Loading...
Note: File does not exist in v3.1.
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/printk.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * Modified to make sys_syslog() more flexible: added commands to
8 * return the last 4k of kernel messages, regardless of whether
9 * they've been read or not. Added option to suppress kernel printk's
10 * to the console. Added hook for sending the console messages
11 * elsewhere, in preparation for a serial line console (someday).
12 * Ted Ts'o, 2/11/93.
13 * Modified for sysctl support, 1/8/97, Chris Horn.
14 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
15 * manfred@colorfullife.com
16 * Rewrote bits to get rid of console_lock
17 * 01Mar01 Andrew Morton
18 */
19
20#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21
22#include <linux/kernel.h>
23#include <linux/mm.h>
24#include <linux/tty.h>
25#include <linux/tty_driver.h>
26#include <linux/console.h>
27#include <linux/init.h>
28#include <linux/jiffies.h>
29#include <linux/nmi.h>
30#include <linux/module.h>
31#include <linux/moduleparam.h>
32#include <linux/delay.h>
33#include <linux/smp.h>
34#include <linux/security.h>
35#include <linux/memblock.h>
36#include <linux/syscalls.h>
37#include <linux/crash_core.h>
38#include <linux/ratelimit.h>
39#include <linux/kmsg_dump.h>
40#include <linux/syslog.h>
41#include <linux/cpu.h>
42#include <linux/rculist.h>
43#include <linux/poll.h>
44#include <linux/irq_work.h>
45#include <linux/ctype.h>
46#include <linux/uio.h>
47#include <linux/sched/clock.h>
48#include <linux/sched/debug.h>
49#include <linux/sched/task_stack.h>
50
51#include <linux/uaccess.h>
52#include <asm/sections.h>
53
54#include <trace/events/initcall.h>
55#define CREATE_TRACE_POINTS
56#include <trace/events/printk.h>
57
58#include "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/*
351 * This is used for debugging the mess that is the VT code by
352 * keeping track if we have the console semaphore held. It's
353 * definitely not the perfect debug tool (we don't know if _WE_
354 * hold it and are racing, but it helps tracking those weird code
355 * paths in the console code where we end up in places I want
356 * locked without the console semaphore held).
357 */
358static int console_locked;
359
360/*
361 * Array of consoles built from command line options (console=)
362 */
363
364#define MAX_CMDLINECONSOLES 8
365
366static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
367
368static int preferred_console = -1;
369int console_set_on_cmdline;
370EXPORT_SYMBOL(console_set_on_cmdline);
371
372/* Flag: console code may call schedule() */
373static int console_may_schedule;
374
375enum con_msg_format_flags {
376 MSG_FORMAT_DEFAULT = 0,
377 MSG_FORMAT_SYSLOG = (1 << 0),
378};
379
380static int console_msg_format = MSG_FORMAT_DEFAULT;
381
382/*
383 * The printk log buffer consists of a sequenced collection of records, each
384 * containing variable length message text. Every record also contains its
385 * own meta-data (@info).
386 *
387 * Every record meta-data carries the timestamp in microseconds, as well as
388 * the standard userspace syslog level and syslog facility. The usual kernel
389 * messages use LOG_KERN; userspace-injected messages always carry a matching
390 * syslog facility, by default LOG_USER. The origin of every message can be
391 * reliably determined that way.
392 *
393 * The human readable log message of a record is available in @text, the
394 * length of the message text in @text_len. The stored message is not
395 * terminated.
396 *
397 * Optionally, a record can carry a dictionary of properties (key/value
398 * pairs), to provide userspace with a machine-readable message context.
399 *
400 * Examples for well-defined, commonly used property names are:
401 * DEVICE=b12:8 device identifier
402 * b12:8 block dev_t
403 * c127:3 char dev_t
404 * n8 netdev ifindex
405 * +sound:card0 subsystem:devname
406 * SUBSYSTEM=pci driver-core subsystem name
407 *
408 * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
409 * and values are terminated by a '\0' character.
410 *
411 * Example of record values:
412 * record.text_buf = "it's a line" (unterminated)
413 * record.info.seq = 56
414 * record.info.ts_nsec = 36863
415 * record.info.text_len = 11
416 * record.info.facility = 0 (LOG_KERN)
417 * record.info.flags = 0
418 * record.info.level = 3 (LOG_ERR)
419 * record.info.caller_id = 299 (task 299)
420 * record.info.dev_info.subsystem = "pci" (terminated)
421 * record.info.dev_info.device = "+pci:0000:00:01.0" (terminated)
422 *
423 * The 'struct printk_info' buffer must never be directly exported to
424 * userspace, it is a kernel-private implementation detail that might
425 * need to be changed in the future, when the requirements change.
426 *
427 * /dev/kmsg exports the structured data in the following line format:
428 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
429 *
430 * Users of the export format should ignore possible additional values
431 * separated by ',', and find the message after the ';' character.
432 *
433 * The optional key/value pairs are attached as continuation lines starting
434 * with a space character and terminated by a newline. All possible
435 * non-prinatable characters are escaped in the "\xff" notation.
436 */
437
438/* syslog_lock protects syslog_* variables and write access to clear_seq. */
439static DEFINE_MUTEX(syslog_lock);
440
441#ifdef CONFIG_PRINTK
442/*
443 * During panic, heavy printk by other CPUs can delay the
444 * panic and risk deadlock on console resources.
445 */
446static int __read_mostly suppress_panic_printk;
447
448DECLARE_WAIT_QUEUE_HEAD(log_wait);
449/* All 3 protected by @syslog_lock. */
450/* the next printk record to read by syslog(READ) or /proc/kmsg */
451static u64 syslog_seq;
452static size_t syslog_partial;
453static bool syslog_time;
454
455struct latched_seq {
456 seqcount_latch_t latch;
457 u64 val[2];
458};
459
460/*
461 * The next printk record to read after the last 'clear' command. There are
462 * two copies (updated with seqcount_latch) so that reads can locklessly
463 * access a valid value. Writers are synchronized by @syslog_lock.
464 */
465static struct latched_seq clear_seq = {
466 .latch = SEQCNT_LATCH_ZERO(clear_seq.latch),
467 .val[0] = 0,
468 .val[1] = 0,
469};
470
471#define LOG_LEVEL(v) ((v) & 0x07)
472#define LOG_FACILITY(v) ((v) >> 3 & 0xff)
473
474/* record buffer */
475#define LOG_ALIGN __alignof__(unsigned long)
476#define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
477#define LOG_BUF_LEN_MAX (u32)(1 << 31)
478static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
479static char *log_buf = __log_buf;
480static u32 log_buf_len = __LOG_BUF_LEN;
481
482/*
483 * Define the average message size. This only affects the number of
484 * descriptors that will be available. Underestimating is better than
485 * overestimating (too many available descriptors is better than not enough).
486 */
487#define PRB_AVGBITS 5 /* 32 character average length */
488
489#if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
490#error CONFIG_LOG_BUF_SHIFT value too small.
491#endif
492_DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
493 PRB_AVGBITS, &__log_buf[0]);
494
495static struct printk_ringbuffer printk_rb_dynamic;
496
497struct printk_ringbuffer *prb = &printk_rb_static;
498
499/*
500 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
501 * per_cpu_areas are initialised. This variable is set to true when
502 * it's safe to access per-CPU data.
503 */
504static bool __printk_percpu_data_ready __ro_after_init;
505
506bool printk_percpu_data_ready(void)
507{
508 return __printk_percpu_data_ready;
509}
510
511/* Must be called under syslog_lock. */
512static void latched_seq_write(struct latched_seq *ls, u64 val)
513{
514 raw_write_seqcount_latch(&ls->latch);
515 ls->val[0] = val;
516 raw_write_seqcount_latch(&ls->latch);
517 ls->val[1] = val;
518}
519
520/* Can be called from any context. */
521static u64 latched_seq_read_nolock(struct latched_seq *ls)
522{
523 unsigned int seq;
524 unsigned int idx;
525 u64 val;
526
527 do {
528 seq = raw_read_seqcount_latch(&ls->latch);
529 idx = seq & 0x1;
530 val = ls->val[idx];
531 } while (raw_read_seqcount_latch_retry(&ls->latch, seq));
532
533 return val;
534}
535
536/* Return log buffer address */
537char *log_buf_addr_get(void)
538{
539 return log_buf;
540}
541
542/* Return log buffer size */
543u32 log_buf_len_get(void)
544{
545 return log_buf_len;
546}
547
548/*
549 * Define how much of the log buffer we could take at maximum. The value
550 * must be greater than two. Note that only half of the buffer is available
551 * when the index points to the middle.
552 */
553#define MAX_LOG_TAKE_PART 4
554static const char trunc_msg[] = "<truncated>";
555
556static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
557{
558 /*
559 * The message should not take the whole buffer. Otherwise, it might
560 * get removed too soon.
561 */
562 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
563
564 if (*text_len > max_text_len)
565 *text_len = max_text_len;
566
567 /* enable the warning message (if there is room) */
568 *trunc_msg_len = strlen(trunc_msg);
569 if (*text_len >= *trunc_msg_len)
570 *text_len -= *trunc_msg_len;
571 else
572 *trunc_msg_len = 0;
573}
574
575int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
576
577static int syslog_action_restricted(int type)
578{
579 if (dmesg_restrict)
580 return 1;
581 /*
582 * Unless restricted, we allow "read all" and "get buffer size"
583 * for everybody.
584 */
585 return type != SYSLOG_ACTION_READ_ALL &&
586 type != SYSLOG_ACTION_SIZE_BUFFER;
587}
588
589static int check_syslog_permissions(int type, int source)
590{
591 /*
592 * If this is from /proc/kmsg and we've already opened it, then we've
593 * already done the capabilities checks at open time.
594 */
595 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
596 goto ok;
597
598 if (syslog_action_restricted(type)) {
599 if (capable(CAP_SYSLOG))
600 goto ok;
601 /*
602 * For historical reasons, accept CAP_SYS_ADMIN too, with
603 * a warning.
604 */
605 if (capable(CAP_SYS_ADMIN)) {
606 pr_warn_once("%s (%d): Attempt to access syslog with "
607 "CAP_SYS_ADMIN but no CAP_SYSLOG "
608 "(deprecated).\n",
609 current->comm, task_pid_nr(current));
610 goto ok;
611 }
612 return -EPERM;
613 }
614ok:
615 return security_syslog(type);
616}
617
618static void append_char(char **pp, char *e, char c)
619{
620 if (*pp < e)
621 *(*pp)++ = c;
622}
623
624static ssize_t info_print_ext_header(char *buf, size_t size,
625 struct printk_info *info)
626{
627 u64 ts_usec = info->ts_nsec;
628 char caller[20];
629#ifdef CONFIG_PRINTK_CALLER
630 u32 id = info->caller_id;
631
632 snprintf(caller, sizeof(caller), ",caller=%c%u",
633 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
634#else
635 caller[0] = '\0';
636#endif
637
638 do_div(ts_usec, 1000);
639
640 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
641 (info->facility << 3) | info->level, info->seq,
642 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
643}
644
645static ssize_t msg_add_ext_text(char *buf, size_t size,
646 const char *text, size_t text_len,
647 unsigned char endc)
648{
649 char *p = buf, *e = buf + size;
650 size_t i;
651
652 /* escape non-printable characters */
653 for (i = 0; i < text_len; i++) {
654 unsigned char c = text[i];
655
656 if (c < ' ' || c >= 127 || c == '\\')
657 p += scnprintf(p, e - p, "\\x%02x", c);
658 else
659 append_char(&p, e, c);
660 }
661 append_char(&p, e, endc);
662
663 return p - buf;
664}
665
666static ssize_t msg_add_dict_text(char *buf, size_t size,
667 const char *key, const char *val)
668{
669 size_t val_len = strlen(val);
670 ssize_t len;
671
672 if (!val_len)
673 return 0;
674
675 len = msg_add_ext_text(buf, size, "", 0, ' '); /* dict prefix */
676 len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
677 len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
678
679 return len;
680}
681
682static ssize_t msg_print_ext_body(char *buf, size_t size,
683 char *text, size_t text_len,
684 struct dev_printk_info *dev_info)
685{
686 ssize_t len;
687
688 len = msg_add_ext_text(buf, size, text, text_len, '\n');
689
690 if (!dev_info)
691 goto out;
692
693 len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
694 dev_info->subsystem);
695 len += msg_add_dict_text(buf + len, size - len, "DEVICE",
696 dev_info->device);
697out:
698 return len;
699}
700
701/* /dev/kmsg - userspace message inject/listen interface */
702struct devkmsg_user {
703 atomic64_t seq;
704 struct ratelimit_state rs;
705 struct mutex lock;
706 struct printk_buffers pbufs;
707};
708
709static __printf(3, 4) __cold
710int devkmsg_emit(int facility, int level, const char *fmt, ...)
711{
712 va_list args;
713 int r;
714
715 va_start(args, fmt);
716 r = vprintk_emit(facility, level, NULL, fmt, args);
717 va_end(args);
718
719 return r;
720}
721
722static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
723{
724 char *buf, *line;
725 int level = default_message_loglevel;
726 int facility = 1; /* LOG_USER */
727 struct file *file = iocb->ki_filp;
728 struct devkmsg_user *user = file->private_data;
729 size_t len = iov_iter_count(from);
730 ssize_t ret = len;
731
732 if (len > PRINTKRB_RECORD_MAX)
733 return -EINVAL;
734
735 /* Ignore when user logging is disabled. */
736 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
737 return len;
738
739 /* Ratelimit when not explicitly enabled. */
740 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
741 if (!___ratelimit(&user->rs, current->comm))
742 return ret;
743 }
744
745 buf = kmalloc(len+1, GFP_KERNEL);
746 if (buf == NULL)
747 return -ENOMEM;
748
749 buf[len] = '\0';
750 if (!copy_from_iter_full(buf, len, from)) {
751 kfree(buf);
752 return -EFAULT;
753 }
754
755 /*
756 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
757 * the decimal value represents 32bit, the lower 3 bit are the log
758 * level, the rest are the log facility.
759 *
760 * If no prefix or no userspace facility is specified, we
761 * enforce LOG_USER, to be able to reliably distinguish
762 * kernel-generated messages from userspace-injected ones.
763 */
764 line = buf;
765 if (line[0] == '<') {
766 char *endp = NULL;
767 unsigned int u;
768
769 u = simple_strtoul(line + 1, &endp, 10);
770 if (endp && endp[0] == '>') {
771 level = LOG_LEVEL(u);
772 if (LOG_FACILITY(u) != 0)
773 facility = LOG_FACILITY(u);
774 endp++;
775 line = endp;
776 }
777 }
778
779 devkmsg_emit(facility, level, "%s", line);
780 kfree(buf);
781 return ret;
782}
783
784static ssize_t devkmsg_read(struct file *file, char __user *buf,
785 size_t count, loff_t *ppos)
786{
787 struct devkmsg_user *user = file->private_data;
788 char *outbuf = &user->pbufs.outbuf[0];
789 struct printk_message pmsg = {
790 .pbufs = &user->pbufs,
791 };
792 ssize_t ret;
793
794 ret = mutex_lock_interruptible(&user->lock);
795 if (ret)
796 return ret;
797
798 if (!printk_get_next_message(&pmsg, atomic64_read(&user->seq), true, false)) {
799 if (file->f_flags & O_NONBLOCK) {
800 ret = -EAGAIN;
801 goto out;
802 }
803
804 /*
805 * Guarantee this task is visible on the waitqueue before
806 * checking the wake condition.
807 *
808 * The full memory barrier within set_current_state() of
809 * prepare_to_wait_event() pairs with the full memory barrier
810 * within wq_has_sleeper().
811 *
812 * This pairs with __wake_up_klogd:A.
