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/syscore_ops.h>
38#include <linux/vmcore_info.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 "printk_ringbuffer.h"
60#include "console_cmdline.h"
61#include "braille.h"
62#include "internal.h"
63
64int console_printk[4] = {
65 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
66 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
67 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
68 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
69};
70EXPORT_SYMBOL_GPL(console_printk);
71
72atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
73EXPORT_SYMBOL(ignore_console_lock_warning);
74
75EXPORT_TRACEPOINT_SYMBOL_GPL(console);
76
77/*
78 * Low level drivers may need that to know if they can schedule in
79 * their unblank() callback or not. So let's export it.
80 */
81int oops_in_progress;
82EXPORT_SYMBOL(oops_in_progress);
83
84/*
85 * console_mutex protects console_list updates and console->flags updates.
86 * The flags are synchronized only for consoles that are registered, i.e.
87 * accessible via the console list.
88 */
89static DEFINE_MUTEX(console_mutex);
90
91/*
92 * console_sem protects updates to console->seq
93 * and also provides serialization for console printing.
94 */
95static DEFINE_SEMAPHORE(console_sem, 1);
96HLIST_HEAD(console_list);
97EXPORT_SYMBOL_GPL(console_list);
98DEFINE_STATIC_SRCU(console_srcu);
99
100/*
101 * System may need to suppress printk message under certain
102 * circumstances, like after kernel panic happens.
103 */
104int __read_mostly suppress_printk;
105
106#ifdef CONFIG_LOCKDEP
107static struct lockdep_map console_lock_dep_map = {
108 .name = "console_lock"
109};
110
111void lockdep_assert_console_list_lock_held(void)
112{
113 lockdep_assert_held(&console_mutex);
114}
115EXPORT_SYMBOL(lockdep_assert_console_list_lock_held);
116#endif
117
118#ifdef CONFIG_DEBUG_LOCK_ALLOC
119bool console_srcu_read_lock_is_held(void)
120{
121 return srcu_read_lock_held(&console_srcu);
122}
123EXPORT_SYMBOL(console_srcu_read_lock_is_held);
124#endif
125
126enum devkmsg_log_bits {
127 __DEVKMSG_LOG_BIT_ON = 0,
128 __DEVKMSG_LOG_BIT_OFF,
129 __DEVKMSG_LOG_BIT_LOCK,
130};
131
132enum devkmsg_log_masks {
133 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
134 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
135 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
136};
137
138/* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
139#define DEVKMSG_LOG_MASK_DEFAULT 0
140
141static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
142
143static int __control_devkmsg(char *str)
144{
145 size_t len;
146
147 if (!str)
148 return -EINVAL;
149
150 len = str_has_prefix(str, "on");
151 if (len) {
152 devkmsg_log = DEVKMSG_LOG_MASK_ON;
153 return len;
154 }
155
156 len = str_has_prefix(str, "off");
157 if (len) {
158 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
159 return len;
160 }
161
162 len = str_has_prefix(str, "ratelimit");
163 if (len) {
164 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
165 return len;
166 }
167
168 return -EINVAL;
169}
170
171static int __init control_devkmsg(char *str)
172{
173 if (__control_devkmsg(str) < 0) {
174 pr_warn("printk.devkmsg: bad option string '%s'\n", str);
175 return 1;
176 }
177
178 /*
179 * Set sysctl string accordingly:
180 */
181 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
182 strscpy(devkmsg_log_str, "on");
183 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
184 strscpy(devkmsg_log_str, "off");
185 /* else "ratelimit" which is set by default. */
186
187 /*
188 * Sysctl cannot change it anymore. The kernel command line setting of
189 * this parameter is to force the setting to be permanent throughout the
190 * runtime of the system. This is a precation measure against userspace
191 * trying to be a smarta** and attempting to change it up on us.
192 */
193 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
194
195 return 1;
196}
197__setup("printk.devkmsg=", control_devkmsg);
198
199char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
200#if defined(CONFIG_PRINTK) && defined(CONFIG_SYSCTL)
201int devkmsg_sysctl_set_loglvl(const struct ctl_table *table, int write,
202 void *buffer, size_t *lenp, loff_t *ppos)
203{
204 char old_str[DEVKMSG_STR_MAX_SIZE];
205 unsigned int old;
206 int err;
207
208 if (write) {
209 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
210 return -EINVAL;
211
212 old = devkmsg_log;
213 strscpy(old_str, devkmsg_log_str);
214 }
215
216 err = proc_dostring(table, write, buffer, lenp, ppos);
217 if (err)
218 return err;
219
220 if (write) {
221 err = __control_devkmsg(devkmsg_log_str);
222
223 /*
224 * Do not accept an unknown string OR a known string with
225 * trailing crap...
226 */
227 if (err < 0 || (err + 1 != *lenp)) {
228
229 /* ... and restore old setting. */
230 devkmsg_log = old;
231 strscpy(devkmsg_log_str, old_str);
232
233 return -EINVAL;
234 }
235 }
236
237 return 0;
238}
239#endif /* CONFIG_PRINTK && CONFIG_SYSCTL */
240
241/**
242 * console_list_lock - Lock the console list
243 *
244 * For console list or console->flags updates
245 */
246void console_list_lock(void)
247{
248 /*
249 * In unregister_console() and console_force_preferred_locked(),
250 * synchronize_srcu() is called with the console_list_lock held.
251 * Therefore it is not allowed that the console_list_lock is taken
252 * with the srcu_lock held.
253 *
254 * Detecting if this context is really in the read-side critical
255 * section is only possible if the appropriate debug options are
256 * enabled.
257 */
258 WARN_ON_ONCE(debug_lockdep_rcu_enabled() &&
259 srcu_read_lock_held(&console_srcu));
260
261 mutex_lock(&console_mutex);
262}
263EXPORT_SYMBOL(console_list_lock);
264
265/**
266 * console_list_unlock - Unlock the console list
267 *
268 * Counterpart to console_list_lock()
269 */
270void console_list_unlock(void)
271{
272 mutex_unlock(&console_mutex);
273}
274EXPORT_SYMBOL(console_list_unlock);
275
276/**
277 * console_srcu_read_lock - Register a new reader for the
278 * SRCU-protected console list
279 *
280 * Use for_each_console_srcu() to iterate the console list
281 *
282 * Context: Any context.
283 * Return: A cookie to pass to console_srcu_read_unlock().
284 */
285int console_srcu_read_lock(void)
286 __acquires(&console_srcu)
287{
288 return srcu_read_lock_nmisafe(&console_srcu);
289}
290EXPORT_SYMBOL(console_srcu_read_lock);
291
292/**
293 * console_srcu_read_unlock - Unregister an old reader from
294 * the SRCU-protected console list
295 * @cookie: cookie returned from console_srcu_read_lock()
296 *
297 * Counterpart to console_srcu_read_lock()
298 */
299void console_srcu_read_unlock(int cookie)
300 __releases(&console_srcu)
301{
302 srcu_read_unlock_nmisafe(&console_srcu, cookie);
303}
304EXPORT_SYMBOL(console_srcu_read_unlock);
305
306/*
307 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
308 * macros instead of functions so that _RET_IP_ contains useful information.
309 */
310#define down_console_sem() do { \
311 down(&console_sem);\
312 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
313} while (0)
314
315static int __down_trylock_console_sem(unsigned long ip)
316{
317 int lock_failed;
318 unsigned long flags;
319
320 /*
321 * Here and in __up_console_sem() we need to be in safe mode,
322 * because spindump/WARN/etc from under console ->lock will
323 * deadlock in printk()->down_trylock_console_sem() otherwise.
324 */
325 printk_safe_enter_irqsave(flags);
326 lock_failed = down_trylock(&console_sem);
327 printk_safe_exit_irqrestore(flags);
328
329 if (lock_failed)
330 return 1;
331 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
332 return 0;
333}
334#define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
335
336static void __up_console_sem(unsigned long ip)
337{
338 unsigned long flags;
339
340 mutex_release(&console_lock_dep_map, ip);
341
342 printk_safe_enter_irqsave(flags);
343 up(&console_sem);
344 printk_safe_exit_irqrestore(flags);
345}
346#define up_console_sem() __up_console_sem(_RET_IP_)
347
348static bool panic_in_progress(void)
349{
350 return unlikely(atomic_read(&panic_cpu) != PANIC_CPU_INVALID);
351}
352
353/* Return true if a panic is in progress on the current CPU. */
354bool this_cpu_in_panic(void)
355{
356 /*
357 * We can use raw_smp_processor_id() here because it is impossible for
358 * the task to be migrated to the panic_cpu, or away from it. If
359 * panic_cpu has already been set, and we're not currently executing on
360 * that CPU, then we never will be.
361 */
362 return unlikely(atomic_read(&panic_cpu) == raw_smp_processor_id());
363}
364
365/*
366 * Return true if a panic is in progress on a remote CPU.
367 *
368 * On true, the local CPU should immediately release any printing resources
369 * that may be needed by the panic CPU.
370 */
371bool other_cpu_in_panic(void)
372{
373 return (panic_in_progress() && !this_cpu_in_panic());
374}
375
376/*
377 * This is used for debugging the mess that is the VT code by
378 * keeping track if we have the console semaphore held. It's
379 * definitely not the perfect debug tool (we don't know if _WE_
380 * hold it and are racing, but it helps tracking those weird code
381 * paths in the console code where we end up in places I want
382 * locked without the console semaphore held).
383 */
384static int console_locked;
385
386/*
387 * Array of consoles built from command line options (console=)
388 */
389
390#define MAX_CMDLINECONSOLES 8
391
392static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
393
394static int preferred_console = -1;
395int console_set_on_cmdline;
396EXPORT_SYMBOL(console_set_on_cmdline);
397
398/* Flag: console code may call schedule() */
399static int console_may_schedule;
400
401enum con_msg_format_flags {
402 MSG_FORMAT_DEFAULT = 0,
403 MSG_FORMAT_SYSLOG = (1 << 0),
404};
405
406static int console_msg_format = MSG_FORMAT_DEFAULT;
407
408/*
409 * The printk log buffer consists of a sequenced collection of records, each
410 * containing variable length message text. Every record also contains its
411 * own meta-data (@info).
412 *
413 * Every record meta-data carries the timestamp in microseconds, as well as
414 * the standard userspace syslog level and syslog facility. The usual kernel
415 * messages use LOG_KERN; userspace-injected messages always carry a matching
416 * syslog facility, by default LOG_USER. The origin of every message can be
417 * reliably determined that way.
418 *
419 * The human readable log message of a record is available in @text, the
420 * length of the message text in @text_len. The stored message is not
421 * terminated.
422 *
423 * Optionally, a record can carry a dictionary of properties (key/value
424 * pairs), to provide userspace with a machine-readable message context.
425 *
426 * Examples for well-defined, commonly used property names are:
427 * DEVICE=b12:8 device identifier
428 * b12:8 block dev_t
429 * c127:3 char dev_t
430 * n8 netdev ifindex
431 * +sound:card0 subsystem:devname
432 * SUBSYSTEM=pci driver-core subsystem name
433 *
434 * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
435 * and values are terminated by a '\0' character.
436 *
437 * Example of record values:
438 * record.text_buf = "it's a line" (unterminated)
439 * record.info.seq = 56
440 * record.info.ts_nsec = 36863
441 * record.info.text_len = 11
442 * record.info.facility = 0 (LOG_KERN)
443 * record.info.flags = 0
444 * record.info.level = 3 (LOG_ERR)
445 * record.info.caller_id = 299 (task 299)
446 * record.info.dev_info.subsystem = "pci" (terminated)
447 * record.info.dev_info.device = "+pci:0000:00:01.0" (terminated)
448 *
449 * The 'struct printk_info' buffer must never be directly exported to
450 * userspace, it is a kernel-private implementation detail that might
451 * need to be changed in the future, when the requirements change.
452 *
453 * /dev/kmsg exports the structured data in the following line format:
454 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
455 *
456 * Users of the export format should ignore possible additional values
457 * separated by ',', and find the message after the ';' character.
458 *
459 * The optional key/value pairs are attached as continuation lines starting
460 * with a space character and terminated by a newline. All possible
461 * non-prinatable characters are escaped in the "\xff" notation.
462 */
463
464/* syslog_lock protects syslog_* variables and write access to clear_seq. */
465static DEFINE_MUTEX(syslog_lock);
466
467/*
468 * Specifies if a legacy console is registered. If legacy consoles are
469 * present, it is necessary to perform the console lock/unlock dance
470 * whenever console flushing should occur.
471 */
472bool have_legacy_console;
473
474/*
475 * Specifies if an nbcon console is registered. If nbcon consoles are present,
476 * synchronous printing of legacy consoles will not occur during panic until
477 * the backtrace has been stored to the ringbuffer.
478 */
479bool have_nbcon_console;
480
481/*
482 * Specifies if a boot console is registered. If boot consoles are present,
483 * nbcon consoles cannot print simultaneously and must be synchronized by
484 * the console lock. This is because boot consoles and nbcon consoles may
485 * have mapped the same hardware.
486 */
487bool have_boot_console;
488
489/* See printk_legacy_allow_panic_sync() for details. */
490bool legacy_allow_panic_sync;
491
492#ifdef CONFIG_PRINTK
493DECLARE_WAIT_QUEUE_HEAD(log_wait);
494static DECLARE_WAIT_QUEUE_HEAD(legacy_wait);
495/* All 3 protected by @syslog_lock. */
496/* the next printk record to read by syslog(READ) or /proc/kmsg */
497static u64 syslog_seq;
498static size_t syslog_partial;
499static bool syslog_time;
500
501/* True when _all_ printer threads are available for printing. */
502bool printk_kthreads_running;
503
504struct latched_seq {
505 seqcount_latch_t latch;
506 u64 val[2];
507};
508
509/*
510 * The next printk record to read after the last 'clear' command. There are
511 * two copies (updated with seqcount_latch) so that reads can locklessly
512 * access a valid value. Writers are synchronized by @syslog_lock.
513 */
514static struct latched_seq clear_seq = {
515 .latch = SEQCNT_LATCH_ZERO(clear_seq.latch),
516 .val[0] = 0,
517 .val[1] = 0,
518};
519
520#define LOG_LEVEL(v) ((v) & 0x07)
521#define LOG_FACILITY(v) ((v) >> 3 & 0xff)
522
523/* record buffer */
524#define LOG_ALIGN __alignof__(unsigned long)
525#define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
526#define LOG_BUF_LEN_MAX ((u32)1 << 31)
527static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
528static char *log_buf = __log_buf;
529static u32 log_buf_len = __LOG_BUF_LEN;
530
531/*
532 * Define the average message size. This only affects the number of
533 * descriptors that will be available. Underestimating is better than
534 * overestimating (too many available descriptors is better than not enough).
535 */
536#define PRB_AVGBITS 5 /* 32 character average length */
537
538#if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
539#error CONFIG_LOG_BUF_SHIFT value too small.
540#endif
541_DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
542 PRB_AVGBITS, &__log_buf[0]);
543
544static struct printk_ringbuffer printk_rb_dynamic;
545
546struct printk_ringbuffer *prb = &printk_rb_static;
547
548/*
549 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
550 * per_cpu_areas are initialised. This variable is set to true when
551 * it's safe to access per-CPU data.
552 */
553static bool __printk_percpu_data_ready __ro_after_init;
554
555bool printk_percpu_data_ready(void)
556{
557 return __printk_percpu_data_ready;
558}
559
560/* Must be called under syslog_lock. */
561static void latched_seq_write(struct latched_seq *ls, u64 val)
562{
563 write_seqcount_latch_begin(&ls->latch);
564 ls->val[0] = val;
565 write_seqcount_latch(&ls->latch);
566 ls->val[1] = val;
567 write_seqcount_latch_end(&ls->latch);
568}
569
570/* Can be called from any context. */
571static u64 latched_seq_read_nolock(struct latched_seq *ls)
572{
573 unsigned int seq;
574 unsigned int idx;
575 u64 val;
576
577 do {
578 seq = read_seqcount_latch(&ls->latch);
579 idx = seq & 0x1;
580 val = ls->val[idx];
581 } while (read_seqcount_latch_retry(&ls->latch, seq));
582
583 return val;
584}
585
586/* Return log buffer address */
587char *log_buf_addr_get(void)
588{
589 return log_buf;
590}
591
592/* Return log buffer size */
593u32 log_buf_len_get(void)
594{
595 return log_buf_len;
596}
597
598/*
599 * Define how much of the log buffer we could take at maximum. The value
600 * must be greater than two. Note that only half of the buffer is available
601 * when the index points to the middle.
602 */
603#define MAX_LOG_TAKE_PART 4
604static const char trunc_msg[] = "<truncated>";
605
606static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
607{
608 /*
609 * The message should not take the whole buffer. Otherwise, it might
610 * get removed too soon.
611 */
612 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
613
614 if (*text_len > max_text_len)
615 *text_len = max_text_len;
616
617 /* enable the warning message (if there is room) */
618 *trunc_msg_len = strlen(trunc_msg);
619 if (*text_len >= *trunc_msg_len)
620 *text_len -= *trunc_msg_len;
621 else
622 *trunc_msg_len = 0;
623}
624
625int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
626
627static int syslog_action_restricted(int type)
628{
629 if (dmesg_restrict)
630 return 1;
631 /*
632 * Unless restricted, we allow "read all" and "get buffer size"
633 * for everybody.
634 */
635 return type != SYSLOG_ACTION_READ_ALL &&
636 type != SYSLOG_ACTION_SIZE_BUFFER;
637}
638
639static int check_syslog_permissions(int type, int source)
640{
641 /*
642 * If this is from /proc/kmsg and we've already opened it, then we've
643 * already done the capabilities checks at open time.
644 */
645 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
646 goto ok;
647
648 if (syslog_action_restricted(type)) {
649 if (capable(CAP_SYSLOG))
650 goto ok;
651 return -EPERM;
652 }
653ok:
654 return security_syslog(type);
655}
656
657static void append_char(char **pp, char *e, char c)
658{
659 if (*pp < e)
660 *(*pp)++ = c;
661}
662
663static ssize_t info_print_ext_header(char *buf, size_t size,
664 struct printk_info *info)
665{
666 u64 ts_usec = info->ts_nsec;
667 char caller[20];
668#ifdef CONFIG_PRINTK_CALLER
669 u32 id = info->caller_id;
670
671 snprintf(caller, sizeof(caller), ",caller=%c%u",
672 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
673#else
674 caller[0] = '\0';
675#endif
676
677 do_div(ts_usec, 1000);
678
679 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
680 (info->facility << 3) | info->level, info->seq,
681 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
682}
683
684static ssize_t msg_add_ext_text(char *buf, size_t size,
685 const char *text, size_t text_len,
686 unsigned char endc)
687{
688 char *p = buf, *e = buf + size;
689 size_t i;
690
691 /* escape non-printable characters */
692 for (i = 0; i < text_len; i++) {
693 unsigned char c = text[i];
694
695 if (c < ' ' || c >= 127 || c == '\\')
696 p += scnprintf(p, e - p, "\\x%02x", c);
697 else
698 append_char(&p, e, c);
699 }
700 append_char(&p, e, endc);
701
702 return p - buf;
703}
704
705static ssize_t msg_add_dict_text(char *buf, size_t size,
706 const char *key, const char *val)
707{
708 size_t val_len = strlen(val);
709 ssize_t len;
710
711 if (!val_len)
712 return 0;
713
714 len = msg_add_ext_text(buf, size, "", 0, ' '); /* dict prefix */
715 len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
716 len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
717
718 return len;
719}
720
721static ssize_t msg_print_ext_body(char *buf, size_t size,
722 char *text, size_t text_len,
723 struct dev_printk_info *dev_info)
724{
725 ssize_t len;
726
727 len = msg_add_ext_text(buf, size, text, text_len, '\n');
728
729 if (!dev_info)
730 goto out;
731
732 len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
733 dev_info->subsystem);
734 len += msg_add_dict_text(buf + len, size - len, "DEVICE",
735 dev_info->device);
736out:
737 return len;
738}
739
740/* /dev/kmsg - userspace message inject/listen interface */
741struct devkmsg_user {
742 atomic64_t seq;
743 struct ratelimit_state rs;
744 struct mutex lock;
745 struct printk_buffers pbufs;
746};
747
748static __printf(3, 4) __cold
749int devkmsg_emit(int facility, int level, const char *fmt, ...)
750{
751 va_list args;
752 int r;
753
754 va_start(args, fmt);
755 r = vprintk_emit(facility, level, NULL, fmt, args);
756 va_end(args);
757
758 return r;
759}
760
761static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
762{
763 char *buf, *line;
764 int level = default_message_loglevel;
765 int facility = 1; /* LOG_USER */
766 struct file *file = iocb->ki_filp;
767 struct devkmsg_user *user = file->private_data;
768 size_t len = iov_iter_count(from);
769 ssize_t ret = len;
770
771 if (len > PRINTKRB_RECORD_MAX)
772 return -EINVAL;
773
774 /* Ignore when user logging is disabled. */
775 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
776 return len;
777
778 /* Ratelimit when not explicitly enabled. */
779 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
780 if (!___ratelimit(&user->rs, current->comm))
781 return ret;
782 }
783
784 buf = kmalloc(len+1, GFP_KERNEL);
785 if (buf == NULL)
786 return -ENOMEM;
787
788 buf[len] = '\0';
789 if (!copy_from_iter_full(buf, len, from)) {
790 kfree(buf);
791 return -EFAULT;
792 }
793
794 /*
795 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
796 * the decimal value represents 32bit, the lower 3 bit are the log
797 * level, the rest are the log facility.
798 *
799 * If no prefix or no userspace facility is specified, we
800 * enforce LOG_USER, to be able to reliably distinguish
801 * kernel-generated messages from userspace-injected ones.
802 */
803 line = buf;
804 if (line[0] == '<') {
805 char *endp = NULL;
806 unsigned int u;
807
808 u = simple_strtoul(line + 1, &endp, 10);
809 if (endp && endp[0] == '>') {
810 level = LOG_LEVEL(u);
811 if (LOG_FACILITY(u) != 0)
812 facility = LOG_FACILITY(u);
813 endp++;
814 line = endp;
815 }
816 }
817
818 devkmsg_emit(facility, level, "%s", line);
819 kfree(buf);
820 return ret;
821}
822
823static ssize_t devkmsg_read(struct file *file, char __user *buf,
824 size_t count, loff_t *ppos)
825{
826 struct devkmsg_user *user = file->private_data;
827 char *outbuf = &user->pbufs.outbuf[0];
828 struct printk_message pmsg = {
829 .pbufs = &user->pbufs,
830 };
831 ssize_t ret;
832
833 ret = mutex_lock_interruptible(&user->lock);
834 if (ret)
835 return ret;
836
837 if (!printk_get_next_message(&pmsg, atomic64_read(&user->seq), true, false)) {
838 if (file->f_flags & O_NONBLOCK) {
839 ret = -EAGAIN;
840 goto out;
841 }
842
843 /*
844 * Guarantee this task is visible on the waitqueue before
845 * checking the wake condition.
846 *
847 * The full memory barrier within set_current_state() of
848 * prepare_to_wait_event() pairs with the full memory barrier
849 * within wq_has_sleeper().
850 *
851 * This pairs with __wake_up_klogd:A.
852 */
853 ret = wait_event_interruptible(log_wait,
854 printk_get_next_message(&pmsg, atomic64_read(&user->seq), true,
855 false)); /* LMM(devkmsg_read:A) */
856 if (ret)
857 goto out;
858 }
859
860 if (pmsg.dropped) {
861 /* our last seen message is gone, return error and reset */
862 atomic64_set(&user->seq, pmsg.seq);
863 ret = -EPIPE;
864 goto out;
865 }
866
867 atomic64_set(&user->seq, pmsg.seq + 1);
868
869 if (pmsg.outbuf_len > count) {
870 ret = -EINVAL;
871 goto out;
872 }
873
874 if (copy_to_user(buf, outbuf, pmsg.outbuf_len)) {
875 ret = -EFAULT;
876 goto out;
877 }
878 ret = pmsg.outbuf_len;
879out:
880 mutex_unlock(&user->lock);
881 return ret;
882}
883
884/*
885 * Be careful when modifying this function!!!
886 *
887 * Only few operations are supported because the device works only with the
888 * entire variable length messages (records). Non-standard values are
889 * returned in the other cases and has been this way for quite some time.
890 * User space applications might depend on this behavior.
891 */
892static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
893{
894 struct devkmsg_user *user = file->private_data;
895 loff_t ret = 0;
896
897 if (offset)
898 return -ESPIPE;
899
900 switch (whence) {
901 case SEEK_SET:
902 /* the first record */
903 atomic64_set(&user->seq, prb_first_valid_seq(prb));
904 break;
905 case SEEK_DATA:
906 /*
907 * The first record after the last SYSLOG_ACTION_CLEAR,
908 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
909 * changes no global state, and does not clear anything.
910 */
911 atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
912 break;
913 case SEEK_END:
914 /* after the last record */
915 atomic64_set(&user->seq, prb_next_seq(prb));
916 break;
917 default:
918 ret = -EINVAL;
919 }
920 return ret;
921}
922
923static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
924{
925 struct devkmsg_user *user = file->private_data;
926 struct printk_info info;
927 __poll_t ret = 0;
928
929 poll_wait(file, &log_wait, wait);
930
931 if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
932 /* return error when data has vanished underneath us */
933 if (info.seq != atomic64_read(&user->seq))
934 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
935 else
936 ret = EPOLLIN|EPOLLRDNORM;
937 }
938
939 return ret;
940}
941
942static int devkmsg_open(struct inode *inode, struct file *file)
943{
944 struct devkmsg_user *user;
945 int err;
946
947 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
948 return -EPERM;
949
950 /* write-only does not need any file context */
951 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
952 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
953 SYSLOG_FROM_READER);
954 if (err)
955 return err;
956 }
957
958 user = kvmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
959 if (!user)
960 return -ENOMEM;
961
962 ratelimit_default_init(&user->rs);
963 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
964
965 mutex_init(&user->lock);
966
967 atomic64_set(&user->seq, prb_first_valid_seq(prb));
968
969 file->private_data = user;
970 return 0;
971}
972
973static int devkmsg_release(struct inode *inode, struct file *file)
974{
975 struct devkmsg_user *user = file->private_data;
976
977 ratelimit_state_exit(&user->rs);
978
979 mutex_destroy(&user->lock);
980 kvfree(user);
981 return 0;
982}
983
984const struct file_operations kmsg_fops = {
985 .open = devkmsg_open,
986 .read = devkmsg_read,
987 .write_iter = devkmsg_write,
988 .llseek = devkmsg_llseek,
989 .poll = devkmsg_poll,
990 .release = devkmsg_release,
991};
992
993#ifdef CONFIG_VMCORE_INFO
994/*
995 * This appends the listed symbols to /proc/vmcore
996 *
997 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
998 * obtain access to symbols that are otherwise very difficult to locate. These
999 * symbols are specifically used so that utilities can access and extract the
1000 * dmesg log from a vmcore file after a crash.
1001 */
1002void log_buf_vmcoreinfo_setup(void)
1003{
1004 struct dev_printk_info *dev_info = NULL;
1005
1006 VMCOREINFO_SYMBOL(prb);
1007 VMCOREINFO_SYMBOL(printk_rb_static);
1008 VMCOREINFO_SYMBOL(clear_seq);
1009
1010 /*
1011 * Export struct size and field offsets. User space tools can
1012 * parse it and detect any changes to structure down the line.
1013 */
1014
1015 VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
1016 VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
1017 VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
1018 VMCOREINFO_OFFSET(printk_ringbuffer, fail);
1019
1020 VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
1021 VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
1022 VMCOREINFO_OFFSET(prb_desc_ring, descs);
1023 VMCOREINFO_OFFSET(prb_desc_ring, infos);
1024 VMCOREINFO_OFFSET(prb_desc_ring, head_id);
1025 VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
1026
1027 VMCOREINFO_STRUCT_SIZE(prb_desc);
1028 VMCOREINFO_OFFSET(prb_desc, state_var);
1029 VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
1030
1031 VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
1032 VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
1033 VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
1034
1035 VMCOREINFO_STRUCT_SIZE(printk_info);
1036 VMCOREINFO_OFFSET(printk_info, seq);
1037 VMCOREINFO_OFFSET(printk_info, ts_nsec);
1038 VMCOREINFO_OFFSET(printk_info, text_len);
1039 VMCOREINFO_OFFSET(printk_info, caller_id);
1040 VMCOREINFO_OFFSET(printk_info, dev_info);
1041
1042 VMCOREINFO_STRUCT_SIZE(dev_printk_info);
1043 VMCOREINFO_OFFSET(dev_printk_info, subsystem);
1044 VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
1045 VMCOREINFO_OFFSET(dev_printk_info, device);
1046 VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
1047
1048 VMCOREINFO_STRUCT_SIZE(prb_data_ring);
1049 VMCOREINFO_OFFSET(prb_data_ring, size_bits);
1050 VMCOREINFO_OFFSET(prb_data_ring, data);
1051 VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
1052 VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
1053
1054 VMCOREINFO_SIZE(atomic_long_t);
1055 VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
1056
1057 VMCOREINFO_STRUCT_SIZE(latched_seq);
1058 VMCOREINFO_OFFSET(latched_seq, val);
1059}
1060#endif
1061
1062/* requested log_buf_len from kernel cmdline */
1063static unsigned long __initdata new_log_buf_len;
1064
1065/* we practice scaling the ring buffer by powers of 2 */
1066static void __init log_buf_len_update(u64 size)
1067{
1068 if (size > (u64)LOG_BUF_LEN_MAX) {
1069 size = (u64)LOG_BUF_LEN_MAX;
1070 pr_err("log_buf over 2G is not supported.\n");
1071 }
1072
1073 if (size)
1074 size = roundup_pow_of_two(size);
1075 if (size > log_buf_len)
1076 new_log_buf_len = (unsigned long)size;
1077}
1078
1079/* save requested log_buf_len since it's too early to process it */
1080static int __init log_buf_len_setup(char *str)
1081{
1082 u64 size;
1083
1084 if (!str)
1085 return -EINVAL;
1086
1087 size = memparse(str, &str);
1088
1089 log_buf_len_update(size);
1090
1091 return 0;
1092}
1093early_param("log_buf_len", log_buf_len_setup);
1094
1095#ifdef CONFIG_SMP
1096#define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1097
1098static void __init log_buf_add_cpu(void)
1099{
1100 unsigned int cpu_extra;
1101
1102 /*
1103 * archs should set up cpu_possible_bits properly with
1104 * set_cpu_possible() after setup_arch() but just in
1105 * case lets ensure this is valid.
1106 */
1107 if (num_possible_cpus() == 1)
1108 return;
1109
1110 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1111
1112 /* by default this will only continue through for large > 64 CPUs */
1113 if (cpu_extra <= __LOG_BUF_LEN / 2)
1114 return;
1115
1116 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1117 __LOG_CPU_MAX_BUF_LEN);
1118 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1119 cpu_extra);
1120 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1121
1122 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1123}
1124#else /* !CONFIG_SMP */
1125static inline void log_buf_add_cpu(void) {}
1126#endif /* CONFIG_SMP */
1127
1128static void __init set_percpu_data_ready(void)
1129{
1130 __printk_percpu_data_ready = true;
1131}
1132
1133static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1134 struct printk_record *r)
1135{
1136 struct prb_reserved_entry e;
1137 struct printk_record dest_r;
1138
1139 prb_rec_init_wr(&dest_r, r->info->text_len);
1140
1141 if (!prb_reserve(&e, rb, &dest_r))
1142 return 0;
1143
1144 memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1145 dest_r.info->text_len = r->info->text_len;
1146 dest_r.info->facility = r->info->facility;
1147 dest_r.info->level = r->info->level;
1148 dest_r.info->flags = r->info->flags;
1149 dest_r.info->ts_nsec = r->info->ts_nsec;
1150 dest_r.info->caller_id = r->info->caller_id;
1151 memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1152
1153 prb_final_commit(&e);
1154
1155 return prb_record_text_space(&e);
1156}
1157
1158static char setup_text_buf[PRINTKRB_RECORD_MAX] __initdata;
1159
1160static void print_log_buf_usage_stats(void)
1161{
1162 unsigned int descs_count = log_buf_len >> PRB_AVGBITS;
1163 size_t meta_data_size;
1164
1165 meta_data_size = descs_count * (sizeof(struct prb_desc) + sizeof(struct printk_info));
1166
1167 pr_info("log buffer data + meta data: %u + %zu = %zu bytes\n",
1168 log_buf_len, meta_data_size, log_buf_len + meta_data_size);
1169}
1170
1171void __init setup_log_buf(int early)
1172{
1173 struct printk_info *new_infos;
1174 unsigned int new_descs_count;
1175 struct prb_desc *new_descs;
1176 struct printk_info info;
1177 struct printk_record r;
1178 unsigned int text_size;
1179 size_t new_descs_size;
1180 size_t new_infos_size;
1181 unsigned long flags;
1182 char *new_log_buf;
1183 unsigned int free;
1184 u64 seq;
1185
1186 /*
1187 * Some archs call setup_log_buf() multiple times - first is very
1188 * early, e.g. from setup_arch(), and second - when percpu_areas
1189 * are initialised.
1190 */
1191 if (!early)
1192 set_percpu_data_ready();
1193
1194 if (log_buf != __log_buf)
1195 return;
1196
1197 if (!early && !new_log_buf_len)
1198 log_buf_add_cpu();
1199
1200 if (!new_log_buf_len) {
1201 /* Show the memory stats only once. */
1202 if (!early)
1203 goto out;
1204
1205 return;
1206 }
1207
1208 new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1209 if (new_descs_count == 0) {
1210 pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1211 goto out;
1212 }
1213
1214 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1215 if (unlikely(!new_log_buf)) {
1216 pr_err("log_buf_len: %lu text bytes not available\n",
1217 new_log_buf_len);
1218 goto out;
1219 }
1220
1221 new_descs_size = new_descs_count * sizeof(struct prb_desc);
1222 new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1223 if (unlikely(!new_descs)) {
1224 pr_err("log_buf_len: %zu desc bytes not available\n",
1225 new_descs_size);
1226 goto err_free_log_buf;
1227 }
1228
1229 new_infos_size = new_descs_count * sizeof(struct printk_info);
1230 new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1231 if (unlikely(!new_infos)) {
1232 pr_err("log_buf_len: %zu info bytes not available\n",
1233 new_infos_size);
1234 goto err_free_descs;
1235 }
1236
1237 prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1238
1239 prb_init(&printk_rb_dynamic,
1240 new_log_buf, ilog2(new_log_buf_len),
1241 new_descs, ilog2(new_descs_count),
1242 new_infos);
1243
1244 local_irq_save(flags);
1245
1246 log_buf_len = new_log_buf_len;
1247 log_buf = new_log_buf;
1248 new_log_buf_len = 0;
1249
1250 free = __LOG_BUF_LEN;
1251 prb_for_each_record(0, &printk_rb_static, seq, &r) {
1252 text_size = add_to_rb(&printk_rb_dynamic, &r);
1253 if (text_size > free)
1254 free = 0;
1255 else
1256 free -= text_size;
1257 }
1258
1259 prb = &printk_rb_dynamic;
1260
1261 local_irq_restore(flags);
1262
1263 /*
1264 * Copy any remaining messages that might have appeared from
1265 * NMI context after copying but before switching to the
1266 * dynamic buffer.
