * @syslog: include the "<4>" prefixes
3148 * @line: buffer to copy the line to
3149 * @size: maximum size of the buffer
3150 * @len: length of line placed into buffer
3151 *
3152 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3153 * record, and copy one record into the provided buffer.
3154 *
3155 * Consecutive calls will return the next available record moving
3156 * towards the end of the buffer with the youngest messages.
3157 *
3158 * A return value of FALSE indicates that there are no more records to
3159 * read.
3160 *
3161 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3162 */
3163bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3164 char *line, size_t size, size_t *len)
3165{
3166 struct printk_log *msg;
3167 size_t l = 0;
3168 bool ret = false;
3169
3170 if (!dumper->active)
3171 goto out;
3172
3173 if (dumper->cur_seq < log_first_seq) {
3174 /* messages are gone, move to first available one */
3175 dumper->cur_seq = log_first_seq;
3176 dumper->cur_idx = log_first_idx;
3177 }
3178
3179 /* last entry */
3180 if (dumper->cur_seq >= log_next_seq)
3181 goto out;
3182
3183 msg = log_from_idx(dumper->cur_idx);
3184 l = msg_print_text(msg, syslog, printk_time, line, size);
3185
3186 dumper->cur_idx = log_next(dumper->cur_idx);
3187 dumper->cur_seq++;
3188 ret = true;
3189out:
3190 if (len)
3191 *len = l;
3192 return ret;
3193}
3194
3195/**
3196 * kmsg_dump_get_line - retrieve one kmsg log line
3197 * @dumper: registered kmsg dumper
3198 * @syslog: include the "<4>" prefixes
3199 * @line: buffer to copy the line to
3200 * @size: maximum size of the buffer
3201 * @len: length of line placed into buffer
3202 *
3203 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3204 * record, and copy one record into the provided buffer.
3205 *
3206 * Consecutive calls will return the next available record moving
3207 * towards the end of the buffer with the youngest messages.
3208 *
3209 * A return value of FALSE indicates that there are no more records to
3210 * read.
3211 */
3212bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3213 char *line, size_t size, size_t *len)
3214{
3215 unsigned long flags;
3216 bool ret;
3217
3218 logbuf_lock_irqsave(flags);
3219 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3220 logbuf_unlock_irqrestore(flags);
3221
3222 return ret;
3223}
3224EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3225
3226/**
3227 * kmsg_dump_get_buffer - copy kmsg log lines
3228 * @dumper: registered kmsg dumper
3229 * @syslog: include the "<4>" prefixes
3230 * @buf: buffer to copy the line to
3231 * @size: maximum size of the buffer
3232 * @len: length of line placed into buffer
3233 *
3234 * Start at the end of the kmsg buffer and fill the provided buffer
3235 * with as many of the the *youngest* kmsg records that fit into it.
3236 * If the buffer is large enough, all available kmsg records will be
3237 * copied with a single call.
3238 *
3239 * Consecutive calls will fill the buffer with the next block of
3240 * available older records, not including the earlier retrieved ones.
3241 *
3242 * A return value of FALSE indicates that there are no more records to
3243 * read.
3244 */
3245bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3246 char *buf, size_t size, size_t *len)
3247{
3248 unsigned long flags;
3249 u64 seq;
3250 u32 idx;
3251 u64 next_seq;
3252 u32 next_idx;
3253 size_t l = 0;
3254 bool ret = false;
3255 bool time = printk_time;
3256
3257 if (!dumper->active)
3258 goto out;
3259
3260 logbuf_lock_irqsave(flags);
3261 if (dumper->cur_seq < log_first_seq) {
3262 /* messages are gone, move to first available one */
3263 dumper->cur_seq = log_first_seq;
3264 dumper->cur_idx = log_first_idx;
3265 }
3266
3267 /* last entry */
3268 if (dumper->cur_seq >= dumper->next_seq) {
3269 logbuf_unlock_irqrestore(flags);
3270 goto out;
3271 }
3272
3273 /* calculate length of entire buffer */
3274 seq = dumper->cur_seq;
3275 idx = dumper->cur_idx;
3276 while (seq < dumper->next_seq) {
3277 struct printk_log *msg = log_from_idx(idx);
3278
3279 l += msg_print_text(msg, true, time, NULL, 0);
3280 idx = log_next(idx);
3281 seq++;
3282 }
3283
3284 /* move first record forward until length fits into the buffer */
3285 seq = dumper->cur_seq;
3286 idx = dumper->cur_idx;
3287 while (l >= size && seq < dumper->next_seq) {
3288 struct printk_log *msg = log_from_idx(idx);
3289
3290 l -= msg_print_text(msg, true, time, NULL, 0);
3291 idx = log_next(idx);
3292 seq++;
3293 }
3294
3295 /* last message in next interation */
3296 next_seq = seq;
3297 next_idx = idx;
3298
3299 l = 0;
3300 while (seq < dumper->next_seq) {
3301 struct printk_log *msg = log_from_idx(idx);
3302
3303 l += msg_print_text(msg, syslog, time, buf + l, size - l);
3304 idx = log_next(idx);
3305 seq++;
3306 }
3307
3308 dumper->next_seq = next_seq;
3309 dumper->next_idx = next_idx;
3310 ret = true;
3311 logbuf_unlock_irqrestore(flags);
3312out:
3313 if (len)
3314 *len = l;
3315 return ret;
3316}
3317EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3318
3319/**
3320 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3321 * @dumper: registered kmsg dumper
3322 *
3323 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3324 * kmsg_dump_get_buffer() can be called again and used multiple
3325 * times within the same dumper.dump() callback.
3326 *
3327 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3328 */
3329void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3330{
3331 dumper->cur_seq = clear_seq;
3332 dumper->cur_idx = clear_idx;
3333 dumper->next_seq = log_next_seq;
3334 dumper->next_idx = log_next_idx;
3335}
3336
3337/**
3338 * kmsg_dump_rewind - reset the interator
3339 * @dumper: registered kmsg dumper
3340 *
3341 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3342 * kmsg_dump_get_buffer() can be called again and used multiple
3343 * times within the same dumper.dump() callback.
3344 */
3345void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3346{
3347 unsigned long flags;
3348
3349 logbuf_lock_irqsave(flags);
3350 kmsg_dump_rewind_nolock(dumper);
3351 logbuf_unlock_irqrestore(flags);
3352}
3353EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3354
3355#endif
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/printk.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * Modified to make sys_syslog() more flexible: added commands to
8 * return the last 4k of kernel messages, regardless of whether
9 * they've been read or not. Added option to suppress kernel printk's
10 * to the console. Added hook for sending the console messages
11 * elsewhere, in preparation for a serial line console (someday).
12 * Ted Ts'o, 2/11/93.
13 * Modified for sysctl support, 1/8/97, Chris Horn.
14 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
15 * manfred@colorfullife.com
16 * Rewrote bits to get rid of console_lock
17 * 01Mar01 Andrew Morton
18 */
19
20#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21
22#include <linux/kernel.h>
23#include <linux/mm.h>
24#include <linux/tty.h>
25#include <linux/tty_driver.h>
26#include <linux/console.h>
27#include <linux/init.h>
28#include <linux/jiffies.h>
29#include <linux/nmi.h>
30#include <linux/module.h>
31#include <linux/moduleparam.h>
32#include <linux/delay.h>
33#include <linux/smp.h>
34#include <linux/security.h>
35#include <linux/memblock.h>
36#include <linux/syscalls.h>
37#include <linux/crash_core.h>
38#include <linux/ratelimit.h>
39#include <linux/kmsg_dump.h>
40#include <linux/syslog.h>
41#include <linux/cpu.h>
42#include <linux/rculist.h>
43#include <linux/poll.h>
44#include <linux/irq_work.h>
45#include <linux/ctype.h>
46#include <linux/uio.h>
47#include <linux/sched/clock.h>
48#include <linux/sched/debug.h>
49#include <linux/sched/task_stack.h>
50
51#include <linux/uaccess.h>
52#include <asm/sections.h>
53
54#include <trace/events/initcall.h>
55#define CREATE_TRACE_POINTS
56#include <trace/events/printk.h>
57
58#include "printk_ringbuffer.h"
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_mutex protects console_list updates and console->flags updates.
83 * The flags are synchronized only for consoles that are registered, i.e.
84 * accessible via the console list.
85 */
86static DEFINE_MUTEX(console_mutex);
87
88/*
89 * console_sem protects updates to console->seq and console_suspended,
90 * and also provides serialization for console printing.
91 */
92static DEFINE_SEMAPHORE(console_sem);
93HLIST_HEAD(console_list);
94EXPORT_SYMBOL_GPL(console_list);
95DEFINE_STATIC_SRCU(console_srcu);
96
97/*
98 * System may need to suppress printk message under certain
99 * circumstances, like after kernel panic happens.
100 */
101int __read_mostly suppress_printk;
102
103/*
104 * During panic, heavy printk by other CPUs can delay the
105 * panic and risk deadlock on console resources.
106 */
107static int __read_mostly suppress_panic_printk;
108
109#ifdef CONFIG_LOCKDEP
110static struct lockdep_map console_lock_dep_map = {
111 .name = "console_lock"
112};
113
114void lockdep_assert_console_list_lock_held(void)
115{
116 lockdep_assert_held(&console_mutex);
117}
118EXPORT_SYMBOL(lockdep_assert_console_list_lock_held);
119#endif
120
121#ifdef CONFIG_DEBUG_LOCK_ALLOC
122bool console_srcu_read_lock_is_held(void)
123{
124 return srcu_read_lock_held(&console_srcu);
125}
126EXPORT_SYMBOL(console_srcu_read_lock_is_held);
127#endif
128
129enum devkmsg_log_bits {
130 __DEVKMSG_LOG_BIT_ON = 0,
131 __DEVKMSG_LOG_BIT_OFF,
132 __DEVKMSG_LOG_BIT_LOCK,
133};
134
135enum devkmsg_log_masks {
136 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
137 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
138 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
139};
140
141/* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
142#define DEVKMSG_LOG_MASK_DEFAULT 0
143
144static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
145
146static int __control_devkmsg(char *str)
147{
148 size_t len;
149
150 if (!str)
151 return -EINVAL;
152
153 len = str_has_prefix(str, "on");
154 if (len) {
155 devkmsg_log = DEVKMSG_LOG_MASK_ON;
156 return len;
157 }
158
159 len = str_has_prefix(str, "off");
160 if (len) {
161 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
162 return len;
163 }
164
165 len = str_has_prefix(str, "ratelimit");
166 if (len) {
167 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
168 return len;
169 }
170
171 return -EINVAL;
172}
173
174static int __init control_devkmsg(char *str)
175{
176 if (__control_devkmsg(str) < 0) {
177 pr_warn("printk.devkmsg: bad option string '%s'\n", str);
178 return 1;
179 }
180
181 /*
182 * Set sysctl string accordingly:
183 */
184 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
185 strcpy(devkmsg_log_str, "on");
186 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
187 strcpy(devkmsg_log_str, "off");
188 /* else "ratelimit" which is set by default. */
189
190 /*
191 * Sysctl cannot change it anymore. The kernel command line setting of
192 * this parameter is to force the setting to be permanent throughout the
193 * runtime of the system. This is a precation measure against userspace
194 * trying to be a smarta** and attempting to change it up on us.
195 */
196 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
197
198 return 1;
199}
200__setup("printk.devkmsg=", control_devkmsg);
201
202char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
203#if defined(CONFIG_PRINTK) && defined(CONFIG_SYSCTL)
204int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
205 void *buffer, size_t *lenp, loff_t *ppos)
206{
207 char old_str[DEVKMSG_STR_MAX_SIZE];
208 unsigned int old;
209 int err;
210
211 if (write) {
212 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
213 return -EINVAL;
214
215 old = devkmsg_log;
216 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
217 }
218
219 err = proc_dostring(table, write, buffer, lenp, ppos);
220 if (err)
221 return err;
222
223 if (write) {
224 err = __control_devkmsg(devkmsg_log_str);
225
226 /*
227 * Do not accept an unknown string OR a known string with
228 * trailing crap...
229 */
230 if (err < 0 || (err + 1 != *lenp)) {
231
232 /* ... and restore old setting. */
233 devkmsg_log = old;
234 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
235
236 return -EINVAL;
237 }
238 }
239
240 return 0;
241}
242#endif /* CONFIG_PRINTK && CONFIG_SYSCTL */
243
244/**
245 * console_list_lock - Lock the console list
246 *
247 * For console list or console->flags updates
248 */
249void console_list_lock(void)
250{
251 /*
252 * In unregister_console() and console_force_preferred_locked(),
253 * synchronize_srcu() is called with the console_list_lock held.
254 * Therefore it is not allowed that the console_list_lock is taken
255 * with the srcu_lock held.
256 *
257 * Detecting if this context is really in the read-side critical
258 * section is only possible if the appropriate debug options are
259 * enabled.
260 */
261 WARN_ON_ONCE(debug_lockdep_rcu_enabled() &&
262 srcu_read_lock_held(&console_srcu));
263
264 mutex_lock(&console_mutex);
265}
266EXPORT_SYMBOL(console_list_lock);
267
268/**
269 * console_list_unlock - Unlock the console list
270 *
271 * Counterpart to console_list_lock()
272 */
273void console_list_unlock(void)
274{
275 mutex_unlock(&console_mutex);
276}
277EXPORT_SYMBOL(console_list_unlock);
278
279/**
280 * console_srcu_read_lock - Register a new reader for the
281 * SRCU-protected console list
282 *
283 * Use for_each_console_srcu() to iterate the console list
284 *
285 * Context: Any context.
286 * Return: A cookie to pass to console_srcu_read_unlock().
287 */
288int console_srcu_read_lock(void)
289{
290 return srcu_read_lock_nmisafe(&console_srcu);
291}
292EXPORT_SYMBOL(console_srcu_read_lock);
293
294/**
295 * console_srcu_read_unlock - Unregister an old reader from
296 * the SRCU-protected console list
297 * @cookie: cookie returned from console_srcu_read_lock()
298 *
299 * Counterpart to console_srcu_read_lock()
300 */
301void console_srcu_read_unlock(int cookie)
302{
303 srcu_read_unlock_nmisafe(&console_srcu, cookie);
304}
305EXPORT_SYMBOL(console_srcu_read_unlock);
306
307/*
308 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
309 * macros instead of functions so that _RET_IP_ contains useful information.
310 */
311#define down_console_sem() do { \
312 down(&console_sem);\
313 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
314} while (0)
315
316static int __down_trylock_console_sem(unsigned long ip)
317{
318 int lock_failed;
319 unsigned long flags;
320
321 /*
322 * Here and in __up_console_sem() we need to be in safe mode,
323 * because spindump/WARN/etc from under console ->lock will
324 * deadlock in printk()->down_trylock_console_sem() otherwise.
325 */
326 printk_safe_enter_irqsave(flags);
327 lock_failed = down_trylock(&console_sem);
328 printk_safe_exit_irqrestore(flags);
329
330 if (lock_failed)
331 return 1;
332 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
333 return 0;
334}
335#define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
336
337static void __up_console_sem(unsigned long ip)
338{
339 unsigned long flags;
340
341 mutex_release(&console_lock_dep_map, ip);
342
343 printk_safe_enter_irqsave(flags);
344 up(&console_sem);
345 printk_safe_exit_irqrestore(flags);
346}
347#define up_console_sem() __up_console_sem(_RET_IP_)
348
349static bool panic_in_progress(void)
350{
351 return unlikely(atomic_read(&panic_cpu) != PANIC_CPU_INVALID);
352}
353
354/*
355 * This is used for debugging the mess that is the VT code by
356 * keeping track if we have the console semaphore held. It's
357 * definitely not the perfect debug tool (we don't know if _WE_
358 * hold it and are racing, but it helps tracking those weird code
359 * paths in the console code where we end up in places I want
360 * locked without the console semaphore held).
361 */
362static int console_locked, console_suspended;
363
364/*
365 * Array of consoles built from command line options (console=)
366 */
367
368#define MAX_CMDLINECONSOLES 8
369
370static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
371
372static int preferred_console = -1;
373int console_set_on_cmdline;
374EXPORT_SYMBOL(console_set_on_cmdline);
375
376/* Flag: console code may call schedule() */
377static int console_may_schedule;
378
379enum con_msg_format_flags {
380 MSG_FORMAT_DEFAULT = 0,
381 MSG_FORMAT_SYSLOG = (1 << 0),
382};
383
384static int console_msg_format = MSG_FORMAT_DEFAULT;
385
386/*
387 * The printk log buffer consists of a sequenced collection of records, each
388 * containing variable length message text. Every record also contains its
389 * own meta-data (@info).
390 *
391 * Every record meta-data carries the timestamp in microseconds, as well as
392 * the standard userspace syslog level and syslog facility. The usual kernel
393 * messages use LOG_KERN; userspace-injected messages always carry a matching
394 * syslog facility, by default LOG_USER. The origin of every message can be
395 * reliably determined that way.