813 */
814 ret = wait_event_interruptible(log_wait,
815 printk_get_next_message(&pmsg, atomic64_read(&user->seq), true,
816 false)); /* LMM(devkmsg_read:A) */
817 if (ret)
818 goto out;
819 }
820
821 if (pmsg.dropped) {
822 /* our last seen message is gone, return error and reset */
823 atomic64_set(&user->seq, pmsg.seq);
824 ret = -EPIPE;
825 goto out;
826 }
827
828 atomic64_set(&user->seq, pmsg.seq + 1);
829
830 if (pmsg.outbuf_len > count) {
831 ret = -EINVAL;
832 goto out;
833 }
834
835 if (copy_to_user(buf, outbuf, pmsg.outbuf_len)) {
836 ret = -EFAULT;
837 goto out;
838 }
839 ret = pmsg.outbuf_len;
840out:
841 mutex_unlock(&user->lock);
842 return ret;
843}
844
845/*
846 * Be careful when modifying this function!!!
847 *
848 * Only few operations are supported because the device works only with the
849 * entire variable length messages (records). Non-standard values are
850 * returned in the other cases and has been this way for quite some time.
851 * User space applications might depend on this behavior.
852 */
853static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
854{
855 struct devkmsg_user *user = file->private_data;
856 loff_t ret = 0;
857
858 if (offset)
859 return -ESPIPE;
860
861 switch (whence) {
862 case SEEK_SET:
863 /* the first record */
864 atomic64_set(&user->seq, prb_first_valid_seq(prb));
865 break;
866 case SEEK_DATA:
867 /*
868 * The first record after the last SYSLOG_ACTION_CLEAR,
869 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
870 * changes no global state, and does not clear anything.
871 */
872 atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
873 break;
874 case SEEK_END:
875 /* after the last record */
876 atomic64_set(&user->seq, prb_next_seq(prb));
877 break;
878 default:
879 ret = -EINVAL;
880 }
881 return ret;
882}
883
884static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
885{
886 struct devkmsg_user *user = file->private_data;
887 struct printk_info info;
888 __poll_t ret = 0;
889
890 poll_wait(file, &log_wait, wait);
891
892 if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
893 /* return error when data has vanished underneath us */
894 if (info.seq != atomic64_read(&user->seq))
895 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
896 else
897 ret = EPOLLIN|EPOLLRDNORM;
898 }
899
900 return ret;
901}
902
903static int devkmsg_open(struct inode *inode, struct file *file)
904{
905 struct devkmsg_user *user;
906 int err;
907
908 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
909 return -EPERM;
910
911 /* write-only does not need any file context */
912 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
913 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
914 SYSLOG_FROM_READER);
915 if (err)
916 return err;
917 }
918
919 user = kvmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
920 if (!user)
921 return -ENOMEM;
922
923 ratelimit_default_init(&user->rs);
924 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
925
926 mutex_init(&user->lock);
927
928 atomic64_set(&user->seq, prb_first_valid_seq(prb));
929
930 file->private_data = user;
931 return 0;
932}
933
934static int devkmsg_release(struct inode *inode, struct file *file)
935{
936 struct devkmsg_user *user = file->private_data;
937
938 ratelimit_state_exit(&user->rs);
939
940 mutex_destroy(&user->lock);
941 kvfree(user);
942 return 0;
943}
944
945const struct file_operations kmsg_fops = {
946 .open = devkmsg_open,
947 .read = devkmsg_read,
948 .write_iter = devkmsg_write,
949 .llseek = devkmsg_llseek,
950 .poll = devkmsg_poll,
951 .release = devkmsg_release,
952};
953
954#ifdef CONFIG_CRASH_CORE
955/*
956 * This appends the listed symbols to /proc/vmcore
957 *
958 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
959 * obtain access to symbols that are otherwise very difficult to locate. These
960 * symbols are specifically used so that utilities can access and extract the
961 * dmesg log from a vmcore file after a crash.
962 */
963void log_buf_vmcoreinfo_setup(void)
964{
965 struct dev_printk_info *dev_info = NULL;
966
967 VMCOREINFO_SYMBOL(prb);
968 VMCOREINFO_SYMBOL(printk_rb_static);
969 VMCOREINFO_SYMBOL(clear_seq);
970
971 /*
972 * Export struct size and field offsets. User space tools can
973 * parse it and detect any changes to structure down the line.
974 */
975
976 VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
977 VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
978 VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
979 VMCOREINFO_OFFSET(printk_ringbuffer, fail);
980
981 VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
982 VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
983 VMCOREINFO_OFFSET(prb_desc_ring, descs);
984 VMCOREINFO_OFFSET(prb_desc_ring, infos);
985 VMCOREINFO_OFFSET(prb_desc_ring, head_id);
986 VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
987
988 VMCOREINFO_STRUCT_SIZE(prb_desc);
989 VMCOREINFO_OFFSET(prb_desc, state_var);
990 VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
991
992 VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
993 VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
994 VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
995
996 VMCOREINFO_STRUCT_SIZE(printk_info);
997 VMCOREINFO_OFFSET(printk_info, seq);
998 VMCOREINFO_OFFSET(printk_info, ts_nsec);
999 VMCOREINFO_OFFSET(printk_info, text_len);
1000 VMCOREINFO_OFFSET(printk_info, caller_id);
1001 VMCOREINFO_OFFSET(printk_info, dev_info);
1002
1003 VMCOREINFO_STRUCT_SIZE(dev_printk_info);
1004 VMCOREINFO_OFFSET(dev_printk_info, subsystem);
1005 VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
1006 VMCOREINFO_OFFSET(dev_printk_info, device);
1007 VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
1008
1009 VMCOREINFO_STRUCT_SIZE(prb_data_ring);
1010 VMCOREINFO_OFFSET(prb_data_ring, size_bits);
1011 VMCOREINFO_OFFSET(prb_data_ring, data);
1012 VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
1013 VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
1014
1015 VMCOREINFO_SIZE(atomic_long_t);
1016 VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
1017
1018 VMCOREINFO_STRUCT_SIZE(latched_seq);
1019 VMCOREINFO_OFFSET(latched_seq, val);
1020}
1021#endif
1022
1023/* requested log_buf_len from kernel cmdline */
1024static unsigned long __initdata new_log_buf_len;
1025
1026/* we practice scaling the ring buffer by powers of 2 */
1027static void __init log_buf_len_update(u64 size)
1028{
1029 if (size > (u64)LOG_BUF_LEN_MAX) {
1030 size = (u64)LOG_BUF_LEN_MAX;
1031 pr_err("log_buf over 2G is not supported.\n");
1032 }
1033
1034 if (size)
1035 size = roundup_pow_of_two(size);
1036 if (size > log_buf_len)
1037 new_log_buf_len = (unsigned long)size;
1038}
1039
1040/* save requested log_buf_len since it's too early to process it */
1041static int __init log_buf_len_setup(char *str)
1042{
1043 u64 size;
1044
1045 if (!str)
1046 return -EINVAL;
1047
1048 size = memparse(str, &str);
1049
1050 log_buf_len_update(size);
1051
1052 return 0;
1053}
1054early_param("log_buf_len", log_buf_len_setup);
1055
1056#ifdef CONFIG_SMP
1057#define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1058
1059static void __init log_buf_add_cpu(void)
1060{
1061 unsigned int cpu_extra;
1062
1063 /*
1064 * archs should set up cpu_possible_bits properly with
1065 * set_cpu_possible() after setup_arch() but just in
1066 * case lets ensure this is valid.
1067 */
1068 if (num_possible_cpus() == 1)
1069 return;
1070
1071 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1072
1073 /* by default this will only continue through for large > 64 CPUs */
1074 if (cpu_extra <= __LOG_BUF_LEN / 2)
1075 return;
1076
1077 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1078 __LOG_CPU_MAX_BUF_LEN);
1079 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1080 cpu_extra);
1081 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1082
1083 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1084}
1085#else /* !CONFIG_SMP */
1086static inline void log_buf_add_cpu(void) {}
1087#endif /* CONFIG_SMP */
1088
1089static void __init set_percpu_data_ready(void)
1090{
1091 __printk_percpu_data_ready = true;
1092}
1093
1094static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1095 struct printk_record *r)
1096{
1097 struct prb_reserved_entry e;
1098 struct printk_record dest_r;
1099
1100 prb_rec_init_wr(&dest_r, r->info->text_len);
1101
1102 if (!prb_reserve(&e, rb, &dest_r))
1103 return 0;
1104
1105 memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1106 dest_r.info->text_len = r->info->text_len;
1107 dest_r.info->facility = r->info->facility;
1108 dest_r.info->level = r->info->level;
1109 dest_r.info->flags = r->info->flags;
1110 dest_r.info->ts_nsec = r->info->ts_nsec;
1111 dest_r.info->caller_id = r->info->caller_id;
1112 memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1113
1114 prb_final_commit(&e);
1115
1116 return prb_record_text_space(&e);
1117}
1118
1119static char setup_text_buf[PRINTKRB_RECORD_MAX] __initdata;
1120
1121void __init setup_log_buf(int early)
1122{
1123 struct printk_info *new_infos;
1124 unsigned int new_descs_count;
1125 struct prb_desc *new_descs;
1126 struct printk_info info;
1127 struct printk_record r;
1128 unsigned int text_size;
1129 size_t new_descs_size;
1130 size_t new_infos_size;
1131 unsigned long flags;
1132 char *new_log_buf;
1133 unsigned int free;
1134 u64 seq;
1135
1136 /*
1137 * Some archs call setup_log_buf() multiple times - first is very
1138 * early, e.g. from setup_arch(), and second - when percpu_areas
1139 * are initialised.
1140 */
1141 if (!early)
1142 set_percpu_data_ready();
1143
1144 if (log_buf != __log_buf)
1145 return;
1146
1147 if (!early && !new_log_buf_len)
1148 log_buf_add_cpu();
1149
1150 if (!new_log_buf_len)
1151 return;
1152
1153 new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1154 if (new_descs_count == 0) {
1155 pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1156 return;
1157 }
1158
1159 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1160 if (unlikely(!new_log_buf)) {
1161 pr_err("log_buf_len: %lu text bytes not available\n",
1162 new_log_buf_len);
1163 return;
1164 }
1165
1166 new_descs_size = new_descs_count * sizeof(struct prb_desc);
1167 new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1168 if (unlikely(!new_descs)) {
1169 pr_err("log_buf_len: %zu desc bytes not available\n",
1170 new_descs_size);
1171 goto err_free_log_buf;
1172 }
1173
1174 new_infos_size = new_descs_count * sizeof(struct printk_info);
1175 new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1176 if (unlikely(!new_infos)) {
1177 pr_err("log_buf_len: %zu info bytes not available\n",
1178 new_infos_size);
1179 goto err_free_descs;
1180 }
1181
1182 prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1183
1184 prb_init(&printk_rb_dynamic,
1185 new_log_buf, ilog2(new_log_buf_len),
1186 new_descs, ilog2(new_descs_count),
1187 new_infos);
1188
1189 local_irq_save(flags);
1190
1191 log_buf_len = new_log_buf_len;
1192 log_buf = new_log_buf;
1193 new_log_buf_len = 0;
1194
1195 free = __LOG_BUF_LEN;
1196 prb_for_each_record(0, &printk_rb_static, seq, &r) {
1197 text_size = add_to_rb(&printk_rb_dynamic, &r);
1198 if (text_size > free)
1199 free = 0;
1200 else
1201 free -= text_size;
1202 }
1203
1204 prb = &printk_rb_dynamic;
1205
1206 local_irq_restore(flags);
1207
1208 /*
1209 * Copy any remaining messages that might have appeared from
1210 * NMI context after copying but before switching to the
1211 * dynamic buffer.
1212 */
1213 prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1214 text_size = add_to_rb(&printk_rb_dynamic, &r);
1215 if (text_size > free)
1216 free = 0;
1217 else
1218 free -= text_size;
1219 }
1220
1221 if (seq != prb_next_seq(&printk_rb_static)) {
1222 pr_err("dropped %llu messages\n",
1223 prb_next_seq(&printk_rb_static) - seq);
1224 }
1225
1226 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1227 pr_info("early log buf free: %u(%u%%)\n",
1228 free, (free * 100) / __LOG_BUF_LEN);
1229 return;
1230
1231err_free_descs:
1232 memblock_free(new_descs, new_descs_size);
1233err_free_log_buf:
1234 memblock_free(new_log_buf, new_log_buf_len);
1235}
1236
1237static bool __read_mostly ignore_loglevel;
1238
1239static int __init ignore_loglevel_setup(char *str)
1240{
1241 ignore_loglevel = true;
1242 pr_info("debug: ignoring loglevel setting.\n");
1243
1244 return 0;
1245}
1246
1247early_param("ignore_loglevel", ignore_loglevel_setup);
1248module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1249MODULE_PARM_DESC(ignore_loglevel,
1250 "ignore loglevel setting (prints all kernel messages to the console)");
1251
1252static bool suppress_message_printing(int level)
1253{
1254 return (level >= console_loglevel && !ignore_loglevel);
1255}
1256
1257#ifdef CONFIG_BOOT_PRINTK_DELAY
1258
1259static int boot_delay; /* msecs delay after each printk during bootup */
1260static unsigned long long loops_per_msec; /* based on boot_delay */
1261
1262static int __init boot_delay_setup(char *str)
1263{
1264 unsigned long lpj;
1265
1266 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1267 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1268
1269 get_option(&str, &boot_delay);
1270 if (boot_delay > 10 * 1000)
1271 boot_delay = 0;
1272
1273 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1274 "HZ: %d, loops_per_msec: %llu\n",
1275 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1276 return 0;
1277}
1278early_param("boot_delay", boot_delay_setup);
1279
1280static void boot_delay_msec(int level)
1281{
1282 unsigned long long k;
1283 unsigned long timeout;
1284
1285 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1286 || suppress_message_printing(level)) {
1287 return;
1288 }
1289
1290 k = (unsigned long long)loops_per_msec * boot_delay;
1291
1292 timeout = jiffies + msecs_to_jiffies(boot_delay);
1293 while (k) {
1294 k--;
1295 cpu_relax();
1296 /*
1297 * use (volatile) jiffies to prevent
1298 * compiler reduction; loop termination via jiffies
1299 * is secondary and may or may not happen.
1300 */
1301 if (time_after(jiffies, timeout))
1302 break;
1303 touch_nmi_watchdog();
1304 }
1305}
1306#else
1307static inline void boot_delay_msec(int level)
1308{
1309}
1310#endif
1311
1312static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1313module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1314
1315static size_t print_syslog(unsigned int level, char *buf)
1316{
1317 return sprintf(buf, "<%u>", level);
1318}
1319
1320static size_t print_time(u64 ts, char *buf)
1321{
1322 unsigned long rem_nsec = do_div(ts, 1000000000);
1323
1324 return sprintf(buf, "[%5lu.%06lu]",
1325 (unsigned long)ts, rem_nsec / 1000);
1326}
1327
1328#ifdef CONFIG_PRINTK_CALLER
1329static size_t print_caller(u32 id, char *buf)
1330{
1331 char caller[12];
1332
1333 snprintf(caller, sizeof(caller), "%c%u",
1334 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1335 return sprintf(buf, "[%6s]", caller);
1336}
1337#else
1338#define print_caller(id, buf) 0
1339#endif
1340
1341static size_t info_print_prefix(const struct printk_info *info, bool syslog,
1342 bool time, char *buf)
1343{
1344 size_t len = 0;
1345
1346 if (syslog)
1347 len = print_syslog((info->facility << 3) | info->level, buf);
1348
1349 if (time)
1350 len += print_time(info->ts_nsec, buf + len);
1351
1352 len += print_caller(info->caller_id, buf + len);
1353
1354 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1355 buf[len++] = ' ';
1356 buf[len] = '\0';
1357 }
1358
1359 return len;
1360}
1361
1362/*
1363 * Prepare the record for printing. The text is shifted within the given
1364 * buffer to avoid a need for another one. The following operations are
1365 * done:
1366 *
1367 * - Add prefix for each line.
1368 * - Drop truncated lines that no longer fit into the buffer.
