1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * linux/kernel/printk.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * Modified to make sys_syslog() more flexible: added commands to
8 * return the last 4k of kernel messages, regardless of whether
9 * they've been read or not. Added option to suppress kernel printk's
10 * to the console. Added hook for sending the console messages
11 * elsewhere, in preparation for a serial line console (someday).
12 * Ted Ts'o, 2/11/93.
13 * Modified for sysctl support, 1/8/97, Chris Horn.
14 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
15 * manfred@colorfullife.com
16 * Rewrote bits to get rid of console_lock
17 * 01Mar01 Andrew Morton
18 */
19
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21
22 #include <linux/kernel.h>
23 #include <linux/mm.h>
24 #include <linux/tty.h>
25 #include <linux/tty_driver.h>
26 #include <linux/console.h>
27 #include <linux/init.h>
28 #include <linux/jiffies.h>
29 #include <linux/nmi.h>
30 #include <linux/module.h>
31 #include <linux/moduleparam.h>
32 #include <linux/delay.h>
33 #include <linux/smp.h>
34 #include <linux/security.h>
35 #include <linux/memblock.h>
36 #include <linux/syscalls.h>
37 #include <linux/vmcore_info.h>
38 #include <linux/ratelimit.h>
39 #include <linux/kmsg_dump.h>
40 #include <linux/syslog.h>
41 #include <linux/cpu.h>
42 #include <linux/rculist.h>
43 #include <linux/poll.h>
44 #include <linux/irq_work.h>
45 #include <linux/ctype.h>
46 #include <linux/uio.h>
47 #include <linux/sched/clock.h>
48 #include <linux/sched/debug.h>
49 #include <linux/sched/task_stack.h>
50
51 #include <linux/uaccess.h>
52 #include <asm/sections.h>
53
54 #include <trace/events/initcall.h>
55 #define CREATE_TRACE_POINTS
56 #include <trace/events/printk.h>
57
58 #include "printk_ringbuffer.h"
59 #include "console_cmdline.h"
60 #include "braille.h"
61 #include "internal.h"
62
63 int 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 };
69 EXPORT_SYMBOL_GPL(console_printk);
70
71 atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
72 EXPORT_SYMBOL(ignore_console_lock_warning);
73
74 EXPORT_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 */
80 int oops_in_progress;
81 EXPORT_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 */
88 static DEFINE_MUTEX(console_mutex);
89
90 /*
91 * console_sem protects updates to console->seq
92 * and also provides serialization for console printing.
93 */
94 static DEFINE_SEMAPHORE(console_sem, 1);
95 HLIST_HEAD(console_list);
96 EXPORT_SYMBOL_GPL(console_list);
97 DEFINE_STATIC_SRCU(console_srcu);
98
99 /*
100 * System may need to suppress printk message under certain
101 * circumstances, like after kernel panic happens.
102 */
103 int __read_mostly suppress_printk;
104
105 #ifdef CONFIG_LOCKDEP
106 static struct lockdep_map console_lock_dep_map = {
107 .name = "console_lock"
108 };
109
lockdep_assert_console_list_lock_held(void)110 void lockdep_assert_console_list_lock_held(void)
111 {
112 lockdep_assert_held(&console_mutex);
113 }
114 EXPORT_SYMBOL(lockdep_assert_console_list_lock_held);
115 #endif
116
117 #ifdef CONFIG_DEBUG_LOCK_ALLOC
console_srcu_read_lock_is_held(void)118 bool console_srcu_read_lock_is_held(void)
119 {
120 return srcu_read_lock_held(&console_srcu);
121 }
122 EXPORT_SYMBOL(console_srcu_read_lock_is_held);
123 #endif
124
125 enum devkmsg_log_bits {
126 __DEVKMSG_LOG_BIT_ON = 0,
127 __DEVKMSG_LOG_BIT_OFF,
128 __DEVKMSG_LOG_BIT_LOCK,
129 };
130
131 enum 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
140 static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
141
__control_devkmsg(char * str)142 static 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
control_devkmsg(char * str)170 static 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 strscpy(devkmsg_log_str, "on");
182 else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
183 strscpy(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
198 char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
199 #if defined(CONFIG_PRINTK) && defined(CONFIG_SYSCTL)
devkmsg_sysctl_set_loglvl(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)200 int 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 strscpy(old_str, devkmsg_log_str);
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 strscpy(devkmsg_log_str, old_str);
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 */
console_list_lock(void)245 void 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 }
262 EXPORT_SYMBOL(console_list_lock);
263
264 /**
265 * console_list_unlock - Unlock the console list
266 *
267 * Counterpart to console_list_lock()
268 */
console_list_unlock(void)269 void console_list_unlock(void)
270 {
271 mutex_unlock(&console_mutex);
272 }
273 EXPORT_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 */
console_srcu_read_lock(void)284 int console_srcu_read_lock(void)
285 {
286 return srcu_read_lock_nmisafe(&console_srcu);
287 }
288 EXPORT_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 */
console_srcu_read_unlock(int cookie)297 void console_srcu_read_unlock(int cookie)
298 {
299 srcu_read_unlock_nmisafe(&console_srcu, cookie);
300 }
301 EXPORT_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
__down_trylock_console_sem(unsigned long ip)312 static 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
__up_console_sem(unsigned long ip)333 static 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
panic_in_progress(void)345 static bool panic_in_progress(void)
346 {
347 return unlikely(atomic_read(&panic_cpu) != PANIC_CPU_INVALID);
348 }
349
350 /* Return true if a panic is in progress on the current CPU. */
this_cpu_in_panic(void)351 bool this_cpu_in_panic(void)
352 {
353 /*
354 * We can use raw_smp_processor_id() here because it is impossible for
355 * the task to be migrated to the panic_cpu, or away from it. If
356 * panic_cpu has already been set, and we're not currently executing on
357 * that CPU, then we never will be.
358 */
359 return unlikely(atomic_read(&panic_cpu) == raw_smp_processor_id());
360 }
361
362 /*
363 * Return true if a panic is in progress on a remote CPU.
364 *
365 * On true, the local CPU should immediately release any printing resources
366 * that may be needed by the panic CPU.
367 */
other_cpu_in_panic(void)368 bool other_cpu_in_panic(void)
369 {
370 return (panic_in_progress() && !this_cpu_in_panic());
371 }
372
373 /*
374 * This is used for debugging the mess that is the VT code by
375 * keeping track if we have the console semaphore held. It's
376 * definitely not the perfect debug tool (we don't know if _WE_
377 * hold it and are racing, but it helps tracking those weird code
378 * paths in the console code where we end up in places I want
379 * locked without the console semaphore held).
380 */
381 static int console_locked;
382
383 /*
384 * Array of consoles built from command line options (console=)
385 */
386 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
387
388 static int preferred_console = -1;
389 int console_set_on_cmdline;
390 EXPORT_SYMBOL(console_set_on_cmdline);
391
392 /* Flag: console code may call schedule() */
393 static int console_may_schedule;
394
395 enum con_msg_format_flags {
396 MSG_FORMAT_DEFAULT = 0,
397 MSG_FORMAT_SYSLOG = (1 << 0),
398 };
399
400 static int console_msg_format = MSG_FORMAT_DEFAULT;
401
402 /*
403 * The printk log buffer consists of a sequenced collection of records, each
404 * containing variable length message text. Every record also contains its
405 * own meta-data (@info).
406 *
407 * Every record meta-data carries the timestamp in microseconds, as well as
408 * the standard userspace syslog level and syslog facility. The usual kernel
409 * messages use LOG_KERN; userspace-injected messages always carry a matching
410 * syslog facility, by default LOG_USER. The origin of every message can be
411 * reliably determined that way.
412 *
413 * The human readable log message of a record is available in @text, the
414 * length of the message text in @text_len. The stored message is not
415 * terminated.
416 *
417 * Optionally, a record can carry a dictionary of properties (key/value
418 * pairs), to provide userspace with a machine-readable message context.
419 *
420 * Examples for well-defined, commonly used property names are:
421 * DEVICE=b12:8 device identifier
422 * b12:8 block dev_t
423 * c127:3 char dev_t
424 * n8 netdev ifindex
425 * +sound:card0 subsystem:devname
426 * SUBSYSTEM=pci driver-core subsystem name
427 *
428 * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
429 * and values are terminated by a '\0' character.
430 *
431 * Example of record values:
432 * record.text_buf = "it's a line" (unterminated)
433 * record.info.seq = 56
434 * record.info.ts_nsec = 36863
435 * record.info.text_len = 11
436 * record.info.facility = 0 (LOG_KERN)
437 * record.info.flags = 0
438 * record.info.level = 3 (LOG_ERR)
439 * record.info.caller_id = 299 (task 299)
440 * record.info.dev_info.subsystem = "pci" (terminated)
441 * record.info.dev_info.device = "+pci:0000:00:01.0" (terminated)
442 *
443 * The 'struct printk_info' buffer must never be directly exported to
444 * userspace, it is a kernel-private implementation detail that might
445 * need to be changed in the future, when the requirements change.
446 *
447 * /dev/kmsg exports the structured data in the following line format:
448 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
449 *
450 * Users of the export format should ignore possible additional values
451 * separated by ',', and find the message after the ';' character.
452 *
453 * The optional key/value pairs are attached as continuation lines starting
454 * with a space character and terminated by a newline. All possible
455 * non-prinatable characters are escaped in the "\xff" notation.
456 */
457
458 /* syslog_lock protects syslog_* variables and write access to clear_seq. */
459 static DEFINE_MUTEX(syslog_lock);
460
461 #ifdef CONFIG_PRINTK
462 DECLARE_WAIT_QUEUE_HEAD(log_wait);
463 /* All 3 protected by @syslog_lock. */
464 /* the next printk record to read by syslog(READ) or /proc/kmsg */
465 static u64 syslog_seq;
466 static size_t syslog_partial;
467 static bool syslog_time;
468
469 struct latched_seq {
470 seqcount_latch_t latch;
471 u64 val[2];
472 };
473
474 /*
475 * The next printk record to read after the last 'clear' command. There are
476 * two copies (updated with seqcount_latch) so that reads can locklessly
477 * access a valid value. Writers are synchronized by @syslog_lock.
478 */
479 static struct latched_seq clear_seq = {
480 .latch = SEQCNT_LATCH_ZERO(clear_seq.latch),
481 .val[0] = 0,
482 .val[1] = 0,
483 };
484
485 #define LOG_LEVEL(v) ((v) & 0x07)
486 #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
487
488 /* record buffer */
489 #define LOG_ALIGN __alignof__(unsigned long)
490 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
491 #define LOG_BUF_LEN_MAX (u32)(1 << 31)
492 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
493 static char *log_buf = __log_buf;
494 static u32 log_buf_len = __LOG_BUF_LEN;
495
496 /*
497 * Define the average message size. This only affects the number of
498 * descriptors that will be available. Underestimating is better than
499 * overestimating (too many available descriptors is better than not enough).
500 */
501 #define PRB_AVGBITS 5 /* 32 character average length */
502
503 #if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
504 #error CONFIG_LOG_BUF_SHIFT value too small.
505 #endif
506 _DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
507 PRB_AVGBITS, &__log_buf[0]);
508
509 static struct printk_ringbuffer printk_rb_dynamic;
510
511 struct printk_ringbuffer *prb = &printk_rb_static;
512
513 /*
514 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
515 * per_cpu_areas are initialised. This variable is set to true when
516 * it's safe to access per-CPU data.
517 */
518 static bool __printk_percpu_data_ready __ro_after_init;
519
printk_percpu_data_ready(void)520 bool printk_percpu_data_ready(void)
521 {
522 return __printk_percpu_data_ready;
523 }
524
525 /* Must be called under syslog_lock. */
latched_seq_write(struct latched_seq * ls,u64 val)526 static void latched_seq_write(struct latched_seq *ls, u64 val)
527 {
528 raw_write_seqcount_latch(&ls->latch);
529 ls->val[0] = val;
530 raw_write_seqcount_latch(&ls->latch);
531 ls->val[1] = val;
532 }
533
534 /* Can be called from any context. */
latched_seq_read_nolock(struct latched_seq * ls)535 static u64 latched_seq_read_nolock(struct latched_seq *ls)
536 {
537 unsigned int seq;
538 unsigned int idx;
539 u64 val;
540
541 do {
542 seq = raw_read_seqcount_latch(&ls->latch);
543 idx = seq & 0x1;
544 val = ls->val[idx];
545 } while (raw_read_seqcount_latch_retry(&ls->latch, seq));
546
547 return val;
548 }
549
550 /* Return log buffer address */
log_buf_addr_get(void)551 char *log_buf_addr_get(void)
552 {
553 return log_buf;
554 }
555
556 /* Return log buffer size */
log_buf_len_get(void)557 u32 log_buf_len_get(void)
558 {
559 return log_buf_len;
560 }
561
562 /*
563 * Define how much of the log buffer we could take at maximum. The value
564 * must be greater than two. Note that only half of the buffer is available
565 * when the index points to the middle.
566 */
567 #define MAX_LOG_TAKE_PART 4
568 static const char trunc_msg[] = "<truncated>";
569
truncate_msg(u16 * text_len,u16 * trunc_msg_len)570 static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
571 {
572 /*
573 * The message should not take the whole buffer. Otherwise, it might
574 * get removed too soon.
575 */
576 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
577
578 if (*text_len > max_text_len)
579 *text_len = max_text_len;
580
581 /* enable the warning message (if there is room) */
582 *trunc_msg_len = strlen(trunc_msg);
583 if (*text_len >= *trunc_msg_len)
584 *text_len -= *trunc_msg_len;
585 else
586 *trunc_msg_len = 0;
587 }
588
589 int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
590
syslog_action_restricted(int type)591 static int syslog_action_restricted(int type)
592 {
593 if (dmesg_restrict)
594 return 1;
595 /*
596 * Unless restricted, we allow "read all" and "get buffer size"
597 * for everybody.
598 */
599 return type != SYSLOG_ACTION_READ_ALL &&
600 type != SYSLOG_ACTION_SIZE_BUFFER;
601 }
602
check_syslog_permissions(int type,int source)603 static int check_syslog_permissions(int type, int source)
604 {
605 /*
606 * If this is from /proc/kmsg and we've already opened it, then we've
607 * already done the capabilities checks at open time.
608 */
609 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
610 goto ok;
611
612 if (syslog_action_restricted(type)) {
613 if (capable(CAP_SYSLOG))
614 goto ok;
615 return -EPERM;
616 }
617 ok:
618 return security_syslog(type);
619 }
620
append_char(char ** pp,char * e,char c)621 static void append_char(char **pp, char *e, char c)
622 {
623 if (*pp < e)
624 *(*pp)++ = c;
625 }
626
info_print_ext_header(char * buf,size_t size,struct printk_info * info)627 static ssize_t info_print_ext_header(char *buf, size_t size,
628 struct printk_info *info)
629 {
630 u64 ts_usec = info->ts_nsec;
631 char caller[20];
632 #ifdef CONFIG_PRINTK_CALLER
633 u32 id = info->caller_id;
634
635 snprintf(caller, sizeof(caller), ",caller=%c%u",
636 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
637 #else
638 caller[0] = '\0';
639 #endif
640
641 do_div(ts_usec, 1000);
642
643 return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
644 (info->facility << 3) | info->level, info->seq,
645 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
646 }
647
msg_add_ext_text(char * buf,size_t size,const char * text,size_t text_len,unsigned char endc)648 static ssize_t msg_add_ext_text(char *buf, size_t size,
649 const char *text, size_t text_len,
650 unsigned char endc)
651 {
652 char *p = buf, *e = buf + size;
653 size_t i;
654
655 /* escape non-printable characters */
656 for (i = 0; i < text_len; i++) {
657 unsigned char c = text[i];
658
659 if (c < ' ' || c >= 127 || c == '\\')
660 p += scnprintf(p, e - p, "\\x%02x", c);
661 else
662 append_char(&p, e, c);
663 }
664 append_char(&p, e, endc);
665
666 return p - buf;
667 }
668
msg_add_dict_text(char * buf,size_t size,const char * key,const char * val)669 static ssize_t msg_add_dict_text(char *buf, size_t size,
670 const char *key, const char *val)
671 {
672 size_t val_len = strlen(val);
673 ssize_t len;
674
675 if (!val_len)
676 return 0;
677
678 len = msg_add_ext_text(buf, size, "", 0, ' '); /* dict prefix */
679 len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
680 len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
681
682 return len;
683 }
684
msg_print_ext_body(char * buf,size_t size,char * text,size_t text_len,struct dev_printk_info * dev_info)685 static ssize_t msg_print_ext_body(char *buf, size_t size,
686 char *text, size_t text_len,
687 struct dev_printk_info *dev_info)
688 {
689 ssize_t len;
690
691 len = msg_add_ext_text(buf, size, text, text_len, '\n');
692
693 if (!dev_info)
694 goto out;
695
696 len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
697 dev_info->subsystem);
698 len += msg_add_dict_text(buf + len, size - len, "DEVICE",
699 dev_info->device);
700 out:
701 return len;
702 }
703
704 /* /dev/kmsg - userspace message inject/listen interface */
705 struct devkmsg_user {
706 atomic64_t seq;
707 struct ratelimit_state rs;
708 struct mutex lock;
709 struct printk_buffers pbufs;
710 };
711
712 static __printf(3, 4) __cold
devkmsg_emit(int facility,int level,const char * fmt,...)713 int devkmsg_emit(int facility, int level, const char *fmt, ...)