396 *
397 * The human readable log message of a record is available in @text, the
398 * length of the message text in @text_len. The stored message is not
399 * terminated.
400 *
401 * Optionally, a record can carry a dictionary of properties (key/value
402 * pairs), to provide userspace with a machine-readable message context.
403 *
404 * Examples for well-defined, commonly used property names are:
405 * DEVICE=b12:8 device identifier
406 * b12:8 block dev_t
407 * c127:3 char dev_t
408 * n8 netdev ifindex
409 * +sound:card0 subsystem:devname
410 * SUBSYSTEM=pci driver-core subsystem name
411 *
412 * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
413 * and values are terminated by a '\0' character.
414 *
415 * Example of record values:
416 * record.text_buf = "it's a line" (unterminated)
417 * record.info.seq = 56
418 * record.info.ts_nsec = 36863
419 * record.info.text_len = 11
420 * record.info.facility = 0 (LOG_KERN)
421 * record.info.flags = 0
422 * record.info.level = 3 (LOG_ERR)
423 * record.info.caller_id = 299 (task 299)
424 * record.info.dev_info.subsystem = "pci" (terminated)
425 * record.info.dev_info.device = "+pci:0000:00:01.0" (terminated)
426 *
427 * The 'struct printk_info' buffer must never be directly exported to
428 * userspace, it is a kernel-private implementation detail that might
429 * need to be changed in the future, when the requirements change.
430 *
431 * /dev/kmsg exports the structured data in the following line format:
432 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
433 *
434 * Users of the export format should ignore possible additional values
435 * separated by ',', and find the message after the ';' character.
436 *
437 * The optional key/value pairs are attached as continuation lines starting
438 * with a space character and terminated by a newline. All possible
439 * non-prinatable characters are escaped in the "\xff" notation.
440 */
441
442/* syslog_lock protects syslog_* variables and write access to clear_seq. */
443static DEFINE_MUTEX(syslog_lock);
444
445#ifdef CONFIG_PRINTK
446DECLARE_WAIT_QUEUE_HEAD(log_wait);
447/* All 3 protected by @syslog_lock. */
448/* the next printk record to read by syslog(READ) or /proc/kmsg */
449static u64 syslog_seq;
450static size_t syslog_partial;
451static bool syslog_time;
452
453struct latched_seq {
454 seqcount_latch_t latch;
455 u64 val[2];
456};
457
458/*
459 * The next printk record to read after the last 'clear' command. There are
460 * two copies (updated with seqcount_latch) so that reads can locklessly
461 * access a valid value. Writers are synchronized by @syslog_lock.
462 */
463static struct latched_seq clear_seq = {
464 .latch = SEQCNT_LATCH_ZERO(clear_seq.latch),
465 .val[0] = 0,
466 .val[1] = 0,
467};
468
469#ifdef CONFIG_PRINTK_CALLER
470#define PREFIX_MAX 48
471#else
472#define PREFIX_MAX 32
473#endif
474
475/* the maximum size of a formatted record (i.e. with prefix added per line) */
476#define CONSOLE_LOG_MAX 1024
477
478/* the maximum size for a dropped text message */
479#define DROPPED_TEXT_MAX 64
480
481/* the maximum size allowed to be reserved for a record */
482#define LOG_LINE_MAX (CONSOLE_LOG_MAX - PREFIX_MAX)
483
484#define LOG_LEVEL(v) ((v) & 0x07)
485#define LOG_FACILITY(v) ((v) >> 3 & 0xff)
486
487/* record buffer */
488#define LOG_ALIGN __alignof__(unsigned long)
489#define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
490#define LOG_BUF_LEN_MAX (u32)(1 << 31)
491static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
492static char *log_buf = __log_buf;
493static u32 log_buf_len = __LOG_BUF_LEN;
494
495/*
496 * Define the average message size. This only affects the number of
497 * descriptors that will be available. Underestimating is better than
498 * overestimating (too many available descriptors is better than not enough).
499 */
500#define PRB_AVGBITS 5 /* 32 character average length */
501
502#if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
503#error CONFIG_LOG_BUF_SHIFT value too small.
504#endif
505_DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
506 PRB_AVGBITS, &__log_buf[0]);
507
508static struct printk_ringbuffer printk_rb_dynamic;
509
510static struct printk_ringbuffer *prb = &printk_rb_static;
511
512/*
513 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
514 * per_cpu_areas are initialised. This variable is set to true when
515 * it's safe to access per-CPU data.
516 */
517static bool __printk_percpu_data_ready __ro_after_init;
518
519bool printk_percpu_data_ready(void)
520{
521 return __printk_percpu_data_ready;
522}
523
524/* Must be called under syslog_lock. */
525static void latched_seq_write(struct latched_seq *ls, u64 val)
526{
527 raw_write_seqcount_latch(&ls->latch);
528 ls->val[0] = val;
529 raw_write_seqcount_latch(&ls->latch);
530 ls->val[1] = val;
531}
532
533/* Can be called from any context. */
534static u64 latched_seq_read_nolock(struct latched_seq *ls)
535{
536 unsigned int seq;
537 unsigned int idx;
538 u64 val;
539
540 do {
541 seq = raw_read_seqcount_latch(&ls->latch);
542 idx = seq & 0x1;
543 val = ls->val[idx];
544 } while (read_seqcount_latch_retry(&ls->latch, seq));
545
546 return val;
547}
548
549/* Return log buffer address */
550char *log_buf_addr_get(void)
551{
552 return log_buf;
553}
554
555/* Return log buffer size */
556u32 log_buf_len_get(void)
557{
558 return log_buf_len;
559}
560
561/*
562 * Define how much of the log buffer we could take at maximum. The value
563 * must be greater than two. Note that only half of the buffer is available
564 * when the index points to the middle.
565 */
566#define MAX_LOG_TAKE_PART 4
567static const char trunc_msg[] = "<truncated>";
568
569static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
570{
571 /*
572 * The message should not take the whole buffer. Otherwise, it might
573 * get removed too soon.
574 */
575 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
576
577 if (*text_len > max_text_len)
578 *text_len = max_text_len;
579
580 /* enable the warning message (if there is room) */
581 *trunc_msg_len = strlen(trunc_msg);
582 if (*text_len >= *trunc_msg_len)
583 *text_len -= *trunc_msg_len;
584 else
585 *trunc_msg_len = 0;
586}
587
588int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
589
590static int syslog_action_restricted(int type)
591{
592 if (dmesg_restrict)
593 return 1;
594 /*
595 * Unless restricted, we allow "read all" and "get buffer size"
596 * for everybody.
597 */
598 return type != SYSLOG_ACTION_READ_ALL &&
599 type != SYSLOG_ACTION_SIZE_BUFFER;
600}
601
602static int check_syslog_permissions(int type, int source)
603{
604 /*
605 * If this is from /proc/kmsg and we've already opened it, then we've
606 * already done the capabilities checks at open time.
607 */
608 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
609 goto ok;
610
611 if (syslog_action_restricted(type)) {
612 if (capable(CAP_SYSLOG))
613 goto ok;
614 /*
615 * For historical reasons, accept CAP_SYS_ADMIN too, with
616 * a warning.
617 */
618 if (capable(CAP_SYS_ADMIN)) {
619 pr_warn_once("%s (%d): Attempt to access syslog with "
620 "CAP_SYS_ADMIN but no CAP_SYSLOG "
621 "(deprecated).\n",
622 current->comm, task_pid_nr(current));
623 goto ok;
624 }
625 return -EPERM;
626 }
627ok:
628 return security_syslog(type);
629}
630
631static void append_char(char **pp, char *e, char c)
632{
633 if (*pp < e)
634 *(*pp)++ = c;
635}
636
637static ssize_t info_print_ext_header(char *buf, size_t size,
638 struct printk_info *info)
639{
640 u64 ts_usec = info->ts_nsec;
641 char caller[20];
642#ifdef CONFIG_PRINTK_CALLER
643 u32 id = info->caller_id;
644
645 snprintf(caller, sizeof(caller), ",caller=%c%u",
646 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
647#else
648 caller[0] = '\0';
649#endif
650
651 do_div(ts_usec, 1000);
652
653 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
654 (info->facility << 3) | info->level, info->seq,
655 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
656}
657
658static ssize_t msg_add_ext_text(char *buf, size_t size,
659 const char *text, size_t text_len,
660 unsigned char endc)
661{
662 char *p = buf, *e = buf + size;
663 size_t i;
664
665 /* escape non-printable characters */
666 for (i = 0; i < text_len; i++) {
667 unsigned char c = text[i];
668
669 if (c < ' ' || c >= 127 || c == '\\')
670 p += scnprintf(p, e - p, "\\x%02x", c);
671 else
672 append_char(&p, e, c);
673 }
674 append_char(&p, e, endc);
675
676 return p - buf;
677}
678
679static ssize_t msg_add_dict_text(char *buf, size_t size,
680 const char *key, const char *val)
681{
682 size_t val_len = strlen(val);
683 ssize_t len;
684
685 if (!val_len)
686 return 0;
687
688 len = msg_add_ext_text(buf, size, "", 0, ' '); /* dict prefix */
689 len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
690 len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
691
692 return len;
693}
694
695static ssize_t msg_print_ext_body(char *buf, size_t size,
696 char *text, size_t text_len,
697 struct dev_printk_info *dev_info)
698{
699 ssize_t len;
700
701 len = msg_add_ext_text(buf, size, text, text_len, '\n');
702
703 if (!dev_info)
704 goto out;
705
706 len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
707 dev_info->subsystem);
708 len += msg_add_dict_text(buf + len, size - len, "DEVICE",
709 dev_info->device);
710out:
711 return len;
712}
713
714/* /dev/kmsg - userspace message inject/listen interface */
715struct devkmsg_user {
716 atomic64_t seq;
717 struct ratelimit_state rs;
718 struct mutex lock;
719 char buf[CONSOLE_EXT_LOG_MAX];
720
721 struct printk_info info;
722 char text_buf[CONSOLE_EXT_LOG_MAX];
723 struct printk_record record;
724};
725
726static __printf(3, 4) __cold
727int devkmsg_emit(int facility, int level, const char *fmt, ...)
728{
729 va_list args;
730 int r;
731
732 va_start(args, fmt);
733 r = vprintk_emit(facility, level, NULL, fmt, args);
734 va_end(args);
735
736 return r;
737}
738
739static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
740{
741 char *buf, *line;
742 int level = default_message_loglevel;
743 int facility = 1; /* LOG_USER */
744 struct file *file = iocb->ki_filp;
745 struct devkmsg_user *user = file->private_data;
746 size_t len = iov_iter_count(from);
747 ssize_t ret = len;
748
749 if (!user || len > LOG_LINE_MAX)
750 return -EINVAL;
751
752 /* Ignore when user logging is disabled. */
753 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
754 return len;
755
756 /* Ratelimit when not explicitly enabled. */
757 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
758 if (!___ratelimit(&user->rs, current->comm))
759 return ret;
760 }
761
762 buf = kmalloc(len+1, GFP_KERNEL);
763 if (buf == NULL)
764 return -ENOMEM;
765
766 buf[len] = '\0';
767 if (!copy_from_iter_full(buf, len, from)) {
768 kfree(buf);
769 return -EFAULT;
770 }
771
772 /*
773 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
774 * the decimal value represents 32bit, the lower 3 bit are the log
775 * level, the rest are the log facility.
776 *
777 * If no prefix or no userspace facility is specified, we
778 * enforce LOG_USER, to be able to reliably distinguish
779 * kernel-generated messages from userspace-injected ones.
780 */
781 line = buf;
782 if (line[0] == '<') {
783 char *endp = NULL;
784 unsigned int u;
785
786 u = simple_strtoul(line + 1, &endp, 10);
787 if (endp && endp[0] == '>') {
788 level = LOG_LEVEL(u);
789 if (LOG_FACILITY(u) != 0)
790 facility = LOG_FACILITY(u);
791 endp++;
792 line = endp;
793 }
794 }
795
796 devkmsg_emit(facility, level, "%s", line);
797 kfree(buf);
798 return ret;
799}
800
801static ssize_t devkmsg_read(struct file *file, char __user *buf,
802 size_t count, loff_t *ppos)
803{
804 struct devkmsg_user *user = file->private_data;
805 struct printk_record *r = &user->record;
806 size_t len;
807 ssize_t ret;
808
809 if (!user)
810 return -EBADF;
811
812 ret = mutex_lock_interruptible(&user->lock);
813 if (ret)
814 return ret;
815
816 if (!prb_read_valid(prb, atomic64_read(&user->seq), r)) {
817 if (file->f_flags & O_NONBLOCK) {
818 ret = -EAGAIN;
819 goto out;
820 }
821
822 /*
823 * Guarantee this task is visible on the waitqueue before
824 * checking the wake condition.
825 *
826 * The full memory barrier within set_current_state() of
827 * prepare_to_wait_event() pairs with the full memory barrier
828 * within wq_has_sleeper().
829 *
830 * This pairs with __wake_up_klogd:A.
831 */
832 ret = wait_event_interruptible(log_wait,
833 prb_read_valid(prb,
834 atomic64_read(&user->seq), r)); /* LMM(devkmsg_read:A) */
835 if (ret)
836 goto out;
837 }
838
839 if (r->info->seq != atomic64_read(&user->seq)) {
840 /* our last seen message is gone, return error and reset */
841 atomic64_set(&user->seq, r->info->seq);
842 ret = -EPIPE;
843 goto out;
844 }
845
846 len = info_print_ext_header(user->buf, sizeof(user->buf), r->info);
847 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
848 &r->text_buf[0], r->info->text_len,
849 &r->info->dev_info);
850
851 atomic64_set(&user->seq, r->info->seq + 1);
852
853 if (len > count) {
854 ret = -EINVAL;
855 goto out;
856 }
857
858 if (copy_to_user(buf, user->buf, len)) {
859 ret = -EFAULT;
860 goto out;
861 }
862 ret = len;
863out:
864 mutex_unlock(&user->lock);
865 return ret;
866}
867
868/*
869 * Be careful when modifying this function!!!
870 *
871 * Only few operations are supported because the device works only with the
872 * entire variable length messages (records). Non-standard values are
873 * returned in the other cases and has been this way for quite some time.
874 * User space applications might depend on this behavior.
875 */
876static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
877{
878 struct devkmsg_user *user = file->private_data;
879 loff_t ret = 0;
880
881 if (!user)
882 return -EBADF;
883 if (offset)
884 return -ESPIPE;
885
886 switch (whence) {
887 case SEEK_SET:
888 /* the first record */
889 atomic64_set(&user->seq, prb_first_valid_seq(prb));
890 break;
891 case SEEK_DATA:
892 /*
893 * The first record after the last SYSLOG_ACTION_CLEAR,
894 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
895 * changes no global state, and does not clear anything.
896 */
897 atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
898 break;
899 case SEEK_END:
900 /* after the last record */
901 atomic64_set(&user->seq, prb_next_seq(prb));
902 break;
903 default:
904 ret = -EINVAL;
905 }
906 return ret;
907}
908
909static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
910{
911 struct devkmsg_user *user = file->private_data;
912 struct printk_info info;
913 __poll_t ret = 0;
914
915 if (!user)
916 return EPOLLERR|EPOLLNVAL;
917
918 poll_wait(file, &log_wait, wait);
919
920 if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
921 /* return error when data has vanished underneath us */
922 if (info.seq != atomic64_read(&user->seq))
923 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
924 else
925 ret = EPOLLIN|EPOLLRDNORM;
926 }
927
928 return ret;
929}
930
931static int devkmsg_open(struct inode *inode, struct file *file)
932{
933 struct devkmsg_user *user;
934 int err;
935
936 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
937 return -EPERM;
938
939 /* write-only does not need any file context */
940 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
941 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
942 SYSLOG_FROM_READER);
943 if (err)
944 return err;
945 }
946
947 user = kvmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
948 if (!user)
949 return -ENOMEM;
950
951 ratelimit_default_init(&user->rs);
952 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
953
954 mutex_init(&user->lock);
955
956 prb_rec_init_rd(&user->record, &user->info,
957 &user->text_buf[0], sizeof(user->text_buf));
958
959 atomic64_set(&user->seq, prb_first_valid_seq(prb));
960
961 file->private_data = user;
962 return 0;
963}
964
965static int devkmsg_release(struct inode *inode, struct file *file)
966{
967 struct devkmsg_user *user = file->private_data;
968
969 if (!user)
970 return 0;
971
972 ratelimit_state_exit(&user->rs);
973
974 mutex_destroy(&user->lock);
975 kvfree(user);
976 return 0;
977}
978
979const struct file_operations kmsg_fops = {
980 .open = devkmsg_open,
981 .read = devkmsg_read,
982 .write_iter = devkmsg_write,
983 .llseek = devkmsg_llseek,
984 .poll = devkmsg_poll,
985 .release = devkmsg_release,
986};
987
988#ifdef CONFIG_CRASH_CORE
989/*
990 * This appends the listed symbols to /proc/vmcore
991 *
992 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
993 * obtain access to symbols that are otherwise very difficult to locate. These
994 * symbols are specifically used so that utilities can access and extract the
995 * dmesg log from a vmcore file after a crash.