1267 */
1268 prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1269 text_size = add_to_rb(&printk_rb_dynamic, &r);
1270 if (text_size > free)
1271 free = 0;
1272 else
1273 free -= text_size;
1274 }
1275
1276 if (seq != prb_next_seq(&printk_rb_static)) {
1277 pr_err("dropped %llu messages\n",
1278 prb_next_seq(&printk_rb_static) - seq);
1279 }
1280
1281 print_log_buf_usage_stats();
1282 pr_info("early log buf free: %u(%u%%)\n",
1283 free, (free * 100) / __LOG_BUF_LEN);
1284 return;
1285
1286err_free_descs:
1287 memblock_free(new_descs, new_descs_size);
1288err_free_log_buf:
1289 memblock_free(new_log_buf, new_log_buf_len);
1290out:
1291 print_log_buf_usage_stats();
1292}
1293
1294static bool __read_mostly ignore_loglevel;
1295
1296static int __init ignore_loglevel_setup(char *str)
1297{
1298 ignore_loglevel = true;
1299 pr_info("debug: ignoring loglevel setting.\n");
1300
1301 return 0;
1302}
1303
1304early_param("ignore_loglevel", ignore_loglevel_setup);
1305module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1306MODULE_PARM_DESC(ignore_loglevel,
1307 "ignore loglevel setting (prints all kernel messages to the console)");
1308
1309static bool suppress_message_printing(int level)
1310{
1311 return (level >= console_loglevel && !ignore_loglevel);
1312}
1313
1314#ifdef CONFIG_BOOT_PRINTK_DELAY
1315
1316static int boot_delay; /* msecs delay after each printk during bootup */
1317static unsigned long long loops_per_msec; /* based on boot_delay */
1318
1319static int __init boot_delay_setup(char *str)
1320{
1321 unsigned long lpj;
1322
1323 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1324 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1325
1326 get_option(&str, &boot_delay);
1327 if (boot_delay > 10 * 1000)
1328 boot_delay = 0;
1329
1330 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1331 "HZ: %d, loops_per_msec: %llu\n",
1332 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1333 return 0;
1334}
1335early_param("boot_delay", boot_delay_setup);
1336
1337static void boot_delay_msec(int level)
1338{
1339 unsigned long long k;
1340 unsigned long timeout;
1341 bool suppress = !is_printk_force_console() &&
1342 suppress_message_printing(level);
1343
1344 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING) || suppress)
1345 return;
1346
1347 k = (unsigned long long)loops_per_msec * boot_delay;
1348
1349 timeout = jiffies + msecs_to_jiffies(boot_delay);
1350 while (k) {
1351 k--;
1352 cpu_relax();
1353 /*
1354 * use (volatile) jiffies to prevent
1355 * compiler reduction; loop termination via jiffies
1356 * is secondary and may or may not happen.
1357 */
1358 if (time_after(jiffies, timeout))
1359 break;
1360 touch_nmi_watchdog();
1361 }
1362}
1363#else
1364static inline void boot_delay_msec(int level)
1365{
1366}
1367#endif
1368
1369static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1370module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1371
1372static size_t print_syslog(unsigned int level, char *buf)
1373{
1374 return sprintf(buf, "<%u>", level);
1375}
1376
1377static size_t print_time(u64 ts, char *buf)
1378{
1379 unsigned long rem_nsec = do_div(ts, 1000000000);
1380
1381 return sprintf(buf, "[%5lu.%06lu]",
1382 (unsigned long)ts, rem_nsec / 1000);
1383}
1384
1385#ifdef CONFIG_PRINTK_CALLER
1386static size_t print_caller(u32 id, char *buf)
1387{
1388 char caller[12];
1389
1390 snprintf(caller, sizeof(caller), "%c%u",
1391 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1392 return sprintf(buf, "[%6s]", caller);
1393}
1394#else
1395#define print_caller(id, buf) 0
1396#endif
1397
1398static size_t info_print_prefix(const struct printk_info *info, bool syslog,
1399 bool time, char *buf)
1400{
1401 size_t len = 0;
1402
1403 if (syslog)
1404 len = print_syslog((info->facility << 3) | info->level, buf);
1405
1406 if (time)
1407 len += print_time(info->ts_nsec, buf + len);
1408
1409 len += print_caller(info->caller_id, buf + len);
1410
1411 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1412 buf[len++] = ' ';
1413 buf[len] = '\0';
1414 }
1415
1416 return len;
1417}
1418
1419/*
1420 * Prepare the record for printing. The text is shifted within the given
1421 * buffer to avoid a need for another one. The following operations are
1422 * done:
1423 *
1424 * - Add prefix for each line.
1425 * - Drop truncated lines that no longer fit into the buffer.
1426 * - Add the trailing newline that has been removed in vprintk_store().
1427 * - Add a string terminator.
1428 *
1429 * Since the produced string is always terminated, the maximum possible
1430 * return value is @r->text_buf_size - 1;
1431 *
1432 * Return: The length of the updated/prepared text, including the added
1433 * prefixes and the newline. The terminator is not counted. The dropped
1434 * line(s) are not counted.
1435 */
1436static size_t record_print_text(struct printk_record *r, bool syslog,
1437 bool time)
1438{
1439 size_t text_len = r->info->text_len;
1440 size_t buf_size = r->text_buf_size;
1441 char *text = r->text_buf;
1442 char prefix[PRINTK_PREFIX_MAX];
1443 bool truncated = false;
1444 size_t prefix_len;
1445 size_t line_len;
1446 size_t len = 0;
1447 char *next;
1448
1449 /*
1450 * If the message was truncated because the buffer was not large
1451 * enough, treat the available text as if it were the full text.
1452 */
1453 if (text_len > buf_size)
1454 text_len = buf_size;
1455
1456 prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1457
1458 /*
1459 * @text_len: bytes of unprocessed text
1460 * @line_len: bytes of current line _without_ newline
1461 * @text: pointer to beginning of current line
1462 * @len: number of bytes prepared in r->text_buf
1463 */
1464 for (;;) {
1465 next = memchr(text, '\n', text_len);
1466 if (next) {
1467 line_len = next - text;
1468 } else {
1469 /* Drop truncated line(s). */
1470 if (truncated)
1471 break;
1472 line_len = text_len;
1473 }
1474
1475 /*
1476 * Truncate the text if there is not enough space to add the
1477 * prefix and a trailing newline and a terminator.
1478 */
1479 if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1480 /* Drop even the current line if no space. */
1481 if (len + prefix_len + line_len + 1 + 1 > buf_size)
1482 break;
1483
1484 text_len = buf_size - len - prefix_len - 1 - 1;
1485 truncated = true;
1486 }
1487
1488 memmove(text + prefix_len, text, text_len);
1489 memcpy(text, prefix, prefix_len);
1490
1491 /*
1492 * Increment the prepared length to include the text and
1493 * prefix that were just moved+copied. Also increment for the
1494 * newline at the end of this line. If this is the last line,
1495 * there is no newline, but it will be added immediately below.
1496 */
1497 len += prefix_len + line_len + 1;
1498 if (text_len == line_len) {
1499 /*
1500 * This is the last line. Add the trailing newline
1501 * removed in vprintk_store().
1502 */
1503 text[prefix_len + line_len] = '\n';
1504 break;
1505 }
1506
1507 /*
1508 * Advance beyond the added prefix and the related line with
1509 * its newline.
1510 */
1511 text += prefix_len + line_len + 1;
1512
1513 /*
1514 * The remaining text has only decreased by the line with its
1515 * newline.
1516 *
1517 * Note that @text_len can become zero. It happens when @text
1518 * ended with a newline (either due to truncation or the
1519 * original string ending with "\n\n"). The loop is correctly
1520 * repeated and (if not truncated) an empty line with a prefix
1521 * will be prepared.
1522 */
1523 text_len -= line_len + 1;
1524 }
1525
1526 /*
1527 * If a buffer was provided, it will be terminated. Space for the
1528 * string terminator is guaranteed to be available. The terminator is
1529 * not counted in the return value.
1530 */
1531 if (buf_size > 0)
1532 r->text_buf[len] = 0;
1533
1534 return len;
1535}
1536
1537static size_t get_record_print_text_size(struct printk_info *info,
1538 unsigned int line_count,
1539 bool syslog, bool time)
1540{
1541 char prefix[PRINTK_PREFIX_MAX];
1542 size_t prefix_len;
1543
1544 prefix_len = info_print_prefix(info, syslog, time, prefix);
1545
1546 /*
1547 * Each line will be preceded with a prefix. The intermediate
1548 * newlines are already within the text, but a final trailing
1549 * newline will be added.
1550 */
1551 return ((prefix_len * line_count) + info->text_len + 1);
1552}
1553
1554/*
1555 * Beginning with @start_seq, find the first record where it and all following
1556 * records up to (but not including) @max_seq fit into @size.
1557 *
1558 * @max_seq is simply an upper bound and does not need to exist. If the caller
1559 * does not require an upper bound, -1 can be used for @max_seq.
1560 */
1561static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1562 bool syslog, bool time)
1563{
1564 struct printk_info info;
1565 unsigned int line_count;
1566 size_t len = 0;
1567 u64 seq;
1568
1569 /* Determine the size of the records up to @max_seq. */
1570 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1571 if (info.seq >= max_seq)
1572 break;
1573 len += get_record_print_text_size(&info, line_count, syslog, time);
1574 }
1575
1576 /*
1577 * Adjust the upper bound for the next loop to avoid subtracting
1578 * lengths that were never added.
1579 */
1580 if (seq < max_seq)
1581 max_seq = seq;
1582
1583 /*
1584 * Move first record forward until length fits into the buffer. Ignore
1585 * newest messages that were not counted in the above cycle. Messages
1586 * might appear and get lost in the meantime. This is a best effort
1587 * that prevents an infinite loop that could occur with a retry.
1588 */
1589 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1590 if (len <= size || info.seq >= max_seq)
1591 break;
1592 len -= get_record_print_text_size(&info, line_count, syslog, time);
1593 }
1594
1595 return seq;
1596}
1597
1598/* The caller is responsible for making sure @size is greater than 0. */
1599static int syslog_print(char __user *buf, int size)
1600{
1601 struct printk_info info;
1602 struct printk_record r;
1603 char *text;
1604 int len = 0;
1605 u64 seq;
1606
1607 text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1608 if (!text)
1609 return -ENOMEM;
1610
1611 prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1612
1613 mutex_lock(&syslog_lock);
1614
1615 /*
1616 * Wait for the @syslog_seq record to be available. @syslog_seq may
1617 * change while waiting.
1618 */
1619 do {
1620 seq = syslog_seq;
1621
1622 mutex_unlock(&syslog_lock);
1623 /*
1624 * Guarantee this task is visible on the waitqueue before
1625 * checking the wake condition.
1626 *
1627 * The full memory barrier within set_current_state() of
1628 * prepare_to_wait_event() pairs with the full memory barrier
1629 * within wq_has_sleeper().
1630 *
1631 * This pairs with __wake_up_klogd:A.
1632 */
1633 len = wait_event_interruptible(log_wait,
1634 prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
1635 mutex_lock(&syslog_lock);
1636
1637 if (len)
1638 goto out;
1639 } while (syslog_seq != seq);
1640
1641 /*
1642 * Copy records that fit into the buffer. The above cycle makes sure
1643 * that the first record is always available.
1644 */
1645 do {
1646 size_t n;
1647 size_t skip;
1648 int err;
1649
1650 if (!prb_read_valid(prb, syslog_seq, &r))
1651 break;
1652
1653 if (r.info->seq != syslog_seq) {
1654 /* message is gone, move to next valid one */
1655 syslog_seq = r.info->seq;
1656 syslog_partial = 0;
1657 }
1658
1659 /*
1660 * To keep reading/counting partial line consistent,
1661 * use printk_time value as of the beginning of a line.
1662 */
1663 if (!syslog_partial)
1664 syslog_time = printk_time;
1665
1666 skip = syslog_partial;
1667 n = record_print_text(&r, true, syslog_time);
1668 if (n - syslog_partial <= size) {
1669 /* message fits into buffer, move forward */
1670 syslog_seq = r.info->seq + 1;
1671 n -= syslog_partial;
1672 syslog_partial = 0;
1673 } else if (!len){
1674 /* partial read(), remember position */
1675 n = size;
1676 syslog_partial += n;
1677 } else
1678 n = 0;
1679
1680 if (!n)
1681 break;
1682
1683 mutex_unlock(&syslog_lock);
1684 err = copy_to_user(buf, text + skip, n);
1685 mutex_lock(&syslog_lock);
1686
1687 if (err) {
1688 if (!len)
1689 len = -EFAULT;
1690 break;
1691 }
1692
1693 len += n;
1694 size -= n;
1695 buf += n;
1696 } while (size);
1697out:
1698 mutex_unlock(&syslog_lock);
1699 kfree(text);
1700 return len;
1701}
1702
1703static int syslog_print_all(char __user *buf, int size, bool clear)
1704{
1705 struct printk_info info;
1706 struct printk_record r;
1707 char *text;
1708 int len = 0;
1709 u64 seq;
1710 bool time;
1711
1712 text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1713 if (!text)
1714 return -ENOMEM;
1715
1716 time = printk_time;
1717 /*
1718 * Find first record that fits, including all following records,
1719 * into the user-provided buffer for this dump.
1720 */
1721 seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1722 size, true, time);
1723
1724 prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1725
1726 prb_for_each_record(seq, prb, seq, &r) {
1727 int textlen;
1728
1729 textlen = record_print_text(&r, true, time);
1730
1731 if (len + textlen > size) {
1732 seq--;
1733 break;
1734 }
1735
1736 if (copy_to_user(buf + len, text, textlen))
1737 len = -EFAULT;
1738 else
1739 len += textlen;
1740
1741 if (len < 0)
1742 break;
1743 }
1744
1745 if (clear) {
1746 mutex_lock(&syslog_lock);
1747 latched_seq_write(&clear_seq, seq);
1748 mutex_unlock(&syslog_lock);
1749 }
1750
1751 kfree(text);
1752 return len;
1753}
1754
1755static void syslog_clear(void)
1756{
1757 mutex_lock(&syslog_lock);
1758 latched_seq_write(&clear_seq, prb_next_seq(prb));
1759 mutex_unlock(&syslog_lock);
1760}
1761
1762int do_syslog(int type, char __user *buf, int len, int source)
1763{
1764 struct printk_info info;
1765 bool clear = false;
1766 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1767 int error;
1768
1769 error = check_syslog_permissions(type, source);
1770 if (error)
1771 return error;
1772
1773 switch (type) {
1774 case SYSLOG_ACTION_CLOSE: /* Close log */
1775 break;
1776 case SYSLOG_ACTION_OPEN: /* Open log */
1777 break;
1778 case SYSLOG_ACTION_READ: /* Read from log */
1779 if (!buf || len < 0)
1780 return -EINVAL;
1781 if (!len)
1782 return 0;
1783 if (!access_ok(buf, len))
1784 return -EFAULT;
1785 error = syslog_print(buf, len);
1786 break;
1787 /* Read/clear last kernel messages */
1788 case SYSLOG_ACTION_READ_CLEAR:
1789 clear = true;
1790 fallthrough;
1791 /* Read last kernel messages */
1792 case SYSLOG_ACTION_READ_ALL:
1793 if (!buf || len < 0)
1794 return -EINVAL;
1795 if (!len)
1796 return 0;
1797 if (!access_ok(buf, len))
1798 return -EFAULT;
1799 error = syslog_print_all(buf, len, clear);
1800 break;
1801 /* Clear ring buffer */
1802 case SYSLOG_ACTION_CLEAR:
1803 syslog_clear();
1804 break;
1805 /* Disable logging to console */
1806 case SYSLOG_ACTION_CONSOLE_OFF:
1807 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1808 saved_console_loglevel = console_loglevel;
1809 console_loglevel = minimum_console_loglevel;
1810 break;
1811 /* Enable logging to console */
1812 case SYSLOG_ACTION_CONSOLE_ON:
1813 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1814 console_loglevel = saved_console_loglevel;
1815 saved_console_loglevel = LOGLEVEL_DEFAULT;
1816 }
1817 break;
1818 /* Set level of messages printed to console */
1819 case SYSLOG_ACTION_CONSOLE_LEVEL:
1820 if (len < 1 || len > 8)
1821 return -EINVAL;
1822 if (len < minimum_console_loglevel)
1823 len = minimum_console_loglevel;
1824 console_loglevel = len;
1825 /* Implicitly re-enable logging to console */
1826 saved_console_loglevel = LOGLEVEL_DEFAULT;
1827 break;
1828 /* Number of chars in the log buffer */
1829 case SYSLOG_ACTION_SIZE_UNREAD:
1830 mutex_lock(&syslog_lock);
1831 if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1832 /* No unread messages. */
1833 mutex_unlock(&syslog_lock);
1834 return 0;
1835 }
1836 if (info.seq != syslog_seq) {
1837 /* messages are gone, move to first one */
1838 syslog_seq = info.seq;
1839 syslog_partial = 0;
1840 }
1841 if (source == SYSLOG_FROM_PROC) {
1842 /*
1843 * Short-cut for poll(/"proc/kmsg") which simply checks
1844 * for pending data, not the size; return the count of
1845 * records, not the length.
1846 */
1847 error = prb_next_seq(prb) - syslog_seq;
1848 } else {
1849 bool time = syslog_partial ? syslog_time : printk_time;
1850 unsigned int line_count;
1851 u64 seq;
1852
1853 prb_for_each_info(syslog_seq, prb, seq, &info,
1854 &line_count) {
1855 error += get_record_print_text_size(&info, line_count,
1856 true, time);
1857 time = printk_time;
1858 }
1859 error -= syslog_partial;
1860 }
1861 mutex_unlock(&syslog_lock);
1862 break;
1863 /* Size of the log buffer */
1864 case SYSLOG_ACTION_SIZE_BUFFER:
1865 error = log_buf_len;
1866 break;
1867 default:
1868 error = -EINVAL;
1869 break;
1870 }
1871
1872 return error;
1873}
1874
1875SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1876{
1877 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1878}
1879
1880/*
1881 * Special console_lock variants that help to reduce the risk of soft-lockups.
1882 * They allow to pass console_lock to another printk() call using a busy wait.
1883 */
1884
1885#ifdef CONFIG_LOCKDEP
1886static struct lockdep_map console_owner_dep_map = {
1887 .name = "console_owner"
1888};
1889#endif
1890
1891static DEFINE_RAW_SPINLOCK(console_owner_lock);
1892static struct task_struct *console_owner;
1893static bool console_waiter;
1894
1895/**
1896 * console_lock_spinning_enable - mark beginning of code where another
1897 * thread might safely busy wait
1898 *
1899 * This basically converts console_lock into a spinlock. This marks
1900 * the section where the console_lock owner can not sleep, because
1901 * there may be a waiter spinning (like a spinlock). Also it must be
1902 * ready to hand over the lock at the end of the section.
1903 */
1904void console_lock_spinning_enable(void)
1905{
1906 /*
1907 * Do not use spinning in panic(). The panic CPU wants to keep the lock.
1908 * Non-panic CPUs abandon the flush anyway.
1909 *
1910 * Just keep the lockdep annotation. The panic-CPU should avoid
1911 * taking console_owner_lock because it might cause a deadlock.
1912 * This looks like the easiest way how to prevent false lockdep
1913 * reports without handling races a lockless way.
1914 */
1915 if (panic_in_progress())
1916 goto lockdep;
1917
1918 raw_spin_lock(&console_owner_lock);
1919 console_owner = current;
1920 raw_spin_unlock(&console_owner_lock);
1921
1922lockdep:
1923 /* The waiter may spin on us after setting console_owner */
1924 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1925}
1926
1927/**
1928 * console_lock_spinning_disable_and_check - mark end of code where another
1929 * thread was able to busy wait and check if there is a waiter
1930 * @cookie: cookie returned from console_srcu_read_lock()
1931 *
1932 * This is called at the end of the section where spinning is allowed.
1933 * It has two functions. First, it is a signal that it is no longer
1934 * safe to start busy waiting for the lock. Second, it checks if
1935 * there is a busy waiter and passes the lock rights to her.
1936 *
1937 * Important: Callers lose both the console_lock and the SRCU read lock if
1938 * there was a busy waiter. They must not touch items synchronized by
1939 * console_lock or SRCU read lock in this case.
1940 *
1941 * Return: 1 if the lock rights were passed, 0 otherwise.
1942 */
1943int console_lock_spinning_disable_and_check(int cookie)
1944{
1945 int waiter;
1946
1947 /*
1948 * Ignore spinning waiters during panic() because they might get stopped
1949 * or blocked at any time,
1950 *
1951 * It is safe because nobody is allowed to start spinning during panic
1952 * in the first place. If there has been a waiter then non panic CPUs
1953 * might stay spinning. They would get stopped anyway. The panic context
1954 * will never start spinning and an interrupted spin on panic CPU will
1955 * never continue.
1956 */
1957 if (panic_in_progress()) {
1958 /* Keep lockdep happy. */
1959 spin_release(&console_owner_dep_map, _THIS_IP_);
1960 return 0;
1961 }
1962
1963 raw_spin_lock(&console_owner_lock);
1964 waiter = READ_ONCE(console_waiter);
1965 console_owner = NULL;
1966 raw_spin_unlock(&console_owner_lock);
1967
1968 if (!waiter) {
1969 spin_release(&console_owner_dep_map, _THIS_IP_);
1970 return 0;
1971 }
1972
1973 /* The waiter is now free to continue */
1974 WRITE_ONCE(console_waiter, false);
1975
1976 spin_release(&console_owner_dep_map, _THIS_IP_);
1977
1978 /*
1979 * Preserve lockdep lock ordering. Release the SRCU read lock before
1980 * releasing the console_lock.
1981 */
1982 console_srcu_read_unlock(cookie);
1983
1984 /*
1985 * Hand off console_lock to waiter. The waiter will perform
1986 * the up(). After this, the waiter is the console_lock owner.
1987 */
1988 mutex_release(&console_lock_dep_map, _THIS_IP_);
1989 return 1;
1990}
1991
1992/**
1993 * console_trylock_spinning - try to get console_lock by busy waiting
1994 *
1995 * This allows to busy wait for the console_lock when the current
1996 * owner is running in specially marked sections. It means that
1997 * the current owner is running and cannot reschedule until it
1998 * is ready to lose the lock.
1999 *
2000 * Return: 1 if we got the lock, 0 othrewise
2001 */
2002static int console_trylock_spinning(void)
2003{
2004 struct task_struct *owner = NULL;
2005 bool waiter;
2006 bool spin = false;
2007 unsigned long flags;
2008
2009 if (console_trylock())
2010 return 1;
2011
2012 /*
2013 * It's unsafe to spin once a panic has begun. If we are the
2014 * panic CPU, we may have already halted the owner of the
2015 * console_sem. If we are not the panic CPU, then we should
2016 * avoid taking console_sem, so the panic CPU has a better
2017 * chance of cleanly acquiring it later.
2018 */
2019 if (panic_in_progress())
2020 return 0;
2021
2022 printk_safe_enter_irqsave(flags);
2023
2024 raw_spin_lock(&console_owner_lock);
2025 owner = READ_ONCE(console_owner);
2026 waiter = READ_ONCE(console_waiter);
2027 if (!waiter && owner && owner != current) {
2028 WRITE_ONCE(console_waiter, true);
2029 spin = true;
2030 }
2031 raw_spin_unlock(&console_owner_lock);
2032
2033 /*
2034 * If there is an active printk() writing to the
2035 * consoles, instead of having it write our data too,
2036 * see if we can offload that load from the active
2037 * printer, and do some printing ourselves.
2038 * Go into a spin only if there isn't already a waiter
2039 * spinning, and there is an active printer, and
2040 * that active printer isn't us (recursive printk?).
2041 */
2042 if (!spin) {
2043 printk_safe_exit_irqrestore(flags);
2044 return 0;
2045 }
2046
2047 /* We spin waiting for the owner to release us */
2048 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
2049 /* Owner will clear console_waiter on hand off */
2050 while (READ_ONCE(console_waiter))
2051 cpu_relax();
2052 spin_release(&console_owner_dep_map, _THIS_IP_);
2053
2054 printk_safe_exit_irqrestore(flags);
2055 /*
2056 * The owner passed the console lock to us.
2057 * Since we did not spin on console lock, annotate
2058 * this as a trylock. Otherwise lockdep will
2059 * complain.
2060 */
2061 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
2062
2063 /*
2064 * Update @console_may_schedule for trylock because the previous
2065 * owner may have been schedulable.
2066 */
2067 console_may_schedule = 0;
2068
2069 return 1;
2070}
2071
2072/*
2073 * Recursion is tracked separately on each CPU. If NMIs are supported, an
2074 * additional NMI context per CPU is also separately tracked. Until per-CPU
2075 * is available, a separate "early tracking" is performed.
2076 */
2077static DEFINE_PER_CPU(u8, printk_count);
2078static u8 printk_count_early;
2079#ifdef CONFIG_HAVE_NMI
2080static DEFINE_PER_CPU(u8, printk_count_nmi);
2081static u8 printk_count_nmi_early;
2082#endif
2083
2084/*
2085 * Recursion is limited to keep the output sane. printk() should not require
2086 * more than 1 level of recursion (allowing, for example, printk() to trigger
2087 * a WARN), but a higher value is used in case some printk-internal errors
2088 * exist, such as the ringbuffer validation checks failing.
2089 */
2090#define PRINTK_MAX_RECURSION 3
2091
2092/*
2093 * Return a pointer to the dedicated counter for the CPU+context of the
2094 * caller.
2095 */
2096static u8 *__printk_recursion_counter(void)
2097{
2098#ifdef CONFIG_HAVE_NMI
2099 if (in_nmi()) {
2100 if (printk_percpu_data_ready())
2101 return this_cpu_ptr(&printk_count_nmi);
2102 return &printk_count_nmi_early;
2103 }
2104#endif
2105 if (printk_percpu_data_ready())
2106 return this_cpu_ptr(&printk_count);
2107 return &printk_count_early;
2108}
2109
2110/*
2111 * Enter recursion tracking. Interrupts are disabled to simplify tracking.
2112 * The caller must check the boolean return value to see if the recursion is
2113 * allowed. On failure, interrupts are not disabled.
2114 *
2115 * @recursion_ptr must be a variable of type (u8 *) and is the same variable
2116 * that is passed to printk_exit_irqrestore().
2117 */
2118#define printk_enter_irqsave(recursion_ptr, flags) \
2119({ \
2120 bool success = true; \
2121 \
2122 typecheck(u8 *, recursion_ptr); \
2123 local_irq_save(flags); \
2124 (recursion_ptr) = __printk_recursion_counter(); \
2125 if (*(recursion_ptr) > PRINTK_MAX_RECURSION) { \
2126 local_irq_restore(flags); \
2127 success = false; \
2128 } else { \
2129 (*(recursion_ptr))++; \
2130 } \
2131 success; \
2132})
2133
2134/* Exit recursion tracking, restoring interrupts. */
2135#define printk_exit_irqrestore(recursion_ptr, flags) \
2136 do { \
2137 typecheck(u8 *, recursion_ptr); \
2138 (*(recursion_ptr))--; \
2139 local_irq_restore(flags); \
2140 } while (0)
2141
2142int printk_delay_msec __read_mostly;
2143
2144static inline void printk_delay(int level)
2145{
2146 boot_delay_msec(level);
2147
2148 if (unlikely(printk_delay_msec)) {
2149 int m = printk_delay_msec;
2150
2151 while (m--) {
2152 mdelay(1);
2153 touch_nmi_watchdog();
2154 }
2155 }
2156}
2157
2158static inline u32 printk_caller_id(void)
2159{
2160 return in_task() ? task_pid_nr(current) :
2161 0x80000000 + smp_processor_id();
2162}
2163
2164/**
2165 * printk_parse_prefix - Parse level and control flags.
2166 *
2167 * @text: The terminated text message.
2168 * @level: A pointer to the current level value, will be updated.
2169 * @flags: A pointer to the current printk_info flags, will be updated.
2170 *
2171 * @level may be NULL if the caller is not interested in the parsed value.
2172 * Otherwise the variable pointed to by @level must be set to
2173 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2174 *
2175 * @flags may be NULL if the caller is not interested in the parsed value.
2176 * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2177 * value.
2178 *
2179 * Return: The length of the parsed level and control flags.
2180 */
2181u16 printk_parse_prefix(const char *text, int *level,
2182 enum printk_info_flags *flags)
2183{
2184 u16 prefix_len = 0;
2185 int kern_level;
2186
2187 while (*text) {
2188 kern_level = printk_get_level(text);
2189 if (!kern_level)
2190 break;
2191
2192 switch (kern_level) {
2193 case '0' ... '7':
2194 if (level && *level == LOGLEVEL_DEFAULT)
2195 *level = kern_level - '0';
2196 break;
2197 case 'c': /* KERN_CONT */
2198 if (flags)
2199 *flags |= LOG_CONT;
2200 }
2201
2202 prefix_len += 2;
2203 text += 2;
2204 }
2205
2206 return prefix_len;
2207}
2208
2209__printf(5, 0)
2210static u16 printk_sprint(char *text, u16 size, int facility,
2211 enum printk_info_flags *flags, const char *fmt,
2212 va_list args)
2213{
2214 u16 text_len;
2215
2216 text_len = vscnprintf(text, size, fmt, args);
2217
2218 /* Mark and strip a trailing newline. */
2219 if (text_len && text[text_len - 1] == '\n') {
2220 text_len--;
2221 *flags |= LOG_NEWLINE;
2222 }
2223
2224 /* Strip log level and control flags. */
2225 if (facility == 0) {
2226 u16 prefix_len;
2227
2228 prefix_len = printk_parse_prefix(text, NULL, NULL);
2229 if (prefix_len) {
2230 text_len -= prefix_len;
2231 memmove(text, text + prefix_len, text_len);
2232 }
2233 }
2234
2235 trace_console(text, text_len);
2236
2237 return text_len;
2238}
2239
2240__printf(4, 0)
2241int vprintk_store(int facility, int level,
2242 const struct dev_printk_info *dev_info,
2243 const char *fmt, va_list args)
2244{
2245 struct prb_reserved_entry e;
2246 enum printk_info_flags flags = 0;
2247 struct printk_record r;
2248 unsigned long irqflags;
2249 u16 trunc_msg_len = 0;
2250 char prefix_buf[8];
2251 u8 *recursion_ptr;
2252 u16 reserve_size;
2253 va_list args2;
2254 u32 caller_id;
2255 u16 text_len;
2256 int ret = 0;
2257 u64 ts_nsec;
2258
2259 if (!printk_enter_irqsave(recursion_ptr, irqflags))
2260 return 0;
2261
2262 /*
2263 * Since the duration of printk() can vary depending on the message
2264 * and state of the ringbuffer, grab the timestamp now so that it is
2265 * close to the call of printk(). This provides a more deterministic
2266 * timestamp with respect to the caller.
2267 */
2268 ts_nsec = local_clock();
2269
2270 caller_id = printk_caller_id();
2271
2272 /*
2273 * The sprintf needs to come first since the syslog prefix might be
2274 * passed in as a parameter. An extra byte must be reserved so that
2275 * later the vscnprintf() into the reserved buffer has room for the
2276 * terminating '\0', which is not counted by vsnprintf().
2277 */
2278 va_copy(args2, args);
2279 reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2280 va_end(args2);
2281
2282 if (reserve_size > PRINTKRB_RECORD_MAX)
2283 reserve_size = PRINTKRB_RECORD_MAX;
2284
2285 /* Extract log level or control flags. */
2286 if (facility == 0)
2287 printk_parse_prefix(&prefix_buf[0], &level, &flags);
2288
2289 if (level == LOGLEVEL_DEFAULT)
2290 level = default_message_loglevel;
2291
2292 if (dev_info)
2293 flags |= LOG_NEWLINE;
2294
2295 if (is_printk_force_console())
2296 flags |= LOG_FORCE_CON;
2297
2298 if (flags & LOG_CONT) {
2299 prb_rec_init_wr(&r, reserve_size);
2300 if (prb_reserve_in_last(&e, prb, &r, caller_id, PRINTKRB_RECORD_MAX)) {
2301 text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2302 facility, &flags, fmt, args);
2303 r.info->text_len += text_len;
2304
2305 if (flags & LOG_FORCE_CON)
2306 r.info->flags |= LOG_FORCE_CON;
2307
2308 if (flags & LOG_NEWLINE) {
2309 r.info->flags |= LOG_NEWLINE;
2310 prb_final_commit(&e);
2311 } else {
2312 prb_commit(&e);
2313 }
2314
2315 ret = text_len;
2316 goto out;
2317 }
2318 }
2319
2320 /*
2321 * Explicitly initialize the record before every prb_reserve() call.
2322 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2323 * structure when they fail.
2324 */
2325 prb_rec_init_wr(&r, reserve_size);
2326 if (!prb_reserve(&e, prb, &r)) {
2327 /* truncate the message if it is too long for empty buffer */
2328 truncate_msg(&reserve_size, &trunc_msg_len);
2329
2330 prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2331 if (!prb_reserve(&e, prb, &r))
2332 goto out;
2333 }
2334
2335 /* fill message */
2336 text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2337 if (trunc_msg_len)
2338 memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2339 r.info->text_len = text_len + trunc_msg_len;
2340 r.info->facility = facility;
2341 r.info->level = level & 7;
2342 r.info->flags = flags & 0x1f;
2343 r.info->ts_nsec = ts_nsec;
2344 r.info->caller_id = caller_id;
2345 if (dev_info)
2346 memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2347
2348 /* A message without a trailing newline can be continued. */
2349 if (!(flags & LOG_NEWLINE))
2350 prb_commit(&e);
2351 else
2352 prb_final_commit(&e);
2353
2354 ret = text_len + trunc_msg_len;
2355out:
2356 printk_exit_irqrestore(recursion_ptr, irqflags);
2357 return ret;
2358}
2359
2360/*
2361 * This acts as a one-way switch to allow legacy consoles to print from
2362 * the printk() caller context on a panic CPU. It also attempts to flush
2363 * the legacy consoles in this context.
2364 */
2365void printk_legacy_allow_panic_sync(void)
2366{
2367 struct console_flush_type ft;
2368
2369 legacy_allow_panic_sync = true;
2370
2371 printk_get_console_flush_type(&ft);
2372 if (ft.legacy_direct) {
2373 if (console_trylock())
2374 console_unlock();
2375 }
2376}
2377
2378asmlinkage int vprintk_emit(int facility, int level,
2379 const struct dev_printk_info *dev_info,
2380 const char *fmt, va_list args)
2381{
2382 struct console_flush_type ft;
2383 int printed_len;
2384
2385 /* Suppress unimportant messages after panic happens */
2386 if (unlikely(suppress_printk))
2387 return 0;
2388
2389 /*
2390 * The messages on the panic CPU are the most important. If
2391 * non-panic CPUs are generating any messages, they will be
2392 * silently dropped.
2393 */
2394 if (other_cpu_in_panic() && !panic_triggering_all_cpu_backtrace)
2395 return 0;
2396
2397 printk_get_console_flush_type(&ft);
2398
2399 /* If called from the scheduler, we can not call up(). */
2400 if (level == LOGLEVEL_SCHED) {
2401 level = LOGLEVEL_DEFAULT;
2402 ft.legacy_offload |= ft.legacy_direct;
2403 ft.legacy_direct = false;
2404 }
2405
2406 printk_delay(level);
2407
2408 printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2409
2410 if (ft.nbcon_atomic)
2411 nbcon_atomic_flush_pending();
2412
2413 if (ft.nbcon_offload)
2414 nbcon_kthreads_wake();
2415
2416 if (ft.legacy_direct) {
2417 /*
2418 * The caller may be holding system-critical or
2419 * timing-sensitive locks. Disable preemption during
2420 * printing of all remaining records to all consoles so that
2421 * this context can return as soon as possible. Hopefully
2422 * another printk() caller will take over the printing.