1369 * - Add the trailing newline that has been removed in vprintk_store().
1370 * - Add a string terminator.
1371 *
1372 * Since the produced string is always terminated, the maximum possible
1373 * return value is @r->text_buf_size - 1;
1374 *
1375 * Return: The length of the updated/prepared text, including the added
1376 * prefixes and the newline. The terminator is not counted. The dropped
1377 * line(s) are not counted.
1378 */
1379static size_t record_print_text(struct printk_record *r, bool syslog,
1380 bool time)
1381{
1382 size_t text_len = r->info->text_len;
1383 size_t buf_size = r->text_buf_size;
1384 char *text = r->text_buf;
1385 char prefix[PRINTK_PREFIX_MAX];
1386 bool truncated = false;
1387 size_t prefix_len;
1388 size_t line_len;
1389 size_t len = 0;
1390 char *next;
1391
1392 /*
1393 * If the message was truncated because the buffer was not large
1394 * enough, treat the available text as if it were the full text.
1395 */
1396 if (text_len > buf_size)
1397 text_len = buf_size;
1398
1399 prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1400
1401 /*
1402 * @text_len: bytes of unprocessed text
1403 * @line_len: bytes of current line _without_ newline
1404 * @text: pointer to beginning of current line
1405 * @len: number of bytes prepared in r->text_buf
1406 */
1407 for (;;) {
1408 next = memchr(text, '\n', text_len);
1409 if (next) {
1410 line_len = next - text;
1411 } else {
1412 /* Drop truncated line(s). */
1413 if (truncated)
1414 break;
1415 line_len = text_len;
1416 }
1417
1418 /*
1419 * Truncate the text if there is not enough space to add the
1420 * prefix and a trailing newline and a terminator.
1421 */
1422 if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1423 /* Drop even the current line if no space. */
1424 if (len + prefix_len + line_len + 1 + 1 > buf_size)
1425 break;
1426
1427 text_len = buf_size - len - prefix_len - 1 - 1;
1428 truncated = true;
1429 }
1430
1431 memmove(text + prefix_len, text, text_len);
1432 memcpy(text, prefix, prefix_len);
1433
1434 /*
1435 * Increment the prepared length to include the text and
1436 * prefix that were just moved+copied. Also increment for the
1437 * newline at the end of this line. If this is the last line,
1438 * there is no newline, but it will be added immediately below.
1439 */
1440 len += prefix_len + line_len + 1;
1441 if (text_len == line_len) {
1442 /*
1443 * This is the last line. Add the trailing newline
1444 * removed in vprintk_store().
1445 */
1446 text[prefix_len + line_len] = '\n';
1447 break;
1448 }
1449
1450 /*
1451 * Advance beyond the added prefix and the related line with
1452 * its newline.
1453 */
1454 text += prefix_len + line_len + 1;
1455
1456 /*
1457 * The remaining text has only decreased by the line with its
1458 * newline.
1459 *
1460 * Note that @text_len can become zero. It happens when @text
1461 * ended with a newline (either due to truncation or the
1462 * original string ending with "\n\n"). The loop is correctly
1463 * repeated and (if not truncated) an empty line with a prefix
1464 * will be prepared.
1465 */
1466 text_len -= line_len + 1;
1467 }
1468
1469 /*
1470 * If a buffer was provided, it will be terminated. Space for the
1471 * string terminator is guaranteed to be available. The terminator is
1472 * not counted in the return value.
1473 */
1474 if (buf_size > 0)
1475 r->text_buf[len] = 0;
1476
1477 return len;
1478}
1479
1480static size_t get_record_print_text_size(struct printk_info *info,
1481 unsigned int line_count,
1482 bool syslog, bool time)
1483{
1484 char prefix[PRINTK_PREFIX_MAX];
1485 size_t prefix_len;
1486
1487 prefix_len = info_print_prefix(info, syslog, time, prefix);
1488
1489 /*
1490 * Each line will be preceded with a prefix. The intermediate
1491 * newlines are already within the text, but a final trailing
1492 * newline will be added.
1493 */
1494 return ((prefix_len * line_count) + info->text_len + 1);
1495}
1496
1497/*
1498 * Beginning with @start_seq, find the first record where it and all following
1499 * records up to (but not including) @max_seq fit into @size.
1500 *
1501 * @max_seq is simply an upper bound and does not need to exist. If the caller
1502 * does not require an upper bound, -1 can be used for @max_seq.
1503 */
1504static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1505 bool syslog, bool time)
1506{
1507 struct printk_info info;
1508 unsigned int line_count;
1509 size_t len = 0;
1510 u64 seq;
1511
1512 /* Determine the size of the records up to @max_seq. */
1513 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1514 if (info.seq >= max_seq)
1515 break;
1516 len += get_record_print_text_size(&info, line_count, syslog, time);
1517 }
1518
1519 /*
1520 * Adjust the upper bound for the next loop to avoid subtracting
1521 * lengths that were never added.
1522 */
1523 if (seq < max_seq)
1524 max_seq = seq;
1525
1526 /*
1527 * Move first record forward until length fits into the buffer. Ignore
1528 * newest messages that were not counted in the above cycle. Messages
1529 * might appear and get lost in the meantime. This is a best effort
1530 * that prevents an infinite loop that could occur with a retry.
1531 */
1532 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1533 if (len <= size || info.seq >= max_seq)
1534 break;
1535 len -= get_record_print_text_size(&info, line_count, syslog, time);
1536 }
1537
1538 return seq;
1539}
1540
1541/* The caller is responsible for making sure @size is greater than 0. */
1542static int syslog_print(char __user *buf, int size)
1543{
1544 struct printk_info info;
1545 struct printk_record r;
1546 char *text;
1547 int len = 0;
1548 u64 seq;
1549
1550 text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1551 if (!text)
1552 return -ENOMEM;
1553
1554 prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1555
1556 mutex_lock(&syslog_lock);
1557
1558 /*
1559 * Wait for the @syslog_seq record to be available. @syslog_seq may
1560 * change while waiting.
1561 */
1562 do {
1563 seq = syslog_seq;
1564
1565 mutex_unlock(&syslog_lock);
1566 /*
1567 * Guarantee this task is visible on the waitqueue before
1568 * checking the wake condition.
1569 *
1570 * The full memory barrier within set_current_state() of
1571 * prepare_to_wait_event() pairs with the full memory barrier
1572 * within wq_has_sleeper().
1573 *
1574 * This pairs with __wake_up_klogd:A.
1575 */
1576 len = wait_event_interruptible(log_wait,
1577 prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
1578 mutex_lock(&syslog_lock);
1579
1580 if (len)
1581 goto out;
1582 } while (syslog_seq != seq);
1583
1584 /*
1585 * Copy records that fit into the buffer. The above cycle makes sure
1586 * that the first record is always available.
1587 */
1588 do {
1589 size_t n;
1590 size_t skip;
1591 int err;
1592
1593 if (!prb_read_valid(prb, syslog_seq, &r))
1594 break;
1595
1596 if (r.info->seq != syslog_seq) {
1597 /* message is gone, move to next valid one */
1598 syslog_seq = r.info->seq;
1599 syslog_partial = 0;
1600 }
1601
1602 /*
1603 * To keep reading/counting partial line consistent,
1604 * use printk_time value as of the beginning of a line.
1605 */
1606 if (!syslog_partial)
1607 syslog_time = printk_time;
1608
1609 skip = syslog_partial;
1610 n = record_print_text(&r, true, syslog_time);
1611 if (n - syslog_partial <= size) {
1612 /* message fits into buffer, move forward */
1613 syslog_seq = r.info->seq + 1;
1614 n -= syslog_partial;
1615 syslog_partial = 0;
1616 } else if (!len){
1617 /* partial read(), remember position */
1618 n = size;
1619 syslog_partial += n;
1620 } else
1621 n = 0;
1622
1623 if (!n)
1624 break;
1625
1626 mutex_unlock(&syslog_lock);
1627 err = copy_to_user(buf, text + skip, n);
1628 mutex_lock(&syslog_lock);
1629
1630 if (err) {
1631 if (!len)
1632 len = -EFAULT;
1633 break;
1634 }
1635
1636 len += n;
1637 size -= n;
1638 buf += n;
1639 } while (size);
1640out:
1641 mutex_unlock(&syslog_lock);
1642 kfree(text);
1643 return len;
1644}
1645
1646static int syslog_print_all(char __user *buf, int size, bool clear)
1647{
1648 struct printk_info info;
1649 struct printk_record r;
1650 char *text;
1651 int len = 0;
1652 u64 seq;
1653 bool time;
1654
1655 text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1656 if (!text)
1657 return -ENOMEM;
1658
1659 time = printk_time;
1660 /*
1661 * Find first record that fits, including all following records,
1662 * into the user-provided buffer for this dump.
1663 */
1664 seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1665 size, true, time);
1666
1667 prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1668
1669 prb_for_each_record(seq, prb, seq, &r) {
1670 int textlen;
1671
1672 textlen = record_print_text(&r, true, time);
1673
1674 if (len + textlen > size) {
1675 seq--;
1676 break;
1677 }
1678
1679 if (copy_to_user(buf + len, text, textlen))
1680 len = -EFAULT;
1681 else
1682 len += textlen;
1683
1684 if (len < 0)
1685 break;
1686 }
1687
1688 if (clear) {
1689 mutex_lock(&syslog_lock);
1690 latched_seq_write(&clear_seq, seq);
1691 mutex_unlock(&syslog_lock);
1692 }
1693
1694 kfree(text);
1695 return len;
1696}
1697
1698static void syslog_clear(void)
1699{
1700 mutex_lock(&syslog_lock);
1701 latched_seq_write(&clear_seq, prb_next_seq(prb));
1702 mutex_unlock(&syslog_lock);
1703}
1704
1705int do_syslog(int type, char __user *buf, int len, int source)
1706{
1707 struct printk_info info;
1708 bool clear = false;
1709 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1710 int error;
1711
1712 error = check_syslog_permissions(type, source);
1713 if (error)
1714 return error;
1715
1716 switch (type) {
1717 case SYSLOG_ACTION_CLOSE: /* Close log */
1718 break;
1719 case SYSLOG_ACTION_OPEN: /* Open log */
1720 break;
1721 case SYSLOG_ACTION_READ: /* Read from log */
1722 if (!buf || len < 0)
1723 return -EINVAL;
1724 if (!len)
1725 return 0;
1726 if (!access_ok(buf, len))
1727 return -EFAULT;
1728 error = syslog_print(buf, len);
1729 break;
1730 /* Read/clear last kernel messages */
1731 case SYSLOG_ACTION_READ_CLEAR:
1732 clear = true;
1733 fallthrough;
1734 /* Read last kernel messages */
1735 case SYSLOG_ACTION_READ_ALL:
1736 if (!buf || len < 0)
1737 return -EINVAL;
1738 if (!len)
1739 return 0;
1740 if (!access_ok(buf, len))
1741 return -EFAULT;
1742 error = syslog_print_all(buf, len, clear);
1743 break;
1744 /* Clear ring buffer */
1745 case SYSLOG_ACTION_CLEAR:
1746 syslog_clear();
1747 break;
1748 /* Disable logging to console */
1749 case SYSLOG_ACTION_CONSOLE_OFF:
1750 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1751 saved_console_loglevel = console_loglevel;
1752 console_loglevel = minimum_console_loglevel;
1753 break;
1754 /* Enable logging to console */
1755 case SYSLOG_ACTION_CONSOLE_ON:
1756 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1757 console_loglevel = saved_console_loglevel;
1758 saved_console_loglevel = LOGLEVEL_DEFAULT;
1759 }
1760 break;
1761 /* Set level of messages printed to console */
1762 case SYSLOG_ACTION_CONSOLE_LEVEL:
1763 if (len < 1 || len > 8)
1764 return -EINVAL;
1765 if (len < minimum_console_loglevel)
1766 len = minimum_console_loglevel;
1767 console_loglevel = len;
1768 /* Implicitly re-enable logging to console */
1769 saved_console_loglevel = LOGLEVEL_DEFAULT;
1770 break;
1771 /* Number of chars in the log buffer */
1772 case SYSLOG_ACTION_SIZE_UNREAD:
1773 mutex_lock(&syslog_lock);
1774 if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1775 /* No unread messages. */
1776 mutex_unlock(&syslog_lock);
1777 return 0;
1778 }
1779 if (info.seq != syslog_seq) {
1780 /* messages are gone, move to first one */
1781 syslog_seq = info.seq;
1782 syslog_partial = 0;
1783 }
1784 if (source == SYSLOG_FROM_PROC) {
1785 /*
1786 * Short-cut for poll(/"proc/kmsg") which simply checks
1787 * for pending data, not the size; return the count of
1788 * records, not the length.
1789 */
1790 error = prb_next_seq(prb) - syslog_seq;
1791 } else {
1792 bool time = syslog_partial ? syslog_time : printk_time;
1793 unsigned int line_count;
1794 u64 seq;
1795
1796 prb_for_each_info(syslog_seq, prb, seq, &info,
1797 &line_count) {
1798 error += get_record_print_text_size(&info, line_count,
1799 true, time);
1800 time = printk_time;
1801 }
1802 error -= syslog_partial;
1803 }
1804 mutex_unlock(&syslog_lock);
1805 break;
1806 /* Size of the log buffer */
1807 case SYSLOG_ACTION_SIZE_BUFFER:
1808 error = log_buf_len;
1809 break;
1810 default:
1811 error = -EINVAL;
1812 break;
1813 }
1814
1815 return error;
1816}
1817
1818SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1819{
1820 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1821}
1822
1823/*
1824 * Special console_lock variants that help to reduce the risk of soft-lockups.
1825 * They allow to pass console_lock to another printk() call using a busy wait.
1826 */
1827
1828#ifdef CONFIG_LOCKDEP
1829static struct lockdep_map console_owner_dep_map = {
1830 .name = "console_owner"
1831};
1832#endif
1833
1834static DEFINE_RAW_SPINLOCK(console_owner_lock);
1835static struct task_struct *console_owner;
1836static bool console_waiter;
1837
1838/**
1839 * console_lock_spinning_enable - mark beginning of code where another
1840 * thread might safely busy wait
1841 *
1842 * This basically converts console_lock into a spinlock. This marks
1843 * the section where the console_lock owner can not sleep, because
1844 * there may be a waiter spinning (like a spinlock). Also it must be
1845 * ready to hand over the lock at the end of the section.
1846 */
1847static void console_lock_spinning_enable(void)
1848{
1849 raw_spin_lock(&console_owner_lock);
1850 console_owner = current;
1851 raw_spin_unlock(&console_owner_lock);
1852
1853 /* The waiter may spin on us after setting console_owner */
1854 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1855}
1856
1857/**
1858 * console_lock_spinning_disable_and_check - mark end of code where another
1859 * thread was able to busy wait and check if there is a waiter
1860 * @cookie: cookie returned from console_srcu_read_lock()
1861 *
1862 * This is called at the end of the section where spinning is allowed.
1863 * It has two functions. First, it is a signal that it is no longer
1864 * safe to start busy waiting for the lock. Second, it checks if
1865 * there is a busy waiter and passes the lock rights to her.
1866 *
1867 * Important: Callers lose both the console_lock and the SRCU read lock if
1868 * there was a busy waiter. They must not touch items synchronized by
1869 * console_lock or SRCU read lock in this case.
1870 *
1871 * Return: 1 if the lock rights were passed, 0 otherwise.
1872 */
1873static int console_lock_spinning_disable_and_check(int cookie)
1874{
1875 int waiter;
1876
1877 raw_spin_lock(&console_owner_lock);
1878 waiter = READ_ONCE(console_waiter);
1879 console_owner = NULL;
1880 raw_spin_unlock(&console_owner_lock);
1881
1882 if (!waiter) {
1883 spin_release(&console_owner_dep_map, _THIS_IP_);
1884 return 0;
1885 }
1886
1887 /* The waiter is now free to continue */
1888 WRITE_ONCE(console_waiter, false);
1889
1890 spin_release(&console_owner_dep_map, _THIS_IP_);
1891
1892 /*
1893 * Preserve lockdep lock ordering. Release the SRCU read lock before
1894 * releasing the console_lock.