996 */
997void log_buf_vmcoreinfo_setup(void)
998{
999 struct dev_printk_info *dev_info = NULL;
1000
1001 VMCOREINFO_SYMBOL(prb);
1002 VMCOREINFO_SYMBOL(printk_rb_static);
1003 VMCOREINFO_SYMBOL(clear_seq);
1004
1005 /*
1006 * Export struct size and field offsets. User space tools can
1007 * parse it and detect any changes to structure down the line.
1008 */
1009
1010 VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
1011 VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
1012 VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
1013 VMCOREINFO_OFFSET(printk_ringbuffer, fail);
1014
1015 VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
1016 VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
1017 VMCOREINFO_OFFSET(prb_desc_ring, descs);
1018 VMCOREINFO_OFFSET(prb_desc_ring, infos);
1019 VMCOREINFO_OFFSET(prb_desc_ring, head_id);
1020 VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
1021
1022 VMCOREINFO_STRUCT_SIZE(prb_desc);
1023 VMCOREINFO_OFFSET(prb_desc, state_var);
1024 VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
1025
1026 VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
1027 VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
1028 VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
1029
1030 VMCOREINFO_STRUCT_SIZE(printk_info);
1031 VMCOREINFO_OFFSET(printk_info, seq);
1032 VMCOREINFO_OFFSET(printk_info, ts_nsec);
1033 VMCOREINFO_OFFSET(printk_info, text_len);
1034 VMCOREINFO_OFFSET(printk_info, caller_id);
1035 VMCOREINFO_OFFSET(printk_info, dev_info);
1036
1037 VMCOREINFO_STRUCT_SIZE(dev_printk_info);
1038 VMCOREINFO_OFFSET(dev_printk_info, subsystem);
1039 VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
1040 VMCOREINFO_OFFSET(dev_printk_info, device);
1041 VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
1042
1043 VMCOREINFO_STRUCT_SIZE(prb_data_ring);
1044 VMCOREINFO_OFFSET(prb_data_ring, size_bits);
1045 VMCOREINFO_OFFSET(prb_data_ring, data);
1046 VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
1047 VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
1048
1049 VMCOREINFO_SIZE(atomic_long_t);
1050 VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
1051
1052 VMCOREINFO_STRUCT_SIZE(latched_seq);
1053 VMCOREINFO_OFFSET(latched_seq, val);
1054}
1055#endif
1056
1057/* requested log_buf_len from kernel cmdline */
1058static unsigned long __initdata new_log_buf_len;
1059
1060/* we practice scaling the ring buffer by powers of 2 */
1061static void __init log_buf_len_update(u64 size)
1062{
1063 if (size > (u64)LOG_BUF_LEN_MAX) {
1064 size = (u64)LOG_BUF_LEN_MAX;
1065 pr_err("log_buf over 2G is not supported.\n");
1066 }
1067
1068 if (size)
1069 size = roundup_pow_of_two(size);
1070 if (size > log_buf_len)
1071 new_log_buf_len = (unsigned long)size;
1072}
1073
1074/* save requested log_buf_len since it's too early to process it */
1075static int __init log_buf_len_setup(char *str)
1076{
1077 u64 size;
1078
1079 if (!str)
1080 return -EINVAL;
1081
1082 size = memparse(str, &str);
1083
1084 log_buf_len_update(size);
1085
1086 return 0;
1087}
1088early_param("log_buf_len", log_buf_len_setup);
1089
1090#ifdef CONFIG_SMP
1091#define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1092
1093static void __init log_buf_add_cpu(void)
1094{
1095 unsigned int cpu_extra;
1096
1097 /*
1098 * archs should set up cpu_possible_bits properly with
1099 * set_cpu_possible() after setup_arch() but just in
1100 * case lets ensure this is valid.
1101 */
1102 if (num_possible_cpus() == 1)
1103 return;
1104
1105 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1106
1107 /* by default this will only continue through for large > 64 CPUs */
1108 if (cpu_extra <= __LOG_BUF_LEN / 2)
1109 return;
1110
1111 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1112 __LOG_CPU_MAX_BUF_LEN);
1113 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1114 cpu_extra);
1115 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1116
1117 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1118}
1119#else /* !CONFIG_SMP */
1120static inline void log_buf_add_cpu(void) {}
1121#endif /* CONFIG_SMP */
1122
1123static void __init set_percpu_data_ready(void)
1124{
1125 __printk_percpu_data_ready = true;
1126}
1127
1128static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1129 struct printk_record *r)
1130{
1131 struct prb_reserved_entry e;
1132 struct printk_record dest_r;
1133
1134 prb_rec_init_wr(&dest_r, r->info->text_len);
1135
1136 if (!prb_reserve(&e, rb, &dest_r))
1137 return 0;
1138
1139 memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1140 dest_r.info->text_len = r->info->text_len;
1141 dest_r.info->facility = r->info->facility;
1142 dest_r.info->level = r->info->level;
1143 dest_r.info->flags = r->info->flags;
1144 dest_r.info->ts_nsec = r->info->ts_nsec;
1145 dest_r.info->caller_id = r->info->caller_id;
1146 memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1147
1148 prb_final_commit(&e);
1149
1150 return prb_record_text_space(&e);
1151}
1152
1153static char setup_text_buf[LOG_LINE_MAX] __initdata;
1154
1155void __init setup_log_buf(int early)
1156{
1157 struct printk_info *new_infos;
1158 unsigned int new_descs_count;
1159 struct prb_desc *new_descs;
1160 struct printk_info info;
1161 struct printk_record r;
1162 unsigned int text_size;
1163 size_t new_descs_size;
1164 size_t new_infos_size;
1165 unsigned long flags;
1166 char *new_log_buf;
1167 unsigned int free;
1168 u64 seq;
1169
1170 /*
1171 * Some archs call setup_log_buf() multiple times - first is very
1172 * early, e.g. from setup_arch(), and second - when percpu_areas
1173 * are initialised.
1174 */
1175 if (!early)
1176 set_percpu_data_ready();
1177
1178 if (log_buf != __log_buf)
1179 return;
1180
1181 if (!early && !new_log_buf_len)
1182 log_buf_add_cpu();
1183
1184 if (!new_log_buf_len)
1185 return;
1186
1187 new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1188 if (new_descs_count == 0) {
1189 pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1190 return;
1191 }
1192
1193 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1194 if (unlikely(!new_log_buf)) {
1195 pr_err("log_buf_len: %lu text bytes not available\n",
1196 new_log_buf_len);
1197 return;
1198 }
1199
1200 new_descs_size = new_descs_count * sizeof(struct prb_desc);
1201 new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1202 if (unlikely(!new_descs)) {
1203 pr_err("log_buf_len: %zu desc bytes not available\n",
1204 new_descs_size);
1205 goto err_free_log_buf;
1206 }
1207
1208 new_infos_size = new_descs_count * sizeof(struct printk_info);
1209 new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1210 if (unlikely(!new_infos)) {
1211 pr_err("log_buf_len: %zu info bytes not available\n",
1212 new_infos_size);
1213 goto err_free_descs;
1214 }
1215
1216 prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1217
1218 prb_init(&printk_rb_dynamic,
1219 new_log_buf, ilog2(new_log_buf_len),
1220 new_descs, ilog2(new_descs_count),
1221 new_infos);
1222
1223 local_irq_save(flags);
1224
1225 log_buf_len = new_log_buf_len;
1226 log_buf = new_log_buf;
1227 new_log_buf_len = 0;
1228
1229 free = __LOG_BUF_LEN;
1230 prb_for_each_record(0, &printk_rb_static, seq, &r) {
1231 text_size = add_to_rb(&printk_rb_dynamic, &r);
1232 if (text_size > free)
1233 free = 0;
1234 else
1235 free -= text_size;
1236 }
1237
1238 prb = &printk_rb_dynamic;
1239
1240 local_irq_restore(flags);
1241
1242 /*
1243 * Copy any remaining messages that might have appeared from
1244 * NMI context after copying but before switching to the
1245 * dynamic buffer.
1246 */
1247 prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1248 text_size = add_to_rb(&printk_rb_dynamic, &r);
1249 if (text_size > free)
1250 free = 0;
1251 else
1252 free -= text_size;
1253 }
1254
1255 if (seq != prb_next_seq(&printk_rb_static)) {
1256 pr_err("dropped %llu messages\n",
1257 prb_next_seq(&printk_rb_static) - seq);
1258 }
1259
1260 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1261 pr_info("early log buf free: %u(%u%%)\n",
1262 free, (free * 100) / __LOG_BUF_LEN);
1263 return;
1264
1265err_free_descs:
1266 memblock_free(new_descs, new_descs_size);
1267err_free_log_buf:
1268 memblock_free(new_log_buf, new_log_buf_len);
1269}
1270
1271static bool __read_mostly ignore_loglevel;
1272
1273static int __init ignore_loglevel_setup(char *str)
1274{
1275 ignore_loglevel = true;
1276 pr_info("debug: ignoring loglevel setting.\n");
1277
1278 return 0;
1279}
1280
1281early_param("ignore_loglevel", ignore_loglevel_setup);
1282module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1283MODULE_PARM_DESC(ignore_loglevel,
1284 "ignore loglevel setting (prints all kernel messages to the console)");
1285
1286static bool suppress_message_printing(int level)
1287{
1288 return (level >= console_loglevel && !ignore_loglevel);
1289}
1290
1291#ifdef CONFIG_BOOT_PRINTK_DELAY
1292
1293static int boot_delay; /* msecs delay after each printk during bootup */
1294static unsigned long long loops_per_msec; /* based on boot_delay */
1295
1296static int __init boot_delay_setup(char *str)
1297{
1298 unsigned long lpj;
1299
1300 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1301 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1302
1303 get_option(&str, &boot_delay);
1304 if (boot_delay > 10 * 1000)
1305 boot_delay = 0;
1306
1307 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1308 "HZ: %d, loops_per_msec: %llu\n",
1309 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1310 return 0;
1311}
1312early_param("boot_delay", boot_delay_setup);
1313
1314static void boot_delay_msec(int level)
1315{
1316 unsigned long long k;
1317 unsigned long timeout;
1318
1319 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1320 || suppress_message_printing(level)) {
1321 return;
1322 }
1323
1324 k = (unsigned long long)loops_per_msec * boot_delay;
1325
1326 timeout = jiffies + msecs_to_jiffies(boot_delay);
1327 while (k) {
1328 k--;
1329 cpu_relax();
1330 /*
1331 * use (volatile) jiffies to prevent
1332 * compiler reduction; loop termination via jiffies
1333 * is secondary and may or may not happen.
1334 */
1335 if (time_after(jiffies, timeout))
1336 break;
1337 touch_nmi_watchdog();
1338 }
1339}
1340#else
1341static inline void boot_delay_msec(int level)
1342{
1343}
1344#endif
1345
1346static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1347module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1348
1349static size_t print_syslog(unsigned int level, char *buf)
1350{
1351 return sprintf(buf, "<%u>", level);
1352}
1353
1354static size_t print_time(u64 ts, char *buf)
1355{
1356 unsigned long rem_nsec = do_div(ts, 1000000000);
1357
1358 return sprintf(buf, "[%5lu.%06lu]",
1359 (unsigned long)ts, rem_nsec / 1000);
1360}
1361
1362#ifdef CONFIG_PRINTK_CALLER
1363static size_t print_caller(u32 id, char *buf)
1364{
1365 char caller[12];
1366
1367 snprintf(caller, sizeof(caller), "%c%u",
1368 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1369 return sprintf(buf, "[%6s]", caller);
1370}
1371#else
1372#define print_caller(id, buf) 0
1373#endif
1374
1375static size_t info_print_prefix(const struct printk_info *info, bool syslog,
1376 bool time, char *buf)
1377{
1378 size_t len = 0;
1379
1380 if (syslog)
1381 len = print_syslog((info->facility << 3) | info->level, buf);
1382
1383 if (time)
1384 len += print_time(info->ts_nsec, buf + len);
1385
1386 len += print_caller(info->caller_id, buf + len);
1387
1388 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1389 buf[len++] = ' ';
1390 buf[len] = '\0';
1391 }
1392
1393 return len;
1394}
1395
1396/*
1397 * Prepare the record for printing. The text is shifted within the given
1398 * buffer to avoid a need for another one. The following operations are
1399 * done:
1400 *
1401 * - Add prefix for each line.
1402 * - Drop truncated lines that no longer fit into the buffer.
1403 * - Add the trailing newline that has been removed in vprintk_store().
1404 * - Add a string terminator.
1405 *
1406 * Since the produced string is always terminated, the maximum possible
1407 * return value is @r->text_buf_size - 1;
1408 *
1409 * Return: The length of the updated/prepared text, including the added
1410 * prefixes and the newline. The terminator is not counted. The dropped
1411 * line(s) are not counted.
1412 */
1413static size_t record_print_text(struct printk_record *r, bool syslog,
1414 bool time)
1415{
1416 size_t text_len = r->info->text_len;
1417 size_t buf_size = r->text_buf_size;
1418 char *text = r->text_buf;
1419 char prefix[PREFIX_MAX];
1420 bool truncated = false;
1421 size_t prefix_len;
1422 size_t line_len;
1423 size_t len = 0;
1424 char *next;
1425
1426 /*
1427 * If the message was truncated because the buffer was not large
1428 * enough, treat the available text as if it were the full text.
1429 */
1430 if (text_len > buf_size)
1431 text_len = buf_size;
1432
1433 prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1434
1435 /*
1436 * @text_len: bytes of unprocessed text
1437 * @line_len: bytes of current line _without_ newline
1438 * @text: pointer to beginning of current line
1439 * @len: number of bytes prepared in r->text_buf
1440 */
1441 for (;;) {
1442 next = memchr(text, '\n', text_len);
1443 if (next) {
1444 line_len = next - text;
1445 } else {
1446 /* Drop truncated line(s). */
1447 if (truncated)
1448 break;
1449 line_len = text_len;
1450 }
1451
1452 /*
1453 * Truncate the text if there is not enough space to add the
1454 * prefix and a trailing newline and a terminator.
1455 */
1456 if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1457 /* Drop even the current line if no space. */
1458 if (len + prefix_len + line_len + 1 + 1 > buf_size)
1459 break;
1460
1461 text_len = buf_size - len - prefix_len - 1 - 1;
1462 truncated = true;
1463 }
1464
1465 memmove(text + prefix_len, text, text_len);
1466 memcpy(text, prefix, prefix_len);
1467
1468 /*
1469 * Increment the prepared length to include the text and
1470 * prefix that were just moved+copied. Also increment for the
1471 * newline at the end of this line. If this is the last line,
1472 * there is no newline, but it will be added immediately below.
1473 */
1474 len += prefix_len + line_len + 1;
1475 if (text_len == line_len) {
1476 /*
1477 * This is the last line. Add the trailing newline
1478 * removed in vprintk_store().
1479 */
1480 text[prefix_len + line_len] = '\n';
1481 break;
1482 }
1483
1484 /*
1485 * Advance beyond the added prefix and the related line with
1486 * its newline.
1487 */
1488 text += prefix_len + line_len + 1;
1489
1490 /*
1491 * The remaining text has only decreased by the line with its
1492 * newline.
1493 *
1494 * Note that @text_len can become zero. It happens when @text
1495 * ended with a newline (either due to truncation or the
1496 * original string ending with "\n\n"). The loop is correctly
1497 * repeated and (if not truncated) an empty line with a prefix
1498 * will be prepared.
1499 */
1500 text_len -= line_len + 1;
1501 }
1502
1503 /*
1504 * If a buffer was provided, it will be terminated. Space for the
1505 * string terminator is guaranteed to be available. The terminator is
1506 * not counted in the return value.
1507 */
1508 if (buf_size > 0)
1509 r->text_buf[len] = 0;
1510
1511 return len;
1512}
1513
1514static size_t get_record_print_text_size(struct printk_info *info,
1515 unsigned int line_count,
1516 bool syslog, bool time)
1517{
1518 char prefix[PREFIX_MAX];
1519 size_t prefix_len;
1520
1521 prefix_len = info_print_prefix(info, syslog, time, prefix);
1522
1523 /*
1524 * Each line will be preceded with a prefix. The intermediate
1525 * newlines are already within the text, but a final trailing
1526 * newline will be added.
1527 */
1528 return ((prefix_len * line_count) + info->text_len + 1);
1529}
1530
1531/*
1532 * Beginning with @start_seq, find the first record where it and all following
1533 * records up to (but not including) @max_seq fit into @size.
1534 *
1535 * @max_seq is simply an upper bound and does not need to exist. If the caller
1536 * does not require an upper bound, -1 can be used for @max_seq.