2423 */
2424 preempt_disable();
2425 /*
2426 * Try to acquire and then immediately release the console
2427 * semaphore. The release will print out buffers. With the
2428 * spinning variant, this context tries to take over the
2429 * printing from another printing context.
2430 */
2431 if (console_trylock_spinning())
2432 console_unlock();
2433 preempt_enable();
2434 }
2435
2436 if (ft.legacy_offload)
2437 defer_console_output();
2438 else
2439 wake_up_klogd();
2440
2441 return printed_len;
2442}
2443EXPORT_SYMBOL(vprintk_emit);
2444
2445int vprintk_default(const char *fmt, va_list args)
2446{
2447 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2448}
2449EXPORT_SYMBOL_GPL(vprintk_default);
2450
2451asmlinkage __visible int _printk(const char *fmt, ...)
2452{
2453 va_list args;
2454 int r;
2455
2456 va_start(args, fmt);
2457 r = vprintk(fmt, args);
2458 va_end(args);
2459
2460 return r;
2461}
2462EXPORT_SYMBOL(_printk);
2463
2464static bool pr_flush(int timeout_ms, bool reset_on_progress);
2465static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress);
2466
2467#else /* CONFIG_PRINTK */
2468
2469#define printk_time false
2470
2471#define prb_read_valid(rb, seq, r) false
2472#define prb_first_valid_seq(rb) 0
2473#define prb_next_seq(rb) 0
2474
2475static u64 syslog_seq;
2476
2477static bool pr_flush(int timeout_ms, bool reset_on_progress) { return true; }
2478static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress) { return true; }
2479
2480#endif /* CONFIG_PRINTK */
2481
2482#ifdef CONFIG_EARLY_PRINTK
2483struct console *early_console;
2484
2485asmlinkage __visible void early_printk(const char *fmt, ...)
2486{
2487 va_list ap;
2488 char buf[512];
2489 int n;
2490
2491 if (!early_console)
2492 return;
2493
2494 va_start(ap, fmt);
2495 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2496 va_end(ap);
2497
2498 early_console->write(early_console, buf, n);
2499}
2500#endif
2501
2502static void set_user_specified(struct console_cmdline *c, bool user_specified)
2503{
2504 if (!user_specified)
2505 return;
2506
2507 /*
2508 * @c console was defined by the user on the command line.
2509 * Do not clear when added twice also by SPCR or the device tree.
2510 */
2511 c->user_specified = true;
2512 /* At least one console defined by the user on the command line. */
2513 console_set_on_cmdline = 1;
2514}
2515
2516static int __add_preferred_console(const char *name, const short idx,
2517 const char *devname, char *options,
2518 char *brl_options, bool user_specified)
2519{
2520 struct console_cmdline *c;
2521 int i;
2522
2523 if (!name && !devname)
2524 return -EINVAL;
2525
2526 /*
2527 * We use a signed short index for struct console for device drivers to
2528 * indicate a not yet assigned index or port. However, a negative index
2529 * value is not valid when the console name and index are defined on
2530 * the command line.
2531 */
2532 if (name && idx < 0)
2533 return -EINVAL;
2534
2535 /*
2536 * See if this tty is not yet registered, and
2537 * if we have a slot free.
2538 */
2539 for (i = 0, c = console_cmdline;
2540 i < MAX_CMDLINECONSOLES && (c->name[0] || c->devname[0]);
2541 i++, c++) {
2542 if ((name && strcmp(c->name, name) == 0 && c->index == idx) ||
2543 (devname && strcmp(c->devname, devname) == 0)) {
2544 if (!brl_options)
2545 preferred_console = i;
2546 set_user_specified(c, user_specified);
2547 return 0;
2548 }
2549 }
2550 if (i == MAX_CMDLINECONSOLES)
2551 return -E2BIG;
2552 if (!brl_options)
2553 preferred_console = i;
2554 if (name)
2555 strscpy(c->name, name);
2556 if (devname)
2557 strscpy(c->devname, devname);
2558 c->options = options;
2559 set_user_specified(c, user_specified);
2560 braille_set_options(c, brl_options);
2561
2562 c->index = idx;
2563 return 0;
2564}
2565
2566static int __init console_msg_format_setup(char *str)
2567{
2568 if (!strcmp(str, "syslog"))
2569 console_msg_format = MSG_FORMAT_SYSLOG;
2570 if (!strcmp(str, "default"))
2571 console_msg_format = MSG_FORMAT_DEFAULT;
2572 return 1;
2573}
2574__setup("console_msg_format=", console_msg_format_setup);
2575
2576/*
2577 * Set up a console. Called via do_early_param() in init/main.c
2578 * for each "console=" parameter in the boot command line.
2579 */
2580static int __init console_setup(char *str)
2581{
2582 static_assert(sizeof(console_cmdline[0].devname) >= sizeof(console_cmdline[0].name) + 4);
2583 char buf[sizeof(console_cmdline[0].devname)];
2584 char *brl_options = NULL;
2585 char *ttyname = NULL;
2586 char *devname = NULL;
2587 char *options;
2588 char *s;
2589 int idx;
2590
2591 /*
2592 * console="" or console=null have been suggested as a way to
2593 * disable console output. Use ttynull that has been created
2594 * for exactly this purpose.
2595 */
2596 if (str[0] == 0 || strcmp(str, "null") == 0) {
2597 __add_preferred_console("ttynull", 0, NULL, NULL, NULL, true);
2598 return 1;
2599 }
2600
2601 if (_braille_console_setup(&str, &brl_options))
2602 return 1;
2603
2604 /* For a DEVNAME:0.0 style console the character device is unknown early */
2605 if (strchr(str, ':'))
2606 devname = buf;
2607 else
2608 ttyname = buf;
2609
2610 /*
2611 * Decode str into name, index, options.
2612 */
2613 if (ttyname && isdigit(str[0]))
2614 scnprintf(buf, sizeof(buf), "ttyS%s", str);
2615 else
2616 strscpy(buf, str);
2617
2618 options = strchr(str, ',');
2619 if (options)
2620 *(options++) = 0;
2621
2622#ifdef __sparc__
2623 if (!strcmp(str, "ttya"))
2624 strscpy(buf, "ttyS0");
2625 if (!strcmp(str, "ttyb"))
2626 strscpy(buf, "ttyS1");
2627#endif
2628
2629 for (s = buf; *s; s++)
2630 if ((ttyname && isdigit(*s)) || *s == ',')
2631 break;
2632
2633 /* @idx will get defined when devname matches. */
2634 if (devname)
2635 idx = -1;
2636 else
2637 idx = simple_strtoul(s, NULL, 10);
2638
2639 *s = 0;
2640
2641 __add_preferred_console(ttyname, idx, devname, options, brl_options, true);
2642 return 1;
2643}
2644__setup("console=", console_setup);
2645
2646/**
2647 * add_preferred_console - add a device to the list of preferred consoles.
2648 * @name: device name
2649 * @idx: device index
2650 * @options: options for this console
2651 *
2652 * The last preferred console added will be used for kernel messages
2653 * and stdin/out/err for init. Normally this is used by console_setup
2654 * above to handle user-supplied console arguments; however it can also
2655 * be used by arch-specific code either to override the user or more
2656 * commonly to provide a default console (ie from PROM variables) when
2657 * the user has not supplied one.
2658 */
2659int add_preferred_console(const char *name, const short idx, char *options)
2660{
2661 return __add_preferred_console(name, idx, NULL, options, NULL, false);
2662}
2663
2664/**
2665 * match_devname_and_update_preferred_console - Update a preferred console
2666 * when matching devname is found.
2667 * @devname: DEVNAME:0.0 style device name
2668 * @name: Name of the corresponding console driver, e.g. "ttyS"
2669 * @idx: Console index, e.g. port number.
2670 *
2671 * The function checks whether a device with the given @devname is
2672 * preferred via the console=DEVNAME:0.0 command line option.
2673 * It fills the missing console driver name and console index
2674 * so that a later register_console() call could find (match)
2675 * and enable this device.
2676 *
2677 * It might be used when a driver subsystem initializes particular
2678 * devices with already known DEVNAME:0.0 style names. And it
2679 * could predict which console driver name and index this device
2680 * would later get associated with.
2681 *
2682 * Return: 0 on success, negative error code on failure.
2683 */
2684int match_devname_and_update_preferred_console(const char *devname,
2685 const char *name,
2686 const short idx)
2687{
2688 struct console_cmdline *c = console_cmdline;
2689 int i;
2690
2691 if (!devname || !strlen(devname) || !name || !strlen(name) || idx < 0)
2692 return -EINVAL;
2693
2694 for (i = 0; i < MAX_CMDLINECONSOLES && (c->name[0] || c->devname[0]);
2695 i++, c++) {
2696 if (!strcmp(devname, c->devname)) {
2697 pr_info("associate the preferred console \"%s\" with \"%s%d\"\n",
2698 devname, name, idx);
2699 strscpy(c->name, name);
2700 c->index = idx;
2701 return 0;
2702 }
2703 }
2704
2705 return -ENOENT;
2706}
2707EXPORT_SYMBOL_GPL(match_devname_and_update_preferred_console);
2708
2709bool console_suspend_enabled = true;
2710EXPORT_SYMBOL(console_suspend_enabled);
2711
2712static int __init console_suspend_disable(char *str)
2713{
2714 console_suspend_enabled = false;
2715 return 1;
2716}
2717__setup("no_console_suspend", console_suspend_disable);
2718module_param_named(console_suspend, console_suspend_enabled,
2719 bool, S_IRUGO | S_IWUSR);
2720MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2721 " and hibernate operations");
2722
2723static bool printk_console_no_auto_verbose;
2724
2725void console_verbose(void)
2726{
2727 if (console_loglevel && !printk_console_no_auto_verbose)
2728 console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2729}
2730EXPORT_SYMBOL_GPL(console_verbose);
2731
2732module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2733MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2734
2735/**
2736 * suspend_console - suspend the console subsystem
2737 *
2738 * This disables printk() while we go into suspend states
2739 */
2740void suspend_console(void)
2741{
2742 struct console *con;
2743
2744 if (!console_suspend_enabled)
2745 return;
2746 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2747 pr_flush(1000, true);
2748
2749 console_list_lock();
2750 for_each_console(con)
2751 console_srcu_write_flags(con, con->flags | CON_SUSPENDED);
2752 console_list_unlock();
2753
2754 /*
2755 * Ensure that all SRCU list walks have completed. All printing
2756 * contexts must be able to see that they are suspended so that it
2757 * is guaranteed that all printing has stopped when this function
2758 * completes.
2759 */
2760 synchronize_srcu(&console_srcu);
2761}
2762
2763void resume_console(void)
2764{
2765 struct console_flush_type ft;
2766 struct console *con;
2767
2768 if (!console_suspend_enabled)
2769 return;
2770
2771 console_list_lock();
2772 for_each_console(con)
2773 console_srcu_write_flags(con, con->flags & ~CON_SUSPENDED);
2774 console_list_unlock();
2775
2776 /*
2777 * Ensure that all SRCU list walks have completed. All printing
2778 * contexts must be able to see they are no longer suspended so
2779 * that they are guaranteed to wake up and resume printing.
2780 */
2781 synchronize_srcu(&console_srcu);
2782
2783 printk_get_console_flush_type(&ft);
2784 if (ft.nbcon_offload)
2785 nbcon_kthreads_wake();
2786 if (ft.legacy_offload)
2787 defer_console_output();
2788
2789 pr_flush(1000, true);
2790}
2791
2792/**
2793 * console_cpu_notify - print deferred console messages after CPU hotplug
2794 * @cpu: unused
2795 *
2796 * If printk() is called from a CPU that is not online yet, the messages
2797 * will be printed on the console only if there are CON_ANYTIME consoles.
2798 * This function is called when a new CPU comes online (or fails to come
2799 * up) or goes offline.
2800 */
2801static int console_cpu_notify(unsigned int cpu)
2802{
2803 struct console_flush_type ft;
2804
2805 if (!cpuhp_tasks_frozen) {
2806 printk_get_console_flush_type(&ft);
2807 if (ft.nbcon_atomic)
2808 nbcon_atomic_flush_pending();
2809 if (ft.legacy_direct) {
2810 if (console_trylock())
2811 console_unlock();
2812 }
2813 }
2814 return 0;
2815}
2816
2817/**
2818 * console_lock - block the console subsystem from printing
2819 *
2820 * Acquires a lock which guarantees that no consoles will
2821 * be in or enter their write() callback.
2822 *
2823 * Can sleep, returns nothing.
2824 */
2825void console_lock(void)
2826{
2827 might_sleep();
2828
2829 /* On panic, the console_lock must be left to the panic cpu. */
2830 while (other_cpu_in_panic())
2831 msleep(1000);
2832
2833 down_console_sem();
2834 console_locked = 1;
2835 console_may_schedule = 1;
2836}
2837EXPORT_SYMBOL(console_lock);
2838
2839/**
2840 * console_trylock - try to block the console subsystem from printing
2841 *
2842 * Try to acquire a lock which guarantees that no consoles will
2843 * be in or enter their write() callback.
2844 *
2845 * returns 1 on success, and 0 on failure to acquire the lock.
2846 */
2847int console_trylock(void)
2848{
2849 /* On panic, the console_lock must be left to the panic cpu. */
2850 if (other_cpu_in_panic())
2851 return 0;
2852 if (down_trylock_console_sem())
2853 return 0;
2854 console_locked = 1;
2855 console_may_schedule = 0;
2856 return 1;
2857}
2858EXPORT_SYMBOL(console_trylock);
2859
2860int is_console_locked(void)
2861{
2862 return console_locked;
2863}
2864EXPORT_SYMBOL(is_console_locked);
2865
2866static void __console_unlock(void)
2867{
2868 console_locked = 0;
2869 up_console_sem();
2870}
2871
2872#ifdef CONFIG_PRINTK
2873
2874/*
2875 * Prepend the message in @pmsg->pbufs->outbuf. This is achieved by shifting
2876 * the existing message over and inserting the scratchbuf message.
2877 *
2878 * @pmsg is the original printk message.
2879 * @fmt is the printf format of the message which will prepend the existing one.
2880 *
2881 * If there is not enough space in @pmsg->pbufs->outbuf, the existing
2882 * message text will be sufficiently truncated.
2883 *
2884 * If @pmsg->pbufs->outbuf is modified, @pmsg->outbuf_len is updated.
2885 */
2886__printf(2, 3)
2887static void console_prepend_message(struct printk_message *pmsg, const char *fmt, ...)
2888{
2889 struct printk_buffers *pbufs = pmsg->pbufs;
2890 const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2891 const size_t outbuf_sz = sizeof(pbufs->outbuf);
2892 char *scratchbuf = &pbufs->scratchbuf[0];
2893 char *outbuf = &pbufs->outbuf[0];
2894 va_list args;
2895 size_t len;
2896
2897 va_start(args, fmt);
2898 len = vscnprintf(scratchbuf, scratchbuf_sz, fmt, args);
2899 va_end(args);
2900
2901 /*
2902 * Make sure outbuf is sufficiently large before prepending.
2903 * Keep at least the prefix when the message must be truncated.
2904 * It is a rather theoretical problem when someone tries to
2905 * use a minimalist buffer.
2906 */
2907 if (WARN_ON_ONCE(len + PRINTK_PREFIX_MAX >= outbuf_sz))
2908 return;
2909
2910 if (pmsg->outbuf_len + len >= outbuf_sz) {
2911 /* Truncate the message, but keep it terminated. */
2912 pmsg->outbuf_len = outbuf_sz - (len + 1);
2913 outbuf[pmsg->outbuf_len] = 0;
2914 }
2915
2916 memmove(outbuf + len, outbuf, pmsg->outbuf_len + 1);
2917 memcpy(outbuf, scratchbuf, len);
2918 pmsg->outbuf_len += len;
2919}
2920
2921/*
2922 * Prepend the message in @pmsg->pbufs->outbuf with a "dropped message".
2923 * @pmsg->outbuf_len is updated appropriately.
2924 *
2925 * @pmsg is the printk message to prepend.
2926 *
2927 * @dropped is the dropped count to report in the dropped message.
2928 */
2929void console_prepend_dropped(struct printk_message *pmsg, unsigned long dropped)
2930{
2931 console_prepend_message(pmsg, "** %lu printk messages dropped **\n", dropped);
2932}
2933
2934/*
2935 * Prepend the message in @pmsg->pbufs->outbuf with a "replay message".
2936 * @pmsg->outbuf_len is updated appropriately.
2937 *
2938 * @pmsg is the printk message to prepend.
2939 */
2940void console_prepend_replay(struct printk_message *pmsg)
2941{
2942 console_prepend_message(pmsg, "** replaying previous printk message **\n");
2943}
2944
2945/*
2946 * Read and format the specified record (or a later record if the specified
2947 * record is not available).
2948 *
2949 * @pmsg will contain the formatted result. @pmsg->pbufs must point to a
2950 * struct printk_buffers.
2951 *
2952 * @seq is the record to read and format. If it is not available, the next
2953 * valid record is read.
2954 *
2955 * @is_extended specifies if the message should be formatted for extended
2956 * console output.
2957 *
2958 * @may_supress specifies if records may be skipped based on loglevel.
2959 *
2960 * Returns false if no record is available. Otherwise true and all fields
2961 * of @pmsg are valid. (See the documentation of struct printk_message
2962 * for information about the @pmsg fields.)
2963 */
2964bool printk_get_next_message(struct printk_message *pmsg, u64 seq,
2965 bool is_extended, bool may_suppress)
2966{
2967 struct printk_buffers *pbufs = pmsg->pbufs;
2968 const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2969 const size_t outbuf_sz = sizeof(pbufs->outbuf);
2970 char *scratchbuf = &pbufs->scratchbuf[0];
2971 char *outbuf = &pbufs->outbuf[0];
2972 struct printk_info info;
2973 struct printk_record r;
2974 size_t len = 0;
2975 bool force_con;
2976
2977 /*
2978 * Formatting extended messages requires a separate buffer, so use the
2979 * scratch buffer to read in the ringbuffer text.
2980 *
2981 * Formatting normal messages is done in-place, so read the ringbuffer
2982 * text directly into the output buffer.
2983 */
2984 if (is_extended)
2985 prb_rec_init_rd(&r, &info, scratchbuf, scratchbuf_sz);
2986 else
2987 prb_rec_init_rd(&r, &info, outbuf, outbuf_sz);
2988
2989 if (!prb_read_valid(prb, seq, &r))
2990 return false;
2991
2992 pmsg->seq = r.info->seq;
2993 pmsg->dropped = r.info->seq - seq;
2994 force_con = r.info->flags & LOG_FORCE_CON;
2995
2996 /*
2997 * Skip records that are not forced to be printed on consoles and that
2998 * has level above the console loglevel.
2999 */
3000 if (!force_con && may_suppress && suppress_message_printing(r.info->level))
3001 goto out;
3002
3003 if (is_extended) {
3004 len = info_print_ext_header(outbuf, outbuf_sz, r.info);
3005 len += msg_print_ext_body(outbuf + len, outbuf_sz - len,
3006 &r.text_buf[0], r.info->text_len, &r.info->dev_info);
3007 } else {
3008 len = record_print_text(&r, console_msg_format & MSG_FORMAT_SYSLOG, printk_time);
3009 }
3010out:
3011 pmsg->outbuf_len = len;
3012 return true;
3013}
3014
3015/*
3016 * Legacy console printing from printk() caller context does not respect
3017 * raw_spinlock/spinlock nesting. For !PREEMPT_RT the lockdep warning is a
3018 * false positive. For PREEMPT_RT the false positive condition does not
3019 * occur.
3020 *
3021 * This map is used to temporarily establish LD_WAIT_SLEEP context for the
3022 * console write() callback when legacy printing to avoid false positive
3023 * lockdep complaints, thus allowing lockdep to continue to function for
3024 * real issues.
3025 */
3026#ifdef CONFIG_PREEMPT_RT
3027static inline void printk_legacy_allow_spinlock_enter(void) { }
3028static inline void printk_legacy_allow_spinlock_exit(void) { }
3029#else
3030static DEFINE_WAIT_OVERRIDE_MAP(printk_legacy_map, LD_WAIT_SLEEP);
3031
3032static inline void printk_legacy_allow_spinlock_enter(void)
3033{
3034 lock_map_acquire_try(&printk_legacy_map);
3035}
3036
3037static inline void printk_legacy_allow_spinlock_exit(void)
3038{
3039 lock_map_release(&printk_legacy_map);
3040}
3041#endif /* CONFIG_PREEMPT_RT */
3042
3043/*
3044 * Used as the printk buffers for non-panic, serialized console printing.
3045 * This is for legacy (!CON_NBCON) as well as all boot (CON_BOOT) consoles.
3046 * Its usage requires the console_lock held.
3047 */
3048struct printk_buffers printk_shared_pbufs;
3049
3050/*
3051 * Print one record for the given console. The record printed is whatever
3052 * record is the next available record for the given console.
3053 *
3054 * @handover will be set to true if a printk waiter has taken over the
3055 * console_lock, in which case the caller is no longer holding both the
3056 * console_lock and the SRCU read lock. Otherwise it is set to false.
3057 *
3058 * @cookie is the cookie from the SRCU read lock.
3059 *
3060 * Returns false if the given console has no next record to print, otherwise
3061 * true.
3062 *
3063 * Requires the console_lock and the SRCU read lock.
3064 */
3065static bool console_emit_next_record(struct console *con, bool *handover, int cookie)
3066{
3067 bool is_extended = console_srcu_read_flags(con) & CON_EXTENDED;
3068 char *outbuf = &printk_shared_pbufs.outbuf[0];
3069 struct printk_message pmsg = {
3070 .pbufs = &printk_shared_pbufs,
3071 };
3072 unsigned long flags;
3073
3074 *handover = false;
3075
3076 if (!printk_get_next_message(&pmsg, con->seq, is_extended, true))
3077 return false;
3078
3079 con->dropped += pmsg.dropped;
3080
3081 /* Skip messages of formatted length 0. */
3082 if (pmsg.outbuf_len == 0) {
3083 con->seq = pmsg.seq + 1;
3084 goto skip;
3085 }
3086
3087 if (con->dropped && !is_extended) {
3088 console_prepend_dropped(&pmsg, con->dropped);
3089 con->dropped = 0;
3090 }
3091
3092 /* Write everything out to the hardware. */
3093
3094 if (force_legacy_kthread() && !panic_in_progress()) {
3095 /*
3096 * With forced threading this function is in a task context
3097 * (either legacy kthread or get_init_console_seq()). There
3098 * is no need for concern about printk reentrance, handovers,
3099 * or lockdep complaints.
3100 */
3101
3102 con->write(con, outbuf, pmsg.outbuf_len);
3103 con->seq = pmsg.seq + 1;
3104 } else {
3105 /*
3106 * While actively printing out messages, if another printk()
3107 * were to occur on another CPU, it may wait for this one to
3108 * finish. This task can not be preempted if there is a
3109 * waiter waiting to take over.
3110 *
3111 * Interrupts are disabled because the hand over to a waiter
3112 * must not be interrupted until the hand over is completed
3113 * (@console_waiter is cleared).
3114 */
3115 printk_safe_enter_irqsave(flags);
3116 console_lock_spinning_enable();
3117
3118 /* Do not trace print latency. */
3119 stop_critical_timings();
3120
3121 printk_legacy_allow_spinlock_enter();
3122 con->write(con, outbuf, pmsg.outbuf_len);
3123 printk_legacy_allow_spinlock_exit();
3124
3125 start_critical_timings();
3126
3127 con->seq = pmsg.seq + 1;
3128
3129 *handover = console_lock_spinning_disable_and_check(cookie);
3130 printk_safe_exit_irqrestore(flags);
3131 }
3132skip:
3133 return true;
3134}
3135
3136#else
3137
3138static bool console_emit_next_record(struct console *con, bool *handover, int cookie)
3139{
3140 *handover = false;
3141 return false;
3142}
3143
3144static inline void printk_kthreads_check_locked(void) { }
3145
3146#endif /* CONFIG_PRINTK */
3147
3148/*
3149 * Print out all remaining records to all consoles.
3150 *
3151 * @do_cond_resched is set by the caller. It can be true only in schedulable
3152 * context.
3153 *
3154 * @next_seq is set to the sequence number after the last available record.
3155 * The value is valid only when this function returns true. It means that all
3156 * usable consoles are completely flushed.
3157 *
3158 * @handover will be set to true if a printk waiter has taken over the
3159 * console_lock, in which case the caller is no longer holding the
3160 * console_lock. Otherwise it is set to false.
3161 *
3162 * Returns true when there was at least one usable console and all messages
3163 * were flushed to all usable consoles. A returned false informs the caller
3164 * that everything was not flushed (either there were no usable consoles or
3165 * another context has taken over printing or it is a panic situation and this
3166 * is not the panic CPU). Regardless the reason, the caller should assume it
3167 * is not useful to immediately try again.
3168 *
3169 * Requires the console_lock.
3170 */
3171static bool console_flush_all(bool do_cond_resched, u64 *next_seq, bool *handover)
3172{
3173 struct console_flush_type ft;
3174 bool any_usable = false;
3175 struct console *con;
3176 bool any_progress;
3177 int cookie;
3178
3179 *next_seq = 0;
3180 *handover = false;
3181
3182 do {
3183 any_progress = false;
3184
3185 printk_get_console_flush_type(&ft);
3186
3187 cookie = console_srcu_read_lock();
3188 for_each_console_srcu(con) {
3189 short flags = console_srcu_read_flags(con);
3190 u64 printk_seq;
3191 bool progress;
3192
3193 /*
3194 * console_flush_all() is only responsible for nbcon
3195 * consoles when the nbcon consoles cannot print via
3196 * their atomic or threaded flushing.
3197 */
3198 if ((flags & CON_NBCON) && (ft.nbcon_atomic || ft.nbcon_offload))
3199 continue;
3200
3201 if (!console_is_usable(con, flags, !do_cond_resched))
3202 continue;
3203 any_usable = true;
3204
3205 if (flags & CON_NBCON) {
3206 progress = nbcon_legacy_emit_next_record(con, handover, cookie,
3207 !do_cond_resched);
3208 printk_seq = nbcon_seq_read(con);
3209 } else {
3210 progress = console_emit_next_record(con, handover, cookie);
3211 printk_seq = con->seq;
3212 }
3213
3214 /*
3215 * If a handover has occurred, the SRCU read lock
3216 * is already released.
3217 */
3218 if (*handover)
3219 return false;
3220
3221 /* Track the next of the highest seq flushed. */
3222 if (printk_seq > *next_seq)
3223 *next_seq = printk_seq;
3224
3225 if (!progress)
3226 continue;
3227 any_progress = true;
3228
3229 /* Allow panic_cpu to take over the consoles safely. */
3230 if (other_cpu_in_panic())
3231 goto abandon;
3232
3233 if (do_cond_resched)
3234 cond_resched();
3235 }
3236 console_srcu_read_unlock(cookie);
3237 } while (any_progress);
3238
3239 return any_usable;
3240
3241abandon:
3242 console_srcu_read_unlock(cookie);
3243 return false;
3244}
3245
3246static void __console_flush_and_unlock(void)
3247{
3248 bool do_cond_resched;
3249 bool handover;
3250 bool flushed;
3251 u64 next_seq;
3252
3253 /*
3254 * Console drivers are called with interrupts disabled, so
3255 * @console_may_schedule should be cleared before; however, we may
3256 * end up dumping a lot of lines, for example, if called from
3257 * console registration path, and should invoke cond_resched()
3258 * between lines if allowable. Not doing so can cause a very long
3259 * scheduling stall on a slow console leading to RCU stall and
3260 * softlockup warnings which exacerbate the issue with more
3261 * messages practically incapacitating the system. Therefore, create
3262 * a local to use for the printing loop.
3263 */
3264 do_cond_resched = console_may_schedule;
3265
3266 do {
3267 console_may_schedule = 0;
3268
3269 flushed = console_flush_all(do_cond_resched, &next_seq, &handover);
3270 if (!handover)
3271 __console_unlock();
3272
3273 /*
3274 * Abort if there was a failure to flush all messages to all
3275 * usable consoles. Either it is not possible to flush (in
3276 * which case it would be an infinite loop of retrying) or
3277 * another context has taken over printing.
3278 */
3279 if (!flushed)
3280 break;
3281
3282 /*
3283 * Some context may have added new records after
3284 * console_flush_all() but before unlocking the console.
3285 * Re-check if there is a new record to flush. If the trylock
3286 * fails, another context is already handling the printing.
3287 */
3288 } while (prb_read_valid(prb, next_seq, NULL) && console_trylock());
3289}
3290
3291/**
3292 * console_unlock - unblock the legacy console subsystem from printing
3293 *
3294 * Releases the console_lock which the caller holds to block printing of
3295 * the legacy console subsystem.
3296 *
3297 * While the console_lock was held, console output may have been buffered
3298 * by printk(). If this is the case, console_unlock() emits the output on
3299 * legacy consoles prior to releasing the lock.
3300 *
3301 * console_unlock(); may be called from any context.
3302 */
3303void console_unlock(void)
3304{
3305 struct console_flush_type ft;
3306
3307 printk_get_console_flush_type(&ft);
3308 if (ft.legacy_direct)
3309 __console_flush_and_unlock();
3310 else
3311 __console_unlock();
3312}
3313EXPORT_SYMBOL(console_unlock);
3314
3315/**
3316 * console_conditional_schedule - yield the CPU if required
3317 *
3318 * If the console code is currently allowed to sleep, and
3319 * if this CPU should yield the CPU to another task, do
3320 * so here.
3321 *
3322 * Must be called within console_lock();.
3323 */
3324void __sched console_conditional_schedule(void)
3325{
3326 if (console_may_schedule)
3327 cond_resched();
3328}
3329EXPORT_SYMBOL(console_conditional_schedule);
3330
3331void console_unblank(void)
3332{
3333 bool found_unblank = false;
3334 struct console *c;
3335 int cookie;
3336
3337 /*
3338 * First check if there are any consoles implementing the unblank()
3339 * callback. If not, there is no reason to continue and take the
3340 * console lock, which in particular can be dangerous if
3341 * @oops_in_progress is set.
3342 */
3343 cookie = console_srcu_read_lock();
3344 for_each_console_srcu(c) {
3345 if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank) {
3346 found_unblank = true;
3347 break;
3348 }
3349 }
3350 console_srcu_read_unlock(cookie);
3351 if (!found_unblank)
3352 return;
3353
3354 /*
3355 * Stop console printing because the unblank() callback may
3356 * assume the console is not within its write() callback.
3357 *
3358 * If @oops_in_progress is set, this may be an atomic context.
3359 * In that case, attempt a trylock as best-effort.
3360 */
3361 if (oops_in_progress) {
3362 /* Semaphores are not NMI-safe. */
3363 if (in_nmi())
3364 return;
3365
3366 /*
3367 * Attempting to trylock the console lock can deadlock
3368 * if another CPU was stopped while modifying the
3369 * semaphore. "Hope and pray" that this is not the
3370 * current situation.
3371 */
3372 if (down_trylock_console_sem() != 0)
3373 return;
3374 } else
3375 console_lock();
3376
3377 console_locked = 1;
3378 console_may_schedule = 0;
3379
3380 cookie = console_srcu_read_lock();
3381 for_each_console_srcu(c) {
3382 if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank)
3383 c->unblank();
3384 }
3385 console_srcu_read_unlock(cookie);
3386
3387 console_unlock();
3388
3389 if (!oops_in_progress)
3390 pr_flush(1000, true);
3391}
3392
3393/*
3394 * Rewind all consoles to the oldest available record.
3395 *
3396 * IMPORTANT: The function is safe only when called under
3397 * console_lock(). It is not enforced because
3398 * it is used as a best effort in panic().
3399 */
3400static void __console_rewind_all(void)
3401{
3402 struct console *c;
3403 short flags;
3404 int cookie;
3405 u64 seq;
3406
3407 seq = prb_first_valid_seq(prb);
3408
3409 cookie = console_srcu_read_lock();
3410 for_each_console_srcu(c) {
3411 flags = console_srcu_read_flags(c);
3412
3413 if (flags & CON_NBCON) {
3414 nbcon_seq_force(c, seq);
3415 } else {
3416 /*
3417 * This assignment is safe only when called under
3418 * console_lock(). On panic, legacy consoles are
3419 * only best effort.
3420 */
3421 c->seq = seq;
3422 }
3423 }
3424 console_srcu_read_unlock(cookie);
3425}
3426
3427/**
3428 * console_flush_on_panic - flush console content on panic
3429 * @mode: flush all messages in buffer or just the pending ones
3430 *
3431 * Immediately output all pending messages no matter what.
3432 */
3433void console_flush_on_panic(enum con_flush_mode mode)
3434{
3435 struct console_flush_type ft;
3436 bool handover;
3437 u64 next_seq;
3438
3439 /*
3440 * Ignore the console lock and flush out the messages. Attempting a
3441 * trylock would not be useful because:
3442 *
3443 * - if it is contended, it must be ignored anyway
3444 * - console_lock() and console_trylock() block and fail
3445 * respectively in panic for non-panic CPUs
3446 * - semaphores are not NMI-safe
3447 */
3448
3449 /*
3450 * If another context is holding the console lock,
3451 * @console_may_schedule might be set. Clear it so that
3452 * this context does not call cond_resched() while flushing.
3453 */
3454 console_may_schedule = 0;
3455
3456 if (mode == CONSOLE_REPLAY_ALL)
3457 __console_rewind_all();
3458
3459 printk_get_console_flush_type(&ft);
3460 if (ft.nbcon_atomic)
3461 nbcon_atomic_flush_pending();
3462
3463 /* Flush legacy consoles once allowed, even when dangerous. */
3464 if (legacy_allow_panic_sync)
3465 console_flush_all(false, &next_seq, &handover);
3466}
3467
3468/*
3469 * Return the console tty driver structure and its associated index
3470 */
3471struct tty_driver *console_device(int *index)
3472{
3473 struct console *c;
3474 struct tty_driver *driver = NULL;
3475 int cookie;
3476
3477 /*
3478 * Take console_lock to serialize device() callback with
3479 * other console operations. For example, fg_console is
3480 * modified under console_lock when switching vt.
3481 */
3482 console_lock();
3483
3484 cookie = console_srcu_read_lock();
3485 for_each_console_srcu(c) {
3486 if (!c->device)
3487 continue;
3488 driver = c->device(c, index);
3489 if (driver)
3490 break;
3491 }
3492 console_srcu_read_unlock(cookie);
3493
3494 console_unlock();
3495 return driver;
3496}
3497
3498/*
3499 * Prevent further output on the passed console device so that (for example)
3500 * serial drivers can disable console output before suspending a port, and can
3501 * re-enable output afterwards.
3502 */
3503void console_stop(struct console *console)
3504{
3505 __pr_flush(console, 1000, true);
3506 console_list_lock();
3507 console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3508 console_list_unlock();
3509
3510 /*
3511 * Ensure that all SRCU list walks have completed. All contexts must
3512 * be able to see that this console is disabled so that (for example)
3513 * the caller can suspend the port without risk of another context
3514 * using the port.