714 {
715 va_list args;
716 int r;
717
718 va_start(args, fmt);
719 r = vprintk_emit(facility, level, NULL, fmt, args);
720 va_end(args);
721
722 return r;
723 }
724
devkmsg_write(struct kiocb * iocb,struct iov_iter * from)725 static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
726 {
727 char *buf, *line;
728 int level = default_message_loglevel;
729 int facility = 1; /* LOG_USER */
730 struct file *file = iocb->ki_filp;
731 struct devkmsg_user *user = file->private_data;
732 size_t len = iov_iter_count(from);
733 ssize_t ret = len;
734
735 if (len > PRINTKRB_RECORD_MAX)
736 return -EINVAL;
737
738 /* Ignore when user logging is disabled. */
739 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
740 return len;
741
742 /* Ratelimit when not explicitly enabled. */
743 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
744 if (!___ratelimit(&user->rs, current->comm))
745 return ret;
746 }
747
748 buf = kmalloc(len+1, GFP_KERNEL);
749 if (buf == NULL)
750 return -ENOMEM;
751
752 buf[len] = '\0';
753 if (!copy_from_iter_full(buf, len, from)) {
754 kfree(buf);
755 return -EFAULT;
756 }
757
758 /*
759 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
760 * the decimal value represents 32bit, the lower 3 bit are the log
761 * level, the rest are the log facility.
762 *
763 * If no prefix or no userspace facility is specified, we
764 * enforce LOG_USER, to be able to reliably distinguish
765 * kernel-generated messages from userspace-injected ones.
766 */
767 line = buf;
768 if (line[0] == '<') {
769 char *endp = NULL;
770 unsigned int u;
771
772 u = simple_strtoul(line + 1, &endp, 10);
773 if (endp && endp[0] == '>') {
774 level = LOG_LEVEL(u);
775 if (LOG_FACILITY(u) != 0)
776 facility = LOG_FACILITY(u);
777 endp++;
778 line = endp;
779 }
780 }
781
782 devkmsg_emit(facility, level, "%s", line);
783 kfree(buf);
784 return ret;
785 }
786
devkmsg_read(struct file * file,char __user * buf,size_t count,loff_t * ppos)787 static ssize_t devkmsg_read(struct file *file, char __user *buf,
788 size_t count, loff_t *ppos)
789 {
790 struct devkmsg_user *user = file->private_data;
791 char *outbuf = &user->pbufs.outbuf[0];
792 struct printk_message pmsg = {
793 .pbufs = &user->pbufs,
794 };
795 ssize_t ret;
796
797 ret = mutex_lock_interruptible(&user->lock);
798 if (ret)
799 return ret;
800
801 if (!printk_get_next_message(&pmsg, atomic64_read(&user->seq), true, false)) {
802 if (file->f_flags & O_NONBLOCK) {
803 ret = -EAGAIN;
804 goto out;
805 }
806
807 /*
808 * Guarantee this task is visible on the waitqueue before
809 * checking the wake condition.
810 *
811 * The full memory barrier within set_current_state() of
812 * prepare_to_wait_event() pairs with the full memory barrier
813 * within wq_has_sleeper().
814 *
815 * This pairs with __wake_up_klogd:A.
816 */
817 ret = wait_event_interruptible(log_wait,
818 printk_get_next_message(&pmsg, atomic64_read(&user->seq), true,
819 false)); /* LMM(devkmsg_read:A) */
820 if (ret)
821 goto out;
822 }
823
824 if (pmsg.dropped) {
825 /* our last seen message is gone, return error and reset */
826 atomic64_set(&user->seq, pmsg.seq);
827 ret = -EPIPE;
828 goto out;
829 }
830
831 atomic64_set(&user->seq, pmsg.seq + 1);
832
833 if (pmsg.outbuf_len > count) {
834 ret = -EINVAL;
835 goto out;
836 }
837
838 if (copy_to_user(buf, outbuf, pmsg.outbuf_len)) {
839 ret = -EFAULT;
840 goto out;
841 }
842 ret = pmsg.outbuf_len;
843 out:
844 mutex_unlock(&user->lock);
845 return ret;
846 }
847
848 /*
849 * Be careful when modifying this function!!!
850 *
851 * Only few operations are supported because the device works only with the
852 * entire variable length messages (records). Non-standard values are
853 * returned in the other cases and has been this way for quite some time.
854 * User space applications might depend on this behavior.
855 */
devkmsg_llseek(struct file * file,loff_t offset,int whence)856 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
857 {
858 struct devkmsg_user *user = file->private_data;
859 loff_t ret = 0;
860
861 if (offset)
862 return -ESPIPE;
863
864 switch (whence) {
865 case SEEK_SET:
866 /* the first record */
867 atomic64_set(&user->seq, prb_first_valid_seq(prb));
868 break;
869 case SEEK_DATA:
870 /*
871 * The first record after the last SYSLOG_ACTION_CLEAR,
872 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
873 * changes no global state, and does not clear anything.
874 */
875 atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
876 break;
877 case SEEK_END:
878 /* after the last record */
879 atomic64_set(&user->seq, prb_next_seq(prb));
880 break;
881 default:
882 ret = -EINVAL;
883 }
884 return ret;
885 }
886
devkmsg_poll(struct file * file,poll_table * wait)887 static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
888 {
889 struct devkmsg_user *user = file->private_data;
890 struct printk_info info;
891 __poll_t ret = 0;
892
893 poll_wait(file, &log_wait, wait);
894
895 if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
896 /* return error when data has vanished underneath us */
897 if (info.seq != atomic64_read(&user->seq))
898 ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
899 else
900 ret = EPOLLIN|EPOLLRDNORM;
901 }
902
903 return ret;
904 }
905
devkmsg_open(struct inode * inode,struct file * file)906 static int devkmsg_open(struct inode *inode, struct file *file)
907 {
908 struct devkmsg_user *user;
909 int err;
910
911 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
912 return -EPERM;
913
914 /* write-only does not need any file context */
915 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
916 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
917 SYSLOG_FROM_READER);
918 if (err)
919 return err;
920 }
921
922 user = kvmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
923 if (!user)
924 return -ENOMEM;
925
926 ratelimit_default_init(&user->rs);
927 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
928
929 mutex_init(&user->lock);
930
931 atomic64_set(&user->seq, prb_first_valid_seq(prb));
932
933 file->private_data = user;
934 return 0;
935 }
936
devkmsg_release(struct inode * inode,struct file * file)937 static int devkmsg_release(struct inode *inode, struct file *file)
938 {
939 struct devkmsg_user *user = file->private_data;
940
941 ratelimit_state_exit(&user->rs);
942
943 mutex_destroy(&user->lock);
944 kvfree(user);
945 return 0;
946 }
947
948 const struct file_operations kmsg_fops = {
949 .open = devkmsg_open,
950 .read = devkmsg_read,
951 .write_iter = devkmsg_write,
952 .llseek = devkmsg_llseek,
953 .poll = devkmsg_poll,
954 .release = devkmsg_release,
955 };
956
957 #ifdef CONFIG_VMCORE_INFO
958 /*
959 * This appends the listed symbols to /proc/vmcore
960 *
961 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
962 * obtain access to symbols that are otherwise very difficult to locate. These
963 * symbols are specifically used so that utilities can access and extract the
964 * dmesg log from a vmcore file after a crash.
965 */
log_buf_vmcoreinfo_setup(void)966 void log_buf_vmcoreinfo_setup(void)
967 {
968 struct dev_printk_info *dev_info = NULL;
969
970 VMCOREINFO_SYMBOL(prb);
971 VMCOREINFO_SYMBOL(printk_rb_static);
972 VMCOREINFO_SYMBOL(clear_seq);
973
974 /*
975 * Export struct size and field offsets. User space tools can
976 * parse it and detect any changes to structure down the line.
977 */
978
979 VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
980 VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
981 VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
982 VMCOREINFO_OFFSET(printk_ringbuffer, fail);
983
984 VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
985 VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
986 VMCOREINFO_OFFSET(prb_desc_ring, descs);
987 VMCOREINFO_OFFSET(prb_desc_ring, infos);
988 VMCOREINFO_OFFSET(prb_desc_ring, head_id);
989 VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
990
991 VMCOREINFO_STRUCT_SIZE(prb_desc);
992 VMCOREINFO_OFFSET(prb_desc, state_var);
993 VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
994
995 VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
996 VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
997 VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
998
999 VMCOREINFO_STRUCT_SIZE(printk_info);
1000 VMCOREINFO_OFFSET(printk_info, seq);
1001 VMCOREINFO_OFFSET(printk_info, ts_nsec);
1002 VMCOREINFO_OFFSET(printk_info, text_len);
1003 VMCOREINFO_OFFSET(printk_info, caller_id);
1004 VMCOREINFO_OFFSET(printk_info, dev_info);
1005
1006 VMCOREINFO_STRUCT_SIZE(dev_printk_info);
1007 VMCOREINFO_OFFSET(dev_printk_info, subsystem);
1008 VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
1009 VMCOREINFO_OFFSET(dev_printk_info, device);
1010 VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
1011
1012 VMCOREINFO_STRUCT_SIZE(prb_data_ring);
1013 VMCOREINFO_OFFSET(prb_data_ring, size_bits);
1014 VMCOREINFO_OFFSET(prb_data_ring, data);
1015 VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
1016 VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
1017
1018 VMCOREINFO_SIZE(atomic_long_t);
1019 VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
1020
1021 VMCOREINFO_STRUCT_SIZE(latched_seq);
1022 VMCOREINFO_OFFSET(latched_seq, val);
1023 }
1024 #endif
1025
1026 /* requested log_buf_len from kernel cmdline */
1027 static unsigned long __initdata new_log_buf_len;
1028
1029 /* we practice scaling the ring buffer by powers of 2 */
log_buf_len_update(u64 size)1030 static void __init log_buf_len_update(u64 size)
1031 {
1032 if (size > (u64)LOG_BUF_LEN_MAX) {
1033 size = (u64)LOG_BUF_LEN_MAX;
1034 pr_err("log_buf over 2G is not supported.\n");
1035 }
1036
1037 if (size)
1038 size = roundup_pow_of_two(size);
1039 if (size > log_buf_len)
1040 new_log_buf_len = (unsigned long)size;
1041 }
1042
1043 /* save requested log_buf_len since it's too early to process it */
log_buf_len_setup(char * str)1044 static int __init log_buf_len_setup(char *str)
1045 {
1046 u64 size;
1047
1048 if (!str)
1049 return -EINVAL;
1050
1051 size = memparse(str, &str);
1052
1053 log_buf_len_update(size);
1054
1055 return 0;
1056 }
1057 early_param("log_buf_len", log_buf_len_setup);
1058
1059 #ifdef CONFIG_SMP
1060 #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1061
log_buf_add_cpu(void)1062 static void __init log_buf_add_cpu(void)
1063 {
1064 unsigned int cpu_extra;
1065
1066 /*
1067 * archs should set up cpu_possible_bits properly with
1068 * set_cpu_possible() after setup_arch() but just in
1069 * case lets ensure this is valid.
1070 */
1071 if (num_possible_cpus() == 1)
1072 return;
1073
1074 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1075
1076 /* by default this will only continue through for large > 64 CPUs */
1077 if (cpu_extra <= __LOG_BUF_LEN / 2)
1078 return;
1079
1080 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1081 __LOG_CPU_MAX_BUF_LEN);
1082 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1083 cpu_extra);
1084 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1085
1086 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1087 }
1088 #else /* !CONFIG_SMP */
log_buf_add_cpu(void)1089 static inline void log_buf_add_cpu(void) {}
1090 #endif /* CONFIG_SMP */
1091
set_percpu_data_ready(void)1092 static void __init set_percpu_data_ready(void)
1093 {
1094 __printk_percpu_data_ready = true;
1095 }
1096
add_to_rb(struct printk_ringbuffer * rb,struct printk_record * r)1097 static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1098 struct printk_record *r)
1099 {
1100 struct prb_reserved_entry e;
1101 struct printk_record dest_r;
1102
1103 prb_rec_init_wr(&dest_r, r->info->text_len);
1104
1105 if (!prb_reserve(&e, rb, &dest_r))
1106 return 0;
1107
1108 memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1109 dest_r.info->text_len = r->info->text_len;
1110 dest_r.info->facility = r->info->facility;
1111 dest_r.info->level = r->info->level;
1112 dest_r.info->flags = r->info->flags;
1113 dest_r.info->ts_nsec = r->info->ts_nsec;
1114 dest_r.info->caller_id = r->info->caller_id;
1115 memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1116
1117 prb_final_commit(&e);
1118
1119 return prb_record_text_space(&e);
1120 }
1121
1122 static char setup_text_buf[PRINTKRB_RECORD_MAX] __initdata;
1123
setup_log_buf(int early)1124 void __init setup_log_buf(int early)
1125 {
1126 struct printk_info *new_infos;
1127 unsigned int new_descs_count;
1128 struct prb_desc *new_descs;
1129 struct printk_info info;
1130 struct printk_record r;
1131 unsigned int text_size;
1132 size_t new_descs_size;
1133 size_t new_infos_size;
1134 unsigned long flags;
1135 char *new_log_buf;
1136 unsigned int free;
1137 u64 seq;
1138
1139 /*
1140 * Some archs call setup_log_buf() multiple times - first is very
1141 * early, e.g. from setup_arch(), and second - when percpu_areas
1142 * are initialised.
1143 */
1144 if (!early)
1145 set_percpu_data_ready();
1146
1147 if (log_buf != __log_buf)
1148 return;
1149
1150 if (!early && !new_log_buf_len)
1151 log_buf_add_cpu();
1152
1153 if (!new_log_buf_len)
1154 return;
1155
1156 new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1157 if (new_descs_count == 0) {
1158 pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1159 return;
1160 }
1161
1162 new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1163 if (unlikely(!new_log_buf)) {
1164 pr_err("log_buf_len: %lu text bytes not available\n",
1165 new_log_buf_len);
1166 return;
1167 }
1168
1169 new_descs_size = new_descs_count * sizeof(struct prb_desc);
1170 new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1171 if (unlikely(!new_descs)) {
1172 pr_err("log_buf_len: %zu desc bytes not available\n",
1173 new_descs_size);
1174 goto err_free_log_buf;
1175 }
1176
1177 new_infos_size = new_descs_count * sizeof(struct printk_info);
1178 new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1179 if (unlikely(!new_infos)) {
1180 pr_err("log_buf_len: %zu info bytes not available\n",
1181 new_infos_size);
1182 goto err_free_descs;
1183 }
1184
1185 prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1186
1187 prb_init(&printk_rb_dynamic,
1188 new_log_buf, ilog2(new_log_buf_len),
1189 new_descs, ilog2(new_descs_count),
1190 new_infos);
1191
1192 local_irq_save(flags);
1193
1194 log_buf_len = new_log_buf_len;
1195 log_buf = new_log_buf;
1196 new_log_buf_len = 0;
1197
1198 free = __LOG_BUF_LEN;
1199 prb_for_each_record(0, &printk_rb_static, seq, &r) {
1200 text_size = add_to_rb(&printk_rb_dynamic, &r);
1201 if (text_size > free)
1202 free = 0;
1203 else
1204 free -= text_size;
1205 }
1206
1207 prb = &printk_rb_dynamic;
1208
1209 local_irq_restore(flags);
1210
1211 /*
1212 * Copy any remaining messages that might have appeared from
1213 * NMI context after copying but before switching to the
1214 * dynamic buffer.
1215 */
1216 prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1217 text_size = add_to_rb(&printk_rb_dynamic, &r);
1218 if (text_size > free)
1219 free = 0;
1220 else
1221 free -= text_size;
1222 }
1223
1224 if (seq != prb_next_seq(&printk_rb_static)) {
1225 pr_err("dropped %llu messages\n",
1226 prb_next_seq(&printk_rb_static) - seq);
1227 }
1228
1229 pr_info("log_buf_len: %u bytes\n", log_buf_len);
1230 pr_info("early log buf free: %u(%u%%)\n",
1231 free, (free * 100) / __LOG_BUF_LEN);
1232 return;
1233
1234 err_free_descs:
1235 memblock_free(new_descs, new_descs_size);
1236 err_free_log_buf:
1237 memblock_free(new_log_buf, new_log_buf_len);
1238 }
1239
1240 static bool __read_mostly ignore_loglevel;
1241
ignore_loglevel_setup(char * str)1242 static int __init ignore_loglevel_setup(char *str)
1243 {
1244 ignore_loglevel = true;
1245 pr_info("debug: ignoring loglevel setting.\n");
1246
1247 return 0;
1248 }
1249
1250 early_param("ignore_loglevel", ignore_loglevel_setup);
1251 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1252 MODULE_PARM_DESC(ignore_loglevel,
1253 "ignore loglevel setting (prints all kernel messages to the console)");
1254
suppress_message_printing(int level)1255 static bool suppress_message_printing(int level)
1256 {
1257 return (level >= console_loglevel && !ignore_loglevel);
1258 }
1259
1260 #ifdef CONFIG_BOOT_PRINTK_DELAY
1261
1262 static int boot_delay; /* msecs delay after each printk during bootup */
1263 static unsigned long long loops_per_msec; /* based on boot_delay */
1264
boot_delay_setup(char * str)1265 static int __init boot_delay_setup(char *str)
1266 {
1267 unsigned long lpj;
1268
1269 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1270 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1271
1272 get_option(&str, &boot_delay);
1273 if (boot_delay > 10 * 1000)
1274 boot_delay = 0;
1275
1276 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1277 "HZ: %d, loops_per_msec: %llu\n",
1278 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1279 return 0;
1280 }
1281 early_param("boot_delay", boot_delay_setup);
1282
boot_delay_msec(int level)1283 static void boot_delay_msec(int level)
1284 {
1285 unsigned long long k;
1286 unsigned long timeout;
1287
1288 if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1289 || suppress_message_printing(level)) {
1290 return;
1291 }
1292
1293 k = (unsigned long long)loops_per_msec * boot_delay;
1294
1295 timeout = jiffies + msecs_to_jiffies(boot_delay);
1296 while (k) {
1297 k--;
1298 cpu_relax();
1299 /*
1300 * use (volatile) jiffies to prevent
1301 * compiler reduction; loop termination via jiffies
1302 * is secondary and may or may not happen.