1537 */
1538static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1539 bool syslog, bool time)
1540{
1541 struct printk_info info;
1542 unsigned int line_count;
1543 size_t len = 0;
1544 u64 seq;
1545
1546 /* Determine the size of the records up to @max_seq. */
1547 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1548 if (info.seq >= max_seq)
1549 break;
1550 len += get_record_print_text_size(&info, line_count, syslog, time);
1551 }
1552
1553 /*
1554 * Adjust the upper bound for the next loop to avoid subtracting
1555 * lengths that were never added.
1556 */
1557 if (seq < max_seq)
1558 max_seq = seq;
1559
1560 /*
1561 * Move first record forward until length fits into the buffer. Ignore
1562 * newest messages that were not counted in the above cycle. Messages
1563 * might appear and get lost in the meantime. This is a best effort
1564 * that prevents an infinite loop that could occur with a retry.
1565 */
1566 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1567 if (len <= size || info.seq >= max_seq)
1568 break;
1569 len -= get_record_print_text_size(&info, line_count, syslog, time);
1570 }
1571
1572 return seq;
1573}
1574
1575/* The caller is responsible for making sure @size is greater than 0. */
1576static int syslog_print(char __user *buf, int size)
1577{
1578 struct printk_info info;
1579 struct printk_record r;
1580 char *text;
1581 int len = 0;
1582 u64 seq;
1583
1584 text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1585 if (!text)
1586 return -ENOMEM;
1587
1588 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1589
1590 mutex_lock(&syslog_lock);
1591
1592 /*
1593 * Wait for the @syslog_seq record to be available. @syslog_seq may
1594 * change while waiting.
1595 */
1596 do {
1597 seq = syslog_seq;
1598
1599 mutex_unlock(&syslog_lock);
1600 /*
1601 * Guarantee this task is visible on the waitqueue before
1602 * checking the wake condition.
1603 *
1604 * The full memory barrier within set_current_state() of
1605 * prepare_to_wait_event() pairs with the full memory barrier
1606 * within wq_has_sleeper().
1607 *
1608 * This pairs with __wake_up_klogd:A.
1609 */
1610 len = wait_event_interruptible(log_wait,
1611 prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
1612 mutex_lock(&syslog_lock);
1613
1614 if (len)
1615 goto out;
1616 } while (syslog_seq != seq);
1617
1618 /*
1619 * Copy records that fit into the buffer. The above cycle makes sure
1620 * that the first record is always available.
1621 */
1622 do {
1623 size_t n;
1624 size_t skip;
1625 int err;
1626
1627 if (!prb_read_valid(prb, syslog_seq, &r))
1628 break;
1629
1630 if (r.info->seq != syslog_seq) {
1631 /* message is gone, move to next valid one */
1632 syslog_seq = r.info->seq;
1633 syslog_partial = 0;
1634 }
1635
1636 /*
1637 * To keep reading/counting partial line consistent,
1638 * use printk_time value as of the beginning of a line.
1639 */
1640 if (!syslog_partial)
1641 syslog_time = printk_time;
1642
1643 skip = syslog_partial;
1644 n = record_print_text(&r, true, syslog_time);
1645 if (n - syslog_partial <= size) {
1646 /* message fits into buffer, move forward */
1647 syslog_seq = r.info->seq + 1;
1648 n -= syslog_partial;
1649 syslog_partial = 0;
1650 } else if (!len){
1651 /* partial read(), remember position */
1652 n = size;
1653 syslog_partial += n;
1654 } else
1655 n = 0;
1656
1657 if (!n)
1658 break;
1659
1660 mutex_unlock(&syslog_lock);
1661 err = copy_to_user(buf, text + skip, n);
1662 mutex_lock(&syslog_lock);
1663
1664 if (err) {
1665 if (!len)
1666 len = -EFAULT;
1667 break;
1668 }
1669
1670 len += n;
1671 size -= n;
1672 buf += n;
1673 } while (size);
1674out:
1675 mutex_unlock(&syslog_lock);
1676 kfree(text);
1677 return len;
1678}
1679
1680static int syslog_print_all(char __user *buf, int size, bool clear)
1681{
1682 struct printk_info info;
1683 struct printk_record r;
1684 char *text;
1685 int len = 0;
1686 u64 seq;
1687 bool time;
1688
1689 text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1690 if (!text)
1691 return -ENOMEM;
1692
1693 time = printk_time;
1694 /*
1695 * Find first record that fits, including all following records,
1696 * into the user-provided buffer for this dump.
1697 */
1698 seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1699 size, true, time);
1700
1701 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1702
1703 len = 0;
1704 prb_for_each_record(seq, prb, seq, &r) {
1705 int textlen;
1706
1707 textlen = record_print_text(&r, true, time);
1708
1709 if (len + textlen > size) {
1710 seq--;
1711 break;
1712 }
1713
1714 if (copy_to_user(buf + len, text, textlen))
1715 len = -EFAULT;
1716 else
1717 len += textlen;
1718
1719 if (len < 0)
1720 break;
1721 }
1722
1723 if (clear) {
1724 mutex_lock(&syslog_lock);
1725 latched_seq_write(&clear_seq, seq);
1726 mutex_unlock(&syslog_lock);
1727 }
1728
1729 kfree(text);
1730 return len;
1731}
1732
1733static void syslog_clear(void)
1734{
1735 mutex_lock(&syslog_lock);
1736 latched_seq_write(&clear_seq, prb_next_seq(prb));
1737 mutex_unlock(&syslog_lock);
1738}
1739
1740int do_syslog(int type, char __user *buf, int len, int source)
1741{
1742 struct printk_info info;
1743 bool clear = false;
1744 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1745 int error;
1746
1747 error = check_syslog_permissions(type, source);
1748 if (error)
1749 return error;
1750
1751 switch (type) {
1752 case SYSLOG_ACTION_CLOSE: /* Close log */
1753 break;
1754 case SYSLOG_ACTION_OPEN: /* Open log */
1755 break;
1756 case SYSLOG_ACTION_READ: /* Read from log */
1757 if (!buf || len < 0)
1758 return -EINVAL;
1759 if (!len)
1760 return 0;
1761 if (!access_ok(buf, len))
1762 return -EFAULT;
1763 error = syslog_print(buf, len);
1764 break;
1765 /* Read/clear last kernel messages */
1766 case SYSLOG_ACTION_READ_CLEAR:
1767 clear = true;
1768 fallthrough;
1769 /* Read last kernel messages */
1770 case SYSLOG_ACTION_READ_ALL:
1771 if (!buf || len < 0)
1772 return -EINVAL;
1773 if (!len)
1774 return 0;
1775 if (!access_ok(buf, len))
1776 return -EFAULT;
1777 error = syslog_print_all(buf, len, clear);
1778 break;
1779 /* Clear ring buffer */
1780 case SYSLOG_ACTION_CLEAR:
1781 syslog_clear();
1782 break;
1783 /* Disable logging to console */
1784 case SYSLOG_ACTION_CONSOLE_OFF:
1785 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1786 saved_console_loglevel = console_loglevel;
1787 console_loglevel = minimum_console_loglevel;
1788 break;
1789 /* Enable logging to console */
1790 case SYSLOG_ACTION_CONSOLE_ON:
1791 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1792 console_loglevel = saved_console_loglevel;
1793 saved_console_loglevel = LOGLEVEL_DEFAULT;
1794 }
1795 break;
1796 /* Set level of messages printed to console */
1797 case SYSLOG_ACTION_CONSOLE_LEVEL:
1798 if (len < 1 || len > 8)
1799 return -EINVAL;
1800 if (len < minimum_console_loglevel)
1801 len = minimum_console_loglevel;
1802 console_loglevel = len;
1803 /* Implicitly re-enable logging to console */
1804 saved_console_loglevel = LOGLEVEL_DEFAULT;
1805 break;
1806 /* Number of chars in the log buffer */
1807 case SYSLOG_ACTION_SIZE_UNREAD:
1808 mutex_lock(&syslog_lock);
1809 if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1810 /* No unread messages. */
1811 mutex_unlock(&syslog_lock);
1812 return 0;
1813 }
1814 if (info.seq != syslog_seq) {
1815 /* messages are gone, move to first one */
1816 syslog_seq = info.seq;
1817 syslog_partial = 0;
1818 }
1819 if (source == SYSLOG_FROM_PROC) {
1820 /*
1821 * Short-cut for poll(/"proc/kmsg") which simply checks
1822 * for pending data, not the size; return the count of
1823 * records, not the length.
1824 */
1825 error = prb_next_seq(prb) - syslog_seq;
1826 } else {
1827 bool time = syslog_partial ? syslog_time : printk_time;
1828 unsigned int line_count;
1829 u64 seq;
1830
1831 prb_for_each_info(syslog_seq, prb, seq, &info,
1832 &line_count) {
1833 error += get_record_print_text_size(&info, line_count,
1834 true, time);
1835 time = printk_time;
1836 }
1837 error -= syslog_partial;
1838 }
1839 mutex_unlock(&syslog_lock);
1840 break;
1841 /* Size of the log buffer */
1842 case SYSLOG_ACTION_SIZE_BUFFER:
1843 error = log_buf_len;
1844 break;
1845 default:
1846 error = -EINVAL;
1847 break;
1848 }
1849
1850 return error;
1851}
1852
1853SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1854{
1855 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1856}
1857
1858/*
1859 * Special console_lock variants that help to reduce the risk of soft-lockups.
1860 * They allow to pass console_lock to another printk() call using a busy wait.
1861 */
1862
1863#ifdef CONFIG_LOCKDEP
1864static struct lockdep_map console_owner_dep_map = {
1865 .name = "console_owner"
1866};
1867#endif
1868
1869static DEFINE_RAW_SPINLOCK(console_owner_lock);
1870static struct task_struct *console_owner;
1871static bool console_waiter;
1872
1873/**
1874 * console_lock_spinning_enable - mark beginning of code where another
1875 * thread might safely busy wait
1876 *
1877 * This basically converts console_lock into a spinlock. This marks
1878 * the section where the console_lock owner can not sleep, because
1879 * there may be a waiter spinning (like a spinlock). Also it must be
1880 * ready to hand over the lock at the end of the section.
1881 */
1882static void console_lock_spinning_enable(void)
1883{
1884 raw_spin_lock(&console_owner_lock);
1885 console_owner = current;
1886 raw_spin_unlock(&console_owner_lock);
1887
1888 /* The waiter may spin on us after setting console_owner */
1889 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1890}
1891
1892/**
1893 * console_lock_spinning_disable_and_check - mark end of code where another
1894 * thread was able to busy wait and check if there is a waiter
1895 * @cookie: cookie returned from console_srcu_read_lock()
1896 *
1897 * This is called at the end of the section where spinning is allowed.
1898 * It has two functions. First, it is a signal that it is no longer
1899 * safe to start busy waiting for the lock. Second, it checks if
1900 * there is a busy waiter and passes the lock rights to her.
1901 *
1902 * Important: Callers lose both the console_lock and the SRCU read lock if
1903 * there was a busy waiter. They must not touch items synchronized by
1904 * console_lock or SRCU read lock in this case.
1905 *
1906 * Return: 1 if the lock rights were passed, 0 otherwise.
1907 */
1908static int console_lock_spinning_disable_and_check(int cookie)
1909{
1910 int waiter;
1911
1912 raw_spin_lock(&console_owner_lock);
1913 waiter = READ_ONCE(console_waiter);
1914 console_owner = NULL;
1915 raw_spin_unlock(&console_owner_lock);
1916
1917 if (!waiter) {
1918 spin_release(&console_owner_dep_map, _THIS_IP_);
1919 return 0;
1920 }
1921
1922 /* The waiter is now free to continue */
1923 WRITE_ONCE(console_waiter, false);
1924
1925 spin_release(&console_owner_dep_map, _THIS_IP_);
1926
1927 /*
1928 * Preserve lockdep lock ordering. Release the SRCU read lock before
1929 * releasing the console_lock.
1930 */
1931 console_srcu_read_unlock(cookie);
1932
1933 /*
1934 * Hand off console_lock to waiter. The waiter will perform
1935 * the up(). After this, the waiter is the console_lock owner.
1936 */
1937 mutex_release(&console_lock_dep_map, _THIS_IP_);
1938 return 1;
1939}
1940
1941/**
1942 * console_trylock_spinning - try to get console_lock by busy waiting
1943 *
1944 * This allows to busy wait for the console_lock when the current
1945 * owner is running in specially marked sections. It means that
1946 * the current owner is running and cannot reschedule until it
1947 * is ready to lose the lock.
1948 *
1949 * Return: 1 if we got the lock, 0 othrewise
1950 */
1951static int console_trylock_spinning(void)
1952{
1953 struct task_struct *owner = NULL;
1954 bool waiter;
1955 bool spin = false;
1956 unsigned long flags;
1957
1958 if (console_trylock())
1959 return 1;
1960
1961 /*
1962 * It's unsafe to spin once a panic has begun. If we are the
1963 * panic CPU, we may have already halted the owner of the
1964 * console_sem. If we are not the panic CPU, then we should
1965 * avoid taking console_sem, so the panic CPU has a better
1966 * chance of cleanly acquiring it later.
1967 */
1968 if (panic_in_progress())
1969 return 0;
1970
1971 printk_safe_enter_irqsave(flags);
1972
1973 raw_spin_lock(&console_owner_lock);
1974 owner = READ_ONCE(console_owner);
1975 waiter = READ_ONCE(console_waiter);
1976 if (!waiter && owner && owner != current) {
1977 WRITE_ONCE(console_waiter, true);
1978 spin = true;
1979 }
1980 raw_spin_unlock(&console_owner_lock);
1981
1982 /*
1983 * If there is an active printk() writing to the
1984 * consoles, instead of having it write our data too,
1985 * see if we can offload that load from the active
1986 * printer, and do some printing ourselves.
1987 * Go into a spin only if there isn't already a waiter
1988 * spinning, and there is an active printer, and
1989 * that active printer isn't us (recursive printk?).
1990 */
1991 if (!spin) {
1992 printk_safe_exit_irqrestore(flags);
1993 return 0;
1994 }
1995
1996 /* We spin waiting for the owner to release us */
1997 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1998 /* Owner will clear console_waiter on hand off */
1999 while (READ_ONCE(console_waiter))
2000 cpu_relax();
2001 spin_release(&console_owner_dep_map, _THIS_IP_);
2002
2003 printk_safe_exit_irqrestore(flags);
2004 /*
2005 * The owner passed the console lock to us.
2006 * Since we did not spin on console lock, annotate
2007 * this as a trylock. Otherwise lockdep will
2008 * complain.
2009 */
2010 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
2011
2012 return 1;
2013}
2014
2015/*
2016 * Call the specified console driver, asking it to write out the specified
2017 * text and length. If @dropped_text is non-NULL and any records have been
2018 * dropped, a dropped message will be written out first.
2019 */
2020static void call_console_driver(struct console *con, const char *text, size_t len,
2021 char *dropped_text)
2022{
2023 size_t dropped_len;
2024
2025 if (con->dropped && dropped_text) {
2026 dropped_len = snprintf(dropped_text, DROPPED_TEXT_MAX,
2027 "** %lu printk messages dropped **\n",
2028 con->dropped);
2029 con->dropped = 0;
2030 con->write(con, dropped_text, dropped_len);
2031 }
2032
2033 con->write(con, text, len);
2034}
2035
2036/*
2037 * Recursion is tracked separately on each CPU. If NMIs are supported, an
2038 * additional NMI context per CPU is also separately tracked. Until per-CPU
2039 * is available, a separate "early tracking" is performed.
2040 */
2041static DEFINE_PER_CPU(u8, printk_count);
2042static u8 printk_count_early;
2043#ifdef CONFIG_HAVE_NMI
2044static DEFINE_PER_CPU(u8, printk_count_nmi);
2045static u8 printk_count_nmi_early;
2046#endif
2047
2048/*
2049 * Recursion is limited to keep the output sane. printk() should not require
2050 * more than 1 level of recursion (allowing, for example, printk() to trigger
2051 * a WARN), but a higher value is used in case some printk-internal errors
2052 * exist, such as the ringbuffer validation checks failing.
2053 */
2054#define PRINTK_MAX_RECURSION 3
2055
2056/*
2057 * Return a pointer to the dedicated counter for the CPU+context of the
2058 * caller.
2059 */
2060static u8 *__printk_recursion_counter(void)
2061{
2062#ifdef CONFIG_HAVE_NMI
2063 if (in_nmi()) {
2064 if (printk_percpu_data_ready())
2065 return this_cpu_ptr(&printk_count_nmi);
2066 return &printk_count_nmi_early;
2067 }
2068#endif
2069 if (printk_percpu_data_ready())
2070 return this_cpu_ptr(&printk_count);
2071 return &printk_count_early;
2072}
2073
2074/*
2075 * Enter recursion tracking. Interrupts are disabled to simplify tracking.