3515 */
3516 synchronize_srcu(&console_srcu);
3517}
3518EXPORT_SYMBOL(console_stop);
3519
3520void console_start(struct console *console)
3521{
3522 struct console_flush_type ft;
3523 bool is_nbcon;
3524
3525 console_list_lock();
3526 console_srcu_write_flags(console, console->flags | CON_ENABLED);
3527 is_nbcon = console->flags & CON_NBCON;
3528 console_list_unlock();
3529
3530 /*
3531 * Ensure that all SRCU list walks have completed. The related
3532 * printing context must be able to see it is enabled so that
3533 * it is guaranteed to wake up and resume printing.
3534 */
3535 synchronize_srcu(&console_srcu);
3536
3537 printk_get_console_flush_type(&ft);
3538 if (is_nbcon && ft.nbcon_offload)
3539 nbcon_kthread_wake(console);
3540 else if (ft.legacy_offload)
3541 defer_console_output();
3542
3543 __pr_flush(console, 1000, true);
3544}
3545EXPORT_SYMBOL(console_start);
3546
3547#ifdef CONFIG_PRINTK
3548static int unregister_console_locked(struct console *console);
3549
3550/* True when system boot is far enough to create printer threads. */
3551static bool printk_kthreads_ready __ro_after_init;
3552
3553static struct task_struct *printk_legacy_kthread;
3554
3555static bool legacy_kthread_should_wakeup(void)
3556{
3557 struct console_flush_type ft;
3558 struct console *con;
3559 bool ret = false;
3560 int cookie;
3561
3562 if (kthread_should_stop())
3563 return true;
3564
3565 printk_get_console_flush_type(&ft);
3566
3567 cookie = console_srcu_read_lock();
3568 for_each_console_srcu(con) {
3569 short flags = console_srcu_read_flags(con);
3570 u64 printk_seq;
3571
3572 /*
3573 * The legacy printer thread is only responsible for nbcon
3574 * consoles when the nbcon consoles cannot print via their
3575 * atomic or threaded flushing.
3576 */
3577 if ((flags & CON_NBCON) && (ft.nbcon_atomic || ft.nbcon_offload))
3578 continue;
3579
3580 if (!console_is_usable(con, flags, false))
3581 continue;
3582
3583 if (flags & CON_NBCON) {
3584 printk_seq = nbcon_seq_read(con);
3585 } else {
3586 /*
3587 * It is safe to read @seq because only this
3588 * thread context updates @seq.
3589 */
3590 printk_seq = con->seq;
3591 }
3592
3593 if (prb_read_valid(prb, printk_seq, NULL)) {
3594 ret = true;
3595 break;
3596 }
3597 }
3598 console_srcu_read_unlock(cookie);
3599
3600 return ret;
3601}
3602
3603static int legacy_kthread_func(void *unused)
3604{
3605 for (;;) {
3606 wait_event_interruptible(legacy_wait, legacy_kthread_should_wakeup());
3607
3608 if (kthread_should_stop())
3609 break;
3610
3611 console_lock();
3612 __console_flush_and_unlock();
3613 }
3614
3615 return 0;
3616}
3617
3618static bool legacy_kthread_create(void)
3619{
3620 struct task_struct *kt;
3621
3622 lockdep_assert_console_list_lock_held();
3623
3624 kt = kthread_run(legacy_kthread_func, NULL, "pr/legacy");
3625 if (WARN_ON(IS_ERR(kt))) {
3626 pr_err("failed to start legacy printing thread\n");
3627 return false;
3628 }
3629
3630 printk_legacy_kthread = kt;
3631
3632 /*
3633 * It is important that console printing threads are scheduled
3634 * shortly after a printk call and with generous runtime budgets.
3635 */
3636 sched_set_normal(printk_legacy_kthread, -20);
3637
3638 return true;
3639}
3640
3641/**
3642 * printk_kthreads_shutdown - shutdown all threaded printers
3643 *
3644 * On system shutdown all threaded printers are stopped. This allows printk
3645 * to transition back to atomic printing, thus providing a robust mechanism
3646 * for the final shutdown/reboot messages to be output.
3647 */
3648static void printk_kthreads_shutdown(void)
3649{
3650 struct console *con;
3651
3652 console_list_lock();
3653 if (printk_kthreads_running) {
3654 printk_kthreads_running = false;
3655
3656 for_each_console(con) {
3657 if (con->flags & CON_NBCON)
3658 nbcon_kthread_stop(con);
3659 }
3660
3661 /*
3662 * The threads may have been stopped while printing a
3663 * backlog. Flush any records left over.
3664 */
3665 nbcon_atomic_flush_pending();
3666 }
3667 console_list_unlock();
3668}
3669
3670static struct syscore_ops printk_syscore_ops = {
3671 .shutdown = printk_kthreads_shutdown,
3672};
3673
3674/*
3675 * If appropriate, start nbcon kthreads and set @printk_kthreads_running.
3676 * If any kthreads fail to start, those consoles are unregistered.
3677 *
3678 * Must be called under console_list_lock().
3679 */
3680static void printk_kthreads_check_locked(void)
3681{
3682 struct hlist_node *tmp;
3683 struct console *con;
3684
3685 lockdep_assert_console_list_lock_held();
3686
3687 if (!printk_kthreads_ready)
3688 return;
3689
3690 if (have_legacy_console || have_boot_console) {
3691 if (!printk_legacy_kthread &&
3692 force_legacy_kthread() &&
3693 !legacy_kthread_create()) {
3694 /*
3695 * All legacy consoles must be unregistered. If there
3696 * are any nbcon consoles, they will set up their own
3697 * kthread.
3698 */
3699 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3700 if (con->flags & CON_NBCON)
3701 continue;
3702
3703 unregister_console_locked(con);
3704 }
3705 }
3706 } else if (printk_legacy_kthread) {
3707 kthread_stop(printk_legacy_kthread);
3708 printk_legacy_kthread = NULL;
3709 }
3710
3711 /*
3712 * Printer threads cannot be started as long as any boot console is
3713 * registered because there is no way to synchronize the hardware
3714 * registers between boot console code and regular console code.
3715 * It can only be known that there will be no new boot consoles when
3716 * an nbcon console is registered.
3717 */
3718 if (have_boot_console || !have_nbcon_console) {
3719 /* Clear flag in case all nbcon consoles unregistered. */
3720 printk_kthreads_running = false;
3721 return;
3722 }
3723
3724 if (printk_kthreads_running)
3725 return;
3726
3727 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3728 if (!(con->flags & CON_NBCON))
3729 continue;
3730
3731 if (!nbcon_kthread_create(con))
3732 unregister_console_locked(con);
3733 }
3734
3735 printk_kthreads_running = true;
3736}
3737
3738static int __init printk_set_kthreads_ready(void)
3739{
3740 register_syscore_ops(&printk_syscore_ops);
3741
3742 console_list_lock();
3743 printk_kthreads_ready = true;
3744 printk_kthreads_check_locked();
3745 console_list_unlock();
3746
3747 return 0;
3748}
3749early_initcall(printk_set_kthreads_ready);
3750#endif /* CONFIG_PRINTK */
3751
3752static int __read_mostly keep_bootcon;
3753
3754static int __init keep_bootcon_setup(char *str)
3755{
3756 keep_bootcon = 1;
3757 pr_info("debug: skip boot console de-registration.\n");
3758
3759 return 0;
3760}
3761
3762early_param("keep_bootcon", keep_bootcon_setup);
3763
3764static int console_call_setup(struct console *newcon, char *options)
3765{
3766 int err;
3767
3768 if (!newcon->setup)
3769 return 0;
3770
3771 /* Synchronize with possible boot console. */
3772 console_lock();
3773 err = newcon->setup(newcon, options);
3774 console_unlock();
3775
3776 return err;
3777}
3778
3779/*
3780 * This is called by register_console() to try to match
3781 * the newly registered console with any of the ones selected
3782 * by either the command line or add_preferred_console() and
3783 * setup/enable it.
3784 *
3785 * Care need to be taken with consoles that are statically
3786 * enabled such as netconsole
3787 */
3788static int try_enable_preferred_console(struct console *newcon,
3789 bool user_specified)
3790{
3791 struct console_cmdline *c;
3792 int i, err;
3793
3794 for (i = 0, c = console_cmdline;
3795 i < MAX_CMDLINECONSOLES && (c->name[0] || c->devname[0]);
3796 i++, c++) {
3797 /* Console not yet initialized? */
3798 if (!c->name[0])
3799 continue;
3800 if (c->user_specified != user_specified)
3801 continue;
3802 if (!newcon->match ||
3803 newcon->match(newcon, c->name, c->index, c->options) != 0) {
3804 /* default matching */
3805 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
3806 if (strcmp(c->name, newcon->name) != 0)
3807 continue;
3808 if (newcon->index >= 0 &&
3809 newcon->index != c->index)
3810 continue;
3811 if (newcon->index < 0)
3812 newcon->index = c->index;
3813
3814 if (_braille_register_console(newcon, c))
3815 return 0;
3816
3817 err = console_call_setup(newcon, c->options);
3818 if (err)
3819 return err;
3820 }
3821 newcon->flags |= CON_ENABLED;
3822 if (i == preferred_console)
3823 newcon->flags |= CON_CONSDEV;
3824 return 0;
3825 }
3826
3827 /*
3828 * Some consoles, such as pstore and netconsole, can be enabled even
3829 * without matching. Accept the pre-enabled consoles only when match()
3830 * and setup() had a chance to be called.
3831 */
3832 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
3833 return 0;
3834
3835 return -ENOENT;
3836}
3837
3838/* Try to enable the console unconditionally */
3839static void try_enable_default_console(struct console *newcon)
3840{
3841 if (newcon->index < 0)
3842 newcon->index = 0;
3843
3844 if (console_call_setup(newcon, NULL) != 0)
3845 return;
3846
3847 newcon->flags |= CON_ENABLED;
3848
3849 if (newcon->device)
3850 newcon->flags |= CON_CONSDEV;
3851}
3852
3853/* Return the starting sequence number for a newly registered console. */
3854static u64 get_init_console_seq(struct console *newcon, bool bootcon_registered)
3855{
3856 struct console *con;
3857 bool handover;
3858 u64 init_seq;
3859
3860 if (newcon->flags & (CON_PRINTBUFFER | CON_BOOT)) {
3861 /* Get a consistent copy of @syslog_seq. */
3862 mutex_lock(&syslog_lock);
3863 init_seq = syslog_seq;
3864 mutex_unlock(&syslog_lock);
3865 } else {
3866 /* Begin with next message added to ringbuffer. */
3867 init_seq = prb_next_seq(prb);
3868
3869 /*
3870 * If any enabled boot consoles are due to be unregistered
3871 * shortly, some may not be caught up and may be the same
3872 * device as @newcon. Since it is not known which boot console
3873 * is the same device, flush all consoles and, if necessary,
3874 * start with the message of the enabled boot console that is
3875 * the furthest behind.
3876 */
3877 if (bootcon_registered && !keep_bootcon) {
3878 /*
3879 * Hold the console_lock to stop console printing and
3880 * guarantee safe access to console->seq.
3881 */
3882 console_lock();
3883
3884 /*
3885 * Flush all consoles and set the console to start at
3886 * the next unprinted sequence number.
3887 */
3888 if (!console_flush_all(true, &init_seq, &handover)) {
3889 /*
3890 * Flushing failed. Just choose the lowest
3891 * sequence of the enabled boot consoles.
3892 */
3893
3894 /*
3895 * If there was a handover, this context no
3896 * longer holds the console_lock.
3897 */
3898 if (handover)
3899 console_lock();
3900
3901 init_seq = prb_next_seq(prb);
3902 for_each_console(con) {
3903 u64 seq;
3904
3905 if (!(con->flags & CON_BOOT) ||
3906 !(con->flags & CON_ENABLED)) {
3907 continue;
3908 }
3909
3910 if (con->flags & CON_NBCON)
3911 seq = nbcon_seq_read(con);
3912 else
3913 seq = con->seq;
3914
3915 if (seq < init_seq)
3916 init_seq = seq;
3917 }
3918 }
3919
3920 console_unlock();
3921 }
3922 }
3923
3924 return init_seq;
3925}
3926
3927#define console_first() \
3928 hlist_entry(console_list.first, struct console, node)
3929
3930static int unregister_console_locked(struct console *console);
3931
3932/*
3933 * The console driver calls this routine during kernel initialization
3934 * to register the console printing procedure with printk() and to
3935 * print any messages that were printed by the kernel before the
3936 * console driver was initialized.
3937 *
3938 * This can happen pretty early during the boot process (because of
3939 * early_printk) - sometimes before setup_arch() completes - be careful
3940 * of what kernel features are used - they may not be initialised yet.
3941 *
3942 * There are two types of consoles - bootconsoles (early_printk) and
3943 * "real" consoles (everything which is not a bootconsole) which are
3944 * handled differently.
3945 * - Any number of bootconsoles can be registered at any time.
3946 * - As soon as a "real" console is registered, all bootconsoles
3947 * will be unregistered automatically.
3948 * - Once a "real" console is registered, any attempt to register a
3949 * bootconsoles will be rejected
3950 */
3951void register_console(struct console *newcon)
3952{
3953 bool use_device_lock = (newcon->flags & CON_NBCON) && newcon->write_atomic;
3954 bool bootcon_registered = false;
3955 bool realcon_registered = false;
3956 struct console *con;
3957 unsigned long flags;
3958 u64 init_seq;
3959 int err;
3960
3961 console_list_lock();
3962
3963 for_each_console(con) {
3964 if (WARN(con == newcon, "console '%s%d' already registered\n",
3965 con->name, con->index)) {
3966 goto unlock;
3967 }
3968
3969 if (con->flags & CON_BOOT)
3970 bootcon_registered = true;
3971 else
3972 realcon_registered = true;
3973 }
3974
3975 /* Do not register boot consoles when there already is a real one. */
3976 if ((newcon->flags & CON_BOOT) && realcon_registered) {
3977 pr_info("Too late to register bootconsole %s%d\n",
3978 newcon->name, newcon->index);
3979 goto unlock;
3980 }
3981
3982 if (newcon->flags & CON_NBCON) {
3983 /*
3984 * Ensure the nbcon console buffers can be allocated
3985 * before modifying any global data.
3986 */
3987 if (!nbcon_alloc(newcon))
3988 goto unlock;
3989 }
3990
3991 /*
3992 * See if we want to enable this console driver by default.
3993 *
3994 * Nope when a console is preferred by the command line, device
3995 * tree, or SPCR.
3996 *
3997 * The first real console with tty binding (driver) wins. More
3998 * consoles might get enabled before the right one is found.
3999 *
4000 * Note that a console with tty binding will have CON_CONSDEV
4001 * flag set and will be first in the list.
4002 */
4003 if (preferred_console < 0) {
4004 if (hlist_empty(&console_list) || !console_first()->device ||
4005 console_first()->flags & CON_BOOT) {
4006 try_enable_default_console(newcon);
4007 }
4008 }
4009
4010 /* See if this console matches one we selected on the command line */
4011 err = try_enable_preferred_console(newcon, true);
4012
4013 /* If not, try to match against the platform default(s) */
4014 if (err == -ENOENT)
4015 err = try_enable_preferred_console(newcon, false);
4016
4017 /* printk() messages are not printed to the Braille console. */
4018 if (err || newcon->flags & CON_BRL) {
4019 if (newcon->flags & CON_NBCON)
4020 nbcon_free(newcon);
4021 goto unlock;
4022 }
4023
4024 /*
4025 * If we have a bootconsole, and are switching to a real console,
4026 * don't print everything out again, since when the boot console, and
4027 * the real console are the same physical device, it's annoying to
4028 * see the beginning boot messages twice
4029 */
4030 if (bootcon_registered &&
4031 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
4032 newcon->flags &= ~CON_PRINTBUFFER;
4033 }
4034
4035 newcon->dropped = 0;
4036 init_seq = get_init_console_seq(newcon, bootcon_registered);
4037
4038 if (newcon->flags & CON_NBCON) {
4039 have_nbcon_console = true;
4040 nbcon_seq_force(newcon, init_seq);
4041 } else {
4042 have_legacy_console = true;
4043 newcon->seq = init_seq;
4044 }
4045
4046 if (newcon->flags & CON_BOOT)
4047 have_boot_console = true;
4048
4049 /*
4050 * If another context is actively using the hardware of this new
4051 * console, it will not be aware of the nbcon synchronization. This
4052 * is a risk that two contexts could access the hardware
4053 * simultaneously if this new console is used for atomic printing
4054 * and the other context is still using the hardware.
4055 *
4056 * Use the driver synchronization to ensure that the hardware is not
4057 * in use while this new console transitions to being registered.
4058 */
4059 if (use_device_lock)
4060 newcon->device_lock(newcon, &flags);
4061
4062 /*
4063 * Put this console in the list - keep the
4064 * preferred driver at the head of the list.
4065 */
4066 if (hlist_empty(&console_list)) {
4067 /* Ensure CON_CONSDEV is always set for the head. */
4068 newcon->flags |= CON_CONSDEV;
4069 hlist_add_head_rcu(&newcon->node, &console_list);
4070
4071 } else if (newcon->flags & CON_CONSDEV) {
4072 /* Only the new head can have CON_CONSDEV set. */
4073 console_srcu_write_flags(console_first(), console_first()->flags & ~CON_CONSDEV);
4074 hlist_add_head_rcu(&newcon->node, &console_list);
4075
4076 } else {
4077 hlist_add_behind_rcu(&newcon->node, console_list.first);
4078 }
4079
4080 /*
4081 * No need to synchronize SRCU here! The caller does not rely
4082 * on all contexts being able to see the new console before
4083 * register_console() completes.
4084 */
4085
4086 /* This new console is now registered. */
4087 if (use_device_lock)
4088 newcon->device_unlock(newcon, flags);
4089
4090 console_sysfs_notify();
4091
4092 /*
4093 * By unregistering the bootconsoles after we enable the real console
4094 * we get the "console xxx enabled" message on all the consoles -
4095 * boot consoles, real consoles, etc - this is to ensure that end
4096 * users know there might be something in the kernel's log buffer that
4097 * went to the bootconsole (that they do not see on the real console)
4098 */
4099 con_printk(KERN_INFO, newcon, "enabled\n");
4100 if (bootcon_registered &&
4101 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
4102 !keep_bootcon) {
4103 struct hlist_node *tmp;
4104
4105 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
4106 if (con->flags & CON_BOOT)
4107 unregister_console_locked(con);
4108 }
4109 }
4110
4111 /* Changed console list, may require printer threads to start/stop. */
4112 printk_kthreads_check_locked();
4113unlock:
4114 console_list_unlock();
4115}
4116EXPORT_SYMBOL(register_console);
4117
4118/* Must be called under console_list_lock(). */
4119static int unregister_console_locked(struct console *console)
4120{
4121 bool use_device_lock = (console->flags & CON_NBCON) && console->write_atomic;
4122 bool found_legacy_con = false;
4123 bool found_nbcon_con = false;
4124 bool found_boot_con = false;
4125 unsigned long flags;
4126 struct console *c;
4127 int res;
4128
4129 lockdep_assert_console_list_lock_held();
4130
4131 con_printk(KERN_INFO, console, "disabled\n");
4132
4133 res = _braille_unregister_console(console);
4134 if (res < 0)
4135 return res;
4136 if (res > 0)
4137 return 0;
4138
4139 if (!console_is_registered_locked(console))
4140 res = -ENODEV;
4141 else if (console_is_usable(console, console->flags, true))
4142 __pr_flush(console, 1000, true);
4143
4144 /* Disable it unconditionally */
4145 console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
4146
4147 if (res < 0)
4148 return res;
4149
4150 /*
4151 * Use the driver synchronization to ensure that the hardware is not
4152 * in use while this console transitions to being unregistered.
4153 */
4154 if (use_device_lock)
4155 console->device_lock(console, &flags);
4156
4157 hlist_del_init_rcu(&console->node);
4158
4159 if (use_device_lock)
4160 console->device_unlock(console, flags);
4161
4162 /*
4163 * <HISTORICAL>
4164 * If this isn't the last console and it has CON_CONSDEV set, we
4165 * need to set it on the next preferred console.
4166 * </HISTORICAL>
4167 *
4168 * The above makes no sense as there is no guarantee that the next
4169 * console has any device attached. Oh well....
4170 */
4171 if (!hlist_empty(&console_list) && console->flags & CON_CONSDEV)
4172 console_srcu_write_flags(console_first(), console_first()->flags | CON_CONSDEV);
4173
4174 /*
4175 * Ensure that all SRCU list walks have completed. All contexts
4176 * must not be able to see this console in the list so that any
4177 * exit/cleanup routines can be performed safely.
4178 */
4179 synchronize_srcu(&console_srcu);
4180
4181 if (console->flags & CON_NBCON)
4182 nbcon_free(console);
4183
4184 console_sysfs_notify();
4185
4186 if (console->exit)
4187 res = console->exit(console);
4188
4189 /*
4190 * With this console gone, the global flags tracking registered
4191 * console types may have changed. Update them.
4192 */
4193 for_each_console(c) {
4194 if (c->flags & CON_BOOT)
4195 found_boot_con = true;
4196
4197 if (c->flags & CON_NBCON)
4198 found_nbcon_con = true;
4199 else
4200 found_legacy_con = true;
4201 }
4202 if (!found_boot_con)
4203 have_boot_console = found_boot_con;
4204 if (!found_legacy_con)
4205 have_legacy_console = found_legacy_con;
4206 if (!found_nbcon_con)
4207 have_nbcon_console = found_nbcon_con;
4208
4209 /* Changed console list, may require printer threads to start/stop. */
4210 printk_kthreads_check_locked();
4211
4212 return res;
4213}
4214
4215int unregister_console(struct console *console)
4216{
4217 int res;
4218
4219 console_list_lock();
4220 res = unregister_console_locked(console);
4221 console_list_unlock();
4222 return res;
4223}
4224EXPORT_SYMBOL(unregister_console);
4225
4226/**
4227 * console_force_preferred_locked - force a registered console preferred
4228 * @con: The registered console to force preferred.
4229 *
4230 * Must be called under console_list_lock().
4231 */
4232void console_force_preferred_locked(struct console *con)
4233{
4234 struct console *cur_pref_con;
4235
4236 if (!console_is_registered_locked(con))
4237 return;
4238
4239 cur_pref_con = console_first();
4240
4241 /* Already preferred? */
4242 if (cur_pref_con == con)
4243 return;
4244
4245 /*
4246 * Delete, but do not re-initialize the entry. This allows the console
4247 * to continue to appear registered (via any hlist_unhashed_lockless()
4248 * checks), even though it was briefly removed from the console list.
4249 */
4250 hlist_del_rcu(&con->node);
4251
4252 /*
4253 * Ensure that all SRCU list walks have completed so that the console
4254 * can be added to the beginning of the console list and its forward
4255 * list pointer can be re-initialized.
4256 */
4257 synchronize_srcu(&console_srcu);
4258
4259 con->flags |= CON_CONSDEV;
4260 WARN_ON(!con->device);
4261
4262 /* Only the new head can have CON_CONSDEV set. */
4263 console_srcu_write_flags(cur_pref_con, cur_pref_con->flags & ~CON_CONSDEV);
4264 hlist_add_head_rcu(&con->node, &console_list);
4265}
4266EXPORT_SYMBOL(console_force_preferred_locked);
4267
4268/*
4269 * Initialize the console device. This is called *early*, so
4270 * we can't necessarily depend on lots of kernel help here.
4271 * Just do some early initializations, and do the complex setup
4272 * later.
4273 */
4274void __init console_init(void)
4275{
4276 int ret;
4277 initcall_t call;
4278 initcall_entry_t *ce;
4279
4280 /* Setup the default TTY line discipline. */
4281 n_tty_init();
4282
4283 /*
4284 * set up the console device so that later boot sequences can
4285 * inform about problems etc..
4286 */
4287 ce = __con_initcall_start;
4288 trace_initcall_level("console");
4289 while (ce < __con_initcall_end) {
4290 call = initcall_from_entry(ce);
4291 trace_initcall_start(call);
4292 ret = call();
4293 trace_initcall_finish(call, ret);
4294 ce++;
4295 }
4296}
4297
4298/*
4299 * Some boot consoles access data that is in the init section and which will
4300 * be discarded after the initcalls have been run. To make sure that no code
4301 * will access this data, unregister the boot consoles in a late initcall.
4302 *
4303 * If for some reason, such as deferred probe or the driver being a loadable
4304 * module, the real console hasn't registered yet at this point, there will
4305 * be a brief interval in which no messages are logged to the console, which
4306 * makes it difficult to diagnose problems that occur during this time.
4307 *
4308 * To mitigate this problem somewhat, only unregister consoles whose memory
4309 * intersects with the init section. Note that all other boot consoles will
4310 * get unregistered when the real preferred console is registered.
4311 */
4312static int __init printk_late_init(void)
4313{
4314 struct hlist_node *tmp;
4315 struct console *con;
4316 int ret;
4317
4318 console_list_lock();
4319 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
4320 if (!(con->flags & CON_BOOT))
4321 continue;
4322
4323 /* Check addresses that might be used for enabled consoles. */
4324 if (init_section_intersects(con, sizeof(*con)) ||
4325 init_section_contains(con->write, 0) ||
4326 init_section_contains(con->read, 0) ||
4327 init_section_contains(con->device, 0) ||
4328 init_section_contains(con->unblank, 0) ||
4329 init_section_contains(con->data, 0)) {
4330 /*
4331 * Please, consider moving the reported consoles out
4332 * of the init section.
4333 */
4334 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
4335 con->name, con->index);
4336 unregister_console_locked(con);
4337 }
4338 }
4339 console_list_unlock();
4340
4341 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
4342 console_cpu_notify);
4343 WARN_ON(ret < 0);
4344 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
4345 console_cpu_notify, NULL);
4346 WARN_ON(ret < 0);
4347 printk_sysctl_init();
4348 return 0;
4349}
4350late_initcall(printk_late_init);
4351
4352#if defined CONFIG_PRINTK
4353/* If @con is specified, only wait for that console. Otherwise wait for all. */
4354static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
4355{
4356 unsigned long timeout_jiffies = msecs_to_jiffies(timeout_ms);
4357 unsigned long remaining_jiffies = timeout_jiffies;
4358 struct console_flush_type ft;
4359 struct console *c;
4360 u64 last_diff = 0;
4361 u64 printk_seq;
4362 short flags;
4363 int cookie;
4364 u64 diff;
4365 u64 seq;
4366
4367 /* Sorry, pr_flush() will not work this early. */
4368 if (system_state < SYSTEM_SCHEDULING)
4369 return false;
4370
4371 might_sleep();
4372
4373 seq = prb_next_reserve_seq(prb);
4374
4375 /* Flush the consoles so that records up to @seq are printed. */
4376 printk_get_console_flush_type(&ft);
4377 if (ft.nbcon_atomic)
4378 nbcon_atomic_flush_pending();
4379 if (ft.legacy_direct) {
4380 console_lock();
4381 console_unlock();
4382 }
4383
4384 for (;;) {
4385 unsigned long begin_jiffies;
4386 unsigned long slept_jiffies;
4387
4388 diff = 0;
4389
4390 /*
4391 * Hold the console_lock to guarantee safe access to
4392 * console->seq. Releasing console_lock flushes more
4393 * records in case @seq is still not printed on all
4394 * usable consoles.
4395 *
4396 * Holding the console_lock is not necessary if there
4397 * are no legacy or boot consoles. However, such a
4398 * console could register at any time. Always hold the
4399 * console_lock as a precaution rather than
4400 * synchronizing against register_console().
4401 */
4402 console_lock();
4403
4404 cookie = console_srcu_read_lock();
4405 for_each_console_srcu(c) {
4406 if (con && con != c)
4407 continue;
4408
4409 flags = console_srcu_read_flags(c);
4410
4411 /*
4412 * If consoles are not usable, it cannot be expected
4413 * that they make forward progress, so only increment
4414 * @diff for usable consoles.
4415 */
4416 if (!console_is_usable(c, flags, true) &&
4417 !console_is_usable(c, flags, false)) {
4418 continue;
4419 }
4420
4421 if (flags & CON_NBCON) {
4422 printk_seq = nbcon_seq_read(c);
4423 } else {
4424 printk_seq = c->seq;
4425 }
4426
4427 if (printk_seq < seq)
4428 diff += seq - printk_seq;
4429 }
4430 console_srcu_read_unlock(cookie);
4431
4432 if (diff != last_diff && reset_on_progress)
4433 remaining_jiffies = timeout_jiffies;
4434
4435 console_unlock();
4436
4437 /* Note: @diff is 0 if there are no usable consoles. */
4438 if (diff == 0 || remaining_jiffies == 0)
4439 break;
4440
4441 /* msleep(1) might sleep much longer. Check time by jiffies. */
4442 begin_jiffies = jiffies;
4443 msleep(1);
4444 slept_jiffies = jiffies - begin_jiffies;
4445
4446 remaining_jiffies -= min(slept_jiffies, remaining_jiffies);
4447
4448 last_diff = diff;
4449 }
4450
4451 return (diff == 0);
4452}
4453
4454/**
4455 * pr_flush() - Wait for printing threads to catch up.
4456 *
4457 * @timeout_ms: The maximum time (in ms) to wait.
4458 * @reset_on_progress: Reset the timeout if forward progress is seen.
4459 *
4460 * A value of 0 for @timeout_ms means no waiting will occur. A value of -1
4461 * represents infinite waiting.
4462 *
4463 * If @reset_on_progress is true, the timeout will be reset whenever any
4464 * printer has been seen to make some forward progress.
4465 *
4466 * Context: Process context. May sleep while acquiring console lock.
4467 * Return: true if all usable printers are caught up.
4468 */
4469static bool pr_flush(int timeout_ms, bool reset_on_progress)
4470{
4471 return __pr_flush(NULL, timeout_ms, reset_on_progress);
4472}
4473
4474/*
4475 * Delayed printk version, for scheduler-internal messages:
4476 */
4477#define PRINTK_PENDING_WAKEUP 0x01
4478#define PRINTK_PENDING_OUTPUT 0x02
4479
4480static DEFINE_PER_CPU(int, printk_pending);
4481
4482static void wake_up_klogd_work_func(struct irq_work *irq_work)
4483{
4484 int pending = this_cpu_xchg(printk_pending, 0);
4485
4486 if (pending & PRINTK_PENDING_OUTPUT) {
4487 if (force_legacy_kthread()) {
4488 if (printk_legacy_kthread)
4489 wake_up_interruptible(&legacy_wait);
4490 } else {
4491 if (console_trylock())
4492 console_unlock();
4493 }
4494 }
4495
4496 if (pending & PRINTK_PENDING_WAKEUP)
4497 wake_up_interruptible(&log_wait);
4498}
4499
4500static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
4501 IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
4502
4503static void __wake_up_klogd(int val)
4504{
4505 if (!printk_percpu_data_ready())
4506 return;
4507
4508 preempt_disable();
4509 /*
4510 * Guarantee any new records can be seen by tasks preparing to wait
4511 * before this context checks if the wait queue is empty.
4512 *
4513 * The full memory barrier within wq_has_sleeper() pairs with the full
4514 * memory barrier within set_current_state() of
4515 * prepare_to_wait_event(), which is called after ___wait_event() adds
4516 * the waiter but before it has checked the wait condition.
4517 *
4518 * This pairs with devkmsg_read:A and syslog_print:A.
4519 */
4520 if (wq_has_sleeper(&log_wait) || /* LMM(__wake_up_klogd:A) */
4521 (val & PRINTK_PENDING_OUTPUT)) {
4522 this_cpu_or(printk_pending, val);
4523 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
4524 }
4525 preempt_enable();
4526}
4527
4528/**
4529 * wake_up_klogd - Wake kernel logging daemon
4530 *
4531 * Use this function when new records have been added to the ringbuffer
4532 * and the console printing of those records has already occurred or is
4533 * known to be handled by some other context. This function will only
4534 * wake the logging daemon.
4535 *
4536 * Context: Any context.
4537 */
4538void wake_up_klogd(void)
4539{
4540 __wake_up_klogd(PRINTK_PENDING_WAKEUP);
4541}
4542
4543/**
4544 * defer_console_output - Wake kernel logging daemon and trigger
4545 * console printing in a deferred context
4546 *
4547 * Use this function when new records have been added to the ringbuffer,
4548 * this context is responsible for console printing those records, but
4549 * the current context is not allowed to perform the console printing.
4550 * Trigger an irq_work context to perform the console printing. This
4551 * function also wakes the logging daemon.
4552 *
4553 * Context: Any context.
4554 */
4555void defer_console_output(void)
4556{
4557 /*
4558 * New messages may have been added directly to the ringbuffer
4559 * using vprintk_store(), so wake any waiters as well.
4560 */
4561 __wake_up_klogd(PRINTK_PENDING_WAKEUP | PRINTK_PENDING_OUTPUT);
4562}
4563
4564void printk_trigger_flush(void)
4565{
4566 defer_console_output();
4567}
4568
4569int vprintk_deferred(const char *fmt, va_list args)
4570{
4571 return vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
4572}
4573
4574int _printk_deferred(const char *fmt, ...)
4575{
4576 va_list args;
4577 int r;
4578
4579 va_start(args, fmt);
4580 r = vprintk_deferred(fmt, args);
4581 va_end(args);
4582
4583 return r;
4584}
4585
4586/*
4587 * printk rate limiting, lifted from the networking subsystem.
4588 *
4589 * This enforces a rate limit: not more than 10 kernel messages
4590 * every 5s to make a denial-of-service attack impossible.
4591 */
4592DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
4593
4594int __printk_ratelimit(const char *func)
4595{
4596 return ___ratelimit(&printk_ratelimit_state, func);
4597}
4598EXPORT_SYMBOL(__printk_ratelimit);
4599
4600/**
4601 * printk_timed_ratelimit - caller-controlled printk ratelimiting
4602 * @caller_jiffies: pointer to caller's state
4603 * @interval_msecs: minimum interval between prints
4604 *
4605 * printk_timed_ratelimit() returns true if more than @interval_msecs
4606 * milliseconds have elapsed since the last time printk_timed_ratelimit()
4607 * returned true.
4608 */
4609bool printk_timed_ratelimit(unsigned long *caller_jiffies,
4610 unsigned int interval_msecs)
4611{
4612 unsigned long elapsed = jiffies - *caller_jiffies;
4613
4614 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
4615 return false;
4616
4617 *caller_jiffies = jiffies;
4618 return true;
4619}
4620EXPORT_SYMBOL(printk_timed_ratelimit);
4621
4622static DEFINE_SPINLOCK(dump_list_lock);
4623static LIST_HEAD(dump_list);
4624
4625/**
4626 * kmsg_dump_register - register a kernel log dumper.
4627 * @dumper: pointer to the kmsg_dumper structure
4628 *
4629 * Adds a kernel log dumper to the system. The dump callback in the
4630 * structure will be called when the kernel oopses or panics and must be
4631 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
4632 */
4633int kmsg_dump_register(struct kmsg_dumper *dumper)
4634{
4635 unsigned long flags;
4636 int err = -EBUSY;
4637
4638 /* The dump callback needs to be set */
4639 if (!dumper->dump)
4640 return -EINVAL;
4641
4642 spin_lock_irqsave(&dump_list_lock, flags);
4643 /* Don't allow registering multiple times */
4644 if (!dumper->registered) {
4645 dumper->registered = 1;
4646 list_add_tail_rcu(&dumper->list, &dump_list);
4647 err = 0;
4648 }
4649 spin_unlock_irqrestore(&dump_list_lock, flags);
4650
4651 return err;
4652}
4653EXPORT_SYMBOL_GPL(kmsg_dump_register);
4654
4655/**
4656 * kmsg_dump_unregister - unregister a kmsg dumper.