1303 */
1304 if (time_after(jiffies, timeout))
1305 break;
1306 touch_nmi_watchdog();
1307 }
1308 }
1309 #else
boot_delay_msec(int level)1310 static inline void boot_delay_msec(int level)
1311 {
1312 }
1313 #endif
1314
1315 static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1316 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1317
print_syslog(unsigned int level,char * buf)1318 static size_t print_syslog(unsigned int level, char *buf)
1319 {
1320 return sprintf(buf, "<%u>", level);
1321 }
1322
print_time(u64 ts,char * buf)1323 static size_t print_time(u64 ts, char *buf)
1324 {
1325 unsigned long rem_nsec = do_div(ts, 1000000000);
1326
1327 return sprintf(buf, "[%5lu.%06lu]",
1328 (unsigned long)ts, rem_nsec / 1000);
1329 }
1330
1331 #ifdef CONFIG_PRINTK_CALLER
print_caller(u32 id,char * buf)1332 static size_t print_caller(u32 id, char *buf)
1333 {
1334 char caller[12];
1335
1336 snprintf(caller, sizeof(caller), "%c%u",
1337 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1338 return sprintf(buf, "[%6s]", caller);
1339 }
1340 #else
1341 #define print_caller(id, buf) 0
1342 #endif
1343
info_print_prefix(const struct printk_info * info,bool syslog,bool time,char * buf)1344 static size_t info_print_prefix(const struct printk_info *info, bool syslog,
1345 bool time, char *buf)
1346 {
1347 size_t len = 0;
1348
1349 if (syslog)
1350 len = print_syslog((info->facility << 3) | info->level, buf);
1351
1352 if (time)
1353 len += print_time(info->ts_nsec, buf + len);
1354
1355 len += print_caller(info->caller_id, buf + len);
1356
1357 if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1358 buf[len++] = ' ';
1359 buf[len] = '\0';
1360 }
1361
1362 return len;
1363 }
1364
1365 /*
1366 * Prepare the record for printing. The text is shifted within the given
1367 * buffer to avoid a need for another one. The following operations are
1368 * done:
1369 *
1370 * - Add prefix for each line.
1371 * - Drop truncated lines that no longer fit into the buffer.
1372 * - Add the trailing newline that has been removed in vprintk_store().
1373 * - Add a string terminator.
1374 *
1375 * Since the produced string is always terminated, the maximum possible
1376 * return value is @r->text_buf_size - 1;
1377 *
1378 * Return: The length of the updated/prepared text, including the added
1379 * prefixes and the newline. The terminator is not counted. The dropped
1380 * line(s) are not counted.
1381 */
record_print_text(struct printk_record * r,bool syslog,bool time)1382 static size_t record_print_text(struct printk_record *r, bool syslog,
1383 bool time)
1384 {
1385 size_t text_len = r->info->text_len;
1386 size_t buf_size = r->text_buf_size;
1387 char *text = r->text_buf;
1388 char prefix[PRINTK_PREFIX_MAX];
1389 bool truncated = false;
1390 size_t prefix_len;
1391 size_t line_len;
1392 size_t len = 0;
1393 char *next;
1394
1395 /*
1396 * If the message was truncated because the buffer was not large
1397 * enough, treat the available text as if it were the full text.
1398 */
1399 if (text_len > buf_size)
1400 text_len = buf_size;
1401
1402 prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1403
1404 /*
1405 * @text_len: bytes of unprocessed text
1406 * @line_len: bytes of current line _without_ newline
1407 * @text: pointer to beginning of current line
1408 * @len: number of bytes prepared in r->text_buf
1409 */
1410 for (;;) {
1411 next = memchr(text, '\n', text_len);
1412 if (next) {
1413 line_len = next - text;
1414 } else {
1415 /* Drop truncated line(s). */
1416 if (truncated)
1417 break;
1418 line_len = text_len;
1419 }
1420
1421 /*
1422 * Truncate the text if there is not enough space to add the
1423 * prefix and a trailing newline and a terminator.
1424 */
1425 if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1426 /* Drop even the current line if no space. */
1427 if (len + prefix_len + line_len + 1 + 1 > buf_size)
1428 break;
1429
1430 text_len = buf_size - len - prefix_len - 1 - 1;
1431 truncated = true;
1432 }
1433
1434 memmove(text + prefix_len, text, text_len);
1435 memcpy(text, prefix, prefix_len);
1436
1437 /*
1438 * Increment the prepared length to include the text and
1439 * prefix that were just moved+copied. Also increment for the
1440 * newline at the end of this line. If this is the last line,
1441 * there is no newline, but it will be added immediately below.
1442 */
1443 len += prefix_len + line_len + 1;
1444 if (text_len == line_len) {
1445 /*
1446 * This is the last line. Add the trailing newline
1447 * removed in vprintk_store().
1448 */
1449 text[prefix_len + line_len] = '\n';
1450 break;
1451 }
1452
1453 /*
1454 * Advance beyond the added prefix and the related line with
1455 * its newline.
1456 */
1457 text += prefix_len + line_len + 1;
1458
1459 /*
1460 * The remaining text has only decreased by the line with its
1461 * newline.
1462 *
1463 * Note that @text_len can become zero. It happens when @text
1464 * ended with a newline (either due to truncation or the
1465 * original string ending with "\n\n"). The loop is correctly
1466 * repeated and (if not truncated) an empty line with a prefix
1467 * will be prepared.
1468 */
1469 text_len -= line_len + 1;
1470 }
1471
1472 /*
1473 * If a buffer was provided, it will be terminated. Space for the
1474 * string terminator is guaranteed to be available. The terminator is
1475 * not counted in the return value.
1476 */
1477 if (buf_size > 0)
1478 r->text_buf[len] = 0;
1479
1480 return len;
1481 }
1482
get_record_print_text_size(struct printk_info * info,unsigned int line_count,bool syslog,bool time)1483 static size_t get_record_print_text_size(struct printk_info *info,
1484 unsigned int line_count,
1485 bool syslog, bool time)
1486 {
1487 char prefix[PRINTK_PREFIX_MAX];
1488 size_t prefix_len;
1489
1490 prefix_len = info_print_prefix(info, syslog, time, prefix);
1491
1492 /*
1493 * Each line will be preceded with a prefix. The intermediate
1494 * newlines are already within the text, but a final trailing
1495 * newline will be added.
1496 */
1497 return ((prefix_len * line_count) + info->text_len + 1);
1498 }
1499
1500 /*
1501 * Beginning with @start_seq, find the first record where it and all following
1502 * records up to (but not including) @max_seq fit into @size.
1503 *
1504 * @max_seq is simply an upper bound and does not need to exist. If the caller
1505 * does not require an upper bound, -1 can be used for @max_seq.
1506 */
find_first_fitting_seq(u64 start_seq,u64 max_seq,size_t size,bool syslog,bool time)1507 static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1508 bool syslog, bool time)
1509 {
1510 struct printk_info info;
1511 unsigned int line_count;
1512 size_t len = 0;
1513 u64 seq;
1514
1515 /* Determine the size of the records up to @max_seq. */
1516 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1517 if (info.seq >= max_seq)
1518 break;
1519 len += get_record_print_text_size(&info, line_count, syslog, time);
1520 }
1521
1522 /*
1523 * Adjust the upper bound for the next loop to avoid subtracting
1524 * lengths that were never added.
1525 */
1526 if (seq < max_seq)
1527 max_seq = seq;
1528
1529 /*
1530 * Move first record forward until length fits into the buffer. Ignore
1531 * newest messages that were not counted in the above cycle. Messages
1532 * might appear and get lost in the meantime. This is a best effort
1533 * that prevents an infinite loop that could occur with a retry.
1534 */
1535 prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1536 if (len <= size || info.seq >= max_seq)
1537 break;
1538 len -= get_record_print_text_size(&info, line_count, syslog, time);
1539 }
1540
1541 return seq;
1542 }
1543
1544 /* The caller is responsible for making sure @size is greater than 0. */
syslog_print(char __user * buf,int size)1545 static int syslog_print(char __user *buf, int size)
1546 {
1547 struct printk_info info;
1548 struct printk_record r;
1549 char *text;
1550 int len = 0;
1551 u64 seq;
1552
1553 text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1554 if (!text)
1555 return -ENOMEM;
1556
1557 prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1558
1559 mutex_lock(&syslog_lock);
1560
1561 /*
1562 * Wait for the @syslog_seq record to be available. @syslog_seq may
1563 * change while waiting.
1564 */
1565 do {
1566 seq = syslog_seq;
1567
1568 mutex_unlock(&syslog_lock);
1569 /*
1570 * Guarantee this task is visible on the waitqueue before
1571 * checking the wake condition.
1572 *
1573 * The full memory barrier within set_current_state() of
1574 * prepare_to_wait_event() pairs with the full memory barrier
1575 * within wq_has_sleeper().
1576 *
1577 * This pairs with __wake_up_klogd:A.
1578 */
1579 len = wait_event_interruptible(log_wait,
1580 prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
1581 mutex_lock(&syslog_lock);
1582
1583 if (len)
1584 goto out;
1585 } while (syslog_seq != seq);
1586
1587 /*
1588 * Copy records that fit into the buffer. The above cycle makes sure
1589 * that the first record is always available.
1590 */
1591 do {
1592 size_t n;
1593 size_t skip;
1594 int err;
1595
1596 if (!prb_read_valid(prb, syslog_seq, &r))
1597 break;
1598
1599 if (r.info->seq != syslog_seq) {
1600 /* message is gone, move to next valid one */
1601 syslog_seq = r.info->seq;
1602 syslog_partial = 0;
1603 }
1604
1605 /*
1606 * To keep reading/counting partial line consistent,
1607 * use printk_time value as of the beginning of a line.
1608 */
1609 if (!syslog_partial)
1610 syslog_time = printk_time;
1611
1612 skip = syslog_partial;
1613 n = record_print_text(&r, true, syslog_time);
1614 if (n - syslog_partial <= size) {
1615 /* message fits into buffer, move forward */
1616 syslog_seq = r.info->seq + 1;
1617 n -= syslog_partial;
1618 syslog_partial = 0;
1619 } else if (!len){
1620 /* partial read(), remember position */
1621 n = size;
1622 syslog_partial += n;
1623 } else
1624 n = 0;
1625
1626 if (!n)
1627 break;
1628
1629 mutex_unlock(&syslog_lock);
1630 err = copy_to_user(buf, text + skip, n);
1631 mutex_lock(&syslog_lock);
1632
1633 if (err) {
1634 if (!len)
1635 len = -EFAULT;
1636 break;
1637 }
1638
1639 len += n;
1640 size -= n;
1641 buf += n;
1642 } while (size);
1643 out:
1644 mutex_unlock(&syslog_lock);
1645 kfree(text);
1646 return len;
1647 }
1648
syslog_print_all(char __user * buf,int size,bool clear)1649 static int syslog_print_all(char __user *buf, int size, bool clear)
1650 {
1651 struct printk_info info;
1652 struct printk_record r;
1653 char *text;
1654 int len = 0;
1655 u64 seq;
1656 bool time;
1657
1658 text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1659 if (!text)
1660 return -ENOMEM;
1661
1662 time = printk_time;
1663 /*
1664 * Find first record that fits, including all following records,
1665 * into the user-provided buffer for this dump.
1666 */
1667 seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1668 size, true, time);
1669
1670 prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1671
1672 prb_for_each_record(seq, prb, seq, &r) {
1673 int textlen;
1674
1675 textlen = record_print_text(&r, true, time);
1676
1677 if (len + textlen > size) {
1678 seq--;
1679 break;
1680 }
1681
1682 if (copy_to_user(buf + len, text, textlen))
1683 len = -EFAULT;
1684 else
1685 len += textlen;
1686
1687 if (len < 0)
1688 break;
1689 }
1690
1691 if (clear) {
1692 mutex_lock(&syslog_lock);
1693 latched_seq_write(&clear_seq, seq);
1694 mutex_unlock(&syslog_lock);
1695 }
1696
1697 kfree(text);
1698 return len;
1699 }
1700
syslog_clear(void)1701 static void syslog_clear(void)
1702 {
1703 mutex_lock(&syslog_lock);
1704 latched_seq_write(&clear_seq, prb_next_seq(prb));
1705 mutex_unlock(&syslog_lock);
1706 }
1707
do_syslog(int type,char __user * buf,int len,int source)1708 int do_syslog(int type, char __user *buf, int len, int source)
1709 {
1710 struct printk_info info;
1711 bool clear = false;
1712 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1713 int error;
1714
1715 error = check_syslog_permissions(type, source);
1716 if (error)
1717 return error;
1718
1719 switch (type) {
1720 case SYSLOG_ACTION_CLOSE: /* Close log */
1721 break;
1722 case SYSLOG_ACTION_OPEN: /* Open log */
1723 break;
1724 case SYSLOG_ACTION_READ: /* Read from log */
1725 if (!buf || len < 0)
1726 return -EINVAL;
1727 if (!len)
1728 return 0;
1729 if (!access_ok(buf, len))
1730 return -EFAULT;
1731 error = syslog_print(buf, len);
1732 break;
1733 /* Read/clear last kernel messages */
1734 case SYSLOG_ACTION_READ_CLEAR:
1735 clear = true;
1736 fallthrough;
1737 /* Read last kernel messages */
1738 case SYSLOG_ACTION_READ_ALL:
1739 if (!buf || len < 0)
1740 return -EINVAL;
1741 if (!len)
1742 return 0;
1743 if (!access_ok(buf, len))
1744 return -EFAULT;
1745 error = syslog_print_all(buf, len, clear);
1746 break;
1747 /* Clear ring buffer */
1748 case SYSLOG_ACTION_CLEAR:
1749 syslog_clear();
1750 break;
1751 /* Disable logging to console */
1752 case SYSLOG_ACTION_CONSOLE_OFF:
1753 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1754 saved_console_loglevel = console_loglevel;
1755 console_loglevel = minimum_console_loglevel;
1756 break;
1757 /* Enable logging to console */
1758 case SYSLOG_ACTION_CONSOLE_ON:
1759 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1760 console_loglevel = saved_console_loglevel;
1761 saved_console_loglevel = LOGLEVEL_DEFAULT;
1762 }
1763 break;
1764 /* Set level of messages printed to console */
1765 case SYSLOG_ACTION_CONSOLE_LEVEL:
1766 if (len < 1 || len > 8)
1767 return -EINVAL;
1768 if (len < minimum_console_loglevel)
1769 len = minimum_console_loglevel;
1770 console_loglevel = len;
1771 /* Implicitly re-enable logging to console */
1772 saved_console_loglevel = LOGLEVEL_DEFAULT;
1773 break;
1774 /* Number of chars in the log buffer */
1775 case SYSLOG_ACTION_SIZE_UNREAD:
1776 mutex_lock(&syslog_lock);
1777 if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1778 /* No unread messages. */
1779 mutex_unlock(&syslog_lock);
1780 return 0;
1781 }
1782 if (info.seq != syslog_seq) {
1783 /* messages are gone, move to first one */
1784 syslog_seq = info.seq;
1785 syslog_partial = 0;
1786 }
1787 if (source == SYSLOG_FROM_PROC) {
1788 /*
1789 * Short-cut for poll(/"proc/kmsg") which simply checks
1790 * for pending data, not the size; return the count of
1791 * records, not the length.
1792 */
1793 error = prb_next_seq(prb) - syslog_seq;
1794 } else {
1795 bool time = syslog_partial ? syslog_time : printk_time;
1796 unsigned int line_count;
1797 u64 seq;
1798
1799 prb_for_each_info(syslog_seq, prb, seq, &info,
1800 &line_count) {
1801 error += get_record_print_text_size(&info, line_count,
1802 true, time);
1803 time = printk_time;
1804 }
1805 error -= syslog_partial;
1806 }
1807 mutex_unlock(&syslog_lock);
1808 break;
1809 /* Size of the log buffer */
1810 case SYSLOG_ACTION_SIZE_BUFFER:
1811 error = log_buf_len;
1812 break;
1813 default:
1814 error = -EINVAL;
1815 break;
1816 }
1817
1818 return error;
1819 }
1820
SYSCALL_DEFINE3(syslog,int,type,char __user *,buf,int,len)1821 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1822 {
1823 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1824 }
1825
1826 /*
1827 * Special console_lock variants that help to reduce the risk of soft-lockups.
1828 * They allow to pass console_lock to another printk() call using a busy wait.
1829 */
1830
1831 #ifdef CONFIG_LOCKDEP
1832 static struct lockdep_map console_owner_dep_map = {
1833 .name = "console_owner"
1834 };
1835 #endif
1836
1837 static DEFINE_RAW_SPINLOCK(console_owner_lock);
1838 static struct task_struct *console_owner;
1839 static bool console_waiter;
1840
1841 /**
1842 * console_lock_spinning_enable - mark beginning of code where another
1843 * thread might safely busy wait
1844 *
1845 * This basically converts console_lock into a spinlock. This marks
1846 * the section where the console_lock owner can not sleep, because
1847 * there may be a waiter spinning (like a spinlock). Also it must be
1848 * ready to hand over the lock at the end of the section.