2076 * The caller must check the boolean return value to see if the recursion is
2077 * allowed. On failure, interrupts are not disabled.
2078 *
2079 * @recursion_ptr must be a variable of type (u8 *) and is the same variable
2080 * that is passed to printk_exit_irqrestore().
2081 */
2082#define printk_enter_irqsave(recursion_ptr, flags) \
2083({ \
2084 bool success = true; \
2085 \
2086 typecheck(u8 *, recursion_ptr); \
2087 local_irq_save(flags); \
2088 (recursion_ptr) = __printk_recursion_counter(); \
2089 if (*(recursion_ptr) > PRINTK_MAX_RECURSION) { \
2090 local_irq_restore(flags); \
2091 success = false; \
2092 } else { \
2093 (*(recursion_ptr))++; \
2094 } \
2095 success; \
2096})
2097
2098/* Exit recursion tracking, restoring interrupts. */
2099#define printk_exit_irqrestore(recursion_ptr, flags) \
2100 do { \
2101 typecheck(u8 *, recursion_ptr); \
2102 (*(recursion_ptr))--; \
2103 local_irq_restore(flags); \
2104 } while (0)
2105
2106int printk_delay_msec __read_mostly;
2107
2108static inline void printk_delay(int level)
2109{
2110 boot_delay_msec(level);
2111
2112 if (unlikely(printk_delay_msec)) {
2113 int m = printk_delay_msec;
2114
2115 while (m--) {
2116 mdelay(1);
2117 touch_nmi_watchdog();
2118 }
2119 }
2120}
2121
2122static inline u32 printk_caller_id(void)
2123{
2124 return in_task() ? task_pid_nr(current) :
2125 0x80000000 + smp_processor_id();
2126}
2127
2128/**
2129 * printk_parse_prefix - Parse level and control flags.
2130 *
2131 * @text: The terminated text message.
2132 * @level: A pointer to the current level value, will be updated.
2133 * @flags: A pointer to the current printk_info flags, will be updated.
2134 *
2135 * @level may be NULL if the caller is not interested in the parsed value.
2136 * Otherwise the variable pointed to by @level must be set to
2137 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2138 *
2139 * @flags may be NULL if the caller is not interested in the parsed value.
2140 * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2141 * value.
2142 *
2143 * Return: The length of the parsed level and control flags.
2144 */
2145u16 printk_parse_prefix(const char *text, int *level,
2146 enum printk_info_flags *flags)
2147{
2148 u16 prefix_len = 0;
2149 int kern_level;
2150
2151 while (*text) {
2152 kern_level = printk_get_level(text);
2153 if (!kern_level)
2154 break;
2155
2156 switch (kern_level) {
2157 case '0' ... '7':
2158 if (level && *level == LOGLEVEL_DEFAULT)
2159 *level = kern_level - '0';
2160 break;
2161 case 'c': /* KERN_CONT */
2162 if (flags)
2163 *flags |= LOG_CONT;
2164 }
2165
2166 prefix_len += 2;
2167 text += 2;
2168 }
2169
2170 return prefix_len;
2171}
2172
2173__printf(5, 0)
2174static u16 printk_sprint(char *text, u16 size, int facility,
2175 enum printk_info_flags *flags, const char *fmt,
2176 va_list args)
2177{
2178 u16 text_len;
2179
2180 text_len = vscnprintf(text, size, fmt, args);
2181
2182 /* Mark and strip a trailing newline. */
2183 if (text_len && text[text_len - 1] == '\n') {
2184 text_len--;
2185 *flags |= LOG_NEWLINE;
2186 }
2187
2188 /* Strip log level and control flags. */
2189 if (facility == 0) {
2190 u16 prefix_len;
2191
2192 prefix_len = printk_parse_prefix(text, NULL, NULL);
2193 if (prefix_len) {
2194 text_len -= prefix_len;
2195 memmove(text, text + prefix_len, text_len);
2196 }
2197 }
2198
2199 trace_console_rcuidle(text, text_len);
2200
2201 return text_len;
2202}
2203
2204__printf(4, 0)
2205int vprintk_store(int facility, int level,
2206 const struct dev_printk_info *dev_info,
2207 const char *fmt, va_list args)
2208{
2209 struct prb_reserved_entry e;
2210 enum printk_info_flags flags = 0;
2211 struct printk_record r;
2212 unsigned long irqflags;
2213 u16 trunc_msg_len = 0;
2214 char prefix_buf[8];
2215 u8 *recursion_ptr;
2216 u16 reserve_size;
2217 va_list args2;
2218 u32 caller_id;
2219 u16 text_len;
2220 int ret = 0;
2221 u64 ts_nsec;
2222
2223 if (!printk_enter_irqsave(recursion_ptr, irqflags))
2224 return 0;
2225
2226 /*
2227 * Since the duration of printk() can vary depending on the message
2228 * and state of the ringbuffer, grab the timestamp now so that it is
2229 * close to the call of printk(). This provides a more deterministic
2230 * timestamp with respect to the caller.
2231 */
2232 ts_nsec = local_clock();
2233
2234 caller_id = printk_caller_id();
2235
2236 /*
2237 * The sprintf needs to come first since the syslog prefix might be
2238 * passed in as a parameter. An extra byte must be reserved so that
2239 * later the vscnprintf() into the reserved buffer has room for the
2240 * terminating '\0', which is not counted by vsnprintf().
2241 */
2242 va_copy(args2, args);
2243 reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2244 va_end(args2);
2245
2246 if (reserve_size > LOG_LINE_MAX)
2247 reserve_size = LOG_LINE_MAX;
2248
2249 /* Extract log level or control flags. */
2250 if (facility == 0)
2251 printk_parse_prefix(&prefix_buf[0], &level, &flags);
2252
2253 if (level == LOGLEVEL_DEFAULT)
2254 level = default_message_loglevel;
2255
2256 if (dev_info)
2257 flags |= LOG_NEWLINE;
2258
2259 if (flags & LOG_CONT) {
2260 prb_rec_init_wr(&r, reserve_size);
2261 if (prb_reserve_in_last(&e, prb, &r, caller_id, LOG_LINE_MAX)) {
2262 text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2263 facility, &flags, fmt, args);
2264 r.info->text_len += text_len;
2265
2266 if (flags & LOG_NEWLINE) {
2267 r.info->flags |= LOG_NEWLINE;
2268 prb_final_commit(&e);
2269 } else {
2270 prb_commit(&e);
2271 }
2272
2273 ret = text_len;
2274 goto out;
2275 }
2276 }
2277
2278 /*
2279 * Explicitly initialize the record before every prb_reserve() call.
2280 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2281 * structure when they fail.
2282 */
2283 prb_rec_init_wr(&r, reserve_size);
2284 if (!prb_reserve(&e, prb, &r)) {
2285 /* truncate the message if it is too long for empty buffer */
2286 truncate_msg(&reserve_size, &trunc_msg_len);
2287
2288 prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2289 if (!prb_reserve(&e, prb, &r))
2290 goto out;
2291 }
2292
2293 /* fill message */
2294 text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2295 if (trunc_msg_len)
2296 memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2297 r.info->text_len = text_len + trunc_msg_len;
2298 r.info->facility = facility;
2299 r.info->level = level & 7;
2300 r.info->flags = flags & 0x1f;
2301 r.info->ts_nsec = ts_nsec;
2302 r.info->caller_id = caller_id;
2303 if (dev_info)
2304 memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2305
2306 /* A message without a trailing newline can be continued. */
2307 if (!(flags & LOG_NEWLINE))
2308 prb_commit(&e);
2309 else
2310 prb_final_commit(&e);
2311
2312 ret = text_len + trunc_msg_len;
2313out:
2314 printk_exit_irqrestore(recursion_ptr, irqflags);
2315 return ret;
2316}
2317
2318asmlinkage int vprintk_emit(int facility, int level,
2319 const struct dev_printk_info *dev_info,
2320 const char *fmt, va_list args)
2321{
2322 int printed_len;
2323 bool in_sched = false;
2324
2325 /* Suppress unimportant messages after panic happens */
2326 if (unlikely(suppress_printk))
2327 return 0;
2328
2329 if (unlikely(suppress_panic_printk) &&
2330 atomic_read(&panic_cpu) != raw_smp_processor_id())
2331 return 0;
2332
2333 if (level == LOGLEVEL_SCHED) {
2334 level = LOGLEVEL_DEFAULT;
2335 in_sched = true;
2336 }
2337
2338 printk_delay(level);
2339
2340 printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2341
2342 /* If called from the scheduler, we can not call up(). */
2343 if (!in_sched) {
2344 /*
2345 * The caller may be holding system-critical or
2346 * timing-sensitive locks. Disable preemption during
2347 * printing of all remaining records to all consoles so that
2348 * this context can return as soon as possible. Hopefully
2349 * another printk() caller will take over the printing.
2350 */
2351 preempt_disable();
2352 /*
2353 * Try to acquire and then immediately release the console
2354 * semaphore. The release will print out buffers. With the
2355 * spinning variant, this context tries to take over the
2356 * printing from another printing context.
2357 */
2358 if (console_trylock_spinning())
2359 console_unlock();
2360 preempt_enable();
2361 }
2362
2363 wake_up_klogd();
2364 return printed_len;
2365}
2366EXPORT_SYMBOL(vprintk_emit);
2367
2368int vprintk_default(const char *fmt, va_list args)
2369{
2370 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2371}
2372EXPORT_SYMBOL_GPL(vprintk_default);
2373
2374asmlinkage __visible int _printk(const char *fmt, ...)
2375{
2376 va_list args;
2377 int r;
2378
2379 va_start(args, fmt);
2380 r = vprintk(fmt, args);
2381 va_end(args);
2382
2383 return r;
2384}
2385EXPORT_SYMBOL(_printk);
2386
2387static bool pr_flush(int timeout_ms, bool reset_on_progress);
2388static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress);
2389
2390#else /* CONFIG_PRINTK */
2391
2392#define CONSOLE_LOG_MAX 0
2393#define DROPPED_TEXT_MAX 0
2394#define printk_time false
2395
2396#define prb_read_valid(rb, seq, r) false
2397#define prb_first_valid_seq(rb) 0
2398#define prb_next_seq(rb) 0
2399
2400static u64 syslog_seq;
2401
2402static size_t record_print_text(const struct printk_record *r,
2403 bool syslog, bool time)
2404{
2405 return 0;
2406}
2407static ssize_t info_print_ext_header(char *buf, size_t size,
2408 struct printk_info *info)
2409{
2410 return 0;
2411}
2412static ssize_t msg_print_ext_body(char *buf, size_t size,
2413 char *text, size_t text_len,
2414 struct dev_printk_info *dev_info) { return 0; }
2415static void console_lock_spinning_enable(void) { }
2416static int console_lock_spinning_disable_and_check(int cookie) { return 0; }
2417static void call_console_driver(struct console *con, const char *text, size_t len,
2418 char *dropped_text)
2419{
2420}
2421static bool suppress_message_printing(int level) { return false; }
2422static bool pr_flush(int timeout_ms, bool reset_on_progress) { return true; }
2423static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress) { return true; }
2424
2425#endif /* CONFIG_PRINTK */
2426
2427#ifdef CONFIG_EARLY_PRINTK
2428struct console *early_console;
2429
2430asmlinkage __visible void early_printk(const char *fmt, ...)
2431{
2432 va_list ap;
2433 char buf[512];
2434 int n;
2435
2436 if (!early_console)
2437 return;
2438
2439 va_start(ap, fmt);
2440 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2441 va_end(ap);
2442
2443 early_console->write(early_console, buf, n);
2444}
2445#endif
2446
2447static void set_user_specified(struct console_cmdline *c, bool user_specified)
2448{
2449 if (!user_specified)
2450 return;
2451
2452 /*
2453 * @c console was defined by the user on the command line.
2454 * Do not clear when added twice also by SPCR or the device tree.
2455 */
2456 c->user_specified = true;
2457 /* At least one console defined by the user on the command line. */
2458 console_set_on_cmdline = 1;
2459}
2460
2461static int __add_preferred_console(char *name, int idx, char *options,
2462 char *brl_options, bool user_specified)
2463{
2464 struct console_cmdline *c;
2465 int i;
2466
2467 /*
2468 * See if this tty is not yet registered, and
2469 * if we have a slot free.
2470 */
2471 for (i = 0, c = console_cmdline;
2472 i < MAX_CMDLINECONSOLES && c->name[0];
2473 i++, c++) {
2474 if (strcmp(c->name, name) == 0 && c->index == idx) {
2475 if (!brl_options)
2476 preferred_console = i;
2477 set_user_specified(c, user_specified);
2478 return 0;
2479 }
2480 }
2481 if (i == MAX_CMDLINECONSOLES)
2482 return -E2BIG;
2483 if (!brl_options)
2484 preferred_console = i;
2485 strscpy(c->name, name, sizeof(c->name));
2486 c->options = options;
2487 set_user_specified(c, user_specified);
2488 braille_set_options(c, brl_options);
2489
2490 c->index = idx;
2491 return 0;
2492}
2493
2494static int __init console_msg_format_setup(char *str)
2495{
2496 if (!strcmp(str, "syslog"))
2497 console_msg_format = MSG_FORMAT_SYSLOG;
2498 if (!strcmp(str, "default"))
2499 console_msg_format = MSG_FORMAT_DEFAULT;
2500 return 1;
2501}
2502__setup("console_msg_format=", console_msg_format_setup);
2503
2504/*
2505 * Set up a console. Called via do_early_param() in init/main.c
2506 * for each "console=" parameter in the boot command line.
2507 */
2508static int __init console_setup(char *str)
2509{
2510 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2511 char *s, *options, *brl_options = NULL;
2512 int idx;
2513
2514 /*
2515 * console="" or console=null have been suggested as a way to
2516 * disable console output. Use ttynull that has been created
2517 * for exactly this purpose.
2518 */
2519 if (str[0] == 0 || strcmp(str, "null") == 0) {
2520 __add_preferred_console("ttynull", 0, NULL, NULL, true);
2521 return 1;
2522 }
2523
2524 if (_braille_console_setup(&str, &brl_options))
2525 return 1;
2526
2527 /*
2528 * Decode str into name, index, options.
2529 */
2530 if (str[0] >= '0' && str[0] <= '9') {
2531 strcpy(buf, "ttyS");
2532 strncpy(buf + 4, str, sizeof(buf) - 5);
2533 } else {
2534 strncpy(buf, str, sizeof(buf) - 1);
2535 }
2536 buf[sizeof(buf) - 1] = 0;
2537 options = strchr(str, ',');
2538 if (options)
2539 *(options++) = 0;
2540#ifdef __sparc__
2541 if (!strcmp(str, "ttya"))
2542 strcpy(buf, "ttyS0");
2543 if (!strcmp(str, "ttyb"))
2544 strcpy(buf, "ttyS1");
2545#endif
2546 for (s = buf; *s; s++)
2547 if (isdigit(*s) || *s == ',')
2548 break;
2549 idx = simple_strtoul(s, NULL, 10);
2550 *s = 0;
2551
2552 __add_preferred_console(buf, idx, options, brl_options, true);
2553 return 1;
2554}
2555__setup("console=", console_setup);
2556
2557/**
2558 * add_preferred_console - add a device to the list of preferred consoles.
2559 * @name: device name
2560 * @idx: device index
2561 * @options: options for this console
2562 *
2563 * The last preferred console added will be used for kernel messages
2564 * and stdin/out/err for init. Normally this is used by console_setup
2565 * above to handle user-supplied console arguments; however it can also
2566 * be used by arch-specific code either to override the user or more
2567 * commonly to provide a default console (ie from PROM variables) when
2568 * the user has not supplied one.
2569 */
2570int add_preferred_console(char *name, int idx, char *options)
2571{
2572 return __add_preferred_console(name, idx, options, NULL, false);
2573}
2574
2575bool console_suspend_enabled = true;
2576EXPORT_SYMBOL(console_suspend_enabled);
2577
2578static int __init console_suspend_disable(char *str)
2579{
2580 console_suspend_enabled = false;
2581 return 1;
2582}
2583__setup("no_console_suspend", console_suspend_disable);
2584module_param_named(console_suspend, console_suspend_enabled,
2585 bool, S_IRUGO | S_IWUSR);
2586MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2587 " and hibernate operations");
2588
2589static bool printk_console_no_auto_verbose;
2590
2591void console_verbose(void)
2592{
2593 if (console_loglevel && !printk_console_no_auto_verbose)
2594 console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2595}
2596EXPORT_SYMBOL_GPL(console_verbose);
2597
2598module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2599MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2600
2601/**
2602 * suspend_console - suspend the console subsystem
2603 *
2604 * This disables printk() while we go into suspend states
2605 */
2606void suspend_console(void)
2607{
2608 if (!console_suspend_enabled)
2609 return;
2610 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2611 pr_flush(1000, true);
2612 console_lock();
2613 console_suspended = 1;
2614 up_console_sem();
2615}
2616
2617void resume_console(void)
2618{
2619 if (!console_suspend_enabled)
2620 return;
2621 down_console_sem();
2622 console_suspended = 0;
2623 console_unlock();
2624 pr_flush(1000, true);
2625}
2626
2627/**
2628 * console_cpu_notify - print deferred console messages after CPU hotplug
2629 * @cpu: unused
2630 *
2631 * If printk() is called from a CPU that is not online yet, the messages
2632 * will be printed on the console only if there are CON_ANYTIME consoles.