1895 */
1896 console_srcu_read_unlock(cookie);
1897
1898 /*
1899 * Hand off console_lock to waiter. The waiter will perform
1900 * the up(). After this, the waiter is the console_lock owner.
1901 */
1902 mutex_release(&console_lock_dep_map, _THIS_IP_);
1903 return 1;
1904}
1905
1906/**
1907 * console_trylock_spinning - try to get console_lock by busy waiting
1908 *
1909 * This allows to busy wait for the console_lock when the current
1910 * owner is running in specially marked sections. It means that
1911 * the current owner is running and cannot reschedule until it
1912 * is ready to lose the lock.
1913 *
1914 * Return: 1 if we got the lock, 0 othrewise
1915 */
1916static int console_trylock_spinning(void)
1917{
1918 struct task_struct *owner = NULL;
1919 bool waiter;
1920 bool spin = false;
1921 unsigned long flags;
1922
1923 if (console_trylock())
1924 return 1;
1925
1926 /*
1927 * It's unsafe to spin once a panic has begun. If we are the
1928 * panic CPU, we may have already halted the owner of the
1929 * console_sem. If we are not the panic CPU, then we should
1930 * avoid taking console_sem, so the panic CPU has a better
1931 * chance of cleanly acquiring it later.
1932 */
1933 if (panic_in_progress())
1934 return 0;
1935
1936 printk_safe_enter_irqsave(flags);
1937
1938 raw_spin_lock(&console_owner_lock);
1939 owner = READ_ONCE(console_owner);
1940 waiter = READ_ONCE(console_waiter);
1941 if (!waiter && owner && owner != current) {
1942 WRITE_ONCE(console_waiter, true);
1943 spin = true;
1944 }
1945 raw_spin_unlock(&console_owner_lock);
1946
1947 /*
1948 * If there is an active printk() writing to the
1949 * consoles, instead of having it write our data too,
1950 * see if we can offload that load from the active
1951 * printer, and do some printing ourselves.
1952 * Go into a spin only if there isn't already a waiter
1953 * spinning, and there is an active printer, and
1954 * that active printer isn't us (recursive printk?).
1955 */
1956 if (!spin) {
1957 printk_safe_exit_irqrestore(flags);
1958 return 0;
1959 }
1960
1961 /* We spin waiting for the owner to release us */
1962 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1963 /* Owner will clear console_waiter on hand off */
1964 while (READ_ONCE(console_waiter))
1965 cpu_relax();
1966 spin_release(&console_owner_dep_map, _THIS_IP_);
1967
1968 printk_safe_exit_irqrestore(flags);
1969 /*
1970 * The owner passed the console lock to us.
1971 * Since we did not spin on console lock, annotate
1972 * this as a trylock. Otherwise lockdep will
1973 * complain.
1974 */
1975 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1976
1977 return 1;
1978}
1979
1980/*
1981 * Recursion is tracked separately on each CPU. If NMIs are supported, an
1982 * additional NMI context per CPU is also separately tracked. Until per-CPU
1983 * is available, a separate "early tracking" is performed.
1984 */
1985static DEFINE_PER_CPU(u8, printk_count);
1986static u8 printk_count_early;
1987#ifdef CONFIG_HAVE_NMI
1988static DEFINE_PER_CPU(u8, printk_count_nmi);
1989static u8 printk_count_nmi_early;
1990#endif
1991
1992/*
1993 * Recursion is limited to keep the output sane. printk() should not require
1994 * more than 1 level of recursion (allowing, for example, printk() to trigger
1995 * a WARN), but a higher value is used in case some printk-internal errors
1996 * exist, such as the ringbuffer validation checks failing.
1997 */
1998#define PRINTK_MAX_RECURSION 3
1999
2000/*
2001 * Return a pointer to the dedicated counter for the CPU+context of the
2002 * caller.
2003 */
2004static u8 *__printk_recursion_counter(void)
2005{
2006#ifdef CONFIG_HAVE_NMI
2007 if (in_nmi()) {
2008 if (printk_percpu_data_ready())
2009 return this_cpu_ptr(&printk_count_nmi);
2010 return &printk_count_nmi_early;
2011 }
2012#endif
2013 if (printk_percpu_data_ready())
2014 return this_cpu_ptr(&printk_count);
2015 return &printk_count_early;
2016}
2017
2018/*
2019 * Enter recursion tracking. Interrupts are disabled to simplify tracking.
2020 * The caller must check the boolean return value to see if the recursion is
2021 * allowed. On failure, interrupts are not disabled.
2022 *
2023 * @recursion_ptr must be a variable of type (u8 *) and is the same variable
2024 * that is passed to printk_exit_irqrestore().
2025 */
2026#define printk_enter_irqsave(recursion_ptr, flags) \
2027({ \
2028 bool success = true; \
2029 \
2030 typecheck(u8 *, recursion_ptr); \
2031 local_irq_save(flags); \
2032 (recursion_ptr) = __printk_recursion_counter(); \
2033 if (*(recursion_ptr) > PRINTK_MAX_RECURSION) { \
2034 local_irq_restore(flags); \
2035 success = false; \
2036 } else { \
2037 (*(recursion_ptr))++; \
2038 } \
2039 success; \
2040})
2041
2042/* Exit recursion tracking, restoring interrupts. */
2043#define printk_exit_irqrestore(recursion_ptr, flags) \
2044 do { \
2045 typecheck(u8 *, recursion_ptr); \
2046 (*(recursion_ptr))--; \
2047 local_irq_restore(flags); \
2048 } while (0)
2049
2050int printk_delay_msec __read_mostly;
2051
2052static inline void printk_delay(int level)
2053{
2054 boot_delay_msec(level);
2055
2056 if (unlikely(printk_delay_msec)) {
2057 int m = printk_delay_msec;
2058
2059 while (m--) {
2060 mdelay(1);
2061 touch_nmi_watchdog();
2062 }
2063 }
2064}
2065
2066static inline u32 printk_caller_id(void)
2067{
2068 return in_task() ? task_pid_nr(current) :
2069 0x80000000 + smp_processor_id();
2070}
2071
2072/**
2073 * printk_parse_prefix - Parse level and control flags.
2074 *
2075 * @text: The terminated text message.
2076 * @level: A pointer to the current level value, will be updated.
2077 * @flags: A pointer to the current printk_info flags, will be updated.
2078 *
2079 * @level may be NULL if the caller is not interested in the parsed value.
2080 * Otherwise the variable pointed to by @level must be set to
2081 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2082 *
2083 * @flags may be NULL if the caller is not interested in the parsed value.
2084 * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2085 * value.
2086 *
2087 * Return: The length of the parsed level and control flags.
2088 */
2089u16 printk_parse_prefix(const char *text, int *level,
2090 enum printk_info_flags *flags)
2091{
2092 u16 prefix_len = 0;
2093 int kern_level;
2094
2095 while (*text) {
2096 kern_level = printk_get_level(text);
2097 if (!kern_level)
2098 break;
2099
2100 switch (kern_level) {
2101 case '0' ... '7':
2102 if (level && *level == LOGLEVEL_DEFAULT)
2103 *level = kern_level - '0';
2104 break;
2105 case 'c': /* KERN_CONT */
2106 if (flags)
2107 *flags |= LOG_CONT;
2108 }
2109
2110 prefix_len += 2;
2111 text += 2;
2112 }
2113
2114 return prefix_len;
2115}
2116
2117__printf(5, 0)
2118static u16 printk_sprint(char *text, u16 size, int facility,
2119 enum printk_info_flags *flags, const char *fmt,
2120 va_list args)
2121{
2122 u16 text_len;
2123
2124 text_len = vscnprintf(text, size, fmt, args);
2125
2126 /* Mark and strip a trailing newline. */
2127 if (text_len && text[text_len - 1] == '\n') {
2128 text_len--;
2129 *flags |= LOG_NEWLINE;
2130 }
2131
2132 /* Strip log level and control flags. */
2133 if (facility == 0) {
2134 u16 prefix_len;
2135
2136 prefix_len = printk_parse_prefix(text, NULL, NULL);
2137 if (prefix_len) {
2138 text_len -= prefix_len;
2139 memmove(text, text + prefix_len, text_len);
2140 }
2141 }
2142
2143 trace_console(text, text_len);
2144
2145 return text_len;
2146}
2147
2148__printf(4, 0)
2149int vprintk_store(int facility, int level,
2150 const struct dev_printk_info *dev_info,
2151 const char *fmt, va_list args)
2152{
2153 struct prb_reserved_entry e;
2154 enum printk_info_flags flags = 0;
2155 struct printk_record r;
2156 unsigned long irqflags;
2157 u16 trunc_msg_len = 0;
2158 char prefix_buf[8];
2159 u8 *recursion_ptr;
2160 u16 reserve_size;
2161 va_list args2;
2162 u32 caller_id;
2163 u16 text_len;
2164 int ret = 0;
2165 u64 ts_nsec;
2166
2167 if (!printk_enter_irqsave(recursion_ptr, irqflags))
2168 return 0;
2169
2170 /*
2171 * Since the duration of printk() can vary depending on the message
2172 * and state of the ringbuffer, grab the timestamp now so that it is
2173 * close to the call of printk(). This provides a more deterministic
2174 * timestamp with respect to the caller.
2175 */
2176 ts_nsec = local_clock();
2177
2178 caller_id = printk_caller_id();
2179
2180 /*
2181 * The sprintf needs to come first since the syslog prefix might be
2182 * passed in as a parameter. An extra byte must be reserved so that
2183 * later the vscnprintf() into the reserved buffer has room for the
2184 * terminating '\0', which is not counted by vsnprintf().
2185 */
2186 va_copy(args2, args);
2187 reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2188 va_end(args2);
2189
2190 if (reserve_size > PRINTKRB_RECORD_MAX)
2191 reserve_size = PRINTKRB_RECORD_MAX;
2192
2193 /* Extract log level or control flags. */
2194 if (facility == 0)
2195 printk_parse_prefix(&prefix_buf[0], &level, &flags);
2196
2197 if (level == LOGLEVEL_DEFAULT)
2198 level = default_message_loglevel;
2199
2200 if (dev_info)
2201 flags |= LOG_NEWLINE;
2202
2203 if (flags & LOG_CONT) {
2204 prb_rec_init_wr(&r, reserve_size);
2205 if (prb_reserve_in_last(&e, prb, &r, caller_id, PRINTKRB_RECORD_MAX)) {
2206 text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2207 facility, &flags, fmt, args);
2208 r.info->text_len += text_len;
2209
2210 if (flags & LOG_NEWLINE) {
2211 r.info->flags |= LOG_NEWLINE;
2212 prb_final_commit(&e);
2213 } else {
2214 prb_commit(&e);
2215 }
2216
2217 ret = text_len;
2218 goto out;
2219 }
2220 }
2221
2222 /*
2223 * Explicitly initialize the record before every prb_reserve() call.
2224 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2225 * structure when they fail.
2226 */
2227 prb_rec_init_wr(&r, reserve_size);
2228 if (!prb_reserve(&e, prb, &r)) {
2229 /* truncate the message if it is too long for empty buffer */
2230 truncate_msg(&reserve_size, &trunc_msg_len);
2231
2232 prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2233 if (!prb_reserve(&e, prb, &r))
2234 goto out;
2235 }
2236
2237 /* fill message */
2238 text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2239 if (trunc_msg_len)
2240 memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2241 r.info->text_len = text_len + trunc_msg_len;
2242 r.info->facility = facility;
2243 r.info->level = level & 7;
2244 r.info->flags = flags & 0x1f;
2245 r.info->ts_nsec = ts_nsec;
2246 r.info->caller_id = caller_id;
2247 if (dev_info)
2248 memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2249
2250 /* A message without a trailing newline can be continued. */
2251 if (!(flags & LOG_NEWLINE))
2252 prb_commit(&e);
2253 else
2254 prb_final_commit(&e);
2255
2256 ret = text_len + trunc_msg_len;
2257out:
2258 printk_exit_irqrestore(recursion_ptr, irqflags);
2259 return ret;
2260}
2261
2262asmlinkage int vprintk_emit(int facility, int level,
2263 const struct dev_printk_info *dev_info,
2264 const char *fmt, va_list args)
2265{
2266 int printed_len;
2267 bool in_sched = false;
2268
2269 /* Suppress unimportant messages after panic happens */
2270 if (unlikely(suppress_printk))
2271 return 0;
2272
2273 if (unlikely(suppress_panic_printk) &&
2274 atomic_read(&panic_cpu) != raw_smp_processor_id())
2275 return 0;
2276
2277 if (level == LOGLEVEL_SCHED) {
2278 level = LOGLEVEL_DEFAULT;
2279 in_sched = true;
2280 }
2281
2282 printk_delay(level);
2283
2284 printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2285
2286 /* If called from the scheduler, we can not call up(). */
2287 if (!in_sched) {
2288 /*
2289 * The caller may be holding system-critical or
2290 * timing-sensitive locks. Disable preemption during
2291 * printing of all remaining records to all consoles so that
2292 * this context can return as soon as possible. Hopefully
2293 * another printk() caller will take over the printing.
2294 */
2295 preempt_disable();
2296 /*
2297 * Try to acquire and then immediately release the console
2298 * semaphore. The release will print out buffers. With the
2299 * spinning variant, this context tries to take over the
2300 * printing from another printing context.
2301 */
2302 if (console_trylock_spinning())
2303 console_unlock();
2304 preempt_enable();
2305 }
2306
2307 if (in_sched)
2308 defer_console_output();
2309 else
2310 wake_up_klogd();
2311
2312 return printed_len;
2313}
2314EXPORT_SYMBOL(vprintk_emit);
2315
2316int vprintk_default(const char *fmt, va_list args)
2317{
2318 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2319}
2320EXPORT_SYMBOL_GPL(vprintk_default);
2321
2322asmlinkage __visible int _printk(const char *fmt, ...)
2323{
2324 va_list args;
2325 int r;
2326
2327 va_start(args, fmt);
2328 r = vprintk(fmt, args);
2329 va_end(args);
2330
2331 return r;
2332}
2333EXPORT_SYMBOL(_printk);
2334
2335static bool pr_flush(int timeout_ms, bool reset_on_progress);
2336static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress);
2337
2338#else /* CONFIG_PRINTK */
2339
2340#define printk_time false
2341
2342#define prb_read_valid(rb, seq, r) false
2343#define prb_first_valid_seq(rb) 0
2344#define prb_next_seq(rb) 0
2345
2346static u64 syslog_seq;
2347
2348static bool pr_flush(int timeout_ms, bool reset_on_progress) { return true; }
2349static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress) { return true; }
2350
2351#endif /* CONFIG_PRINTK */
2352
2353#ifdef CONFIG_EARLY_PRINTK
2354struct console *early_console;
2355
2356asmlinkage __visible void early_printk(const char *fmt, ...)
2357{
2358 va_list ap;
2359 char buf[512];
2360 int n;
2361
2362 if (!early_console)
2363 return;
2364
2365 va_start(ap, fmt);
2366 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2367 va_end(ap);
2368
2369 early_console->write(early_console, buf, n);
2370}
2371#endif
2372
2373static void set_user_specified(struct console_cmdline *c, bool user_specified)
2374{
2375 if (!user_specified)
2376 return;
2377
2378 /*
2379 * @c console was defined by the user on the command line.
2380 * Do not clear when added twice also by SPCR or the device tree.
2381 */
2382 c->user_specified = true;
2383 /* At least one console defined by the user on the command line. */
2384 console_set_on_cmdline = 1;
2385}
2386
2387static int __add_preferred_console(const char *name, const short idx, char *options,
2388 char *brl_options, bool user_specified)
2389{
2390 struct console_cmdline *c;
2391 int i;
2392
2393 /*
2394 * We use a signed short index for struct console for device drivers to
2395 * indicate a not yet assigned index or port. However, a negative index
2396 * value is not valid for preferred console.