1849 */
console_lock_spinning_enable(void)1850 static void console_lock_spinning_enable(void)
1851 {
1852 /*
1853 * Do not use spinning in panic(). The panic CPU wants to keep the lock.
1854 * Non-panic CPUs abandon the flush anyway.
1855 *
1856 * Just keep the lockdep annotation. The panic-CPU should avoid
1857 * taking console_owner_lock because it might cause a deadlock.
1858 * This looks like the easiest way how to prevent false lockdep
1859 * reports without handling races a lockless way.
1860 */
1861 if (panic_in_progress())
1862 goto lockdep;
1863
1864 raw_spin_lock(&console_owner_lock);
1865 console_owner = current;
1866 raw_spin_unlock(&console_owner_lock);
1867
1868 lockdep:
1869 /* The waiter may spin on us after setting console_owner */
1870 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1871 }
1872
1873 /**
1874 * console_lock_spinning_disable_and_check - mark end of code where another
1875 * thread was able to busy wait and check if there is a waiter
1876 * @cookie: cookie returned from console_srcu_read_lock()
1877 *
1878 * This is called at the end of the section where spinning is allowed.
1879 * It has two functions. First, it is a signal that it is no longer
1880 * safe to start busy waiting for the lock. Second, it checks if
1881 * there is a busy waiter and passes the lock rights to her.
1882 *
1883 * Important: Callers lose both the console_lock and the SRCU read lock if
1884 * there was a busy waiter. They must not touch items synchronized by
1885 * console_lock or SRCU read lock in this case.
1886 *
1887 * Return: 1 if the lock rights were passed, 0 otherwise.
1888 */
console_lock_spinning_disable_and_check(int cookie)1889 static int console_lock_spinning_disable_and_check(int cookie)
1890 {
1891 int waiter;
1892
1893 /*
1894 * Ignore spinning waiters during panic() because they might get stopped
1895 * or blocked at any time,
1896 *
1897 * It is safe because nobody is allowed to start spinning during panic
1898 * in the first place. If there has been a waiter then non panic CPUs
1899 * might stay spinning. They would get stopped anyway. The panic context
1900 * will never start spinning and an interrupted spin on panic CPU will
1901 * never continue.
1902 */
1903 if (panic_in_progress()) {
1904 /* Keep lockdep happy. */
1905 spin_release(&console_owner_dep_map, _THIS_IP_);
1906 return 0;
1907 }
1908
1909 raw_spin_lock(&console_owner_lock);
1910 waiter = READ_ONCE(console_waiter);
1911 console_owner = NULL;
1912 raw_spin_unlock(&console_owner_lock);
1913
1914 if (!waiter) {
1915 spin_release(&console_owner_dep_map, _THIS_IP_);
1916 return 0;
1917 }
1918
1919 /* The waiter is now free to continue */
1920 WRITE_ONCE(console_waiter, false);
1921
1922 spin_release(&console_owner_dep_map, _THIS_IP_);
1923
1924 /*
1925 * Preserve lockdep lock ordering. Release the SRCU read lock before
1926 * releasing the console_lock.
1927 */
1928 console_srcu_read_unlock(cookie);
1929
1930 /*
1931 * Hand off console_lock to waiter. The waiter will perform
1932 * the up(). After this, the waiter is the console_lock owner.
1933 */
1934 mutex_release(&console_lock_dep_map, _THIS_IP_);
1935 return 1;
1936 }
1937
1938 /**
1939 * console_trylock_spinning - try to get console_lock by busy waiting
1940 *
1941 * This allows to busy wait for the console_lock when the current
1942 * owner is running in specially marked sections. It means that
1943 * the current owner is running and cannot reschedule until it
1944 * is ready to lose the lock.
1945 *
1946 * Return: 1 if we got the lock, 0 othrewise
1947 */
console_trylock_spinning(void)1948 static int console_trylock_spinning(void)
1949 {
1950 struct task_struct *owner = NULL;
1951 bool waiter;
1952 bool spin = false;
1953 unsigned long flags;
1954
1955 if (console_trylock())
1956 return 1;
1957
1958 /*
1959 * It's unsafe to spin once a panic has begun. If we are the
1960 * panic CPU, we may have already halted the owner of the
1961 * console_sem. If we are not the panic CPU, then we should
1962 * avoid taking console_sem, so the panic CPU has a better
1963 * chance of cleanly acquiring it later.
1964 */
1965 if (panic_in_progress())
1966 return 0;
1967
1968 printk_safe_enter_irqsave(flags);
1969
1970 raw_spin_lock(&console_owner_lock);
1971 owner = READ_ONCE(console_owner);
1972 waiter = READ_ONCE(console_waiter);
1973 if (!waiter && owner && owner != current) {
1974 WRITE_ONCE(console_waiter, true);
1975 spin = true;
1976 }
1977 raw_spin_unlock(&console_owner_lock);
1978
1979 /*
1980 * If there is an active printk() writing to the
1981 * consoles, instead of having it write our data too,
1982 * see if we can offload that load from the active
1983 * printer, and do some printing ourselves.
1984 * Go into a spin only if there isn't already a waiter
1985 * spinning, and there is an active printer, and
1986 * that active printer isn't us (recursive printk?).
1987 */
1988 if (!spin) {
1989 printk_safe_exit_irqrestore(flags);
1990 return 0;
1991 }
1992
1993 /* We spin waiting for the owner to release us */
1994 spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1995 /* Owner will clear console_waiter on hand off */
1996 while (READ_ONCE(console_waiter))
1997 cpu_relax();
1998 spin_release(&console_owner_dep_map, _THIS_IP_);
1999
2000 printk_safe_exit_irqrestore(flags);
2001 /*
2002 * The owner passed the console lock to us.
2003 * Since we did not spin on console lock, annotate
2004 * this as a trylock. Otherwise lockdep will
2005 * complain.
2006 */
2007 mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
2008
2009 /*
2010 * Update @console_may_schedule for trylock because the previous
2011 * owner may have been schedulable.
2012 */
2013 console_may_schedule = 0;
2014
2015 return 1;
2016 }
2017
2018 /*
2019 * Recursion is tracked separately on each CPU. If NMIs are supported, an
2020 * additional NMI context per CPU is also separately tracked. Until per-CPU
2021 * is available, a separate "early tracking" is performed.
2022 */
2023 static DEFINE_PER_CPU(u8, printk_count);
2024 static u8 printk_count_early;
2025 #ifdef CONFIG_HAVE_NMI
2026 static DEFINE_PER_CPU(u8, printk_count_nmi);
2027 static u8 printk_count_nmi_early;
2028 #endif
2029
2030 /*
2031 * Recursion is limited to keep the output sane. printk() should not require
2032 * more than 1 level of recursion (allowing, for example, printk() to trigger
2033 * a WARN), but a higher value is used in case some printk-internal errors
2034 * exist, such as the ringbuffer validation checks failing.
2035 */
2036 #define PRINTK_MAX_RECURSION 3
2037
2038 /*
2039 * Return a pointer to the dedicated counter for the CPU+context of the
2040 * caller.
2041 */
__printk_recursion_counter(void)2042 static u8 *__printk_recursion_counter(void)
2043 {
2044 #ifdef CONFIG_HAVE_NMI
2045 if (in_nmi()) {
2046 if (printk_percpu_data_ready())
2047 return this_cpu_ptr(&printk_count_nmi);
2048 return &printk_count_nmi_early;
2049 }
2050 #endif
2051 if (printk_percpu_data_ready())
2052 return this_cpu_ptr(&printk_count);
2053 return &printk_count_early;
2054 }
2055
2056 /*
2057 * Enter recursion tracking. Interrupts are disabled to simplify tracking.
2058 * The caller must check the boolean return value to see if the recursion is
2059 * allowed. On failure, interrupts are not disabled.
2060 *
2061 * @recursion_ptr must be a variable of type (u8 *) and is the same variable
2062 * that is passed to printk_exit_irqrestore().
2063 */
2064 #define printk_enter_irqsave(recursion_ptr, flags) \
2065 ({ \
2066 bool success = true; \
2067 \
2068 typecheck(u8 *, recursion_ptr); \
2069 local_irq_save(flags); \
2070 (recursion_ptr) = __printk_recursion_counter(); \
2071 if (*(recursion_ptr) > PRINTK_MAX_RECURSION) { \
2072 local_irq_restore(flags); \
2073 success = false; \
2074 } else { \
2075 (*(recursion_ptr))++; \
2076 } \
2077 success; \
2078 })
2079
2080 /* Exit recursion tracking, restoring interrupts. */
2081 #define printk_exit_irqrestore(recursion_ptr, flags) \
2082 do { \
2083 typecheck(u8 *, recursion_ptr); \
2084 (*(recursion_ptr))--; \
2085 local_irq_restore(flags); \
2086 } while (0)
2087
2088 int printk_delay_msec __read_mostly;
2089
printk_delay(int level)2090 static inline void printk_delay(int level)
2091 {
2092 boot_delay_msec(level);
2093
2094 if (unlikely(printk_delay_msec)) {
2095 int m = printk_delay_msec;
2096
2097 while (m--) {
2098 mdelay(1);
2099 touch_nmi_watchdog();
2100 }
2101 }
2102 }
2103
printk_caller_id(void)2104 static inline u32 printk_caller_id(void)
2105 {
2106 return in_task() ? task_pid_nr(current) :
2107 0x80000000 + smp_processor_id();
2108 }
2109
2110 /**
2111 * printk_parse_prefix - Parse level and control flags.
2112 *
2113 * @text: The terminated text message.
2114 * @level: A pointer to the current level value, will be updated.
2115 * @flags: A pointer to the current printk_info flags, will be updated.
2116 *
2117 * @level may be NULL if the caller is not interested in the parsed value.
2118 * Otherwise the variable pointed to by @level must be set to
2119 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2120 *
2121 * @flags may be NULL if the caller is not interested in the parsed value.
2122 * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2123 * value.
2124 *
2125 * Return: The length of the parsed level and control flags.
2126 */
printk_parse_prefix(const char * text,int * level,enum printk_info_flags * flags)2127 u16 printk_parse_prefix(const char *text, int *level,
2128 enum printk_info_flags *flags)
2129 {
2130 u16 prefix_len = 0;
2131 int kern_level;
2132
2133 while (*text) {
2134 kern_level = printk_get_level(text);
2135 if (!kern_level)
2136 break;
2137
2138 switch (kern_level) {
2139 case '0' ... '7':
2140 if (level && *level == LOGLEVEL_DEFAULT)
2141 *level = kern_level - '0';
2142 break;
2143 case 'c': /* KERN_CONT */
2144 if (flags)
2145 *flags |= LOG_CONT;
2146 }
2147
2148 prefix_len += 2;
2149 text += 2;
2150 }
2151
2152 return prefix_len;
2153 }
2154
2155 __printf(5, 0)
printk_sprint(char * text,u16 size,int facility,enum printk_info_flags * flags,const char * fmt,va_list args)2156 static u16 printk_sprint(char *text, u16 size, int facility,
2157 enum printk_info_flags *flags, const char *fmt,
2158 va_list args)
2159 {
2160 u16 text_len;
2161
2162 text_len = vscnprintf(text, size, fmt, args);
2163
2164 /* Mark and strip a trailing newline. */
2165 if (text_len && text[text_len - 1] == '\n') {
2166 text_len--;
2167 *flags |= LOG_NEWLINE;
2168 }
2169
2170 /* Strip log level and control flags. */
2171 if (facility == 0) {
2172 u16 prefix_len;
2173
2174 prefix_len = printk_parse_prefix(text, NULL, NULL);
2175 if (prefix_len) {
2176 text_len -= prefix_len;
2177 memmove(text, text + prefix_len, text_len);
2178 }
2179 }
2180
2181 trace_console(text, text_len);
2182
2183 return text_len;
2184 }
2185
2186 __printf(4, 0)
vprintk_store(int facility,int level,const struct dev_printk_info * dev_info,const char * fmt,va_list args)2187 int vprintk_store(int facility, int level,
2188 const struct dev_printk_info *dev_info,
2189 const char *fmt, va_list args)
2190 {
2191 struct prb_reserved_entry e;
2192 enum printk_info_flags flags = 0;
2193 struct printk_record r;
2194 unsigned long irqflags;
2195 u16 trunc_msg_len = 0;
2196 char prefix_buf[8];
2197 u8 *recursion_ptr;
2198 u16 reserve_size;
2199 va_list args2;
2200 u32 caller_id;
2201 u16 text_len;
2202 int ret = 0;
2203 u64 ts_nsec;
2204
2205 if (!printk_enter_irqsave(recursion_ptr, irqflags))
2206 return 0;
2207
2208 /*
2209 * Since the duration of printk() can vary depending on the message
2210 * and state of the ringbuffer, grab the timestamp now so that it is
2211 * close to the call of printk(). This provides a more deterministic
2212 * timestamp with respect to the caller.
2213 */
2214 ts_nsec = local_clock();
2215
2216 caller_id = printk_caller_id();
2217
2218 /*
2219 * The sprintf needs to come first since the syslog prefix might be
2220 * passed in as a parameter. An extra byte must be reserved so that
2221 * later the vscnprintf() into the reserved buffer has room for the
2222 * terminating '\0', which is not counted by vsnprintf().
2223 */
2224 va_copy(args2, args);
2225 reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2226 va_end(args2);
2227
2228 if (reserve_size > PRINTKRB_RECORD_MAX)
2229 reserve_size = PRINTKRB_RECORD_MAX;
2230
2231 /* Extract log level or control flags. */
2232 if (facility == 0)
2233 printk_parse_prefix(&prefix_buf[0], &level, &flags);
2234
2235 if (level == LOGLEVEL_DEFAULT)
2236 level = default_message_loglevel;
2237
2238 if (dev_info)
2239 flags |= LOG_NEWLINE;
2240
2241 if (flags & LOG_CONT) {
2242 prb_rec_init_wr(&r, reserve_size);
2243 if (prb_reserve_in_last(&e, prb, &r, caller_id, PRINTKRB_RECORD_MAX)) {
2244 text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2245 facility, &flags, fmt, args);
2246 r.info->text_len += text_len;
2247
2248 if (flags & LOG_NEWLINE) {
2249 r.info->flags |= LOG_NEWLINE;
2250 prb_final_commit(&e);
2251 } else {
2252 prb_commit(&e);
2253 }
2254
2255 ret = text_len;
2256 goto out;
2257 }
2258 }
2259
2260 /*
2261 * Explicitly initialize the record before every prb_reserve() call.
2262 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2263 * structure when they fail.
2264 */
2265 prb_rec_init_wr(&r, reserve_size);
2266 if (!prb_reserve(&e, prb, &r)) {
2267 /* truncate the message if it is too long for empty buffer */
2268 truncate_msg(&reserve_size, &trunc_msg_len);
2269
2270 prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2271 if (!prb_reserve(&e, prb, &r))
2272 goto out;
2273 }
2274
2275 /* fill message */
2276 text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2277 if (trunc_msg_len)
2278 memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2279 r.info->text_len = text_len + trunc_msg_len;
2280 r.info->facility = facility;
2281 r.info->level = level & 7;
2282 r.info->flags = flags & 0x1f;
2283 r.info->ts_nsec = ts_nsec;
2284 r.info->caller_id = caller_id;
2285 if (dev_info)
2286 memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2287
2288 /* A message without a trailing newline can be continued. */
2289 if (!(flags & LOG_NEWLINE))
2290 prb_commit(&e);
2291 else
2292 prb_final_commit(&e);
2293
2294 ret = text_len + trunc_msg_len;
2295 out:
2296 printk_exit_irqrestore(recursion_ptr, irqflags);
2297 return ret;
2298 }
2299
vprintk_emit(int facility,int level,const struct dev_printk_info * dev_info,const char * fmt,va_list args)2300 asmlinkage int vprintk_emit(int facility, int level,
2301 const struct dev_printk_info *dev_info,
2302 const char *fmt, va_list args)
2303 {
2304 int printed_len;
2305 bool in_sched = false;
2306
2307 /* Suppress unimportant messages after panic happens */
2308 if (unlikely(suppress_printk))
2309 return 0;
2310
2311 /*
2312 * The messages on the panic CPU are the most important. If
2313 * non-panic CPUs are generating any messages, they will be
2314 * silently dropped.
2315 */
2316 if (other_cpu_in_panic())
2317 return 0;
2318
2319 if (level == LOGLEVEL_SCHED) {
2320 level = LOGLEVEL_DEFAULT;
2321 in_sched = true;
2322 }
2323
2324 printk_delay(level);
2325
2326 printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2327
2328 /* If called from the scheduler, we can not call up(). */
2329 if (!in_sched) {
2330 /*
2331 * The caller may be holding system-critical or
2332 * timing-sensitive locks. Disable preemption during
2333 * printing of all remaining records to all consoles so that
2334 * this context can return as soon as possible. Hopefully
2335 * another printk() caller will take over the printing.
2336 */
2337 preempt_disable();
2338 /*
2339 * Try to acquire and then immediately release the console
2340 * semaphore. The release will print out buffers. With the
2341 * spinning variant, this context tries to take over the
2342 * printing from another printing context.