4657 * @dumper: pointer to the kmsg_dumper structure
4658 *
4659 * Removes a dump device from the system. Returns zero on success and
4660 * %-EINVAL otherwise.
4661 */
4662int kmsg_dump_unregister(struct kmsg_dumper *dumper)
4663{
4664 unsigned long flags;
4665 int err = -EINVAL;
4666
4667 spin_lock_irqsave(&dump_list_lock, flags);
4668 if (dumper->registered) {
4669 dumper->registered = 0;
4670 list_del_rcu(&dumper->list);
4671 err = 0;
4672 }
4673 spin_unlock_irqrestore(&dump_list_lock, flags);
4674 synchronize_rcu();
4675
4676 return err;
4677}
4678EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
4679
4680static bool always_kmsg_dump;
4681module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
4682
4683const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
4684{
4685 switch (reason) {
4686 case KMSG_DUMP_PANIC:
4687 return "Panic";
4688 case KMSG_DUMP_OOPS:
4689 return "Oops";
4690 case KMSG_DUMP_EMERG:
4691 return "Emergency";
4692 case KMSG_DUMP_SHUTDOWN:
4693 return "Shutdown";
4694 default:
4695 return "Unknown";
4696 }
4697}
4698EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
4699
4700/**
4701 * kmsg_dump_desc - dump kernel log to kernel message dumpers.
4702 * @reason: the reason (oops, panic etc) for dumping
4703 * @desc: a short string to describe what caused the panic or oops. Can be NULL
4704 * if no additional description is available.
4705 *
4706 * Call each of the registered dumper's dump() callback, which can
4707 * retrieve the kmsg records with kmsg_dump_get_line() or
4708 * kmsg_dump_get_buffer().
4709 */
4710void kmsg_dump_desc(enum kmsg_dump_reason reason, const char *desc)
4711{
4712 struct kmsg_dumper *dumper;
4713 struct kmsg_dump_detail detail = {
4714 .reason = reason,
4715 .description = desc};
4716
4717 rcu_read_lock();
4718 list_for_each_entry_rcu(dumper, &dump_list, list) {
4719 enum kmsg_dump_reason max_reason = dumper->max_reason;
4720
4721 /*
4722 * If client has not provided a specific max_reason, default
4723 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
4724 */
4725 if (max_reason == KMSG_DUMP_UNDEF) {
4726 max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
4727 KMSG_DUMP_OOPS;
4728 }
4729 if (reason > max_reason)
4730 continue;
4731
4732 /* invoke dumper which will iterate over records */
4733 dumper->dump(dumper, &detail);
4734 }
4735 rcu_read_unlock();
4736}
4737
4738/**
4739 * kmsg_dump_get_line - retrieve one kmsg log line
4740 * @iter: kmsg dump iterator
4741 * @syslog: include the "<4>" prefixes
4742 * @line: buffer to copy the line to
4743 * @size: maximum size of the buffer
4744 * @len: length of line placed into buffer
4745 *
4746 * Start at the beginning of the kmsg buffer, with the oldest kmsg
4747 * record, and copy one record into the provided buffer.
4748 *
4749 * Consecutive calls will return the next available record moving
4750 * towards the end of the buffer with the youngest messages.
4751 *
4752 * A return value of FALSE indicates that there are no more records to
4753 * read.
4754 */
4755bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
4756 char *line, size_t size, size_t *len)
4757{
4758 u64 min_seq = latched_seq_read_nolock(&clear_seq);
4759 struct printk_info info;
4760 unsigned int line_count;
4761 struct printk_record r;
4762 size_t l = 0;
4763 bool ret = false;
4764
4765 if (iter->cur_seq < min_seq)
4766 iter->cur_seq = min_seq;
4767
4768 prb_rec_init_rd(&r, &info, line, size);
4769
4770 /* Read text or count text lines? */
4771 if (line) {
4772 if (!prb_read_valid(prb, iter->cur_seq, &r))
4773 goto out;
4774 l = record_print_text(&r, syslog, printk_time);
4775 } else {
4776 if (!prb_read_valid_info(prb, iter->cur_seq,
4777 &info, &line_count)) {
4778 goto out;
4779 }
4780 l = get_record_print_text_size(&info, line_count, syslog,
4781 printk_time);
4782
4783 }
4784
4785 iter->cur_seq = r.info->seq + 1;
4786 ret = true;
4787out:
4788 if (len)
4789 *len = l;
4790 return ret;
4791}
4792EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
4793
4794/**
4795 * kmsg_dump_get_buffer - copy kmsg log lines
4796 * @iter: kmsg dump iterator
4797 * @syslog: include the "<4>" prefixes
4798 * @buf: buffer to copy the line to
4799 * @size: maximum size of the buffer
4800 * @len_out: length of line placed into buffer
4801 *
4802 * Start at the end of the kmsg buffer and fill the provided buffer
4803 * with as many of the *youngest* kmsg records that fit into it.
4804 * If the buffer is large enough, all available kmsg records will be
4805 * copied with a single call.
4806 *
4807 * Consecutive calls will fill the buffer with the next block of
4808 * available older records, not including the earlier retrieved ones.
4809 *
4810 * A return value of FALSE indicates that there are no more records to
4811 * read.
4812 */
4813bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
4814 char *buf, size_t size, size_t *len_out)
4815{
4816 u64 min_seq = latched_seq_read_nolock(&clear_seq);
4817 struct printk_info info;
4818 struct printk_record r;
4819 u64 seq;
4820 u64 next_seq;
4821 size_t len = 0;
4822 bool ret = false;
4823 bool time = printk_time;
4824
4825 if (!buf || !size)
4826 goto out;
4827
4828 if (iter->cur_seq < min_seq)
4829 iter->cur_seq = min_seq;
4830
4831 if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
4832 if (info.seq != iter->cur_seq) {
4833 /* messages are gone, move to first available one */
4834 iter->cur_seq = info.seq;
4835 }
4836 }
4837
4838 /* last entry */
4839 if (iter->cur_seq >= iter->next_seq)
4840 goto out;
4841
4842 /*
4843 * Find first record that fits, including all following records,
4844 * into the user-provided buffer for this dump. Pass in size-1
4845 * because this function (by way of record_print_text()) will
4846 * not write more than size-1 bytes of text into @buf.
4847 */
4848 seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
4849 size - 1, syslog, time);
4850
4851 /*
4852 * Next kmsg_dump_get_buffer() invocation will dump block of
4853 * older records stored right before this one.
4854 */
4855 next_seq = seq;
4856
4857 prb_rec_init_rd(&r, &info, buf, size);
4858
4859 prb_for_each_record(seq, prb, seq, &r) {
4860 if (r.info->seq >= iter->next_seq)
4861 break;
4862
4863 len += record_print_text(&r, syslog, time);
4864
4865 /* Adjust record to store to remaining buffer space. */
4866 prb_rec_init_rd(&r, &info, buf + len, size - len);
4867 }
4868
4869 iter->next_seq = next_seq;
4870 ret = true;
4871out:
4872 if (len_out)
4873 *len_out = len;
4874 return ret;
4875}
4876EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
4877
4878/**
4879 * kmsg_dump_rewind - reset the iterator
4880 * @iter: kmsg dump iterator
4881 *
4882 * Reset the dumper's iterator so that kmsg_dump_get_line() and
4883 * kmsg_dump_get_buffer() can be called again and used multiple
4884 * times within the same dumper.dump() callback.
4885 */
4886void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
4887{
4888 iter->cur_seq = latched_seq_read_nolock(&clear_seq);
4889 iter->next_seq = prb_next_seq(prb);
4890}
4891EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
4892
4893/**
4894 * console_try_replay_all - try to replay kernel log on consoles
4895 *
4896 * Try to obtain lock on console subsystem and replay all
4897 * available records in printk buffer on the consoles.
4898 * Does nothing if lock is not obtained.
4899 *
4900 * Context: Any, except for NMI.
4901 */
4902void console_try_replay_all(void)
4903{
4904 struct console_flush_type ft;
4905
4906 printk_get_console_flush_type(&ft);
4907 if (console_trylock()) {
4908 __console_rewind_all();
4909 if (ft.nbcon_atomic)
4910 nbcon_atomic_flush_pending();
4911 if (ft.nbcon_offload)
4912 nbcon_kthreads_wake();
4913 if (ft.legacy_offload)
4914 defer_console_output();
4915 /* Consoles are flushed as part of console_unlock(). */
4916 console_unlock();
4917 }
4918}
4919#endif
4920
4921#ifdef CONFIG_SMP
4922static atomic_t printk_cpu_sync_owner = ATOMIC_INIT(-1);
4923static atomic_t printk_cpu_sync_nested = ATOMIC_INIT(0);
4924
4925bool is_printk_cpu_sync_owner(void)
4926{
4927 return (atomic_read(&printk_cpu_sync_owner) == raw_smp_processor_id());
4928}
4929
4930/**
4931 * __printk_cpu_sync_wait() - Busy wait until the printk cpu-reentrant
4932 * spinning lock is not owned by any CPU.
4933 *
4934 * Context: Any context.
4935 */
4936void __printk_cpu_sync_wait(void)
4937{
4938 do {
4939 cpu_relax();
4940 } while (atomic_read(&printk_cpu_sync_owner) != -1);
4941}
4942EXPORT_SYMBOL(__printk_cpu_sync_wait);
4943
4944/**
4945 * __printk_cpu_sync_try_get() - Try to acquire the printk cpu-reentrant
4946 * spinning lock.
4947 *
4948 * If no processor has the lock, the calling processor takes the lock and
4949 * becomes the owner. If the calling processor is already the owner of the
4950 * lock, this function succeeds immediately.
4951 *
4952 * Context: Any context. Expects interrupts to be disabled.
4953 * Return: 1 on success, otherwise 0.
4954 */
4955int __printk_cpu_sync_try_get(void)
4956{
4957 int cpu;
4958 int old;
4959
4960 cpu = smp_processor_id();
4961
4962 /*
4963 * Guarantee loads and stores from this CPU when it is the lock owner
4964 * are _not_ visible to the previous lock owner. This pairs with
4965 * __printk_cpu_sync_put:B.
4966 *
4967 * Memory barrier involvement:
4968 *
4969 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4970 * then __printk_cpu_sync_put:A can never read from
4971 * __printk_cpu_sync_try_get:B.
4972 *
4973 * Relies on:
4974 *
4975 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4976 * of the previous CPU
4977 * matching
4978 * ACQUIRE from __printk_cpu_sync_try_get:A to
4979 * __printk_cpu_sync_try_get:B of this CPU
4980 */
4981 old = atomic_cmpxchg_acquire(&printk_cpu_sync_owner, -1,
4982 cpu); /* LMM(__printk_cpu_sync_try_get:A) */
4983 if (old == -1) {
4984 /*
4985 * This CPU is now the owner and begins loading/storing
4986 * data: LMM(__printk_cpu_sync_try_get:B)
4987 */
4988 return 1;
4989
4990 } else if (old == cpu) {
4991 /* This CPU is already the owner. */
4992 atomic_inc(&printk_cpu_sync_nested);
4993 return 1;
4994 }
4995
4996 return 0;
4997}
4998EXPORT_SYMBOL(__printk_cpu_sync_try_get);
4999
5000/**
5001 * __printk_cpu_sync_put() - Release the printk cpu-reentrant spinning lock.
5002 *
5003 * The calling processor must be the owner of the lock.
5004 *
5005 * Context: Any context. Expects interrupts to be disabled.
5006 */
5007void __printk_cpu_sync_put(void)
5008{
5009 if (atomic_read(&printk_cpu_sync_nested)) {
5010 atomic_dec(&printk_cpu_sync_nested);
5011 return;
5012 }
5013
5014 /*
5015 * This CPU is finished loading/storing data:
5016 * LMM(__printk_cpu_sync_put:A)
5017 */
5018
5019 /*
5020 * Guarantee loads and stores from this CPU when it was the
5021 * lock owner are visible to the next lock owner. This pairs
5022 * with __printk_cpu_sync_try_get:A.
5023 *
5024 * Memory barrier involvement:
5025 *
5026 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
5027 * then __printk_cpu_sync_try_get:B reads from __printk_cpu_sync_put:A.
5028 *
5029 * Relies on:
5030 *
5031 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
5032 * of this CPU
5033 * matching
5034 * ACQUIRE from __printk_cpu_sync_try_get:A to
5035 * __printk_cpu_sync_try_get:B of the next CPU
5036 */
5037 atomic_set_release(&printk_cpu_sync_owner,
5038 -1); /* LMM(__printk_cpu_sync_put:B) */
5039}
5040EXPORT_SYMBOL(__printk_cpu_sync_put);
5041#endif /* CONFIG_SMP */
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/vmcore_info.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
74EXPORT_TRACEPOINT_SYMBOL_GPL(console);
75
76/*
77 * Low level drivers may need that to know if they can schedule in
78 * their unblank() callback or not. So let's export it.
79 */
80int oops_in_progress;
81EXPORT_SYMBOL(oops_in_progress);
82
83/*
84 * console_mutex protects console_list updates and console->flags updates.
85 * The flags are synchronized only for consoles that are registered, i.e.
86 * accessible via the console list.
87 */
88static DEFINE_MUTEX(console_mutex);
89
90/*
91 * console_sem protects updates to console->seq
92 * and also provides serialization for console printing.
93 */
94static DEFINE_SEMAPHORE(console_sem, 1);
95HLIST_HEAD(console_list);
96EXPORT_SYMBOL_GPL(console_list);
97DEFINE_STATIC_SRCU(console_srcu);
98
99/*
100 * System may need to suppress printk message under certain
101 * circumstances, like after kernel panic happens.
102 */
103int __read_mostly suppress_printk;
104
105#ifdef CONFIG_LOCKDEP
106static struct lockdep_map console_lock_dep_map = {
107 .name = "console_lock"
108};
109
110void lockdep_assert_console_list_lock_held(void)
111{
112 lockdep_assert_held(&console_mutex);
113}
114EXPORT_SYMBOL(lockdep_assert_console_list_lock_held);
115#endif
116
117#ifdef CONFIG_DEBUG_LOCK_ALLOC
118bool console_srcu_read_lock_is_held(void)
119{
120 return srcu_read_lock_held(&console_srcu);
121}
122EXPORT_SYMBOL(console_srcu_read_lock_is_held);
123#endif
124
125enum devkmsg_log_bits {
126 __DEVKMSG_LOG_BIT_ON = 0,
127 __DEVKMSG_LOG_BIT_OFF,
128 __DEVKMSG_LOG_BIT_LOCK,
129};
130
131enum devkmsg_log_masks {
132 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
133 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
134 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
135};
136
137/* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
138#define DEVKMSG_LOG_MASK_DEFAULT 0
139
140static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
141
142static int __control_devkmsg(char *str)
143{
144 size_t len;
145
146 if (!str)
147 return -EINVAL;
148
149 len = str_has_prefix(str, "on");
150 if (len) {
151 devkmsg_log = DEVKMSG_LOG_MASK_ON;
152 return len;
153 }
154
155 len = str_has_prefix(str, "off");
156 if (len) {
157 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
158 return len;
159 }
160
161 len = str_has_prefix(str, "ratelimit");
162 if (len) {
163 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
164 return len;
165 }
166
167 return -EINVAL;
168}
169
170static int __init control_devkmsg(char *str)
171{
172 if (__control_devkmsg(str) < 0) {
173 pr_warn("printk.devkmsg: bad option string '%s'\n", str);
174 return 1;
175 }
176
177 /*
178 * Set sysctl string accordingly:
179 */
180 if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
181 strcpy(devkmsg_log_str, "on");
182 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
183 strcpy(devkmsg_log_str, "off");
184 /* else "ratelimit" which is set by default. */
185
186 /*
187 * Sysctl cannot change it anymore. The kernel command line setting of
188 * this parameter is to force the setting to be permanent throughout the
189 * runtime of the system. This is a precation measure against userspace
190 * trying to be a smarta** and attempting to change it up on us.
191 */
192 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
193
194 return 1;
195}
196__setup("printk.devkmsg=", control_devkmsg);
197
198char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
199#if defined(CONFIG_PRINTK) && defined(CONFIG_SYSCTL)
200int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
201 void *buffer, size_t *lenp, loff_t *ppos)
202{
203 char old_str[DEVKMSG_STR_MAX_SIZE];
204 unsigned int old;
205 int err;
206
207 if (write) {
208 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
209 return -EINVAL;
210
211 old = devkmsg_log;
212 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
213 }
214
215 err = proc_dostring(table, write, buffer, lenp, ppos);
216 if (err)
217 return err;
218
219 if (write) {
220 err = __control_devkmsg(devkmsg_log_str);
221
222 /*
223 * Do not accept an unknown string OR a known string with
224 * trailing crap...
225 */
226 if (err < 0 || (err + 1 != *lenp)) {
227
228 /* ... and restore old setting. */
229 devkmsg_log = old;
230 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
231
232 return -EINVAL;
233 }
234 }
235
236 return 0;
237}
238#endif /* CONFIG_PRINTK && CONFIG_SYSCTL */
239
240/**
241 * console_list_lock - Lock the console list
242 *
243 * For console list or console->flags updates
244 */
245void console_list_lock(void)
246{
247 /*
248 * In unregister_console() and console_force_preferred_locked(),
249 * synchronize_srcu() is called with the console_list_lock held.
250 * Therefore it is not allowed that the console_list_lock is taken
251 * with the srcu_lock held.
252 *
253 * Detecting if this context is really in the read-side critical
254 * section is only possible if the appropriate debug options are
255 * enabled.
256 */
257 WARN_ON_ONCE(debug_lockdep_rcu_enabled() &&
258 srcu_read_lock_held(&console_srcu));
259
260 mutex_lock(&console_mutex);
261}
262EXPORT_SYMBOL(console_list_lock);
263
264/**
265 * console_list_unlock - Unlock the console list
266 *
267 * Counterpart to console_list_lock()
268 */
269void console_list_unlock(void)
270{
271 mutex_unlock(&console_mutex);
272}
273EXPORT_SYMBOL(console_list_unlock);
274
275/**
276 * console_srcu_read_lock - Register a new reader for the
277 * SRCU-protected console list
278 *
279 * Use for_each_console_srcu() to iterate the console list
280 *
281 * Context: Any context.
282 * Return: A cookie to pass to console_srcu_read_unlock().
283 */
284int console_srcu_read_lock(void)
285{
286 return srcu_read_lock_nmisafe(&console_srcu);
287}
288EXPORT_SYMBOL(console_srcu_read_lock);
289
290/**
291 * console_srcu_read_unlock - Unregister an old reader from
292 * the SRCU-protected console list
293 * @cookie: cookie returned from console_srcu_read_lock()
294 *
295 * Counterpart to console_srcu_read_lock()
296 */
297void console_srcu_read_unlock(int cookie)
298{
299 srcu_read_unlock_nmisafe(&console_srcu, cookie);
300}
301EXPORT_SYMBOL(console_srcu_read_unlock);
302
303/*
304 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
305 * macros instead of functions so that _RET_IP_ contains useful information.
306 */
307#define down_console_sem() do { \
308 down(&console_sem);\
309 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
310} while (0)
311
312static int __down_trylock_console_sem(unsigned long ip)
313{
314 int lock_failed;
315 unsigned long flags;
316
317 /*
318 * Here and in __up_console_sem() we need to be in safe mode,
319 * because spindump/WARN/etc from under console ->lock will
320 * deadlock in printk()->down_trylock_console_sem() otherwise.
321 */
322 printk_safe_enter_irqsave(flags);
323 lock_failed = down_trylock(&console_sem);
324 printk_safe_exit_irqrestore(flags);
325
326 if (lock_failed)
327 return 1;
328 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
329 return 0;
330}
331#define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
332
333static void __up_console_sem(unsigned long ip)
334{
335 unsigned long flags;
336
337 mutex_release(&console_lock_dep_map, ip);
338
339 printk_safe_enter_irqsave(flags);
340 up(&console_sem);
341 printk_safe_exit_irqrestore(flags);
342}
343#define up_console_sem() __up_console_sem(_RET_IP_)
344
345static bool panic_in_progress(void)
346{
347 return unlikely(atomic_read(&panic_cpu) != PANIC_CPU_INVALID);
348}
349
350/* Return true if a panic is in progress on the current CPU. */
351bool this_cpu_in_panic(void)
352{
353 /*
354 * We can use raw_smp_processor_id() here because it is impossible for
355 * the task to be migrated to the panic_cpu, or away from it. If
356 * panic_cpu has already been set, and we're not currently executing on
357 * that CPU, then we never will be.
358 */
359 return unlikely(atomic_read(&panic_cpu) == raw_smp_processor_id());
360}
361
362/*
363 * Return true if a panic is in progress on a remote CPU.
364 *
365 * On true, the local CPU should immediately release any printing resources
366 * that may be needed by the panic CPU.
367 */
368bool other_cpu_in_panic(void)
369{
370 return (panic_in_progress() && !this_cpu_in_panic());
371}
372
373/*
374 * This is used for debugging the mess that is the VT code by
375 * keeping track if we have the console semaphore held. It's
376 * definitely not the perfect debug tool (we don't know if _WE_
377 * hold it and are racing, but it helps tracking those weird code
378 * paths in the console code where we end up in places I want
379 * locked without the console semaphore held).
380 */
381static int console_locked;
382
383/*
384 * Array of consoles built from command line options (console=)
385 */
386
387#define MAX_CMDLINECONSOLES 8
388
389static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
390
391static int preferred_console = -1;
392int console_set_on_cmdline;
393EXPORT_SYMBOL(console_set_on_cmdline);
394
395/* Flag: console code may call schedule() */
396static int console_may_schedule;
397
398enum con_msg_format_flags {
399 MSG_FORMAT_DEFAULT = 0,
400 MSG_FORMAT_SYSLOG = (1 << 0),
401};
402
403static int console_msg_format = MSG_FORMAT_DEFAULT;
404
405/*
406 * The printk log buffer consists of a sequenced collection of records, each
407 * containing variable length message text. Every record also contains its
408 * own meta-data (@info).
409 *
410 * Every record meta-data carries the timestamp in microseconds, as well as
411 * the standard userspace syslog level and syslog facility. The usual kernel
412 * messages use LOG_KERN; userspace-injected messages always carry a matching
413 * syslog facility, by default LOG_USER. The origin of every message can be
414 * reliably determined that way.
415 *
416 * The human readable log message of a record is available in @text, the
417 * length of the message text in @text_len. The stored message is not
418 * terminated.
419 *
420 * Optionally, a record can carry a dictionary of properties (key/value
421 * pairs), to provide userspace with a machine-readable message context.
422 *
423 * Examples for well-defined, commonly used property names are:
424 * DEVICE=b12:8 device identifier
425 * b12:8 block dev_t
426 * c127:3 char dev_t
427 * n8 netdev ifindex
428 * +sound:card0 subsystem:devname
429 * SUBSYSTEM=pci driver-core subsystem name
430 *
431 * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
432 * and values are terminated by a '\0' character.
433 *
434 * Example of record values:
435 * record.text_buf = "it's a line" (unterminated)
436 * record.info.seq = 56
437 * record.info.ts_nsec = 36863
438 * record.info.text_len = 11
439 * record.info.facility = 0 (LOG_KERN)
440 * record.info.flags = 0
441 * record.info.level = 3 (LOG_ERR)
442 * record.info.caller_id = 299 (task 299)
443 * record.info.dev_info.subsystem = "pci" (terminated)
444 * record.info.dev_info.device = "+pci:0000:00:01.0" (terminated)
445 *
446 * The 'struct printk_info' buffer must never be directly exported to
447 * userspace, it is a kernel-private implementation detail that might
448 * need to be changed in the future, when the requirements change.
449 *
450 * /dev/kmsg exports the structured data in the following line format:
451 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
452 *
453 * Users of the export format should ignore possible additional values
454 * separated by ',', and find the message after the ';' character.
455 *
456 * The optional key/value pairs are attached as continuation lines starting
457 * with a space character and terminated by a newline. All possible
458 * non-prinatable characters are escaped in the "\xff" notation.
459 */
460
461/* syslog_lock protects syslog_* variables and write access to clear_seq. */
462static DEFINE_MUTEX(syslog_lock);
463
464#ifdef CONFIG_PRINTK
465DECLARE_WAIT_QUEUE_HEAD(log_wait);
466/* All 3 protected by @syslog_lock. */
467/* the next printk record to read by syslog(READ) or /proc/kmsg */
468static u64 syslog_seq;
469static size_t syslog_partial;
470static bool syslog_time;
471
472struct latched_seq {
473 seqcount_latch_t latch;
474 u64 val[2];
475};
476
477/*
478 * The next printk record to read after the last 'clear' command. There are
479 * two copies (updated with seqcount_latch) so that reads can locklessly
480 * access a valid value. Writers are synchronized by @syslog_lock.
481 */
482static struct latched_seq clear_seq = {
483 .latch = SEQCNT_LATCH_ZERO(clear_seq.latch),
484 .val[0] = 0,
485 .val[1] = 0,
486};
487
488#define LOG_LEVEL(v) ((v) & 0x07)
489#define LOG_FACILITY(v) ((v) >> 3 & 0xff)
490
491/* record buffer */
492#define LOG_ALIGN __alignof__(unsigned long)
493#define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
494#define LOG_BUF_LEN_MAX (u32)(1 << 31)
495static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
496static char *log_buf = __log_buf;
497static u32 log_buf_len = __LOG_BUF_LEN;
498
499/*
500 * Define the average message size. This only affects the number of
501 * descriptors that will be available. Underestimating is better than
502 * overestimating (too many available descriptors is better than not enough).
503 */
504#define PRB_AVGBITS 5 /* 32 character average length */
505
506#if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
507#error CONFIG_LOG_BUF_SHIFT value too small.
508#endif
509_DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
510 PRB_AVGBITS, &__log_buf[0]);
511
512static struct printk_ringbuffer printk_rb_dynamic;
513
514struct printk_ringbuffer *prb = &printk_rb_static;
515
516/*
517 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
518 * per_cpu_areas are initialised. This variable is set to true when
519 * it's safe to access per-CPU data.
520 */
521static bool __printk_percpu_data_ready __ro_after_init;
522
523bool printk_percpu_data_ready(void)
524{
525 return __printk_percpu_data_ready;
526}
527
528/* Must be called under syslog_lock. */
529static void latched_seq_write(struct latched_seq *ls, u64 val)
530{
531 raw_write_seqcount_latch(&ls->latch);
532 ls->val[0] = val;
533 raw_write_seqcount_latch(&ls->latch);
534 ls->val[1] = val;
535}
536
537/* Can be called from any context. */
538static u64 latched_seq_read_nolock(struct latched_seq *ls)
539{
540 unsigned int seq;
541 unsigned int idx;
542 u64 val;
543
544 do {
545 seq = raw_read_seqcount_latch(&ls->latch);
546 idx = seq & 0x1;
547 val = ls->val[idx];
548 } while (raw_read_seqcount_latch_retry(&ls->latch, seq));
549
550 return val;
551}
552
553/* Return log buffer address */
554char *log_buf_addr_get(void)
555{
556 return log_buf;
557}
558
559/* Return log buffer size */
560u32 log_buf_len_get(void)
561{
562 return log_buf_len;
563}
564
565/*
566 * Define how much of the log buffer we could take at maximum. The value
567 * must be greater than two. Note that only half of the buffer is available
568 * when the index points to the middle.
569 */
570#define MAX_LOG_TAKE_PART 4
571static const char trunc_msg[] = "<truncated>";
572
573static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
574{
575 /*
576 * The message should not take the whole buffer. Otherwise, it might
577 * get removed too soon.
578 */
579 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
580
581 if (*text_len > max_text_len)
582 *text_len = max_text_len;
583
584 /* enable the warning message (if there is room) */
585 *trunc_msg_len = strlen(trunc_msg);
586 if (*text_len >= *trunc_msg_len)
587 *text_len -= *trunc_msg_len;
588 else
589 *trunc_msg_len = 0;
590}
591
592int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
593
594static int syslog_action_restricted(int type)
595{
596 if (dmesg_restrict)
597 return 1;
598 /*
599 * Unless restricted, we allow "read all" and "get buffer size"
600 * for everybody.
601 */
602 return type != SYSLOG_ACTION_READ_ALL &&
603 type != SYSLOG_ACTION_SIZE_BUFFER;
604}
605
606static int check_syslog_permissions(int type, int source)
607{
608 /*
609 * If this is from /proc/kmsg and we've already opened it, then we've
610 * already done the capabilities checks at open time.
611 */
612 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
613 goto ok;
614
615 if (syslog_action_restricted(type)) {
616 if (capable(CAP_SYSLOG))
617 goto ok;
618 return -EPERM;
619 }
620ok:
621 return security_syslog(type);
622}
623
624static void append_char(char **pp, char *e, char c)
625{
626 if (*pp < e)
627 *(*pp)++ = c;
628}
629
630static ssize_t info_print_ext_header(char *buf, size_t size,
631 struct printk_info *info)
632{
633 u64 ts_usec = info->ts_nsec;
634 char caller[20];
635#ifdef CONFIG_PRINTK_CALLER
636 u32 id = info->caller_id;
637
638 snprintf(caller, sizeof(caller), ",caller=%c%u",
639 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
640#else
641 caller[0] = '\0';
642#endif
643
644 do_div(ts_usec, 1000);
645
646 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
647 (info->facility << 3) | info->level, info->seq,
648 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
649}
650
651static ssize_t msg_add_ext_text(char *buf, size_t size,
652 const char *text, size_t text_len,
653 unsigned char endc)
654{
655 char *p = buf, *e = buf + size;
656 size_t i;
657
658 /* escape non-printable characters */
659 for (i = 0; i < text_len; i++) {
660 unsigned char c = text[i];
661
662 if (c < ' ' || c >= 127 || c == '\\')
663 p += scnprintf(p, e - p, "\\x%02x", c);
664 else
665 append_char(&p, e, c);
666 }
667 append_char(&p, e, endc);
668
669 return p - buf;
670}
671
672static ssize_t msg_add_dict_text(char *buf, size_t size,
673 const char *key, const char *val)
674{
675 size_t val_len = strlen(val);
676 ssize_t len;
677
678 if (!val_len)
679 return 0;
680
681 len = msg_add_ext_text(buf, size, "", 0, ' '); /* dict prefix */
682 len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
683 len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
684
685 return len;
686}
687
688static ssize_t msg_print_ext_body(char *buf, size_t size,
689 char *text, size_t text_len,
690 struct dev_printk_info *dev_info)
691{
692 ssize_t len;
693
694 len = msg_add_ext_text(buf, size, text, text_len, '\n');
695
696 if (!dev_info)
697 goto out;
698
699 len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
700 dev_info->subsystem);
701 len += msg_add_dict_text(buf + len, size - len, "DEVICE",
702 dev_info->device);
703out:
704 return len;
705}
706
707/* /dev/kmsg - userspace message inject/listen interface */
708struct devkmsg_user {
709 atomic64_t seq;
710 struct ratelimit_state rs;
711 struct mutex lock;
712 struct printk_buffers pbufs;
713};
714
715static __printf(3, 4) __cold
716int devkmsg_emit(int facility, int level, const char *fmt, ...)
717{
718 va_list args;
719 int r;
720
721 va_start(args, fmt);
722 r = vprintk_emit(facility, level, NULL, fmt, args);
723 va_end(args);
724
725 return r;
726}
727
728static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
729{
730 char *buf, *line;
731 int level = default_message_loglevel;
732 int facility = 1; /* LOG_USER */
733 struct file *file = iocb->ki_filp;
734 struct devkmsg_user *user = file->private_data;
735 size_t len = iov_iter_count(from);
736 ssize_t ret = len;
737
738 if (len > PRINTKRB_RECORD_MAX)
739 return -EINVAL;
740
741 /* Ignore when user logging is disabled. */
742 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
743 return len;
744
745 /* Ratelimit when not explicitly enabled. */
746 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
747 if (!___ratelimit(&user->rs, current->comm))
748 return ret;
749 }
750
751 buf = kmalloc(len+1, GFP_KERNEL);
752 if (buf == NULL)
753 return -ENOMEM;
754
755 buf[len] = '\0';
756 if (!copy_from_iter_full(buf, len, from)) {
757 kfree(buf);
758 return -EFAULT;
759 }
760
761 /*
762 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
763 * the decimal value represents 32bit, the lower 3 bit are the log
764 * level, the rest are the log facility.
765 *
766 * If no prefix or no userspace facility is specified, we
767 * enforce LOG_USER, to be able to reliably distinguish
768 * kernel-generated messages from userspace-injected ones.
769 */
770 line = buf;
771 if (line[0] == '<') {
772 char *endp = NULL;
773 unsigned int u;
774
775 u = simple_strtoul(line + 1, &endp, 10);
776 if (endp && endp[0] == '>') {
777 level = LOG_LEVEL(u);
778 if (LOG_FACILITY(u) != 0)
779 facility = LOG_FACILITY(u);
780 endp++;
781 line = endp;
782 }
783 }
784
785 devkmsg_emit(facility, level, "%s", line);
786 kfree(buf);
787 return ret;
788}
789
790static ssize_t devkmsg_read(struct file *file, char __user *buf,
791 size_t count, loff_t *ppos)
792{
793 struct devkmsg_user *user = file->private_data;
794 char *outbuf = &user->pbufs.outbuf[0];
795 struct printk_message pmsg = {
796 .pbufs = &user->pbufs,
797 };
798 ssize_t ret;
799
800 ret = mutex_lock_interruptible(&user->lock);
801 if (ret)
802 return ret;
803
804 if (!printk_get_next_message(&pmsg, atomic64_read(&user->seq), true, false)) {
805 if (file->f_flags & O_NONBLOCK) {
806 ret = -EAGAIN;
807 goto out;
808 }
809
810 /*
811 * Guarantee this task is visible on the waitqueue before
812 * checking the wake condition.
813 *
814 * The full memory barrier within set_current_state() of
815 * prepare_to_wait_event() pairs with the full memory barrier
816 * within wq_has_sleeper().
817 *
818 * This pairs with __wake_up_klogd:A.
819 */
820 ret = wait_event_interruptible(log_wait,
821 printk_get_next_message(&pmsg, atomic64_read(&user->seq), true,
822 false)); /* LMM(devkmsg_read:A) */
823 if (ret)
824 goto out;
825 }
826
827 if (pmsg.dropped) {
828 /* our last seen message is gone, return error and reset */
829 atomic64_set(&user->seq, pmsg.seq);
830 ret = -EPIPE;
831 goto out;
832 }
833
834 atomic64_set(&user->seq, pmsg.seq + 1);
835
836 if (pmsg.outbuf_len > count) {
837 ret = -EINVAL;
838 goto out;
839 }
840
841 if (copy_to_user(buf, outbuf, pmsg.outbuf_len)) {
842 ret = -EFAULT;
843 goto out;
844 }
845 ret = pmsg.outbuf_len;
846out:
847 mutex_unlock(&user->lock);
848 return ret;
849}
850
851/*
852 * Be careful when modifying this function!!!
853 *
854 * Only few operations are supported because the device works only with the
855 * entire variable length messages (records). Non-standard values are
856 * returned in the other cases and has been this way for quite some time.
857 * User space applications might depend on this behavior.
858 */
859static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
860{
861 struct devkmsg_user *user = file->private_data;
862 loff_t ret = 0;
863
864 if (offset)
865 return -ESPIPE;
866
867 switch (whence) {
868 case SEEK_SET:
869 /* the first record */
870 atomic64_set(&user->seq, prb_first_valid_seq(prb));
871 break;
872 case SEEK_DATA:
873 /*
874 * The first record after the last SYSLOG_ACTION_CLEAR,
875 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
876 * changes no global state, and does not clear anything.