2633 * This function is called when a new CPU comes online (or fails to come
2634 * up) or goes offline.
2635 */
2636static int console_cpu_notify(unsigned int cpu)
2637{
2638 if (!cpuhp_tasks_frozen) {
2639 /* If trylock fails, someone else is doing the printing */
2640 if (console_trylock())
2641 console_unlock();
2642 }
2643 return 0;
2644}
2645
2646/**
2647 * console_lock - block the console subsystem from printing
2648 *
2649 * Acquires a lock which guarantees that no consoles will
2650 * be in or enter their write() callback.
2651 *
2652 * Can sleep, returns nothing.
2653 */
2654void console_lock(void)
2655{
2656 might_sleep();
2657
2658 down_console_sem();
2659 if (console_suspended)
2660 return;
2661 console_locked = 1;
2662 console_may_schedule = 1;
2663}
2664EXPORT_SYMBOL(console_lock);
2665
2666/**
2667 * console_trylock - try to block the console subsystem from printing
2668 *
2669 * Try to acquire a lock which guarantees that no consoles will
2670 * be in or enter their write() callback.
2671 *
2672 * returns 1 on success, and 0 on failure to acquire the lock.
2673 */
2674int console_trylock(void)
2675{
2676 if (down_trylock_console_sem())
2677 return 0;
2678 if (console_suspended) {
2679 up_console_sem();
2680 return 0;
2681 }
2682 console_locked = 1;
2683 console_may_schedule = 0;
2684 return 1;
2685}
2686EXPORT_SYMBOL(console_trylock);
2687
2688int is_console_locked(void)
2689{
2690 return console_locked;
2691}
2692EXPORT_SYMBOL(is_console_locked);
2693
2694/*
2695 * Return true when this CPU should unlock console_sem without pushing all
2696 * messages to the console. This reduces the chance that the console is
2697 * locked when the panic CPU tries to use it.
2698 */
2699static bool abandon_console_lock_in_panic(void)
2700{
2701 if (!panic_in_progress())
2702 return false;
2703
2704 /*
2705 * We can use raw_smp_processor_id() here because it is impossible for
2706 * the task to be migrated to the panic_cpu, or away from it. If
2707 * panic_cpu has already been set, and we're not currently executing on
2708 * that CPU, then we never will be.
2709 */
2710 return atomic_read(&panic_cpu) != raw_smp_processor_id();
2711}
2712
2713/*
2714 * Check if the given console is currently capable and allowed to print
2715 * records.
2716 *
2717 * Requires the console_srcu_read_lock.
2718 */
2719static inline bool console_is_usable(struct console *con)
2720{
2721 short flags = console_srcu_read_flags(con);
2722
2723 if (!(flags & CON_ENABLED))
2724 return false;
2725
2726 if (!con->write)
2727 return false;
2728
2729 /*
2730 * Console drivers may assume that per-cpu resources have been
2731 * allocated. So unless they're explicitly marked as being able to
2732 * cope (CON_ANYTIME) don't call them until this CPU is officially up.
2733 */
2734 if (!cpu_online(raw_smp_processor_id()) && !(flags & CON_ANYTIME))
2735 return false;
2736
2737 return true;
2738}
2739
2740static void __console_unlock(void)
2741{
2742 console_locked = 0;
2743 up_console_sem();
2744}
2745
2746/*
2747 * Print one record for the given console. The record printed is whatever
2748 * record is the next available record for the given console.
2749 *
2750 * @text is a buffer of size CONSOLE_LOG_MAX.
2751 *
2752 * If extended messages should be printed, @ext_text is a buffer of size
2753 * CONSOLE_EXT_LOG_MAX. Otherwise @ext_text must be NULL.
2754 *
2755 * If dropped messages should be printed, @dropped_text is a buffer of size
2756 * DROPPED_TEXT_MAX. Otherwise @dropped_text must be NULL.
2757 *
2758 * @handover will be set to true if a printk waiter has taken over the
2759 * console_lock, in which case the caller is no longer holding both the
2760 * console_lock and the SRCU read lock. Otherwise it is set to false.
2761 *
2762 * @cookie is the cookie from the SRCU read lock.
2763 *
2764 * Returns false if the given console has no next record to print, otherwise
2765 * true.
2766 *
2767 * Requires the console_lock and the SRCU read lock.
2768 */
2769static bool console_emit_next_record(struct console *con, char *text, char *ext_text,
2770 char *dropped_text, bool *handover, int cookie)
2771{
2772 static int panic_console_dropped;
2773 struct printk_info info;
2774 struct printk_record r;
2775 unsigned long flags;
2776 char *write_text;
2777 size_t len;
2778
2779 prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
2780
2781 *handover = false;
2782
2783 if (!prb_read_valid(prb, con->seq, &r))
2784 return false;
2785
2786 if (con->seq != r.info->seq) {
2787 con->dropped += r.info->seq - con->seq;
2788 con->seq = r.info->seq;
2789 if (panic_in_progress() && panic_console_dropped++ > 10) {
2790 suppress_panic_printk = 1;
2791 pr_warn_once("Too many dropped messages. Suppress messages on non-panic CPUs to prevent livelock.\n");
2792 }
2793 }
2794
2795 /* Skip record that has level above the console loglevel. */
2796 if (suppress_message_printing(r.info->level)) {
2797 con->seq++;
2798 goto skip;
2799 }
2800
2801 if (ext_text) {
2802 write_text = ext_text;
2803 len = info_print_ext_header(ext_text, CONSOLE_EXT_LOG_MAX, r.info);
2804 len += msg_print_ext_body(ext_text + len, CONSOLE_EXT_LOG_MAX - len,
2805 &r.text_buf[0], r.info->text_len, &r.info->dev_info);
2806 } else {
2807 write_text = text;
2808 len = record_print_text(&r, console_msg_format & MSG_FORMAT_SYSLOG, printk_time);
2809 }
2810
2811 /*
2812 * While actively printing out messages, if another printk()
2813 * were to occur on another CPU, it may wait for this one to
2814 * finish. This task can not be preempted if there is a
2815 * waiter waiting to take over.
2816 *
2817 * Interrupts are disabled because the hand over to a waiter
2818 * must not be interrupted until the hand over is completed
2819 * (@console_waiter is cleared).
2820 */
2821 printk_safe_enter_irqsave(flags);
2822 console_lock_spinning_enable();
2823
2824 stop_critical_timings(); /* don't trace print latency */
2825 call_console_driver(con, write_text, len, dropped_text);
2826 start_critical_timings();
2827
2828 con->seq++;
2829
2830 *handover = console_lock_spinning_disable_and_check(cookie);
2831 printk_safe_exit_irqrestore(flags);
2832skip:
2833 return true;
2834}
2835
2836/*
2837 * Print out all remaining records to all consoles.
2838 *
2839 * @do_cond_resched is set by the caller. It can be true only in schedulable
2840 * context.
2841 *
2842 * @next_seq is set to the sequence number after the last available record.
2843 * The value is valid only when this function returns true. It means that all
2844 * usable consoles are completely flushed.
2845 *
2846 * @handover will be set to true if a printk waiter has taken over the
2847 * console_lock, in which case the caller is no longer holding the
2848 * console_lock. Otherwise it is set to false.
2849 *
2850 * Returns true when there was at least one usable console and all messages
2851 * were flushed to all usable consoles. A returned false informs the caller
2852 * that everything was not flushed (either there were no usable consoles or
2853 * another context has taken over printing or it is a panic situation and this
2854 * is not the panic CPU). Regardless the reason, the caller should assume it
2855 * is not useful to immediately try again.
2856 *
2857 * Requires the console_lock.
2858 */
2859static bool console_flush_all(bool do_cond_resched, u64 *next_seq, bool *handover)
2860{
2861 static char dropped_text[DROPPED_TEXT_MAX];
2862 static char ext_text[CONSOLE_EXT_LOG_MAX];
2863 static char text[CONSOLE_LOG_MAX];
2864 bool any_usable = false;
2865 struct console *con;
2866 bool any_progress;
2867 int cookie;
2868
2869 *next_seq = 0;
2870 *handover = false;
2871
2872 do {
2873 any_progress = false;
2874
2875 cookie = console_srcu_read_lock();
2876 for_each_console_srcu(con) {
2877 bool progress;
2878
2879 if (!console_is_usable(con))
2880 continue;
2881 any_usable = true;
2882
2883 if (console_srcu_read_flags(con) & CON_EXTENDED) {
2884 /* Extended consoles do not print "dropped messages". */
2885 progress = console_emit_next_record(con, &text[0],
2886 &ext_text[0], NULL,
2887 handover, cookie);
2888 } else {
2889 progress = console_emit_next_record(con, &text[0],
2890 NULL, &dropped_text[0],
2891 handover, cookie);
2892 }
2893
2894 /*
2895 * If a handover has occurred, the SRCU read lock
2896 * is already released.
2897 */
2898 if (*handover)
2899 return false;
2900
2901 /* Track the next of the highest seq flushed. */
2902 if (con->seq > *next_seq)
2903 *next_seq = con->seq;
2904
2905 if (!progress)
2906 continue;
2907 any_progress = true;
2908
2909 /* Allow panic_cpu to take over the consoles safely. */
2910 if (abandon_console_lock_in_panic())
2911 goto abandon;
2912
2913 if (do_cond_resched)
2914 cond_resched();
2915 }
2916 console_srcu_read_unlock(cookie);
2917 } while (any_progress);
2918
2919 return any_usable;
2920
2921abandon:
2922 console_srcu_read_unlock(cookie);
2923 return false;
2924}
2925
2926/**
2927 * console_unlock - unblock the console subsystem from printing
2928 *
2929 * Releases the console_lock which the caller holds to block printing of
2930 * the console subsystem.
2931 *
2932 * While the console_lock was held, console output may have been buffered
2933 * by printk(). If this is the case, console_unlock(); emits
2934 * the output prior to releasing the lock.
2935 *
2936 * console_unlock(); may be called from any context.
2937 */
2938void console_unlock(void)
2939{
2940 bool do_cond_resched;
2941 bool handover;
2942 bool flushed;
2943 u64 next_seq;
2944
2945 if (console_suspended) {
2946 up_console_sem();
2947 return;
2948 }
2949
2950 /*
2951 * Console drivers are called with interrupts disabled, so
2952 * @console_may_schedule should be cleared before; however, we may
2953 * end up dumping a lot of lines, for example, if called from
2954 * console registration path, and should invoke cond_resched()
2955 * between lines if allowable. Not doing so can cause a very long
2956 * scheduling stall on a slow console leading to RCU stall and
2957 * softlockup warnings which exacerbate the issue with more
2958 * messages practically incapacitating the system. Therefore, create
2959 * a local to use for the printing loop.
2960 */
2961 do_cond_resched = console_may_schedule;
2962
2963 do {
2964 console_may_schedule = 0;
2965
2966 flushed = console_flush_all(do_cond_resched, &next_seq, &handover);
2967 if (!handover)
2968 __console_unlock();
2969
2970 /*
2971 * Abort if there was a failure to flush all messages to all
2972 * usable consoles. Either it is not possible to flush (in
2973 * which case it would be an infinite loop of retrying) or
2974 * another context has taken over printing.
2975 */
2976 if (!flushed)
2977 break;
2978
2979 /*
2980 * Some context may have added new records after
2981 * console_flush_all() but before unlocking the console.
2982 * Re-check if there is a new record to flush. If the trylock
2983 * fails, another context is already handling the printing.
2984 */
2985 } while (prb_read_valid(prb, next_seq, NULL) && console_trylock());
2986}
2987EXPORT_SYMBOL(console_unlock);
2988
2989/**
2990 * console_conditional_schedule - yield the CPU if required
2991 *
2992 * If the console code is currently allowed to sleep, and
2993 * if this CPU should yield the CPU to another task, do
2994 * so here.
2995 *
2996 * Must be called within console_lock();.
2997 */
2998void __sched console_conditional_schedule(void)
2999{
3000 if (console_may_schedule)
3001 cond_resched();
3002}
3003EXPORT_SYMBOL(console_conditional_schedule);
3004
3005void console_unblank(void)
3006{
3007 struct console *c;
3008 int cookie;
3009
3010 /*
3011 * Stop console printing because the unblank() callback may
3012 * assume the console is not within its write() callback.
3013 *
3014 * If @oops_in_progress is set, this may be an atomic context.
3015 * In that case, attempt a trylock as best-effort.
3016 */
3017 if (oops_in_progress) {
3018 if (down_trylock_console_sem() != 0)
3019 return;
3020 } else
3021 console_lock();
3022
3023 console_locked = 1;
3024 console_may_schedule = 0;
3025
3026 cookie = console_srcu_read_lock();
3027 for_each_console_srcu(c) {
3028 if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank)
3029 c->unblank();
3030 }
3031 console_srcu_read_unlock(cookie);
3032
3033 console_unlock();
3034
3035 if (!oops_in_progress)
3036 pr_flush(1000, true);
3037}
3038
3039/**
3040 * console_flush_on_panic - flush console content on panic
3041 * @mode: flush all messages in buffer or just the pending ones
3042 *
3043 * Immediately output all pending messages no matter what.
3044 */
3045void console_flush_on_panic(enum con_flush_mode mode)
3046{
3047 /*
3048 * If someone else is holding the console lock, trylock will fail
3049 * and may_schedule may be set. Ignore and proceed to unlock so
3050 * that messages are flushed out. As this can be called from any
3051 * context and we don't want to get preempted while flushing,
3052 * ensure may_schedule is cleared.
3053 */
3054 console_trylock();
3055 console_may_schedule = 0;
3056
3057 if (mode == CONSOLE_REPLAY_ALL) {
3058 struct console *c;
3059 int cookie;
3060 u64 seq;
3061
3062 seq = prb_first_valid_seq(prb);
3063
3064 cookie = console_srcu_read_lock();
3065 for_each_console_srcu(c) {
3066 /*
3067 * If the above console_trylock() failed, this is an
3068 * unsynchronized assignment. But in that case, the
3069 * kernel is in "hope and pray" mode anyway.
3070 */
3071 c->seq = seq;
3072 }
3073 console_srcu_read_unlock(cookie);
3074 }
3075 console_unlock();
3076}
3077
3078/*
3079 * Return the console tty driver structure and its associated index
3080 */
3081struct tty_driver *console_device(int *index)
3082{
3083 struct console *c;
3084 struct tty_driver *driver = NULL;
3085 int cookie;
3086
3087 /*
3088 * Take console_lock to serialize device() callback with
3089 * other console operations. For example, fg_console is
3090 * modified under console_lock when switching vt.
3091 */
3092 console_lock();
3093
3094 cookie = console_srcu_read_lock();
3095 for_each_console_srcu(c) {
3096 if (!c->device)
3097 continue;
3098 driver = c->device(c, index);
3099 if (driver)
3100 break;
3101 }
3102 console_srcu_read_unlock(cookie);
3103
3104 console_unlock();
3105 return driver;
3106}
3107
3108/*
3109 * Prevent further output on the passed console device so that (for example)
3110 * serial drivers can disable console output before suspending a port, and can
3111 * re-enable output afterwards.
3112 */
3113void console_stop(struct console *console)
3114{
3115 __pr_flush(console, 1000, true);
3116 console_list_lock();
3117 console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3118 console_list_unlock();
3119
3120 /*
3121 * Ensure that all SRCU list walks have completed. All contexts must
3122 * be able to see that this console is disabled so that (for example)
3123 * the caller can suspend the port without risk of another context
3124 * using the port.
3125 */
3126 synchronize_srcu(&console_srcu);
3127}
3128EXPORT_SYMBOL(console_stop);
3129
3130void console_start(struct console *console)
3131{
3132 console_list_lock();
3133 console_srcu_write_flags(console, console->flags | CON_ENABLED);
3134 console_list_unlock();
3135 __pr_flush(console, 1000, true);
3136}
3137EXPORT_SYMBOL(console_start);
3138
3139static int __read_mostly keep_bootcon;
3140
3141static int __init keep_bootcon_setup(char *str)
3142{
3143 keep_bootcon = 1;
3144 pr_info("debug: skip boot console de-registration.\n");
3145
3146 return 0;
3147}
3148
3149early_param("keep_bootcon", keep_bootcon_setup);
3150
3151/*
3152 * This is called by register_console() to try to match
3153 * the newly registered console with any of the ones selected
3154 * by either the command line or add_preferred_console() and
3155 * setup/enable it.