2343 */
2344 if (console_trylock_spinning())
2345 console_unlock();
2346 preempt_enable();
2347 }
2348
2349 if (in_sched)
2350 defer_console_output();
2351 else
2352 wake_up_klogd();
2353
2354 return printed_len;
2355 }
2356 EXPORT_SYMBOL(vprintk_emit);
2357
vprintk_default(const char * fmt,va_list args)2358 int vprintk_default(const char *fmt, va_list args)
2359 {
2360 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2361 }
2362 EXPORT_SYMBOL_GPL(vprintk_default);
2363
_printk(const char * fmt,...)2364 asmlinkage __visible int _printk(const char *fmt, ...)
2365 {
2366 va_list args;
2367 int r;
2368
2369 va_start(args, fmt);
2370 r = vprintk(fmt, args);
2371 va_end(args);
2372
2373 return r;
2374 }
2375 EXPORT_SYMBOL(_printk);
2376
2377 static bool pr_flush(int timeout_ms, bool reset_on_progress);
2378 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress);
2379
2380 #else /* CONFIG_PRINTK */
2381
2382 #define printk_time false
2383
2384 #define prb_read_valid(rb, seq, r) false
2385 #define prb_first_valid_seq(rb) 0
2386 #define prb_next_seq(rb) 0
2387
2388 static u64 syslog_seq;
2389
pr_flush(int timeout_ms,bool reset_on_progress)2390 static bool pr_flush(int timeout_ms, bool reset_on_progress) { return true; }
__pr_flush(struct console * con,int timeout_ms,bool reset_on_progress)2391 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress) { return true; }
2392
2393 #endif /* CONFIG_PRINTK */
2394
2395 #ifdef CONFIG_EARLY_PRINTK
2396 struct console *early_console;
2397
early_printk(const char * fmt,...)2398 asmlinkage __visible void early_printk(const char *fmt, ...)
2399 {
2400 va_list ap;
2401 char buf[512];
2402 int n;
2403
2404 if (!early_console)
2405 return;
2406
2407 va_start(ap, fmt);
2408 n = vscnprintf(buf, sizeof(buf), fmt, ap);
2409 va_end(ap);
2410
2411 early_console->write(early_console, buf, n);
2412 }
2413 #endif
2414
set_user_specified(struct console_cmdline * c,bool user_specified)2415 static void set_user_specified(struct console_cmdline *c, bool user_specified)
2416 {
2417 if (!user_specified)
2418 return;
2419
2420 /*
2421 * @c console was defined by the user on the command line.
2422 * Do not clear when added twice also by SPCR or the device tree.
2423 */
2424 c->user_specified = true;
2425 /* At least one console defined by the user on the command line. */
2426 console_set_on_cmdline = 1;
2427 }
2428
__add_preferred_console(const char * name,const short idx,char * options,char * brl_options,bool user_specified)2429 static int __add_preferred_console(const char *name, const short idx, char *options,
2430 char *brl_options, bool user_specified)
2431 {
2432 struct console_cmdline *c;
2433 int i;
2434
2435 /*
2436 * We use a signed short index for struct console for device drivers to
2437 * indicate a not yet assigned index or port. However, a negative index
2438 * value is not valid for preferred console.
2439 */
2440 if (idx < 0)
2441 return -EINVAL;
2442
2443 /*
2444 * See if this tty is not yet registered, and
2445 * if we have a slot free.
2446 */
2447 for (i = 0, c = console_cmdline;
2448 i < MAX_CMDLINECONSOLES && c->name[0];
2449 i++, c++) {
2450 if (strcmp(c->name, name) == 0 && c->index == idx) {
2451 if (!brl_options)
2452 preferred_console = i;
2453 set_user_specified(c, user_specified);
2454 return 0;
2455 }
2456 }
2457 if (i == MAX_CMDLINECONSOLES)
2458 return -E2BIG;
2459 if (!brl_options)
2460 preferred_console = i;
2461 strscpy(c->name, name, sizeof(c->name));
2462 c->options = options;
2463 set_user_specified(c, user_specified);
2464 braille_set_options(c, brl_options);
2465
2466 c->index = idx;
2467 return 0;
2468 }
2469
console_msg_format_setup(char * str)2470 static int __init console_msg_format_setup(char *str)
2471 {
2472 if (!strcmp(str, "syslog"))
2473 console_msg_format = MSG_FORMAT_SYSLOG;
2474 if (!strcmp(str, "default"))
2475 console_msg_format = MSG_FORMAT_DEFAULT;
2476 return 1;
2477 }
2478 __setup("console_msg_format=", console_msg_format_setup);
2479
2480 /*
2481 * Set up a console. Called via do_early_param() in init/main.c
2482 * for each "console=" parameter in the boot command line.
2483 */
console_setup(char * str)2484 static int __init console_setup(char *str)
2485 {
2486 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2487 char *s, *options, *brl_options = NULL;
2488 int idx;
2489
2490 /*
2491 * console="" or console=null have been suggested as a way to
2492 * disable console output. Use ttynull that has been created
2493 * for exactly this purpose.
2494 */
2495 if (str[0] == 0 || strcmp(str, "null") == 0) {
2496 __add_preferred_console("ttynull", 0, NULL, NULL, true);
2497 return 1;
2498 }
2499
2500 if (_braille_console_setup(&str, &brl_options))
2501 return 1;
2502
2503 /* Save the console for driver subsystem use */
2504 if (console_opt_save(str, brl_options))
2505 return 1;
2506
2507 /* Flag register_console() to not call try_enable_default_console() */
2508 console_set_on_cmdline = 1;
2509
2510 /* Don't attempt to parse a DEVNAME:0.0 style console */
2511 if (strchr(str, ':'))
2512 return 1;
2513
2514 /*
2515 * Decode str into name, index, options.
2516 */
2517 if (isdigit(str[0]))
2518 scnprintf(buf, sizeof(buf), "ttyS%s", str);
2519 else
2520 strscpy(buf, str);
2521
2522 options = strchr(str, ',');
2523 if (options)
2524 *(options++) = 0;
2525
2526 #ifdef __sparc__
2527 if (!strcmp(str, "ttya"))
2528 strscpy(buf, "ttyS0");
2529 if (!strcmp(str, "ttyb"))
2530 strscpy(buf, "ttyS1");
2531 #endif
2532
2533 for (s = buf; *s; s++)
2534 if (isdigit(*s) || *s == ',')
2535 break;
2536 idx = simple_strtoul(s, NULL, 10);
2537 *s = 0;
2538
2539 __add_preferred_console(buf, idx, options, brl_options, true);
2540 return 1;
2541 }
2542 __setup("console=", console_setup);
2543
2544 /* Only called from add_preferred_console_match() */
console_opt_add_preferred_console(const char * name,const short idx,char * options,char * brl_options)2545 int console_opt_add_preferred_console(const char *name, const short idx,
2546 char *options, char *brl_options)
2547 {
2548 return __add_preferred_console(name, idx, options, brl_options, true);
2549 }
2550
2551 /**
2552 * add_preferred_console - add a device to the list of preferred consoles.
2553 * @name: device name
2554 * @idx: device index
2555 * @options: options for this console
2556 *
2557 * The last preferred console added will be used for kernel messages
2558 * and stdin/out/err for init. Normally this is used by console_setup
2559 * above to handle user-supplied console arguments; however it can also
2560 * be used by arch-specific code either to override the user or more
2561 * commonly to provide a default console (ie from PROM variables) when
2562 * the user has not supplied one.
2563 */
add_preferred_console(const char * name,const short idx,char * options)2564 int add_preferred_console(const char *name, const short idx, char *options)
2565 {
2566 return __add_preferred_console(name, idx, options, NULL, false);
2567 }
2568
2569 bool console_suspend_enabled = true;
2570 EXPORT_SYMBOL(console_suspend_enabled);
2571
console_suspend_disable(char * str)2572 static int __init console_suspend_disable(char *str)
2573 {
2574 console_suspend_enabled = false;
2575 return 1;
2576 }
2577 __setup("no_console_suspend", console_suspend_disable);
2578 module_param_named(console_suspend, console_suspend_enabled,
2579 bool, S_IRUGO | S_IWUSR);
2580 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2581 " and hibernate operations");
2582
2583 static bool printk_console_no_auto_verbose;
2584
console_verbose(void)2585 void console_verbose(void)
2586 {
2587 if (console_loglevel && !printk_console_no_auto_verbose)
2588 console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2589 }
2590 EXPORT_SYMBOL_GPL(console_verbose);
2591
2592 module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2593 MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2594
2595 /**
2596 * suspend_console - suspend the console subsystem
2597 *
2598 * This disables printk() while we go into suspend states
2599 */
suspend_console(void)2600 void suspend_console(void)
2601 {
2602 struct console *con;
2603
2604 if (!console_suspend_enabled)
2605 return;
2606 pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2607 pr_flush(1000, true);
2608
2609 console_list_lock();
2610 for_each_console(con)
2611 console_srcu_write_flags(con, con->flags | CON_SUSPENDED);
2612 console_list_unlock();
2613
2614 /*
2615 * Ensure that all SRCU list walks have completed. All printing
2616 * contexts must be able to see that they are suspended so that it
2617 * is guaranteed that all printing has stopped when this function
2618 * completes.
2619 */
2620 synchronize_srcu(&console_srcu);
2621 }
2622
resume_console(void)2623 void resume_console(void)
2624 {
2625 struct console *con;
2626
2627 if (!console_suspend_enabled)
2628 return;
2629
2630 console_list_lock();
2631 for_each_console(con)
2632 console_srcu_write_flags(con, con->flags & ~CON_SUSPENDED);
2633 console_list_unlock();
2634
2635 /*
2636 * Ensure that all SRCU list walks have completed. All printing
2637 * contexts must be able to see they are no longer suspended so
2638 * that they are guaranteed to wake up and resume printing.
2639 */
2640 synchronize_srcu(&console_srcu);
2641
2642 pr_flush(1000, true);
2643 }
2644
2645 /**
2646 * console_cpu_notify - print deferred console messages after CPU hotplug
2647 * @cpu: unused
2648 *
2649 * If printk() is called from a CPU that is not online yet, the messages
2650 * will be printed on the console only if there are CON_ANYTIME consoles.
2651 * This function is called when a new CPU comes online (or fails to come
2652 * up) or goes offline.
2653 */
console_cpu_notify(unsigned int cpu)2654 static int console_cpu_notify(unsigned int cpu)
2655 {
2656 if (!cpuhp_tasks_frozen) {
2657 /* If trylock fails, someone else is doing the printing */
2658 if (console_trylock())
2659 console_unlock();
2660 }
2661 return 0;
2662 }
2663
2664 /**
2665 * console_lock - block the console subsystem from printing
2666 *
2667 * Acquires a lock which guarantees that no consoles will
2668 * be in or enter their write() callback.
2669 *
2670 * Can sleep, returns nothing.
2671 */
console_lock(void)2672 void console_lock(void)
2673 {
2674 might_sleep();
2675
2676 /* On panic, the console_lock must be left to the panic cpu. */
2677 while (other_cpu_in_panic())
2678 msleep(1000);
2679
2680 down_console_sem();
2681 console_locked = 1;
2682 console_may_schedule = 1;
2683 }
2684 EXPORT_SYMBOL(console_lock);
2685
2686 /**
2687 * console_trylock - try to block the console subsystem from printing
2688 *
2689 * Try to acquire a lock which guarantees that no consoles will
2690 * be in or enter their write() callback.
2691 *
2692 * returns 1 on success, and 0 on failure to acquire the lock.
2693 */
console_trylock(void)2694 int console_trylock(void)
2695 {
2696 /* On panic, the console_lock must be left to the panic cpu. */
2697 if (other_cpu_in_panic())
2698 return 0;
2699 if (down_trylock_console_sem())
2700 return 0;
2701 console_locked = 1;
2702 console_may_schedule = 0;
2703 return 1;
2704 }
2705 EXPORT_SYMBOL(console_trylock);
2706
is_console_locked(void)2707 int is_console_locked(void)
2708 {
2709 return console_locked;
2710 }
2711 EXPORT_SYMBOL(is_console_locked);
2712
2713 /*
2714 * Check if the given console is currently capable and allowed to print
2715 * records.
2716 *
2717 * Requires the console_srcu_read_lock.
2718 */
console_is_usable(struct console * con)2719 static inline bool console_is_usable(struct console *con)
2720 {
2721 short flags = console_srcu_read_flags(con);
2722
2723 if (!(flags & CON_ENABLED))
2724 return false;
2725
2726 if ((flags & CON_SUSPENDED))
2727 return false;
2728
2729 if (!con->write)
2730 return false;
2731
2732 /*
2733 * Console drivers may assume that per-cpu resources have been
2734 * allocated. So unless they're explicitly marked as being able to
2735 * cope (CON_ANYTIME) don't call them until this CPU is officially up.
2736 */
2737 if (!cpu_online(raw_smp_processor_id()) && !(flags & CON_ANYTIME))
2738 return false;
2739
2740 return true;
2741 }
2742
__console_unlock(void)2743 static void __console_unlock(void)
2744 {
2745 console_locked = 0;
2746 up_console_sem();
2747 }
2748
2749 #ifdef CONFIG_PRINTK
2750
2751 /*
2752 * Prepend the message in @pmsg->pbufs->outbuf with a "dropped message". This
2753 * is achieved by shifting the existing message over and inserting the dropped
2754 * message.
2755 *
2756 * @pmsg is the printk message to prepend.
2757 *
2758 * @dropped is the dropped count to report in the dropped message.
2759 *
2760 * If the message text in @pmsg->pbufs->outbuf does not have enough space for
2761 * the dropped message, the message text will be sufficiently truncated.
2762 *
2763 * If @pmsg->pbufs->outbuf is modified, @pmsg->outbuf_len is updated.
2764 */
console_prepend_dropped(struct printk_message * pmsg,unsigned long dropped)2765 void console_prepend_dropped(struct printk_message *pmsg, unsigned long dropped)
2766 {
2767 struct printk_buffers *pbufs = pmsg->pbufs;
2768 const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2769 const size_t outbuf_sz = sizeof(pbufs->outbuf);
2770 char *scratchbuf = &pbufs->scratchbuf[0];
2771 char *outbuf = &pbufs->outbuf[0];
2772 size_t len;
2773
2774 len = scnprintf(scratchbuf, scratchbuf_sz,
2775 "** %lu printk messages dropped **\n", dropped);
2776
2777 /*
2778 * Make sure outbuf is sufficiently large before prepending.
2779 * Keep at least the prefix when the message must be truncated.
2780 * It is a rather theoretical problem when someone tries to
2781 * use a minimalist buffer.
2782 */
2783 if (WARN_ON_ONCE(len + PRINTK_PREFIX_MAX >= outbuf_sz))
2784 return;
2785
2786 if (pmsg->outbuf_len + len >= outbuf_sz) {
2787 /* Truncate the message, but keep it terminated. */
2788 pmsg->outbuf_len = outbuf_sz - (len + 1);
2789 outbuf[pmsg->outbuf_len] = 0;
2790 }
2791
2792 memmove(outbuf + len, outbuf, pmsg->outbuf_len + 1);
2793 memcpy(outbuf, scratchbuf, len);
2794 pmsg->outbuf_len += len;
2795 }
2796
2797 /*
2798 * Read and format the specified record (or a later record if the specified
2799 * record is not available).
2800 *
2801 * @pmsg will contain the formatted result. @pmsg->pbufs must point to a
2802 * struct printk_buffers.
2803 *
2804 * @seq is the record to read and format. If it is not available, the next
2805 * valid record is read.
2806 *
2807 * @is_extended specifies if the message should be formatted for extended
2808 * console output.
2809 *
2810 * @may_supress specifies if records may be skipped based on loglevel.
2811 *
2812 * Returns false if no record is available. Otherwise true and all fields
2813 * of @pmsg are valid. (See the documentation of struct printk_message
2814 * for information about the @pmsg fields.)
2815 */
printk_get_next_message(struct printk_message * pmsg,u64 seq,bool is_extended,bool may_suppress)2816 bool printk_get_next_message(struct printk_message *pmsg, u64 seq,
2817 bool is_extended, bool may_suppress)
2818 {
2819 struct printk_buffers *pbufs = pmsg->pbufs;
2820 const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2821 const size_t outbuf_sz = sizeof(pbufs->outbuf);
2822 char *scratchbuf = &pbufs->scratchbuf[0];
2823 char *outbuf = &pbufs->outbuf[0];
2824 struct printk_info info;
2825 struct printk_record r;
2826 size_t len = 0;
2827
2828 /*
2829 * Formatting extended messages requires a separate buffer, so use the
2830 * scratch buffer to read in the ringbuffer text.
2831 *
2832 * Formatting normal messages is done in-place, so read the ringbuffer
2833 * text directly into the output buffer.
2834 */
2835 if (is_extended)
2836 prb_rec_init_rd(&r, &info, scratchbuf, scratchbuf_sz);
2837 else
2838 prb_rec_init_rd(&r, &info, outbuf, outbuf_sz);
2839
2840 if (!prb_read_valid(prb, seq, &r))
2841 return false;
2842
2843 pmsg->seq = r.info->seq;
2844 pmsg->dropped = r.info->seq - seq;
2845
2846 /* Skip record that has level above the console loglevel. */
2847 if (may_suppress && suppress_message_printing(r.info->level))
2848 goto out;
2849
2850 if (is_extended) {
2851 len = info_print_ext_header(outbuf, outbuf_sz, r.info);
2852 len += msg_print_ext_body(outbuf + len, outbuf_sz - len,
2853 &r.text_buf[0], r.info->text_len, &r.info->dev_info);
2854 } else {
2855 len = record_print_text(&r, console_msg_format & MSG_FORMAT_SYSLOG, printk_time);
2856 }
2857 out:
2858 pmsg->outbuf_len = len;
2859 return true;
2860 }
2861
2862 /*
2863 * Used as the printk buffers for non-panic, serialized console printing.