877 */
878 atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
879 break;
880 case SEEK_END:
881 /* after the last record */
882 atomic64_set(&user->seq, prb_next_seq(prb));
883 break;
884 default:
885 ret = -EINVAL;
886 }
887 return ret;
888}
889
890static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
891{
892 struct devkmsg_user *user = file->private_data;
893 struct printk_info info;
894 __poll_t ret = 0;
895
896 poll_wait(file, &log_wait, wait);
897
898 if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
899 /* return error when data has vanished underneath us */
900 if (info.seq != atomic64_read(&user->seq))
901 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
902 else
903 ret = EPOLLIN|EPOLLRDNORM;
904 }
905
906 return ret;
907}
908
909static int devkmsg_open(struct inode *inode, struct file *file)
910{
911 struct devkmsg_user *user;
912 int err;
913
914 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
915 return -EPERM;
916
917 /* write-only does not need any file context */
918 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
919 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
920 SYSLOG_FROM_READER);
921 if (err)
922 return err;
923 }
924
925 user = kvmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
926 if (!user)
927 return -ENOMEM;
928
929 ratelimit_default_init(&user->rs);
930 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
931
932 mutex_init(&user->lock);
933
934 atomic64_set(&user->seq, prb_first_valid_seq(prb));
935
936 file->private_data = user;
937 return 0;
938}
939
940static int devkmsg_release(struct inode *inode, struct file *file)
941{
942 struct devkmsg_user *user = file->private_data;
943
944 ratelimit_state_exit(&user->rs);
945
946 mutex_destroy(&user->lock);
947 kvfree(user);
948 return 0;
949}
950
951const struct file_operations kmsg_fops = {
952 .open = devkmsg_open,
953 .read = devkmsg_read,
954 .write_iter = devkmsg_write,
955 .llseek = devkmsg_llseek,
956 .poll = devkmsg_poll,
957 .release = devkmsg_release,
958};
959
960#ifdef CONFIG_VMCORE_INFO
961/*
962 * This appends the listed symbols to /proc/vmcore
963 *
964 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
965 * obtain access to symbols that are otherwise very difficult to locate. These
966 * symbols are specifically used so that utilities can access and extract the
967 * dmesg log from a vmcore file after a crash.
968 */
969void log_buf_vmcoreinfo_setup(void)
970{
971 struct dev_printk_info *dev_info = NULL;
972
973 VMCOREINFO_SYMBOL(prb);
974 VMCOREINFO_SYMBOL(printk_rb_static);
975 VMCOREINFO_SYMBOL(clear_seq);
976
977 /*
978 * Export struct size and field offsets. User space tools can
979 * parse it and detect any changes to structure down the line.
980 */
981
982 VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
983 VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
984 VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
985 VMCOREINFO_OFFSET(printk_ringbuffer, fail);
986
987 VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
988 VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
989 VMCOREINFO_OFFSET(prb_desc_ring, descs);
990 VMCOREINFO_OFFSET(prb_desc_ring, infos);
991 VMCOREINFO_OFFSET(prb_desc_ring, head_id);
992 VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
993
994 VMCOREINFO_STRUCT_SIZE(prb_desc);
995 VMCOREINFO_OFFSET(prb_desc, state_var);
996 VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
997
998 VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
999 VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
1000 VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
1001
1002 VMCOREINFO_STRUCT_SIZE(printk_info);
1003 VMCOREINFO_OFFSET(printk_info, seq);
1004 VMCOREINFO_OFFSET(printk_info, ts_nsec);
1005 VMCOREINFO_OFFSET(printk_info, text_len);
1006 VMCOREINFO_OFFSET(printk_info, caller_id);
1007 VMCOREINFO_OFFSET(printk_info, dev_info);
1008
1009 VMCOREINFO_STRUCT_SIZE(dev_printk_info);
1010 VMCOREINFO_OFFSET(dev_printk_info, subsystem);
1011 VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
1012 VMCOREINFO_OFFSET(dev_printk_info, device);
1013 VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
1014
1015 VMCOREINFO_STRUCT_SIZE(prb_data_ring);
1016 VMCOREINFO_OFFSET(prb_data_ring, size_bits);
1017 VMCOREINFO_OFFSET(prb_data_ring, data);
1018 VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
1019 VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
1020
1021 VMCOREINFO_SIZE(atomic_long_t);
1022 VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
1023
1024 VMCOREINFO_STRUCT_SIZE(latched_seq);
1025 VMCOREINFO_OFFSET(latched_seq, val);
1026}
1027#endif
1028
1029/* requested log_buf_len from kernel cmdline */
1030static unsigned long __initdata new_log_buf_len;
1031
1032/* we practice scaling the ring buffer by powers of 2 */
1033static void __init log_buf_len_update(u64 size)
1034{
1035 if (size > (u64)LOG_BUF_LEN_MAX) {
1036 size = (u64)LOG_BUF_LEN_MAX;
1037 pr_err("log_buf over 2G is not supported.\n");
1038 }
1039
1040 if (size)
1041 size = roundup_pow_of_two(size);
1042 if (size > log_buf_len)
1043 new_log_buf_len = (unsigned long)size;
1044}
1045
1046/* save requested log_buf_len since it's too early to process it */
1047static int __init log_buf_len_setup(char *str)
1048{
1049 u64 size;
1050
1051 if (!str)
1052 return -EINVAL;
1053
1054 size = memparse(str, &str);
1055
1056 log_buf_len_update(size);
1057
1058 return 0;
1059}
1060early_param("log_buf_len", log_buf_len_setup);
1061
1062#ifdef CONFIG_SMP
1063#define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1064
1065static void __init log_buf_add_cpu(void)
1066{
1067 unsigned int cpu_extra;
1068
1069 /*
1070 * archs should set up cpu_possible_bits properly with
1071 * set_cpu_possible() after setup_arch() but just in
1072 * case lets ensure this is valid.
1073 */
1074 if (num_possible_cpus() == 1)
1075 return;
1076
1077 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1078
1079 /* by default this will only continue through for large > 64 CPUs */
1080 if (cpu_extra <= __LOG_BUF_LEN / 2)
1081 return;
1082
1083 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1084 __LOG_CPU_MAX_BUF_LEN);
1085 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1086 cpu_extra);
1087 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1088
1089 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1090}
1091#else /* !CONFIG_SMP */
1092static inline void log_buf_add_cpu(void) {}
1093#endif /* CONFIG_SMP */
1094
1095static void __init set_percpu_data_ready(void)
1096{
1097 __printk_percpu_data_ready = true;
1098}
1099
1100static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1101 struct printk_record *r)
1102{
1103 struct prb_reserved_entry e;
1104 struct printk_record dest_r;
1105
1106 prb_rec_init_wr(&dest_r, r->info->text_len);
1107
1108 if (!prb_reserve(&e, rb, &dest_r))
1109 return 0;
1110
1111 memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1112 dest_r.info->text_len = r->info->text_len;
1113 dest_r.info->facility = r->info->facility;
1114 dest_r.info->level = r->info->level;
1115 dest_r.info->flags = r->info->flags;
1116 dest_r.info->ts_nsec = r->info->ts_nsec;
1117 dest_r.info->caller_id = r->info->caller_id;
1118 memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1119
1120 prb_final_commit(&e);
1121
1122 return prb_record_text_space(&e);
1123}
1124
1125static char setup_text_buf[PRINTKRB_RECORD_MAX] __initdata;
1126
1127void __init setup_log_buf(int early)
1128{
1129 struct printk_info *new_infos;
1130 unsigned int new_descs_count;
1131 struct prb_desc *new_descs;
1132 struct printk_info info;
1133 struct printk_record r;
1134 unsigned int text_size;
1135 size_t new_descs_size;
1136 size_t new_infos_size;
1137 unsigned long flags;
1138 char *new_log_buf;
1139 unsigned int free;
1140 u64 seq;
1141
1142 /*
1143 * Some archs call setup_log_buf() multiple times - first is very
1144 * early, e.g. from setup_arch(), and second - when percpu_areas
1145 * are initialised.
1146 */
1147 if (!early)
1148 set_percpu_data_ready();
1149
1150 if (log_buf != __log_buf)
1151 return;
1152
1153 if (!early && !new_log_buf_len)
1154 log_buf_add_cpu();
1155
1156 if (!new_log_buf_len)
1157 return;
1158
1159 new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1160 if (new_descs_count == 0) {
1161 pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1162 return;
1163 }
1164
1165 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1166 if (unlikely(!new_log_buf)) {
1167 pr_err("log_buf_len: %lu text bytes not available\n",
1168 new_log_buf_len);
1169 return;
1170 }
1171
1172 new_descs_size = new_descs_count * sizeof(struct prb_desc);
1173 new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1174 if (unlikely(!new_descs)) {
1175 pr_err("log_buf_len: %zu desc bytes not available\n",
1176 new_descs_size);
1177 goto err_free_log_buf;
1178 }
1179
1180 new_infos_size = new_descs_count * sizeof(struct printk_info);
1181 new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1182 if (unlikely(!new_infos)) {
1183 pr_err("log_buf_len: %zu info bytes not available\n",
1184 new_infos_size);
1185 goto err_free_descs;
1186 }
1187
1188 prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1189
1190 prb_init(&printk_rb_dynamic,
1191 new_log_buf, ilog2(new_log_buf_len),
1192 new_descs, ilog2(new_descs_count),
1193 new_infos);
1194
1195 local_irq_save(flags);
1196
1197 log_buf_len = new_log_buf_len;
1198 log_buf = new_log_buf;
1199 new_log_buf_len = 0;
1200
1201 free = __LOG_BUF_LEN;
1202 prb_for_each_record(0, &printk_rb_static, seq, &r) {
1203 text_size = add_to_rb(&printk_rb_dynamic, &r);
1204 if (text_size > free)
1205 free = 0;
1206 else
1207 free -= text_size;
1208 }
1209
1210 prb = &printk_rb_dynamic;
1211
1212 local_irq_restore(flags);
1213
1214 /*
1215 * Copy any remaining messages that might have appeared from
1216 * NMI context after copying but before switching to the
1217 * dynamic buffer.
1218 */
1219 prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1220 text_size = add_to_rb(&printk_rb_dynamic, &r);
1221 if (text_size > free)
1222 free = 0;
1223 else
1224 free -= text_size;
1225 }
1226
1227 if (seq != prb_next_seq(&printk_rb_static)) {
1228 pr_err("dropped %llu messages\n",
1229 prb_next_seq(&printk_rb_static) - seq);
1230 }
1231
1232 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1233 pr_info("early log buf free: %u(%u%%)\n",
1234 free, (free * 100) / __LOG_BUF_LEN);
1235 return;
1236
1237err_free_descs:
1238 memblock_free(new_descs, new_descs_size);
1239err_free_log_buf:
1240 memblock_free(new_log_buf, new_log_buf_len);
1241}
1242
1243static bool __read_mostly ignore_loglevel;
1244
1245static int __init ignore_loglevel_setup(char *str)
1246{
1247 ignore_loglevel = true;
1248 pr_info("debug: ignoring loglevel setting.\n");
1249
1250 return 0;
1251}
1252
1253early_param("ignore_loglevel", ignore_loglevel_setup);
1254module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1255MODULE_PARM_DESC(ignore_loglevel,
1256 "ignore loglevel setting (prints all kernel messages to the console)");
1257
1258static bool suppress_message_printing(int level)
1259{
1260 return (level >= console_loglevel && !ignore_loglevel);
1261}
1262
1263#ifdef CONFIG_BOOT_PRINTK_DELAY
1264
1265static int boot_delay; /* msecs delay after each printk during bootup */
1266static unsigned long long loops_per_msec; /* based on boot_delay */
1267
1268static int __init boot_delay_setup(char *str)
1269{
1270 unsigned long lpj;
1271
1272 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1273 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1274
1275 get_option(&str, &boot_delay);
1276 if (boot_delay > 10 * 1000)
1277 boot_delay = 0;
1278
1279 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1280 "HZ: %d, loops_per_msec: %llu\n",
1281 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1282 return 0;
1283}
1284early_param("boot_delay", boot_delay_setup);
1285
1286static void boot_delay_msec(int level)
1287{
1288 unsigned long long k;
1289 unsigned long timeout;
1290
1291 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1292 || suppress_message_printing(level)) {
1293 return;
1294 }
1295
1296 k = (unsigned long long)loops_per_msec * boot_delay;
1297
1298 timeout = jiffies + msecs_to_jiffies(boot_delay);
1299 while (k) {
1300 k--;
1301 cpu_relax();
1302 /*
1303 * use (volatile) jiffies to prevent
1304 * compiler reduction; loop termination via jiffies
1305 * is secondary and may or may not happen.
1306 */
1307 if (time_after(jiffies, timeout))
1308 break;
1309 touch_nmi_watchdog();
1310 }
1311}
1312#else
1313static inline void boot_delay_msec(int level)
1314{
1315}
1316#endif
1317
1318static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1319module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1320
1321static size_t print_syslog(unsigned int level, char *buf)
1322{
1323 return sprintf(buf, "<%u>", level);
1324}
1325
1326static size_t print_time(u64 ts, char *buf)
1327{
1328 unsigned long rem_nsec = do_div(ts, 1000000000);
1329
1330 return sprintf(buf, "[%5lu.%06lu]",
1331 (unsigned long)ts, rem_nsec / 1000);
1332}
1333
1334#ifdef CONFIG_PRINTK_CALLER
1335static size_t print_caller(u32 id, char *buf)
1336{
1337 char caller[12];
1338
1339 snprintf(caller, sizeof(caller), "%c%u",
1340 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1341 return sprintf(buf, "[%6s]", caller);
1342}
1343#else
1344#define print_caller(id, buf) 0
1345#endif
1346
1347static size_t info_print_prefix(const struct printk_info *info, bool syslog,
1348 bool time, char *buf)
1349{
1350 size_t len = 0;
1351
1352 if (syslog)
1353 len = print_syslog((info->facility << 3) | info->level, buf);
1354
1355 if (time)
1356 len += print_time(info->ts_nsec, buf + len);
1357
1358 len += print_caller(info->caller_id, buf + len);
1359
1360 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1361 buf[len++] = ' ';
1362 buf[len] = '\0';
1363 }
1364
1365 return len;
1366}
1367
1368/*
1369 * Prepare the record for printing. The text is shifted within the given
1370 * buffer to avoid a need for another one. The following operations are
1371 * done:
1372 *
1373 * - Add prefix for each line.
1374 * - Drop truncated lines that no longer fit into the buffer.
1375 * - Add the trailing newline that has been removed in vprintk_store().
1376 * - Add a string terminator.
1377 *
1378 * Since the produced string is always terminated, the maximum possible
1379 * return value is @r->text_buf_size - 1;
1380 *
1381 * Return: The length of the updated/prepared text, including the added
1382 * prefixes and the newline. The terminator is not counted. The dropped
1383 * line(s) are not counted.
1384 */
1385static size_t record_print_text(struct printk_record *r, bool syslog,
1386 bool time)
1387{
1388 size_t text_len = r->info->text_len;
1389 size_t buf_size = r->text_buf_size;
1390 char *text = r->text_buf;
1391 char prefix[PRINTK_PREFIX_MAX];
1392 bool truncated = false;
1393 size_t prefix_len;
1394 size_t line_len;
1395 size_t len = 0;
1396 char *next;
1397
1398 /*
1399 * If the message was truncated because the buffer was not large
1400 * enough, treat the available text as if it were the full text.
1401 */
1402 if (text_len > buf_size)
1403 text_len = buf_size;
1404
1405 prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1406
1407 /*
1408 * @text_len: bytes of unprocessed text
1409 * @line_len: bytes of current line _without_ newline
1410 * @text: pointer to beginning of current line
1411 * @len: number of bytes prepared in r->text_buf
1412 */
1413 for (;;) {
1414 next = memchr(text, '\n', text_len);
1415 if (next) {
1416 line_len = next - text;
1417 } else {
1418 /* Drop truncated line(s). */
1419 if (truncated)
1420 break;
1421 line_len = text_len;
1422 }
1423
1424 /*
1425 * Truncate the text if there is not enough space to add the
1426 * prefix and a trailing newline and a terminator.
1427 */
1428 if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1429 /* Drop even the current line if no space. */
1430 if (len + prefix_len + line_len + 1 + 1 > buf_size)
1431 break;
1432
1433 text_len = buf_size - len - prefix_len - 1 - 1;
1434 truncated = true;
1435 }
1436
1437 memmove(text + prefix_len, text, text_len);
1438 memcpy(text, prefix, prefix_len);
1439
1440 /*
1441 * Increment the prepared length to include the text and
1442 * prefix that were just moved+copied. Also increment for the
1443 * newline at the end of this line. If this is the last line,
1444 * there is no newline, but it will be added immediately below.
1445 */
1446 len += prefix_len + line_len + 1;
1447 if (text_len == line_len) {
1448 /*
1449 * This is the last line. Add the trailing newline
1450 * removed in vprintk_store().
1451 */
1452 text[prefix_len + line_len] = '\n';
1453 break;
1454 }
1455
1456 /*
1457 * Advance beyond the added prefix and the related line with
1458 * its newline.
1459 */
1460 text += prefix_len + line_len + 1;
1461
1462 /*
1463 * The remaining text has only decreased by the line with its
1464 * newline.
1465 *
1466 * Note that @text_len can become zero. It happens when @text
1467 * ended with a newline (either due to truncation or the
1468 * original string ending with "\n\n"). The loop is correctly
1469 * repeated and (if not truncated) an empty line with a prefix
1470 * will be prepared.
1471 */
1472 text_len -= line_len + 1;
1473 }
1474
1475 /*
1476 * If a buffer was provided, it will be terminated. Space for the
1477 * string terminator is guaranteed to be available. The terminator is
1478 * not counted in the return value.
1479 */
1480 if (buf_size > 0)
1481 r->text_buf[len] = 0;
1482
1483 return len;
1484}
1485
1486static size_t get_record_print_text_size(struct printk_info *info,
1487 unsigned int line_count,
1488 bool syslog, bool time)
1489{
1490 char prefix[PRINTK_PREFIX_MAX];
1491 size_t prefix_len;
1492
1493 prefix_len = info_print_prefix(info, syslog, time, prefix);
1494
1495 /*
1496 * Each line will be preceded with a prefix. The intermediate
1497 * newlines are already within the text, but a final trailing
1498 * newline will be added.
1499 */
1500 return ((prefix_len * line_count) + info->text_len + 1);
1501}
1502
1503/*
1504 * Beginning with @start_seq, find the first record where it and all following
1505 * records up to (but not including) @max_seq fit into @size.
1506 *
1507 * @max_seq is simply an upper bound and does not need to exist. If the caller
1508 * does not require an upper bound, -1 can be used for @max_seq.
1509 */
1510static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1511 bool syslog, bool time)
1512{
1513 struct printk_info info;
1514 unsigned int line_count;
1515 size_t len = 0;
1516 u64 seq;
1517
1518 /* Determine the size of the records up to @max_seq. */
1519 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1520 if (info.seq >= max_seq)
1521 break;
1522 len += get_record_print_text_size(&info, line_count, syslog, time);
1523 }
1524
1525 /*
1526 * Adjust the upper bound for the next loop to avoid subtracting
1527 * lengths that were never added.
1528 */
1529 if (seq < max_seq)
1530 max_seq = seq;
1531
1532 /*
1533 * Move first record forward until length fits into the buffer. Ignore
1534 * newest messages that were not counted in the above cycle. Messages
1535 * might appear and get lost in the meantime. This is a best effort
1536 * that prevents an infinite loop that could occur with a retry.
1537 */
1538 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1539 if (len <= size || info.seq >= max_seq)
1540 break;
1541 len -= get_record_print_text_size(&info, line_count, syslog, time);
1542 }
1543
1544 return seq;
1545}
1546
1547/* The caller is responsible for making sure @size is greater than 0. */
1548static int syslog_print(char __user *buf, int size)
1549{
1550 struct printk_info info;
1551 struct printk_record r;
1552 char *text;
1553 int len = 0;
1554 u64 seq;
1555
1556 text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1557 if (!text)
1558 return -ENOMEM;
1559
1560 prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1561
1562 mutex_lock(&syslog_lock);
1563
1564 /*
1565 * Wait for the @syslog_seq record to be available. @syslog_seq may
1566 * change while waiting.
1567 */
1568 do {
1569 seq = syslog_seq;
1570
1571 mutex_unlock(&syslog_lock);
1572 /*
1573 * Guarantee this task is visible on the waitqueue before
1574 * checking the wake condition.
1575 *
1576 * The full memory barrier within set_current_state() of
1577 * prepare_to_wait_event() pairs with the full memory barrier
1578 * within wq_has_sleeper().
1579 *
1580 * This pairs with __wake_up_klogd:A.
1581 */
1582 len = wait_event_interruptible(log_wait,
1583 prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
1584 mutex_lock(&syslog_lock);
1585
1586 if (len)
1587 goto out;
1588 } while (syslog_seq != seq);
1589
1590 /*
1591 * Copy records that fit into the buffer. The above cycle makes sure
1592 * that the first record is always available.
1593 */
1594 do {
1595 size_t n;
1596 size_t skip;
1597 int err;
1598
1599 if (!prb_read_valid(prb, syslog_seq, &r))
1600 break;
1601
1602 if (r.info->seq != syslog_seq) {
1603 /* message is gone, move to next valid one */
1604 syslog_seq = r.info->seq;
1605 syslog_partial = 0;
1606 }
1607
1608 /*
1609 * To keep reading/counting partial line consistent,
1610 * use printk_time value as of the beginning of a line.
1611 */
1612 if (!syslog_partial)
1613 syslog_time = printk_time;
1614
1615 skip = syslog_partial;
1616 n = record_print_text(&r, true, syslog_time);
1617 if (n - syslog_partial <= size) {
1618 /* message fits into buffer, move forward */
1619 syslog_seq = r.info->seq + 1;
1620 n -= syslog_partial;
1621 syslog_partial = 0;
1622 } else if (!len){
1623 /* partial read(), remember position */
1624 n = size;
1625 syslog_partial += n;
1626 } else
1627 n = 0;
1628
1629 if (!n)
1630 break;
1631
1632 mutex_unlock(&syslog_lock);
1633 err = copy_to_user(buf, text + skip, n);
1634 mutex_lock(&syslog_lock);
1635
1636 if (err) {
1637 if (!len)
1638 len = -EFAULT;
1639 break;
1640 }
1641
1642 len += n;
1643 size -= n;
1644 buf += n;
1645 } while (size);
1646out:
1647 mutex_unlock(&syslog_lock);
1648 kfree(text);
1649 return len;
1650}
1651
1652static int syslog_print_all(char __user *buf, int size, bool clear)
1653{
1654 struct printk_info info;
1655 struct printk_record r;
1656 char *text;
1657 int len = 0;
1658 u64 seq;
1659 bool time;
1660
1661 text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1662 if (!text)
1663 return -ENOMEM;
1664
1665 time = printk_time;
1666 /*
1667 * Find first record that fits, including all following records,
1668 * into the user-provided buffer for this dump.
1669 */
1670 seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1671 size, true, time);
1672
1673 prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1674
1675 prb_for_each_record(seq, prb, seq, &r) {
1676 int textlen;
1677
1678 textlen = record_print_text(&r, true, time);
1679
1680 if (len + textlen > size) {
1681 seq--;
1682 break;
1683 }
1684
1685 if (copy_to_user(buf + len, text, textlen))
1686 len = -EFAULT;
1687 else
1688 len += textlen;
1689
1690 if (len < 0)
1691 break;
1692 }
1693
1694 if (clear) {
1695 mutex_lock(&syslog_lock);
1696 latched_seq_write(&clear_seq, seq);
1697 mutex_unlock(&syslog_lock);
1698 }
1699
1700 kfree(text);
1701 return len;
1702}
1703
1704static void syslog_clear(void)
1705{
1706 mutex_lock(&syslog_lock);
1707 latched_seq_write(&clear_seq, prb_next_seq(prb));
1708 mutex_unlock(&syslog_lock);
1709}
1710
1711int do_syslog(int type, char __user *buf, int len, int source)
1712{
1713 struct printk_info info;
1714 bool clear = false;
1715 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1716 int error;
1717
1718 error = check_syslog_permissions(type, source);
1719 if (error)
1720 return error;
1721
1722 switch (type) {
1723 case SYSLOG_ACTION_CLOSE: /* Close log */
1724 break;
1725 case SYSLOG_ACTION_OPEN: /* Open log */
1726 break;
1727 case SYSLOG_ACTION_READ: /* Read from log */
1728 if (!buf || len < 0)
1729 return -EINVAL;
1730 if (!len)
1731 return 0;
1732 if (!access_ok(buf, len))
1733 return -EFAULT;
1734 error = syslog_print(buf, len);
1735 break;
1736 /* Read/clear last kernel messages */
1737 case SYSLOG_ACTION_READ_CLEAR:
1738 clear = true;
1739 fallthrough;
1740 /* Read last kernel messages */
1741 case SYSLOG_ACTION_READ_ALL:
1742 if (!buf || len < 0)
1743 return -EINVAL;
1744 if (!len)
1745 return 0;
1746 if (!access_ok(buf, len))
1747 return -EFAULT;
1748 error = syslog_print_all(buf, len, clear);
1749 break;
1750 /* Clear ring buffer */
1751 case SYSLOG_ACTION_CLEAR:
1752 syslog_clear();
1753 break;
1754 /* Disable logging to console */
1755 case SYSLOG_ACTION_CONSOLE_OFF:
1756 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1757 saved_console_loglevel = console_loglevel;
1758 console_loglevel = minimum_console_loglevel;
1759 break;
1760 /* Enable logging to console */
1761 case SYSLOG_ACTION_CONSOLE_ON:
1762 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1763 console_loglevel = saved_console_loglevel;
1764 saved_console_loglevel = LOGLEVEL_DEFAULT;
1765 }
1766 break;
1767 /* Set level of messages printed to console */
1768 case SYSLOG_ACTION_CONSOLE_LEVEL:
1769 if (len < 1 || len > 8)
1770 return -EINVAL;
1771 if (len < minimum_console_loglevel)
1772 len = minimum_console_loglevel;
1773 console_loglevel = len;
1774 /* Implicitly re-enable logging to console */
1775 saved_console_loglevel = LOGLEVEL_DEFAULT;
1776 break;
1777 /* Number of chars in the log buffer */
1778 case SYSLOG_ACTION_SIZE_UNREAD:
1779 mutex_lock(&syslog_lock);
1780 if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1781 /* No unread messages. */
1782 mutex_unlock(&syslog_lock);
1783 return 0;
1784 }
1785 if (info.seq != syslog_seq) {
1786 /* messages are gone, move to first one */
1787 syslog_seq = info.seq;
1788 syslog_partial = 0;
1789 }
1790 if (source == SYSLOG_FROM_PROC) {
1791 /*
1792 * Short-cut for poll(/"proc/kmsg") which simply checks
1793 * for pending data, not the size; return the count of
1794 * records, not the length.
1795 */
1796 error = prb_next_seq(prb) - syslog_seq;
1797 } else {
1798 bool time = syslog_partial ? syslog_time : printk_time;
1799 unsigned int line_count;
1800 u64 seq;
1801
1802 prb_for_each_info(syslog_seq, prb, seq, &info,
1803 &line_count) {
1804 error += get_record_print_text_size(&info, line_count,
1805 true, time);
1806 time = printk_time;
1807 }
1808 error -= syslog_partial;
1809 }
1810 mutex_unlock(&syslog_lock);
1811 break;
1812 /* Size of the log buffer */
1813 case SYSLOG_ACTION_SIZE_BUFFER:
1814 error = log_buf_len;
1815 break;
1816 default:
1817 error = -EINVAL;
1818 break;
1819 }
1820
1821 return error;
1822}
1823
1824SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1825{
1826 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1827}
1828
1829/*
1830 * Special console_lock variants that help to reduce the risk of soft-lockups.
1831 * They allow to pass console_lock to another printk() call using a busy wait.
1832 */
1833
1834#ifdef CONFIG_LOCKDEP
1835static struct lockdep_map console_owner_dep_map = {
1836 .name = "console_owner"
1837};
1838#endif
1839
1840static DEFINE_RAW_SPINLOCK(console_owner_lock);
1841static struct task_struct *console_owner;
1842static bool console_waiter;
1843
1844/**
1845 * console_lock_spinning_enable - mark beginning of code where another
1846 * thread might safely busy wait
1847 *
1848 * This basically converts console_lock into a spinlock. This marks
1849 * the section where the console_lock owner can not sleep, because
1850 * there may be a waiter spinning (like a spinlock). Also it must be
1851 * ready to hand over the lock at the end of the section.
1852 */
1853static void console_lock_spinning_enable(void)
1854{
1855 /*
1856 * Do not use spinning in panic(). The panic CPU wants to keep the lock.
1857 * Non-panic CPUs abandon the flush anyway.
1858 *
1859 * Just keep the lockdep annotation. The panic-CPU should avoid
1860 * taking console_owner_lock because it might cause a deadlock.
1861 * This looks like the easiest way how to prevent false lockdep
1862 * reports without handling races a lockless way.
1863 */
1864 if (panic_in_progress())
1865 goto lockdep;
1866
1867 raw_spin_lock(&console_owner_lock);
1868 console_owner = current;
1869 raw_spin_unlock(&console_owner_lock);
1870
1871lockdep:
1872 /* The waiter may spin on us after setting console_owner */
1873 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1874}
1875
1876/**
1877 * console_lock_spinning_disable_and_check - mark end of code where another
1878 * thread was able to busy wait and check if there is a waiter
1879 * @cookie: cookie returned from console_srcu_read_lock()
1880 *
1881 * This is called at the end of the section where spinning is allowed.
1882 * It has two functions. First, it is a signal that it is no longer
1883 * safe to start busy waiting for the lock. Second, it checks if
1884 * there is a busy waiter and passes the lock rights to her.
1885 *
1886 * Important: Callers lose both the console_lock and the SRCU read lock if
1887 * there was a busy waiter. They must not touch items synchronized by
1888 * console_lock or SRCU read lock in this case.
1889 *
1890 * Return: 1 if the lock rights were passed, 0 otherwise.
1891 */
1892static int console_lock_spinning_disable_and_check(int cookie)
1893{
1894 int waiter;
1895
1896 /*
1897 * Ignore spinning waiters during panic() because they might get stopped
1898 * or blocked at any time,
1899 *
1900 * It is safe because nobody is allowed to start spinning during panic
1901 * in the first place. If there has been a waiter then non panic CPUs
1902 * might stay spinning. They would get stopped anyway. The panic context
1903 * will never start spinning and an interrupted spin on panic CPU will
1904 * never continue.
1905 */
1906 if (panic_in_progress()) {
1907 /* Keep lockdep happy. */
1908 spin_release(&console_owner_dep_map, _THIS_IP_);
1909 return 0;
1910 }
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 /*
2013 * Update @console_may_schedule for trylock because the previous
2014 * owner may have been schedulable.
2015 */
2016 console_may_schedule = 0;
2017
2018 return 1;
2019}
2020
2021/*
2022 * Recursion is tracked separately on each CPU. If NMIs are supported, an
2023 * additional NMI context per CPU is also separately tracked. Until per-CPU
2024 * is available, a separate "early tracking" is performed.
2025 */
2026static DEFINE_PER_CPU(u8, printk_count);
2027static u8 printk_count_early;
2028#ifdef CONFIG_HAVE_NMI
2029static DEFINE_PER_CPU(u8, printk_count_nmi);
2030static u8 printk_count_nmi_early;
2031#endif
2032
2033/*
2034 * Recursion is limited to keep the output sane. printk() should not require
2035 * more than 1 level of recursion (allowing, for example, printk() to trigger
2036 * a WARN), but a higher value is used in case some printk-internal errors
2037 * exist, such as the ringbuffer validation checks failing.
2038 */
2039#define PRINTK_MAX_RECURSION 3
2040
2041/*
2042 * Return a pointer to the dedicated counter for the CPU+context of the
2043 * caller.
2044 */
2045static u8 *__printk_recursion_counter(void)
2046{
2047#ifdef CONFIG_HAVE_NMI
2048 if (in_nmi()) {
2049 if (printk_percpu_data_ready())
2050 return this_cpu_ptr(&printk_count_nmi);
2051 return &printk_count_nmi_early;
2052 }
2053#endif
2054 if (printk_percpu_data_ready())
2055 return this_cpu_ptr(&printk_count);
2056 return &printk_count_early;
2057}
2058
2059/*
2060 * Enter recursion tracking. Interrupts are disabled to simplify tracking.
2061 * The caller must check the boolean return value to see if the recursion is
2062 * allowed. On failure, interrupts are not disabled.
2063 *
2064 * @recursion_ptr must be a variable of type (u8 *) and is the same variable
2065 * that is passed to printk_exit_irqrestore().
2066 */
2067#define printk_enter_irqsave(recursion_ptr, flags) \
2068({ \
2069 bool success = true; \
2070 \
2071 typecheck(u8 *, recursion_ptr); \
2072 local_irq_save(flags); \
2073 (recursion_ptr) = __printk_recursion_counter(); \
2074 if (*(recursion_ptr) > PRINTK_MAX_RECURSION) { \
2075 local_irq_restore(flags); \
2076 success = false; \
2077 } else { \
2078 (*(recursion_ptr))++; \
2079 } \
2080 success; \
2081})
2082
2083/* Exit recursion tracking, restoring interrupts. */
2084#define printk_exit_irqrestore(recursion_ptr, flags) \
2085 do { \
2086 typecheck(u8 *, recursion_ptr); \
2087 (*(recursion_ptr))--; \
2088 local_irq_restore(flags); \
2089 } while (0)
2090
2091int printk_delay_msec __read_mostly;
2092
2093static inline void printk_delay(int level)
2094{
2095 boot_delay_msec(level);
2096
2097 if (unlikely(printk_delay_msec)) {
2098 int m = printk_delay_msec;
2099
2100 while (m--) {
2101 mdelay(1);
2102 touch_nmi_watchdog();
2103 }
2104 }
2105}
2106
2107static inline u32 printk_caller_id(void)
2108{
2109 return in_task() ? task_pid_nr(current) :
2110 0x80000000 + smp_processor_id();
2111}
2112
2113/**
2114 * printk_parse_prefix - Parse level and control flags.
2115 *
2116 * @text: The terminated text message.
2117 * @level: A pointer to the current level value, will be updated.
2118 * @flags: A pointer to the current printk_info flags, will be updated.
2119 *
2120 * @level may be NULL if the caller is not interested in the parsed value.
2121 * Otherwise the variable pointed to by @level must be set to
2122 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2123 *
2124 * @flags may be NULL if the caller is not interested in the parsed value.
2125 * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2126 * value.
2127 *
2128 * Return: The length of the parsed level and control flags.