2397 */
2398 if (idx < 0)
2399 return -EINVAL;
2400
2401 /*
2402 * See if this tty is not yet registered, and
2403 * if we have a slot free.
2404 */
2405 for (i = 0, c = console_cmdline;
2406 i < MAX_CMDLINECONSOLES && c->name[0];
2407 i++, c++) {
2408 if (strcmp(c->name, name) == 0 && c->index == idx) {
2409 if (!brl_options)
2410 preferred_console = i;
2411 set_user_specified(c, user_specified);
2412 return 0;
2413 }
2414 }
2415 if (i == MAX_CMDLINECONSOLES)
2416 return -E2BIG;
2417 if (!brl_options)
2418 preferred_console = i;
2419 strscpy(c->name, name, sizeof(c->name));
2420 c->options = options;
2421 set_user_specified(c, user_specified);
2422 braille_set_options(c, brl_options);
2423
2424 c->index = idx;
2425 return 0;
2426}
2427
2428static int __init console_msg_format_setup(char *str)
2429{
2430 if (!strcmp(str, "syslog"))
2431 console_msg_format = MSG_FORMAT_SYSLOG;
2432 if (!strcmp(str, "default"))
2433 console_msg_format = MSG_FORMAT_DEFAULT;
2434 return 1;
2435}
2436__setup("console_msg_format=", console_msg_format_setup);
2437
2438/*
2439 * Set up a console. Called via do_early_param() in init/main.c
2440 * for each "console=" parameter in the boot command line.
2441 */
2442static int __init console_setup(char *str)
2443{
2444 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2445 char *s, *options, *brl_options = NULL;
2446 int idx;
2447
2448 /*
2449 * console="" or console=null have been suggested as a way to
2450 * disable console output. Use ttynull that has been created
2451 * for exactly this purpose.
2452 */
2453 if (str[0] == 0 || strcmp(str, "null") == 0) {
2454 __add_preferred_console("ttynull", 0, NULL, NULL, true);
2455 return 1;
2456 }
2457
2458 if (_braille_console_setup(&str, &brl_options))
2459 return 1;
2460
2461 /*
2462 * Decode str into name, index, options.
2463 */
2464 if (str[0] >= '0' && str[0] <= '9') {
2465 strcpy(buf, "ttyS");
2466 strncpy(buf + 4, str, sizeof(buf) - 5);
2467 } else {
2468 strncpy(buf, str, sizeof(buf) - 1);
2469 }
2470 buf[sizeof(buf) - 1] = 0;
2471 options = strchr(str, ',');
2472 if (options)
2473 *(options++) = 0;
2474#ifdef __sparc__
2475 if (!strcmp(str, "ttya"))
2476 strcpy(buf, "ttyS0");
2477 if (!strcmp(str, "ttyb"))
2478 strcpy(buf, "ttyS1");
2479#endif
2480 for (s = buf; *s; s++)
2481 if (isdigit(*s) || *s == ',')
2482 break;
2483 idx = simple_strtoul(s, NULL, 10);
2484 *s = 0;
2485
2486 __add_preferred_console(buf, idx, options, brl_options, true);
2487 return 1;
2488}
2489__setup("console=", console_setup);
2490
2491/**
2492 * add_preferred_console - add a device to the list of preferred consoles.
2493 * @name: device name
2494 * @idx: device index
2495 * @options: options for this console
2496 *
2497 * The last preferred console added will be used for kernel messages
2498 * and stdin/out/err for init. Normally this is used by console_setup
2499 * above to handle user-supplied console arguments; however it can also
2500 * be used by arch-specific code either to override the user or more
2501 * commonly to provide a default console (ie from PROM variables) when
2502 * the user has not supplied one.
2503 */
2504int add_preferred_console(const char *name, const short idx, char *options)
2505{
2506 return __add_preferred_console(name, idx, options, NULL, false);
2507}
2508
2509bool console_suspend_enabled = true;
2510EXPORT_SYMBOL(console_suspend_enabled);
2511
2512static int __init console_suspend_disable(char *str)
2513{
2514 console_suspend_enabled = false;
2515 return 1;
2516}
2517__setup("no_console_suspend", console_suspend_disable);
2518module_param_named(console_suspend, console_suspend_enabled,
2519 bool, S_IRUGO | S_IWUSR);
2520MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2521 " and hibernate operations");
2522
2523static bool printk_console_no_auto_verbose;
2524
2525void console_verbose(void)
2526{
2527 if (console_loglevel && !printk_console_no_auto_verbose)
2528 console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2529}
2530EXPORT_SYMBOL_GPL(console_verbose);
2531
2532module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2533MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2534
2535/**
2536 * suspend_console - suspend the console subsystem
2537 *
2538 * This disables printk() while we go into suspend states
2539 */
2540void suspend_console(void)
2541{
2542 struct console *con;
2543
2544 if (!console_suspend_enabled)
2545 return;
2546 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2547 pr_flush(1000, true);
2548
2549 console_list_lock();
2550 for_each_console(con)
2551 console_srcu_write_flags(con, con->flags | CON_SUSPENDED);
2552 console_list_unlock();
2553
2554 /*
2555 * Ensure that all SRCU list walks have completed. All printing
2556 * contexts must be able to see that they are suspended so that it
2557 * is guaranteed that all printing has stopped when this function
2558 * completes.
2559 */
2560 synchronize_srcu(&console_srcu);
2561}
2562
2563void resume_console(void)
2564{
2565 struct console *con;
2566
2567 if (!console_suspend_enabled)
2568 return;
2569
2570 console_list_lock();
2571 for_each_console(con)
2572 console_srcu_write_flags(con, con->flags & ~CON_SUSPENDED);
2573 console_list_unlock();
2574
2575 /*
2576 * Ensure that all SRCU list walks have completed. All printing
2577 * contexts must be able to see they are no longer suspended so
2578 * that they are guaranteed to wake up and resume printing.
2579 */
2580 synchronize_srcu(&console_srcu);
2581
2582 pr_flush(1000, true);
2583}
2584
2585/**
2586 * console_cpu_notify - print deferred console messages after CPU hotplug
2587 * @cpu: unused
2588 *
2589 * If printk() is called from a CPU that is not online yet, the messages
2590 * will be printed on the console only if there are CON_ANYTIME consoles.
2591 * This function is called when a new CPU comes online (or fails to come
2592 * up) or goes offline.
2593 */
2594static int console_cpu_notify(unsigned int cpu)
2595{
2596 if (!cpuhp_tasks_frozen) {
2597 /* If trylock fails, someone else is doing the printing */
2598 if (console_trylock())
2599 console_unlock();
2600 }
2601 return 0;
2602}
2603
2604/*
2605 * Return true if a panic is in progress on a remote CPU.
2606 *
2607 * On true, the local CPU should immediately release any printing resources
2608 * that may be needed by the panic CPU.
2609 */
2610bool other_cpu_in_panic(void)
2611{
2612 if (!panic_in_progress())
2613 return false;
2614
2615 /*
2616 * We can use raw_smp_processor_id() here because it is impossible for
2617 * the task to be migrated to the panic_cpu, or away from it. If
2618 * panic_cpu has already been set, and we're not currently executing on
2619 * that CPU, then we never will be.
2620 */
2621 return atomic_read(&panic_cpu) != raw_smp_processor_id();
2622}
2623
2624/**
2625 * console_lock - block the console subsystem from printing
2626 *
2627 * Acquires a lock which guarantees that no consoles will
2628 * be in or enter their write() callback.
2629 *
2630 * Can sleep, returns nothing.
2631 */
2632void console_lock(void)
2633{
2634 might_sleep();
2635
2636 /* On panic, the console_lock must be left to the panic cpu. */
2637 while (other_cpu_in_panic())
2638 msleep(1000);
2639
2640 down_console_sem();
2641 console_locked = 1;
2642 console_may_schedule = 1;
2643}
2644EXPORT_SYMBOL(console_lock);
2645
2646/**
2647 * console_trylock - try to block the console subsystem from printing
2648 *
2649 * Try to acquire a lock which guarantees that no consoles will
2650 * be in or enter their write() callback.
2651 *
2652 * returns 1 on success, and 0 on failure to acquire the lock.
2653 */
2654int console_trylock(void)
2655{
2656 /* On panic, the console_lock must be left to the panic cpu. */
2657 if (other_cpu_in_panic())
2658 return 0;
2659 if (down_trylock_console_sem())
2660 return 0;
2661 console_locked = 1;
2662 console_may_schedule = 0;
2663 return 1;
2664}
2665EXPORT_SYMBOL(console_trylock);
2666
2667int is_console_locked(void)
2668{
2669 return console_locked;
2670}
2671EXPORT_SYMBOL(is_console_locked);
2672
2673/*
2674 * Check if the given console is currently capable and allowed to print
2675 * records.
2676 *
2677 * Requires the console_srcu_read_lock.
2678 */
2679static inline bool console_is_usable(struct console *con)
2680{
2681 short flags = console_srcu_read_flags(con);
2682
2683 if (!(flags & CON_ENABLED))
2684 return false;
2685
2686 if ((flags & CON_SUSPENDED))
2687 return false;
2688
2689 if (!con->write)
2690 return false;
2691
2692 /*
2693 * Console drivers may assume that per-cpu resources have been
2694 * allocated. So unless they're explicitly marked as being able to
2695 * cope (CON_ANYTIME) don't call them until this CPU is officially up.
2696 */
2697 if (!cpu_online(raw_smp_processor_id()) && !(flags & CON_ANYTIME))
2698 return false;
2699
2700 return true;
2701}
2702
2703static void __console_unlock(void)
2704{
2705 console_locked = 0;
2706 up_console_sem();
2707}
2708
2709#ifdef CONFIG_PRINTK
2710
2711/*
2712 * Prepend the message in @pmsg->pbufs->outbuf with a "dropped message". This
2713 * is achieved by shifting the existing message over and inserting the dropped
2714 * message.
2715 *
2716 * @pmsg is the printk message to prepend.
2717 *
2718 * @dropped is the dropped count to report in the dropped message.
2719 *
2720 * If the message text in @pmsg->pbufs->outbuf does not have enough space for
2721 * the dropped message, the message text will be sufficiently truncated.
2722 *
2723 * If @pmsg->pbufs->outbuf is modified, @pmsg->outbuf_len is updated.
2724 */
2725void console_prepend_dropped(struct printk_message *pmsg, unsigned long dropped)
2726{
2727 struct printk_buffers *pbufs = pmsg->pbufs;
2728 const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2729 const size_t outbuf_sz = sizeof(pbufs->outbuf);
2730 char *scratchbuf = &pbufs->scratchbuf[0];
2731 char *outbuf = &pbufs->outbuf[0];
2732 size_t len;
2733
2734 len = scnprintf(scratchbuf, scratchbuf_sz,
2735 "** %lu printk messages dropped **\n", dropped);
2736
2737 /*
2738 * Make sure outbuf is sufficiently large before prepending.
2739 * Keep at least the prefix when the message must be truncated.
2740 * It is a rather theoretical problem when someone tries to
2741 * use a minimalist buffer.
2742 */
2743 if (WARN_ON_ONCE(len + PRINTK_PREFIX_MAX >= outbuf_sz))
2744 return;
2745
2746 if (pmsg->outbuf_len + len >= outbuf_sz) {
2747 /* Truncate the message, but keep it terminated. */
2748 pmsg->outbuf_len = outbuf_sz - (len + 1);
2749 outbuf[pmsg->outbuf_len] = 0;
2750 }
2751
2752 memmove(outbuf + len, outbuf, pmsg->outbuf_len + 1);
2753 memcpy(outbuf, scratchbuf, len);
2754 pmsg->outbuf_len += len;
2755}
2756
2757/*
2758 * Read and format the specified record (or a later record if the specified
2759 * record is not available).
2760 *
2761 * @pmsg will contain the formatted result. @pmsg->pbufs must point to a
2762 * struct printk_buffers.
2763 *
2764 * @seq is the record to read and format. If it is not available, the next
2765 * valid record is read.
2766 *
2767 * @is_extended specifies if the message should be formatted for extended
2768 * console output.
2769 *
2770 * @may_supress specifies if records may be skipped based on loglevel.
2771 *
2772 * Returns false if no record is available. Otherwise true and all fields
2773 * of @pmsg are valid. (See the documentation of struct printk_message
2774 * for information about the @pmsg fields.)
2775 */
2776bool printk_get_next_message(struct printk_message *pmsg, u64 seq,
2777 bool is_extended, bool may_suppress)
2778{
2779 static int panic_console_dropped;
2780
2781 struct printk_buffers *pbufs = pmsg->pbufs;
2782 const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2783 const size_t outbuf_sz = sizeof(pbufs->outbuf);
2784 char *scratchbuf = &pbufs->scratchbuf[0];
2785 char *outbuf = &pbufs->outbuf[0];
2786 struct printk_info info;
2787 struct printk_record r;
2788 size_t len = 0;
2789
2790 /*
2791 * Formatting extended messages requires a separate buffer, so use the
2792 * scratch buffer to read in the ringbuffer text.
2793 *
2794 * Formatting normal messages is done in-place, so read the ringbuffer
2795 * text directly into the output buffer.
2796 */
2797 if (is_extended)
2798 prb_rec_init_rd(&r, &info, scratchbuf, scratchbuf_sz);
2799 else
2800 prb_rec_init_rd(&r, &info, outbuf, outbuf_sz);
2801
2802 if (!prb_read_valid(prb, seq, &r))
2803 return false;
2804
2805 pmsg->seq = r.info->seq;
2806 pmsg->dropped = r.info->seq - seq;
2807
2808 /*
2809 * Check for dropped messages in panic here so that printk
2810 * suppression can occur as early as possible if necessary.
2811 */
2812 if (pmsg->dropped &&
2813 panic_in_progress() &&
2814 panic_console_dropped++ > 10) {
2815 suppress_panic_printk = 1;
2816 pr_warn_once("Too many dropped messages. Suppress messages on non-panic CPUs to prevent livelock.\n");
2817 }
2818
2819 /* Skip record that has level above the console loglevel. */
2820 if (may_suppress && suppress_message_printing(r.info->level))
2821 goto out;
2822
2823 if (is_extended) {
2824 len = info_print_ext_header(outbuf, outbuf_sz, r.info);
2825 len += msg_print_ext_body(outbuf + len, outbuf_sz - len,
2826 &r.text_buf[0], r.info->text_len, &r.info->dev_info);
2827 } else {
2828 len = record_print_text(&r, console_msg_format & MSG_FORMAT_SYSLOG, printk_time);
2829 }
2830out:
2831 pmsg->outbuf_len = len;
2832 return true;
2833}
2834
2835/*
2836 * Used as the printk buffers for non-panic, serialized console printing.
2837 * This is for legacy (!CON_NBCON) as well as all boot (CON_BOOT) consoles.
2838 * Its usage requires the console_lock held.
2839 */
2840struct printk_buffers printk_shared_pbufs;
2841
2842/*
2843 * Print one record for the given console. The record printed is whatever
2844 * record is the next available record for the given console.
2845 *
2846 * @handover will be set to true if a printk waiter has taken over the
2847 * console_lock, in which case the caller is no longer holding both the
2848 * console_lock and the SRCU read lock. Otherwise it is set to false.
2849 *
2850 * @cookie is the cookie from the SRCU read lock.
2851 *
2852 * Returns false if the given console has no next record to print, otherwise
2853 * true.
2854 *
2855 * Requires the console_lock and the SRCU read lock.