3156 *
3157 * Care need to be taken with consoles that are statically
3158 * enabled such as netconsole
3159 */
3160static int try_enable_preferred_console(struct console *newcon,
3161 bool user_specified)
3162{
3163 struct console_cmdline *c;
3164 int i, err;
3165
3166 for (i = 0, c = console_cmdline;
3167 i < MAX_CMDLINECONSOLES && c->name[0];
3168 i++, c++) {
3169 if (c->user_specified != user_specified)
3170 continue;
3171 if (!newcon->match ||
3172 newcon->match(newcon, c->name, c->index, c->options) != 0) {
3173 /* default matching */
3174 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
3175 if (strcmp(c->name, newcon->name) != 0)
3176 continue;
3177 if (newcon->index >= 0 &&
3178 newcon->index != c->index)
3179 continue;
3180 if (newcon->index < 0)
3181 newcon->index = c->index;
3182
3183 if (_braille_register_console(newcon, c))
3184 return 0;
3185
3186 if (newcon->setup &&
3187 (err = newcon->setup(newcon, c->options)) != 0)
3188 return err;
3189 }
3190 newcon->flags |= CON_ENABLED;
3191 if (i == preferred_console)
3192 newcon->flags |= CON_CONSDEV;
3193 return 0;
3194 }
3195
3196 /*
3197 * Some consoles, such as pstore and netconsole, can be enabled even
3198 * without matching. Accept the pre-enabled consoles only when match()
3199 * and setup() had a chance to be called.
3200 */
3201 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
3202 return 0;
3203
3204 return -ENOENT;
3205}
3206
3207/* Try to enable the console unconditionally */
3208static void try_enable_default_console(struct console *newcon)
3209{
3210 if (newcon->index < 0)
3211 newcon->index = 0;
3212
3213 if (newcon->setup && newcon->setup(newcon, NULL) != 0)
3214 return;
3215
3216 newcon->flags |= CON_ENABLED;
3217
3218 if (newcon->device)
3219 newcon->flags |= CON_CONSDEV;
3220}
3221
3222#define con_printk(lvl, con, fmt, ...) \
3223 printk(lvl pr_fmt("%sconsole [%s%d] " fmt), \
3224 (con->flags & CON_BOOT) ? "boot" : "", \
3225 con->name, con->index, ##__VA_ARGS__)
3226
3227static void console_init_seq(struct console *newcon, bool bootcon_registered)
3228{
3229 struct console *con;
3230 bool handover;
3231
3232 if (newcon->flags & (CON_PRINTBUFFER | CON_BOOT)) {
3233 /* Get a consistent copy of @syslog_seq. */
3234 mutex_lock(&syslog_lock);
3235 newcon->seq = syslog_seq;
3236 mutex_unlock(&syslog_lock);
3237 } else {
3238 /* Begin with next message added to ringbuffer. */
3239 newcon->seq = prb_next_seq(prb);
3240
3241 /*
3242 * If any enabled boot consoles are due to be unregistered
3243 * shortly, some may not be caught up and may be the same
3244 * device as @newcon. Since it is not known which boot console
3245 * is the same device, flush all consoles and, if necessary,
3246 * start with the message of the enabled boot console that is
3247 * the furthest behind.
3248 */
3249 if (bootcon_registered && !keep_bootcon) {
3250 /*
3251 * Hold the console_lock to stop console printing and
3252 * guarantee safe access to console->seq.
3253 */
3254 console_lock();
3255
3256 /*
3257 * Flush all consoles and set the console to start at
3258 * the next unprinted sequence number.
3259 */
3260 if (!console_flush_all(true, &newcon->seq, &handover)) {
3261 /*
3262 * Flushing failed. Just choose the lowest
3263 * sequence of the enabled boot consoles.
3264 */
3265
3266 /*
3267 * If there was a handover, this context no
3268 * longer holds the console_lock.
3269 */
3270 if (handover)
3271 console_lock();
3272
3273 newcon->seq = prb_next_seq(prb);
3274 for_each_console(con) {
3275 if ((con->flags & CON_BOOT) &&
3276 (con->flags & CON_ENABLED) &&
3277 con->seq < newcon->seq) {
3278 newcon->seq = con->seq;
3279 }
3280 }
3281 }
3282
3283 console_unlock();
3284 }
3285 }
3286}
3287
3288#define console_first() \
3289 hlist_entry(console_list.first, struct console, node)
3290
3291static int unregister_console_locked(struct console *console);
3292
3293/*
3294 * The console driver calls this routine during kernel initialization
3295 * to register the console printing procedure with printk() and to
3296 * print any messages that were printed by the kernel before the
3297 * console driver was initialized.
3298 *
3299 * This can happen pretty early during the boot process (because of
3300 * early_printk) - sometimes before setup_arch() completes - be careful
3301 * of what kernel features are used - they may not be initialised yet.
3302 *
3303 * There are two types of consoles - bootconsoles (early_printk) and
3304 * "real" consoles (everything which is not a bootconsole) which are
3305 * handled differently.
3306 * - Any number of bootconsoles can be registered at any time.
3307 * - As soon as a "real" console is registered, all bootconsoles
3308 * will be unregistered automatically.
3309 * - Once a "real" console is registered, any attempt to register a
3310 * bootconsoles will be rejected
3311 */
3312void register_console(struct console *newcon)
3313{
3314 struct console *con;
3315 bool bootcon_registered = false;
3316 bool realcon_registered = false;
3317 int err;
3318
3319 console_list_lock();
3320
3321 for_each_console(con) {
3322 if (WARN(con == newcon, "console '%s%d' already registered\n",
3323 con->name, con->index)) {
3324 goto unlock;
3325 }
3326
3327 if (con->flags & CON_BOOT)
3328 bootcon_registered = true;
3329 else
3330 realcon_registered = true;
3331 }
3332
3333 /* Do not register boot consoles when there already is a real one. */
3334 if ((newcon->flags & CON_BOOT) && realcon_registered) {
3335 pr_info("Too late to register bootconsole %s%d\n",
3336 newcon->name, newcon->index);
3337 goto unlock;
3338 }
3339
3340 /*
3341 * See if we want to enable this console driver by default.
3342 *
3343 * Nope when a console is preferred by the command line, device
3344 * tree, or SPCR.
3345 *
3346 * The first real console with tty binding (driver) wins. More
3347 * consoles might get enabled before the right one is found.
3348 *
3349 * Note that a console with tty binding will have CON_CONSDEV
3350 * flag set and will be first in the list.
3351 */
3352 if (preferred_console < 0) {
3353 if (hlist_empty(&console_list) || !console_first()->device ||
3354 console_first()->flags & CON_BOOT) {
3355 try_enable_default_console(newcon);
3356 }
3357 }
3358
3359 /* See if this console matches one we selected on the command line */
3360 err = try_enable_preferred_console(newcon, true);
3361
3362 /* If not, try to match against the platform default(s) */
3363 if (err == -ENOENT)
3364 err = try_enable_preferred_console(newcon, false);
3365
3366 /* printk() messages are not printed to the Braille console. */
3367 if (err || newcon->flags & CON_BRL)
3368 goto unlock;
3369
3370 /*
3371 * If we have a bootconsole, and are switching to a real console,
3372 * don't print everything out again, since when the boot console, and
3373 * the real console are the same physical device, it's annoying to
3374 * see the beginning boot messages twice
3375 */
3376 if (bootcon_registered &&
3377 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
3378 newcon->flags &= ~CON_PRINTBUFFER;
3379 }
3380
3381 newcon->dropped = 0;
3382 console_init_seq(newcon, bootcon_registered);
3383
3384 /*
3385 * Put this console in the list - keep the
3386 * preferred driver at the head of the list.
3387 */
3388 if (hlist_empty(&console_list)) {
3389 /* Ensure CON_CONSDEV is always set for the head. */
3390 newcon->flags |= CON_CONSDEV;
3391 hlist_add_head_rcu(&newcon->node, &console_list);
3392
3393 } else if (newcon->flags & CON_CONSDEV) {
3394 /* Only the new head can have CON_CONSDEV set. */
3395 console_srcu_write_flags(console_first(), console_first()->flags & ~CON_CONSDEV);
3396 hlist_add_head_rcu(&newcon->node, &console_list);
3397
3398 } else {
3399 hlist_add_behind_rcu(&newcon->node, console_list.first);
3400 }
3401
3402 /*
3403 * No need to synchronize SRCU here! The caller does not rely
3404 * on all contexts being able to see the new console before
3405 * register_console() completes.
3406 */
3407
3408 console_sysfs_notify();
3409
3410 /*
3411 * By unregistering the bootconsoles after we enable the real console
3412 * we get the "console xxx enabled" message on all the consoles -
3413 * boot consoles, real consoles, etc - this is to ensure that end
3414 * users know there might be something in the kernel's log buffer that
3415 * went to the bootconsole (that they do not see on the real console)
3416 */
3417 con_printk(KERN_INFO, newcon, "enabled\n");
3418 if (bootcon_registered &&
3419 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
3420 !keep_bootcon) {
3421 struct hlist_node *tmp;
3422
3423 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3424 if (con->flags & CON_BOOT)
3425 unregister_console_locked(con);
3426 }
3427 }
3428unlock:
3429 console_list_unlock();
3430}
3431EXPORT_SYMBOL(register_console);
3432
3433/* Must be called under console_list_lock(). */
3434static int unregister_console_locked(struct console *console)
3435{
3436 int res;
3437
3438 lockdep_assert_console_list_lock_held();
3439
3440 con_printk(KERN_INFO, console, "disabled\n");
3441
3442 res = _braille_unregister_console(console);
3443 if (res < 0)
3444 return res;
3445 if (res > 0)
3446 return 0;
3447
3448 /* Disable it unconditionally */
3449 console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3450
3451 if (!console_is_registered_locked(console))
3452 return -ENODEV;
3453
3454 hlist_del_init_rcu(&console->node);
3455
3456 /*
3457 * <HISTORICAL>
3458 * If this isn't the last console and it has CON_CONSDEV set, we
3459 * need to set it on the next preferred console.
3460 * </HISTORICAL>
3461 *
3462 * The above makes no sense as there is no guarantee that the next
3463 * console has any device attached. Oh well....
3464 */
3465 if (!hlist_empty(&console_list) && console->flags & CON_CONSDEV)
3466 console_srcu_write_flags(console_first(), console_first()->flags | CON_CONSDEV);
3467
3468 /*
3469 * Ensure that all SRCU list walks have completed. All contexts
3470 * must not be able to see this console in the list so that any
3471 * exit/cleanup routines can be performed safely.
3472 */
3473 synchronize_srcu(&console_srcu);
3474
3475 console_sysfs_notify();
3476
3477 if (console->exit)
3478 res = console->exit(console);
3479
3480 return res;
3481}
3482
3483int unregister_console(struct console *console)
3484{
3485 int res;
3486
3487 console_list_lock();
3488 res = unregister_console_locked(console);
3489 console_list_unlock();
3490 return res;
3491}
3492EXPORT_SYMBOL(unregister_console);
3493
3494/**
3495 * console_force_preferred_locked - force a registered console preferred
3496 * @con: The registered console to force preferred.
3497 *
3498 * Must be called under console_list_lock().
3499 */
3500void console_force_preferred_locked(struct console *con)
3501{
3502 struct console *cur_pref_con;
3503
3504 if (!console_is_registered_locked(con))
3505 return;
3506
3507 cur_pref_con = console_first();
3508
3509 /* Already preferred? */
3510 if (cur_pref_con == con)
3511 return;
3512
3513 /*
3514 * Delete, but do not re-initialize the entry. This allows the console
3515 * to continue to appear registered (via any hlist_unhashed_lockless()
3516 * checks), even though it was briefly removed from the console list.
3517 */
3518 hlist_del_rcu(&con->node);
3519
3520 /*
3521 * Ensure that all SRCU list walks have completed so that the console
3522 * can be added to the beginning of the console list and its forward
3523 * list pointer can be re-initialized.
3524 */
3525 synchronize_srcu(&console_srcu);
3526
3527 con->flags |= CON_CONSDEV;
3528 WARN_ON(!con->device);
3529
3530 /* Only the new head can have CON_CONSDEV set. */
3531 console_srcu_write_flags(cur_pref_con, cur_pref_con->flags & ~CON_CONSDEV);
3532 hlist_add_head_rcu(&con->node, &console_list);
3533}
3534EXPORT_SYMBOL(console_force_preferred_locked);
3535
3536/*
3537 * Initialize the console device. This is called *early*, so
3538 * we can't necessarily depend on lots of kernel help here.
3539 * Just do some early initializations, and do the complex setup
3540 * later.
3541 */
3542void __init console_init(void)
3543{
3544 int ret;
3545 initcall_t call;
3546 initcall_entry_t *ce;
3547
3548 /* Setup the default TTY line discipline. */
3549 n_tty_init();
3550
3551 /*
3552 * set up the console device so that later boot sequences can
3553 * inform about problems etc..
3554 */
3555 ce = __con_initcall_start;
3556 trace_initcall_level("console");
3557 while (ce < __con_initcall_end) {
3558 call = initcall_from_entry(ce);
3559 trace_initcall_start(call);
3560 ret = call();
3561 trace_initcall_finish(call, ret);
3562 ce++;
3563 }
3564}
3565
3566/*
3567 * Some boot consoles access data that is in the init section and which will
3568 * be discarded after the initcalls have been run. To make sure that no code
3569 * will access this data, unregister the boot consoles in a late initcall.
3570 *
3571 * If for some reason, such as deferred probe or the driver being a loadable
3572 * module, the real console hasn't registered yet at this point, there will
3573 * be a brief interval in which no messages are logged to the console, which
3574 * makes it difficult to diagnose problems that occur during this time.
3575 *
3576 * To mitigate this problem somewhat, only unregister consoles whose memory
3577 * intersects with the init section. Note that all other boot consoles will
3578 * get unregistered when the real preferred console is registered.
3579 */
3580static int __init printk_late_init(void)
3581{
3582 struct hlist_node *tmp;
3583 struct console *con;
3584 int ret;
3585
3586 console_list_lock();
3587 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3588 if (!(con->flags & CON_BOOT))
3589 continue;
3590
3591 /* Check addresses that might be used for enabled consoles. */
3592 if (init_section_intersects(con, sizeof(*con)) ||
3593 init_section_contains(con->write, 0) ||
3594 init_section_contains(con->read, 0) ||
3595 init_section_contains(con->device, 0) ||
3596 init_section_contains(con->unblank, 0) ||
3597 init_section_contains(con->data, 0)) {
3598 /*
3599 * Please, consider moving the reported consoles out
3600 * of the init section.
3601 */
3602 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3603 con->name, con->index);
3604 unregister_console_locked(con);
3605 }
3606 }
3607 console_list_unlock();
3608
3609 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3610 console_cpu_notify);
3611 WARN_ON(ret < 0);
3612 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3613 console_cpu_notify, NULL);
3614 WARN_ON(ret < 0);
3615 printk_sysctl_init();
3616 return 0;
3617}
3618late_initcall(printk_late_init);
3619
3620#if defined CONFIG_PRINTK
3621/* If @con is specified, only wait for that console. Otherwise wait for all. */
3622static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
3623{
3624 int remaining = timeout_ms;
3625 struct console *c;
3626 u64 last_diff = 0;
3627 u64 printk_seq;
3628 int cookie;
3629 u64 diff;
3630 u64 seq;
3631
3632 might_sleep();
3633
3634 seq = prb_next_seq(prb);
3635
3636 for (;;) {
3637 diff = 0;
3638
3639 /*
3640 * Hold the console_lock to guarantee safe access to
3641 * console->seq and to prevent changes to @console_suspended
3642 * until all consoles have been processed.
3643 */
3644 console_lock();
3645
3646 cookie = console_srcu_read_lock();
3647 for_each_console_srcu(c) {
3648 if (con && con != c)
3649 continue;
3650 if (!console_is_usable(c))
3651 continue;
3652 printk_seq = c->seq;
3653 if (printk_seq < seq)
3654 diff += seq - printk_seq;
3655 }
3656 console_srcu_read_unlock(cookie);
3657
3658 /*
3659 * If consoles are suspended, it cannot be expected that they
3660 * make forward progress, so timeout immediately. @diff is
3661 * still used to return a valid flush status.
3662 */
3663 if (console_suspended)
3664 remaining = 0;
3665 else if (diff != last_diff && reset_on_progress)
3666 remaining = timeout_ms;
3667
3668 console_unlock();
3669
3670 if (diff == 0 || remaining == 0)
3671 break;
3672
3673 if (remaining < 0) {
3674 /* no timeout limit */
3675 msleep(100);
3676 } else if (remaining < 100) {
3677 msleep(remaining);
3678 remaining = 0;
3679 } else {
3680 msleep(100);
3681 remaining -= 100;
3682 }
3683
3684 last_diff = diff;
3685 }
3686
3687 return (diff == 0);
3688}
3689
3690/**
3691 * pr_flush() - Wait for printing threads to catch up.