2864 * This is for legacy (!CON_NBCON) as well as all boot (CON_BOOT) consoles.
2865 * Its usage requires the console_lock held.
2866 */
2867 struct printk_buffers printk_shared_pbufs;
2868
2869 /*
2870 * Print one record for the given console. The record printed is whatever
2871 * record is the next available record for the given console.
2872 *
2873 * @handover will be set to true if a printk waiter has taken over the
2874 * console_lock, in which case the caller is no longer holding both the
2875 * console_lock and the SRCU read lock. Otherwise it is set to false.
2876 *
2877 * @cookie is the cookie from the SRCU read lock.
2878 *
2879 * Returns false if the given console has no next record to print, otherwise
2880 * true.
2881 *
2882 * Requires the console_lock and the SRCU read lock.
2883 */
console_emit_next_record(struct console * con,bool * handover,int cookie)2884 static bool console_emit_next_record(struct console *con, bool *handover, int cookie)
2885 {
2886 bool is_extended = console_srcu_read_flags(con) & CON_EXTENDED;
2887 char *outbuf = &printk_shared_pbufs.outbuf[0];
2888 struct printk_message pmsg = {
2889 .pbufs = &printk_shared_pbufs,
2890 };
2891 unsigned long flags;
2892
2893 *handover = false;
2894
2895 if (!printk_get_next_message(&pmsg, con->seq, is_extended, true))
2896 return false;
2897
2898 con->dropped += pmsg.dropped;
2899
2900 /* Skip messages of formatted length 0. */
2901 if (pmsg.outbuf_len == 0) {
2902 con->seq = pmsg.seq + 1;
2903 goto skip;
2904 }
2905
2906 if (con->dropped && !is_extended) {
2907 console_prepend_dropped(&pmsg, con->dropped);
2908 con->dropped = 0;
2909 }
2910
2911 /*
2912 * While actively printing out messages, if another printk()
2913 * were to occur on another CPU, it may wait for this one to
2914 * finish. This task can not be preempted if there is a
2915 * waiter waiting to take over.
2916 *
2917 * Interrupts are disabled because the hand over to a waiter
2918 * must not be interrupted until the hand over is completed
2919 * (@console_waiter is cleared).
2920 */
2921 printk_safe_enter_irqsave(flags);
2922 console_lock_spinning_enable();
2923
2924 /* Do not trace print latency. */
2925 stop_critical_timings();
2926
2927 /* Write everything out to the hardware. */
2928 con->write(con, outbuf, pmsg.outbuf_len);
2929
2930 start_critical_timings();
2931
2932 con->seq = pmsg.seq + 1;
2933
2934 *handover = console_lock_spinning_disable_and_check(cookie);
2935 printk_safe_exit_irqrestore(flags);
2936 skip:
2937 return true;
2938 }
2939
2940 #else
2941
console_emit_next_record(struct console * con,bool * handover,int cookie)2942 static bool console_emit_next_record(struct console *con, bool *handover, int cookie)
2943 {
2944 *handover = false;
2945 return false;
2946 }
2947
2948 #endif /* CONFIG_PRINTK */
2949
2950 /*
2951 * Print out all remaining records to all consoles.
2952 *
2953 * @do_cond_resched is set by the caller. It can be true only in schedulable
2954 * context.
2955 *
2956 * @next_seq is set to the sequence number after the last available record.
2957 * The value is valid only when this function returns true. It means that all
2958 * usable consoles are completely flushed.
2959 *
2960 * @handover will be set to true if a printk waiter has taken over the
2961 * console_lock, in which case the caller is no longer holding the
2962 * console_lock. Otherwise it is set to false.
2963 *
2964 * Returns true when there was at least one usable console and all messages
2965 * were flushed to all usable consoles. A returned false informs the caller
2966 * that everything was not flushed (either there were no usable consoles or
2967 * another context has taken over printing or it is a panic situation and this
2968 * is not the panic CPU). Regardless the reason, the caller should assume it
2969 * is not useful to immediately try again.
2970 *
2971 * Requires the console_lock.
2972 */
console_flush_all(bool do_cond_resched,u64 * next_seq,bool * handover)2973 static bool console_flush_all(bool do_cond_resched, u64 *next_seq, bool *handover)
2974 {
2975 bool any_usable = false;
2976 struct console *con;
2977 bool any_progress;
2978 int cookie;
2979
2980 *next_seq = 0;
2981 *handover = false;
2982
2983 do {
2984 any_progress = false;
2985
2986 cookie = console_srcu_read_lock();
2987 for_each_console_srcu(con) {
2988 bool progress;
2989
2990 if (!console_is_usable(con))
2991 continue;
2992 any_usable = true;
2993
2994 progress = console_emit_next_record(con, handover, cookie);
2995
2996 /*
2997 * If a handover has occurred, the SRCU read lock
2998 * is already released.
2999 */
3000 if (*handover)
3001 return false;
3002
3003 /* Track the next of the highest seq flushed. */
3004 if (con->seq > *next_seq)
3005 *next_seq = con->seq;
3006
3007 if (!progress)
3008 continue;
3009 any_progress = true;
3010
3011 /* Allow panic_cpu to take over the consoles safely. */
3012 if (other_cpu_in_panic())
3013 goto abandon;
3014
3015 if (do_cond_resched)
3016 cond_resched();
3017 }
3018 console_srcu_read_unlock(cookie);
3019 } while (any_progress);
3020
3021 return any_usable;
3022
3023 abandon:
3024 console_srcu_read_unlock(cookie);
3025 return false;
3026 }
3027
3028 /**
3029 * console_unlock - unblock the console subsystem from printing
3030 *
3031 * Releases the console_lock which the caller holds to block printing of
3032 * the console subsystem.
3033 *
3034 * While the console_lock was held, console output may have been buffered
3035 * by printk(). If this is the case, console_unlock(); emits
3036 * the output prior to releasing the lock.
3037 *
3038 * console_unlock(); may be called from any context.
3039 */
console_unlock(void)3040 void console_unlock(void)
3041 {
3042 bool do_cond_resched;
3043 bool handover;
3044 bool flushed;
3045 u64 next_seq;
3046
3047 /*
3048 * Console drivers are called with interrupts disabled, so
3049 * @console_may_schedule should be cleared before; however, we may
3050 * end up dumping a lot of lines, for example, if called from
3051 * console registration path, and should invoke cond_resched()
3052 * between lines if allowable. Not doing so can cause a very long
3053 * scheduling stall on a slow console leading to RCU stall and
3054 * softlockup warnings which exacerbate the issue with more
3055 * messages practically incapacitating the system. Therefore, create
3056 * a local to use for the printing loop.
3057 */
3058 do_cond_resched = console_may_schedule;
3059
3060 do {
3061 console_may_schedule = 0;
3062
3063 flushed = console_flush_all(do_cond_resched, &next_seq, &handover);
3064 if (!handover)
3065 __console_unlock();
3066
3067 /*
3068 * Abort if there was a failure to flush all messages to all
3069 * usable consoles. Either it is not possible to flush (in
3070 * which case it would be an infinite loop of retrying) or
3071 * another context has taken over printing.
3072 */
3073 if (!flushed)
3074 break;
3075
3076 /*
3077 * Some context may have added new records after
3078 * console_flush_all() but before unlocking the console.
3079 * Re-check if there is a new record to flush. If the trylock
3080 * fails, another context is already handling the printing.
3081 */
3082 } while (prb_read_valid(prb, next_seq, NULL) && console_trylock());
3083 }
3084 EXPORT_SYMBOL(console_unlock);
3085
3086 /**
3087 * console_conditional_schedule - yield the CPU if required
3088 *
3089 * If the console code is currently allowed to sleep, and
3090 * if this CPU should yield the CPU to another task, do
3091 * so here.
3092 *
3093 * Must be called within console_lock();.
3094 */
console_conditional_schedule(void)3095 void __sched console_conditional_schedule(void)
3096 {
3097 if (console_may_schedule)
3098 cond_resched();
3099 }
3100 EXPORT_SYMBOL(console_conditional_schedule);
3101
console_unblank(void)3102 void console_unblank(void)
3103 {
3104 bool found_unblank = false;
3105 struct console *c;
3106 int cookie;
3107
3108 /*
3109 * First check if there are any consoles implementing the unblank()
3110 * callback. If not, there is no reason to continue and take the
3111 * console lock, which in particular can be dangerous if
3112 * @oops_in_progress is set.
3113 */
3114 cookie = console_srcu_read_lock();
3115 for_each_console_srcu(c) {
3116 if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank) {
3117 found_unblank = true;
3118 break;
3119 }
3120 }
3121 console_srcu_read_unlock(cookie);
3122 if (!found_unblank)
3123 return;
3124
3125 /*
3126 * Stop console printing because the unblank() callback may
3127 * assume the console is not within its write() callback.
3128 *
3129 * If @oops_in_progress is set, this may be an atomic context.
3130 * In that case, attempt a trylock as best-effort.
3131 */
3132 if (oops_in_progress) {
3133 /* Semaphores are not NMI-safe. */
3134 if (in_nmi())
3135 return;
3136
3137 /*
3138 * Attempting to trylock the console lock can deadlock
3139 * if another CPU was stopped while modifying the
3140 * semaphore. "Hope and pray" that this is not the
3141 * current situation.
3142 */
3143 if (down_trylock_console_sem() != 0)
3144 return;
3145 } else
3146 console_lock();
3147
3148 console_locked = 1;
3149 console_may_schedule = 0;
3150
3151 cookie = console_srcu_read_lock();
3152 for_each_console_srcu(c) {
3153 if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank)
3154 c->unblank();
3155 }
3156 console_srcu_read_unlock(cookie);
3157
3158 console_unlock();
3159
3160 if (!oops_in_progress)
3161 pr_flush(1000, true);
3162 }
3163
3164 /*
3165 * Rewind all consoles to the oldest available record.
3166 *
3167 * IMPORTANT: The function is safe only when called under
3168 * console_lock(). It is not enforced because
3169 * it is used as a best effort in panic().
3170 */
__console_rewind_all(void)3171 static void __console_rewind_all(void)
3172 {
3173 struct console *c;
3174 short flags;
3175 int cookie;
3176 u64 seq;
3177
3178 seq = prb_first_valid_seq(prb);
3179
3180 cookie = console_srcu_read_lock();
3181 for_each_console_srcu(c) {
3182 flags = console_srcu_read_flags(c);
3183
3184 if (flags & CON_NBCON) {
3185 nbcon_seq_force(c, seq);
3186 } else {
3187 /*
3188 * This assignment is safe only when called under
3189 * console_lock(). On panic, legacy consoles are
3190 * only best effort.
3191 */
3192 c->seq = seq;
3193 }
3194 }
3195 console_srcu_read_unlock(cookie);
3196 }
3197
3198 /**
3199 * console_flush_on_panic - flush console content on panic
3200 * @mode: flush all messages in buffer or just the pending ones
3201 *
3202 * Immediately output all pending messages no matter what.
3203 */
console_flush_on_panic(enum con_flush_mode mode)3204 void console_flush_on_panic(enum con_flush_mode mode)
3205 {
3206 bool handover;
3207 u64 next_seq;
3208
3209 /*
3210 * Ignore the console lock and flush out the messages. Attempting a
3211 * trylock would not be useful because:
3212 *
3213 * - if it is contended, it must be ignored anyway
3214 * - console_lock() and console_trylock() block and fail
3215 * respectively in panic for non-panic CPUs
3216 * - semaphores are not NMI-safe
3217 */
3218
3219 /*
3220 * If another context is holding the console lock,
3221 * @console_may_schedule might be set. Clear it so that
3222 * this context does not call cond_resched() while flushing.
3223 */
3224 console_may_schedule = 0;
3225
3226 if (mode == CONSOLE_REPLAY_ALL)
3227 __console_rewind_all();
3228
3229 console_flush_all(false, &next_seq, &handover);
3230 }
3231
3232 /*
3233 * Return the console tty driver structure and its associated index
3234 */
console_device(int * index)3235 struct tty_driver *console_device(int *index)
3236 {
3237 struct console *c;
3238 struct tty_driver *driver = NULL;
3239 int cookie;
3240
3241 /*
3242 * Take console_lock to serialize device() callback with
3243 * other console operations. For example, fg_console is
3244 * modified under console_lock when switching vt.
3245 */
3246 console_lock();
3247
3248 cookie = console_srcu_read_lock();
3249 for_each_console_srcu(c) {
3250 if (!c->device)
3251 continue;
3252 driver = c->device(c, index);
3253 if (driver)
3254 break;
3255 }
3256 console_srcu_read_unlock(cookie);
3257
3258 console_unlock();
3259 return driver;
3260 }
3261
3262 /*
3263 * Prevent further output on the passed console device so that (for example)
3264 * serial drivers can disable console output before suspending a port, and can
3265 * re-enable output afterwards.
3266 */
console_stop(struct console * console)3267 void console_stop(struct console *console)
3268 {
3269 __pr_flush(console, 1000, true);
3270 console_list_lock();
3271 console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3272 console_list_unlock();
3273
3274 /*
3275 * Ensure that all SRCU list walks have completed. All contexts must
3276 * be able to see that this console is disabled so that (for example)
3277 * the caller can suspend the port without risk of another context
3278 * using the port.
3279 */
3280 synchronize_srcu(&console_srcu);
3281 }
3282 EXPORT_SYMBOL(console_stop);
3283
console_start(struct console * console)3284 void console_start(struct console *console)
3285 {
3286 console_list_lock();
3287 console_srcu_write_flags(console, console->flags | CON_ENABLED);
3288 console_list_unlock();
3289 __pr_flush(console, 1000, true);
3290 }
3291 EXPORT_SYMBOL(console_start);
3292
3293 static int __read_mostly keep_bootcon;
3294
keep_bootcon_setup(char * str)3295 static int __init keep_bootcon_setup(char *str)
3296 {
3297 keep_bootcon = 1;
3298 pr_info("debug: skip boot console de-registration.\n");
3299
3300 return 0;
3301 }
3302
3303 early_param("keep_bootcon", keep_bootcon_setup);
3304
console_call_setup(struct console * newcon,char * options)3305 static int console_call_setup(struct console *newcon, char *options)
3306 {
3307 int err;
3308
3309 if (!newcon->setup)
3310 return 0;
3311
3312 /* Synchronize with possible boot console. */
3313 console_lock();
3314 err = newcon->setup(newcon, options);
3315 console_unlock();
3316
3317 return err;
3318 }
3319
3320 /*
3321 * This is called by register_console() to try to match
3322 * the newly registered console with any of the ones selected
3323 * by either the command line or add_preferred_console() and
3324 * setup/enable it.
3325 *
3326 * Care need to be taken with consoles that are statically
3327 * enabled such as netconsole
3328 */
try_enable_preferred_console(struct console * newcon,bool user_specified)3329 static int try_enable_preferred_console(struct console *newcon,
3330 bool user_specified)
3331 {
3332 struct console_cmdline *c;
3333 int i, err;
3334
3335 for (i = 0, c = console_cmdline;
3336 i < MAX_CMDLINECONSOLES && c->name[0];
3337 i++, c++) {
3338 if (c->user_specified != user_specified)
3339 continue;
3340 if (!newcon->match ||
3341 newcon->match(newcon, c->name, c->index, c->options) != 0) {
3342 /* default matching */
3343 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
3344 if (strcmp(c->name, newcon->name) != 0)
3345 continue;
3346 if (newcon->index >= 0 &&
3347 newcon->index != c->index)
3348 continue;
3349 if (newcon->index < 0)
3350 newcon->index = c->index;
3351
3352 if (_braille_register_console(newcon, c))
3353 return 0;
3354
3355 err = console_call_setup(newcon, c->options);
3356 if (err)
3357 return err;
3358 }
3359 newcon->flags |= CON_ENABLED;
3360 if (i == preferred_console)
3361 newcon->flags |= CON_CONSDEV;
3362 return 0;
3363 }
3364
3365 /*
3366 * Some consoles, such as pstore and netconsole, can be enabled even
3367 * without matching. Accept the pre-enabled consoles only when match()
3368 * and setup() had a chance to be called.
3369 */
3370 if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
3371 return 0;
3372
3373 return -ENOENT;
3374 }
3375
3376 /* Try to enable the console unconditionally */
try_enable_default_console(struct console * newcon)3377 static void try_enable_default_console(struct console *newcon)
3378 {
3379 if (newcon->index < 0)
3380 newcon->index = 0;
3381
3382 if (console_call_setup(newcon, NULL) != 0)
3383 return;
3384
3385 newcon->flags |= CON_ENABLED;
3386
3387 if (newcon->device)
3388 newcon->flags |= CON_CONSDEV;
3389 }
3390
console_init_seq(struct console * newcon,bool bootcon_registered)3391 static void console_init_seq(struct console *newcon, bool bootcon_registered)
3392 {
3393 struct console *con;
3394 bool handover;
3395
3396 if (newcon->flags & (CON_PRINTBUFFER | CON_BOOT)) {
3397 /* Get a consistent copy of @syslog_seq. */
3398 mutex_lock(&syslog_lock);
3399 newcon->seq = syslog_seq;
3400 mutex_unlock(&syslog_lock);
3401 } else {
3402 /* Begin with next message added to ringbuffer. */
3403 newcon->seq = prb_next_seq(prb);
3404
3405 /*
3406 * If any enabled boot consoles are due to be unregistered
3407 * shortly, some may not be caught up and may be the same
3408 * device as @newcon. Since it is not known which boot console
3409 * is the same device, flush all consoles and, if necessary,
3410 * start with the message of the enabled boot console that is
3411 * the furthest behind.