2856 */
2857static bool console_emit_next_record(struct console *con, bool *handover, int cookie)
2858{
2859 bool is_extended = console_srcu_read_flags(con) & CON_EXTENDED;
2860 char *outbuf = &printk_shared_pbufs.outbuf[0];
2861 struct printk_message pmsg = {
2862 .pbufs = &printk_shared_pbufs,
2863 };
2864 unsigned long flags;
2865
2866 *handover = false;
2867
2868 if (!printk_get_next_message(&pmsg, con->seq, is_extended, true))
2869 return false;
2870
2871 con->dropped += pmsg.dropped;
2872
2873 /* Skip messages of formatted length 0. */
2874 if (pmsg.outbuf_len == 0) {
2875 con->seq = pmsg.seq + 1;
2876 goto skip;
2877 }
2878
2879 if (con->dropped && !is_extended) {
2880 console_prepend_dropped(&pmsg, con->dropped);
2881 con->dropped = 0;
2882 }
2883
2884 /*
2885 * While actively printing out messages, if another printk()
2886 * were to occur on another CPU, it may wait for this one to
2887 * finish. This task can not be preempted if there is a
2888 * waiter waiting to take over.
2889 *
2890 * Interrupts are disabled because the hand over to a waiter
2891 * must not be interrupted until the hand over is completed
2892 * (@console_waiter is cleared).
2893 */
2894 printk_safe_enter_irqsave(flags);
2895 console_lock_spinning_enable();
2896
2897 /* Do not trace print latency. */
2898 stop_critical_timings();
2899
2900 /* Write everything out to the hardware. */
2901 con->write(con, outbuf, pmsg.outbuf_len);
2902
2903 start_critical_timings();
2904
2905 con->seq = pmsg.seq + 1;
2906
2907 *handover = console_lock_spinning_disable_and_check(cookie);
2908 printk_safe_exit_irqrestore(flags);
2909skip:
2910 return true;
2911}
2912
2913#else
2914
2915static bool console_emit_next_record(struct console *con, bool *handover, int cookie)
2916{
2917 *handover = false;
2918 return false;
2919}
2920
2921#endif /* CONFIG_PRINTK */
2922
2923/*
2924 * Print out all remaining records to all consoles.
2925 *
2926 * @do_cond_resched is set by the caller. It can be true only in schedulable
2927 * context.
2928 *
2929 * @next_seq is set to the sequence number after the last available record.
2930 * The value is valid only when this function returns true. It means that all
2931 * usable consoles are completely flushed.
2932 *
2933 * @handover will be set to true if a printk waiter has taken over the
2934 * console_lock, in which case the caller is no longer holding the
2935 * console_lock. Otherwise it is set to false.
2936 *
2937 * Returns true when there was at least one usable console and all messages
2938 * were flushed to all usable consoles. A returned false informs the caller
2939 * that everything was not flushed (either there were no usable consoles or
2940 * another context has taken over printing or it is a panic situation and this
2941 * is not the panic CPU). Regardless the reason, the caller should assume it
2942 * is not useful to immediately try again.
2943 *
2944 * Requires the console_lock.
2945 */
2946static bool console_flush_all(bool do_cond_resched, u64 *next_seq, bool *handover)
2947{
2948 bool any_usable = false;
2949 struct console *con;
2950 bool any_progress;
2951 int cookie;
2952
2953 *next_seq = 0;
2954 *handover = false;
2955
2956 do {
2957 any_progress = false;
2958
2959 cookie = console_srcu_read_lock();
2960 for_each_console_srcu(con) {
2961 bool progress;
2962
2963 if (!console_is_usable(con))
2964 continue;
2965 any_usable = true;
2966
2967 progress = console_emit_next_record(con, handover, cookie);
2968
2969 /*
2970 * If a handover has occurred, the SRCU read lock
2971 * is already released.
2972 */
2973 if (*handover)
2974 return false;
2975
2976 /* Track the next of the highest seq flushed. */
2977 if (con->seq > *next_seq)
2978 *next_seq = con->seq;
2979
2980 if (!progress)
2981 continue;
2982 any_progress = true;
2983
2984 /* Allow panic_cpu to take over the consoles safely. */
2985 if (other_cpu_in_panic())
2986 goto abandon;
2987
2988 if (do_cond_resched)
2989 cond_resched();
2990 }
2991 console_srcu_read_unlock(cookie);
2992 } while (any_progress);
2993
2994 return any_usable;
2995
2996abandon:
2997 console_srcu_read_unlock(cookie);
2998 return false;
2999}
3000
3001/**
3002 * console_unlock - unblock the console subsystem from printing
3003 *
3004 * Releases the console_lock which the caller holds to block printing of
3005 * the console subsystem.
3006 *
3007 * While the console_lock was held, console output may have been buffered
3008 * by printk(). If this is the case, console_unlock(); emits
3009 * the output prior to releasing the lock.
3010 *
3011 * console_unlock(); may be called from any context.
3012 */
3013void console_unlock(void)
3014{
3015 bool do_cond_resched;
3016 bool handover;
3017 bool flushed;
3018 u64 next_seq;
3019
3020 /*
3021 * Console drivers are called with interrupts disabled, so
3022 * @console_may_schedule should be cleared before; however, we may
3023 * end up dumping a lot of lines, for example, if called from
3024 * console registration path, and should invoke cond_resched()
3025 * between lines if allowable. Not doing so can cause a very long
3026 * scheduling stall on a slow console leading to RCU stall and
3027 * softlockup warnings which exacerbate the issue with more
3028 * messages practically incapacitating the system. Therefore, create
3029 * a local to use for the printing loop.
3030 */
3031 do_cond_resched = console_may_schedule;
3032
3033 do {
3034 console_may_schedule = 0;
3035
3036 flushed = console_flush_all(do_cond_resched, &next_seq, &handover);
3037 if (!handover)
3038 __console_unlock();
3039
3040 /*
3041 * Abort if there was a failure to flush all messages to all
3042 * usable consoles. Either it is not possible to flush (in
3043 * which case it would be an infinite loop of retrying) or
3044 * another context has taken over printing.
3045 */
3046 if (!flushed)
3047 break;
3048
3049 /*
3050 * Some context may have added new records after
3051 * console_flush_all() but before unlocking the console.
3052 * Re-check if there is a new record to flush. If the trylock
3053 * fails, another context is already handling the printing.
3054 */
3055 } while (prb_read_valid(prb, next_seq, NULL) && console_trylock());
3056}
3057EXPORT_SYMBOL(console_unlock);
3058
3059/**
3060 * console_conditional_schedule - yield the CPU if required
3061 *
3062 * If the console code is currently allowed to sleep, and
3063 * if this CPU should yield the CPU to another task, do
3064 * so here.
3065 *
3066 * Must be called within console_lock();.
3067 */
3068void __sched console_conditional_schedule(void)
3069{
3070 if (console_may_schedule)
3071 cond_resched();
3072}
3073EXPORT_SYMBOL(console_conditional_schedule);
3074
3075void console_unblank(void)
3076{
3077 bool found_unblank = false;
3078 struct console *c;
3079 int cookie;
3080
3081 /*
3082 * First check if there are any consoles implementing the unblank()
3083 * callback. If not, there is no reason to continue and take the
3084 * console lock, which in particular can be dangerous if
3085 * @oops_in_progress is set.
3086 */
3087 cookie = console_srcu_read_lock();
3088 for_each_console_srcu(c) {
3089 if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank) {
3090 found_unblank = true;
3091 break;
3092 }
3093 }
3094 console_srcu_read_unlock(cookie);
3095 if (!found_unblank)
3096 return;
3097
3098 /*
3099 * Stop console printing because the unblank() callback may
3100 * assume the console is not within its write() callback.
3101 *
3102 * If @oops_in_progress is set, this may be an atomic context.
3103 * In that case, attempt a trylock as best-effort.
3104 */
3105 if (oops_in_progress) {
3106 /* Semaphores are not NMI-safe. */
3107 if (in_nmi())
3108 return;
3109
3110 /*
3111 * Attempting to trylock the console lock can deadlock
3112 * if another CPU was stopped while modifying the
3113 * semaphore. "Hope and pray" that this is not the
3114 * current situation.
3115 */
3116 if (down_trylock_console_sem() != 0)
3117 return;
3118 } else
3119 console_lock();
3120
3121 console_locked = 1;
3122 console_may_schedule = 0;
3123
3124 cookie = console_srcu_read_lock();
3125 for_each_console_srcu(c) {
3126 if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank)
3127 c->unblank();
3128 }
3129 console_srcu_read_unlock(cookie);
3130
3131 console_unlock();
3132
3133 if (!oops_in_progress)
3134 pr_flush(1000, true);
3135}
3136
3137/**
3138 * console_flush_on_panic - flush console content on panic
3139 * @mode: flush all messages in buffer or just the pending ones
3140 *
3141 * Immediately output all pending messages no matter what.
3142 */
3143void console_flush_on_panic(enum con_flush_mode mode)
3144{
3145 bool handover;
3146 u64 next_seq;
3147
3148 /*
3149 * Ignore the console lock and flush out the messages. Attempting a
3150 * trylock would not be useful because:
3151 *
3152 * - if it is contended, it must be ignored anyway
3153 * - console_lock() and console_trylock() block and fail
3154 * respectively in panic for non-panic CPUs
3155 * - semaphores are not NMI-safe
3156 */
3157
3158 /*
3159 * If another context is holding the console lock,
3160 * @console_may_schedule might be set. Clear it so that
3161 * this context does not call cond_resched() while flushing.
3162 */
3163 console_may_schedule = 0;
3164
3165 if (mode == CONSOLE_REPLAY_ALL) {
3166 struct console *c;
3167 short flags;
3168 int cookie;
3169 u64 seq;
3170
3171 seq = prb_first_valid_seq(prb);
3172
3173 cookie = console_srcu_read_lock();
3174 for_each_console_srcu(c) {
3175 flags = console_srcu_read_flags(c);
3176
3177 if (flags & CON_NBCON) {
3178 nbcon_seq_force(c, seq);
3179 } else {
3180 /*
3181 * This is an unsynchronized assignment. On
3182 * panic legacy consoles are only best effort.
3183 */
3184 c->seq = seq;
3185 }
3186 }
3187 console_srcu_read_unlock(cookie);
3188 }
3189
3190 console_flush_all(false, &next_seq, &handover);
3191}
3192
3193/*
3194 * Return the console tty driver structure and its associated index
3195 */
3196struct tty_driver *console_device(int *index)
3197{
3198 struct console *c;
3199 struct tty_driver *driver = NULL;
3200 int cookie;
3201
3202 /*
3203 * Take console_lock to serialize device() callback with
3204 * other console operations. For example, fg_console is
3205 * modified under console_lock when switching vt.
3206 */
3207 console_lock();
3208
3209 cookie = console_srcu_read_lock();
3210 for_each_console_srcu(c) {
3211 if (!c->device)
3212 continue;
3213 driver = c->device(c, index);
3214 if (driver)
3215 break;
3216 }
3217 console_srcu_read_unlock(cookie);
3218
3219 console_unlock();
3220 return driver;
3221}
3222
3223/*
3224 * Prevent further output on the passed console device so that (for example)
3225 * serial drivers can disable console output before suspending a port, and can
3226 * re-enable output afterwards.
3227 */
3228void console_stop(struct console *console)
3229{
3230 __pr_flush(console, 1000, true);
3231 console_list_lock();
3232 console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3233 console_list_unlock();
3234
3235 /*
3236 * Ensure that all SRCU list walks have completed. All contexts must
3237 * be able to see that this console is disabled so that (for example)
3238 * the caller can suspend the port without risk of another context
3239 * using the port.
3240 */
3241 synchronize_srcu(&console_srcu);
3242}
3243EXPORT_SYMBOL(console_stop);
3244
3245void console_start(struct console *console)
3246{
3247 console_list_lock();
3248 console_srcu_write_flags(console, console->flags | CON_ENABLED);
3249 console_list_unlock();
3250 __pr_flush(console, 1000, true);
3251}
3252EXPORT_SYMBOL(console_start);
3253
3254static int __read_mostly keep_bootcon;
3255
3256static int __init keep_bootcon_setup(char *str)
3257{
3258 keep_bootcon = 1;
3259 pr_info("debug: skip boot console de-registration.\n");
3260
3261 return 0;
3262}
3263
3264early_param("keep_bootcon", keep_bootcon_setup);
3265
3266/*
3267 * This is called by register_console() to try to match
3268 * the newly registered console with any of the ones selected
3269 * by either the command line or add_preferred_console() and
3270 * setup/enable it.
3271 *
3272 * Care need to be taken with consoles that are statically
3273 * enabled such as netconsole
3274 */
3275static int try_enable_preferred_console(struct console *newcon,
3276 bool user_specified)
3277{
3278 struct console_cmdline *c;
3279 int i, err;
3280
3281 for (i = 0, c = console_cmdline;
3282 i < MAX_CMDLINECONSOLES && c->name[0];
3283 i++, c++) {
3284 if (c->user_specified != user_specified)
3285 continue;
3286 if (!newcon->match ||
3287 newcon->match(newcon, c->name, c->index, c->options) != 0) {
3288 /* default matching */
3289 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
3290 if (strcmp(c->name, newcon->name) != 0)
3291 continue;
3292 if (newcon->index >= 0 &&
3293 newcon->index != c->index)
3294 continue;
3295 if (newcon->index < 0)
3296 newcon->index = c->index;
3297
3298 if (_braille_register_console(newcon, c))
3299 return 0;
3300
3301 if (newcon->setup &&
3302 (err = newcon->setup(newcon, c->options)) != 0)
3303 return err;
3304 }
3305 newcon->flags |= CON_ENABLED;
3306 if (i == preferred_console)
3307 newcon->flags |= CON_CONSDEV;
3308 return 0;
3309 }
3310
3311 /*
3312 * Some consoles, such as pstore and netconsole, can be enabled even
3313 * without matching. Accept the pre-enabled consoles only when match()
3314 * and setup() had a chance to be called.
3315 */
3316 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
3317 return 0;
3318
3319 return -ENOENT;
3320}
3321
3322/* Try to enable the console unconditionally */
3323static void try_enable_default_console(struct console *newcon)
3324{
3325 if (newcon->index < 0)
3326 newcon->index = 0;
3327
3328 if (newcon->setup && newcon->setup(newcon, NULL) != 0)
3329 return;
3330
3331 newcon->flags |= CON_ENABLED;
3332
3333 if (newcon->device)
3334 newcon->flags |= CON_CONSDEV;
3335}
3336
3337static void console_init_seq(struct console *newcon, bool bootcon_registered)
3338{
3339 struct console *con;
3340 bool handover;
3341
3342 if (newcon->flags & (CON_PRINTBUFFER | CON_BOOT)) {
3343 /* Get a consistent copy of @syslog_seq. */
3344 mutex_lock(&syslog_lock);
3345 newcon->seq = syslog_seq;
3346 mutex_unlock(&syslog_lock);
3347 } else {
3348 /* Begin with next message added to ringbuffer. */
3349 newcon->seq = prb_next_seq(prb);
3350
3351 /*
3352 * If any enabled boot consoles are due to be unregistered
3353 * shortly, some may not be caught up and may be the same
3354 * device as @newcon. Since it is not known which boot console
3355 * is the same device, flush all consoles and, if necessary,
3356 * start with the message of the enabled boot console that is
3357 * the furthest behind.
3358 */
3359 if (bootcon_registered && !keep_bootcon) {
3360 /*
3361 * Hold the console_lock to stop console printing and
3362 * guarantee safe access to console->seq.
3363 */
3364 console_lock();
3365
3366 /*
3367 * Flush all consoles and set the console to start at
3368 * the next unprinted sequence number.
3369 */
3370 if (!console_flush_all(true, &newcon->seq, &handover)) {
3371 /*
3372 * Flushing failed. Just choose the lowest
3373 * sequence of the enabled boot consoles.
3374 */
3375
3376 /*
3377 * If there was a handover, this context no
3378 * longer holds the console_lock.