2129 */
2130u16 printk_parse_prefix(const char *text, int *level,
2131 enum printk_info_flags *flags)
2132{
2133 u16 prefix_len = 0;
2134 int kern_level;
2135
2136 while (*text) {
2137 kern_level = printk_get_level(text);
2138 if (!kern_level)
2139 break;
2140
2141 switch (kern_level) {
2142 case '0' ... '7':
2143 if (level && *level == LOGLEVEL_DEFAULT)
2144 *level = kern_level - '0';
2145 break;
2146 case 'c': /* KERN_CONT */
2147 if (flags)
2148 *flags |= LOG_CONT;
2149 }
2150
2151 prefix_len += 2;
2152 text += 2;
2153 }
2154
2155 return prefix_len;
2156}
2157
2158__printf(5, 0)
2159static u16 printk_sprint(char *text, u16 size, int facility,
2160 enum printk_info_flags *flags, const char *fmt,
2161 va_list args)
2162{
2163 u16 text_len;
2164
2165 text_len = vscnprintf(text, size, fmt, args);
2166
2167 /* Mark and strip a trailing newline. */
2168 if (text_len && text[text_len - 1] == '\n') {
2169 text_len--;
2170 *flags |= LOG_NEWLINE;
2171 }
2172
2173 /* Strip log level and control flags. */
2174 if (facility == 0) {
2175 u16 prefix_len;
2176
2177 prefix_len = printk_parse_prefix(text, NULL, NULL);
2178 if (prefix_len) {
2179 text_len -= prefix_len;
2180 memmove(text, text + prefix_len, text_len);
2181 }
2182 }
2183
2184 trace_console(text, text_len);
2185
2186 return text_len;
2187}
2188
2189__printf(4, 0)
2190int vprintk_store(int facility, int level,
2191 const struct dev_printk_info *dev_info,
2192 const char *fmt, va_list args)
2193{
2194 struct prb_reserved_entry e;
2195 enum printk_info_flags flags = 0;
2196 struct printk_record r;
2197 unsigned long irqflags;
2198 u16 trunc_msg_len = 0;
2199 char prefix_buf[8];
2200 u8 *recursion_ptr;
2201 u16 reserve_size;
2202 va_list args2;
2203 u32 caller_id;
2204 u16 text_len;
2205 int ret = 0;
2206 u64 ts_nsec;
2207
2208 if (!printk_enter_irqsave(recursion_ptr, irqflags))
2209 return 0;
2210
2211 /*
2212 * Since the duration of printk() can vary depending on the message
2213 * and state of the ringbuffer, grab the timestamp now so that it is
2214 * close to the call of printk(). This provides a more deterministic
2215 * timestamp with respect to the caller.
2216 */
2217 ts_nsec = local_clock();
2218
2219 caller_id = printk_caller_id();
2220
2221 /*
2222 * The sprintf needs to come first since the syslog prefix might be
2223 * passed in as a parameter. An extra byte must be reserved so that
2224 * later the vscnprintf() into the reserved buffer has room for the
2225 * terminating '\0', which is not counted by vsnprintf().
2226 */
2227 va_copy(args2, args);
2228 reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2229 va_end(args2);
2230
2231 if (reserve_size > PRINTKRB_RECORD_MAX)
2232 reserve_size = PRINTKRB_RECORD_MAX;
2233
2234 /* Extract log level or control flags. */
2235 if (facility == 0)
2236 printk_parse_prefix(&prefix_buf[0], &level, &flags);
2237
2238 if (level == LOGLEVEL_DEFAULT)
2239 level = default_message_loglevel;
2240
2241 if (dev_info)
2242 flags |= LOG_NEWLINE;
2243
2244 if (flags & LOG_CONT) {
2245 prb_rec_init_wr(&r, reserve_size);
2246 if (prb_reserve_in_last(&e, prb, &r, caller_id, PRINTKRB_RECORD_MAX)) {
2247 text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2248 facility, &flags, fmt, args);
2249 r.info->text_len += text_len;
2250
2251 if (flags & LOG_NEWLINE) {
2252 r.info->flags |= LOG_NEWLINE;
2253 prb_final_commit(&e);
2254 } else {
2255 prb_commit(&e);
2256 }
2257
2258 ret = text_len;
2259 goto out;
2260 }
2261 }
2262
2263 /*
2264 * Explicitly initialize the record before every prb_reserve() call.
2265 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2266 * structure when they fail.
2267 */
2268 prb_rec_init_wr(&r, reserve_size);
2269 if (!prb_reserve(&e, prb, &r)) {
2270 /* truncate the message if it is too long for empty buffer */
2271 truncate_msg(&reserve_size, &trunc_msg_len);
2272
2273 prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2274 if (!prb_reserve(&e, prb, &r))
2275 goto out;
2276 }
2277
2278 /* fill message */
2279 text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2280 if (trunc_msg_len)
2281 memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2282 r.info->text_len = text_len + trunc_msg_len;
2283 r.info->facility = facility;
2284 r.info->level = level & 7;
2285 r.info->flags = flags & 0x1f;
2286 r.info->ts_nsec = ts_nsec;
2287 r.info->caller_id = caller_id;
2288 if (dev_info)
2289 memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2290
2291 /* A message without a trailing newline can be continued. */
2292 if (!(flags & LOG_NEWLINE))
2293 prb_commit(&e);
2294 else
2295 prb_final_commit(&e);
2296
2297 ret = text_len + trunc_msg_len;
2298out:
2299 printk_exit_irqrestore(recursion_ptr, irqflags);
2300 return ret;
2301}
2302
2303asmlinkage int vprintk_emit(int facility, int level,
2304 const struct dev_printk_info *dev_info,
2305 const char *fmt, va_list args)
2306{
2307 int printed_len;
2308 bool in_sched = false;
2309
2310 /* Suppress unimportant messages after panic happens */
2311 if (unlikely(suppress_printk))
2312 return 0;
2313
2314 /*
2315 * The messages on the panic CPU are the most important. If
2316 * non-panic CPUs are generating any messages, they will be
2317 * silently dropped.
2318 */
2319 if (other_cpu_in_panic())
2320 return 0;
2321
2322 if (level == LOGLEVEL_SCHED) {
2323 level = LOGLEVEL_DEFAULT;
2324 in_sched = true;
2325 }
2326
2327 printk_delay(level);
2328
2329 printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2330
2331 /* If called from the scheduler, we can not call up(). */
2332 if (!in_sched) {
2333 /*
2334 * The caller may be holding system-critical or
2335 * timing-sensitive locks. Disable preemption during
2336 * printing of all remaining records to all consoles so that
2337 * this context can return as soon as possible. Hopefully
2338 * another printk() caller will take over the printing.
2339 */
2340 preempt_disable();
2341 /*
2342 * Try to acquire and then immediately release the console
2343 * semaphore. The release will print out buffers. With the
2344 * spinning variant, this context tries to take over the
2345 * printing from another printing context.
2346 */
2347 if (console_trylock_spinning())
2348 console_unlock();
2349 preempt_enable();
2350 }
2351
2352 if (in_sched)
2353 defer_console_output();
2354 else
2355 wake_up_klogd();
2356
2357 return printed_len;
2358}
2359EXPORT_SYMBOL(vprintk_emit);
2360
2361int vprintk_default(const char *fmt, va_list args)
2362{
2363 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2364}
2365EXPORT_SYMBOL_GPL(vprintk_default);
2366
2367asmlinkage __visible int _printk(const char *fmt, ...)
2368{
2369 va_list args;
2370 int r;
2371
2372 va_start(args, fmt);
2373 r = vprintk(fmt, args);
2374 va_end(args);
2375
2376 return r;
2377}
2378EXPORT_SYMBOL(_printk);
2379
2380static bool pr_flush(int timeout_ms, bool reset_on_progress);
2381static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress);
2382
2383#else /* CONFIG_PRINTK */
2384
2385#define printk_time false
2386
2387#define prb_read_valid(rb, seq, r) false
2388#define prb_first_valid_seq(rb) 0
2389#define prb_next_seq(rb) 0
2390
2391static u64 syslog_seq;
2392
2393static bool pr_flush(int timeout_ms, bool reset_on_progress) { return true; }
2394static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress) { return true; }
2395
2396#endif /* CONFIG_PRINTK */
2397
2398#ifdef CONFIG_EARLY_PRINTK
2399struct console *early_console;
2400
2401asmlinkage __visible void early_printk(const char *fmt, ...)
2402{
2403 va_list ap;
2404 char buf[512];
2405 int n;
2406
2407 if (!early_console)
2408 return;
2409
2410 va_start(ap, fmt);
2411 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2412 va_end(ap);
2413
2414 early_console->write(early_console, buf, n);
2415}
2416#endif
2417
2418static void set_user_specified(struct console_cmdline *c, bool user_specified)
2419{
2420 if (!user_specified)
2421 return;
2422
2423 /*
2424 * @c console was defined by the user on the command line.
2425 * Do not clear when added twice also by SPCR or the device tree.
2426 */
2427 c->user_specified = true;
2428 /* At least one console defined by the user on the command line. */
2429 console_set_on_cmdline = 1;
2430}
2431
2432static int __add_preferred_console(const char *name, const short idx, char *options,
2433 char *brl_options, bool user_specified)
2434{
2435 struct console_cmdline *c;
2436 int i;
2437
2438 /*
2439 * We use a signed short index for struct console for device drivers to
2440 * indicate a not yet assigned index or port. However, a negative index
2441 * value is not valid for preferred console.
2442 */
2443 if (idx < 0)
2444 return -EINVAL;
2445
2446 /*
2447 * See if this tty is not yet registered, and
2448 * if we have a slot free.
2449 */
2450 for (i = 0, c = console_cmdline;
2451 i < MAX_CMDLINECONSOLES && c->name[0];
2452 i++, c++) {
2453 if (strcmp(c->name, name) == 0 && c->index == idx) {
2454 if (!brl_options)
2455 preferred_console = i;
2456 set_user_specified(c, user_specified);
2457 return 0;
2458 }
2459 }
2460 if (i == MAX_CMDLINECONSOLES)
2461 return -E2BIG;
2462 if (!brl_options)
2463 preferred_console = i;
2464 strscpy(c->name, name, sizeof(c->name));
2465 c->options = options;
2466 set_user_specified(c, user_specified);
2467 braille_set_options(c, brl_options);
2468
2469 c->index = idx;
2470 return 0;
2471}
2472
2473static int __init console_msg_format_setup(char *str)
2474{
2475 if (!strcmp(str, "syslog"))
2476 console_msg_format = MSG_FORMAT_SYSLOG;
2477 if (!strcmp(str, "default"))
2478 console_msg_format = MSG_FORMAT_DEFAULT;
2479 return 1;
2480}
2481__setup("console_msg_format=", console_msg_format_setup);
2482
2483/*
2484 * Set up a console. Called via do_early_param() in init/main.c
2485 * for each "console=" parameter in the boot command line.
2486 */
2487static int __init console_setup(char *str)
2488{
2489 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2490 char *s, *options, *brl_options = NULL;
2491 int idx;
2492
2493 /*
2494 * console="" or console=null have been suggested as a way to
2495 * disable console output. Use ttynull that has been created
2496 * for exactly this purpose.
2497 */
2498 if (str[0] == 0 || strcmp(str, "null") == 0) {
2499 __add_preferred_console("ttynull", 0, NULL, NULL, true);
2500 return 1;
2501 }
2502
2503 if (_braille_console_setup(&str, &brl_options))
2504 return 1;
2505
2506 /*
2507 * Decode str into name, index, options.
2508 */
2509 if (str[0] >= '0' && str[0] <= '9') {
2510 strcpy(buf, "ttyS");
2511 strncpy(buf + 4, str, sizeof(buf) - 5);
2512 } else {
2513 strncpy(buf, str, sizeof(buf) - 1);
2514 }
2515 buf[sizeof(buf) - 1] = 0;
2516 options = strchr(str, ',');
2517 if (options)
2518 *(options++) = 0;
2519#ifdef __sparc__
2520 if (!strcmp(str, "ttya"))
2521 strcpy(buf, "ttyS0");
2522 if (!strcmp(str, "ttyb"))
2523 strcpy(buf, "ttyS1");
2524#endif
2525 for (s = buf; *s; s++)
2526 if (isdigit(*s) || *s == ',')
2527 break;
2528 idx = simple_strtoul(s, NULL, 10);
2529 *s = 0;
2530
2531 __add_preferred_console(buf, idx, options, brl_options, true);
2532 return 1;
2533}
2534__setup("console=", console_setup);
2535
2536/**
2537 * add_preferred_console - add a device to the list of preferred consoles.
2538 * @name: device name
2539 * @idx: device index
2540 * @options: options for this console
2541 *
2542 * The last preferred console added will be used for kernel messages
2543 * and stdin/out/err for init. Normally this is used by console_setup
2544 * above to handle user-supplied console arguments; however it can also
2545 * be used by arch-specific code either to override the user or more
2546 * commonly to provide a default console (ie from PROM variables) when
2547 * the user has not supplied one.
2548 */
2549int add_preferred_console(const char *name, const short idx, char *options)
2550{
2551 return __add_preferred_console(name, idx, options, NULL, false);
2552}
2553
2554bool console_suspend_enabled = true;
2555EXPORT_SYMBOL(console_suspend_enabled);
2556
2557static int __init console_suspend_disable(char *str)
2558{
2559 console_suspend_enabled = false;
2560 return 1;
2561}
2562__setup("no_console_suspend", console_suspend_disable);
2563module_param_named(console_suspend, console_suspend_enabled,
2564 bool, S_IRUGO | S_IWUSR);
2565MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2566 " and hibernate operations");
2567
2568static bool printk_console_no_auto_verbose;
2569
2570void console_verbose(void)
2571{
2572 if (console_loglevel && !printk_console_no_auto_verbose)
2573 console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2574}
2575EXPORT_SYMBOL_GPL(console_verbose);
2576
2577module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2578MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2579
2580/**
2581 * suspend_console - suspend the console subsystem
2582 *
2583 * This disables printk() while we go into suspend states
2584 */
2585void suspend_console(void)
2586{
2587 struct console *con;
2588
2589 if (!console_suspend_enabled)
2590 return;
2591 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2592 pr_flush(1000, true);
2593
2594 console_list_lock();
2595 for_each_console(con)
2596 console_srcu_write_flags(con, con->flags | CON_SUSPENDED);
2597 console_list_unlock();
2598
2599 /*
2600 * Ensure that all SRCU list walks have completed. All printing
2601 * contexts must be able to see that they are suspended so that it
2602 * is guaranteed that all printing has stopped when this function
2603 * completes.
2604 */
2605 synchronize_srcu(&console_srcu);
2606}
2607
2608void resume_console(void)
2609{
2610 struct console *con;
2611
2612 if (!console_suspend_enabled)
2613 return;
2614
2615 console_list_lock();
2616 for_each_console(con)
2617 console_srcu_write_flags(con, con->flags & ~CON_SUSPENDED);
2618 console_list_unlock();
2619
2620 /*
2621 * Ensure that all SRCU list walks have completed. All printing
2622 * contexts must be able to see they are no longer suspended so
2623 * that they are guaranteed to wake up and resume printing.
2624 */
2625 synchronize_srcu(&console_srcu);
2626
2627 pr_flush(1000, true);
2628}
2629
2630/**
2631 * console_cpu_notify - print deferred console messages after CPU hotplug
2632 * @cpu: unused
2633 *
2634 * If printk() is called from a CPU that is not online yet, the messages
2635 * will be printed on the console only if there are CON_ANYTIME consoles.
2636 * This function is called when a new CPU comes online (or fails to come
2637 * up) or goes offline.
2638 */
2639static int console_cpu_notify(unsigned int cpu)
2640{
2641 if (!cpuhp_tasks_frozen) {
2642 /* If trylock fails, someone else is doing the printing */
2643 if (console_trylock())
2644 console_unlock();
2645 }
2646 return 0;
2647}
2648
2649/**
2650 * console_lock - block the console subsystem from printing
2651 *
2652 * Acquires a lock which guarantees that no consoles will
2653 * be in or enter their write() callback.
2654 *
2655 * Can sleep, returns nothing.
2656 */
2657void console_lock(void)
2658{
2659 might_sleep();
2660
2661 /* On panic, the console_lock must be left to the panic cpu. */
2662 while (other_cpu_in_panic())
2663 msleep(1000);
2664
2665 down_console_sem();
2666 console_locked = 1;
2667 console_may_schedule = 1;
2668}
2669EXPORT_SYMBOL(console_lock);
2670
2671/**
2672 * console_trylock - try to block the console subsystem from printing
2673 *
2674 * Try to acquire a lock which guarantees that no consoles will
2675 * be in or enter their write() callback.
2676 *
2677 * returns 1 on success, and 0 on failure to acquire the lock.
2678 */
2679int console_trylock(void)
2680{
2681 /* On panic, the console_lock must be left to the panic cpu. */
2682 if (other_cpu_in_panic())
2683 return 0;
2684 if (down_trylock_console_sem())
2685 return 0;
2686 console_locked = 1;
2687 console_may_schedule = 0;
2688 return 1;
2689}
2690EXPORT_SYMBOL(console_trylock);
2691
2692int is_console_locked(void)
2693{
2694 return console_locked;
2695}
2696EXPORT_SYMBOL(is_console_locked);
2697
2698/*
2699 * Check if the given console is currently capable and allowed to print
2700 * records.
2701 *
2702 * Requires the console_srcu_read_lock.
2703 */
2704static inline bool console_is_usable(struct console *con)
2705{
2706 short flags = console_srcu_read_flags(con);
2707
2708 if (!(flags & CON_ENABLED))
2709 return false;
2710
2711 if ((flags & CON_SUSPENDED))
2712 return false;
2713
2714 if (!con->write)
2715 return false;
2716
2717 /*
2718 * Console drivers may assume that per-cpu resources have been
2719 * allocated. So unless they're explicitly marked as being able to
2720 * cope (CON_ANYTIME) don't call them until this CPU is officially up.
2721 */
2722 if (!cpu_online(raw_smp_processor_id()) && !(flags & CON_ANYTIME))
2723 return false;
2724
2725 return true;
2726}
2727
2728static void __console_unlock(void)
2729{
2730 console_locked = 0;
2731 up_console_sem();
2732}
2733
2734#ifdef CONFIG_PRINTK
2735
2736/*
2737 * Prepend the message in @pmsg->pbufs->outbuf with a "dropped message". This
2738 * is achieved by shifting the existing message over and inserting the dropped
2739 * message.
2740 *
2741 * @pmsg is the printk message to prepend.
2742 *
2743 * @dropped is the dropped count to report in the dropped message.
2744 *
2745 * If the message text in @pmsg->pbufs->outbuf does not have enough space for
2746 * the dropped message, the message text will be sufficiently truncated.
2747 *
2748 * If @pmsg->pbufs->outbuf is modified, @pmsg->outbuf_len is updated.
2749 */
2750void console_prepend_dropped(struct printk_message *pmsg, unsigned long dropped)
2751{
2752 struct printk_buffers *pbufs = pmsg->pbufs;
2753 const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2754 const size_t outbuf_sz = sizeof(pbufs->outbuf);
2755 char *scratchbuf = &pbufs->scratchbuf[0];
2756 char *outbuf = &pbufs->outbuf[0];
2757 size_t len;
2758
2759 len = scnprintf(scratchbuf, scratchbuf_sz,
2760 "** %lu printk messages dropped **\n", dropped);
2761
2762 /*
2763 * Make sure outbuf is sufficiently large before prepending.
2764 * Keep at least the prefix when the message must be truncated.
2765 * It is a rather theoretical problem when someone tries to
2766 * use a minimalist buffer.
2767 */
2768 if (WARN_ON_ONCE(len + PRINTK_PREFIX_MAX >= outbuf_sz))
2769 return;
2770
2771 if (pmsg->outbuf_len + len >= outbuf_sz) {
2772 /* Truncate the message, but keep it terminated. */
2773 pmsg->outbuf_len = outbuf_sz - (len + 1);
2774 outbuf[pmsg->outbuf_len] = 0;
2775 }
2776
2777 memmove(outbuf + len, outbuf, pmsg->outbuf_len + 1);
2778 memcpy(outbuf, scratchbuf, len);
2779 pmsg->outbuf_len += len;
2780}
2781
2782/*
2783 * Read and format the specified record (or a later record if the specified
2784 * record is not available).
2785 *
2786 * @pmsg will contain the formatted result. @pmsg->pbufs must point to a
2787 * struct printk_buffers.
2788 *
2789 * @seq is the record to read and format. If it is not available, the next
2790 * valid record is read.
2791 *
2792 * @is_extended specifies if the message should be formatted for extended
2793 * console output.
2794 *
2795 * @may_supress specifies if records may be skipped based on loglevel.
2796 *
2797 * Returns false if no record is available. Otherwise true and all fields
2798 * of @pmsg are valid. (See the documentation of struct printk_message
2799 * for information about the @pmsg fields.)
2800 */
2801bool printk_get_next_message(struct printk_message *pmsg, u64 seq,
2802 bool is_extended, bool may_suppress)
2803{
2804 struct printk_buffers *pbufs = pmsg->pbufs;
2805 const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2806 const size_t outbuf_sz = sizeof(pbufs->outbuf);
2807 char *scratchbuf = &pbufs->scratchbuf[0];
2808 char *outbuf = &pbufs->outbuf[0];
2809 struct printk_info info;
2810 struct printk_record r;
2811 size_t len = 0;
2812
2813 /*
2814 * Formatting extended messages requires a separate buffer, so use the
2815 * scratch buffer to read in the ringbuffer text.
2816 *
2817 * Formatting normal messages is done in-place, so read the ringbuffer
2818 * text directly into the output buffer.
2819 */
2820 if (is_extended)
2821 prb_rec_init_rd(&r, &info, scratchbuf, scratchbuf_sz);
2822 else
2823 prb_rec_init_rd(&r, &info, outbuf, outbuf_sz);
2824
2825 if (!prb_read_valid(prb, seq, &r))
2826 return false;
2827
2828 pmsg->seq = r.info->seq;
2829 pmsg->dropped = r.info->seq - seq;
2830
2831 /* Skip record that has level above the console loglevel. */
2832 if (may_suppress && suppress_message_printing(r.info->level))
2833 goto out;
2834
2835 if (is_extended) {
2836 len = info_print_ext_header(outbuf, outbuf_sz, r.info);
2837 len += msg_print_ext_body(outbuf + len, outbuf_sz - len,
2838 &r.text_buf[0], r.info->text_len, &r.info->dev_info);
2839 } else {
2840 len = record_print_text(&r, console_msg_format & MSG_FORMAT_SYSLOG, printk_time);
2841 }
2842out:
2843 pmsg->outbuf_len = len;
2844 return true;
2845}
2846
2847/*
2848 * Used as the printk buffers for non-panic, serialized console printing.
2849 * This is for legacy (!CON_NBCON) as well as all boot (CON_BOOT) consoles.
2850 * Its usage requires the console_lock held.
2851 */
2852struct printk_buffers printk_shared_pbufs;
2853
2854/*
2855 * Print one record for the given console. The record printed is whatever
2856 * record is the next available record for the given console.
2857 *
2858 * @handover will be set to true if a printk waiter has taken over the
2859 * console_lock, in which case the caller is no longer holding both the
2860 * console_lock and the SRCU read lock. Otherwise it is set to false.
2861 *
2862 * @cookie is the cookie from the SRCU read lock.
2863 *
2864 * Returns false if the given console has no next record to print, otherwise
2865 * true.
2866 *
2867 * Requires the console_lock and the SRCU read lock.
2868 */
2869static bool console_emit_next_record(struct console *con, bool *handover, int cookie)
2870{
2871 bool is_extended = console_srcu_read_flags(con) & CON_EXTENDED;
2872 char *outbuf = &printk_shared_pbufs.outbuf[0];
2873 struct printk_message pmsg = {
2874 .pbufs = &printk_shared_pbufs,
2875 };
2876 unsigned long flags;
2877
2878 *handover = false;
2879
2880 if (!printk_get_next_message(&pmsg, con->seq, is_extended, true))
2881 return false;
2882
2883 con->dropped += pmsg.dropped;
2884
2885 /* Skip messages of formatted length 0. */
2886 if (pmsg.outbuf_len == 0) {
2887 con->seq = pmsg.seq + 1;
2888 goto skip;
2889 }
2890
2891 if (con->dropped && !is_extended) {
2892 console_prepend_dropped(&pmsg, con->dropped);
2893 con->dropped = 0;
2894 }
2895
2896 /*
2897 * While actively printing out messages, if another printk()
2898 * were to occur on another CPU, it may wait for this one to
2899 * finish. This task can not be preempted if there is a
2900 * waiter waiting to take over.
2901 *
2902 * Interrupts are disabled because the hand over to a waiter
2903 * must not be interrupted until the hand over is completed
2904 * (@console_waiter is cleared).
2905 */
2906 printk_safe_enter_irqsave(flags);
2907 console_lock_spinning_enable();
2908
2909 /* Do not trace print latency. */
2910 stop_critical_timings();
2911
2912 /* Write everything out to the hardware. */
2913 con->write(con, outbuf, pmsg.outbuf_len);
2914
2915 start_critical_timings();
2916
2917 con->seq = pmsg.seq + 1;
2918
2919 *handover = console_lock_spinning_disable_and_check(cookie);
2920 printk_safe_exit_irqrestore(flags);
2921skip:
2922 return true;
2923}
2924
2925#else
2926
2927static bool console_emit_next_record(struct console *con, bool *handover, int cookie)
2928{
2929 *handover = false;
2930 return false;
2931}
2932
2933#endif /* CONFIG_PRINTK */
2934
2935/*
2936 * Print out all remaining records to all consoles.
2937 *
2938 * @do_cond_resched is set by the caller. It can be true only in schedulable
2939 * context.
2940 *
2941 * @next_seq is set to the sequence number after the last available record.
2942 * The value is valid only when this function returns true. It means that all
2943 * usable consoles are completely flushed.
2944 *
2945 * @handover will be set to true if a printk waiter has taken over the
2946 * console_lock, in which case the caller is no longer holding the
2947 * console_lock. Otherwise it is set to false.
2948 *
2949 * Returns true when there was at least one usable console and all messages
2950 * were flushed to all usable consoles. A returned false informs the caller
2951 * that everything was not flushed (either there were no usable consoles or
2952 * another context has taken over printing or it is a panic situation and this
2953 * is not the panic CPU). Regardless the reason, the caller should assume it
2954 * is not useful to immediately try again.
2955 *
2956 * Requires the console_lock.
2957 */
2958static bool console_flush_all(bool do_cond_resched, u64 *next_seq, bool *handover)
2959{
2960 bool any_usable = false;
2961 struct console *con;
2962 bool any_progress;
2963 int cookie;
2964
2965 *next_seq = 0;
2966 *handover = false;
2967
2968 do {
2969 any_progress = false;
2970
2971 cookie = console_srcu_read_lock();
2972 for_each_console_srcu(con) {
2973 bool progress;
2974
2975 if (!console_is_usable(con))
2976 continue;
2977 any_usable = true;
2978
2979 progress = console_emit_next_record(con, handover, cookie);
2980
2981 /*
2982 * If a handover has occurred, the SRCU read lock
2983 * is already released.
2984 */
2985 if (*handover)
2986 return false;
2987
2988 /* Track the next of the highest seq flushed. */
2989 if (con->seq > *next_seq)
2990 *next_seq = con->seq;
2991
2992 if (!progress)
2993 continue;
2994 any_progress = true;
2995
2996 /* Allow panic_cpu to take over the consoles safely. */
2997 if (other_cpu_in_panic())
2998 goto abandon;
2999
3000 if (do_cond_resched)
3001 cond_resched();
3002 }
3003 console_srcu_read_unlock(cookie);
3004 } while (any_progress);
3005
3006 return any_usable;
3007
3008abandon:
3009 console_srcu_read_unlock(cookie);
3010 return false;
3011}
3012
3013/**
3014 * console_unlock - unblock the console subsystem from printing
3015 *
3016 * Releases the console_lock which the caller holds to block printing of
3017 * the console subsystem.
3018 *
3019 * While the console_lock was held, console output may have been buffered
3020 * by printk(). If this is the case, console_unlock(); emits
3021 * the output prior to releasing the lock.
3022 *
3023 * console_unlock(); may be called from any context.
3024 */
3025void console_unlock(void)
3026{
3027 bool do_cond_resched;
3028 bool handover;
3029 bool flushed;
3030 u64 next_seq;
3031
3032 /*
3033 * Console drivers are called with interrupts disabled, so
3034 * @console_may_schedule should be cleared before; however, we may
3035 * end up dumping a lot of lines, for example, if called from
3036 * console registration path, and should invoke cond_resched()
3037 * between lines if allowable. Not doing so can cause a very long
3038 * scheduling stall on a slow console leading to RCU stall and
3039 * softlockup warnings which exacerbate the issue with more
3040 * messages practically incapacitating the system. Therefore, create
3041 * a local to use for the printing loop.
3042 */
3043 do_cond_resched = console_may_schedule;
3044
3045 do {
3046 console_may_schedule = 0;
3047
3048 flushed = console_flush_all(do_cond_resched, &next_seq, &handover);
3049 if (!handover)
3050 __console_unlock();
3051
3052 /*
3053 * Abort if there was a failure to flush all messages to all
3054 * usable consoles. Either it is not possible to flush (in
3055 * which case it would be an infinite loop of retrying) or
3056 * another context has taken over printing.
3057 */
3058 if (!flushed)
3059 break;
3060
3061 /*
3062 * Some context may have added new records after
3063 * console_flush_all() but before unlocking the console.
3064 * Re-check if there is a new record to flush. If the trylock
3065 * fails, another context is already handling the printing.
3066 */
3067 } while (prb_read_valid(prb, next_seq, NULL) && console_trylock());
3068}
3069EXPORT_SYMBOL(console_unlock);
3070
3071/**
3072 * console_conditional_schedule - yield the CPU if required
3073 *
3074 * If the console code is currently allowed to sleep, and
3075 * if this CPU should yield the CPU to another task, do
3076 * so here.
3077 *
3078 * Must be called within console_lock();.
3079 */
3080void __sched console_conditional_schedule(void)
3081{
3082 if (console_may_schedule)
3083 cond_resched();
3084}
3085EXPORT_SYMBOL(console_conditional_schedule);
3086
3087void console_unblank(void)
3088{
3089 bool found_unblank = false;
3090 struct console *c;
3091 int cookie;
3092
3093 /*
3094 * First check if there are any consoles implementing the unblank()
3095 * callback. If not, there is no reason to continue and take the
3096 * console lock, which in particular can be dangerous if
3097 * @oops_in_progress is set.
3098 */
3099 cookie = console_srcu_read_lock();
3100 for_each_console_srcu(c) {
3101 if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank) {
3102 found_unblank = true;
3103 break;
3104 }
3105 }
3106 console_srcu_read_unlock(cookie);
3107 if (!found_unblank)
3108 return;
3109
3110 /*
3111 * Stop console printing because the unblank() callback may
3112 * assume the console is not within its write() callback.
3113 *
3114 * If @oops_in_progress is set, this may be an atomic context.
3115 * In that case, attempt a trylock as best-effort.
3116 */
3117 if (oops_in_progress) {
3118 /* Semaphores are not NMI-safe. */
3119 if (in_nmi())
3120 return;
3121
3122 /*
3123 * Attempting to trylock the console lock can deadlock
3124 * if another CPU was stopped while modifying the
3125 * semaphore. "Hope and pray" that this is not the
3126 * current situation.
3127 */
3128 if (down_trylock_console_sem() != 0)
3129 return;
3130 } else
3131 console_lock();
3132
3133 console_locked = 1;
3134 console_may_schedule = 0;
3135
3136 cookie = console_srcu_read_lock();
3137 for_each_console_srcu(c) {
3138 if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank)
3139 c->unblank();
3140 }
3141 console_srcu_read_unlock(cookie);
3142
3143 console_unlock();
3144
3145 if (!oops_in_progress)
3146 pr_flush(1000, true);
3147}
3148
3149/**
3150 * console_flush_on_panic - flush console content on panic
3151 * @mode: flush all messages in buffer or just the pending ones
3152 *
3153 * Immediately output all pending messages no matter what.
3154 */
3155void console_flush_on_panic(enum con_flush_mode mode)
3156{
3157 bool handover;
3158 u64 next_seq;
3159
3160 /*
3161 * Ignore the console lock and flush out the messages. Attempting a
3162 * trylock would not be useful because:
3163 *
3164 * - if it is contended, it must be ignored anyway
3165 * - console_lock() and console_trylock() block and fail
3166 * respectively in panic for non-panic CPUs
3167 * - semaphores are not NMI-safe
3168 */
3169
3170 /*
3171 * If another context is holding the console lock,
3172 * @console_may_schedule might be set. Clear it so that
3173 * this context does not call cond_resched() while flushing.
3174 */
3175 console_may_schedule = 0;
3176
3177 if (mode == CONSOLE_REPLAY_ALL) {
3178 struct console *c;
3179 short flags;
3180 int cookie;
3181 u64 seq;
3182
3183 seq = prb_first_valid_seq(prb);
3184
3185 cookie = console_srcu_read_lock();
3186 for_each_console_srcu(c) {
3187 flags = console_srcu_read_flags(c);
3188
3189 if (flags & CON_NBCON) {
3190 nbcon_seq_force(c, seq);
3191 } else {
3192 /*
3193 * This is an unsynchronized assignment. On
3194 * panic legacy consoles are only best effort.
3195 */
3196 c->seq = seq;
3197 }
3198 }
3199 console_srcu_read_unlock(cookie);
3200 }
3201
3202 console_flush_all(false, &next_seq, &handover);
3203}
3204
3205/*
3206 * Return the console tty driver structure and its associated index
3207 */
3208struct tty_driver *console_device(int *index)
3209{
3210 struct console *c;
3211 struct tty_driver *driver = NULL;
3212 int cookie;
3213
3214 /*
3215 * Take console_lock to serialize device() callback with
3216 * other console operations. For example, fg_console is
3217 * modified under console_lock when switching vt.
3218 */
3219 console_lock();
3220
3221 cookie = console_srcu_read_lock();
3222 for_each_console_srcu(c) {
3223 if (!c->device)
3224 continue;
3225 driver = c->device(c, index);
3226 if (driver)
3227 break;
3228 }
3229 console_srcu_read_unlock(cookie);
3230
3231 console_unlock();
3232 return driver;
3233}
3234
3235/*
3236 * Prevent further output on the passed console device so that (for example)
3237 * serial drivers can disable console output before suspending a port, and can
3238 * re-enable output afterwards.
3239 */
3240void console_stop(struct console *console)
3241{
3242 __pr_flush(console, 1000, true);
3243 console_list_lock();
3244 console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3245 console_list_unlock();
3246
3247 /*
3248 * Ensure that all SRCU list walks have completed. All contexts must
3249 * be able to see that this console is disabled so that (for example)
3250 * the caller can suspend the port without risk of another context
3251 * using the port.
3252 */
3253 synchronize_srcu(&console_srcu);
3254}
3255EXPORT_SYMBOL(console_stop);
3256
3257void console_start(struct console *console)
3258{
3259 console_list_lock();
3260 console_srcu_write_flags(console, console->flags | CON_ENABLED);
3261 console_list_unlock();
3262 __pr_flush(console, 1000, true);
3263}
3264EXPORT_SYMBOL(console_start);
3265
3266static int __read_mostly keep_bootcon;
3267
3268static int __init keep_bootcon_setup(char *str)
3269{
3270 keep_bootcon = 1;
3271 pr_info("debug: skip boot console de-registration.\n");
3272
3273 return 0;
3274}
3275
3276early_param("keep_bootcon", keep_bootcon_setup);
3277
3278static int console_call_setup(struct console *newcon, char *options)
3279{
3280 int err;
3281
3282 if (!newcon->setup)
3283 return 0;
3284
3285 /* Synchronize with possible boot console. */
3286 console_lock();
3287 err = newcon->setup(newcon, options);
3288 console_unlock();
3289
3290 return err;
3291}
3292
3293/*
3294 * This is called by register_console() to try to match
3295 * the newly registered console with any of the ones selected
3296 * by either the command line or add_preferred_console() and
3297 * setup/enable it.