3412 */
3413 if (bootcon_registered && !keep_bootcon) {
3414 /*
3415 * Hold the console_lock to stop console printing and
3416 * guarantee safe access to console->seq.
3417 */
3418 console_lock();
3419
3420 /*
3421 * Flush all consoles and set the console to start at
3422 * the next unprinted sequence number.
3423 */
3424 if (!console_flush_all(true, &newcon->seq, &handover)) {
3425 /*
3426 * Flushing failed. Just choose the lowest
3427 * sequence of the enabled boot consoles.
3428 */
3429
3430 /*
3431 * If there was a handover, this context no
3432 * longer holds the console_lock.
3433 */
3434 if (handover)
3435 console_lock();
3436
3437 newcon->seq = prb_next_seq(prb);
3438 for_each_console(con) {
3439 if ((con->flags & CON_BOOT) &&
3440 (con->flags & CON_ENABLED) &&
3441 con->seq < newcon->seq) {
3442 newcon->seq = con->seq;
3443 }
3444 }
3445 }
3446
3447 console_unlock();
3448 }
3449 }
3450 }
3451
3452 #define console_first() \
3453 hlist_entry(console_list.first, struct console, node)
3454
3455 static int unregister_console_locked(struct console *console);
3456
3457 /*
3458 * The console driver calls this routine during kernel initialization
3459 * to register the console printing procedure with printk() and to
3460 * print any messages that were printed by the kernel before the
3461 * console driver was initialized.
3462 *
3463 * This can happen pretty early during the boot process (because of
3464 * early_printk) - sometimes before setup_arch() completes - be careful
3465 * of what kernel features are used - they may not be initialised yet.
3466 *
3467 * There are two types of consoles - bootconsoles (early_printk) and
3468 * "real" consoles (everything which is not a bootconsole) which are
3469 * handled differently.
3470 * - Any number of bootconsoles can be registered at any time.
3471 * - As soon as a "real" console is registered, all bootconsoles
3472 * will be unregistered automatically.
3473 * - Once a "real" console is registered, any attempt to register a
3474 * bootconsoles will be rejected
3475 */
register_console(struct console * newcon)3476 void register_console(struct console *newcon)
3477 {
3478 struct console *con;
3479 bool bootcon_registered = false;
3480 bool realcon_registered = false;
3481 int err;
3482
3483 console_list_lock();
3484
3485 for_each_console(con) {
3486 if (WARN(con == newcon, "console '%s%d' already registered\n",
3487 con->name, con->index)) {
3488 goto unlock;
3489 }
3490
3491 if (con->flags & CON_BOOT)
3492 bootcon_registered = true;
3493 else
3494 realcon_registered = true;
3495 }
3496
3497 /* Do not register boot consoles when there already is a real one. */
3498 if ((newcon->flags & CON_BOOT) && realcon_registered) {
3499 pr_info("Too late to register bootconsole %s%d\n",
3500 newcon->name, newcon->index);
3501 goto unlock;
3502 }
3503
3504 if (newcon->flags & CON_NBCON) {
3505 /*
3506 * Ensure the nbcon console buffers can be allocated
3507 * before modifying any global data.
3508 */
3509 if (!nbcon_alloc(newcon))
3510 goto unlock;
3511 }
3512
3513 /*
3514 * See if we want to enable this console driver by default.
3515 *
3516 * Nope when a console is preferred by the command line, device
3517 * tree, or SPCR.
3518 *
3519 * The first real console with tty binding (driver) wins. More
3520 * consoles might get enabled before the right one is found.
3521 *
3522 * Note that a console with tty binding will have CON_CONSDEV
3523 * flag set and will be first in the list.
3524 */
3525 if (preferred_console < 0 && !console_set_on_cmdline) {
3526 if (hlist_empty(&console_list) || !console_first()->device ||
3527 console_first()->flags & CON_BOOT) {
3528 try_enable_default_console(newcon);
3529 }
3530 }
3531
3532 /* See if this console matches one we selected on the command line */
3533 err = try_enable_preferred_console(newcon, true);
3534
3535 /* If not, try to match against the platform default(s) */
3536 if (err == -ENOENT)
3537 err = try_enable_preferred_console(newcon, false);
3538
3539 /* printk() messages are not printed to the Braille console. */
3540 if (err || newcon->flags & CON_BRL) {
3541 if (newcon->flags & CON_NBCON)
3542 nbcon_free(newcon);
3543 goto unlock;
3544 }
3545
3546 /*
3547 * If we have a bootconsole, and are switching to a real console,
3548 * don't print everything out again, since when the boot console, and
3549 * the real console are the same physical device, it's annoying to
3550 * see the beginning boot messages twice
3551 */
3552 if (bootcon_registered &&
3553 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
3554 newcon->flags &= ~CON_PRINTBUFFER;
3555 }
3556
3557 newcon->dropped = 0;
3558 console_init_seq(newcon, bootcon_registered);
3559
3560 if (newcon->flags & CON_NBCON)
3561 nbcon_init(newcon);
3562
3563 /*
3564 * Put this console in the list - keep the
3565 * preferred driver at the head of the list.
3566 */
3567 if (hlist_empty(&console_list)) {
3568 /* Ensure CON_CONSDEV is always set for the head. */
3569 newcon->flags |= CON_CONSDEV;
3570 hlist_add_head_rcu(&newcon->node, &console_list);
3571
3572 } else if (newcon->flags & CON_CONSDEV) {
3573 /* Only the new head can have CON_CONSDEV set. */
3574 console_srcu_write_flags(console_first(), console_first()->flags & ~CON_CONSDEV);
3575 hlist_add_head_rcu(&newcon->node, &console_list);
3576
3577 } else {
3578 hlist_add_behind_rcu(&newcon->node, console_list.first);
3579 }
3580
3581 /*
3582 * No need to synchronize SRCU here! The caller does not rely
3583 * on all contexts being able to see the new console before
3584 * register_console() completes.
3585 */
3586
3587 console_sysfs_notify();
3588
3589 /*
3590 * By unregistering the bootconsoles after we enable the real console
3591 * we get the "console xxx enabled" message on all the consoles -
3592 * boot consoles, real consoles, etc - this is to ensure that end
3593 * users know there might be something in the kernel's log buffer that
3594 * went to the bootconsole (that they do not see on the real console)
3595 */
3596 con_printk(KERN_INFO, newcon, "enabled\n");
3597 if (bootcon_registered &&
3598 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
3599 !keep_bootcon) {
3600 struct hlist_node *tmp;
3601
3602 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3603 if (con->flags & CON_BOOT)
3604 unregister_console_locked(con);
3605 }
3606 }
3607 unlock:
3608 console_list_unlock();
3609 }
3610 EXPORT_SYMBOL(register_console);
3611
3612 /* Must be called under console_list_lock(). */
unregister_console_locked(struct console * console)3613 static int unregister_console_locked(struct console *console)
3614 {
3615 int res;
3616
3617 lockdep_assert_console_list_lock_held();
3618
3619 con_printk(KERN_INFO, console, "disabled\n");
3620
3621 res = _braille_unregister_console(console);
3622 if (res < 0)
3623 return res;
3624 if (res > 0)
3625 return 0;
3626
3627 /* Disable it unconditionally */
3628 console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3629
3630 if (!console_is_registered_locked(console))
3631 return -ENODEV;
3632
3633 hlist_del_init_rcu(&console->node);
3634
3635 /*
3636 * <HISTORICAL>
3637 * If this isn't the last console and it has CON_CONSDEV set, we
3638 * need to set it on the next preferred console.
3639 * </HISTORICAL>
3640 *
3641 * The above makes no sense as there is no guarantee that the next
3642 * console has any device attached. Oh well....
3643 */
3644 if (!hlist_empty(&console_list) && console->flags & CON_CONSDEV)
3645 console_srcu_write_flags(console_first(), console_first()->flags | CON_CONSDEV);
3646
3647 /*
3648 * Ensure that all SRCU list walks have completed. All contexts
3649 * must not be able to see this console in the list so that any
3650 * exit/cleanup routines can be performed safely.
3651 */
3652 synchronize_srcu(&console_srcu);
3653
3654 if (console->flags & CON_NBCON)
3655 nbcon_free(console);
3656
3657 console_sysfs_notify();
3658
3659 if (console->exit)
3660 res = console->exit(console);
3661
3662 return res;
3663 }
3664
unregister_console(struct console * console)3665 int unregister_console(struct console *console)
3666 {
3667 int res;
3668
3669 console_list_lock();
3670 res = unregister_console_locked(console);
3671 console_list_unlock();
3672 return res;
3673 }
3674 EXPORT_SYMBOL(unregister_console);
3675
3676 /**
3677 * console_force_preferred_locked - force a registered console preferred
3678 * @con: The registered console to force preferred.
3679 *
3680 * Must be called under console_list_lock().
3681 */
console_force_preferred_locked(struct console * con)3682 void console_force_preferred_locked(struct console *con)
3683 {
3684 struct console *cur_pref_con;
3685
3686 if (!console_is_registered_locked(con))
3687 return;
3688
3689 cur_pref_con = console_first();
3690
3691 /* Already preferred? */
3692 if (cur_pref_con == con)
3693 return;
3694
3695 /*
3696 * Delete, but do not re-initialize the entry. This allows the console
3697 * to continue to appear registered (via any hlist_unhashed_lockless()
3698 * checks), even though it was briefly removed from the console list.
3699 */
3700 hlist_del_rcu(&con->node);
3701
3702 /*
3703 * Ensure that all SRCU list walks have completed so that the console
3704 * can be added to the beginning of the console list and its forward
3705 * list pointer can be re-initialized.
3706 */
3707 synchronize_srcu(&console_srcu);
3708
3709 con->flags |= CON_CONSDEV;
3710 WARN_ON(!con->device);
3711
3712 /* Only the new head can have CON_CONSDEV set. */
3713 console_srcu_write_flags(cur_pref_con, cur_pref_con->flags & ~CON_CONSDEV);
3714 hlist_add_head_rcu(&con->node, &console_list);
3715 }
3716 EXPORT_SYMBOL(console_force_preferred_locked);
3717
3718 /*
3719 * Initialize the console device. This is called *early*, so
3720 * we can't necessarily depend on lots of kernel help here.
3721 * Just do some early initializations, and do the complex setup
3722 * later.
3723 */
console_init(void)3724 void __init console_init(void)
3725 {
3726 int ret;
3727 initcall_t call;
3728 initcall_entry_t *ce;
3729
3730 /* Setup the default TTY line discipline. */
3731 n_tty_init();
3732
3733 /*
3734 * set up the console device so that later boot sequences can
3735 * inform about problems etc..
3736 */
3737 ce = __con_initcall_start;
3738 trace_initcall_level("console");
3739 while (ce < __con_initcall_end) {
3740 call = initcall_from_entry(ce);
3741 trace_initcall_start(call);
3742 ret = call();
3743 trace_initcall_finish(call, ret);
3744 ce++;
3745 }
3746 }
3747
3748 /*
3749 * Some boot consoles access data that is in the init section and which will
3750 * be discarded after the initcalls have been run. To make sure that no code
3751 * will access this data, unregister the boot consoles in a late initcall.
3752 *
3753 * If for some reason, such as deferred probe or the driver being a loadable
3754 * module, the real console hasn't registered yet at this point, there will
3755 * be a brief interval in which no messages are logged to the console, which
3756 * makes it difficult to diagnose problems that occur during this time.
3757 *
3758 * To mitigate this problem somewhat, only unregister consoles whose memory
3759 * intersects with the init section. Note that all other boot consoles will
3760 * get unregistered when the real preferred console is registered.
3761 */
printk_late_init(void)3762 static int __init printk_late_init(void)
3763 {
3764 struct hlist_node *tmp;
3765 struct console *con;
3766 int ret;
3767
3768 console_list_lock();
3769 hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3770 if (!(con->flags & CON_BOOT))
3771 continue;
3772
3773 /* Check addresses that might be used for enabled consoles. */
3774 if (init_section_intersects(con, sizeof(*con)) ||
3775 init_section_contains(con->write, 0) ||
3776 init_section_contains(con->read, 0) ||
3777 init_section_contains(con->device, 0) ||
3778 init_section_contains(con->unblank, 0) ||
3779 init_section_contains(con->data, 0)) {
3780 /*
3781 * Please, consider moving the reported consoles out
3782 * of the init section.
3783 */
3784 pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3785 con->name, con->index);
3786 unregister_console_locked(con);
3787 }
3788 }
3789 console_list_unlock();
3790
3791 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3792 console_cpu_notify);
3793 WARN_ON(ret < 0);
3794 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3795 console_cpu_notify, NULL);
3796 WARN_ON(ret < 0);
3797 printk_sysctl_init();
3798 return 0;
3799 }
3800 late_initcall(printk_late_init);
3801
3802 #if defined CONFIG_PRINTK
3803 /* If @con is specified, only wait for that console. Otherwise wait for all. */
__pr_flush(struct console * con,int timeout_ms,bool reset_on_progress)3804 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
3805 {
3806 unsigned long timeout_jiffies = msecs_to_jiffies(timeout_ms);
3807 unsigned long remaining_jiffies = timeout_jiffies;
3808 struct console *c;
3809 u64 last_diff = 0;
3810 u64 printk_seq;
3811 short flags;
3812 int cookie;
3813 u64 diff;
3814 u64 seq;
3815
3816 might_sleep();
3817
3818 seq = prb_next_reserve_seq(prb);
3819
3820 /* Flush the consoles so that records up to @seq are printed. */
3821 console_lock();
3822 console_unlock();
3823
3824 for (;;) {
3825 unsigned long begin_jiffies;
3826 unsigned long slept_jiffies;
3827
3828 diff = 0;
3829
3830 /*
3831 * Hold the console_lock to guarantee safe access to
3832 * console->seq. Releasing console_lock flushes more
3833 * records in case @seq is still not printed on all
3834 * usable consoles.
3835 */
3836 console_lock();
3837
3838 cookie = console_srcu_read_lock();
3839 for_each_console_srcu(c) {
3840 if (con && con != c)
3841 continue;
3842
3843 flags = console_srcu_read_flags(c);
3844
3845 /*
3846 * If consoles are not usable, it cannot be expected
3847 * that they make forward progress, so only increment
3848 * @diff for usable consoles.
3849 */
3850 if (!console_is_usable(c))
3851 continue;
3852
3853 if (flags & CON_NBCON) {
3854 printk_seq = nbcon_seq_read(c);
3855 } else {
3856 printk_seq = c->seq;
3857 }
3858
3859 if (printk_seq < seq)
3860 diff += seq - printk_seq;
3861 }
3862 console_srcu_read_unlock(cookie);
3863
3864 if (diff != last_diff && reset_on_progress)
3865 remaining_jiffies = timeout_jiffies;
3866
3867 console_unlock();
3868
3869 /* Note: @diff is 0 if there are no usable consoles. */
3870 if (diff == 0 || remaining_jiffies == 0)
3871 break;
3872
3873 /* msleep(1) might sleep much longer. Check time by jiffies. */
3874 begin_jiffies = jiffies;
3875 msleep(1);
3876 slept_jiffies = jiffies - begin_jiffies;
3877
3878 remaining_jiffies -= min(slept_jiffies, remaining_jiffies);
3879
3880 last_diff = diff;
3881 }
3882
3883 return (diff == 0);
3884 }
3885
3886 /**
3887 * pr_flush() - Wait for printing threads to catch up.
3888 *
3889 * @timeout_ms: The maximum time (in ms) to wait.
3890 * @reset_on_progress: Reset the timeout if forward progress is seen.
3891 *
3892 * A value of 0 for @timeout_ms means no waiting will occur. A value of -1
3893 * represents infinite waiting.
3894 *
3895 * If @reset_on_progress is true, the timeout will be reset whenever any
3896 * printer has been seen to make some forward progress.
3897 *
3898 * Context: Process context. May sleep while acquiring console lock.
3899 * Return: true if all usable printers are caught up.
3900 */
pr_flush(int timeout_ms,bool reset_on_progress)3901 static bool pr_flush(int timeout_ms, bool reset_on_progress)
3902 {
3903 return __pr_flush(NULL, timeout_ms, reset_on_progress);
3904 }
3905
3906 /*
3907 * Delayed printk version, for scheduler-internal messages:
3908 */
3909 #define PRINTK_PENDING_WAKEUP 0x01
3910 #define PRINTK_PENDING_OUTPUT 0x02
3911
3912 static DEFINE_PER_CPU(int, printk_pending);
3913
wake_up_klogd_work_func(struct irq_work * irq_work)3914 static void wake_up_klogd_work_func(struct irq_work *irq_work)
3915 {
3916 int pending = this_cpu_xchg(printk_pending, 0);
3917
3918 if (pending & PRINTK_PENDING_OUTPUT) {
3919 /* If trylock fails, someone else is doing the printing */
3920 if (console_trylock())
3921 console_unlock();
3922 }
3923
3924 if (pending & PRINTK_PENDING_WAKEUP)
3925 wake_up_interruptible(&log_wait);
3926 }
3927
3928 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3929 IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3930
__wake_up_klogd(int val)3931 static void __wake_up_klogd(int val)
3932 {
3933 if (!printk_percpu_data_ready())
3934 return;
3935
3936 preempt_disable();
3937 /*
3938 * Guarantee any new records can be seen by tasks preparing to wait
3939 * before this context checks if the wait queue is empty.