3379 */
3380 if (handover)
3381 console_lock();
3382
3383 newcon->seq = prb_next_seq(prb);
3384 for_each_console(con) {
3385 if ((con->flags & CON_BOOT) &&
3386 (con->flags & CON_ENABLED) &&
3387 con->seq < newcon->seq) {
3388 newcon->seq = con->seq;
3389 }
3390 }
3391 }
3392
3393 console_unlock();
3394 }
3395 }
3396}
3397
3398#define console_first() \
3399 hlist_entry(console_list.first, struct console, node)
3400
3401static int unregister_console_locked(struct console *console);
3402
3403/*
3404 * The console driver calls this routine during kernel initialization
3405 * to register the console printing procedure with printk() and to
3406 * print any messages that were printed by the kernel before the
3407 * console driver was initialized.
3408 *
3409 * This can happen pretty early during the boot process (because of
3410 * early_printk) - sometimes before setup_arch() completes - be careful
3411 * of what kernel features are used - they may not be initialised yet.
3412 *
3413 * There are two types of consoles - bootconsoles (early_printk) and
3414 * "real" consoles (everything which is not a bootconsole) which are
3415 * handled differently.
3416 * - Any number of bootconsoles can be registered at any time.
3417 * - As soon as a "real" console is registered, all bootconsoles
3418 * will be unregistered automatically.
3419 * - Once a "real" console is registered, any attempt to register a
3420 * bootconsoles will be rejected
3421 */
3422void register_console(struct console *newcon)
3423{
3424 struct console *con;
3425 bool bootcon_registered = false;
3426 bool realcon_registered = false;
3427 int err;
3428
3429 console_list_lock();
3430
3431 for_each_console(con) {
3432 if (WARN(con == newcon, "console '%s%d' already registered\n",
3433 con->name, con->index)) {
3434 goto unlock;
3435 }
3436
3437 if (con->flags & CON_BOOT)
3438 bootcon_registered = true;
3439 else
3440 realcon_registered = true;
3441 }
3442
3443 /* Do not register boot consoles when there already is a real one. */
3444 if ((newcon->flags & CON_BOOT) && realcon_registered) {
3445 pr_info("Too late to register bootconsole %s%d\n",
3446 newcon->name, newcon->index);
3447 goto unlock;
3448 }
3449
3450 if (newcon->flags & CON_NBCON) {
3451 /*
3452 * Ensure the nbcon console buffers can be allocated
3453 * before modifying any global data.
3454 */
3455 if (!nbcon_alloc(newcon))
3456 goto unlock;
3457 }
3458
3459 /*
3460 * See if we want to enable this console driver by default.
3461 *
3462 * Nope when a console is preferred by the command line, device
3463 * tree, or SPCR.
3464 *
3465 * The first real console with tty binding (driver) wins. More
3466 * consoles might get enabled before the right one is found.
3467 *
3468 * Note that a console with tty binding will have CON_CONSDEV
3469 * flag set and will be first in the list.
3470 */
3471 if (preferred_console < 0) {
3472 if (hlist_empty(&console_list) || !console_first()->device ||
3473 console_first()->flags & CON_BOOT) {
3474 try_enable_default_console(newcon);
3475 }
3476 }
3477
3478 /* See if this console matches one we selected on the command line */
3479 err = try_enable_preferred_console(newcon, true);
3480
3481 /* If not, try to match against the platform default(s) */
3482 if (err == -ENOENT)
3483 err = try_enable_preferred_console(newcon, false);
3484
3485 /* printk() messages are not printed to the Braille console. */
3486 if (err || newcon->flags & CON_BRL) {
3487 if (newcon->flags & CON_NBCON)
3488 nbcon_free(newcon);
3489 goto unlock;
3490 }
3491
3492 /*
3493 * If we have a bootconsole, and are switching to a real console,
3494 * don't print everything out again, since when the boot console, and
3495 * the real console are the same physical device, it's annoying to
3496 * see the beginning boot messages twice
3497 */
3498 if (bootcon_registered &&
3499 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
3500 newcon->flags &= ~CON_PRINTBUFFER;
3501 }
3502
3503 newcon->dropped = 0;
3504 console_init_seq(newcon, bootcon_registered);
3505
3506 if (newcon->flags & CON_NBCON)
3507 nbcon_init(newcon);
3508
3509 /*
3510 * Put this console in the list - keep the
3511 * preferred driver at the head of the list.
3512 */
3513 if (hlist_empty(&console_list)) {
3514 /* Ensure CON_CONSDEV is always set for the head. */
3515 newcon->flags |= CON_CONSDEV;
3516 hlist_add_head_rcu(&newcon->node, &console_list);
3517
3518 } else if (newcon->flags & CON_CONSDEV) {
3519 /* Only the new head can have CON_CONSDEV set. */
3520 console_srcu_write_flags(console_first(), console_first()->flags & ~CON_CONSDEV);
3521 hlist_add_head_rcu(&newcon->node, &console_list);
3522
3523 } else {
3524 hlist_add_behind_rcu(&newcon->node, console_list.first);
3525 }
3526
3527 /*
3528 * No need to synchronize SRCU here! The caller does not rely
3529 * on all contexts being able to see the new console before
3530 * register_console() completes.
3531 */
3532
3533 console_sysfs_notify();
3534
3535 /*
3536 * By unregistering the bootconsoles after we enable the real console
3537 * we get the "console xxx enabled" message on all the consoles -
3538 * boot consoles, real consoles, etc - this is to ensure that end
3539 * users know there might be something in the kernel's log buffer that
3540 * went to the bootconsole (that they do not see on the real console)
3541 */
3542 con_printk(KERN_INFO, newcon, "enabled\n");
3543 if (bootcon_registered &&
3544 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
3545 !keep_bootcon) {
3546 struct hlist_node *tmp;
3547
3548 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3549 if (con->flags & CON_BOOT)
3550 unregister_console_locked(con);
3551 }
3552 }
3553unlock:
3554 console_list_unlock();
3555}
3556EXPORT_SYMBOL(register_console);
3557
3558/* Must be called under console_list_lock(). */
3559static int unregister_console_locked(struct console *console)
3560{
3561 int res;
3562
3563 lockdep_assert_console_list_lock_held();
3564
3565 con_printk(KERN_INFO, console, "disabled\n");
3566
3567 res = _braille_unregister_console(console);
3568 if (res < 0)
3569 return res;
3570 if (res > 0)
3571 return 0;
3572
3573 /* Disable it unconditionally */
3574 console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3575
3576 if (!console_is_registered_locked(console))
3577 return -ENODEV;
3578
3579 hlist_del_init_rcu(&console->node);
3580
3581 /*
3582 * <HISTORICAL>
3583 * If this isn't the last console and it has CON_CONSDEV set, we
3584 * need to set it on the next preferred console.
3585 * </HISTORICAL>
3586 *
3587 * The above makes no sense as there is no guarantee that the next
3588 * console has any device attached. Oh well....
3589 */
3590 if (!hlist_empty(&console_list) && console->flags & CON_CONSDEV)
3591 console_srcu_write_flags(console_first(), console_first()->flags | CON_CONSDEV);
3592
3593 /*
3594 * Ensure that all SRCU list walks have completed. All contexts
3595 * must not be able to see this console in the list so that any
3596 * exit/cleanup routines can be performed safely.
3597 */
3598 synchronize_srcu(&console_srcu);
3599
3600 if (console->flags & CON_NBCON)
3601 nbcon_free(console);
3602
3603 console_sysfs_notify();
3604
3605 if (console->exit)
3606 res = console->exit(console);
3607
3608 return res;
3609}
3610
3611int unregister_console(struct console *console)
3612{
3613 int res;
3614
3615 console_list_lock();
3616 res = unregister_console_locked(console);
3617 console_list_unlock();
3618 return res;
3619}
3620EXPORT_SYMBOL(unregister_console);
3621
3622/**
3623 * console_force_preferred_locked - force a registered console preferred
3624 * @con: The registered console to force preferred.
3625 *
3626 * Must be called under console_list_lock().
3627 */
3628void console_force_preferred_locked(struct console *con)
3629{
3630 struct console *cur_pref_con;
3631
3632 if (!console_is_registered_locked(con))
3633 return;
3634
3635 cur_pref_con = console_first();
3636
3637 /* Already preferred? */
3638 if (cur_pref_con == con)
3639 return;
3640
3641 /*
3642 * Delete, but do not re-initialize the entry. This allows the console
3643 * to continue to appear registered (via any hlist_unhashed_lockless()
3644 * checks), even though it was briefly removed from the console list.
3645 */
3646 hlist_del_rcu(&con->node);
3647
3648 /*
3649 * Ensure that all SRCU list walks have completed so that the console
3650 * can be added to the beginning of the console list and its forward
3651 * list pointer can be re-initialized.
3652 */
3653 synchronize_srcu(&console_srcu);
3654
3655 con->flags |= CON_CONSDEV;
3656 WARN_ON(!con->device);
3657
3658 /* Only the new head can have CON_CONSDEV set. */
3659 console_srcu_write_flags(cur_pref_con, cur_pref_con->flags & ~CON_CONSDEV);
3660 hlist_add_head_rcu(&con->node, &console_list);
3661}
3662EXPORT_SYMBOL(console_force_preferred_locked);
3663
3664/*
3665 * Initialize the console device. This is called *early*, so
3666 * we can't necessarily depend on lots of kernel help here.
3667 * Just do some early initializations, and do the complex setup
3668 * later.
3669 */
3670void __init console_init(void)
3671{
3672 int ret;
3673 initcall_t call;
3674 initcall_entry_t *ce;
3675
3676 /* Setup the default TTY line discipline. */
3677 n_tty_init();
3678
3679 /*
3680 * set up the console device so that later boot sequences can
3681 * inform about problems etc..
3682 */
3683 ce = __con_initcall_start;
3684 trace_initcall_level("console");
3685 while (ce < __con_initcall_end) {
3686 call = initcall_from_entry(ce);
3687 trace_initcall_start(call);
3688 ret = call();
3689 trace_initcall_finish(call, ret);
3690 ce++;
3691 }
3692}
3693
3694/*
3695 * Some boot consoles access data that is in the init section and which will
3696 * be discarded after the initcalls have been run. To make sure that no code
3697 * will access this data, unregister the boot consoles in a late initcall.
3698 *
3699 * If for some reason, such as deferred probe or the driver being a loadable
3700 * module, the real console hasn't registered yet at this point, there will
3701 * be a brief interval in which no messages are logged to the console, which
3702 * makes it difficult to diagnose problems that occur during this time.
3703 *
3704 * To mitigate this problem somewhat, only unregister consoles whose memory
3705 * intersects with the init section. Note that all other boot consoles will
3706 * get unregistered when the real preferred console is registered.
3707 */
3708static int __init printk_late_init(void)
3709{
3710 struct hlist_node *tmp;
3711 struct console *con;
3712 int ret;
3713
3714 console_list_lock();
3715 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3716 if (!(con->flags & CON_BOOT))
3717 continue;
3718
3719 /* Check addresses that might be used for enabled consoles. */
3720 if (init_section_intersects(con, sizeof(*con)) ||
3721 init_section_contains(con->write, 0) ||
3722 init_section_contains(con->read, 0) ||
3723 init_section_contains(con->device, 0) ||
3724 init_section_contains(con->unblank, 0) ||
3725 init_section_contains(con->data, 0)) {
3726 /*
3727 * Please, consider moving the reported consoles out
3728 * of the init section.
3729 */
3730 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3731 con->name, con->index);
3732 unregister_console_locked(con);
3733 }
3734 }
3735 console_list_unlock();
3736
3737 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3738 console_cpu_notify);
3739 WARN_ON(ret < 0);
3740 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3741 console_cpu_notify, NULL);
3742 WARN_ON(ret < 0);
3743 printk_sysctl_init();
3744 return 0;
3745}
3746late_initcall(printk_late_init);
3747
3748#if defined CONFIG_PRINTK
3749/* If @con is specified, only wait for that console. Otherwise wait for all. */
3750static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
3751{
3752 unsigned long timeout_jiffies = msecs_to_jiffies(timeout_ms);
3753 unsigned long remaining_jiffies = timeout_jiffies;
3754 struct console *c;
3755 u64 last_diff = 0;
3756 u64 printk_seq;
3757 short flags;
3758 int cookie;
3759 u64 diff;
3760 u64 seq;
3761
3762 might_sleep();
3763
3764 seq = prb_next_seq(prb);
3765
3766 /* Flush the consoles so that records up to @seq are printed. */
3767 console_lock();
3768 console_unlock();
3769
3770 for (;;) {
3771 unsigned long begin_jiffies;
3772 unsigned long slept_jiffies;
3773
3774 diff = 0;
3775
3776 /*
3777 * Hold the console_lock to guarantee safe access to
3778 * console->seq. Releasing console_lock flushes more
3779 * records in case @seq is still not printed on all
3780 * usable consoles.
3781 */
3782 console_lock();
3783
3784 cookie = console_srcu_read_lock();
3785 for_each_console_srcu(c) {
3786 if (con && con != c)
3787 continue;
3788
3789 flags = console_srcu_read_flags(c);
3790
3791 /*
3792 * If consoles are not usable, it cannot be expected
3793 * that they make forward progress, so only increment
3794 * @diff for usable consoles.
3795 */
3796 if (!console_is_usable(c))
3797 continue;
3798
3799 if (flags & CON_NBCON) {
3800 printk_seq = nbcon_seq_read(c);
3801 } else {
3802 printk_seq = c->seq;
3803 }
3804
3805 if (printk_seq < seq)
3806 diff += seq - printk_seq;
3807 }
3808 console_srcu_read_unlock(cookie);
3809
3810 if (diff != last_diff && reset_on_progress)
3811 remaining_jiffies = timeout_jiffies;
3812
3813 console_unlock();
3814
3815 /* Note: @diff is 0 if there are no usable consoles. */
3816 if (diff == 0 || remaining_jiffies == 0)
3817 break;
3818
3819 /* msleep(1) might sleep much longer. Check time by jiffies. */
3820 begin_jiffies = jiffies;
3821 msleep(1);
3822 slept_jiffies = jiffies - begin_jiffies;
3823
3824 remaining_jiffies -= min(slept_jiffies, remaining_jiffies);
3825
3826 last_diff = diff;
3827 }
3828
3829 return (diff == 0);
3830}
3831
3832/**
3833 * pr_flush() - Wait for printing threads to catch up.
3834 *
3835 * @timeout_ms: The maximum time (in ms) to wait.
3836 * @reset_on_progress: Reset the timeout if forward progress is seen.
3837 *
3838 * A value of 0 for @timeout_ms means no waiting will occur. A value of -1
3839 * represents infinite waiting.
3840 *
3841 * If @reset_on_progress is true, the timeout will be reset whenever any
3842 * printer has been seen to make some forward progress.
3843 *
3844 * Context: Process context. May sleep while acquiring console lock.
3845 * Return: true if all usable printers are caught up.
3846 */
3847static bool pr_flush(int timeout_ms, bool reset_on_progress)
3848{
3849 return __pr_flush(NULL, timeout_ms, reset_on_progress);
3850}
3851
3852/*
3853 * Delayed printk version, for scheduler-internal messages:
3854 */
3855#define PRINTK_PENDING_WAKEUP 0x01
3856#define PRINTK_PENDING_OUTPUT 0x02
3857
3858static DEFINE_PER_CPU(int, printk_pending);
3859
3860static void wake_up_klogd_work_func(struct irq_work *irq_work)
3861{
3862 int pending = this_cpu_xchg(printk_pending, 0);
3863
3864 if (pending & PRINTK_PENDING_OUTPUT) {
3865 /* If trylock fails, someone else is doing the printing */
3866 if (console_trylock())
3867 console_unlock();
3868 }
3869
3870 if (pending & PRINTK_PENDING_WAKEUP)
3871 wake_up_interruptible(&log_wait);
3872}
3873
3874static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3875 IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3876
3877static void __wake_up_klogd(int val)
3878{
3879 if (!printk_percpu_data_ready())
3880 return;
3881
3882 preempt_disable();
3883 /*
3884 * Guarantee any new records can be seen by tasks preparing to wait
3885 * before this context checks if the wait queue is empty.