3692 *
3693 * @timeout_ms: The maximum time (in ms) to wait.
3694 * @reset_on_progress: Reset the timeout if forward progress is seen.
3695 *
3696 * A value of 0 for @timeout_ms means no waiting will occur. A value of -1
3697 * represents infinite waiting.
3698 *
3699 * If @reset_on_progress is true, the timeout will be reset whenever any
3700 * printer has been seen to make some forward progress.
3701 *
3702 * Context: Process context. May sleep while acquiring console lock.
3703 * Return: true if all enabled printers are caught up.
3704 */
3705static bool pr_flush(int timeout_ms, bool reset_on_progress)
3706{
3707 return __pr_flush(NULL, timeout_ms, reset_on_progress);
3708}
3709
3710/*
3711 * Delayed printk version, for scheduler-internal messages:
3712 */
3713#define PRINTK_PENDING_WAKEUP 0x01
3714#define PRINTK_PENDING_OUTPUT 0x02
3715
3716static DEFINE_PER_CPU(int, printk_pending);
3717
3718static void wake_up_klogd_work_func(struct irq_work *irq_work)
3719{
3720 int pending = this_cpu_xchg(printk_pending, 0);
3721
3722 if (pending & PRINTK_PENDING_OUTPUT) {
3723 /* If trylock fails, someone else is doing the printing */
3724 if (console_trylock())
3725 console_unlock();
3726 }
3727
3728 if (pending & PRINTK_PENDING_WAKEUP)
3729 wake_up_interruptible(&log_wait);
3730}
3731
3732static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3733 IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3734
3735static void __wake_up_klogd(int val)
3736{
3737 if (!printk_percpu_data_ready())
3738 return;
3739
3740 preempt_disable();
3741 /*
3742 * Guarantee any new records can be seen by tasks preparing to wait
3743 * before this context checks if the wait queue is empty.
3744 *
3745 * The full memory barrier within wq_has_sleeper() pairs with the full
3746 * memory barrier within set_current_state() of
3747 * prepare_to_wait_event(), which is called after ___wait_event() adds
3748 * the waiter but before it has checked the wait condition.
3749 *
3750 * This pairs with devkmsg_read:A and syslog_print:A.
3751 */
3752 if (wq_has_sleeper(&log_wait) || /* LMM(__wake_up_klogd:A) */
3753 (val & PRINTK_PENDING_OUTPUT)) {
3754 this_cpu_or(printk_pending, val);
3755 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3756 }
3757 preempt_enable();
3758}
3759
3760void wake_up_klogd(void)
3761{
3762 __wake_up_klogd(PRINTK_PENDING_WAKEUP);
3763}
3764
3765void defer_console_output(void)
3766{
3767 /*
3768 * New messages may have been added directly to the ringbuffer
3769 * using vprintk_store(), so wake any waiters as well.
3770 */
3771 __wake_up_klogd(PRINTK_PENDING_WAKEUP | PRINTK_PENDING_OUTPUT);
3772}
3773
3774void printk_trigger_flush(void)
3775{
3776 defer_console_output();
3777}
3778
3779int vprintk_deferred(const char *fmt, va_list args)
3780{
3781 int r;
3782
3783 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3784 defer_console_output();
3785
3786 return r;
3787}
3788
3789int _printk_deferred(const char *fmt, ...)
3790{
3791 va_list args;
3792 int r;
3793
3794 va_start(args, fmt);
3795 r = vprintk_deferred(fmt, args);
3796 va_end(args);
3797
3798 return r;
3799}
3800
3801/*
3802 * printk rate limiting, lifted from the networking subsystem.
3803 *
3804 * This enforces a rate limit: not more than 10 kernel messages
3805 * every 5s to make a denial-of-service attack impossible.
3806 */
3807DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3808
3809int __printk_ratelimit(const char *func)
3810{
3811 return ___ratelimit(&printk_ratelimit_state, func);
3812}
3813EXPORT_SYMBOL(__printk_ratelimit);
3814
3815/**
3816 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3817 * @caller_jiffies: pointer to caller's state
3818 * @interval_msecs: minimum interval between prints
3819 *
3820 * printk_timed_ratelimit() returns true if more than @interval_msecs
3821 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3822 * returned true.
3823 */
3824bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3825 unsigned int interval_msecs)
3826{
3827 unsigned long elapsed = jiffies - *caller_jiffies;
3828
3829 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3830 return false;
3831
3832 *caller_jiffies = jiffies;
3833 return true;
3834}
3835EXPORT_SYMBOL(printk_timed_ratelimit);
3836
3837static DEFINE_SPINLOCK(dump_list_lock);
3838static LIST_HEAD(dump_list);
3839
3840/**
3841 * kmsg_dump_register - register a kernel log dumper.
3842 * @dumper: pointer to the kmsg_dumper structure
3843 *
3844 * Adds a kernel log dumper to the system. The dump callback in the
3845 * structure will be called when the kernel oopses or panics and must be
3846 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3847 */
3848int kmsg_dump_register(struct kmsg_dumper *dumper)
3849{
3850 unsigned long flags;
3851 int err = -EBUSY;
3852
3853 /* The dump callback needs to be set */
3854 if (!dumper->dump)
3855 return -EINVAL;
3856
3857 spin_lock_irqsave(&dump_list_lock, flags);
3858 /* Don't allow registering multiple times */
3859 if (!dumper->registered) {
3860 dumper->registered = 1;
3861 list_add_tail_rcu(&dumper->list, &dump_list);
3862 err = 0;
3863 }
3864 spin_unlock_irqrestore(&dump_list_lock, flags);
3865
3866 return err;
3867}
3868EXPORT_SYMBOL_GPL(kmsg_dump_register);
3869
3870/**
3871 * kmsg_dump_unregister - unregister a kmsg dumper.
3872 * @dumper: pointer to the kmsg_dumper structure
3873 *
3874 * Removes a dump device from the system. Returns zero on success and
3875 * %-EINVAL otherwise.
3876 */
3877int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3878{
3879 unsigned long flags;
3880 int err = -EINVAL;
3881
3882 spin_lock_irqsave(&dump_list_lock, flags);
3883 if (dumper->registered) {
3884 dumper->registered = 0;
3885 list_del_rcu(&dumper->list);
3886 err = 0;
3887 }
3888 spin_unlock_irqrestore(&dump_list_lock, flags);
3889 synchronize_rcu();
3890
3891 return err;
3892}
3893EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3894
3895static bool always_kmsg_dump;
3896module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3897
3898const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
3899{
3900 switch (reason) {
3901 case KMSG_DUMP_PANIC:
3902 return "Panic";
3903 case KMSG_DUMP_OOPS:
3904 return "Oops";
3905 case KMSG_DUMP_EMERG:
3906 return "Emergency";
3907 case KMSG_DUMP_SHUTDOWN:
3908 return "Shutdown";
3909 default:
3910 return "Unknown";
3911 }
3912}
3913EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
3914
3915/**
3916 * kmsg_dump - dump kernel log to kernel message dumpers.
3917 * @reason: the reason (oops, panic etc) for dumping
3918 *
3919 * Call each of the registered dumper's dump() callback, which can
3920 * retrieve the kmsg records with kmsg_dump_get_line() or
3921 * kmsg_dump_get_buffer().
3922 */
3923void kmsg_dump(enum kmsg_dump_reason reason)
3924{
3925 struct kmsg_dumper *dumper;
3926
3927 rcu_read_lock();
3928 list_for_each_entry_rcu(dumper, &dump_list, list) {
3929 enum kmsg_dump_reason max_reason = dumper->max_reason;
3930
3931 /*
3932 * If client has not provided a specific max_reason, default
3933 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
3934 */
3935 if (max_reason == KMSG_DUMP_UNDEF) {
3936 max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
3937 KMSG_DUMP_OOPS;
3938 }
3939 if (reason > max_reason)
3940 continue;
3941
3942 /* invoke dumper which will iterate over records */
3943 dumper->dump(dumper, reason);
3944 }
3945 rcu_read_unlock();
3946}
3947
3948/**
3949 * kmsg_dump_get_line - retrieve one kmsg log line
3950 * @iter: kmsg dump iterator
3951 * @syslog: include the "<4>" prefixes
3952 * @line: buffer to copy the line to
3953 * @size: maximum size of the buffer
3954 * @len: length of line placed into buffer
3955 *
3956 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3957 * record, and copy one record into the provided buffer.
3958 *
3959 * Consecutive calls will return the next available record moving
3960 * towards the end of the buffer with the youngest messages.
3961 *
3962 * A return value of FALSE indicates that there are no more records to
3963 * read.
3964 */
3965bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
3966 char *line, size_t size, size_t *len)
3967{
3968 u64 min_seq = latched_seq_read_nolock(&clear_seq);
3969 struct printk_info info;
3970 unsigned int line_count;
3971 struct printk_record r;
3972 size_t l = 0;
3973 bool ret = false;
3974
3975 if (iter->cur_seq < min_seq)
3976 iter->cur_seq = min_seq;
3977
3978 prb_rec_init_rd(&r, &info, line, size);
3979
3980 /* Read text or count text lines? */
3981 if (line) {
3982 if (!prb_read_valid(prb, iter->cur_seq, &r))
3983 goto out;
3984 l = record_print_text(&r, syslog, printk_time);
3985 } else {
3986 if (!prb_read_valid_info(prb, iter->cur_seq,
3987 &info, &line_count)) {
3988 goto out;
3989 }
3990 l = get_record_print_text_size(&info, line_count, syslog,
3991 printk_time);
3992
3993 }
3994
3995 iter->cur_seq = r.info->seq + 1;
3996 ret = true;
3997out:
3998 if (len)
3999 *len = l;
4000 return ret;
4001}
4002EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
4003
4004/**
4005 * kmsg_dump_get_buffer - copy kmsg log lines
4006 * @iter: kmsg dump iterator
4007 * @syslog: include the "<4>" prefixes
4008 * @buf: buffer to copy the line to
4009 * @size: maximum size of the buffer
4010 * @len_out: length of line placed into buffer
4011 *
4012 * Start at the end of the kmsg buffer and fill the provided buffer
4013 * with as many of the *youngest* kmsg records that fit into it.
4014 * If the buffer is large enough, all available kmsg records will be
4015 * copied with a single call.
4016 *
4017 * Consecutive calls will fill the buffer with the next block of
4018 * available older records, not including the earlier retrieved ones.
4019 *
4020 * A return value of FALSE indicates that there are no more records to
4021 * read.
4022 */
4023bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
4024 char *buf, size_t size, size_t *len_out)
4025{
4026 u64 min_seq = latched_seq_read_nolock(&clear_seq);
4027 struct printk_info info;
4028 struct printk_record r;
4029 u64 seq;
4030 u64 next_seq;
4031 size_t len = 0;
4032 bool ret = false;
4033 bool time = printk_time;
4034
4035 if (!buf || !size)
4036 goto out;
4037
4038 if (iter->cur_seq < min_seq)
4039 iter->cur_seq = min_seq;
4040
4041 if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
4042 if (info.seq != iter->cur_seq) {
4043 /* messages are gone, move to first available one */
4044 iter->cur_seq = info.seq;
4045 }
4046 }
4047
4048 /* last entry */
4049 if (iter->cur_seq >= iter->next_seq)
4050 goto out;
4051
4052 /*
4053 * Find first record that fits, including all following records,
4054 * into the user-provided buffer for this dump. Pass in size-1
4055 * because this function (by way of record_print_text()) will
4056 * not write more than size-1 bytes of text into @buf.
4057 */
4058 seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
4059 size - 1, syslog, time);
4060
4061 /*
4062 * Next kmsg_dump_get_buffer() invocation will dump block of
4063 * older records stored right before this one.
4064 */
4065 next_seq = seq;
4066
4067 prb_rec_init_rd(&r, &info, buf, size);
4068
4069 len = 0;
4070 prb_for_each_record(seq, prb, seq, &r) {
4071 if (r.info->seq >= iter->next_seq)
4072 break;
4073
4074 len += record_print_text(&r, syslog, time);
4075
4076 /* Adjust record to store to remaining buffer space. */
4077 prb_rec_init_rd(&r, &info, buf + len, size - len);
4078 }
4079
4080 iter->next_seq = next_seq;
4081 ret = true;
4082out:
4083 if (len_out)
4084 *len_out = len;
4085 return ret;
4086}
4087EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
4088
4089/**
4090 * kmsg_dump_rewind - reset the iterator
4091 * @iter: kmsg dump iterator
4092 *
4093 * Reset the dumper's iterator so that kmsg_dump_get_line() and
4094 * kmsg_dump_get_buffer() can be called again and used multiple
4095 * times within the same dumper.dump() callback.
4096 */
4097void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
4098{
4099 iter->cur_seq = latched_seq_read_nolock(&clear_seq);
4100 iter->next_seq = prb_next_seq(prb);
4101}
4102EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
4103
4104#endif
4105
4106#ifdef CONFIG_SMP
4107static atomic_t printk_cpu_sync_owner = ATOMIC_INIT(-1);
4108static atomic_t printk_cpu_sync_nested = ATOMIC_INIT(0);
4109
4110/**
4111 * __printk_cpu_sync_wait() - Busy wait until the printk cpu-reentrant
4112 * spinning lock is not owned by any CPU.
4113 *
4114 * Context: Any context.
4115 */
4116void __printk_cpu_sync_wait(void)
4117{
4118 do {
4119 cpu_relax();
4120 } while (atomic_read(&printk_cpu_sync_owner) != -1);
4121}
4122EXPORT_SYMBOL(__printk_cpu_sync_wait);
4123
4124/**
4125 * __printk_cpu_sync_try_get() - Try to acquire the printk cpu-reentrant
4126 * spinning lock.
4127 *
4128 * If no processor has the lock, the calling processor takes the lock and
4129 * becomes the owner. If the calling processor is already the owner of the
4130 * lock, this function succeeds immediately.
4131 *
4132 * Context: Any context. Expects interrupts to be disabled.
4133 * Return: 1 on success, otherwise 0.
4134 */
4135int __printk_cpu_sync_try_get(void)
4136{
4137 int cpu;
4138 int old;
4139
4140 cpu = smp_processor_id();
4141
4142 /*
4143 * Guarantee loads and stores from this CPU when it is the lock owner
4144 * are _not_ visible to the previous lock owner. This pairs with
4145 * __printk_cpu_sync_put:B.
4146 *
4147 * Memory barrier involvement:
4148 *
4149 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4150 * then __printk_cpu_sync_put:A can never read from
4151 * __printk_cpu_sync_try_get:B.
4152 *
4153 * Relies on:
4154 *
4155 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4156 * of the previous CPU
4157 * matching
4158 * ACQUIRE from __printk_cpu_sync_try_get:A to
4159 * __printk_cpu_sync_try_get:B of this CPU
4160 */
4161 old = atomic_cmpxchg_acquire(&printk_cpu_sync_owner, -1,
4162 cpu); /* LMM(__printk_cpu_sync_try_get:A) */
4163 if (old == -1) {
4164 /*
4165 * This CPU is now the owner and begins loading/storing
4166 * data: LMM(__printk_cpu_sync_try_get:B)
4167 */
4168 return 1;
4169
4170 } else if (old == cpu) {
4171 /* This CPU is already the owner. */
4172 atomic_inc(&printk_cpu_sync_nested);
4173 return 1;
4174 }
4175
4176 return 0;
4177}
4178EXPORT_SYMBOL(__printk_cpu_sync_try_get);
4179
4180/**
4181 * __printk_cpu_sync_put() - Release the printk cpu-reentrant spinning lock.
4182 *
4183 * The calling processor must be the owner of the lock.
4184 *
4185 * Context: Any context. Expects interrupts to be disabled.
4186 */
4187void __printk_cpu_sync_put(void)
4188{
4189 if (atomic_read(&printk_cpu_sync_nested)) {
4190 atomic_dec(&printk_cpu_sync_nested);
4191 return;
4192 }
4193
4194 /*
4195 * This CPU is finished loading/storing data:
4196 * LMM(__printk_cpu_sync_put:A)
4197 */
4198
4199 /*
4200 * Guarantee loads and stores from this CPU when it was the
4201 * lock owner are visible to the next lock owner. This pairs
4202 * with __printk_cpu_sync_try_get:A.
4203 *
4204 * Memory barrier involvement:
4205 *
4206 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4207 * then __printk_cpu_sync_try_get:B reads from __printk_cpu_sync_put:A.
4208 *
4209 * Relies on:
4210 *
4211 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4212 * of this CPU
4213 * matching
4214 * ACQUIRE from __printk_cpu_sync_try_get:A to
4215 * __printk_cpu_sync_try_get:B of the next CPU
4216 */
4217 atomic_set_release(&printk_cpu_sync_owner,
4218 -1); /* LMM(__printk_cpu_sync_put:B) */
4219}
4220EXPORT_SYMBOL(__printk_cpu_sync_put);
4221#endif /* CONFIG_SMP */