3940 *
3941 * The full memory barrier within wq_has_sleeper() pairs with the full
3942 * memory barrier within set_current_state() of
3943 * prepare_to_wait_event(), which is called after ___wait_event() adds
3944 * the waiter but before it has checked the wait condition.
3945 *
3946 * This pairs with devkmsg_read:A and syslog_print:A.
3947 */
3948 if (wq_has_sleeper(&log_wait) || /* LMM(__wake_up_klogd:A) */
3949 (val & PRINTK_PENDING_OUTPUT)) {
3950 this_cpu_or(printk_pending, val);
3951 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3952 }
3953 preempt_enable();
3954 }
3955
3956 /**
3957 * wake_up_klogd - Wake kernel logging daemon
3958 *
3959 * Use this function when new records have been added to the ringbuffer
3960 * and the console printing of those records has already occurred or is
3961 * known to be handled by some other context. This function will only
3962 * wake the logging daemon.
3963 *
3964 * Context: Any context.
3965 */
wake_up_klogd(void)3966 void wake_up_klogd(void)
3967 {
3968 __wake_up_klogd(PRINTK_PENDING_WAKEUP);
3969 }
3970
3971 /**
3972 * defer_console_output - Wake kernel logging daemon and trigger
3973 * console printing in a deferred context
3974 *
3975 * Use this function when new records have been added to the ringbuffer,
3976 * this context is responsible for console printing those records, but
3977 * the current context is not allowed to perform the console printing.
3978 * Trigger an irq_work context to perform the console printing. This
3979 * function also wakes the logging daemon.
3980 *
3981 * Context: Any context.
3982 */
defer_console_output(void)3983 void defer_console_output(void)
3984 {
3985 /*
3986 * New messages may have been added directly to the ringbuffer
3987 * using vprintk_store(), so wake any waiters as well.
3988 */
3989 __wake_up_klogd(PRINTK_PENDING_WAKEUP | PRINTK_PENDING_OUTPUT);
3990 }
3991
printk_trigger_flush(void)3992 void printk_trigger_flush(void)
3993 {
3994 defer_console_output();
3995 }
3996
vprintk_deferred(const char * fmt,va_list args)3997 int vprintk_deferred(const char *fmt, va_list args)
3998 {
3999 return vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
4000 }
4001
_printk_deferred(const char * fmt,...)4002 int _printk_deferred(const char *fmt, ...)
4003 {
4004 va_list args;
4005 int r;
4006
4007 va_start(args, fmt);
4008 r = vprintk_deferred(fmt, args);
4009 va_end(args);
4010
4011 return r;
4012 }
4013
4014 /*
4015 * printk rate limiting, lifted from the networking subsystem.
4016 *
4017 * This enforces a rate limit: not more than 10 kernel messages
4018 * every 5s to make a denial-of-service attack impossible.
4019 */
4020 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
4021
__printk_ratelimit(const char * func)4022 int __printk_ratelimit(const char *func)
4023 {
4024 return ___ratelimit(&printk_ratelimit_state, func);
4025 }
4026 EXPORT_SYMBOL(__printk_ratelimit);
4027
4028 /**
4029 * printk_timed_ratelimit - caller-controlled printk ratelimiting
4030 * @caller_jiffies: pointer to caller's state
4031 * @interval_msecs: minimum interval between prints
4032 *
4033 * printk_timed_ratelimit() returns true if more than @interval_msecs
4034 * milliseconds have elapsed since the last time printk_timed_ratelimit()
4035 * returned true.
4036 */
printk_timed_ratelimit(unsigned long * caller_jiffies,unsigned int interval_msecs)4037 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
4038 unsigned int interval_msecs)
4039 {
4040 unsigned long elapsed = jiffies - *caller_jiffies;
4041
4042 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
4043 return false;
4044
4045 *caller_jiffies = jiffies;
4046 return true;
4047 }
4048 EXPORT_SYMBOL(printk_timed_ratelimit);
4049
4050 static DEFINE_SPINLOCK(dump_list_lock);
4051 static LIST_HEAD(dump_list);
4052
4053 /**
4054 * kmsg_dump_register - register a kernel log dumper.
4055 * @dumper: pointer to the kmsg_dumper structure
4056 *
4057 * Adds a kernel log dumper to the system. The dump callback in the
4058 * structure will be called when the kernel oopses or panics and must be
4059 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
4060 */
kmsg_dump_register(struct kmsg_dumper * dumper)4061 int kmsg_dump_register(struct kmsg_dumper *dumper)
4062 {
4063 unsigned long flags;
4064 int err = -EBUSY;
4065
4066 /* The dump callback needs to be set */
4067 if (!dumper->dump)
4068 return -EINVAL;
4069
4070 spin_lock_irqsave(&dump_list_lock, flags);
4071 /* Don't allow registering multiple times */
4072 if (!dumper->registered) {
4073 dumper->registered = 1;
4074 list_add_tail_rcu(&dumper->list, &dump_list);
4075 err = 0;
4076 }
4077 spin_unlock_irqrestore(&dump_list_lock, flags);
4078
4079 return err;
4080 }
4081 EXPORT_SYMBOL_GPL(kmsg_dump_register);
4082
4083 /**
4084 * kmsg_dump_unregister - unregister a kmsg dumper.
4085 * @dumper: pointer to the kmsg_dumper structure
4086 *
4087 * Removes a dump device from the system. Returns zero on success and
4088 * %-EINVAL otherwise.
4089 */
kmsg_dump_unregister(struct kmsg_dumper * dumper)4090 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
4091 {
4092 unsigned long flags;
4093 int err = -EINVAL;
4094
4095 spin_lock_irqsave(&dump_list_lock, flags);
4096 if (dumper->registered) {
4097 dumper->registered = 0;
4098 list_del_rcu(&dumper->list);
4099 err = 0;
4100 }
4101 spin_unlock_irqrestore(&dump_list_lock, flags);
4102 synchronize_rcu();
4103
4104 return err;
4105 }
4106 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
4107
4108 static bool always_kmsg_dump;
4109 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
4110
kmsg_dump_reason_str(enum kmsg_dump_reason reason)4111 const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
4112 {
4113 switch (reason) {
4114 case KMSG_DUMP_PANIC:
4115 return "Panic";
4116 case KMSG_DUMP_OOPS:
4117 return "Oops";
4118 case KMSG_DUMP_EMERG:
4119 return "Emergency";
4120 case KMSG_DUMP_SHUTDOWN:
4121 return "Shutdown";
4122 default:
4123 return "Unknown";
4124 }
4125 }
4126 EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
4127
4128 /**
4129 * kmsg_dump - dump kernel log to kernel message dumpers.
4130 * @reason: the reason (oops, panic etc) for dumping
4131 *
4132 * Call each of the registered dumper's dump() callback, which can
4133 * retrieve the kmsg records with kmsg_dump_get_line() or
4134 * kmsg_dump_get_buffer().
4135 */
kmsg_dump(enum kmsg_dump_reason reason)4136 void kmsg_dump(enum kmsg_dump_reason reason)
4137 {
4138 struct kmsg_dumper *dumper;
4139
4140 rcu_read_lock();
4141 list_for_each_entry_rcu(dumper, &dump_list, list) {
4142 enum kmsg_dump_reason max_reason = dumper->max_reason;
4143
4144 /*
4145 * If client has not provided a specific max_reason, default
4146 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
4147 */
4148 if (max_reason == KMSG_DUMP_UNDEF) {
4149 max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
4150 KMSG_DUMP_OOPS;
4151 }
4152 if (reason > max_reason)
4153 continue;
4154
4155 /* invoke dumper which will iterate over records */
4156 dumper->dump(dumper, reason);
4157 }
4158 rcu_read_unlock();
4159 }
4160
4161 /**
4162 * kmsg_dump_get_line - retrieve one kmsg log line
4163 * @iter: kmsg dump iterator
4164 * @syslog: include the "<4>" prefixes
4165 * @line: buffer to copy the line to
4166 * @size: maximum size of the buffer
4167 * @len: length of line placed into buffer
4168 *
4169 * Start at the beginning of the kmsg buffer, with the oldest kmsg
4170 * record, and copy one record into the provided buffer.
4171 *
4172 * Consecutive calls will return the next available record moving
4173 * towards the end of the buffer with the youngest messages.
4174 *
4175 * A return value of FALSE indicates that there are no more records to
4176 * read.
4177 */
kmsg_dump_get_line(struct kmsg_dump_iter * iter,bool syslog,char * line,size_t size,size_t * len)4178 bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
4179 char *line, size_t size, size_t *len)
4180 {
4181 u64 min_seq = latched_seq_read_nolock(&clear_seq);
4182 struct printk_info info;
4183 unsigned int line_count;
4184 struct printk_record r;
4185 size_t l = 0;
4186 bool ret = false;
4187
4188 if (iter->cur_seq < min_seq)
4189 iter->cur_seq = min_seq;
4190
4191 prb_rec_init_rd(&r, &info, line, size);
4192
4193 /* Read text or count text lines? */
4194 if (line) {
4195 if (!prb_read_valid(prb, iter->cur_seq, &r))
4196 goto out;
4197 l = record_print_text(&r, syslog, printk_time);
4198 } else {
4199 if (!prb_read_valid_info(prb, iter->cur_seq,
4200 &info, &line_count)) {
4201 goto out;
4202 }
4203 l = get_record_print_text_size(&info, line_count, syslog,
4204 printk_time);
4205
4206 }
4207
4208 iter->cur_seq = r.info->seq + 1;
4209 ret = true;
4210 out:
4211 if (len)
4212 *len = l;
4213 return ret;
4214 }
4215 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
4216
4217 /**
4218 * kmsg_dump_get_buffer - copy kmsg log lines
4219 * @iter: kmsg dump iterator
4220 * @syslog: include the "<4>" prefixes
4221 * @buf: buffer to copy the line to
4222 * @size: maximum size of the buffer
4223 * @len_out: length of line placed into buffer
4224 *
4225 * Start at the end of the kmsg buffer and fill the provided buffer
4226 * with as many of the *youngest* kmsg records that fit into it.
4227 * If the buffer is large enough, all available kmsg records will be
4228 * copied with a single call.
4229 *
4230 * Consecutive calls will fill the buffer with the next block of
4231 * available older records, not including the earlier retrieved ones.
4232 *
4233 * A return value of FALSE indicates that there are no more records to
4234 * read.
4235 */
kmsg_dump_get_buffer(struct kmsg_dump_iter * iter,bool syslog,char * buf,size_t size,size_t * len_out)4236 bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
4237 char *buf, size_t size, size_t *len_out)
4238 {
4239 u64 min_seq = latched_seq_read_nolock(&clear_seq);
4240 struct printk_info info;
4241 struct printk_record r;
4242 u64 seq;
4243 u64 next_seq;
4244 size_t len = 0;
4245 bool ret = false;
4246 bool time = printk_time;
4247
4248 if (!buf || !size)
4249 goto out;
4250
4251 if (iter->cur_seq < min_seq)
4252 iter->cur_seq = min_seq;
4253
4254 if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
4255 if (info.seq != iter->cur_seq) {
4256 /* messages are gone, move to first available one */
4257 iter->cur_seq = info.seq;
4258 }
4259 }
4260
4261 /* last entry */
4262 if (iter->cur_seq >= iter->next_seq)
4263 goto out;
4264
4265 /*
4266 * Find first record that fits, including all following records,
4267 * into the user-provided buffer for this dump. Pass in size-1
4268 * because this function (by way of record_print_text()) will
4269 * not write more than size-1 bytes of text into @buf.
4270 */
4271 seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
4272 size - 1, syslog, time);
4273
4274 /*
4275 * Next kmsg_dump_get_buffer() invocation will dump block of
4276 * older records stored right before this one.
4277 */
4278 next_seq = seq;
4279
4280 prb_rec_init_rd(&r, &info, buf, size);
4281
4282 prb_for_each_record(seq, prb, seq, &r) {
4283 if (r.info->seq >= iter->next_seq)
4284 break;
4285
4286 len += record_print_text(&r, syslog, time);
4287
4288 /* Adjust record to store to remaining buffer space. */
4289 prb_rec_init_rd(&r, &info, buf + len, size - len);
4290 }
4291
4292 iter->next_seq = next_seq;
4293 ret = true;
4294 out:
4295 if (len_out)
4296 *len_out = len;
4297 return ret;
4298 }
4299 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
4300
4301 /**
4302 * kmsg_dump_rewind - reset the iterator
4303 * @iter: kmsg dump iterator
4304 *
4305 * Reset the dumper's iterator so that kmsg_dump_get_line() and
4306 * kmsg_dump_get_buffer() can be called again and used multiple
4307 * times within the same dumper.dump() callback.
4308 */
kmsg_dump_rewind(struct kmsg_dump_iter * iter)4309 void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
4310 {
4311 iter->cur_seq = latched_seq_read_nolock(&clear_seq);
4312 iter->next_seq = prb_next_seq(prb);
4313 }
4314 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
4315
4316 /**
4317 * console_replay_all - replay kernel log on consoles
4318 *
4319 * Try to obtain lock on console subsystem and replay all
4320 * available records in printk buffer on the consoles.
4321 * Does nothing if lock is not obtained.
4322 *
4323 * Context: Any context.
4324 */
console_replay_all(void)4325 void console_replay_all(void)
4326 {
4327 if (console_trylock()) {
4328 __console_rewind_all();
4329 /* Consoles are flushed as part of console_unlock(). */
4330 console_unlock();
4331 }
4332 }
4333 #endif
4334
4335 #ifdef CONFIG_SMP
4336 static atomic_t printk_cpu_sync_owner = ATOMIC_INIT(-1);
4337 static atomic_t printk_cpu_sync_nested = ATOMIC_INIT(0);
4338
4339 /**
4340 * __printk_cpu_sync_wait() - Busy wait until the printk cpu-reentrant
4341 * spinning lock is not owned by any CPU.
4342 *
4343 * Context: Any context.
4344 */
__printk_cpu_sync_wait(void)4345 void __printk_cpu_sync_wait(void)
4346 {
4347 do {
4348 cpu_relax();
4349 } while (atomic_read(&printk_cpu_sync_owner) != -1);
4350 }
4351 EXPORT_SYMBOL(__printk_cpu_sync_wait);
4352
4353 /**
4354 * __printk_cpu_sync_try_get() - Try to acquire the printk cpu-reentrant
4355 * spinning lock.
4356 *
4357 * If no processor has the lock, the calling processor takes the lock and
4358 * becomes the owner. If the calling processor is already the owner of the
4359 * lock, this function succeeds immediately.
4360 *
4361 * Context: Any context. Expects interrupts to be disabled.
4362 * Return: 1 on success, otherwise 0.
4363 */
__printk_cpu_sync_try_get(void)4364 int __printk_cpu_sync_try_get(void)
4365 {
4366 int cpu;
4367 int old;
4368
4369 cpu = smp_processor_id();
4370
4371 /*
4372 * Guarantee loads and stores from this CPU when it is the lock owner
4373 * are _not_ visible to the previous lock owner. This pairs with
4374 * __printk_cpu_sync_put:B.
4375 *
4376 * Memory barrier involvement:
4377 *
4378 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4379 * then __printk_cpu_sync_put:A can never read from
4380 * __printk_cpu_sync_try_get:B.
4381 *
4382 * Relies on:
4383 *
4384 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4385 * of the previous CPU
4386 * matching
4387 * ACQUIRE from __printk_cpu_sync_try_get:A to
4388 * __printk_cpu_sync_try_get:B of this CPU
4389 */
4390 old = atomic_cmpxchg_acquire(&printk_cpu_sync_owner, -1,
4391 cpu); /* LMM(__printk_cpu_sync_try_get:A) */
4392 if (old == -1) {
4393 /*
4394 * This CPU is now the owner and begins loading/storing
4395 * data: LMM(__printk_cpu_sync_try_get:B)
4396 */
4397 return 1;
4398
4399 } else if (old == cpu) {
4400 /* This CPU is already the owner. */
4401 atomic_inc(&printk_cpu_sync_nested);
4402 return 1;
4403 }
4404
4405 return 0;
4406 }
4407 EXPORT_SYMBOL(__printk_cpu_sync_try_get);
4408
4409 /**
4410 * __printk_cpu_sync_put() - Release the printk cpu-reentrant spinning lock.
4411 *
4412 * The calling processor must be the owner of the lock.
4413 *
4414 * Context: Any context. Expects interrupts to be disabled.
4415 */
__printk_cpu_sync_put(void)4416 void __printk_cpu_sync_put(void)
4417 {
4418 if (atomic_read(&printk_cpu_sync_nested)) {
4419 atomic_dec(&printk_cpu_sync_nested);
4420 return;
4421 }
4422
4423 /*
4424 * This CPU is finished loading/storing data:
4425 * LMM(__printk_cpu_sync_put:A)
4426 */
4427
4428 /*
4429 * Guarantee loads and stores from this CPU when it was the
4430 * lock owner are visible to the next lock owner. This pairs
4431 * with __printk_cpu_sync_try_get:A.
4432 *
4433 * Memory barrier involvement:
4434 *
4435 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4436 * then __printk_cpu_sync_try_get:B reads from __printk_cpu_sync_put:A.
4437 *
4438 * Relies on:
4439 *
4440 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4441 * of this CPU
4442 * matching
4443 * ACQUIRE from __printk_cpu_sync_try_get:A to
4444 * __printk_cpu_sync_try_get:B of the next CPU
4445 */
4446 atomic_set_release(&printk_cpu_sync_owner,
4447 -1); /* LMM(__printk_cpu_sync_put:B) */
4448 }
4449 EXPORT_SYMBOL(__printk_cpu_sync_put);
4450 #endif /* CONFIG_SMP */
4451