3886 *
3887 * The full memory barrier within wq_has_sleeper() pairs with the full
3888 * memory barrier within set_current_state() of
3889 * prepare_to_wait_event(), which is called after ___wait_event() adds
3890 * the waiter but before it has checked the wait condition.
3891 *
3892 * This pairs with devkmsg_read:A and syslog_print:A.
3893 */
3894 if (wq_has_sleeper(&log_wait) || /* LMM(__wake_up_klogd:A) */
3895 (val & PRINTK_PENDING_OUTPUT)) {
3896 this_cpu_or(printk_pending, val);
3897 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3898 }
3899 preempt_enable();
3900}
3901
3902/**
3903 * wake_up_klogd - Wake kernel logging daemon
3904 *
3905 * Use this function when new records have been added to the ringbuffer
3906 * and the console printing of those records has already occurred or is
3907 * known to be handled by some other context. This function will only
3908 * wake the logging daemon.
3909 *
3910 * Context: Any context.
3911 */
3912void wake_up_klogd(void)
3913{
3914 __wake_up_klogd(PRINTK_PENDING_WAKEUP);
3915}
3916
3917/**
3918 * defer_console_output - Wake kernel logging daemon and trigger
3919 * console printing in a deferred context
3920 *
3921 * Use this function when new records have been added to the ringbuffer,
3922 * this context is responsible for console printing those records, but
3923 * the current context is not allowed to perform the console printing.
3924 * Trigger an irq_work context to perform the console printing. This
3925 * function also wakes the logging daemon.
3926 *
3927 * Context: Any context.
3928 */
3929void defer_console_output(void)
3930{
3931 /*
3932 * New messages may have been added directly to the ringbuffer
3933 * using vprintk_store(), so wake any waiters as well.
3934 */
3935 __wake_up_klogd(PRINTK_PENDING_WAKEUP | PRINTK_PENDING_OUTPUT);
3936}
3937
3938void printk_trigger_flush(void)
3939{
3940 defer_console_output();
3941}
3942
3943int vprintk_deferred(const char *fmt, va_list args)
3944{
3945 return vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3946}
3947
3948int _printk_deferred(const char *fmt, ...)
3949{
3950 va_list args;
3951 int r;
3952
3953 va_start(args, fmt);
3954 r = vprintk_deferred(fmt, args);
3955 va_end(args);
3956
3957 return r;
3958}
3959
3960/*
3961 * printk rate limiting, lifted from the networking subsystem.
3962 *
3963 * This enforces a rate limit: not more than 10 kernel messages
3964 * every 5s to make a denial-of-service attack impossible.
3965 */
3966DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3967
3968int __printk_ratelimit(const char *func)
3969{
3970 return ___ratelimit(&printk_ratelimit_state, func);
3971}
3972EXPORT_SYMBOL(__printk_ratelimit);
3973
3974/**
3975 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3976 * @caller_jiffies: pointer to caller's state
3977 * @interval_msecs: minimum interval between prints
3978 *
3979 * printk_timed_ratelimit() returns true if more than @interval_msecs
3980 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3981 * returned true.
3982 */
3983bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3984 unsigned int interval_msecs)
3985{
3986 unsigned long elapsed = jiffies - *caller_jiffies;
3987
3988 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3989 return false;
3990
3991 *caller_jiffies = jiffies;
3992 return true;
3993}
3994EXPORT_SYMBOL(printk_timed_ratelimit);
3995
3996static DEFINE_SPINLOCK(dump_list_lock);
3997static LIST_HEAD(dump_list);
3998
3999/**
4000 * kmsg_dump_register - register a kernel log dumper.
4001 * @dumper: pointer to the kmsg_dumper structure
4002 *
4003 * Adds a kernel log dumper to the system. The dump callback in the
4004 * structure will be called when the kernel oopses or panics and must be
4005 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
4006 */
4007int kmsg_dump_register(struct kmsg_dumper *dumper)
4008{
4009 unsigned long flags;
4010 int err = -EBUSY;
4011
4012 /* The dump callback needs to be set */
4013 if (!dumper->dump)
4014 return -EINVAL;
4015
4016 spin_lock_irqsave(&dump_list_lock, flags);
4017 /* Don't allow registering multiple times */
4018 if (!dumper->registered) {
4019 dumper->registered = 1;
4020 list_add_tail_rcu(&dumper->list, &dump_list);
4021 err = 0;
4022 }
4023 spin_unlock_irqrestore(&dump_list_lock, flags);
4024
4025 return err;
4026}
4027EXPORT_SYMBOL_GPL(kmsg_dump_register);
4028
4029/**
4030 * kmsg_dump_unregister - unregister a kmsg dumper.
4031 * @dumper: pointer to the kmsg_dumper structure
4032 *
4033 * Removes a dump device from the system. Returns zero on success and
4034 * %-EINVAL otherwise.
4035 */
4036int kmsg_dump_unregister(struct kmsg_dumper *dumper)
4037{
4038 unsigned long flags;
4039 int err = -EINVAL;
4040
4041 spin_lock_irqsave(&dump_list_lock, flags);
4042 if (dumper->registered) {
4043 dumper->registered = 0;
4044 list_del_rcu(&dumper->list);
4045 err = 0;
4046 }
4047 spin_unlock_irqrestore(&dump_list_lock, flags);
4048 synchronize_rcu();
4049
4050 return err;
4051}
4052EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
4053
4054static bool always_kmsg_dump;
4055module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
4056
4057const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
4058{
4059 switch (reason) {
4060 case KMSG_DUMP_PANIC:
4061 return "Panic";
4062 case KMSG_DUMP_OOPS:
4063 return "Oops";
4064 case KMSG_DUMP_EMERG:
4065 return "Emergency";
4066 case KMSG_DUMP_SHUTDOWN:
4067 return "Shutdown";
4068 default:
4069 return "Unknown";
4070 }
4071}
4072EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
4073
4074/**
4075 * kmsg_dump - dump kernel log to kernel message dumpers.
4076 * @reason: the reason (oops, panic etc) for dumping
4077 *
4078 * Call each of the registered dumper's dump() callback, which can
4079 * retrieve the kmsg records with kmsg_dump_get_line() or
4080 * kmsg_dump_get_buffer().
4081 */
4082void kmsg_dump(enum kmsg_dump_reason reason)
4083{
4084 struct kmsg_dumper *dumper;
4085
4086 rcu_read_lock();
4087 list_for_each_entry_rcu(dumper, &dump_list, list) {
4088 enum kmsg_dump_reason max_reason = dumper->max_reason;
4089
4090 /*
4091 * If client has not provided a specific max_reason, default
4092 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
4093 */
4094 if (max_reason == KMSG_DUMP_UNDEF) {
4095 max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
4096 KMSG_DUMP_OOPS;
4097 }
4098 if (reason > max_reason)
4099 continue;
4100
4101 /* invoke dumper which will iterate over records */
4102 dumper->dump(dumper, reason);
4103 }
4104 rcu_read_unlock();
4105}
4106
4107/**
4108 * kmsg_dump_get_line - retrieve one kmsg log line
4109 * @iter: kmsg dump iterator
4110 * @syslog: include the "<4>" prefixes
4111 * @line: buffer to copy the line to
4112 * @size: maximum size of the buffer
4113 * @len: length of line placed into buffer
4114 *
4115 * Start at the beginning of the kmsg buffer, with the oldest kmsg
4116 * record, and copy one record into the provided buffer.
4117 *
4118 * Consecutive calls will return the next available record moving
4119 * towards the end of the buffer with the youngest messages.
4120 *
4121 * A return value of FALSE indicates that there are no more records to
4122 * read.
4123 */
4124bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
4125 char *line, size_t size, size_t *len)
4126{
4127 u64 min_seq = latched_seq_read_nolock(&clear_seq);
4128 struct printk_info info;
4129 unsigned int line_count;
4130 struct printk_record r;
4131 size_t l = 0;
4132 bool ret = false;
4133
4134 if (iter->cur_seq < min_seq)
4135 iter->cur_seq = min_seq;
4136
4137 prb_rec_init_rd(&r, &info, line, size);
4138
4139 /* Read text or count text lines? */
4140 if (line) {
4141 if (!prb_read_valid(prb, iter->cur_seq, &r))
4142 goto out;
4143 l = record_print_text(&r, syslog, printk_time);
4144 } else {
4145 if (!prb_read_valid_info(prb, iter->cur_seq,
4146 &info, &line_count)) {
4147 goto out;
4148 }
4149 l = get_record_print_text_size(&info, line_count, syslog,
4150 printk_time);
4151
4152 }
4153
4154 iter->cur_seq = r.info->seq + 1;
4155 ret = true;
4156out:
4157 if (len)
4158 *len = l;
4159 return ret;
4160}
4161EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
4162
4163/**
4164 * kmsg_dump_get_buffer - copy kmsg log lines
4165 * @iter: kmsg dump iterator
4166 * @syslog: include the "<4>" prefixes
4167 * @buf: buffer to copy the line to
4168 * @size: maximum size of the buffer
4169 * @len_out: length of line placed into buffer
4170 *
4171 * Start at the end of the kmsg buffer and fill the provided buffer
4172 * with as many of the *youngest* kmsg records that fit into it.
4173 * If the buffer is large enough, all available kmsg records will be
4174 * copied with a single call.
4175 *
4176 * Consecutive calls will fill the buffer with the next block of
4177 * available older records, not including the earlier retrieved ones.
4178 *
4179 * A return value of FALSE indicates that there are no more records to
4180 * read.
4181 */
4182bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
4183 char *buf, size_t size, size_t *len_out)
4184{
4185 u64 min_seq = latched_seq_read_nolock(&clear_seq);
4186 struct printk_info info;
4187 struct printk_record r;
4188 u64 seq;
4189 u64 next_seq;
4190 size_t len = 0;
4191 bool ret = false;
4192 bool time = printk_time;
4193
4194 if (!buf || !size)
4195 goto out;
4196
4197 if (iter->cur_seq < min_seq)
4198 iter->cur_seq = min_seq;
4199
4200 if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
4201 if (info.seq != iter->cur_seq) {
4202 /* messages are gone, move to first available one */
4203 iter->cur_seq = info.seq;
4204 }
4205 }
4206
4207 /* last entry */
4208 if (iter->cur_seq >= iter->next_seq)
4209 goto out;
4210
4211 /*
4212 * Find first record that fits, including all following records,
4213 * into the user-provided buffer for this dump. Pass in size-1
4214 * because this function (by way of record_print_text()) will
4215 * not write more than size-1 bytes of text into @buf.
4216 */
4217 seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
4218 size - 1, syslog, time);
4219
4220 /*
4221 * Next kmsg_dump_get_buffer() invocation will dump block of
4222 * older records stored right before this one.
4223 */
4224 next_seq = seq;
4225
4226 prb_rec_init_rd(&r, &info, buf, size);
4227
4228 prb_for_each_record(seq, prb, seq, &r) {
4229 if (r.info->seq >= iter->next_seq)
4230 break;
4231
4232 len += record_print_text(&r, syslog, time);
4233
4234 /* Adjust record to store to remaining buffer space. */
4235 prb_rec_init_rd(&r, &info, buf + len, size - len);
4236 }
4237
4238 iter->next_seq = next_seq;
4239 ret = true;
4240out:
4241 if (len_out)
4242 *len_out = len;
4243 return ret;
4244}
4245EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
4246
4247/**
4248 * kmsg_dump_rewind - reset the iterator
4249 * @iter: kmsg dump iterator
4250 *
4251 * Reset the dumper's iterator so that kmsg_dump_get_line() and
4252 * kmsg_dump_get_buffer() can be called again and used multiple
4253 * times within the same dumper.dump() callback.
4254 */
4255void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
4256{
4257 iter->cur_seq = latched_seq_read_nolock(&clear_seq);
4258 iter->next_seq = prb_next_seq(prb);
4259}
4260EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
4261
4262#endif
4263
4264#ifdef CONFIG_SMP
4265static atomic_t printk_cpu_sync_owner = ATOMIC_INIT(-1);
4266static atomic_t printk_cpu_sync_nested = ATOMIC_INIT(0);
4267
4268/**
4269 * __printk_cpu_sync_wait() - Busy wait until the printk cpu-reentrant
4270 * spinning lock is not owned by any CPU.
4271 *
4272 * Context: Any context.
4273 */
4274void __printk_cpu_sync_wait(void)
4275{
4276 do {
4277 cpu_relax();
4278 } while (atomic_read(&printk_cpu_sync_owner) != -1);
4279}
4280EXPORT_SYMBOL(__printk_cpu_sync_wait);
4281
4282/**
4283 * __printk_cpu_sync_try_get() - Try to acquire the printk cpu-reentrant
4284 * spinning lock.
4285 *
4286 * If no processor has the lock, the calling processor takes the lock and
4287 * becomes the owner. If the calling processor is already the owner of the
4288 * lock, this function succeeds immediately.
4289 *
4290 * Context: Any context. Expects interrupts to be disabled.
4291 * Return: 1 on success, otherwise 0.
4292 */
4293int __printk_cpu_sync_try_get(void)
4294{
4295 int cpu;
4296 int old;
4297
4298 cpu = smp_processor_id();
4299
4300 /*
4301 * Guarantee loads and stores from this CPU when it is the lock owner
4302 * are _not_ visible to the previous lock owner. This pairs with
4303 * __printk_cpu_sync_put:B.
4304 *
4305 * Memory barrier involvement:
4306 *
4307 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4308 * then __printk_cpu_sync_put:A can never read from
4309 * __printk_cpu_sync_try_get:B.
4310 *
4311 * Relies on:
4312 *
4313 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4314 * of the previous CPU
4315 * matching
4316 * ACQUIRE from __printk_cpu_sync_try_get:A to
4317 * __printk_cpu_sync_try_get:B of this CPU
4318 */
4319 old = atomic_cmpxchg_acquire(&printk_cpu_sync_owner, -1,
4320 cpu); /* LMM(__printk_cpu_sync_try_get:A) */
4321 if (old == -1) {
4322 /*
4323 * This CPU is now the owner and begins loading/storing
4324 * data: LMM(__printk_cpu_sync_try_get:B)
4325 */
4326 return 1;
4327
4328 } else if (old == cpu) {
4329 /* This CPU is already the owner. */
4330 atomic_inc(&printk_cpu_sync_nested);
4331 return 1;
4332 }
4333
4334 return 0;
4335}
4336EXPORT_SYMBOL(__printk_cpu_sync_try_get);
4337
4338/**
4339 * __printk_cpu_sync_put() - Release the printk cpu-reentrant spinning lock.
4340 *
4341 * The calling processor must be the owner of the lock.
4342 *
4343 * Context: Any context. Expects interrupts to be disabled.
4344 */
4345void __printk_cpu_sync_put(void)
4346{
4347 if (atomic_read(&printk_cpu_sync_nested)) {
4348 atomic_dec(&printk_cpu_sync_nested);
4349 return;
4350 }
4351
4352 /*
4353 * This CPU is finished loading/storing data:
4354 * LMM(__printk_cpu_sync_put:A)
4355 */
4356
4357 /*
4358 * Guarantee loads and stores from this CPU when it was the
4359 * lock owner are visible to the next lock owner. This pairs
4360 * with __printk_cpu_sync_try_get:A.
4361 *
4362 * Memory barrier involvement:
4363 *
4364 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4365 * then __printk_cpu_sync_try_get:B reads from __printk_cpu_sync_put:A.
4366 *
4367 * Relies on:
4368 *
4369 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4370 * of this CPU
4371 * matching
4372 * ACQUIRE from __printk_cpu_sync_try_get:A to
4373 * __printk_cpu_sync_try_get:B of the next CPU
4374 */
4375 atomic_set_release(&printk_cpu_sync_owner,
4376 -1); /* LMM(__printk_cpu_sync_put:B) */
4377}
4378EXPORT_SYMBOL(__printk_cpu_sync_put);
4379#endif /* CONFIG_SMP */