3298 *
3299 * Care need to be taken with consoles that are statically
3300 * enabled such as netconsole
3301 */
3302static int try_enable_preferred_console(struct console *newcon,
3303 bool user_specified)
3304{
3305 struct console_cmdline *c;
3306 int i, err;
3307
3308 for (i = 0, c = console_cmdline;
3309 i < MAX_CMDLINECONSOLES && c->name[0];
3310 i++, c++) {
3311 if (c->user_specified != user_specified)
3312 continue;
3313 if (!newcon->match ||
3314 newcon->match(newcon, c->name, c->index, c->options) != 0) {
3315 /* default matching */
3316 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
3317 if (strcmp(c->name, newcon->name) != 0)
3318 continue;
3319 if (newcon->index >= 0 &&
3320 newcon->index != c->index)
3321 continue;
3322 if (newcon->index < 0)
3323 newcon->index = c->index;
3324
3325 if (_braille_register_console(newcon, c))
3326 return 0;
3327
3328 err = console_call_setup(newcon, c->options);
3329 if (err)
3330 return err;
3331 }
3332 newcon->flags |= CON_ENABLED;
3333 if (i == preferred_console)
3334 newcon->flags |= CON_CONSDEV;
3335 return 0;
3336 }
3337
3338 /*
3339 * Some consoles, such as pstore and netconsole, can be enabled even
3340 * without matching. Accept the pre-enabled consoles only when match()
3341 * and setup() had a chance to be called.
3342 */
3343 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
3344 return 0;
3345
3346 return -ENOENT;
3347}
3348
3349/* Try to enable the console unconditionally */
3350static void try_enable_default_console(struct console *newcon)
3351{
3352 if (newcon->index < 0)
3353 newcon->index = 0;
3354
3355 if (console_call_setup(newcon, NULL) != 0)
3356 return;
3357
3358 newcon->flags |= CON_ENABLED;
3359
3360 if (newcon->device)
3361 newcon->flags |= CON_CONSDEV;
3362}
3363
3364static void console_init_seq(struct console *newcon, bool bootcon_registered)
3365{
3366 struct console *con;
3367 bool handover;
3368
3369 if (newcon->flags & (CON_PRINTBUFFER | CON_BOOT)) {
3370 /* Get a consistent copy of @syslog_seq. */
3371 mutex_lock(&syslog_lock);
3372 newcon->seq = syslog_seq;
3373 mutex_unlock(&syslog_lock);
3374 } else {
3375 /* Begin with next message added to ringbuffer. */
3376 newcon->seq = prb_next_seq(prb);
3377
3378 /*
3379 * If any enabled boot consoles are due to be unregistered
3380 * shortly, some may not be caught up and may be the same
3381 * device as @newcon. Since it is not known which boot console
3382 * is the same device, flush all consoles and, if necessary,
3383 * start with the message of the enabled boot console that is
3384 * the furthest behind.
3385 */
3386 if (bootcon_registered && !keep_bootcon) {
3387 /*
3388 * Hold the console_lock to stop console printing and
3389 * guarantee safe access to console->seq.
3390 */
3391 console_lock();
3392
3393 /*
3394 * Flush all consoles and set the console to start at
3395 * the next unprinted sequence number.
3396 */
3397 if (!console_flush_all(true, &newcon->seq, &handover)) {
3398 /*
3399 * Flushing failed. Just choose the lowest
3400 * sequence of the enabled boot consoles.
3401 */
3402
3403 /*
3404 * If there was a handover, this context no
3405 * longer holds the console_lock.
3406 */
3407 if (handover)
3408 console_lock();
3409
3410 newcon->seq = prb_next_seq(prb);
3411 for_each_console(con) {
3412 if ((con->flags & CON_BOOT) &&
3413 (con->flags & CON_ENABLED) &&
3414 con->seq < newcon->seq) {
3415 newcon->seq = con->seq;
3416 }
3417 }
3418 }
3419
3420 console_unlock();
3421 }
3422 }
3423}
3424
3425#define console_first() \
3426 hlist_entry(console_list.first, struct console, node)
3427
3428static int unregister_console_locked(struct console *console);
3429
3430/*
3431 * The console driver calls this routine during kernel initialization
3432 * to register the console printing procedure with printk() and to
3433 * print any messages that were printed by the kernel before the
3434 * console driver was initialized.
3435 *
3436 * This can happen pretty early during the boot process (because of
3437 * early_printk) - sometimes before setup_arch() completes - be careful
3438 * of what kernel features are used - they may not be initialised yet.
3439 *
3440 * There are two types of consoles - bootconsoles (early_printk) and
3441 * "real" consoles (everything which is not a bootconsole) which are
3442 * handled differently.
3443 * - Any number of bootconsoles can be registered at any time.
3444 * - As soon as a "real" console is registered, all bootconsoles
3445 * will be unregistered automatically.
3446 * - Once a "real" console is registered, any attempt to register a
3447 * bootconsoles will be rejected
3448 */
3449void register_console(struct console *newcon)
3450{
3451 struct console *con;
3452 bool bootcon_registered = false;
3453 bool realcon_registered = false;
3454 int err;
3455
3456 console_list_lock();
3457
3458 for_each_console(con) {
3459 if (WARN(con == newcon, "console '%s%d' already registered\n",
3460 con->name, con->index)) {
3461 goto unlock;
3462 }
3463
3464 if (con->flags & CON_BOOT)
3465 bootcon_registered = true;
3466 else
3467 realcon_registered = true;
3468 }
3469
3470 /* Do not register boot consoles when there already is a real one. */
3471 if ((newcon->flags & CON_BOOT) && realcon_registered) {
3472 pr_info("Too late to register bootconsole %s%d\n",
3473 newcon->name, newcon->index);
3474 goto unlock;
3475 }
3476
3477 if (newcon->flags & CON_NBCON) {
3478 /*
3479 * Ensure the nbcon console buffers can be allocated
3480 * before modifying any global data.
3481 */
3482 if (!nbcon_alloc(newcon))
3483 goto unlock;
3484 }
3485
3486 /*
3487 * See if we want to enable this console driver by default.
3488 *
3489 * Nope when a console is preferred by the command line, device
3490 * tree, or SPCR.
3491 *
3492 * The first real console with tty binding (driver) wins. More
3493 * consoles might get enabled before the right one is found.
3494 *
3495 * Note that a console with tty binding will have CON_CONSDEV
3496 * flag set and will be first in the list.
3497 */
3498 if (preferred_console < 0) {
3499 if (hlist_empty(&console_list) || !console_first()->device ||
3500 console_first()->flags & CON_BOOT) {
3501 try_enable_default_console(newcon);
3502 }
3503 }
3504
3505 /* See if this console matches one we selected on the command line */
3506 err = try_enable_preferred_console(newcon, true);
3507
3508 /* If not, try to match against the platform default(s) */
3509 if (err == -ENOENT)
3510 err = try_enable_preferred_console(newcon, false);
3511
3512 /* printk() messages are not printed to the Braille console. */
3513 if (err || newcon->flags & CON_BRL) {
3514 if (newcon->flags & CON_NBCON)
3515 nbcon_free(newcon);
3516 goto unlock;
3517 }
3518
3519 /*
3520 * If we have a bootconsole, and are switching to a real console,
3521 * don't print everything out again, since when the boot console, and
3522 * the real console are the same physical device, it's annoying to
3523 * see the beginning boot messages twice
3524 */
3525 if (bootcon_registered &&
3526 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
3527 newcon->flags &= ~CON_PRINTBUFFER;
3528 }
3529
3530 newcon->dropped = 0;
3531 console_init_seq(newcon, bootcon_registered);
3532
3533 if (newcon->flags & CON_NBCON)
3534 nbcon_init(newcon);
3535
3536 /*
3537 * Put this console in the list - keep the
3538 * preferred driver at the head of the list.
3539 */
3540 if (hlist_empty(&console_list)) {
3541 /* Ensure CON_CONSDEV is always set for the head. */
3542 newcon->flags |= CON_CONSDEV;
3543 hlist_add_head_rcu(&newcon->node, &console_list);
3544
3545 } else if (newcon->flags & CON_CONSDEV) {
3546 /* Only the new head can have CON_CONSDEV set. */
3547 console_srcu_write_flags(console_first(), console_first()->flags & ~CON_CONSDEV);
3548 hlist_add_head_rcu(&newcon->node, &console_list);
3549
3550 } else {
3551 hlist_add_behind_rcu(&newcon->node, console_list.first);
3552 }
3553
3554 /*
3555 * No need to synchronize SRCU here! The caller does not rely
3556 * on all contexts being able to see the new console before
3557 * register_console() completes.
3558 */
3559
3560 console_sysfs_notify();
3561
3562 /*
3563 * By unregistering the bootconsoles after we enable the real console
3564 * we get the "console xxx enabled" message on all the consoles -
3565 * boot consoles, real consoles, etc - this is to ensure that end
3566 * users know there might be something in the kernel's log buffer that
3567 * went to the bootconsole (that they do not see on the real console)
3568 */
3569 con_printk(KERN_INFO, newcon, "enabled\n");
3570 if (bootcon_registered &&
3571 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
3572 !keep_bootcon) {
3573 struct hlist_node *tmp;
3574
3575 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3576 if (con->flags & CON_BOOT)
3577 unregister_console_locked(con);
3578 }
3579 }
3580unlock:
3581 console_list_unlock();
3582}
3583EXPORT_SYMBOL(register_console);
3584
3585/* Must be called under console_list_lock(). */
3586static int unregister_console_locked(struct console *console)
3587{
3588 int res;
3589
3590 lockdep_assert_console_list_lock_held();
3591
3592 con_printk(KERN_INFO, console, "disabled\n");
3593
3594 res = _braille_unregister_console(console);
3595 if (res < 0)
3596 return res;
3597 if (res > 0)
3598 return 0;
3599
3600 /* Disable it unconditionally */
3601 console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3602
3603 if (!console_is_registered_locked(console))
3604 return -ENODEV;
3605
3606 hlist_del_init_rcu(&console->node);
3607
3608 /*
3609 * <HISTORICAL>
3610 * If this isn't the last console and it has CON_CONSDEV set, we
3611 * need to set it on the next preferred console.
3612 * </HISTORICAL>
3613 *
3614 * The above makes no sense as there is no guarantee that the next
3615 * console has any device attached. Oh well....
3616 */
3617 if (!hlist_empty(&console_list) && console->flags & CON_CONSDEV)
3618 console_srcu_write_flags(console_first(), console_first()->flags | CON_CONSDEV);
3619
3620 /*
3621 * Ensure that all SRCU list walks have completed. All contexts
3622 * must not be able to see this console in the list so that any
3623 * exit/cleanup routines can be performed safely.
3624 */
3625 synchronize_srcu(&console_srcu);
3626
3627 if (console->flags & CON_NBCON)
3628 nbcon_free(console);
3629
3630 console_sysfs_notify();
3631
3632 if (console->exit)
3633 res = console->exit(console);
3634
3635 return res;
3636}
3637
3638int unregister_console(struct console *console)
3639{
3640 int res;
3641
3642 console_list_lock();
3643 res = unregister_console_locked(console);
3644 console_list_unlock();
3645 return res;
3646}
3647EXPORT_SYMBOL(unregister_console);
3648
3649/**
3650 * console_force_preferred_locked - force a registered console preferred
3651 * @con: The registered console to force preferred.
3652 *
3653 * Must be called under console_list_lock().
3654 */
3655void console_force_preferred_locked(struct console *con)
3656{
3657 struct console *cur_pref_con;
3658
3659 if (!console_is_registered_locked(con))
3660 return;
3661
3662 cur_pref_con = console_first();
3663
3664 /* Already preferred? */
3665 if (cur_pref_con == con)
3666 return;
3667
3668 /*
3669 * Delete, but do not re-initialize the entry. This allows the console
3670 * to continue to appear registered (via any hlist_unhashed_lockless()
3671 * checks), even though it was briefly removed from the console list.
3672 */
3673 hlist_del_rcu(&con->node);
3674
3675 /*
3676 * Ensure that all SRCU list walks have completed so that the console
3677 * can be added to the beginning of the console list and its forward
3678 * list pointer can be re-initialized.
3679 */
3680 synchronize_srcu(&console_srcu);
3681
3682 con->flags |= CON_CONSDEV;
3683 WARN_ON(!con->device);
3684
3685 /* Only the new head can have CON_CONSDEV set. */
3686 console_srcu_write_flags(cur_pref_con, cur_pref_con->flags & ~CON_CONSDEV);
3687 hlist_add_head_rcu(&con->node, &console_list);
3688}
3689EXPORT_SYMBOL(console_force_preferred_locked);
3690
3691/*
3692 * Initialize the console device. This is called *early*, so
3693 * we can't necessarily depend on lots of kernel help here.
3694 * Just do some early initializations, and do the complex setup
3695 * later.
3696 */
3697void __init console_init(void)
3698{
3699 int ret;
3700 initcall_t call;
3701 initcall_entry_t *ce;
3702
3703 /* Setup the default TTY line discipline. */
3704 n_tty_init();
3705
3706 /*
3707 * set up the console device so that later boot sequences can
3708 * inform about problems etc..
3709 */
3710 ce = __con_initcall_start;
3711 trace_initcall_level("console");
3712 while (ce < __con_initcall_end) {
3713 call = initcall_from_entry(ce);
3714 trace_initcall_start(call);
3715 ret = call();
3716 trace_initcall_finish(call, ret);
3717 ce++;
3718 }
3719}
3720
3721/*
3722 * Some boot consoles access data that is in the init section and which will
3723 * be discarded after the initcalls have been run. To make sure that no code
3724 * will access this data, unregister the boot consoles in a late initcall.
3725 *
3726 * If for some reason, such as deferred probe or the driver being a loadable
3727 * module, the real console hasn't registered yet at this point, there will
3728 * be a brief interval in which no messages are logged to the console, which
3729 * makes it difficult to diagnose problems that occur during this time.
3730 *
3731 * To mitigate this problem somewhat, only unregister consoles whose memory
3732 * intersects with the init section. Note that all other boot consoles will
3733 * get unregistered when the real preferred console is registered.
3734 */
3735static int __init printk_late_init(void)
3736{
3737 struct hlist_node *tmp;
3738 struct console *con;
3739 int ret;
3740
3741 console_list_lock();
3742 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3743 if (!(con->flags & CON_BOOT))
3744 continue;
3745
3746 /* Check addresses that might be used for enabled consoles. */
3747 if (init_section_intersects(con, sizeof(*con)) ||
3748 init_section_contains(con->write, 0) ||
3749 init_section_contains(con->read, 0) ||
3750 init_section_contains(con->device, 0) ||
3751 init_section_contains(con->unblank, 0) ||
3752 init_section_contains(con->data, 0)) {
3753 /*
3754 * Please, consider moving the reported consoles out
3755 * of the init section.
3756 */
3757 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3758 con->name, con->index);
3759 unregister_console_locked(con);
3760 }
3761 }
3762 console_list_unlock();
3763
3764 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3765 console_cpu_notify);
3766 WARN_ON(ret < 0);
3767 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3768 console_cpu_notify, NULL);
3769 WARN_ON(ret < 0);
3770 printk_sysctl_init();
3771 return 0;
3772}
3773late_initcall(printk_late_init);
3774
3775#if defined CONFIG_PRINTK
3776/* If @con is specified, only wait for that console. Otherwise wait for all. */
3777static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
3778{
3779 unsigned long timeout_jiffies = msecs_to_jiffies(timeout_ms);
3780 unsigned long remaining_jiffies = timeout_jiffies;
3781 struct console *c;
3782 u64 last_diff = 0;
3783 u64 printk_seq;
3784 short flags;
3785 int cookie;
3786 u64 diff;
3787 u64 seq;
3788
3789 might_sleep();
3790
3791 seq = prb_next_reserve_seq(prb);
3792
3793 /* Flush the consoles so that records up to @seq are printed. */
3794 console_lock();
3795 console_unlock();
3796
3797 for (;;) {
3798 unsigned long begin_jiffies;
3799 unsigned long slept_jiffies;
3800
3801 diff = 0;
3802
3803 /*
3804 * Hold the console_lock to guarantee safe access to
3805 * console->seq. Releasing console_lock flushes more
3806 * records in case @seq is still not printed on all
3807 * usable consoles.
3808 */
3809 console_lock();
3810
3811 cookie = console_srcu_read_lock();
3812 for_each_console_srcu(c) {
3813 if (con && con != c)
3814 continue;
3815
3816 flags = console_srcu_read_flags(c);
3817
3818 /*
3819 * If consoles are not usable, it cannot be expected
3820 * that they make forward progress, so only increment
3821 * @diff for usable consoles.
3822 */
3823 if (!console_is_usable(c))
3824 continue;
3825
3826 if (flags & CON_NBCON) {
3827 printk_seq = nbcon_seq_read(c);
3828 } else {
3829 printk_seq = c->seq;
3830 }
3831
3832 if (printk_seq < seq)
3833 diff += seq - printk_seq;
3834 }
3835 console_srcu_read_unlock(cookie);
3836
3837 if (diff != last_diff && reset_on_progress)
3838 remaining_jiffies = timeout_jiffies;
3839
3840 console_unlock();
3841
3842 /* Note: @diff is 0 if there are no usable consoles. */
3843 if (diff == 0 || remaining_jiffies == 0)
3844 break;
3845
3846 /* msleep(1) might sleep much longer. Check time by jiffies. */
3847 begin_jiffies = jiffies;
3848 msleep(1);
3849 slept_jiffies = jiffies - begin_jiffies;
3850
3851 remaining_jiffies -= min(slept_jiffies, remaining_jiffies);
3852
3853 last_diff = diff;
3854 }
3855
3856 return (diff == 0);
3857}
3858
3859/**
3860 * pr_flush() - Wait for printing threads to catch up.
3861 *
3862 * @timeout_ms: The maximum time (in ms) to wait.
3863 * @reset_on_progress: Reset the timeout if forward progress is seen.
3864 *
3865 * A value of 0 for @timeout_ms means no waiting will occur. A value of -1
3866 * represents infinite waiting.
3867 *
3868 * If @reset_on_progress is true, the timeout will be reset whenever any
3869 * printer has been seen to make some forward progress.
3870 *
3871 * Context: Process context. May sleep while acquiring console lock.
3872 * Return: true if all usable printers are caught up.
3873 */
3874static bool pr_flush(int timeout_ms, bool reset_on_progress)
3875{
3876 return __pr_flush(NULL, timeout_ms, reset_on_progress);
3877}
3878
3879/*
3880 * Delayed printk version, for scheduler-internal messages:
3881 */
3882#define PRINTK_PENDING_WAKEUP 0x01
3883#define PRINTK_PENDING_OUTPUT 0x02
3884
3885static DEFINE_PER_CPU(int, printk_pending);
3886
3887static void wake_up_klogd_work_func(struct irq_work *irq_work)
3888{
3889 int pending = this_cpu_xchg(printk_pending, 0);
3890
3891 if (pending & PRINTK_PENDING_OUTPUT) {
3892 /* If trylock fails, someone else is doing the printing */
3893 if (console_trylock())
3894 console_unlock();
3895 }
3896
3897 if (pending & PRINTK_PENDING_WAKEUP)
3898 wake_up_interruptible(&log_wait);
3899}
3900
3901static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3902 IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3903
3904static void __wake_up_klogd(int val)
3905{
3906 if (!printk_percpu_data_ready())
3907 return;
3908
3909 preempt_disable();
3910 /*
3911 * Guarantee any new records can be seen by tasks preparing to wait
3912 * before this context checks if the wait queue is empty.
3913 *
3914 * The full memory barrier within wq_has_sleeper() pairs with the full
3915 * memory barrier within set_current_state() of
3916 * prepare_to_wait_event(), which is called after ___wait_event() adds
3917 * the waiter but before it has checked the wait condition.
3918 *
3919 * This pairs with devkmsg_read:A and syslog_print:A.
3920 */
3921 if (wq_has_sleeper(&log_wait) || /* LMM(__wake_up_klogd:A) */
3922 (val & PRINTK_PENDING_OUTPUT)) {
3923 this_cpu_or(printk_pending, val);
3924 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3925 }
3926 preempt_enable();
3927}
3928
3929/**
3930 * wake_up_klogd - Wake kernel logging daemon
3931 *
3932 * Use this function when new records have been added to the ringbuffer
3933 * and the console printing of those records has already occurred or is
3934 * known to be handled by some other context. This function will only
3935 * wake the logging daemon.
3936 *
3937 * Context: Any context.
3938 */
3939void wake_up_klogd(void)
3940{
3941 __wake_up_klogd(PRINTK_PENDING_WAKEUP);
3942}
3943
3944/**
3945 * defer_console_output - Wake kernel logging daemon and trigger
3946 * console printing in a deferred context
3947 *
3948 * Use this function when new records have been added to the ringbuffer,
3949 * this context is responsible for console printing those records, but
3950 * the current context is not allowed to perform the console printing.
3951 * Trigger an irq_work context to perform the console printing. This
3952 * function also wakes the logging daemon.
3953 *
3954 * Context: Any context.
3955 */
3956void defer_console_output(void)
3957{
3958 /*
3959 * New messages may have been added directly to the ringbuffer
3960 * using vprintk_store(), so wake any waiters as well.
3961 */
3962 __wake_up_klogd(PRINTK_PENDING_WAKEUP | PRINTK_PENDING_OUTPUT);
3963}
3964
3965void printk_trigger_flush(void)
3966{
3967 defer_console_output();
3968}
3969
3970int vprintk_deferred(const char *fmt, va_list args)
3971{
3972 return vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3973}
3974
3975int _printk_deferred(const char *fmt, ...)
3976{
3977 va_list args;
3978 int r;
3979
3980 va_start(args, fmt);
3981 r = vprintk_deferred(fmt, args);
3982 va_end(args);
3983
3984 return r;
3985}
3986
3987/*
3988 * printk rate limiting, lifted from the networking subsystem.
3989 *
3990 * This enforces a rate limit: not more than 10 kernel messages
3991 * every 5s to make a denial-of-service attack impossible.
3992 */
3993DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3994
3995int __printk_ratelimit(const char *func)
3996{
3997 return ___ratelimit(&printk_ratelimit_state, func);
3998}
3999EXPORT_SYMBOL(__printk_ratelimit);
4000
4001/**
4002 * printk_timed_ratelimit - caller-controlled printk ratelimiting
4003 * @caller_jiffies: pointer to caller's state
4004 * @interval_msecs: minimum interval between prints
4005 *
4006 * printk_timed_ratelimit() returns true if more than @interval_msecs
4007 * milliseconds have elapsed since the last time printk_timed_ratelimit()
4008 * returned true.
4009 */
4010bool printk_timed_ratelimit(unsigned long *caller_jiffies,
4011 unsigned int interval_msecs)
4012{
4013 unsigned long elapsed = jiffies - *caller_jiffies;
4014
4015 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
4016 return false;
4017
4018 *caller_jiffies = jiffies;
4019 return true;
4020}
4021EXPORT_SYMBOL(printk_timed_ratelimit);
4022
4023static DEFINE_SPINLOCK(dump_list_lock);
4024static LIST_HEAD(dump_list);
4025
4026/**
4027 * kmsg_dump_register - register a kernel log dumper.
4028 * @dumper: pointer to the kmsg_dumper structure
4029 *
4030 * Adds a kernel log dumper to the system. The dump callback in the
4031 * structure will be called when the kernel oopses or panics and must be
4032 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
4033 */
4034int kmsg_dump_register(struct kmsg_dumper *dumper)
4035{
4036 unsigned long flags;
4037 int err = -EBUSY;
4038
4039 /* The dump callback needs to be set */
4040 if (!dumper->dump)
4041 return -EINVAL;
4042
4043 spin_lock_irqsave(&dump_list_lock, flags);
4044 /* Don't allow registering multiple times */
4045 if (!dumper->registered) {
4046 dumper->registered = 1;
4047 list_add_tail_rcu(&dumper->list, &dump_list);
4048 err = 0;
4049 }
4050 spin_unlock_irqrestore(&dump_list_lock, flags);
4051
4052 return err;
4053}
4054EXPORT_SYMBOL_GPL(kmsg_dump_register);
4055
4056/**
4057 * kmsg_dump_unregister - unregister a kmsg dumper.
4058 * @dumper: pointer to the kmsg_dumper structure
4059 *
4060 * Removes a dump device from the system. Returns zero on success and
4061 * %-EINVAL otherwise.
4062 */
4063int kmsg_dump_unregister(struct kmsg_dumper *dumper)
4064{
4065 unsigned long flags;
4066 int err = -EINVAL;
4067
4068 spin_lock_irqsave(&dump_list_lock, flags);
4069 if (dumper->registered) {
4070 dumper->registered = 0;
4071 list_del_rcu(&dumper->list);
4072 err = 0;
4073 }
4074 spin_unlock_irqrestore(&dump_list_lock, flags);
4075 synchronize_rcu();
4076
4077 return err;
4078}
4079EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
4080
4081static bool always_kmsg_dump;
4082module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
4083
4084const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
4085{
4086 switch (reason) {
4087 case KMSG_DUMP_PANIC:
4088 return "Panic";
4089 case KMSG_DUMP_OOPS:
4090 return "Oops";
4091 case KMSG_DUMP_EMERG:
4092 return "Emergency";
4093 case KMSG_DUMP_SHUTDOWN:
4094 return "Shutdown";
4095 default:
4096 return "Unknown";
4097 }
4098}
4099EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
4100
4101/**
4102 * kmsg_dump - dump kernel log to kernel message dumpers.
4103 * @reason: the reason (oops, panic etc) for dumping
4104 *
4105 * Call each of the registered dumper's dump() callback, which can
4106 * retrieve the kmsg records with kmsg_dump_get_line() or
4107 * kmsg_dump_get_buffer().
4108 */
4109void kmsg_dump(enum kmsg_dump_reason reason)
4110{
4111 struct kmsg_dumper *dumper;
4112
4113 rcu_read_lock();
4114 list_for_each_entry_rcu(dumper, &dump_list, list) {
4115 enum kmsg_dump_reason max_reason = dumper->max_reason;
4116
4117 /*
4118 * If client has not provided a specific max_reason, default
4119 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
4120 */
4121 if (max_reason == KMSG_DUMP_UNDEF) {
4122 max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
4123 KMSG_DUMP_OOPS;
4124 }
4125 if (reason > max_reason)
4126 continue;
4127
4128 /* invoke dumper which will iterate over records */
4129 dumper->dump(dumper, reason);
4130 }
4131 rcu_read_unlock();
4132}
4133
4134/**
4135 * kmsg_dump_get_line - retrieve one kmsg log line
4136 * @iter: kmsg dump iterator
4137 * @syslog: include the "<4>" prefixes
4138 * @line: buffer to copy the line to
4139 * @size: maximum size of the buffer
4140 * @len: length of line placed into buffer
4141 *
4142 * Start at the beginning of the kmsg buffer, with the oldest kmsg
4143 * record, and copy one record into the provided buffer.
4144 *
4145 * Consecutive calls will return the next available record moving
4146 * towards the end of the buffer with the youngest messages.
4147 *
4148 * A return value of FALSE indicates that there are no more records to
4149 * read.
4150 */
4151bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
4152 char *line, size_t size, size_t *len)
4153{
4154 u64 min_seq = latched_seq_read_nolock(&clear_seq);
4155 struct printk_info info;
4156 unsigned int line_count;
4157 struct printk_record r;
4158 size_t l = 0;
4159 bool ret = false;
4160
4161 if (iter->cur_seq < min_seq)
4162 iter->cur_seq = min_seq;
4163
4164 prb_rec_init_rd(&r, &info, line, size);
4165
4166 /* Read text or count text lines? */
4167 if (line) {
4168 if (!prb_read_valid(prb, iter->cur_seq, &r))
4169 goto out;
4170 l = record_print_text(&r, syslog, printk_time);
4171 } else {
4172 if (!prb_read_valid_info(prb, iter->cur_seq,
4173 &info, &line_count)) {
4174 goto out;
4175 }
4176 l = get_record_print_text_size(&info, line_count, syslog,
4177 printk_time);
4178
4179 }
4180
4181 iter->cur_seq = r.info->seq + 1;
4182 ret = true;
4183out:
4184 if (len)
4185 *len = l;
4186 return ret;
4187}
4188EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
4189
4190/**
4191 * kmsg_dump_get_buffer - copy kmsg log lines
4192 * @iter: kmsg dump iterator
4193 * @syslog: include the "<4>" prefixes
4194 * @buf: buffer to copy the line to
4195 * @size: maximum size of the buffer
4196 * @len_out: length of line placed into buffer
4197 *
4198 * Start at the end of the kmsg buffer and fill the provided buffer
4199 * with as many of the *youngest* kmsg records that fit into it.
4200 * If the buffer is large enough, all available kmsg records will be
4201 * copied with a single call.
4202 *
4203 * Consecutive calls will fill the buffer with the next block of
4204 * available older records, not including the earlier retrieved ones.
4205 *
4206 * A return value of FALSE indicates that there are no more records to
4207 * read.
4208 */
4209bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
4210 char *buf, size_t size, size_t *len_out)
4211{
4212 u64 min_seq = latched_seq_read_nolock(&clear_seq);
4213 struct printk_info info;
4214 struct printk_record r;
4215 u64 seq;
4216 u64 next_seq;
4217 size_t len = 0;
4218 bool ret = false;
4219 bool time = printk_time;
4220
4221 if (!buf || !size)
4222 goto out;
4223
4224 if (iter->cur_seq < min_seq)
4225 iter->cur_seq = min_seq;
4226
4227 if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
4228 if (info.seq != iter->cur_seq) {
4229 /* messages are gone, move to first available one */
4230 iter->cur_seq = info.seq;
4231 }
4232 }
4233
4234 /* last entry */
4235 if (iter->cur_seq >= iter->next_seq)
4236 goto out;
4237
4238 /*
4239 * Find first record that fits, including all following records,
4240 * into the user-provided buffer for this dump. Pass in size-1
4241 * because this function (by way of record_print_text()) will
4242 * not write more than size-1 bytes of text into @buf.
4243 */
4244 seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
4245 size - 1, syslog, time);
4246
4247 /*
4248 * Next kmsg_dump_get_buffer() invocation will dump block of
4249 * older records stored right before this one.
4250 */
4251 next_seq = seq;
4252
4253 prb_rec_init_rd(&r, &info, buf, size);
4254
4255 prb_for_each_record(seq, prb, seq, &r) {
4256 if (r.info->seq >= iter->next_seq)
4257 break;
4258
4259 len += record_print_text(&r, syslog, time);
4260
4261 /* Adjust record to store to remaining buffer space. */
4262 prb_rec_init_rd(&r, &info, buf + len, size - len);
4263 }
4264
4265 iter->next_seq = next_seq;
4266 ret = true;
4267out:
4268 if (len_out)
4269 *len_out = len;
4270 return ret;
4271}
4272EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
4273
4274/**
4275 * kmsg_dump_rewind - reset the iterator
4276 * @iter: kmsg dump iterator
4277 *
4278 * Reset the dumper's iterator so that kmsg_dump_get_line() and
4279 * kmsg_dump_get_buffer() can be called again and used multiple
4280 * times within the same dumper.dump() callback.
4281 */
4282void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
4283{
4284 iter->cur_seq = latched_seq_read_nolock(&clear_seq);
4285 iter->next_seq = prb_next_seq(prb);
4286}
4287EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
4288
4289#endif
4290
4291#ifdef CONFIG_SMP
4292static atomic_t printk_cpu_sync_owner = ATOMIC_INIT(-1);
4293static atomic_t printk_cpu_sync_nested = ATOMIC_INIT(0);
4294
4295/**
4296 * __printk_cpu_sync_wait() - Busy wait until the printk cpu-reentrant
4297 * spinning lock is not owned by any CPU.
4298 *
4299 * Context: Any context.
4300 */
4301void __printk_cpu_sync_wait(void)
4302{
4303 do {
4304 cpu_relax();
4305 } while (atomic_read(&printk_cpu_sync_owner) != -1);
4306}
4307EXPORT_SYMBOL(__printk_cpu_sync_wait);
4308
4309/**
4310 * __printk_cpu_sync_try_get() - Try to acquire the printk cpu-reentrant
4311 * spinning lock.
4312 *
4313 * If no processor has the lock, the calling processor takes the lock and
4314 * becomes the owner. If the calling processor is already the owner of the
4315 * lock, this function succeeds immediately.
4316 *
4317 * Context: Any context. Expects interrupts to be disabled.
4318 * Return: 1 on success, otherwise 0.
4319 */
4320int __printk_cpu_sync_try_get(void)
4321{
4322 int cpu;
4323 int old;
4324
4325 cpu = smp_processor_id();
4326
4327 /*
4328 * Guarantee loads and stores from this CPU when it is the lock owner
4329 * are _not_ visible to the previous lock owner. This pairs with
4330 * __printk_cpu_sync_put:B.
4331 *
4332 * Memory barrier involvement:
4333 *
4334 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4335 * then __printk_cpu_sync_put:A can never read from
4336 * __printk_cpu_sync_try_get:B.
4337 *
4338 * Relies on:
4339 *
4340 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4341 * of the previous CPU
4342 * matching
4343 * ACQUIRE from __printk_cpu_sync_try_get:A to
4344 * __printk_cpu_sync_try_get:B of this CPU
4345 */
4346 old = atomic_cmpxchg_acquire(&printk_cpu_sync_owner, -1,
4347 cpu); /* LMM(__printk_cpu_sync_try_get:A) */
4348 if (old == -1) {
4349 /*
4350 * This CPU is now the owner and begins loading/storing
4351 * data: LMM(__printk_cpu_sync_try_get:B)
4352 */
4353 return 1;
4354
4355 } else if (old == cpu) {
4356 /* This CPU is already the owner. */
4357 atomic_inc(&printk_cpu_sync_nested);
4358 return 1;
4359 }
4360
4361 return 0;
4362}
4363EXPORT_SYMBOL(__printk_cpu_sync_try_get);
4364
4365/**
4366 * __printk_cpu_sync_put() - Release the printk cpu-reentrant spinning lock.
4367 *
4368 * The calling processor must be the owner of the lock.
4369 *
4370 * Context: Any context. Expects interrupts to be disabled.
4371 */
4372void __printk_cpu_sync_put(void)
4373{
4374 if (atomic_read(&printk_cpu_sync_nested)) {
4375 atomic_dec(&printk_cpu_sync_nested);
4376 return;
4377 }
4378
4379 /*
4380 * This CPU is finished loading/storing data:
4381 * LMM(__printk_cpu_sync_put:A)
4382 */
4383
4384 /*
4385 * Guarantee loads and stores from this CPU when it was the
4386 * lock owner are visible to the next lock owner. This pairs
4387 * with __printk_cpu_sync_try_get:A.
4388 *
4389 * Memory barrier involvement:
4390 *
4391 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4392 * then __printk_cpu_sync_try_get:B reads from __printk_cpu_sync_put:A.
4393 *
4394 * Relies on:
4395 *
4396 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4397 * of this CPU
4398 * matching
4399 * ACQUIRE from __printk_cpu_sync_try_get:A to
4400 * __printk_cpu_sync_try_get:B of the next CPU
4401 */
4402 atomic_set_release(&printk_cpu_sync_owner,
4403 -1); /* LMM(__printk_cpu_sync_put:B) */
4404}
4405EXPORT_SYMBOL(__printk_cpu_sync_put);
4406#endif /* CONFIG_SMP */